Split (3)

train · 4.35k rows

article stringlengths 1.98k 169k	summary stringlengths 1.01k 4.15k	section_headings listlengths 2 38	keywords listlengths 0 12	year stringclasses 11 values	title stringlengths 30 189
ApoE4 genotype is the most prevalent and also clinically most important risk factor for late-onset Alzheimer’s disease ( AD ) . Available evidence suggests that the root cause for this increased risk is a trafficking defect at the level of the early endosome . ApoE4 differs from the most common ApoE3 isoform by a single amino acid that increases its isoelectric point and promotes unfolding of ApoE4 upon endosomal vesicle acidification . We found that pharmacological and genetic inhibition of NHE6 , the primary proton leak channel in the early endosome , in rodents completely reverses the ApoE4-induced recycling block of the ApoE receptor Apoer2/Lrp8 and the AMPA- and NMDA-type glutamate receptors that are regulated by , and co-endocytosed in a complex with , Apoer2 . Moreover , NHE6 inhibition restores the Reelin-mediated modulation of excitatory synapses that is impaired by ApoE4 . Our findings suggest a novel potential approach for the prevention of late-onset AD . Over the course of the last 30 years , we have learned much about the genetics of Alzheimer’s disease ( AD ) , yet the pathological mechanisms that cause the onset of the disease and that define its progression culminating in massive neurodegeneration and debilitating dementia remain poorly understood . We know that amyloid-β ( Aβ ) , an aggregation-prone proteolytic processing product that consists of juxtamembrane sequences and approximately 2/3 of the transmembrane segment of the larger amyloid precursor protein ( APP ) , plays a central role early on , while during later stages of the disease – through mechanisms that are completely obscure – the microtubule-associated protein τ begins to form intraneuronal aggregates , that is the neurofibrillary tangles , that can spread transsynaptically ( Selkoe and Hardy , 2016 ) . It is this τ-aggregation , not the amyloid plaques , which occur early in the disease process , that is thought to be primarily responsible for the neuronal cell death and brain mass loss and that most closely tracks with dementia progression . Before they become visible as the telltale plaques and tangles that cement the AD diagnosis for the pathologist , Aβ and τ are present as smaller oligomeric aggregates that can directly and profoundly impair neuronal functions , by disrupting synaptic Ca2+ homeostasis ( Kuchibhotla et al . , 2008 ) . The resulting synaptic dysfunction is widely thought to represent the earliest stage of the AD pathogenic process and to be a major cause of its clinical manifestation as mild cognitive impairment ( MCI ) ( Palop and Mucke , 2010; Shankar and Walsh , 2009 ) . Although AD almost certainly involves early Aβ accumulation , only a vanishingly small number of individuals suffering from the vicious and dominant early-onset form of AD carry mutations in APP ( Campion et al . , 1999 ) . These missense mutations invariably increase Aβ production and lead to early plaque deposition , while a similarly small number of people have a genetic variation in APP that lowers Aβ production and hence protects from the much more frequent late-onset form of AD ( LOAD ) ( Jonsson et al . , 2012 ) . Numerous other genetic determinants contribute to the bulk of LOAD , which typically develops after the 6th decade . The most important of these is Apolipoprotein E ε4 ( ApoE4 ) genotype ( Corder et al . , 1993; Strittmatter et al . , 1993 ) . ApoE is a lipid and cholesterol carrying protein that is primarily produced by the liver and is responsible for plasma lipid homeostasis ( Mahley , 1988 ) . It occurs in three major isoforms in humans known as ApoE2 , ApoE3 and ApoE4 , with ApoE3 being the most frequent allele ( ~77% homozygosity ) followed by ApoE4 ( ~15 – 20% allele frequency ) which is present in >50% of LOAD ( Liu et al . , 2013 ) . The effect of ApoE4 on Aβ accumulation through impaired Aβ turnover , increased aggregation and thus plaque formation is allele dosage-dependent and this can partly explain its effect on the earlier age of disease onset ( Corder et al . , 1993 ) . However , ApoE4 can independently impair synapse function and Ca2+ homeostasis by disrupting the endocytic transport and recycling of synaptic ApoE receptors and the excitatory AMPA and NMDA type glutamate receptors that are regulated by those ApoE receptors and that are consequently trapped with them in the same vesicles ( Chen et al . , 2010 ) . Most ApoE receptors , which are all members of the low-density lipoprotein ( LDL ) receptor gene family , are expressed in the brain and several are intrinsic components of excitatory synapses where they are present in the presynaptic and postsynaptic compartments ( reviewed in Lane-Donovan et al . , 2014; Pohlkamp et al . , 2017 ) . Of these , ApoE receptor-2 ( Apoer2 , a . k . a . LRP8 ) is the best characterized . It is present both pre- as well as postsynaptically where it primarily functions as a receptor for Reelin ( Bal et al . , 2013; Beffert et al . , 2005; Lane-Donovan and Herz , 2017 ) . Reelin is a large secreted protein that is essential for the formation of cortical layers during embryonic brain development where it serves as a guidance molecule in the regulation of neuronal migration ( D'Arcangelo et al . , 1995; Del Río et al . , 1997 ) . As the brain continues to develop and mature postnatally , its expression pattern changes and Reelin is now produced by a subset of GABAergic interneurons that are interspersed throughout the neocortex and the hippocampus ( Alcántara et al . , 1998; Pesold et al . , 1998; Pohlkamp et al . , 2014 ) . In the adult brain , this secreted Reelin now functions as a neuromodulator by signaling through Apoer2 and its closely related family member Vldlr to activate Src-family tyrosine kinases directly in the synapse , which results in increased Ca2+ influx through NMDA receptors and thus the robust elevation and maintenance of synaptic potentiation ( Chen et al . , 2005; Hiesberger et al . , 1999; Wasser and Herz , 2017 ) . This is the key event in the maintenance of synaptic homeostasis that is impaired by ApoE4 and which occurs independent of Aβ accumulation ( Chen et al . , 2005 ) . Indeed , ApoE4-specific alterations in brain structure have been found in <2 year old children ( Dean et al . , 2014; Shaw et al . , 2007 ) . The molecular basis by which ApoE4 causes the disruption of normal endosomal vesicle transport and recycling is most likely the result of its propensity to unfold and assume a ‘molten-globule’ conformation upon entering an acidic environment ( Morrow et al . , 2002 ) . ApoE4 differs from ApoE3 by a single amino acid , which alters its isoelectric point to coincide with the pH of ~6 . 5 that is present in the early endosome ( Casey et al . , 2010; Ordovas et al . , 1987 ) . We hypothesized that this isoelectric charge neutralization would make ApoE4 prone to aggregation , which could be the molecular basis for the ApoE4-induced and gene dosage-dependent recycling defect . pH in the early endosome is maintained by the opposing functions of the proton pump , which decreases vesicular pH , and the Na+/H+ exchanger NHE6 , which increases it ( Fuster and Alexander , 2014 ) . Here , we have investigated the role of NHE6 inhibition as a means of lowering endosomal pH , away from the isoelectric point of ApoE4 . We found that this simple pharmacological intervention releases the endosomal ApoE4 block , restores the normal trafficking of ApoE receptors and glutamate receptors in neurons and corrects the functional defects in vitro and in vivo . These findings suggest NHE6 inhibition as a novel rational therapeutic approach for reversing the AD risk imposed by ApoE4 . ApoE generically interacts with cysteine-rich ligand binding-type repeats that are ubiquitously present in all LDL receptor family members ( Blacklow , 2007 ) . We first investigated the interaction between ApoE and its ligand-receptor Apoer2 using a solid phase interaction assay . For this purpose , we used naturally secreted ApoE particles containing the three common ApoE isoforms in humans ( ApoE2 , ApoE3 and ApoE4 ) . We found that ApoE3 and ApoE4 strongly interact with Apoer2 , whereas ApoE2 binding was much weaker ( Figure 1A and B ) . These results are consistent with the established similarly high-affinity binding of ApoE3 and ApoE4 to the LDL receptor and the 100-fold reduced affinity of ApoE2 ( Rall and Mahley , 1992; Weisgraber , 1994 ) . Thus , Apoer2 is a high-affinity ApoE receptor and immunofluorescence analysis of primary neurons treated with GFP-tagged ApoE3 accordingly showed co-localization of Apoer2 and ApoE3 in endosomes ( Figure 1C , Videos 1 and 2 ) . We have previously reported that the presence of receptor binding competent ApoE4 particles at physiological concentrations impairs the recycling and consequently surface expression of neuronal Apoer2 ( Chen et al . , 2010 ) . To test whether other unrelated receptors that do not interact with ApoE may be also affected in their plasma membrane expression and recycling properties , we performed the cell surface biotinylation experiments outlined in Figure 2A . Briefly , primary rat cortical neurons were incubated in the absence ( Figure 2B , lane 1 ) or presence of cell-derived , naturally secreted recombinant ApoE3 ( lane 2 ) or ApoE4 ( lane 3 ) particles for 1 hr at 37˚C , purified Reelin was then added to induce the rapid endocytosis of Apoer2 , and after 30 min the cells were transferred to 4˚C and washed with ice-cold PBS . Cell surface biotinylation was performed , biotinylated proteins were isolated and detected by immunoblotting . While Apoer2 quickly recycled in the presence of ApoE3 , its reappearance on the cell surface was greatly delayed in the presence of ApoE4 ( Figure 2B , C ) . By contrast , other endocytic cell surface receptors that do not bind ApoE , such as the insulin receptor ( IR ) or the transferrin receptor ( TfR ) , or that do not interact with Reelin and therefore do not undergo ligand-induced endocytosis ( low-density lipoprotein receptor-related protein 1 ( Lrp1 ) and low-density lipoprotein receptor ( Ldlr ) ) , were not significantly affected by the presence of ApoE ( Figure 2B , C ) . The delayed recycling of Apoer2 was also apparent by the prolonged retention of cell-derived ApoE4 compared to cell-derived ApoE3 in Apoer2-containing intracellular compartments as shown by co-immunoprecipitation of neuronal lysates ( Figure 2D , E ) . That this prolonged retention may be caused by the partial unfolding of ApoE4 is further supported by an experiment in which increasing amounts of naturally secreted , receptor binding-competent ApoE4 particles were added to primary cortical neurons and tyrosine phosphorylation of Dab1 was measured . Dab1 binds to the NPxY motif in the cytoplasmic domain of Apoer2 and when the receptors are clustered , for example by interacting with Reelin , Dab1 undergoes transphosphorylation on tyrosine residues ( Hiesberger et al . , 1999; Howell et al . , 1997 ) . We hypothesized that ApoE4 in its molten-globule state , that is in acidic endosomes , might similarly induce receptor clustering in a dose-dependent manner , whereas ApoE3 would not . When we treated primary neurons with ApoE4 , Dab1 phosphorylation was indeed increased as expected ( Figure 2F , G ) . The pioneering work of Weisgraber and colleagues revealed the propensity of ApoE4 to become structurally labile and undergo transformation to a molten-globule state in a low pH environment , while ApoE2 and ApoE3 were far more resistant to low pH-induced unfolding ( Morrow et al . , 2002 ) . In addition , we noticed that the isoelectric point ( IEP ) of ApoE4 lies close to the pH in the early endosome ( Figure 3A ) . Many proteins are known to lose hydrophilicity near their IEP . Indeed , the first purification of insulin depended upon this biophysical phenomenon ( Wintersteiner and Abramson , 1933 ) . We thus hypothesized that the structural lability of ApoE4 , combined with a reduced solubility in the acidic endosomal environment , might be a driver for the resulting recycling block . pH regulation in the endosome is achieved by a combination of two primary mechanisms: the activity of the proton pump , that is v-ATPase and the exchange of protons for Na+ or K+ through the activity of Na+/H+ exchangers ( NHEs ) ( Figure 3B ) . Functional disruption of the endosomal NHE6 isoform thus lowers endosomal pH ( Brett et al . , 2002 ) and would be predicted to restore ApoE4 solubility and thus vesicle trafficking ( Figure 3C ) . NHE6-specific inhibitors do not presently exist , however , there is a large number of inhibitors for the abundant NHE1 isoform , which is expressed on the plasma membrane of most cell types and which regulates cytosolic pH ( Masereel et al . , 2003 ) . NHE1 inhibitors , such as the epithelial sodium channel blocker and guanidinium derivative amiloride , have been used in clinical practice for half a century as diuretics , and numerous analogues have been developed over the years . We reasoned that some of these analogues might cross-inhibit NHE6 and thus decided to test them in our Apoer2 recycling assay . We found several amiloride analogues to be effective at restoring Apoer2 recycling and chose EMD87580 for detailed analysis ( Figure 4A ) . Increasing concentrations progressively restored normal surface recycling of Apoer2 in the presence of ApoE4 ( Figure 4B , C ) . As shown in Figure 4D and E , physiological concentrations of ApoE3 ( 5 µg/ml ) also impaired Apoer2 recycling to a small , but significant extent . This was also prevented by EMD87580 at the same concentration ( 3 µM ) at which the effect of ApoE4 on Apoer2 recycling was completely neutralized ( Figure 4D , E ) , suggesting that ApoE3 also , although to a minor extent , may lose solubility upon entering the early endosome , and that this is also prevented by pH lowering . To confirm that the effect of the NHE inhibitor was caused by altering the pH and not by an unrelated mode of action we did the converse experiment . We incubated the neurons with ApoE4 in the presence of 3 µM EMD87580 and increasing concentrations of Bafilomycin , an inhibitor of the proton pump . Figure 4F and G show that the resulting increase of the endosomal pH reverses the correction of the ApoE4 recycling deficit by EMD87580 and results in renewed impairment of Apoer2 trafficking . This was further confirmed by an additional experiment in which we incubated the neurons in the presence of a fixed concentration of ApoE4 in the presence or absence of 50 nM Bafilomycin . In the presence of this partial inhibition of proton pump activity , higher concentrations of EMD87580 were required to restore normal Apoer2 recycling ( Figure 4H ) as evident by the right-shift of the dose-response curve ( Figure 4I ) . Taken together , these data show that vesicular pH is the primary driving factor that determines to what extent ApoE4 alters endosomal trafficking . More than 10 different NHEs exist in mammals ( Fuster and Alexander , 2014 ) , which function in intracellular , organellar and extracellular pH regulation in all cells of the body and in a variety of organs . EMD87580 , as most other commercially available NHE inhibitors , was developed with the goal of inhibiting NHE1 ( Chen et al . , 2004 ) and its ability to cross inhibit other NHE forms is unknown . Because ApoE4 blocks Apoer2 and glutamate receptor recycling , and the early endosome is the first acidic organelle ApoE4 encounters during endocytosis , we suspected that the effect of EMD87580 to restore normal receptor trafficking is due to NHE6 inhibition . However , we could not exclude that other vesicular NHEs , in particular the NHE9 , which resides in the Golgi and the late endosome , participate in the release of the recycling block . To determine whether NHE6 inhibition is sufficient to restore normal Apoer2 recycling , we used shRNAs designed against all the vesicular NHEs present in intracellular compartments which ApoE and Apoer2 might encounter during their passage through the recycling pathway . Figure 5A–D shows that of all NHEs ( 1 and 5 through 9 ) that were targeted by shRNA inhibition , only NHE6-specific shRNAs were able to completely restore Apoer2 expression at the plasma membrane in the presence of ApoE4 ( Figure 5A , lane 6 , 5C , lanes 8 , 10 , 12 , and quantified in Panels B and D ) . Three different shRNAs directed against NHE6 were used in Figure 5C and D . None of the shRNAs directed against NHE1 , 5 , 7 , 8 or 9 had any effect on surface Apoer2 expression , neither in the absence of ApoE4 or in its presence . Next , we determined whether shRNA-mediated inhibition of NHE6 function would also restore the normal recycling of AMPA and NMDA-type glutamate receptors in primary neurons treated with ApoE4 and Reelin . Three different shRNAs which efficiently reduced NHE6 protein expression by at least 90% ( Figure 6A ) and a scrambled control shRNA were used in the receptor surface recycling assay ( Figure 6B , lanes 4 – 8 ) . Pharmacological inhibition of NHE function with EMD87580 was used as a control ( lane 3 ) . EMD87580 and the NHE6-specific shRNAs completely restored Apoer2 and glutamate receptor surface expression , while the scrambled shRNA had no effect ( Panel B and quantified in Panels C-F ) . These findings suggested that NHE6 inhibition might be effective in restoring a normal synaptic response in ApoE4 targeted replacement mice , which we have previously shown are completely resistant to long-term potentiation ( LTP ) enhancement by Reelin ( Chen et al . , 2010; Lane-Donovan et al . , 2014 ) . To test if restored glutamate receptor trafficking by NHE6 inhibition improves synapse function , we measured hippocampal LTP in acute slices of human ApoE targeted replacement mice: ApoE3-knockin ( ApoE3-KI ) and ApoE4-knockin ( ApoE4-KI ) . Mice were treated with or without EMD87580 by simultaneous intraperitoneal and intranasal application , the brains were subsequently harvested and electrophysiological field recordings were performed . We chose to complement the intraperitoneal injections with intranasal delivery since EMD87580 , as all existing guanidine-based NHEs , have poor blood-brain-barrier penetration and intranasal delivery of small molecules and peptides including insulin has been shown to increase their biological effect in the brain ( Grassin-Delyle et al . , 2012 ) . Consistent with previous results ( Rönicke et al . , 2009 ) , we found that EMD87580 increased input-output ( I/O ) ratios in the ApoE3-KI mice ( Figure 7A ) . In the ApoE4-KI mice , baseline I/O ratios were higher and did not respond to EMD87580 ( Figure 7B ) . As shown previously ( Durakoglugil et al . , 2009 ) , LTP was increased in ApoE3-KI slices treated with Reelin ( Figure 7C ) . ApoE3-KI slices treated with EMD87580 also showed increased LTP ( Figure 7C ) . Interestingly , Reelin and EMD87580 have no additional synergistic effect and in fact increase LTP to a lesser extent than either EMD or Reelin alone . As shown previously ( Durakoglugil et al . , 2009 ) , Reelin had no effect on LTP in ApoE4-KI slices ( Figure 7D ) . In contrast to ApoE3-KI , ApoE4-KI slices treated with EMD87580 exhibited reduced LTP . Importantly , ApoE4-KI slices with EMD87580 readily responded to Reelin , and LTP was increased . Therefore , on the ApoE4 knockin background , EMD87580 restores electrophysiological parameters comparable to the ApoE3 and wild-type background . These data suggest that endosomal NHE inhibition can neutralize the effect of ApoE4 on vesicle trafficking and concomitant synaptic dysfunction . Can it also reverse the persistent synaptic suppression caused by oligomeric β-amyloid in the presence of ApoE4 ? Aβ42 oligomers potently suppress synaptic potentiation ( Townsend et al . , 2006 ) , but this can be averted by preincubation of hippocampal slices with Reelin , which can by itself potentiate the synapse and thus counteract the Aβ induced suppression ( Durakoglugil et al . , 2009 ) . In Figure 8 , we repeated these experiments again in the absence as well as in the presence of EMD87580 . As we had found previously , AD patient brain extracts containing Aβ oligomers , but not control brain extracts , potently suppressed LTP in hippocampal slices from ApoE3-KI and in ApoE4-KI mice . Reelin prevented this suppression in the ApoE3 slices , while the slices from ApoE4 mice were almost completely resistant to Reelin and LTP remained suppressed ( Figure 8 , solid triangles in Panels A and B ) . By striking contrast , this LTP suppression in the presence of Aβ and Reelin in ApoE4 slices was completely abolished when the slices were perfused with EMD87580 for 4 hr prior to LTP induction . In data not shown here we observed at 30-min preperfusion with EMD87580 a trend toward alleviating the ApoE4-mediated Reelin resistance that was , however , not yet significant . This may suggest that relief of the ApoE4 endosomal recycling block requires some time , perhaps to ‘flush out’ the vesicles that are already stuck and clog up the recycling route . We have used insights into the molecular structure and biophysical properties of ApoE isoforms to develop a novel rational drug identification approach to reverse the increased AD risk inherent to the ApoE4 allele . In earlier studies , we found that ApoE4 impairs endosomal vesicle recycling ( Chen et al . , 2010 ) . While investigating the molecular basis for this trafficking delay , we recognized that the predicted isoelectric point of ApoE4 closely matches the prevailing pH in the early endosome . We hypothesized that ApoE4 , which is known to assume a molten-globule state under low pH conditions , might lose solubility as it enters the lower pH environment of the early endosome . This in turn could impair vesicle propagation through the endosomal recycling pathway and result in the observed sequestration of ApoE receptors and associated excitatory neurotransmitter receptors in cortical neurons ( Chen et al . , 2010 ) . We predicted that changing endosomal pH away from the isoelectric point of ApoE4 should prevent this isoelectric precipitation and resolve the recycling block . Since raising endosomal pH using alkalinizing agents , such as ammonium chloride or chloroquine , is known to arrest endosomal trafficking by preventing lipoprotein release from lipoprotein receptors ( Goldstein et al . , 1985 ) , we investigated approaches to lower endosomal pH instead . Two possible mechanisms for this are i ) activation of the proton pump , or ii ) preventing proton efflux from the endosome by inhibiting the endosomal sodium/potassium hydrogen exchanger NHE6 . Our results now show that pharmacological as well as genetic inhibition of NHE activity in the early endosome is sufficient to completely resolve the ApoE4 induced endosomal recycling block and restore the normal cell surface recycling rate of the synaptic ApoE receptor Apoer2 and the excitatory AMPA- and NMDA-type glutamate receptors that are regulated by Apoer2 and that traffic together with Apoer2 through the endosomal recycling compartments ( illustrated by the model shown in Figure 9 ) . Previously , we and others have shown that ApoE4-KI mice exhibit enhanced LTP while the neuromodulator Reelin has no potentiating effect on LTP expression in this genotype ( Durakoglugil et al . , 2009 ) , suggesting that ApoE4 impairs the physiological response to Reelin . Here , we show that NHE inhibition with EMD87580 in ApoE4-KI mice reverses the ApoE4 induced Reelin resistance ( Figure 7 ) and restores the ability of Reelin to protect against Aβ toxicity ( Figure 8 ) . ApoE3-KI slices treated with EMD87580 exhibit increased LTP consistent with previous findings that have shown that the NHE inhibitor ethyl-isopropyl amiloride ( EIPA ) can enhance theta-burst-induced LTP ( Rönicke et al . , 2009 ) . NHE activity was proposed to be a negative feedback mechanism that can regulate neuronal excitability as well as plasticity . In ApoE4-KI mice , EMD87580 decreased LTP to baseline levels of control ApoE3-KI mice and importantly LTP became now responsive to Reelin-facilitation . This observation suggests that it is the restoration of normal release of ApoE4 from Apoer2 in acidified endosomal compartments and the subsequent normalization of endosomal trafficking that reestablishes optimal synaptic homeostasis in the presence of ApoE4 . Although more than 20 genetic loci that modify the risk for late-onset AD have been discovered to date ( Karch et al . , 2014 ) , ApoE4 genotype is by far the major genetic risk factor for late-onset AD besides aging , affecting almost 1/5th of the human population , and hence it is clinically the most important one . The molecular mechanisms by which ApoE4 imposes this risk remain under debate . Early work following the seminal discovery of this striking genetic association by the Roses group ( Corder et al . , 1993 ) focused on the differential ability of ApoE isoforms to interact with Aβ and affect fibril formation . Efforts in our own laboratory were based on the rationale that ApoE receptors , that is the LDL receptor gene family , are highly likely to be involved in the disease process . This hypothesis was the initial driver that defined a plethora of surprising functions of LDL receptor-related proteins ( LRPs ) in the central and peripheral nervous system ( Bal et al . , 2013; Beffert et al . , 2005; Bell et al . , 2012; Choi et al . , 2013; Kim et al . , 2008; Lane-Donovan and Herz , 2017; Liu et al . , 2013; Liu et al . , 2007; Liu et al . , 2011; May et al . , 2004; Nakajima et al . , 2013; Pohlkamp et al . , 2015; Pohlkamp et al . , 2017; Trommsdorff et al . , 1999; Wasser and Herz , 2017; Wasser et al . , 2014; Weeber et al . , 2002; Zhang et al . , 2008; Zhao et al . , 2017 ) . They included unprecedented roles as direct signal-transducing receptors ( Hiesberger et al . , 1999; Trommsdorff et al . , 1999 ) and regulators of central and peripheral synaptic transmission ( Beffert et al . , 2005; Choi et al . , 2013; Weeber et al . , 2002 ) and have established a strong rationale and mechanistic basis by which ApoE isoforms and ApoE receptors can directly affect synaptic homeostasis , neuronal survival and thus the cognitive impairment and progressive neurodegeneration that underlie LOAD . Presynaptic and postsynaptic vesicle recycling is a central element of synaptic transmission ( Harris et al . , 2012; Kawasaki et al . , 2000; Robinson et al . , 1993; Sontag et al . , 1994; Sudhof , 2004 ) . Intriguingly , enlarged endosomal compartments and impaired endolysosomal functions are also a prominent feature of APP expression , processing and early AD ( Cataldo et al . , 2000; Decourt et al . , 2013; Ishigaki et al . , 2000; Nixon et al . , 2001; Salehi et al . , 2006 ) . Our original finding that ApoE isoforms can differentially impair synapse functions by trapping postsynaptic ( and possibly also presynaptic ) recycling vesicles that contain ApoE receptors was inspired by the original observations of Heeren , Beisiegel and colleagues who first described a prolonged intracellular retention of ApoE4 in a hepatoma cell line ( Heeren et al . , 2004 ) . In a series of experiments , we showed that ApoE4 impairs NMDA receptor activation and neuronal Ca2+ conductance by the neuromodulator and ApoE receptor ligand Reelin . This was caused by the dramatically delayed recycling of the Reelin receptor and regulator of glutamate receptor trafficking Apoer2 in the presence of ApoE4 and , notably , also to a smaller extent by ApoE3 and less by ApoE2 ( Chen et al . , 2010 ) . The dramatically altered recycling kinetics in the presence of ApoE4 lead to an altered state of synaptic homeostasis , which we have termed ‘Reelin resistance’ ( Lane-Donovan et al . , 2014 ) and which prevents the synapse from adequately adapting to the rising levels of synapse-suppressing Aβ oligomers as they accumulate in the aging brain . The compensatory increase in synaptic and network activity ( Palop and Mucke , 2010 ) would further drive Aβ production ( Cirrito et al . , 2008 ) , thereby accelerating a self-reinforcing cycle that would be predicted to contribute to the earlier amyloid accumulation in ApoE4 carriers . Moreover , a second mechanism by which impaired ApoE4-containing vesicle recycling would be predicted to contribute to accelerated amyloid accumulation and plaque deposition ( Fagan et al . , 2002 ) involves the reduced turnover rate of Aβ in the brains of ApoE4 targeted replacement mice ( Castellano et al . , 2011 ) and humans ( Wildsmith et al . , 2012 ) . Recently , the Bu laboratory reported a direct interaction of ApoE with insulin receptors in the brain , which also resulted in their intracellular retention and impaired insulin signaling ( Zhao et al . , 2017 ) . Although in our experiments we did not detect impaired insulin receptor trafficking ( Figure 2B and C ) , and there is also no evidence that ApoE4 carriers are predisposed to the predicted insulin resistance that would be expected to result from intracellular insulin receptor trapping , these data add nevertheless further support to our model which postulates impaired neuronal endosomal vesicle trafficking as the root cause for the increased AD risk imposed by ApoE4 ( Chen et al . , 2005; Chen et al . , 2010; Lane-Donovan and Herz , 2017; Lane-Donovan et al . , 2014 ) . Our data indicate that ApoE4 induces endosomal trafficking deficits . Consistently , alkalinizing drugs , such as ammonium chloride or chloroquine , which increase vesicular pH in the cell , lead to an arrest of endosomal trafficking by preventing lipoprotein release from their receptors ( Goldstein et al . , 1985 ) . We therefore chose to explore the effect of acidification of endosomes as a means to overcome the trafficking impairments caused by ApoE4 . By contrast , a recent study in astrocytes has proposed that the presence of ApoE4 results in endosomal acidification caused by NHE6 reduction , leading to impaired Aβ-clearance ( Prasad and Rao , 2018 ) . These authors further reported that increased NHE6 expression , which would elevate endosomal pH , induced by HDAC-inhibition alleviated the impaired of Aβ-clearance . We show here that Aβ-induced synaptic impairment could be abolished by the NHE inhibitor EMD87580 ( Figure 8 ) . Both studies show that manipulation of endosomal pH can affect the endosomal trafficking of ApoE . We show that in neurons this alters cell surface Apoer2 expression levels and thus synaptic plasticity . Prasad and Rao ( 2018 ) showed that in astrocytes a similar mechanism may mediate Aβ clearance involving the ApoE receptor Lrp1 . The isoelectric point of ApoE4 , but not ApoE2 and ApoE3 , matches the physiological pH present in early endosomes and reduced solubility at or near their isoelectric point is a general property of many proteins including insulin ( Wintersteiner and Abramson , 1933 ) . In contrast to ApoE2 and ApoE3 , the conformation of ApoE4 in the early endosome might be particularly vulnerable , because it alone has been shown to be prone to unfolding under low pH conditions resulting in a molten-globule state ( Morrow et al . , 2002 ) . Along the same lines , ApoE4 increased the number and size of early endosomes in AD patients ( Cataldo et al . , 2000 ) and in neurons derived from induced pluripotent stem cells ( Lin et al . , 2018 ) . These data suggest that manipulating endosomal pH represents a promising therapeutic target to improve endosomal trafficking deficits induced by ApoE4 . In order to fully address the physiological functions of NHE6 and understand its role in ApoE4 induced neurodegenerative processes , it will be necessary to develop NHE6 specific inhibitors that can pharmacologically modulate NHE6 - as opposed to switching it on or off in a binary fashion - to achieve the optimal pH for ApoE4 trafficking without undue interference with essential cellular transport processes or the function of other NHEs . Such inhibitors must also be capable of crossing the blood brain barrier , something the presently available non-specific inhibitor classes are incapable of doing . In summary , we have shown that the conformational change of ApoE4 to a molten-globule state ( Morrow et al . , 2002 ) in a low pH environment , which correlates with the impaired endosomal vesicle recycling in the presence of ApoE4 ( Chen et al . , 2010; Heeren et al . , 2004; Lane-Donovan and Herz , 2017; Lane-Donovan et al . , 2014; Rellin et al . , 2008 ) , can be reversed by changing endosomal pH ( Herz et al . , 2010 ) . As a mechanistic basis we propose the propensity of many proteins to lose hydrophilicity at or near their isoelectric point , which makes them prone to aggregation and precipitation , a seminal discovery that made the purification of insulin on an industrial scale possible ( Wintersteiner and Abramson , 1933 ) . By altering endosomal pH , ApoE4 maintains its solubility and the recycling block is avoided . This simple biophysical property can explain in a straightforward manner many observations by which ApoE isoforms differentially affect neuronal functions and AD-relevant mechanisms . Our findings also point toward NHE6-specific inhibitors as a rational basis for a novel approach to erase the AD risk imposed by ApoE4 . Request for reagents should be directed to Joachim Herz ( Joachim . Herz@utsouthwestern . edu ) . ApoE3-KI or ApoE4-KI mice backcrossed to C57BL/6 were generously provided by Nobuyo Maeda and Patrick Sullivan ( Knouff et al . , 1999; Sullivan et al . , 1997 ) . Animals were group-housed on a standard 12 hr light/dark cycle and fed ad libitum standard mouse chow ( Diet 7001; Harlan Teklad , Madison , WI , USA ) . Ethics Statement: All experimental procedures were performed according to the approved guidelines for Institutional Animal Care and Use Committee ( IACUC ) at the University of Texas Southwestern Medical Center at Dallas ( Approval Number: A3472-01; 2015-101088 ) . Deidentified human cortical brain extracts ( IRB exempt ) from a non-AD , normal subject ( Control ) and a clinically and histopathologically confirmed Alzheimer’s disease ( AD ) case were prepared as described previously ( Durakoglugil et al . , 2009 ) . Briefly , control brain extract contained monomeric Aβ , but no detectable oligomers and only a trace amount of higher order aggregates; by contrast , AD brain extract contained monomeric Aβ in addition to Aβ dimers and higher order aggregates . One gram of brain tissue was homogenized in 4 ml of Tris buffered saline and centrifuged at 175 , 000 x g as described by Shankar et al . ( 2008 ) . The supernatant was designated ‘extract’ . The genotype of the brain tissue was ApoE3/3 for the control and ApoE3/4 for the AD tissue . Primary rat ( Sprague-Dawley ) cortical neurons ( E18 ) were prepared as described previously ( Chen et al . , 2005 ) and cultured in six-well plates ( 1 million neurons/9 cm2 ) or on Poly-D-Lysine coated coverslips ( 30 , 000 neurons/1 . 1 cm2 ) in Neurobasal/B27 medium at 37°C and 5% CO2 . At 9–14 days in vitro ( DIV ) primary neurons were used for experiments . All primers used for cloning are listed in the Key Resources Table . pcDNA3 . 1-Apoer2-Fc: The mouse ApoER2-Fc construct ( secreted Apoer2 ectodomain ) , tagged with the V5 epitope and Fc , was described previously ( Hiesberger et al . , 1999 ) . To produce recombinant Apoer2-Fc , HEK 293 cells were transfected and the medium was harvested as described ( Chen et al . , 2010 ) . pLKO . 1-shRNA constructs: pLKO . 1-constructs containing shRNA targeting different NHE subtypes were purchased from Sigma or created by inserting aligned oligos ( for shRNA sequence refer to Key Resources Table ) into the pLKO . 1-TRC , as described elsewhere ( Moffat et al . , 2006 ) . pLVX-mCherry-Apoer2: SalI and XbaI cloning sites were inserted into a plasmid containing the Apoer2 full-length cDNA immediately downstream of the Apoer2 signal peptide ( N-terminus , NT ) by site-directed mutagenesis . In a second step , PCR-amplified mCherry was inserted into the newly created SalI and XbaI sites . NT-mCherry-Apoer2 was then amplified by PCR and cloned into the NheI and EcoRI sites of lentiviral vector that was generated by modifying pLVXCMV100 by removing an NheI site through site-directed mutagenesis and replacing the truncated CMV100 with a full length CMV promoter ( inserted into the ClaI and NheI sites ) . pcDNA3 . 1-ApoE3-GFP: ApoE3 was PCR-amplified and inserted into pEGFP-N1 using EcoR1 and BamHI cloning sites . After treatment , neurons were washed three times with cold PBS , and lysed in RIPA buffer ( 50 mM Tris-HCl , pH 8 . 0; 150 mM NaCl; 1% Nonidet P-40; phosphatase and protease inhibitors ) for 20 min on ice . Cellular debris was removed by centrifugation for 10 min at 14 , 000 rpm and 4°C in an Eppendorff centrifuge . Protein concentrations were measured using the Bradford Protein Assay ( Bio-Rad ) . After adding 4x SDS loading buffer ( 0 . 1 M Tris-HCl , pH 6 . 8 , 2% SDS , 5% β-mercaptoethanol , 10% glycerol , and 0 . 05% bromphenol blue ) the samples were boiled at 95°C for 10 min . For immunoblotting of NHE6 , samples were incubated for 30 min at room temperature instead of boiling . 4x SDS loading buffer was added to the media to detect secreted proteins . After boiling ( 95°C for 10 min ) samples were loaded on SDS-PAGE . Proteins were transferred to a nitrocellulose membrane for western blotting with the indicated antibodies . For imaging DIV9 neurons on 12 mm coverslips were infected with lentiviral DNA encoding NT-mCherry-Apoer2 . Lentivirus containing medium was removed 14 hr after infection . On DIV12 ApoE3-GFP containing supernatant of 293 cells transfected with pcDNA3 . 1-ApoE3-GFP ( FuGENE ) was added . On DIV13 neurons were washed 2x with PBS and fixed with 4% PFA . Coverslips were mounted using Vectashield Antifade Mounting Medium with DAPI . Z-stack images were obtained using a Confocal Zeiss LSM880 Airyscan microscope and a 63x objective and a step size of 1 µm . 3D projections and orthogonal views were generated using NIH Fiji/ImageJ software . Primary neurons were pre-treated for 30 min with ApoE-conditioned medium ( 5 μg/ml unless stated differently ) , then incubated with Reelin ( 2 μg/ml ) for an additional 30 min ( see timeline in Figure 2A ) . After treatment cells were washed with cold PBS and incubated in PBS containing sulfo-NHS-SS-biotin ( 1 . 0 mg/ml ) for 30 min at 4°C . Excess reagent was quenched by rinsing the neurons with cold PBS containing 100 mM glycine . Neurons were lysed in 160 µl/9 cm2 lysis buffer ( PBS with 0 . 1% SDS , 1% Triton X-100 , and protease inhibitors ) at 4°C for 20 min . Cell debris was removed by centrifugation for 10 min at 14 , 000 rpm at 4°C in an Eppendorff centrifuge . The protein concentration was measured using the Bradford Protein Assay ( Bio-Rad ) and 100 µg of total protein was incubated with 50 µl of NeutrAvidin agarose at 4°C for 1 hr . Agarose pellets were washed three times using washing buffer ( 500 mM NaCl; 15 mM Tris-HCl , pH 8 . 0; 0 . 5% Triton X-100 ) , biotinylated surface proteins were eluted from agarose beads by boiling in 2x SDS sample loading buffer and loaded on SDS-PAGE for western blot analysis . For drug treatments , cells were pre-incubated with EMD87580 and/or Bafilomycin for 1 hr prior to ApoE and Reelin addition . 3 μM EMD87580 ( 3 mM stock in PBS ) was used unless stated differently . Apoer2-Fc/ApoE interaction: Protein G Sepharose ( 50 µl ) was added to 1 ml culture supernatant containing ApoER2-Fc from HEK 293 cells at 4°C overnight . Beads were then sedimented by brief centrifugation , and then 500 µl culture supernatant containing ApoE and CaCl2 ( final concentration 1 mM ) was added to the beads . The mixture was incubated for an additional 4 hr at 4°C . Beads were washed three times using washing buffer ( 500 mM NaCl , 15 mM Tris . HCl , 0 . 5% Triton X-100 ( pH8 . 0 ) ) , and bound proteins were separated on 4 – 15% SDS-PAGE and immunoblotted for ApoE . Apoer2-ApoE interaction: ApoE-treated ( 5 µg/ml for 3 hr at 37°C ) neurons were washed with ice cold PBS and lysed in RIPA buffer . For immunoprecipitation 600 µg of lysate were co-incubated with anti-Apoer2 rabbit serum or control rabbit serum and protein A-Sepharose beads at 4°C overnight . Precipitated beads were washed 3x in RIPA buffer , resuspended in 2x SDS sample buffer and boiled at 95°C for 10 min . Eluted proteins were probed for Apoer2 and ApoE . Primary neurons were treated with Reelin ( 2 µg/ml for 30 min ) or ApoE3 or ApoE4 at 37°C for 3 hr . After treatment , neurons were lysed and prepared for western blotting as described above and probed with antibodies raised against total Dab1 and phospho-Tyrosine ( 4G10 ) to identify phospho-Dab1 ( Masereel et al . , 2003 ) . HEK 293 T cells were co-transfected with psPAX2 , pMD2 . G , and the individual transfer constructs ( pLKO . 1-shRNA or pLVX-mCherry-Apoer2 ) . Media was replaced after 12 – 15 hr . Viral particle containing media was collected and cell debris spun down . The virus was concentrated by ultra-centrifugation and re-suspension in DMEM ( 1/10th volume ) . To transfect neurons 100 µl of concentrated virus was added per ml of medium in the culture dish . Transduction media were replaced 12 – 15 hr after infection and neurons were incubated for 3 days before experiments were conducted . For in vivo treatment , mice were intraperitoneally injected with EMD87580 ( 1 mg/ml in PBS ) at a dose of 5 mg/kg . Additionally , mice received intranasal application of 10 µl of 1 mg/ml EMD87580 . Animals were treated twice a day for 2 consecutive days ( in the morning and evening , 12 hr interval ) . On the third day , mice were treated with EMD87580 in the morning and sacrificed 2 hr later for extracellular field recordings of hippocampal slices . Hippocampal slices were prepared from 2 to 3 months old ApoE3-KI or ApoE4-KI mice . Brains were quickly removed and placed in cold high sucrose cutting solution ( 110 mM sucrose , 60 mM NaCl , 3 mM KCl , 1 . 25 mM NaH2PO4 , 28 mM NaHCO3 , 0 . 5 mM CaCl2 , 5 mM glucose , 0 . 6 mM Ascorbic acid , 7 mM MgSO4 ) . 400 µm transverse sections were cut using a vibratome . Slices were then transferred into an incubation chamber containing 50% aCSF ( 124 mM NaCl , 3 mM KCl , 1 . 25 mM NaH2PO4 , 26 mM NaHCO3 , 10 mM D-glucose , 2 mM CaCl2 , 1 mM MgSO4 ) and 50% sucrose cutting solution . For electrophysiological recordings a final concentration of 3 µM EMD87580 was used . For combined treatment with Reelin , AD brain extract and EMD87580 ( Figure 8 ) , slices were pretreated with EMD87580 for 3 hr and that concentration was maintained during recording in the presence or absence of Reelin and/or AD extract . Slices were transferred to the interface recording chamber where they were kept in aCSF at 31°C and a flow rate of 2 – 3 ml/min . In the recording chamber , different combinations of treatment with AD brain extract , Reelin , and EMD87580 were used . Slices were perfused with the different final components for an additional ~30 min until they stabilized in the recording chamber before application of theta burst stimulation and one hour thereafter throughout the recording period . For stimulation concentric bipolar electrodes were used ( FHC , Catalog no CBBRC75 , 1201 Main St Bowdoin , ME 04287 , USA ) and placed into the stratum radiatum . Stimulus intensity was set at 40 – 60% maximum response and delivered through an Isolated Pulse Stimulator ( A-M Systems , Model 2100 . A custom written program in Labview 7 . 0 was used for recording and analysis of LTP experiments ( courtesy of Dr Jay Gibson ) . A theta burst ( TBS; train of 4 pulses at 100 Hz repeated 10 times with 200 ms intervals and again repeated 5 times at 10 s intervals ) was used as conditioning stimulus . For input/output analysis data were binned and fitted linearly . Slopes were calculated using regression analysis . Data were expressed as the mean ± SEM and evaluated using two-tailed Student’s t test for two groups with one variable tested and equal variances , or one-way analysis of variance ( ANOVA ) with Dunnett’s post-hoc for multiple groups with only variable tested . The differences were considered to be significant at p<0 . 05 ( p<0 . 05 , p<0 . 01 , **p<0 . 001 ) .	Alzheimer’s disease is a degenerative condition that destroys connections between brain cells leading to memory loss , confusion and difficulties in thinking . Apolipoprotein E is a protein that carries fatty substances called lipids and cholesterol around the brain , and plays an important role in repair mechanisms . There are three major forms of Apolipoprotein E , and individuals who carry a version known as ApoE4 are up to 10 times more likely to develop Alzheimer’s disease than those who carry other variations . In nerve cells , or neurons , Apolipoprotein E binds to a specific family of receptors . One of these receptors , called Apoer2 , is found in the synaptic gap between neurons , where it regulates their activities . Both Apolipoprotein E and Apoer2 are taken into the cell within compartments known as endosomal vesicles . Usually , the Apoer2 receptor is quickly recycled back to the surface of the cell , but this recycling process is delayed in individuals with the ApoE4 version of Apolipoprotein E . Apoer2 is just one of many different receptors on the surface of neurons that are taken into vesicles before being recycled back to the cell surface . The fluid inside these vesicles becomes progressively more acidic as they move through the cell . This process helps to control the interaction of these receptors with their binding partners and to regulate their movement and recycling . Here , Xian , Pohlkamp et al . investigated whether changing the acidity of vesicles in rat neurons could overcome the block in recycling Apoer2 – and other receptors that travel with Apoer2 in the same compartments – in the presence of ApoE4 . A protein called NHE6 is embedded in the membrane of vesicles called early endosomes and acts to make the vesicles less acidic . Xian , Pohlkamp et al . used drugs to block the activity of NHE6 , which led to the vesicles becoming more acidic and allowed Apoer2 to be recycled faster . Using a genetic approach known as siRNA knockdown to decrease the amount of NHE6 produced in neurons also had a similar effect on Apoer2 recycling . Together these findings suggest that drugs that make vesicles in neurons more acidic may have the potential to help prevent individuals that carry the ApoE4 protein from developing Alzheimer’s disease . Current drugs that target NHE6 also affect other molecules , which can often lead to side effects . A next step will be to develop tailor-made , small molecule drugs that can enter the brain efficiently and selectively block NHE6 .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Reversal of ApoE4-induced recycling block as a novel prevention approach for Alzheimer’s disease
Repetitive proteins are thought to have arisen through the amplification of subdomain-sized peptides . Many of these originated in a non-repetitive context as cofactors of RNA-based replication and catalysis , and required the RNA to assume their active conformation . In search of the origins of one of the most widespread repeat protein families , the tetratricopeptide repeat ( TPR ) , we identified several potential homologs of its repeated helical hairpin in non-repetitive proteins , including the putatively ancient ribosomal protein S20 ( RPS20 ) , which only becomes structured in the context of the ribosome . We evaluated the ability of the RPS20 hairpin to form a TPR fold by amplification and obtained structures identical to natural TPRs for variants with 2–5 point mutations per repeat . The mutations were neutral in the parent organism , suggesting that they could have been sampled in the course of evolution . TPRs could thus have plausibly arisen by amplification from an ancestral helical hairpin . Most present-day proteins arose through the combinatorial shuffling and differentiation of a set of domain prototypes . In many cases , these prototypes can be traced back to the root of cellular life and have since acted as the primary unit of protein evolution ( Anantharaman et al . , 2001; Apic et al . , 2001; Koonin , 2003; Kyrpides et al . , 1999; Orengo and Thornton , 2005; Ponting and Russell , 2002; Ranea et al . , 2006 ) . The mechanisms by which they themselves arose are however still poorly understood . We have proposed that the first folded domains emerged through the repetition , fusion , recombination , and accretion of an ancestral set of peptides , which supported RNA-based replication and catalysis ( the RNA world Bernhardt , 2012; Gilbert , 1986 ) ( Alva et al . , 2015; Lupas et al . , 2001; Söding and Lupas , 2003 ) . Repetition would have been a particularly prominent mechanism by which these peptides yielded folds; six of the ten most populated folds in the Structural Classification of Proteins ( SCOP ) ( Murzin et al . , 1995 ) – including the five most frequent ones – have repetitive structures . In all cases , their amplification from subdomain-sized fragments can also be retraced at the sequence level in at least some of their members . One of these highly populated repetitive folds is the αα-solenoid ( SCOP a . 118 ) , whose most widespread superfamily is the tetratricopeptide repeat ( TPR; a . 118 . 8 ) . This was originally identified as a repeating 34 amino-acid motif in Cdc23p of Saccharomyces cerevisiae ( Sikorski et al . , 1990 ) – hence its name . Since then , TPR-containing proteins have been discovered in all kingdoms of life , where they mediate protein-protein interactions in a broad range of biological processes , such as cell cycle control , transcription , protein translocation , protein folding , signal transduction and innate immunity ( Cortajarena and Regan , 2006; Dunin-Horkawicz et al . , 2014; Katibah et al . , 2014; Keiski et al . , 2010; Kyrpides and Woese , 1998; Lamb et al . , 1995; Sikorski et al . , 1990 ) . The ﬁrst crystal structure of a TPR domain ( Das et al . , 1998 ) showed that the repeat units are helical hairpins , stacked into a continuous , right-handed superhelical architecture with an inner groove that mediates the interaction with target proteins ( Forrer et al . , 2004 ) . The hairpins interact via a specific geometry involving knobs-into-holes packing ( Crick , 1953 ) and burying about 40% of their surface between repeat units . This tightly packed , superhelical arrangement of a repeating structural unit is typical of all αα-solenoid proteins ( Di Domenico et al . , 2014; Kajava , 2012; Kobe and Kajava , 2000 ) . Comparison of TPRs from a variety of proteins reveals a high degree of sequence diversity , with conservation observed mainly in the size of the repeating unit and the hydrophobicity of a few key residues ( D'Andrea and Regan , 2003; Magliery and Regan , 2004 ) . Nevertheless , almost all known TPR-containing proteins can be detected using a single sequence proﬁle ( Karpenahalli et al . , 2007 ) , underscoring their homologous origin . As their name implies , TPR proteins generally contain at least two unit hairpins in a repeated fashion . The few that have only one hairpin , notably the mitochondrial import protein Tom20 ( Abe et al . , 2000 ) , are clearly not ancestral based on their phylogenetic distribution and functionality , implying that the ancestor of the superfamily already had a repeated structure . In searching for the origin of TPRs , we hypothesized that the hairpin at the root of the fold might either have been part of a different , non-repetitive fold or have given rise to both repetitive and non-repetitive folds at the origin of folded domains . Either way we hoped that we might find α-hairpins in non-repetitive proteins that are similar in both sequence and structure to the TPR unit , suggesting a common origin . Here we show that such hairpins are detectable and that one of them , from the ribosomal protein RPS20 ( Schluenzen et al . , 2000 ) , can be customized to yield a TPR fold by repetition , with only a small number of point mutations that are neutral for the parent organism . Ribosomal proteins most likely constitute some of the oldest proteins observable today and are still intimately involved in an RNA-driven process: translation ( Fox , 2010; Hsiao et al . , 2009 ) . They are mostly incapable of assuming their folds outside the ribosomal context ( Peng et al . , 2014 ) and thus belong to a class of intrinsically disordered proteins that become structured upon binding to a macromolecular scaffold ( Dyson and Wright , 2005; Habchi et al . , 2014; Oldfield and Dunker , 2014; Peng et al . , 2014; Varadi et al . , 2014 ) . This hairpin therefore plausibly retains today many of the properties likely to have been present in the ancestral peptide that gave rise to the TPR fold . Repetitive folds with variable numbers of repeats , such as HEAT , LRR , TPR or β-propellers , usually have some members with a high level of sequence identity between their repeat units ( Dunin-Horkawicz et al . , 2014 ) . In these proteins , the units are more similar to each other than to any other unit in the protein sequence database , showing that they were recently amplified . In a detailed study of β-propellers ( Chaudhuri et al . , 2008 ) , we found that this process of amplification and differentiation has been ongoing since the origin of the fold . TPR proteins show a similar evolutionary history . In some proteins , most of the repeats can be seen to have been amplified separately and to a different extent in each ortholog , pointing to their recent origin ( Figure 1a ) ; in others , the amplification must have occurred much earlier , as their ancestor already had fully differentiated repeats ( Figure 1b ) . In recently amplified proteins , such as the ones shown in Figure 1a , within which repeats frequently have >80% pairwise sequence identity , tracking the probable α-hairpin at the root of the amplification is a fairly straightforward proposition . We wondered , however , whether it might be possible to go much further back in time and track the original α-hairpin from which the first TPR protein was amplified . We therefore searched for TPR-like α-hairpins in non-repetitive proteins as present-day descendants of the original hairpin . 10 . 7554/eLife . 16761 . 003Figure 1 . Two evolutionary scenarios for TPRs , illustrated by neighbor-joining phylogenetic trees . ( a ) Amplification from single helical hairpin , as seen in TPR proteins from Cyanobacteria . ( b ) Divergent evolution of a TPR with multiple repeat units , as seen in the TPR domains of Serine/threonine-protein phosphatase 5 ( Ara: Arabidopsis thaliana , Dan: Danio rerio , Hom: Homo sapiens , Mus: Musca domestica , Sac: Saccharomyces cerevisiae , The: Theileria annulata , Xen: Xenopus ( Silurana ) tropicalis ) . Since evolutionary reconstructions are subject to Occam’s razor and reflect the hypothesis with the fewest assumptions , we have postulated here one amplification event from one precursor hairpin . Our findings would however also be fully compatible with the precursor hairpin yielding a population of homologous variants , some of which were independently amplified to TPR-like folds; one or more survivors among these would have become the ancestor ( s ) of today’s TPR proteins . In this more complex scenario , the homology of TPR proteins , which we trace through the comparison of individual hairpins , is still given , but the TPR fold could have arisen from several independent amplifications , and not just a single one . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 00310 . 7554/eLife . 16761 . 004Figure 1—figure supplement 1 . Multiple sequence alignments of recently amplified TPR repeat units . ( a ) Alignments of the TPR units used for the phylogeny in Figure 1a . Residues different from the most common one in each column are shown in bold face and highlighted in yellow . Abbreviations: Ana: Anabaena sp . 90 ( gi: 752818954 , accession: WP_041458168 . 1 ) ; Cal: Calothrix sp . 336/3 ( gi: 821031795 , accession: WP_046815017 . 1 ) ; Cya: Cyanothece sp . PCC 8801 ( gi: 501590504 , accession: WP_012594639 . 1 ) ; fil: filamentous cyanobacterium ESFC-1 ( gi: 740500649 , accession: WP_038331513 . 1 ) ; Mic: Microcystis aeruginosa SPC777 ( gi: 513477764 , accession: EPF24195 . 1 ) . ( b ) The corresponding alignment of the DNA sequences for the most recently amplified TPR units , Cal4-Cal18 , of which the central repeats , Cal9-Cal16 , are fully identical . Synonymous mutations ( highlighted in gray ) are found at less than 1% of the nucleotides , illustrating the recent time point of the amplification . Non-synonymous mutations ( highlighted in yellow ) are about 2 . 5 times as frequent as synonymous ones . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 00410 . 7554/eLife . 16761 . 005Figure 1—figure supplement 2 . Multiple sequence alignments of the three TPR repeat units in serine/threonine-protein phosphatase 5 from seven taxa . Columns with identify ≥80% are highlighted in black and marked by vertical bars ( \| ) ; column with identify <80% but ≥50% are highlighted in gray and marked by dots ( . ) . Abbreviations: Ara: Arabidopsis thaliana ( gi: 18406066 , accession: NP_565985 . 1 ) ; Dan: Danio rerio ( gi: 126158897 , accession: NP_001014372 . 2 ) ; Hom: Homo sapiens ( gi: 5453958 , accession: NP_006238 . 1 ) ; Mus: Musca domestica ( gi: 557765703 , accession: XP_005182549 . 1 ) ; Sac: Saccharomyces cerevisiae S288c ( gi: 398365781 , accession: NP_011639 . 3 ) ; The: Theileria annulata strain Ankara ( gi: 84994100 , accession: XP_951772 . 1 ) ; Xen: Xenopus tropicalis ( gi: 56118654 , accession: NP_001007891 . 1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 005 We had previously developed a proﬁle-based method , named TPRpred , specially designed for the detection of TPRs and related repeat proteins with high sensitivity from sequence data ( Karpenahalli et al . , 2007 ) . Here , in a first step , we used TPRpred to scan protein sequences in the Protein Data Bank ( PDB ) ( Berman et al . , 2000 ) for peptides that share statistically significant similarity to the TPR sequence profile and yet have not been annotated as TPR in Pfam ( Finn et al . , 2014 ) ; we used a p-value cutoff = 1 . 0e−4 , which leads to an estimated false discovery rate of 1 . 0% , see Materials and methods . We ignored tandem repeats in the hit list and focused only on the singleton cases . Subsequently , we compared the structures of these helical hairpin singletons to the average TPR hairpin and removed non-hairpin-like structures . This yielded 31 helical hairpins that are similar to the TPR unit with respect to both sequence and structure . Among them , 22 are part of solenoid-like structures and were discarded . The remaining nine hits belong to three families: ( I ) mitochondrial import receptor subunit Tom20; ( II ) microtubule interacting and transport ( MIT ) domain including katanin ( Iwaya et al . , 2010 ) ; and ( III ) 30S ribosomal protein S20 ( RPS20 ) ( Figure 2 ) . 10 . 7554/eLife . 16761 . 006Figure 2 . TPR-like hairpins found in non-repetitive proteins in the PDB . ( a ) Structure gallery of non-repetitive helical hairpins in the PDB that share both sequence and structure similarity to TPR unit hairpin . Only the 34 amino-acid helical hairpins are shown . The helical hairpins in 30S ribosomal protein s20 ( RPS20 ) , mitochondrial import receptor subunit ( Tom20 ) , and microtubule interacting and transport domain ( MIT ) are depicted in cyan , green , and yellow , respectively . The structure of a TPR with a consensus sequence , CTPR3 , is shown in the center with the middle TPR unit highlighted in red . PDB IDs and chain names of the proteins are given in parentheses . In the superposition , all helical hairpins are superimposed onto the middle TPR unit of CTPR3 . ( b ) Multiple sequence alignment of the helical hairpin sequences listed in ( a ) . The eight TPR signature positions are marked by dots in CTPR3 . Columns with sequence identity ≥ 80% are in black , and columns with sequence identity ≥ 50% are in gray . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 006 The similarity of Tom20 and MIT domains to TPR proteins has been noted before ( Abe et al . , 2000; Iwaya et al . , 2010; Scott et al . , 2005 ) , but the similarity of RPS20 was surprising and drew our attention particularly due to the ancestrality attributed to ribosomal proteins . To further explore the similarity between the helical hairpin in RPS20 ( in short , RPS20-hh ) and TPR , we used TPRpred to rank the RPS20 sequences in Pfam ( Finn et al . , 2014 ) . The top-scoring hit was RPS20-hh from Thermus aquaticus ( NCBI accession number = WP_003044315 . 1 , UniProt id = B7A5L8_THEAQ ) , which matches the TPR unit sequence proﬁle at a p-value of 5 . 4e−07 , almost an order of magnitude better than the second hit ( see Supplementary file 1D ) . Furthermore , we examined the surface residues of RPS20-hh fragments to assess their suitability to occur in a tandem repeat mode , as in TPRs . To this end , we first defined five interface positions on the TPR helical hairpin and transferred the definition to RPS20-hh according to their structure alignment ( positions 3 , 7 , 10 , 21 and 28 using TPR unit numbering ) . Then , we searched for RPS20-hhs with as many hydrophobic residues as possible at these interface positions . We found 42 RPS20-hhs that contain at least three hydrophobic residues out of the five interface positions . Among them , the only RPS20-hh predicted to match the TPR unit proﬁle above a p-value of 1 . 0e−4 was again the RPS20 from T . aquaticus , in which three of the ﬁve interface residues are hydrophobic ( L10 , I21 and V28 ) . We therefore chose this helical hairpin ( RPS20-hhta ) to construct a TPR-like solenoid by amplification ( Figure 3 ) . 10 . 7554/eLife . 16761 . 007Figure 3 . The design of TPR using RPS20 . RPS20-hh is identified by TPRpred to match the sequence profile of TPR units . Their structures are also very similar ( helices are shown as cylinders ) , except for the last four residues ( colored in light and dark magenta ) . We designed a TPR protein using a RPS20-hh with up to five mutations ( yellow strips ) in each repeat unit . The C-terminal loop in the TPR unit ( dark magenta loop ) is used to replace the corresponding C-terminus ( light magenta cylinder ) of RPS20-hh to connect adjacent repeats . The C-terminal helix in RPS20 ( white cylinder ) was used as the stop helix in the design . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 007 We focused on the construction of three-repeat TPRs , which represent the most common form of this fold ( D'Andrea and Regan , 2003; Sawyer et al . , 2013 ) . For instance , 18 of the 54 non-identical TPR domains in the extended Structural Classification of Proteins database ( SCOPe v2 . 05 ) ( Fox et al . , 2014 ) have three repeats . A previously designed three-repeat TPR protein , CTPR3 , was also demonstrated to be highly stable , even more so than natural three-repeat TPR proteins ( Main et al . , 2003b ) . We concatenated three copies of RPS20-hhta as an initial construct , connected by the TPR consensus loop sequence ( DPNN ) . We annotate the two helices in each repeat unit as helix Ai and Bi , where i is the index of the repeat unit ( i = 1 , 2 or 3 ) ( Figure 3 ) . Under the hypothesis of common descent between TPR and RPS20 from the same ancestral peptide and retention of ancestral features in RPS20 , this basic construct would fold as a TPR solenoid with a minimal number of mutations , ideally none . When we experimentally made a construct containing no mutations ( M0 , Table 1 ) , it was soluble but remained unfolded under all conditions tested ( see Section 2 . 4 ) . We therefore introduced point mutations into the sequence of RPS20-hhta , aimed at favoring the target structure . Here , we followed the principle of consensus design ( Forrer et al . , 2004; Main et al . , 2003a ) , which requires the mutation positions to be occupied by the most commonly observed residues in homologous proteins ( Forrer et al . , 2004 ) . Consensus design methods have been successful in engineering several different repeat proteins with solenoid folds , including ankyrin repeats ( Binz et al . , 2003; Kohl et al . , 2003; Mosavi et al . , 2002 ) , TPRs ( Doyle et al . , 2015; Kajander et al . , 2007; Main et al . , 2003b ) , pentatricopeptide repeats ( PPRs ) ( Coquille et al . , 2014; Shen et al . , 2016 ) and leucine rich repeats ( Rämisch et al . , 2014; Stumpp et al . , 2003 ) . Following these principles , four different sites of mutation ( L4W , K7L/R , V9N , I23D/Y , see Figure 4 ) were considered to improve interface hydrophobicity or preserve coevolved positions observed in TPRs ( Sawyer et al . , 2013 ) ( see Materials and methods ) . Furthermore , as natural TPR proteins tend to exhibit zero net charge ( Magliery and Regan , 2004 ) , four positively charged residues were also targeted ( K2E , K6N , K22E , R25Q/E , see Figure 4 ) . This resulted in a set of eight candidate mutation sites . In order to preserve the character of the RPS20-hhta sequence , we restricted the number of mutations in any repeat unit to be at most five . 10 . 7554/eLife . 16761 . 008Table 1 . The primary structures of the six designed proteins using RPS20-hhta tested in vitro . Point mutations introduced into RPS20-hhta are shown in bold and underlined . The C-terminal four residues in RPS20-hhta were replaced by the consensus loop sequence DPNN in TPRs ( underlined ) . The sequence of the stop helix is italicized . M4NΔC is M4N without stop helix . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 008NameMutationsSequenceM0-NS IKTLSKKAVLLAQEGKAEEAIKIMRKAVSLDPNN IKTLSKKAVLLAQEGKAEEAIKIMRKAVSLDPNN IKTLSKKAVLLAQEGKAEEAIKIMRKAVSLIDKA AKGSTLHKNAAARRKSRLMRKVQKL M2K7L , I23YNS IKTLSKLAVLLAQEGKAEEAIKYMRKAVSLDPNN IKTLSKLAVLLAQEGKAEEAIKYMRKAVSLDPNN IKTLSKLAVLLAQEGKAEEAIKYMRKAVSLIDKA AKGSTLHKNAAARRKSRLMRKVQKL M4EK2E , K7L , V9N , I23YNS IETLSKLANLLAQEGKAEEAIKYMRKAVSLDPNN IETLSKLANLLAQEGKAEEAIKYMRKAVSLDPNN IETLSKLAVLLAQEGKAEEAIKYMRKAVSLIDKA AKGSTLHKNAAARRKSRLMRKVQKL M4NK6N , K7L , V9N , I23YNS IKTLSNLANLLAQEGKAEEAIKYMRKAVSLDPNN IKTLSNLANLLAQEGKAEEAIKYMRKAVSLDPNN IKTLSNLAVLLAQEGKAEEAIKYMRKAVSLIDKA AKGSTLHKNAAARRKSRLMRKVQKL M4RDK2E , K7R , V9N , I23DNS IETLSKRANLLAQEGKAEEAIKDMRKAVSLDPNN IETLSKRANLLAQEGKAEEAIKDMRKAVSLDPNN IETLSKRAVLLAQEGKAEEAIKDMRKAVSLIDKA AKGSTLHKNAAARRKSRLMRKVQKL M5K2E , L4W , K7L , V9N , I23YNS IETLSKLANLLAQEGKAEEAIKYMRKAVSLDPNN IETWSKLANLLAQEGKAEEAIKYMRKAVSLDPNN IETWSKLAVLLAQEGKAEEAIKYMRKAVSLIDKA AKGSTLHKNAAARRKSRLMRKVQKL M4NΔCK6N , K7L , V9N , I23YNS IKTLSNLANLLAQEGKAEEAIKYMRKAVSLDPNN IKTLSNLANLLAQEGKAEEAIKYMRKAVSLDPNN IKTLSNLAVLLAQEGKAEEAIKYMRKAVSLIDKA AK 10 . 7554/eLife . 16761 . 009Figure 4 . Sequence positions considered for optimizing the designed proteins . ( a ) Sequence logo of the TPR motif . A TPR consensus sequence ( Main et al . , 2003b ) ( PDB: 1na0 , chain A ) and its secondary structure determined by DSSP ( Kabsch and Sander , 1983 ) are aligned below the sequence logo . The eight TPR signature positions are underscored in the consensus sequence . The five interface positions are highlighted in yellow . ( b ) Sequence logo of RPS20-hh . The RPS20-hhta sequence and its predicted secondary structure using Quick2D ( Biegert et al . , 2006 ) is aligned below the sequence logo . The derived interface positions are highlighted in yellow . The four residues subjected to mutations are colored in red . The four positively charged residues selected for mutation to lower the surface charge are in blue . ( c ) The locations of the interface positions displayed on a TPR ( left ) and a RPS20 structure ( right ) . In both structures , the interface positions are labeled and highlighted as yellow spheres . The TPR structure is CTPR3 ( PDB: 1na0 , chain A ) , which is shown as a cartoon and is colored using the same scheme as the secondary structure representation in ( a ) . The stop helix is in gray . The RPS20 structure is from T . thermophilus ( PDB: 4gkj , chain T ) , in which the RPS20-hh fragment is colored using the same scheme as the secondary structure representation in ( b ) . The sequence logos were generated using WebLogo ( Crooks et al . , 2004 ) . Sequences from representative proteome 75% ( Chen et al . , 2011 ) downloaded from Pfam families TPR_1 and Ribosomal_S20p were used as input to WebLogo ( 9338 and 972 sequences , respectively ) . The structures were rendered using PyMOL ( Schrödinger , 2010 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 00910 . 7554/eLife . 16761 . 010Figure 4—figure supplement 1 . Mutual information plot ( a and b ) and direct coupling analysis plot ( c and d ) for TPR repeat sequences . The subfigures ( a ) and ( c ) were generated using the seed alignment sequences from Pfam family TPR_1 ( 558 sequences . Sequence Q29585_PIG/28–61 was removed as it contains unknown residue X ) . The largest mutual information value is observed between position 7 and 23 . The subfigures ( b ) and ( d ) were generated using the multiple alignment of representative proteomes rp75 sequences from Pfam family TPR_1 ( 9338 sequences ) . The largest non-local mutual information value was observed between position 24 and 47 , corresponding to position 7 and 23 using TPR repeat numbering . Alignments were taken from Pfam 27 . 0 . Subfigures ( a ) and ( b ) were generated using MatrixPlot . Subfigures ( c ) and ( d ) were generated using DCA Workbench ( http://dca . rice . edu/portal/dca/workbench ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 01010 . 7554/eLife . 16761 . 011Figure 4—figure supplement 2 . Rosetta energy scores ( fixbb+relax ) for TPR designs based on RPS20-hhta sequence and various sets of mutations . The scores for the designs are shown in two groups: the group to the left are combinations involving only primary mutations ( see Supplementary file 1E ) . The group to the right are designs involving both primary and secondary mutations ( Supplementary file 1E ) . The design variants are sorted by the average of the lowest 10% scores . The designs tested in the lab are marked by red arrows ( M2 , M4E , M5 , M4N , M4RD ) . The in silico simulation was performed using Rosetta 3 . 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 01110 . 7554/eLife . 16761 . 012Figure 4—figure supplement 3 . Prediction of intrinsically disordered regions in RPS20 of Thermus aquaticus ( NCBI gi: 489134531 , accession: WP_003044315 . 1 ) using a ) IUPred ( http://iupred . enzim . hu/ ) ; b ) DisEMBL ( http://dis . embl . de/ ) and c ) PONDR ( http://www . pondr . com/ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 012 In most TPR proteins , there is an α-helix at the C-terminus , which interacts with the last TPR unit by covering the hydrophobic surface . This so-called C-terminal 'stop helix' had been observed in all known TPR structures and was considered essential for the solubility of natural TPR proteins ( D'Andrea and Regan , 2003; Das et al . , 1998; Main et al . , 2003b ) . Most other designed TPRs employ purpose-designed stop helix sequences . Here , we chose to use the RPS20 C-terminal helix to become a natural stop helix , since it is already known to interact favorably with RPS20-hhta ( Figure 3 ) . Further , we inserted two residues ( Asn-Ser ) before the ﬁrst TPR unit as an N-terminal cap to the first helix ( Aurora and Rose , 1998; Kumar and Bansal , 1998 ) , in analogy to a previously designed idealized TPR protein , CTPR3 ( Main et al . , 2003b ) . To model the structure of the designed proteins in silico , we fused two structures to create a hybrid template: We used CTPR3 ( PDB id: 1na0 chain A ) as the structural template for the three RPS20-hhta fragments , and the best-resolved RPS20 structure ( PDB id: 2vqe chain T; 2 . 5 Å ) for helix B3 and the stop helix . We built structural models on this hybrid template and tested a variety of mutants using the Rosetta programs ﬁxbb and relax , which perform fixed-backbone design and structural refinement ( Das and Baker , 2008; Doyle et al . , 2015; Park et al . , 2015; Parmeggiani et al . , 2015 ) . The Rosetta energy score of the models calculated for all mutants is depicted in a boxplot ( Figure 4—figure supplement 2 ) . Among them , five were selected for further testing in vitro ( see Materials and methods ) . These five tested mutants are termed M2 , M4E , M4N , M4RD and M5 . Their primary structures are listed in Table 1 . We cloned the five TPR designs plus the unmutated construct M0 into pET vectors for expression in Escherichia coli . Three proteins ( M0 , M4RD and M5 ) could be purified from soluble extracts; the other constructs were insoluble and were refolded from inclusion bodies . In far UV circular dichroism ( CD ) spectra , all proteins displayed a strong alpha-helical pattern , except M0 and M4RD , which appeared to be unfolded , but not prone to aggregation and precipitation , even at high concentrations . When we studied the melting curves , M4N showed cooperative unfolding with a Tm of 77°C ( Supplementary file 1F ) , while the unfolding of M2 , M4E and M5 did not conform to a classical two-state transition , consistent with an unstable molten globule-like state . On the other hand , non-cooperative unfolding processes have been demonstrated for perfectly stable TPR repeats and suggested to be common for various types of repeat proteins ( Cortajarena and Regan , 2006; Kajander et al . , 2007; Stumpp et al . , 2003 ) . To clarify this point , urea-induced unfolding transitions were monitored by CD . Like M4N , the three variants M2 , M4E and M5 yielded typical cooperative denaturation curves , indicative of folded polypeptides ( Figure 5—figure supplement 2 ) . The ∆GU-FH2O values agree well with those reported for other designed TPRs ( Supplementary file 1F ) ( Main et al . , 2005 ) . In line with these findings , M5 , the only protein containing tryptophan residues , had a λmax of 336 nm in fluorescence emission spectra , as expected for partially shielded aromatic residues . We conclude that four of the five designed TPR variants , M2 , M4E , M4N and M5 , result in well-folded repeat proteins . To determine the oligomeric state of our folded proteins , we performed static light scattering experiments . Surprisingly , all four constructs were exclusively dimers ( Supplementary file 1F ) . We also examined the ribosomal parent protein RPS20 . Within the ribosome , RPS20 is partially embedded in the 16S rRNA , making many nucleic acid contacts . Like many other ribosomal proteins , it is not expected to adopt a stable structure in isolation . Indeed , it has a biased amino acid composition and is predicted to be largely unstructured by many prediction programs ( Figure 4—figure supplement 1 , see also Supplementary file 1J ) . It had been shown previously that isolated RPS20 exhibits only one third helical content by CD ( Paterakis et al . , 1983 ) . For Thermus RPS20 specifically , simulations predict a flexible conformation in solution ( Burton et al . , 2012 ) . We cloned RPS20 from T . aquaticus and its close relative T . thermophilus . Upon expression , both proteins were insoluble and had to be refolded . In static light scattering measurements , both proteins behaved as monomers ( Supplementary file 1F ) . Based on CD spectra , which showed a high proportion of random structure , and the absence of defined melting and urea-denaturation curves ( Supplementary file 1F ) , we conclude that RPS20 indeed exhibits considerable conformational variation in solution . To obtain high-resolution structural information on our designed proteins , we set up crystallization trials for all four folded constructs . We obtained crystals and solved the structure of M4N to a resolution of 2 . 2 Å ( Figure 5a ) . The asymmetric unit ( ASU ) contains three polypeptide chains of almost identical structure ( all pairwise Cα RMSD values below 1 . 4 Å ) . Notably , all three chains exhibit the desired TPR architecture with three repetitive hairpins , which interact via knobs-into-holes packing between helices Ai and B ( i-1 ) , as is characteristic of TPR hairpins . A superposition to the CTPR3 structure yields Cα RMSD values below 2 . 6 Å ( supplementary file 1I ) . An unexpected difference to the canonical TPR structure is that the stop helix of M4N is not resolved in any of the three chains . However , this missing helix is compensated for by a specific dimerization mode of two M4N protomers . Therein , the C-terminal TPR units of the two protomers form a tight interface , in which the B3 helix of each chain substitutes for the stop helix of the other , mimicking the capping effect of the stop helix ( Figure 6 ) . A superposition of this mimicry to the last TPR unit and stop helix of CTPR3 yields Cα RMSD values as low as 1 . 2 Å over 44 residues . The third chain of the ASU , however , was found as a monomer , capping its C-terminal TPR unit in a more unspecific manner by packing it orthogonally against the two A1 helices of the dimer ( Figure 5a ) . 10 . 7554/eLife . 16761 . 013Figure 5 . The X-ray structure of M4N . ( a ) The three chains A , B and C in the asymmetric unit are colored green , blue and yellow , respectively . Chains A and B form a dimer . ( b ) Superposition of the three chains . Only Cα traces are shown for clarity . ( c ) Superposition of M4N ( chain A , green ) and the designed consensus TPR CTPR3 ( PDB: 1na0 , chain A , gray ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 01310 . 7554/eLife . 16761 . 014Figure 5—figure supplement 1 . The interaction of M4N molecules in the crystal . ( a ) Five adjacent ASUs are depicted . Chain A ( green ) and B ( blue ) form a dimer , while chain C ( yellow ) packs its C-terminus to the N-termini of chains A and B . ( b ) Top view . ( c ) An additional ASU ( top-left ) is shown to demonstrate the packing of N-termini of chains C . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 01410 . 7554/eLife . 16761 . 015Figure 5—figure supplement 2 . Urea denaturation of designed TPR repeats . Urea-induced equilibrium unfolding at 23°C was monitored by circular dichroism at 222 nm . Data were converted to the fraction of unfolded protein fU and fitted to a two-state model . The protein concentration was 15 µM . See Supplement file 1F for obtained parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 01510 . 7554/eLife . 16761 . 016Figure 5—figure supplement 3 . Mass spectrometry ( MS ) analysis of M4N . The M4N fragment with a mass of 12733 . 533 Da in MS is underlined and highlighted in blue ( theoretical mass 12733 . 77 Da ) . The C-terminus of M4N as observed in the crystal structure is marked by a red arrow . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 01610 . 7554/eLife . 16761 . 017Figure 6 . Mimicry of the stop helix in the M4N dimer . The C-terminal TPR unit in chain A ( green ) and the C-terminal helix B3 in chain B ( blue ) are superposed to the last TPR unit plus the stop helix in CTPR3 ( gray ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 017 Analysis by mass spectrometry revealed that the M4N stop helix had been partially proteolyzed upon expression of the protein ( Figure 5—figure supplement 3 ) . Although we did not observe proteolysis in the other folded constructs ( M2 , M4E and M5 ) , which were also all dimeric , we analyzed whether proteolysis might have favored the dimerization of M4N . Extending the stop helix with a C-terminal His6-tag prevented proteolysis , but did not affect stability or dimerization ( M4N-His; Supplementary file 1F ) . We conclude that in the amplified constructs , the observed interactions are more favorable than the interaction with the native stop helix , releasing it and rendering it prone to degradation . This led us to ask whether this helix is in fact dispensable . Indeed , an M4NΔC construct , which terminates with the B3 helix , showed the same stability and dimerization as M4N . We obtained two structures for M4NΔC from different crystal forms at 2 . 0 Å and 1 . 7 Å resolution , respectively , the first ( CF I ) with two dimers in the ASU and the second ( CF II ) with a single chain in the ASU , for which we constructed the dimer by crystallographic symmetry . All three dimers superimpose to the M4N dimer with Cα RMSD below 1 . 9 Å ( Figure 7 , Supplementary file 1I ) . We conclude that the stop helix is dispensable for folding , dimerization and the stability of our designed constructs . 10 . 7554/eLife . 16761 . 018Figure 7 . M4NΔC structures of two different crystal forms and their comparison to the M4N dimer . ( a ) Two dimers in the ASU of M4NΔC CF I . ( b ) Dimer constructed by applying the crystallographic symmetry to the single chain in the ASU of M4NΔC CF II . ( c ) Superposition of all the four M4N and M4NΔC dimers . The M4N dimer is in green and blue . The three M4NΔC dimers are in different shades of gray as in ( a ) and ( b ) . ( d ) Superposition of all the chains in the M4N and M4NΔC dimers ( eight chains in total ) . Only Cα traces of proteins are shown for clarity . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 018 The geometry of dimerization in M4N has not been observed in TPR structures before . Although there have been reports on the self-association of TPR-containing proteins involved in various regulatory biological processes ( Bansal et al . , 2009a , 2009b; Ramarao et al . , 2001; Serasinghe and Yoon , 2008 ) , only a small number of oligomeric TPR structures have been determined to date ( Krachler et al . , 2010; Lunelli et al . , 2009; Zeytuni et al . , 2012 , 2015; Zhang et al . , 2010 ) . None of these resemble the ring-shaped dimer of M4N . The results shown above suggest that the mutations we made to RPS20-hhta were crucial for obtaining the TPR fold . If RPS20 and TPR proteins indeed share a common ancestor , such mutations may have been sampled in the course of evolution . Since we cannot reconstruct the ancestor and do not know what its function was beyond a general expectation of RNA binding , we decided to test whether the mutations we introduced impaired the interaction between RPS20 and its cognate RNA , as an indication of their compatibility with RNA interaction . Each mutation in M2 and M4N occurs in natural RPS20 sequences ( see Supplementary file 1A ) , but no RPS20 sequence has all four mutations simultaneously and we therefore tested if they can be tolerated in vivo . As genetic engineering in T . aquaticus turned out to be unfeasible , we performed these tests in T . thermophilus HB8 , which is a well-established model organism . The RPS20 helical hairpins in T . aquaticus and T . thermophilus differ only at four positions , of which two are highly conservative substitutions ( Figure 8a ) . 10 . 7554/eLife . 16761 . 019Figure 8 . RPS20 variants M2 and M4N are functional proteins . ( a ) The 34 amino-acid long RPS20-hh fragments in T . aquaticus and T . thermophilus differ only at four positions , including two conservative mutations ( V9I and I21L ) . ( b ) Scheme of the rpsT region before ( upper ) and after ( lower ) substitution of rpsT with the kanamycin resistance cassette ( kat ) . Base pair ( bp ) values indicate the PCR products that can be amplified . Regions depicted with the same pattern are identical . Regions in solid black and gray also contain genes which are not marked for clarity . ( c ) PCR to detect substitution of rps20 by the kat gene and ( d ) PCR to detect the presence of chromosomal rpsT in T . thermophilus strains ( WT: T . thermophilus HB8; KM4:T . thermophilus KM4 ) carrying various plasmids ( TT: pJJSpro-rpsTTt; E: pJJSpro; TA: pJJSpro-rpsTTa; TA2: pJJSpro-rpsTTaM2; TA4: pJJSpro-rpsTTaM4N; -: No plasmid ) after sequential grow under different selective pressures ( 1: 30 µg/ml kanamycin; 2: 120 µg/ml kanamycin; 3: 0 µg/ml kanamycin ) . ( e ) Corresponding growth curves of the host bacteria with various substitutions and plasmids . DOI: http://dx . doi . org/10 . 7554/eLife . 16761 . 019 We first attempted to substitute the chromosomal RPS20-encoding gene , rpsT , with a kanamycin resistance cassette , to obtain T . thermophilus strain KM4 ( Figure 8b ) . For complementation we introduced plasmids bearing wild type rpsT from T . thermophilus ( TT ) or T . aquaticus ( TA ) , rpsT from T . aquaticus carrying the mutations from M2 ( TA2 ) or M4N ( TA4 ) , or merely empty plasmids as negative control ( E ) . We monitored the substitution of rpsT by a PCR screening protocol , which will amplify a 1500 bp region if WT rpsT is substituted and an 800 bp region otherwise ( Figure 8b ) . Under selection pressure from kanamycin , only the 1500 bp product was obtained in all cases where plasmid-borne rpsT was introduced , whether in wild-type or mutated form ( Figure 8c panels 1 and 2 , lanes TT , TA , TA2 and TA4 ) , showing that the chromosomal gene had been fully substituted . In contrast , PCR screening of strain KM4 complemented with an empty plasmid produced both 800 bp and 1500 bp fragments ( Figure 8c panels 1 and 2 , lane E ) . Since T . thermophilus HB8 is a polyploid organism ( minimally tetraploid [Ohtani et al . , 2010] ) , this result shows that rpsT can be reduced in copy number , but not fully eliminated , suggesting that the gene is essential . To assess the level of substitution achieved with the various plasmids , we designed a second PCR screening protocol to specifically detect chromosomal rpsT via a 300 bp product . At low kanamycin concentrations this protocol always generated a product ( Figure 8d panel 1 ) , but at increased kanamycin concentration we did not obtain product for any rpsT allele ( Figure 8d panel 2 , lanes TT , TA , TA2 and TA4 ) . This demonstrates that plasmid-borne rpsT and its mutants were able to complement the chromosomal rpsT and that the latter was displaced from the population to a level that left it undetectable by PCR . In contrast , we could never completely suppress chromosomal rpsT in strain KM4 complemented with an empty plasmid , even under high kanamycin conditions ( 120 µg/ml ) . In E . coli and Salmonella enterica , rpsT has been reported to be non-essential , but its deletion significantly lowers growth rates ( Bubunenko et al . , 2007; Tobin et al . , 2010 ) . We found that rpsT is essential in T . thermophilus , but that its loss could be complemented by wild-type and mutant T . aquaticus rpsT , and that this complementation restored wild-type levels of growth ( Figure 8e ) . Moreover , when the selection pressure from kanamycin was removed , no reversal in the PCR products was detected for any strain ( Figure 8c and d , panel 3 ) , which conﬁrms that chromosomal rpsT was substantially displaced during kanamycin treatment . We conclude that rpsT from T . aquaticus and its two mutated alleles are neutral with respect to survival and growth for T . thermophilus . This demonstrates that the mutations we introduced do not affect negatively the interaction between RPS20 and its cognate RNA , and that therefore such mutations could have been sampled multiply and in a cumulative fashion by neutral drift during the course of evolution . Proteins are the most complex macromolecules synthesized in nature and by and large need to assume defined structures for their activity . This folding process is complicated and easily disrupted , witness the elaborate systems for protein folding , quality control and degradation universal to all living beings . Despite the widespread problems to reach and maintain the folded state , natural proteins nevertheless form a best-case group , since the overwhelming majority of random polypeptides do not appear to have a folded structure ( Keefe and Szostak , 2001; Wei et al . , 2003 ) . It thus seems impossible that , at the origin of life , the prototypes for the folded proteins we see today could have arisen by random concatenation of amino acids . We have proposed that folding resulted from the increasing complexity of peptides that supported RNA replication and catalysis , and that these peptides assumed their structure through the interaction with the RNA scaffold ( Lupas et al . , 2001; Söding and Lupas , 2003 ) . In this view , protein folding was an emergent property of RNA-peptide coevolution . We have recently described 40 such peptides whose conservation in diverse folds suggests that they predated folded proteins ( Alva et al . , 2015 ) . These peptides are enriched for nucleic-acid binders , particularly in the context of the ribosome . Due to its extremely slow rate of change , the ribosome essentially represents a living fossil , providing the possibility to study the chronology of ancient events in molecular evolution ( Hsiao et al . , 2009 ) . Thus , core ribosomal proteins offer a window into the time when proteins were acquiring the ability to fold . Those close to the catalytic center almost entirely lack secondary structure . Further away from the center , their secondary structure content gradually increases and at the periphery , these secondary structure elements become arranged into topologies that parallel those seen in cytosolic proteins ( Hsiao et al . , 2009 ) . Collectively , the structures of ribosomal proteins chart a path of progressive emancipation from the RNA scaffold . Even the peripheral proteins , however , still mostly assume their structure only in the context of the ribosomal RNA , as exemplified by RPS20 in our study ( Supplementary file 1F , see also Paterakis et al . , 1983 ) . The simplest mechanism to achieve an increase in complexity is the repetition of building blocks and nature provides many examples for this , at all levels of organization . The dominant role of repetition in the genesis of protein folds has been documented in many publications since the 1960s ( Alva et al . , 2007; Blundell et al . , 1979; Broom et al . , 2012; Eck and Dayhoff , 1966; Kopec and Lupas , 2013; Lee and Blaber , 2011; McLachlan , 1972 , 1987; Remmert et al . , 2010; Söding et al . , 2006 ) . As a test of this mechanism , we explored whether a peptide originating from a ribosomal protein that is disordered outside the context of the ribosome , could form a folded protein through an increase in complexity afforded by repetition . For this , we chose a present-day representative of one of the 40 fragments we reconstructed ( Alva et al . , 2015 ) ; this fragment is naturally found in a single copy in several different folds , including that of ribosomal protein RPS20 , and repetitively in one fold , TPR . Simple repetition was not sufficient in our case , but the repeat protein was so close to a folded structure that only two point mutations per repeat were necessary to allow it to fold reliably . The mutations needed for this transition did not appear to affect negatively the interaction with the RNA scaffold , raising the possibility that they could have been among the variants sampled multiply in the course of evolution . Our experiments recapitulate a scenario for the emergence of a protein fold by a widespread and well-documented mechanism , and show that this could have proceeded in a straightforward way . These experiments represent a proof of concept , starting with a modern peptide likely to still retain many features of an ancestral αα-hairpin that gave rise to a number of folds , including TPR . Rather than proposing proto-RPS20 as the parent of TPR domains , we see it as one of many proteins emerging from this ancestral hairpin . Given the ease with which repetition of the RPS20 hairpin yielded a TPR-like fold , we consider it likely that the hairpins belonging to the ancestral group were amplified many times during the emergence of folded proteins to yield a range of TPR-like offspring , of which only one may have survived to this day ( but see also the figure legend to Figure 1 ) . The reason for this limited survival may lie in the fact that a structure is a prerequisite for protein function , but it is the function that is under biological selection . It could be that the newly emerged TPR-like folds required many additional changes to achieve a useful activity and that therefore only very few – possibly just one – survived . We consider a different scenario more probable , however . All present-day TPR domains whose function has been characterized mediate protein-protein interactions by binding to linear sequence motifs in unstructured polypeptide segments ( D'Andrea and Regan , 2003; Zeytuni and Zarivach , 2012 ) . This activity would have been particularly relevant at a time of transition from peptides dependent on RNA scaffolds for their structure , to autonomously folded polypeptides . Functions relevant in this context would have been to prevent aggregation and increase the solubility of newly emerging ( poly ) peptides , to promote autonomous folding , to serve as assembling factors for RNA-protein and protein-protein complexes , and to recognize targeting sequences in the emerging cellular networks . It therefore seems likely to us that many of the newly evolved TPR-like folds became established in one or the other of these activities , only to be subsequently displaced by folding becoming a general property of cellular polypeptides and by more advanced , energy-dependent folding factors , which offered much better regulation . Exploring the extent to which our new TPR protein could fulfill such functions represents the next frontier in our studies . All sequence similarity searches in this work were performed using the Web BLAST ( RRID:SCR_004870 ) from the National Institute for Biotechnology Information ( NCBI; http://blast . ncbi . nlm . nih . gov; Boratyn et al . , 2013 ) and in the MPI Bioinformatics Toolkit ( RRID:SCR_010277 , https://toolkit . tuebingen . mpg . de/; Alva et al . , 2016 ) . Examples of recently amplified repeat units in TPR were taken from a previous investigation ( Dunin-Horkawicz et al . , 2014 ) . The TPR domain in serine/threonine-protein phosphatase 5 was chosen as a representative three-repeat TPR , the most common TPR form in natural proteins ( D'Andrea and Regan , 2003; Sawyer et al . , 2013 ) , to study divergent evolution of TPR . We ran BLAST on the non-redundant protein sequence database ( nr ) with an E-value threshold of 0 . 05 using the TPR domain of serine/threonine-protein phosphatase 5 from Homo sapiens as query ( Das et al . , 1998 ) . From the results , we chose seven taxa to cover a diverse range of life . TPRpred program ( Karpenahalli et al . , 2007 ) was used to help identify tandem repeats of TPR units . The construction of multiple sequence alignments ( MSAs ) for TPR units was straightforward as all TPR units are of the same size ( 34 aa ) and no indels were allowed in the MSAs . We used Clustal X 2 . 1 ( Larkin et al . , 2007 ) to build phylogenetic trees using the neighbor-joining clustering algorithm and 1000 bootstrap trials ( Bootstrap N-J Tree ) . SplitsTree4 ( Huson and Bryant , 2006 ) was used to render the phylogenetic trees . To find proteins homologous to the TPR unit , we first employed the TPRpred program ( Karpenahalli et al . , 2007 ) to identify proteins that share significant sequence similarity to the TPR sequence profile , then filtered them by comparing to the TPR structures . First , TPRpred program with TPR proﬁle tpr2 . 8 was used to identify TPR unit like sequences from all protein sequences of known structures in the Protein Data Bank ( PDB , RRID:SCR_012820 ) ( Berman et al . , 2000 ) . Protein sequences from the SEQRES record in PDB files were downloaded from the PDB . We only considered sequences with at least 34 residues , which is the length of the TPR unit . Redundancy was removed by keeping only non-identical sequences . In total , 68 , 197 sequences were scanned by using TPRpred with default parameters . Only fragments predicted to be TPR with a p-value lower than 1 . 0e−4 were retained ( 646 hits ) . We estimated the false discovery rate ( FDR ) ( Noble , 2009 ) associated with this p-value cutoff using a simulated sequence dataset generated by using the amino-acid composition derived from the PDB sequences . The dataset contains the same number of sequences of the same length distribution as the PDB sequences . The FDR was estimated to be the ratio of the number of hits in the simulated dataset to the number of detected hits in the PDB sequences ( Noble , 2009 ) . We repeated the simulation 100 times and the FDR was estimated to be 1 . 0 ± 0 . 4% . Within the 646 hits , we kept only TPR unit singletons , which are TPR units that have no other TPR units close to them within a distance of 10 residues in the same sequence . TPR units of identical sequences are considered only once . Subsequently , these TPR unit singletons were filtered by removing those annotated to belong to clan TPR ( CL0020 ) in Pfam 27 . 0 ( RRID:SCR_004726 ) . The structures of the predicted TPR units obtained from the previous step were then compared to an average TPR unit structure . A predicted TPR unit was discarded if the Cα RMSD of the 34 residues is greater than 2 . 0 Å after superposition . The average TPR unit structure was generated by considering all proteins belonging to family tetratricopeptide repeat ( TPR ) ( a . 118 . 8 . 1 ) in SCOP 1 . 75 ( RRID:SCR_007039 ) ( Murzin et al . , 1995 ) . TPR repeats in these proteins were again detected using TPRpred and a per-repeat p-value cutoff of 1 . 0e−4 was used . In total , 50 non-redundant TPR repeat fragments were identiﬁed and superposed using a multiple structure alignment tool MultiProt ( Shatsky et al . , 2004 ) . The average Cα positions were calculated from the 50 structures after superposition . We obtained 31 fragments after the structure filtering step ( Supplementary file 1C ) . We then inspected the protein structures using PyMOL ( RRID:SCR_000305 ) ( Schrödinger , 2010 ) . Among them , 22 were observed to be involved in the formation of solenoid or tandem repeat structures and were thus not further considered . We applied TPRpred to scan all RPS20 sequences belonging to Pfam 27 . 0 family Ribosomal S20p ( PF01649 ) , including sequences from both datasets 'full' and 'ncbi' . There are 4402 and 2284 sequences in the two sets . We merged the two sets and removed identical sequences to create a dataset of 3742 RPS20 sequences . TPRpred was used to detect TPR unit homologs in them . We obtained 24 hits in these RPS20 sequences predicted by TPRpred to match TPR unit proﬁle with a p-value smaller than 1 . 0e−4 ( see Supplementary file 1D ) . We defined 'interface positions' in the TPR unit and then transferred the definition to RPS20-hh according to their structure superposition . We considered the residues on the outer side of the two helices facing neighboring TPR units . Both helix A and helix B in the TPR unit are α-helices , which have on average 3 . 6 residues per turn . Thus , every third or fourth residue always appears on the same side of the helix . They are positions 3 , 7 and 10 in helix A and positions 17 , 21 , 24 and 28 in helix B . According to the TPR sequence proﬁle compiled by Main et al . ( Main et al . , 2003b ) , the most common residues at these positions are hydrophobic except for positions 17 and 24 , where the most common residues are both Tyr ( see also Figure 4a ) . Therefore , positions 17 and 24 were not included in the definition of interface positions . Furthermore , the residue at position equivalent to position 24 in RPS20 structure faces its C-terminal helix and is already an interface residue ( Figure 4c ) . Thus , it was not considered as an interface position to be checked in the study . In the end , only positions 3 , 7 , 10 , 21 and 28 in RPS20-hh were deﬁned to be interface positions to be examined , because they are exposed to the solvent or interact with the RNA molecules in the ribosome , but would interact with neighboring repeats in the TPR fold . We searched all RPS20 sequences in Pfam 27 . 0 family Ribosomal_S20p ( PF01649 ) , including both datasets 'full' and 'ncbi' , for candidates in which the interface positions are occupied by as many hydrophobic residues as possible . In the MSA provided by Pfam , we extracted the 34 columns that correspond to the sequence fragment of RPS20-hh from Thermus aquaticus , which was found by TPRpred to be the hit with the best p-value and was thus used as the reference RPS20-hh . We obtained 1370 sequence fragments that do not contain any indels , in which the interface positions were examined for hydrophobicity . Here , Ala , Ile , Leu , Met , Phe , Val were considered as hydrophobic residues . Trp was not included as its side chain may be too large to be accommodated at the interface . We employed several low-complexity / intrinsically disordered region prediction methods ( SEG [Wootton , 1994] , PONDR [Romero et al . , 2001] , DisEMBL [Linding et al . , 2003] , IUPred [Dosztányi et al . , 2005a , 2005b] ) to investigate putative intrinsically disordered regions in the RPS20 of Thermus aquaticus . We ran SEG with three sets of recommended parameters ( Wootton and Federhen , 1996 ) and the other approaches with default parameters . We considered eight positions ( 2 , 4 , 6 , 7 , 9 , 22 , 23 and 25 ) in RPS20-hhta for optimization apart from the four residues at the C-terminus . Main et al . ( Main et al . , 2003b ) discovered a set of eight 'TPR signature residues' in the consensus design: W4 , L7 , G8 , Y11 , A20 , Y24 , A27 and P32 . Six of them are missing in RPS20-hhta except A20 and A27 . Following the principle of consensus design , we introduced L4W and K7L into RPS20-hhta . K7 is also one of the interface positions that ought to be mutated to hydrophobic residue for better packing at interfaces . A8 and L11 were not optimized because they are the second and third most common residues at positions 8 and 11 in the TPR profile , respectively . M24 was also retained because it seems long hydrophobic side chains are favored at position 24 though Met is not one of the three most common residues ( YFL ) . P32 was introduced to replace D32 in RPS20-hhta as part of the C-terminal consensus loop ( DPNN ) between repeats . Co-evolution is commonly observed between physically interacting residues ( de Juan et al . , 2013 ) . We investigated if any positions we optimized are involved in a co-evolution relationship so that we can preserve such correlations . We performed a direct coupling analysis ( Morcos et al . , 2011 ) and computed the mutual information using MatrixPlot ( Gorodkin et al . , 1999 ) between all positions in TPR repeat sequences . The results of both approaches revealed that the highest correlation occurs between positions 7 and 23 ( Figure 4—figure supplement 1 ) , with the most commonly observed combinations being R7-D23 and L7-Y23 . Therefore , we always mutated I23 to the most commonly observed residue tyrosine ( I23Y ) in the TPR consensus sequence together with aforementioned mutation K7L . In addition , we considered combination K7R and I23D together . Combination K7-I23D was also tested because of highly similar physicochemical properties between Lys and Arg side chains . The hydrophobic side chain of valine at position 9 in RPS20-hhta is buried between helices in RPS20 , but would be exposed on the surface of the designed protein except in the last repeat , in which V9 interacts with the stop helix . Therefore , it is considered to be mutated to the most common residue asparagine ( V9N ) in the TPR repeat consensus except in the last repeat ( Figure 4c ) . RPS20-hhta sequence and surface is enriched with positively charged residues ( Figure 4b ) . This would lead to the exceedingly high theoretical iso-electric point ( pI ) of the designed proteins . Natural TPR proteins tend to exhibit zero net charge ( Magliery and Regan , 2004 ) . Hence , we decided to randomly mutate the positively charged residues ( Lys and Arg ) in the two helices of RPS20-hhta to the corresponding most common residues in TPR sequence profile ( K2E , K6N , K22E , R25Q/E ) . K26 was not mutated as Lys is already the most common residue in the TPR profile . At the C-terminus of the designed TPR , the last four residue of RPS20-hhta ( IDKA ) were replaced with the TPR consensus loop sequence ( DPNN ) between repeat units . The reason is as follows . The secondary structure of the TPR unit is helix ( 13 aa ) – loop ( 3aa ) – helix ( 14 aa ) – loop ( 4aa ) , while the secondary structure of the RPS20-hhta identiﬁed to be homologous to TPR unit is helix ( 13 aa ) – loop ( 3 aa ) – helix ( 18 aa ) ( Figures 2 and 4 ) . The last four residues may have been included in the prediction by TPRpred merely to fulﬁll the size requirement of TPR repeat ( 34 aa ) . Indeed , when we scanned RPS20-hhta sequence using the hidden Markov model constructed for Pfam family TPR_1 , only positions 2–28 were found to be similar to the TPR_1 proﬁle using HMMER 3 . 0 ( RRID:SCR_005305 ) ( Eddy , 2009 ) , even if all filters were switched oﬀ . So the four very C-terminal residues in RPS20-hhta were not used in the designed TPR between repeat units . They were not replaced in the last repeat unit ( Figure 3 ) . CTPR3 structure of an idealized TPR repeat ( Main et al . , 2003b ) ( PDB id: 1na0 , chain A ) was taken as the main template to build an initial TPR structure model using RPS20-hhta . Helix B3 and the stop helix in our designed protein are different from natural TPRs , but more similar to natural RPS20s . So we also used a RPS20 protein as the structure template for the last repeat and the stop helix . The structure of RPS20 from Thermus thermophilus HB8 ( PDB id: 2vqe , chain T ) was used because it was the structure with the best resolution ( 2 . 5 Å ) . The C-terminal loop in 2vqeT was discarded . The two structures 1na0A and 2vqeT were merged into a hybrid template based on the superposition of their homologous helical hairpins: the third TPR unit in 1na0A and the RPS20-hh in 2vqeT ( the very C-terminal four residues were not used ) . We then modeled the designed TPR sequences using RPS20-hhta onto the hybrid structure template using Rosetta programs fixbb and relax ( Das and Baker , 2008 ) . The Rosetta ﬁxed backbone design application ﬁxbb was used to make the initial model . Subsequently , these models were relaxed using the Rosetta structure refinement application relax . The two steps were iterated three times . See the Supplementary file 1E for the command lines . Rosetta 3 . 4 was used in the work . We selected five constructs for further testing in vitro ( Table 1 ) . They are among the best-scoring constructs according to the in silico simulation ( Figure 4—figure supplement 2 ) . If two constructs have comparable scores ( they are adjacent in the score ranking ) , the one with fewer mutations was preferred . The selected constructs all differ at least at two positions in their sequences . When we searched these optimized RPS20-hhta fragments in the NCBI nr database using BLAST ( Camacho et al . , 2009 ) , the top hits were still RPS20s . DNA sequences coding for the designed TPR repeats were gene-synthesized in codon-optimized form ( Eurofins ) and cloned into vector pET-28b ( Novagen ) using NcoI/HindIII restriction sites , and into pETHis_1a to generate proteins with an N-terminal cleavable His6-tag . RPS20 T . aquaticus and T . thermophilus genes were amplified from genomic DNA and cloned likewise . Recombinant plasmids were transformed into E . coli strain BL21-Gold ( DE3 ) grown on LB agar plates containing 50 µg/ml kanamycin . For expression , cells were cultured at 25°C and induced with 1 mM isopropyl-D-thiogalactopyranoside ( IPTG ) at an OD600 of 0 . 6 for continued growth overnight . Bacterial cell pellets were resuspended in buffer A ( 50 mM Tris pH 8 , 150 mM NaCl ) , supplemented with 5 mM MgCl2 , DNaseI ( Applichem ) and protease inhibitor cocktail ( cOmplete , Roche ) . After breaking the cells in a French Press , the suspension was centrifuged twice at 37 , 000 g . Soluble His6-tagged proteins were purified by binding proteins to Ni-NTA columns ( GE Healthcare ) in buffer A ( 50 mM Tris pH 8 . 0 , 300 mM NaCl ) and elution with increasing concentrations of imidazole up to 0 . 6 M . Eluted proteins were dialyzed against buffer A for cleavage by His6-TEV-protease ( 50 U/mg protein ) . Cleavage leaves two additional residues ( Gly-Ala ) as N-terminal extension to all proteins produced in this manner . After incubation overnight , cleaved proteins were re-run on Ni-NTA columns and collected in the flow-through . They were finally purified by gel size exclusion chromatography ( Superdex G75 , GE Healthcare ) in buffer A containing 0 . 5 mM EDTA . Insoluble proteins were dissolved in 6 M guanidinium chloride and refolded by dialysis overnight against buffer A . Refolded proteins were further purified by sequential anion-exchange ( Q Sepharose HP ) and cation-exchange ( SP Sepharose HP ) chromatography using 0–500 mM NaCl salt gradients in buffer D ( 20 mM Tris pH 8 , 1 mM EDTA ) , and by gel size exclusion chromatography ( Superdex G75 ) in buffer A . To determine the native molecular mass of designed TPR repeats , static light scattering experiment was performed by applying samples onto a superdex S200 gel size exclusion column to which a miniDAWN Tristar Laser photometer ( Wyatt ) and an RI 2031 differential refractometer ( JASCO ) were coupled . Runs were performed in buffer A . Data analysis and molecular mass calculations were carried out with ASTRA V software ( Wyatt ) . Tryptophan fluorescence spectra were recorded on a Jasco FP-6500 spectrofluorometer at 23°C; excitation was at 280 nm , emission spectra were collected from 300–400 nm . Circular dichroism ( CD ) spectra from 200–250 nm were recorded with a Jasco J-810 spectropolarimeter at 23°C in buffer E ( 30 mM MOPS pH 7 . 2 , 150 mM NaCl ) . Cuvettes of 1 mm path length were used in all measurements . For melting curves and determination of Tm , CD measurements were recorded at 222 nm from 20–95°C , the temperature change was set to 1°C per minute , using a Peltier-controlled sample holder unit . For equilibrium-unfolding experiments performed at 23°C , native protein was mixed with different concentrations of urea in buffer A . After equilibration , circular dichroism was monitored at 222 nm . The fraction of unfolded protein fU was determined based on fu = ( yF – y ) / ( yF – yU ) , where yF and yU are the values of y typical of the folded and unfolded states . Data were fitted to a two-state model with the software ProFit ( 6 . 1 ) as described ( Grimsley et al . , 2013 ) , assuming a linear urea [D] dependence of ∆G: ∆GU-FD = ∆GU-FH2O – m[D] , where ∆GU-FD is the free energy change at a given denaturant concentration , ∆GU-FH2O the free energy change in the absence of denaturant , and m the sensitivity of the transition to denaturant . Fragment sizes of M4N were determined by ESI-micrOTOF mass spectrometry ( Bruker Daltonics , Max Planck Institute core facility Martinsried ) , followed by bioinformatic analysis using the Find-Pept tool ( ExPASy ) . For crystallization , the M4N and M4NΔC protein solutions were concentrated to 70 and 30 mg/ml , respectively , in buffer A . The buffer for M4NΔC additionally contained 0 . 5 mM EDTA . Crystallization trials were performed at 295 K in 96-well sitting-drop vapor-diffusion plates with 50 µl of reservoir solution and drops consisting of 300 nl protein solution and 300 nl reservoir solution in the case of M4N , and 400 nl protein solution and 200 nl reservoir solution in the case of M4NΔC . Crystallization conditions for the crystals used in the diffraction experiments are listed in Supplementary file 1H together with the solutions used for cryo-protection . Single crystals were transferred into a droplet of cryo-protectant before loop-mounting and flash-cooling in liquid nitrogen . For experimental phasing , crystals of M4N were soaked overnight in a droplet containing reservoir solution supplemented with 5 mM K2PtCl4 prior to cryo-protection and flash-cooling . All data were collected at beamline X10SA ( PXII ) at the Swiss Light Source ( Paul Scherrer Institute , Villigen , Switzerland ) at 100 K using a PILATUS 6M detector ( DECTRIS ) at the wavelengths indicated in Supplementary file 1H . Diffraction images were processed and scaled using the XDS program suite ( Kabsch , 1993 ) . Using SHELXD ( Sheldrick , 2008 ) , three strong Pt-sites were identified in the M4N derivative dataset . After density modification with SHELXE , the resulting electron density map could be traced by Buccaneer ( Cowtan , 2006 ) to large extents , and revealed three chains of M4N in the asymmetric unit ( ASU ) , organized as one dimer and one monomer . Refinement was continued using the native dataset . The two different crystal forms of M4NΔC , CF I and CF II , were solved by molecular replacement on the basis of the refined M4N coordinates . Using MOLREP ( Vagin and Teplyakov , 2000 ) , two copies of the dimeric assembly of the M4N structure were located in the ASU of CF I , and one monomer in the ASU of CF II . All models were completed by cyclic manual modeling with Coot ( Emsley and Cowtan , 2004 ) and refinement with PHENIX ( RRID:SCR_014224 ) ( Adams et al . , 2010 ) . Analysis with PROCHECK ( Laskowski et al . , 1993 ) showed excellent geometries for all structures . Data collection and refinement statistics are summarized in Supplementary file 1H . The three structures are deposited in the PDB ( Berman et al . , 2000 ) with accession codes: 5FZQ ( M4N ) , 5FZR ( M4NΔC CF I ) , 5FZS ( M4NΔC CF II ) . T . thermophilus HB8 and T . aquaticus YT-1 were obtained from the German Collection of Microorganisms and Cell Cultures ( DSMZ ) . Growth in liquid medium was performed under mild stirring ( 160 rpm ) in long necked ﬂasks at 68°C with DSMZ Medium 74 for T . thermophilus and DSMZ Medium 878 for T . aquaticus . Agar ( 1 . 6% w/v ) was added to the medium for growth on plates . When required , kanamycin ( 30 µg/ml ) and bleocin ( 10 µg/ml ) were added to the media . For puriﬁcation experiments 25 ml cultures were grown to an optic density of 0 . 7 OD600 ( ~12 hr ) and then re-inoculated in the same volume to an optical density of 0 . 035 OD600 . The process was repeated serially three times and two 5 ml samples were taken in each step for glycerol stocks and DNA purification . Transformation of T . thermophilus was performed as described previously ( Nguyen and Silberg , 2010 ) . Genomic and plasmid DNA from Thermus were puriﬁed from 5 ml cultures using the QIAamp DNA Mini Kit and the QIAprep Spin Miniprep Kit , respectively . T . thermophilus KM4 strain was generated by gene replacement as follows: two PCR products comprising each one 1 Kb of DNA upstream and downstream of rpsT were ampliﬁed from T . thermophilus HB8 genomic DNA and then fused by overlapping PCR . The resulting fragment , in which rpsT is substituted by a PstI site , was cloned in the KpnI/XbaI sites of plasmid pBlueScript II SK ( + ) . Next , a fragment from plasmid pKT1 ( Biotools , Spain ) , which contains the thermostable kanamycin resistance Kat gene under the control of the constitutive PslpA promoter , was inserted into the new PstI site . Direction of the Kat cassette insertion was selected , so transcription from the PslpA promoter continues through thx , a gene that is located downstream and is predicted to form an operon with rpsT . The 3 Kb ﬁnal construct cloned in pBluescript was subsequently amplified by PCR and the linear product was puriﬁed and transformed by electroporation in T . thermophilus HB8 . Integration of the Kat cassette was selected by growth in kanamycin . For the complementation in trans of rpsT from T . thermophilus , a PCR product of rpsT was ampliﬁed from genomic DNA and cloned in the SpeI/PstI sites of plasmid pJJSpro ( Nguyen and Silberg , 2010 ) generating plasmid pJJSpro-rpsTTt . The same approach was followed for rpsT in T . aquaticus ( pJJSpro-rpsTTa ) and in T . aquaticus rpsT alleles with two ( pJJSpro-rpsTTaM2 ) and four ( pJJSpro-rpsTTaM4N ) amino-acid substitutions . The PCR product for the two later constructs was ampliﬁed using the plasmids in which the synthesized genes were delivered as a template .	All life is built upon the chemical activity of proteins . For this activity , proteins need to fold into specific 3D structures . Protein folding is complicated and easily disrupted , and its evolutionary origin remains poorly understood . A possibility is that folded proteins arose through different genetic processes from shorter pieces of protein called peptides , which participated in an ancient , primordial form of life . One of these processes involves the same peptide being repeated within one protein chain . In 2015 , researchers identified 40 primordial peptides whose sequences appear in seemingly unrelated proteins . The study suggested that repetition allows peptides that are unable to fold by themselves to yield folded proteins . Now , Zhu et al . – who are members of the same research group who performed the 2015 study – have explored experimentally whether one of these peptides could indeed yield a folded protein by repetition . The studied primordial peptide gave rise to several protein folds seen today , including – by repetition – a type of fold called TPR . Zhu et al . tried to retrace the emergence of the TPR fold by taking a descendant of the primordial peptide from a ribosomal protein , which is unable to fold without the assistance of an RNA scaffold , and repeating it three times within the same protein chain . The ribosome is a central component of all living cells and evolves very slowly , and so the peptide Zhu et al . took from it is likely to retain many properties of its primordial ancestor . Further experiments found that the repeated peptide was indeed able to fold into a TPR-like structure , but needed several mutations to do so . Introducing these mutations back into the ribosomal protein , however , did not affect the survival and growth of the cell . Thus , they could have occurred without adverse effects during evolution . Structure is a prerequisite for chemical activity , but it is activity that is under selection in living beings . Having produced a new protein , Zhu et al . will now explore ways of endowing it with a selectable activity .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology" ]	2016	Origin of a folded repeat protein from an intrinsically disordered ancestor
Sexual maturation must occur on a controlled developmental schedule . In mammals , Makorin3 ( MKRN3 ) and the miRNA regulators LIN28A/B are key regulators of this process , but how they act is unclear . In C . elegans , sexual maturation of the nervous system includes the functional remodeling of postmitotic neurons and the onset of adult-specific behaviors . Here , we find that the lin-28–let-7 axis ( the ‘heterochronic pathway’ ) determines the timing of these events . Upstream of lin-28 , the Makorin lep-2 and the lncRNA lep-5 regulate maturation cell-autonomously , indicating that distributed clocks , not a central timer , coordinate sexual differentiation of the C . elegans nervous system . Overexpression of human MKRN3 delays aspects of C . elegans sexual maturation , suggesting the conservation of Makorin function . These studies reveal roles for a Makorin and a lncRNA in timing of sexual differentiation; moreover , they demonstrate deep conservation of the lin-28–let-7 system in controlling the functional maturation of the nervous system . The timing of sexual maturation is subject to complex regulation that reflects competing demands on biological systems ( Bogin et al . , 2011; Gluckman and Hanson , 2006; Parent et al . , 2003 ) . Over evolutionary time , the pressure on a species to reproduce rapidly is countered by the need to allow sufficient time for robust juvenile development . Optimum balance of these pressures is encoded by internal , genetic mechanisms that guide the timing of the juvenile-to-adult transition ( Zhu et al . , 2018 ) . At the level of an individual , internal and external stimuli—for example the presence of mates and food , as well as health and nutritional status—interact with these timers to determine the onset of sexual maturation ( Abreu and Kaiser , 2016; Avendaño et al . , 2017; Livadas and Chrousos , 2016; Plant , 2015 ) . In general , the nature of these genetic timers and their interaction with other signals is not well understood . Furthermore , while the hallmark of the juvenile-to-adult transition is the functional maturation of the germline and genitalia , reproductive maturation is typically accompanied by a much broader suite of changes in morphology , physiology , and behavior . These behavioral changes entail not only the activation of copulatory programs themselves , but also include modifications to decision-making circuits , allowing animals to incorporate reproductive drives when prioritizing behavioral programs . Whether genetic timers function at a single hub to coordinate behavioral transitions , or whether instead they may have distributed functions across different cell types , is unclear . The timing of sexual maturation has been extensively studied in mammals , where it is coordinated by the temporally controlled production of gonadal steroids at puberty ( Abreu and Kaiser , 2016; Avendaño et al . , 2017; Livadas and Chrousos , 2016; Plant , 2015 ) . This process is activated by the hypothalamic-pituitary-gonadal ( HPG ) axis , in which the pulsatile release of gonadotropin-releasing hormone ( GnRH ) by neurons in the hypothalamus activates gonadotropin production in the pituitary gland . This , in turn , triggers gonadal maturation and subsequent steroid hormone production . Upstream of HPG axis activation , a complex set of regulatory inputs converges on the release of the neuropeptide kisspeptin in the arcuate nucleus of the hypothalamus to stimulate GnRH production . Multiple studies suggest that kisspeptin release is influenced by environmental signals as well as internal timing cues ( Harter et al . , 2018 ) . Recent studies in mice and humans have identified a number of genes important for this process , including the miRNA regulators LIN28A/B , their target LET7 , and the Makorin MKRN3 , whose loss causes Central Precocious Puberty in humans ( Abreu et al . , 2013; Corre et al . , 2016; Gajdos et al . , 2010; Ong et al . , 2009; Park et al . , 2012; Yi et al . , 2018; Zhu et al . , 2010; Zhu et al . , 2018 ) . However , where and how these genes act to regulate the onset of reproductive maturation remains unclear . Moreover , some aspects of this model are likely to be specific to mammals , highlighting the importance of studying these questions in diverse animals . In the nematode C . elegans , juveniles pass through four larval stages before becoming sexually mature adults , offering an ideal opportunity to understand the mechanisms that regulate the juvenile-to-adult transition . Most studies addressing this question have focused on stage-specific events during larval development , particularly in the hypodermal seam cells , epidermal-like cells that lie along the sides of the body . In seam cells , stage-specific patterns of proliferation and differentiation are controlled by a complex mechanism known as the ‘heterochronic pathway’ ( Rougvie and Moss , 2013 ) . In this mechanism , temporally controlled waves of miRNA expression ( including the first two miRNAs discovered in any system , lin-4 and let-7 ) act on a variety of targets to specify stage-specific events and regulate the transition from one stage to the next . The conserved miRNA regulator lin-28 is also a key component of this mechanism . The heterochronic pathway also controls some events that occur during the larval-to-adult transition , including the cessation of molting , the fusion of the seam cells , and male-specific tail tip morphogenesis ( TTM ) , a remodeling of four postmitotic posterior hypodermal cells ( Del Rio-Albrechtsen et al . , 2006; Rougvie and Moss , 2013 ) . In some heterochronic mutants ( ‘precocious’ mutants ) , particular stage-specific events are skipped or occur too early; in others ( ‘retarded’ or ‘delayed’ mutants ) , events are delayed or absent . Recent studies of TTM have identified two additional heterochronic genes , lep-2/Makorin and the lncRNA lep-5 , that influence developmental timing by regulating LIN-28 stability ( Herrera et al . , 2016; Kiontke et al . , 2019 ) . Mutations in these genes cause a delayed phenotype , the retention of a larval-like , pointed ( ‘Lep’ ) tail tip in males . These mutants also have developmental defects in the body hypodermis , but they lack the seam cell lineage disruptions of other heterochronic mutants . Interestingly , lep-2 mutant males have defects in mating behavior , suggesting a role for lep-2 in the sexual maturation of the nervous system ( Herrera et al . , 2016 ) . While the maturation of the hypodermis has been well studied , the mechanisms that control the timing of adult-specific sexual behavior in C . elegans are poorly understood . In both sexes , males and hermaphrodites ( somatic females that produce both eggs and sperm ) , sexual behavior requires larval neurogenesis that produces sex-specific circuits for copulation and egg-laying behavior ( Emmons , 2018; García and Portman , 2016 ) . The development of these circuits is independent of the gonad , relying instead on intrinsic lineage programs in which chromosomal sex acts via the sex-determination hierarchy to regulate neurogenesis and cell fate specification ( Barr et al . , 2018 ) . Furthermore , multiple aspects of adult male behavior also require changes in synaptic connectivity and the functional modulation of pre-existing , post-mitotic neurons and circuits at the juvenile-to-adult transition ( Portman , 2017 ) . For example , the AWA olfactory neurons express the food-associated chemoreceptor odr-10 in larvae of both sexes , but in males , odr-10 expression is repressed at the larval-to-adult transition . This repression desensitizes males to food stimuli and promotes mate-searching behavior ( Ryan et al . , 2014 ) . In other cases , adult-specific activation of gene expression in males , such as the expression of daf-7 in the ASJ neurons ( Hilbert and Kim , 2017 ) facilitates adult behavior . The extent to which sex-shared , postmitotic neurons are functionally modulated at the juvenile-to-adult transition remains unclear , and the mechanisms that control the timing of this modulation are unknown . The striking involvement in mammalian puberty of orthologs of three C . elegans heterochronic genes ( lep-2 , lin-28 , and let-7 ) , together with the emerging appreciation that C . elegans sexual differentiation involves functional transitions in post-mitotic neurons , led us to explore the role of the heterochronic pathway in the timing of these changes . We find that the lep-2/lep-5 branch of the heterochronic pathway plays a key role in timing the sexual differentiation of the C . elegans nervous system , demonstrating conservation of mechanism across species and raising the possibility that an unidentified lep-5-like lncRNA may play a key role in mammalian sexual maturation . Moreover , we find that this pathway acts cell-autonomously in multiple neurons , suggesting that genetic timing information is widely distributed across the nervous system . Multiple targets of lin-41 , including mab-3 and lin-29a , integrate the heterochronic and sex-determination pathways to regulate downstream effector genes and the activation of multiple adult behaviors . Interestingly , although the mammalian Makorin MKRN3 functions in the opposite direction as its C . elegans ortholog lep-2 , we find that MRKN3 expression in the C . elegans nervous system is sufficient to delay sexual maturation , indicating that the pathway in which it interacts is functionally conserved across species . These findings identify the heterochronic pathway as an ancient regulator of sexual differentiation of the nervous system , provide specific mechanistic hypotheses for the functions of these genes in the mammalian brain , and suggest an important role for a yet-unidentified mammalian lncRNA in the timing of puberty . To visualize the sexual maturation of the C . elegans nervous system , we studied the developmental dynamics of the expression of five genes known to display sexually dimorphic expression in adults ( Figure 1 ) . Four of these reflect sexual dimorphism in shared ( i . e . , non-sex-specific ) neurons: srj-54 , a GPCR of unknown function expressed male-specifically in the AIM interneurons ( Lee and Portman , 2007; Portman , 2007 ) ; odr-10 , a chemoreceptor whose male-specific downregulation in the AWA olfactory neurons is associated with decreased diacetyl and food detection ( Lee and Portman , 2007; Ryan et al . , 2014; Sengupta et al . , 1996 ) ; mab-3 , a doublesex-family transcription factor that is male-specifically expressed in the ADF chemosensory neurons , where it promotes the detection of hermaphrodite sex pheromone ( Fagan et al . , 2018; Yi et al . , 2000 ) ; and daf-7 , a TGFβ-superfamily ligand that is male-specifically expressed in the ASJ chemosensory neurons and promotes the food-leaving behavior of adult males ( Hilbert and Kim , 2017; Ren et al . , 1996 ) . We also studied pkd-2 , a marker for the male-specific CEM pheromone-detection neurons , which exist in an undifferentiated state until late larval development ( Barr and Sternberg , 1999; Narayan et al . , 2016; Srinivasan et al . , 2008; Sulston et al . , 1980; Sulston et al . , 1983 ) . Importantly , all of these neurons are generated during embryogenesis and , with the exception of the CEMs , are thought to be functional components of larval circuits . For all of the markers we examined , we found that sex-specific expression patterns were absent in young larvae , emerging only late in larval development ( Figure 1 ) . In L3 males , we detected virtually no expression of srj-54 in AIM , mab-3 in ADF , or , consistent with a previous report ( Hilbert and Kim , 2017 ) , daf-7 in ASJ . Moderate expression of these markers was observed in L4 males , but the full extent of sex differences in expression was not fully apparent until adulthood ( Figure 1A , C , D ) . With regard to odr-10 expression in AWA , we observed the opposite pattern , confirming our earlier findings ( Ryan et al . , 2014 ) : reporter expression was readily detectable in L3 males , as it is in adult hermaphrodites , and diminished in males through L4 into adulthood ( Figure 1B ) . Thus , at least with respect to these markers , the AWA , AIM , ADF , and ASJ neurons in larval males exist in a hermaphrodite-like ‘ground state’ before they acquire adult-specific characteristics during the juvenile-to-adult transition . The CEMs , which are absent in hermaphrodites , underwent maturation with similar timing . As expected ( Barr and Sternberg , 1999 ) , we detected no expression of pkd-2 in L3 males; intermediate expression was observed in L4 animals and was further strengthened in day-one adults ( Figure 1E ) . This timing is consistent with that previously reported for other CEM-specific genes ( Barr et al . , 2001; Portman and Emmons , 2004; Wang et al . , 2015 ) as well as the cholinergic maturation of these neurons ( Pereira et al . , 2015 ) . Together , these findings reinforce the idea that the larval-to-adult transition involves not only the incorporation of new male-specific neurons into the nervous system ( Barr et al . , 2018 ) , but also the molecular maturation of multiple classes of pre-existing neurons . To ask whether the heterochronic pathway has a role in the timing of these molecular maturation events , we examined animals carrying mutations in lin-28 and let-7 , as these genes are at the core of the late heterochronic timer ( Rougvie and Moss , 2013 ) ( Figure 2A , B ) . In the hypodermal seam and the male tail tip hypodermis , lin-28 mutants precociously execute late-larval patterns of proliferation , differentiation , and morphogenesis ( Ambros , 1989; Ambros and Horvitz , 1984; Herrera et al . , 2016; Moss et al . , 1997 ) ; conversely , let-7 mutants retain larval-specific fates and male tail morphology as adults ( Reinhart et al . , 2000; Slack et al . , 2000; Vadla et al . , 2012 ) . We found that gene expression dynamics in the nervous system were similarly altered in these mutants ( Figure 2C–G ) . lin-28 mutant males exhibited precocious activation of srj-54 in AIM , with clear reporter expression in over half of L3 males and over 80% of L4 males . In contrast , the sexual maturation of AIM was delayed in let-7 mutant males , with very little srj-54 expression detectable in L4 and only moderate expression in one-day adults ( Figure 2C ) . ( Since let-7 null mutants are inviable , these experiments used the hypomorphic allele n2853; null phenotypes would likely be stronger ( Reinhart et al . , 2000 ) . ) We observed similar effects of lin-28 and let-7 mutations on pkd-2 reporter expression in the CEMs ( Figure 2G ) : loss of lin-28 resulted in precocious expression in L3 males , while loss of let-7 slightly reduced pkd-2 expression in young adults . let-7 mutant adults also retained larval-like expression of odr-10 in AWA . Unexpectedly , however , odr-10 downregulation did not occur precociously in lin-28 mutant males; instead , levels increased in L4 ( Figure 2D ) . Because odr-10 downregulation does depend on genes downstream of lin-28 ( see below ) , some adult-specific characteristic ( s ) that are independent of lin-28 are likely to be required for odr-10 downregulation ( e . g . , signals from male-specific components of the adult nervous system ) . Finally , we also observed precocious ADF expression of mab-3 in lin-28 mutant males ( Figure 2E ) , with strong expression initiating in L3 . Together , these results demonstrate that the lin-28–let-7 regulatory axis governs the timing of key events in the sexual maturation of the C . elegans nervous system . In the hypodermis , a key target of let-7 is lin-41 , which encodes an RBCC protein of the NHL family ( Slack et al . , 2000 ) . We found that loss of lin-41 function resulted in precocious activation of srj-54 in AIM and pkd-2 in CEM , and precocious loss of odr-10 in AWA ( Figure 2C , D , G ) . Reciprocally , a lin-41 ( gf ) allele ( Del Rio-Albrechtsen et al . , 2006 ) caused somewhat delayed maturation of the expression of srj-54 , though it had no statistically significant effects on odr-10 or pkd-2 expression ( Figure 2C , D , G ) . This discrepancy could reflect cell-type-specific thresholds for the degree of lin-41 function necessary to elicit a gain-of-function phenotype . Together , these results indicate that lin-41 is an important effector of let-7 function in the nervous system , as it is in the hypodermis . In the timing of male tail tip morphogenesis , the Makorin lep-2 and the lncRNA lep-5 act upstream of lin-28 ( Herrera et al . , 2016; Kiontke et al . , 2019 ) . Because loss-of-function mutations in the lep-2 ortholog MKRN3 cause Central Precocious Puberty in humans ( Abreu et al . , 2013 ) and because of the sexual behavior defects in C . elegans lep-2 mutant males ( Herrera et al . , 2016 ) , we considered the possibility that the lep-2/lep-5 branch of the heterochronic pathway might function in sexual differentiation of the C . elegans nervous system . Consistent with this idea , we found that these animals exhibited striking defects in neuronal maturation: for srj-54 , odr-10 , and mab-3 , day one adult lep-5 males retained the expression pattern typically seen in wild-type L3 males ( Figure 2C , D , E ) . We observed weaker phenotypes for daf-7 and pkd-2: in both cases , expression increased during larval development , but failed to reach the level typically seen in young adults ( Figure 2F , G ) . Thus , lep-2 is required for the sexual differentiation of AIM , AWA , and ADF , and also contributes to this process in ASJ and CEM . Intriguingly , the absence of sexually mature characteristics in young lep-2 adult males stands in contrast to the Central Precocious Puberty that results from loss of human MKRN3 function , a distinction to which we return below . As is the case for tail tip morphogenesis ( Kiontke et al . , 2019 ) , the lep-5 mutant phenotype almost completely phenocopied that of lep-2 . Mutant males failed to adopt adult-specific gene expression patterns in AIM , AWA , and ADF , but had only modest defects in marker expression in ASJ and CEMs ( Figure 2C–G ) . lep-5 encodes a lncRNA that is predicted to adopt a compact secondary structure with prominent stem-loop structures near its 5´ and 3´ ends , as well as a two-part central double-stranded ‘zipper’ region ( Figure 2—figure supplement 1A ) ( Kiontke et al . , 2019 ) . lep-5 has been proposed to function as an RNA scaffold , forming a tripartite complex with LEP-2 and LIN-28 to promote LIN-28 degradation ( Kiontke et al . , 2019 ) . We found that a Crispr-mediated deletion of the predicted 3´ hairpin , lep-5 ( fs20 ) , phenocopied the lep-5 null allele , showing a nearly complete lack of srj-54 expression in AIM in adult males ( Figure 2—figure supplement 1B ) . lep-5 ( fs8 ) , a point mutation that disrupts the 5´ hairpin , as well as lep-5 ( fs21 ) , a Crispr-engineered allele bearing a cluster of five point mutations predicted to disrupt the central zipper , caused a complete loss of srj-54 expression in adults ( Figure 2—figure supplement 1B , C ) . Furthermore , lep-5 ( fs21fs25 ) , a double mutant that carries a set of compensatory changes predicted to restore the central zipper , completely restored adult expression of srj-54 ( Figure 2—figure supplement 1B , C ) . The phenotypes of these mutants are consistent with those seen in male tail tip morphogenesis ( Kiontke et al . , 2019 ) , indicating that lep-5 likely acts through a similar mechanism in both tail tip morphogenesis and neuronal maturation . We considered several scenarios by which heterochronic genes could control the timing of neuronal maturation ( Figure 3A ) . In one , heterochronic genes might act in the gonad to control the secretion of hormonal signals . However , this seems unlikely , as the development of the C . elegans gonad is largely independent of the heterochronic pathway ( Rougvie and Moss , 2013 ) and , with the exception of vulva and hindgut development , ( Kimble , 1981; Sulston et al . , 1980 ) , the gonad does not play a significant role in the sexual differentiation of the C . elegans soma ( Barr et al . , 2018 ) . Alternatively , the heterochronic pathway might act as a central somatic timer ( e . g . , in seam cells or a master ‘clock’ neuron ) from which it broadcasts temporal information to the nervous system . Finally , individual neurons might possess their own timers , such that temporal control is distributed throughout the nervous system . To address this , we first confirmed earlier reports suggesting that lin-28 , lep-2 , and lep-5 are broadly expressed in the nervous system ( Herrera et al . , 2016; Kiontke et al . , 2019; Moss et al . , 1997 ) . Using a translational LIN-28::GFP reporter , we observed widespread expression in L1 and L2 animals , including extensive expression in the nervous system ( Figure 3B ) . LIN-28::GFP abundance decreased during L2 in the nervous system and elsewhere , and by L3 , expression was virtually undetectable , consistent with previous reports ( Moss et al . , 1997 ) . We observed an inverse pattern of Plep-5::GFP expression in the nervous system , consistent with its previously described dynamics in the hypodermis ( Kiontke et al . , 2019 ) . Plep-5::GFP was detectable in very few neurons in L1 larvae , but broad neuronal expression was apparent by L2 and L3 ( Figure 3C ) . Interestingly , lep-5 reporter expression appears to be restricted to a subset of the nervous system ( Figure 3C ) , an important issue for future work . With a Plep-2::lep-2::GFP translational reporter ( Herrera et al . , 2016 ) , we observed expression in many head neurons in early larvae ( not shown ) and in L3 , L4 , and adult males ( Figure 3C ) . Because lep-2 expression levels remain essentially constant during larval development ( Herrera et al . , 2016 ) , lep-2 likely does not itself provide timing information but rather seems to provide a permissive cue for the transition into adulthood . Instead , lep-5 likely provides an instructive signal important for determining the onset of the sexual differentiation of the nervous system , as it does in the hypodermis ( Kiontke et al . , 2019 ) . To ask whether the heterochronic pathway functions in the nervous system itself , we restored expression of wild-type lep-2 or lin-28 in the nervous system of lep-2 or lin-28 mutants . Expression of lep-2 under the control of the AIM-specific promoter Peat-4prom11 ( Serrano-Saiz et al . , 2017 ) restored srj-54 expression to lep-2 mutant adults , but had no effect on odr-10 expression in AWA ( Figure 3E , F ) , indicating that lep-2 acts cell-autonomously to control timing in AIM . Furthermore , Peat-4prom11::lin-28 rescued the precocious maturation of AIM in lin-28 mutant larvae ( Figure 3G ) . Thus , timing information intrinsic to AIM itself is critical for the timely control of AIM’s maturation . This property is not limited to AIM , as AWA-specific expression of lep-2 ( Pgpa-4∆6::lep-2 ( + ) ) rescued the lack of odr-10 downregulation in AWA in lep-2 mutants ( Figure 3H ) . Together , these findings indicate that the heterochronic pathway does not control an organism-wide temporal signal that is broadcast from a central timer; rather , timing information is distributed , such that individual neurons maintain their own internal clocks . Multiple studies have shown that lin-28 negatively regulates the biogenesis and activity of multiple miRNA targets , including let-7 , in C . elegans , Drosophila , and mammals ( Tsialikas and Romer-Seibert , 2015 ) . The decay of lin-28 activity is a key regulatory point in the progression into adulthood , allowing the activation of let-7 and the subsequent downregulation of its targets . In the hypodermal seam , lin-28 decay is controlled at both the mRNA and protein levels ( Huang et al . , 2011; Morita and Han , 2006; Seggerson et al . , 2002; Weaver et al . , 2017; Weaver et al . , 2014 ) . In the male tail tip , lep-2 and lep-5 function together to promote the degradation of LIN-28 protein during L3 ( Herrera et al . , 2016; Kiontke et al . , 2019 ) . We carried out a series of genetic and molecular epistasis experiments to ask which aspects of this regulatory logic extend to the nervous system . Unlike wild-type males , in which neuronal LIN-28::GFP decays during L2 and is virtually absent by L3 , lep-2 mutant males exhibited strong neuronal LIN-28::GFP that persisted into L3 and L4 ( Figure 4A and data not shown ) . Thus , consistent with its role in the tail tip and other hypodermal cells ( Herrera et al . , 2016 ) , lep-2 promotes the timely decay of LIN-28 in the nervous system . We also examined expression of the AIM maturation marker srj-54 in lin-28; lep-2 double mutants . Like lin-28 single mutants , these animals exhibited precocious AIM maturation ( Figure 4B ) , confirming that lep-2 acts genetically upstream of lin-28 . We observed a similar phenotype in lin-28; lep-5 double mutants ( Figure 4C ) , indicating that lep-5 also acts upstream of lin-28 with respect to nervous system maturation . Finally , we confirmed that lin-28 acts upstream of let-7 in the nervous system by examining srj-54 expression in lin-28; let-7 mutants . As expected , AIM maturation was delayed compared to wild-type ( Figure 4D ) . Thus , lep-2 and lep-5 regulate the lin-28–let-7 regulatory module to control the onset of sexual differentiation of the C . elegans nervous system . Together with previous work ( Kiontke et al . , 2019 ) , our results support the idea that the onset of lep-5 expression is a critical step in determining the timing of LIN-28 decay and the subsequent progression into adulthood . To investigate mechanisms controlling the timing of lep-5 activation , we considered a role for genes involved in the timing of early larval development . The miRNA lin-4 and its key target , the transcription factor lin-14 , regulate stage-specific development in L1 and L2 ( Rougvie and Moss , 2013 ) . We found that neuronal expression of Plep-5::GFP was significantly disrupted in lin-4 mutants: very little reporter expression was detectable in the nervous system of larvae of any stage ( Figure 3—figure supplement 1 ) . This suggests that lin-4 is necessary to trigger the onset of lep-5 expression in the nervous system , consistent with its role in promoting the L1 to L2 transition ( Rougvie and Moss , 2013 ) . Because the lep-2 and lin-28 orthologs MKRN3 and LIN28A/B are both implicated in the timing of human puberty , our results and those of Herrera et al . ( 2016 ) raise the possibility that MKRN3 might act upstream of LIN28A/B in the mammalian nervous system . However , while loss of the C . elegans Makorin lep-2 causes a delay in sexual differentiation , loss-of-function mutations in MKRN3 are associated with the opposite phenotype , Central Precocious Puberty ( Abreu et al . , 2013 ) . Several models can explain this apparent discrepancy . For example , it could be the case that both MKRN3 and LEP-2 inhibit LIN28A/B and LIN-28 , but LIN28A/B might have multiple functions in mammalian puberty , acting both to promote and inhibit its onset ( Corre et al . , 2016 ) . Alternatively , MKRN3 might act to stabilize LIN28A/B , possibly by inhibiting another mammalian Makorin that might function more like C . elegans lep-2 . Finally , MKRN3 might regulate other substrates that function downstream of , or in parallel with , mammalian LIN28A/B in regulating puberty . The first model predicts that MKRN3 might be able to functionally substitute for lep-2 . However , we found that pan-neural expression of MKRN3 in lep-2 mutants had no apparent effect on srj-54 expression ( Figure 4—figure supplement 1A ) . Though we cannot rule out the possibility that technical issues hindered rescue , this result suggests that LEP-2 and MKRN3 do not have equivalent biochemical functions . Because wild-type MKRN3 inhibits sexual differentiation in humans , we next asked whether it might have the same effect in C . elegans . Remarkably , expression of MKRN3 from a pan-neuronal promoter in a wild-type background caused a modest but statistically significant delay in the activation of srj-54 expression in AIM and the downregulation of odr-10 expression in AWA ( Figure 4E , F ) . However , we observed no effect of MKRN3 overexpression on pkd-2 expression ( Figure 4G ) . The observation that MKRN3 retains its ability to inhibit some aspects of sexual maturation when expressed in the C . elegans nervous system raises the intriguing possibility that MKRN3 exerts an inhibitory effect on the conserved lin-28–let-7-dependent heterochronic timer and , moreover , that there is deep functional conservation of the mechanism in which it participates . The lack of an apparent effect of MKRN3 overexpression on CEM maturation is consistent with our finding that these neurons rely less on lep-2 and lep-5 function ( Figure 2G ) , indicating that their maturation occurs through mechanisms that are at least partially distinct from those that operate in AWA and AIM . Our survey of heterochronic mutants ( Figure 2 ) indicated that the lin-28–let-7 pathway controls sexual maturity via the RNA-binding protein LIN-41 , a direct target of let-7 ( Slack et al . , 2000 ) . LIN-41 regulates gene expression by binding to the 5´ or 3´ UTR of multiple target genes , including lin-29a , mab-3 , dmd-3 , and mab-10 ( Aeschimann et al . , 2017 ) . In the case of ADF , whose maturation is marked by the activation of mab-3 , this provides a continuous molecular pathway connecting the lep-2–lep-5–lin-28 timer to the sexual differentiation of ADF . With regard to the other neurons we studied , mab-3 and lin-29a appear to have overlapping but distinct roles in sexual maturation . Studies carried out in parallel with this work found that lin-29a regulates expression of srj-54 in AIM and daf-7 in ASJ ( Pereira et al . , 2019 ) . We found that the full extent of srj-54 expression in AIM also depended on mab-3 ( Figure 4—figure supplement 1B ) . Thus , both lin-29a and mab-3 promote AIM maturation , though the role of mab-3 may be secondary and/or indirect . Furthermore , we found that the adult-specific repression of odr-10 in AWA was lost in mab-3 mutants but was unaffected by loss of lin-29a ( Figure 4—figure supplement 1C , D ) . Thus , with respect to the molecular maturation events we have investigated here , both mab-3 and lin-29a link the heterochronic pathway to downstream effectors of sexual differentiation . Because both mab-3 and lin-29a are also targets of the master sexual regulator tra-1 ( Fagan et al . , 2018; Pereira et al . , 2019; Yi et al . , 2000 ) , these genes integrate sexual and temporal information to specify male-specific aspects of the sexual maturation of the C . elegans nervous system . With the exception of srj-54 , whose function is unknown , each of the molecular markers used in this work is associated with adult-specific male behavior . Indeed , previous research has found that lep-2 mutants have defects in male mating and food-leaving behaviors ( Herrera et al . , 2016 ) , although the underlying mechanisms were unclear . We therefore surveyed these and other aspects of adult male behavior in both lep-2 and lep-5 mutants . While many heterochronic mutants have drastic , pleiotropic effects that interfere with behavioral assays , lep-2 and lep-5 males are , with the exception of a moderate defect in tail tip retraction , morphologically wild type . The retention of a pointed tail tip into adulthood is in itself thought not to significantly compromise male sexual behavior ( Del Rio-Albrechtsen et al . , 2006 ) . Food-leaving behavior , the propensity of well-fed males to leave a food source in search of mates , is manifested only in adulthood ( Lipton et al . , 2004 ) . Multiple mechanisms promote male food-leaving , including signals from the male-specific ray sensory neurons , the neuropeptide receptor PDFR-1 , the TGFβ-superfamily ligand DAF-7 , and downregulation of the food-associated chemoreceptor odr-10 ( Barrios et al . , 2012; Barrios et al . , 2008; Hilbert and Kim , 2017; Lipton et al . , 2004; Ryan et al . , 2014 ) . We found that lep-2 and lep-5 mutant males exhibited markedly decreased food-leaving behavior as adults ( Figure 5A ) , indicating that both of these genes are important for the acquisition of male sexual drive . We also examined attraction to ascaroside sex pheromones , a male-specific behavior that depends partially on mab-3 function in ADF ( Fagan et al . , 2018; Srinivasan et al . , 2008 ) . We found that ascaroside attraction was absent in male larvae: L3 males were slightly repelled by an ascr#2/#3/#8 blend , reminiscent of the behavior of adult hermaphrodites ( Fagan et al . , 2018 ) ( Figure 5B ) . The transition to the adult behavioral state requires lep-2 and lep-5 , as pheromone attraction was abolished in these mutants ( Figure 5B ) . In contrast , lin-29a mutant males exhibited no apparent defects in ascaroside attraction ( Figure 3—figure supplement 1D ) , suggesting that mab-3 is the primary effector of the sexual maturation of ADF . Finally , we examined two sub-steps of male mating , response behavior ( ‘Rsp’ , the typical backward-locomotion response of males to physical contact with hermaphrodites ) and the subsequent vulva-location behavior ( ‘Lov’ , the ability of the male to locate the hermaphrodite vulva ) ( Liu and Sternberg , 1995 ) . While these behaviors depend primarily on male-specific sensorimotor circuits in the tail , sex-specific characteristics of shared neurons are also likely to be important for their execution ( Barr and Sternberg , 1999; Koo et al . , 2011; Liu and Sternberg , 1995; Liu et al . , 2011; Sherlekar et al . , 2013 ) ( R . M . M . and D . S . P . , unpublished results ) . We found that response behavior was reduced in both lep-2 and lep-5 mutant males , while vulva-location behavior was almost completely absent ( Figure 5C , D ) . Together , these findings demonstrate that the Makorin lep-2 and the lncRNA lep-5 , and by extension , the heterochronic pathway , are important for implementing adult-specific sexual behavior in C . elegans males . Although it has long been appreciated that animal behavior undergoes functional transitions commensurate with sexual maturation ( Spear , 2000 ) , the mechanisms that determine the timing of these changes are poorly understood . Indeed , in humans , adolescence is a period of heightened susceptibility to a variety of neuropsychiatric disorders ( Jones , 2013; Walker et al . , 2017 ) , highlighting the importance of understanding the regulatory mechanisms that control these transitions . Here , we unite disparate findings from studies of mammalian puberty and C . elegans developmental biology to establish that a conserved developmental timer functions cell-autonomously to coordinate the sexual maturation of the C . elegans nervous system ( Figure 6 ) . Central to this mechanism is the temporal control provided by the lncRNA lep-5 , which , together with the Makorin lep-2 , promotes the timely decay of the conserved miRNA inhibitor LIN-28 ( Kiontke et al . , 2019 ) . This decay permits the biosynthesis of mature let-7 miRNA and , through the pathway shown in Figure 6 , promotes the juvenile-to-adult transition . The maturation events we have studied here entail not the incorporation of new neurons and circuits into the nervous system , but rather gene expression and functional changes in pre-existing , postmitotic neurons . As such , these events parallel similar changes in the mammalian hypothalamus , where both gene expression and physiological changes determine the timing of HPG axis maturation and the onset of puberty ( Abreu and Kaiser , 2016; Avendaño et al . , 2017; Livadas and Chrousos , 2016; Plant , 2015 ) . Because MKRN3 , LIN28A/B , and LET7 are expressed in the hypothalamus and elsewhere in the mammalian brain , our results indicate that the neuronal function of the heterochronic pathway is an ancient regulator of the juvenile-to-adult transition . In this view , the activational effects of gonadal steroids at puberty ( McCarthy et al . , 2012 ) are likely to be a recent evolutionary adaptation that reinforces sexual differentiation of the mammalian nervous system via an endocrine feedback loop . Moreover , it suggests that puberty-independent events outside the hypothalamus that contribute to behavioral maturation during mammalian adolescence , such as the transition from play to aggression in Siberian hamsters ( Paul et al . , 2018 ) , could also be timed by the heterochronic pathway . In addition to identifying neuronal functions for lep-2 , lin-28 , and let-7 , we also show that the recently identified lncRNA lep-5 functions in this pathway . lep-5 was identified in forward genetic screens for male tail morphogenesis defects; in lep-5 mutants , LIN-28 protein persists longer than in wild-type , delaying the onset of expression of dmd-3 , the master regulator of tail tip morphogenesis ( Kiontke et al . , 2019 ) . Because lep-5 is expressed in a temporal wave whose rise corresponds to the time at which lin-28 expression decays , lep-5 is thought to provide an instructive temporal cue for the destabilization of LIN-28 and , in turn , allow production of the let-7 miRNA and the progression through late-larval development . Because lep-2 mutants have a phenotype essentially identical to that of lep-5 , Kiontke et al . , 2019 have proposed that lep-5 serves as an RNA scaffold that recruits both LEP-2 and LIN-28; such a model is supported by in vivo crosslinking experiments demonstrating that lep-5 binds to both LEP-2 and LIN-28 . As a member of the Makorin family , LEP-2 may possess both RNA-binding and putative E3 ubiquitin ligase activities ( Herrera et al . , 2016 ) , suggesting that a tripartite LEP-2–lep-5–LIN-28 complex could destabilize LIN-28 through ubiquitination by LEP-2 ( Kiontke et al . , 2019 ) . Sequence homologs of lep-5 have not been found outside of nematodes; however , lncRNAs are known to evolve rapidly , usually making it impossible to identify orthologs based on primary sequence alone ( Diederichs , 2014 ) . It is tempting to speculate that structural and/or functional orthologs of lep-5 are present in mammals and could have an important role in the regulation of LIN28 and its control of puberty . Intriguingly , mammalian MKRN3 has the opposite effect on developmental timing as its C . elegans ortholog lep-2 . While lep-2 mutant adults exhibit juvenile characteristics in the male tail tip ( Herrera et al . , 2016 ) and nervous system ( this work ) , null or hypomorphic mutations in human MKRN3 lead to Central Precocious Puberty ( Abreu et al . , 2013 ) , in which children younger than eight ( girls ) or nine ( boys ) years of age develop secondary sexual characteristics . This may reflect a generally more complex role for a putative MKRN3–LIN28–LET7 regulatory module in mammalian puberty compared to C . elegans sexual maturation . For example , overexpression of Lin28a delays puberty in male and female mice ( Corre et al . , 2016; Zhu et al . , 2010 ) , consistent with the maturation-inhibiting function of C . elegans lin-28 . However , Lin28b–/– mice , as well as mice overexpressing let-7 , also exhibit delayed sexual maturation , in contrast to the roles of lin-28 and let-7 in C . elegans; moreover , these effects were seen only in males ( Corre et al . , 2016 ) . In mammals , these genes can be easily imagined to function at multiple steps in sexual development , such that apparently antagonistic activities could arise from roles for the mammalian heterochronic pathway in the production of both maturation-promoting and maturation-inhibiting signals . Clarification of these issues will require regionally and temporally controlled manipulations of mammalian gene activity . Here , we found that human MKRN3 was not able to rescue the defects of C . elegans lep-2 mutants . Remarkably , however , MKRN3 overexpression caused a slight delay in the sexual maturation of two neuron types , AIM and AWA , in the C . elegans nervous system . This strongly suggests that the biochemical functions of LEP-2 and MKRN3 are not equivalent , and , moreover , that the mechanism by which MKRN3 regulates pubertal timing is conserved in C . elegans . Several potential models are consistent with these observations ( Figure 6 ) . MKRN3 might inhibit LEP-2 , perhaps through a dominant-negative mechanism in which MKRN3 binds to lep-5 but is unable to recognize C . elegans LIN-28 . Alternatively , MKRN3 might somehow stabilize LIN-28 , rather than promoting its degradation , or , less likely , MKRN3 could function in parallel with the lep-2—lin-28 module . Further , we cannot rule out the possibility that MKRN3 overexpression artifactually inhibits lep-2 function . Ours are not the first studies to examine roles for the heterochronic pathway in the C . elegans nervous system . Early in larval development , the timing of synaptic remodeling of a class of GABAergic motorneurons is controlled by the heterochronic gene lin-14 , which specifies multiple aspects of the L1 stage ( Hallam and Jin , 1998 ) . Later , a hermaphrodite-specific aspect of nervous system maturation , the outgrowth of the HSN axon , is controlled by lin-14 and lin-28 ( Olsson-Carter and Slack , 2010 ) . However , in neither case have the relevant targets of heterochronic function been identified . Interestingly , the heterochronic pathway , particularly let-7 and lin-41 , also has a role in regulating damaged-induced regeneration of the AVM axon in adults , an ability that declines with age ( Zou et al . , 2013 ) . Moreover , studies in Drosophila also support a role for the heterochronic pathway in the timing of major transitions in the nervous system ( Faunes and Larraín , 2016 ) . let-7 and related miRNAs control temporal specification of neuronal fate at the larval-to-pupal transition ( Chawla et al . , 2016; Marchetti and Tavosanis , 2017; Wu et al . , 2012 ) ; they also regulate remodeling of neuromuscular anatomy and function at the pupal-to-adult transition ( Caygill and Johnston , 2008; Sokol et al . , 2008 ) . Recently , Drosophila lin-28 has been found to be widely expressed in the nervous system in late third instar larvae , and loss of lin-28 function accelerates the larval-to-pupal transition ( González-Itier et al . , 2018 ) . Together , these findings indicate an ancient role for this regulatory module in orchestrating developmental transitions in the nervous system . Work carried out in parallel with the studies reported here ( Pereira et al . , 2019 ) found that the heterochronic pathway acts cell-autonomously to control several other aspects of male-specific sexual differentiation , including a neurotransmitter fate switch and changes in synaptic connectivity . Interestingly , Pereira et al . identified lin-29a , a male-specific isoform of lin-29 , as a key mediator of many of these transitions , acting to integrate temporal and sexual signals . Here , our studies consider other aspects of male-specific functional maturation and focus on upstream components of the heterochronic pathway , particularly lep-2 and lep-5 , best known for their roles in male tail development ( Herrera et al . , 2016; Kiontke et al . , 2019 ) . The strong nervous system phenotypes of these mutants indicate that this specialized branch of the heterochronic pathway is critical for the sexual maturation of the nervous system . Moreover , these studies strongly suggest that the function of MKRN3 in regulating human puberty is related to the LIN28–LET7 axis and predict the central involvement of a yet-unidentified lncRNA in this process . Though all of the maturation events we have studied here are male-specific , the heterochronic pathway almost certainly has a role in hermaphrodite-specific and non-sex-specific maturation of the C . elegans nervous system as well . Indeed , as mentioned above , differentiation of the hermaphrodite-specific HSN neuron requires heterochronic function ( Olsson-Carter and Slack , 2010 ) ; however , little is known about non-male-specific changes in neuronal function that accompany sexual maturation . Interestingly , two recent studies have demonstrated functional transitions in chemosensory coding and behavior that occur at the juvenile-to-adult transition in hermaphrodites ( Fujiwara et al . , 2016; Hale et al . , 2016 ) . In one of these cases ( Fujiwara et al . , 2016 ) , signaling from the gonad plays an important role , raising the more general possibility that signals from the gonad could intersect with cell-autonomous timers to provide more flexible control of behavior . Although the phenomena we have studied here derive their sex-specificity by the integration of temporal and sexual signaling at the level of mab-3 and lin-29a , other interactions between sex-determination and heterochronic timing are likely . By ChIP-seq studies , lin-28 has been identified as a direct target of tra-1 ( Berkseth et al . , 2013 ) ; both tra-1 and tra-2 are potential targets of the miRNA let-7 ( targetscan . org ) ; the abundance of TRA-1 protein has been shown to be developmentally regulated ( Weinberg et al . , 2018 ) ( H . L . and D . S . P . , in prep . ) ; and some phenotypes of heterochronic mutants can be considered sexual transformations ( Ambros and Horvitz , 1984 ) . Interestingly , many aspects of the timing of mammalian puberty also appear to be sex-specific ( Abreu and Kaiser , 2016; Cousminer et al . , 2016 ) . Together , these observations blur the distinctions between sexual differentiation mechanisms and developmental timing pathways , helping provide a mechanistic understanding of the observation that sexual dimorphism often arises through heterochrony ( McNamara , 2012 ) . Because the deep crosstalk between sexual and temporal information is unlikely to be limited to C . elegans , reproductive maturation disorders and sex-biased disorders of the nervous system in humans will both be informed by a deeper appreciation of these intersections . In general , C . elegans were cultured using standard methods ( Stiernagle , 2006 ) . All strains used here carried the mutation him-5 ( e1490 ) to increase the numbers of males , such that we refer to him-5 ( e1490 ) as the wild-type . All strains used in this work are listed in the Key resources table . . To isolate populations at specific larval stages , adult hermaphrodites were allowed to lay eggs for two hours , roughly synchronizing their progeny . Because male tail morphogenesis cannot be used to reliably stage some heterochronic mutant males under low magnification , sibling hermaphrodites were used as proxies to roughly stage animals . For all individuals scored , larval stage was definitively determined by the progression of gonadal development using DIC microscopy ( Zeiss Axioplan 2 , 63x Plan-Apo objective ) . Males in which the linker cell had migrated dorsally but not yet ventrally were staged as L3 larvae . Males in which the linker cell had migrated ventrally and extended posteriorly were staged as L4 larvae . One-day old adults were obtained by picking L3-L4 larvae and allowing them to mature overnight . Because lin-28 hermaphrodites have severe egg-laying defects , we passed all strains containing lin-28 mutations through the dauer stage to suppresses this phenotype ( Liu and Ambros , 1991 ) , allowing us to use timed egg-laying for rough synchronization . Animals were mounted on 4% agarose pads in M9 buffer supplemented with levamisole . GFP and DIC brightfield images were obtained using a 63x PlanApo objective on a Zeiss Axioplan 2 . GFP intensity was qualitatively determined to be bright ( 3 ) , moderate ( 2 ) , faint ( 1 ) , or absent ( 0 ) . For rescue and overexpression experiments , the experimenter was blind to the presence or absence of the expression array when scoring reporter fluorescence . Animals not bearing the array were used as non-transgenic sibling controls . Representative images were taken using the same illumination intensity and exposure time . Images of ksIs2 were captured using an Apotome . Fluorescence intensity data were compared using Mann-Whitney tests ( for pairwise analyses ) or Kruskal-Wallis tests with Dunn’s correction ( for multiple comparisons ) . For these and all other statistical analyses reported in this work , statistical significance is indicated in the figures as follows: 0 . 01 < p < 0 . 05; 0 . 001 < p < 0 . 01; *p<0 . 001 . Animals were mounted as described above and images were obtained using a Leica SP5 Confocal Microscope with a HyD hybrid detector and 63x objective . Z-series were obtained starting at the most superficial neuron and ending at the deepest neuron as determined by expression of the pan-neural nuclear marker otIs355[Prab-3::tagRFP::2xNLS] . Images were taken at 0 . 5 µm steps , 200 Hz , and 2x zoom . FIJI ( Schindelin et al . , 2012 ) was used to create max-intensity z-projections of the entire z-stack . Images were pseudocolored and merged using Adobe Photoshop . Quantitation and statistical analysis was carried out as described in the preceding paragraph . Animals were mounted as above but were anesthetized with 20 mM NaN3 . Images were obtained using a Zeiss AxioImager equipped with an Apotome using the 40x objective . Ten to fifteen slices , taken with a spacing of 0 . 5 µm , were merged into a Z projection using Image J . cDNAs were amplified from RNA extracted from mixed-stage him-5 cultures . lep-2 and lin-28 cDNAs were amplified using primers shown in Table S1 . Human MKRN3 cDNA was amplified from the plasmid SC319872 ( Origene ) using primers shown in Table S1 . The resulting cDNAs were cloned into pDONR221 and recombined via Multisite Gateway Cloning ( Invitrogen ) to make following expression clones: Transgenes were injected into young adult hermaphrodites using standard methods ( Evans , 2006 ) to generate fsEx transgenic lines . The Plep-5::GFP reporter was constructed by overlap extension PCR ( Nelson and Fitch , 2011 ) using primers KKOlp-5_1 and plep-5_GFP_B to amplify C . elegans genomic DNA ( for fragment A ) and primers pPD95 . 75 . C and pPD95 . 75D to amplify from plasmid pPD95 . 75 ( for fragment B ) . Final extension used primers KKlep5_expr-9 and pPD95 . 75 . D . Transgenes were injected with standard methods using the co-injection marker pRF4[rol-6 ( d ) ] .	Most animals develop from juveniles , which cannot reproduce , to sexually mature adults . The most obvious signs of this transition are changes in body shape and size . However , changes also take place in the brain that enable the animals to adapt their behavior to the demands of adulthood . For example , fully fed adult male roundworms will leave a food source to search for mates , whereas juvenile males will continue feeding . The transition to sexual maturity needs to be carefully timed . Too early , and the animal risks compromising key stages of development . Too late , and the animal may be less competitive in the quest for reproductive success . Cues in the environment , such as the presence of food and mates , interact with timing mechanisms in the brain to trigger sexual maturity . But how these mechanisms work – in particular where and how an animal keeps track of its developmental stage – is not well understood . In the roundworm species Caenorhabditis elegans , waves of gene activity , known collectively as the heterochronic pathway , determine patterns of cell growth as animals mature . Through further studies of these worms , Lawson et al . now show that these waves also control the time at which neural circuits mature . In addition , the waves of activity occur inside the nervous system itself , rather than in a tissue that sends signals to the nervous system . Moreover , they occur independently inside many different neurons . Each neuron thus has its own molecular clock for keeping track of development . Several of the genes critical for developmental timekeeping in worms are also found in mammals , including two genes that help to control when puberty starts in humans . If one of these genes – called MKRN3 – does not work correctly , it can lead to a condition that causes individuals to go through puberty several years earlier than normal . Studying the mechanisms identified in roundworms may help us to better understand this disorder . More generally , future work that builds on the results presented by Lawson et al . will help to reveal how environmental cues and gene activity interact to control when we become adults .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2019	The Makorin lep-2 and the lncRNA lep-5 regulate lin-28 to schedule sexual maturation of the C. elegans nervous system
The human cytomegalovirus ( HCMV ) US12 family consists of ten sequentially arranged genes ( US12-21 ) with poorly characterized function . We now identify novel natural killer ( NK ) cell evasion functions for four members: US12 , US14 , US18 and US20 . Using a systematic multiplexed proteomics approach to quantify ~1300 cell surface and ~7200 whole cell proteins , we demonstrate that the US12 family selectively targets plasma membrane proteins and plays key roles in regulating NK ligands , adhesion molecules and cytokine receptors . US18 and US20 work in concert to suppress cell surface expression of the critical NKp30 ligand B7-H6 thus inhibiting NK cell activation . The US12 family is therefore identified as a major new hub of immune regulation . At 236 kb the human cytomegalovirus ( HCMV ) genome is the largest of any characterized human virus and is comprised of long and short unique regions ( UL and US ) , each flanked by inverted terminal repeats . HCMV codes for around of 170 canonical protein-coding genes with 39 herpesvirus ‘core’ genes concentrated in the center of the UL region ( Dolan et al . , 2004 ) . The core genes mainly encode structural components of the virion and proteins required for virus DNA replication and have orthologues in the other human herpesviruses . The vast majority of the remaining HCMV genes are not essential for virus replication in vitro ( Dunn et al . , 2003 ) yet are replete with accessory functions , many of which have been implicated in suppressing host immune responses . Unusually , HCMV encodes 15 gene families of variable size that are often clustered on the genome ( Davison et al . , 2002; Holzerlandt et al . , 2002; Chee et al . , 1990; Dolan et al . , 2004; Davison et al . , 2003 ) . Many of these gene families exhibit homology with cellular genes and are conserved to various extents in other primate CMVs . Consequently , these primate CMV gene families are likely to have arisen through gene capture and amplification driven by differential selective pressures in their various primate hosts over millennia ( Davison et al . , 2013 , 2003 ) . The US12 gene family consists of 10 genes , designated US12 to US21 , arranged sequentially in the US region and transcribed in the same orientation ( Chee et al . , 1990; Dolan et al . , 2004 ) . The genetic arrangement of the US12 family is reminiscent of ‘accordion’ gene expansions , which are generated when a cellular or virus resistance function is placed under strong selective pressure ( Filée , 2013 ) . Such an expansion was recently exemplified experimentally using a poxvirus interferon resistance function ( Elde et al . , 2012 ) . The US12 family encodes a series of 7-transmembrane spanning proteins with low-level homology to the cellular transmembrane bax-inhibitor one motif-containing proteins ( TMBIM ) . While not essential for virus replication , the US12 family has been implicated in HCMV tropism , virion maturation and immune evasion ( Das and Pellett , 2007; Cavaletto et al . , 2015; Bronzini et al . , 2012; Hai et al . , 2006; Gurczynski et al . , 2014; Fielding et al . , 2014 ) . Natural Killer ( NK ) cells play a critical role in controlling HCMV infections , and the virus invests a substantial proportion of its coding capacity to inhibit NK cell activation ( Wilkinson et al . , 2013 ) . We previously observed that US18 and US20 suppress cell surface expression of the NK cell-activating ligand MICA ( Fielding et al . , 2014 ) and posited that the synergistic action of US18 and US20 may be the vestige of an immune selective pressure that drove the original expansion of the US12 family . These data show that multiple US12 family members can co-operate to target the same cellular protein . Therefore individual functions , as identified with single gene viral mutants , may not be readily replicated by expressing these same viral genes in isolation , i . e . these viral genes may work more efficiently in the context of HCMV productive infection . To investigate the function of all US12 family genes , we undertook a systematic functional analysis that showed four members were NK immunevasins . Conventional biochemical investigations on US12 family proteins are rendered problematic due to their extreme hydrophobicity . We therefore undertook multiplexed Tandem Mass Tag ( TMT ) -based proteomic analyses to systematically evaluate the capacity of all US12 family genes to modulate the cellular proteome , both individually and in concert . Such an approach has been enabled by our recent development of Plasma Membrane Profiling ( PMP ) to identify novel cell surface targets for the HCMV latency protein UL138 ( Weekes et al . , 2013 ) and individual viral immunevasins UL141 and US2 ( Hsu et al . , 2015 ) . Quantitative Temporal Viromics ( QTV ) allowed >8000 cellular and 153 viral proteins to be tracked throughout the course of productive HCMV infection , thus building a comprehensive picture of cellular control by the virus ( Weekes et al . , 2014 ) . Through comparative analysis of ~1300 cell surface and ~7200 cellular proteins during infection with HCMV US12 family deletion mutants , we now describe in detail how this family has a profound influence not only on NK cell recognition but other key functions that impact on immunity including cellular adhesion and cytokine signalling . To determine whether the US12 family has a broader role in modulating NK cell responses , a series of HCMV US12 family deletion mutants were generated ( 10 single deletion mutants and the US12-21 ‘block’ deletion ) . The HCMV genome was manipulated by DNA recombineering in a strain Merlin BAC that did not express RL13 and UL128 ( Stanton et al . , 2010 ) . Viruses were generated by DNA transfection and the complete genomic sequence of the virus stocks was validated by deep sequencing ( Table 1 ) . In NK functional assays , the △US12-21 block deletion mutant induced substantially higher levels of NK cell degranulation compared to the parent HCMV with all four different donors tested ( Figure 1B ) . Significantly increased levels of NK activation were detected in assays using deletion mutants of 5 different US12 family members: US12 ( 3 of 4 donors ) , US14 ( 1 of 4 donors , with a trend towards increased NK activation in the other three donors ) , US18 ( 3 of 4 donors ) , US20 ( 4 of 4 donors ) , US21 ( 4 of 4 donors ) ( Figure 1B ) , while three US12 family deletion mutants ( US15 , US16 , US19 ) reduced the level of NK cell activation in some donors ( Figure 1B ) . Although members of the US12 family are capable individually of either activating or suppressing NK cell function , the net effect of the complete US12 gene family is clearly to inhibit NK cell recognition . 10 . 7554/eLife . 22206 . 003Figure 1 . Multiple US12 family proteins regulate NK activation . ( A ) Fibroblasts ( HF-TERTs ) were mock infected or infected with the Merlin strain of HCMV or the series of US12 family deletion mutants for 72 hr . Infected cells were incubated with donor PBMC for 5 hr and NK degranulation assessed by % CD107+ cells within the CD3- , CD56+ population by flow cytometry . ( B ) CD107 assay results ( mean and SD ) are shown from four separate donors performed in duplicate or triplicate and analysed by unpaired ordinary one way ANOVA with Dunnett’s test for multiple comparisons against the HCMV control p<0 . 05 , p<0 . 01 , p<0 . 005 p<0 . 001 ) . Infected cells were assessed by the % cells with down-regulated MHC I compared to the mock-infected cells ( for the experiment using donor #23 and # 385 , HCMV 93% , ΔUS12 94% , ΔUS13 98% , ΔUS14 94% , ΔUS15 98% , ΔUS16 99% , ΔUS17 98% , ΔUS18 97% , ΔUS19 99% , ΔUS20 97% , ΔUS21 91% , ΔUS12-21 94%; for the experiment using donor #91 and # 285 , HCMV 94% , ΔUS12 94% , ΔUS13 94% , ΔUS14 83% , ΔUS15 95% , ΔUS16 97% , ΔUS17 85% , ΔUS18 94% , ΔUS19 96% , ΔUS20 94% , ΔUS21 94% , ΔUS12-21 96% ) . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 00310 . 7554/eLife . 22206 . 004Table 1 . HCMV constructs used in the study . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 004VirusBAC #Accession no . Cassette usedModificationPrevious referenceHCMV1111GU179001 . 1NoneRL13- , UL128-Stanton et al . ( 2010 ) △US121810KU221097GalKUS12 CDS deletedNone△US131831KU221099GalKUS13 CDS deletedNone△US141798KU221093GalKUS14 CDS deletedNone△US151800KU221094GalKUS15 CDS deletedNone△US161802KU221095GalKUS16 CDS deletedNone△US171804KU221096GalKUS17 CDS deletedNone△US181654KU221091RpsL-Neo-LacZUS18 CDS deletedFielding et al . ( 2014 ) △US191796KU221092RpsL-Neo-LacZUS19 CDS deletedNone△US201595KU221090SacB-AmpR-LacZUS20 CDS deletedFielding et al . ( 2014 ) △US211871KU221100RpsL-Neo-LacZUS21 CDS deletedNone△US12-211815KU221098RpsL-Neo-LacZUS12-21 deletedNone To provide an unbiased systematic analysis of the entire US12 gene family , we employed 10-plex tandem-mass tags ( TMT ) with MS3/Multinotch mass spectrometry to quantify whole cell ( WCL ) and plasma membrane ( PM ) proteomes in fibroblasts infected with the panel of HCMV US12 family mutants . The proteomic analyses were performed in two parts to permit inclusion of appropriate controls . Samples analyzed in Proteomic Series 1 included mock-infected controls , the parental HCMV strain , the US12-21 block deletion and defined mutants in US18 , US19 and US20 ( Figure 2 , Figure 3 , Figure 4A ) . Mass spectrometry quantified 7215 WCL and 1281 PM proteins . The extremely dynamic modulation of the host cell proteome observed during productive HCMV infection was consistent with previous findings ( Figure 3 ) ( Weekes et al . , 2014 ) . The role of the entire US12 gene family was assessed by comparison of the HCMV US12-21 block deletion mutant with its parental virus . The impact was most noticeably focused on the PM , in that the majority of WCL proteins affected were also PM ‘hits’ ( Figure 2 ) . The heat map illustrating changes in protein abundance correspondingly appears more dynamic i . e . a higher proportion of proteins detected were regulated in response to the deletion of individual US12 family genes for the PM proteome than for the much larger set of proteins quantified in WCL samples ( Figure 3 ) . For a given PM protein modulated by the US12 gene family ( Figure 4 ) , comparable results were observed in WCL samples ( Figure 5 ) , suggesting that the US12 family regulates protein expression or stability of cell surface proteins . 10 . 7554/eLife . 22206 . 005Figure 2 . The US12 family targets numerous plasma membrane proteins . Cells infected with HCMV ( Merlin ) or HCMV △US12-21 mutant were processed to give PM or WCL fractions and analyzed by TMT mass spectrometry . Scatter plot of proteins identified in the PM ( panel A ) or WCL fractions ( panel B ) respectively and quantified by 2 or more unique peptides . Fold change ( △US12-21-infected fibroblasts/HCMV-infected cells ) is shown as the log2 ratio on the x-axis and the signal:noise on the y-axis as log10 . Proteins unaltered by the US21-21 deletion locate at the center of the plots ( 0 log2/1 fold-change ) , whereas proteins to the left or right of center represent proteins down- or up-regulated by the US12-21 deletion respectively . Significance B was used to estimate p values ( Cox et al . , 2009 ) . The 2 different alleles of MICA present in HFs were detected by this analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 00510 . 7554/eLife . 22206 . 006Figure 3 . Hierarchical cluster analysis of all proteins quantified in proteomic series 1 and 2 . Zoomed regions are shown for clusters of interest . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 00610 . 7554/eLife . 22206 . 007Figure 4 . Proteomic analysis of cellular proteins targeted by US12 family members in PM samples . ( A-B ) Workflows of proteomic series 1 and 2 respectively . ( C-E ) Quantitation of PM proteins in enriched KEGG pathways identified using DAVID . Relative abundance of each protein is expressed relative to the sample with the highest abundance ( set to 1 ) . NQ – not quantified . A number of the US12-21 mutant targets were also regulated by the US18 and US20 mutants ( B7-H6 , ULBP2 , IL6ST , KIT , KITLG JAM3 , ACVR1 , ACVRL1 , IFNGR1 , JAM3 , MPZL1 , CXADR , ALCAM , SDC4 , CD99 , SDC1 ) . A subset of these proteins were also regulated by the US14 and/or US16 mutants ( KIT , KITLG , ACVR1 , IFNGR1 , JAM3 , MPZL1 , CD99 , SDC1 ) . We estimated p values for the ratios of each mutant compared to HCMV Merlin using Benjamini-Hochberg corrected Significance B values ( Cox and Mann , 2008 ) : p<0 . 05 , *p<0 . 0001 . For proteomic series 1 , ratios were calculated as US12 family deletion mutant / average HCMV and for proteomic series 2 , US12 family deletion mutant / HCMV . All proteins quantified by 2 or more peptides were included in this calculation . SDC1 was quantified by a single peptide in proteomic series 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 00710 . 7554/eLife . 22206 . 008Figure 5 . Individual US12-family proteins target key natural killer cell ligands , cell adhesion molecules and cytokines and their receptors in WCL samples . ( A–C ) Quantitation of key US12 family targets in WCL samples ( proteomic series 1 and 2 - comparative analysis of these proteins from PM samples is shown in Figure 3 ) . We generally observed similar results for proteins quantified in both PM and WCL samples . Relative abundance of each protein is expressed relative to the sample with the highest abundance ( set to 1 ) . NQ – not quantified . P values were calculated as described in Figure 4: p<0 . 05 , *p<0 . 0001 . MICB was only quantified by 1 peptide in proteomic series 1 , and MICA , JAM3 and SDC4 by one peptide in proteomic series 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 008 NK cells continually monitor the levels of inhibitory and activating ligands on the surface of potential targets , thus one of the roles of the US12 family in re-modeling the PM proteome is compatible with its role in directly impacting NK cell recognition . To gain an overview of pathways targeted , the online bioinformatics resource DAVID ( Database for Annotation , Visualisation and Intergrated Discovery; https://david . ncifcrf . gov/ ) was used to perform a KEGG ( Kyoto Encyclopedia of Genes and Genomes ) pathway enrichment analysis on proteins rescued >3 fold by the deletion of the US12-21 block ( Dennis et al . , 2003; Huang et al . , 2009 ) . Multiple KEGG pathways were significantly enriched ( data not shown ) , including natural killer cell-mediated cytotoxicity ( MICA , MICB , ULBP2 , IFNGR1 ) , cytokine-cytokine receptor interaction ( IL6ST , KIT , KITLG , ACVRL1 , ACVR1 , IFNGR1 ) and cell adhesion molecules ( CAMs; JAM3 , ICOSLG , PTPRM , MPZL1 , CXADR , ALCAM , SDC4 , CD99 , SDC1 ) . The modulation of cell adhesion molecules and chemokines/cytokines may additionally impact NK cell recognition . A further NK ligand , B7-H6 , was not identified by DAVID analysis , but was also regulated by the US12-21 mutant . While MICA is a recognized target of US18 and US20 ( Fielding et al . , 2014 ) , the regulation of the NK cell activating ligands B7H6 , MICB and ULBP2 by the US12 family are novel . A substantial subset of the proteins regulated by the US12-21 block mutant were similarly modulated by US18 and US20 ( Figure 4 , Figure 5 ) , whereas US19 specifically targeted RALGPS2 ( Figure 6 ) . To determine the contribution of each individual US12 family member to the overall effects observed with the ‘block’ deletion mutant , Proteomic Series two compared infection with all the single gene deletion mutants ( except the US19 deletion ) to the parent HCMV ( Figure 4B ) . Mass spectrometry quantified 7156 whole cell and 1312 PM proteins . We re-examined key candidate molecules identified by bioinformatic analysis of the US12-21 block deletion ( Figure 4C–E ) . A striking feature of this data was that the majority of these target proteins were regulated by both US18 and US20 , with a subset additionally regulated to a lesser degree by US14 and/or US16 ( Figure 4C–E ) . Other US12 family members also had important effects , with US12 and US13 mutants regulating the NKG2D ligands ( NKG2DL ) ULBP2 and MICB respectively , US14 mutant regulating PTPRM , and the US15 mutant regulating IL6ST ( Figure 4C–E and Figure 6 ) . Overall , the contribution made by a given family member varied dramatically ranging from the highly focused impact of US13 and US19 ( one cellular target each ) to the exceptionally promiscuous US20 ( 54 cellular targets ) ( Figure 6 ) . Where a given protein was quantified both in PM and WCL samples , generally similar changes were observed in both samples ( Figure 4 and 5 ) . Interestingly , MICA was regulated by the deletion of US12-21 block to a much greater extent than any of the single deletion mutants , implying that multiple US12 family members may need to act in concert to optimize control over certain cellular targets ( Figure 4C ) . 10 . 7554/eLife . 22206 . 009Figure 6 . Co-regulation of multiple cell surface proteins by >1 US12 family gene . Proteins that were rescued by deletion of each US12 family member in the PM fraction of proteomic Series 2 ( i . e . not including US19 ) were analyzed to determine which were additionally targeted by one or more US12 family members . Proteins were plotted in a matrix with unique protein targets on the lower diagonal of the plot ( US12 family member compared with itself ) and common protein targets in intersections with other US12 members . To identify the highest confidence targets of the US12-21 family for the purposes of this analysis and to generate a shortlist of US12-21 family PM ‘hits’ , we employed the following strategy: we included proteins ( a ) exhibiting at least 3-fold rescue upon deletion of a given US12 family member , quantified by at least two peptides and annotated by GO to indicate a PM location . ( b ) validated by a corresponding >2 fold change in at least one of ( i ) deletion of the US12-21 block ( proteomic series 1 or 3 ) , ( ii ) the biological repeat of US18 or US20 in proteomic series 1 , ( iii ) the corresponding gene deletion in WCL proteomic series 2 . This strategy identified all proteins shown in Figure 3 . US19 data was used from proteomic series 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 009 Infection with the US21 deletion mutant resulted in selective impairment of many late phase Tp3 and Tp5 ( Temporal protein profile 3 and 5 ) HCMV proteins ( Weekes et al . , 2014 ) , for example UL32 and UL99/pp28 , and elevated expression of pUS20 , whereas the HCMV US12-21 block mutant did not recapitulate this effect ( Figure 7 , with further proteomic data searchable in Excel spreadsheet Supplemental file 1 ) . Our data are consistent with deletion of the US21 structural gene impacting on the transcriptional control of US20 ( data not shown ) , and may contribute to a potential growth defect observed for this particular mutant . Therefore the activation of NK cells in response to US21 mutant-infected targets cannot be assigned directly to an effect of US21 ( Figure 1 ) . A definitive assessment of the function of US21 will require an alternative approach . 10 . 7554/eLife . 22206 . 010Figure 7 . Altered expression of US20 , UL32 and UL99 levels in the US21 deletion mutant-infected fibroblasts . Quantitation of US20 , UL32 and UL99 in proteomic series 1–3 in both PM and WCL . Abundance of each protein is expressed relative to the sample with the highest abundance ( set to 1 ) . For proteomic series 1 and 2 , p values were calculated as described in Figure 4: p<0 . 05 , p<0 . 0001 . US20 was only quantified by one peptide in proteomic series 1 , WCL experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 010 Of all PM proteins regulated >3 fold by a member of the US12-21 family , 29% were regulated by 2 or more family members and 6% by 3 or more family members ( Figure 6 ) , although using a stringent 3-fold cutoff may underestimate the true incidence of co-regulation . Such co-regulation may suggest evolutionary pressure underpins the expansion of this gene family . The majority of proteins that the US12 family down regulates from the PM are also lost from the WCL , consistent with post-translational proteolysis or an overall reduction in expression . We previously found that US20 targets the NKG2DL MICA for degradation in lysosomes ( Fielding et al . , 2014 ) . The DAVID pathway analysis identified increases in lysosomal proteins within the PM fraction of △US20-infected cells; the list of lysosomal proteins included a number of cathepsins and ATPases ( PSAP , CTSD , ATP6V1E1 , ATP6V1H , NPC2 ) ( data not shown ) . These data suggest a role for US20 in regulating intracellular endo-lysosomal vesicular transport . To determine if this is a more general mechanism for the whole US12-21 family , HFFs were infected with HCMV , the US12-21 block deletion mutant or HCMV in the presence of the lysosomal protease inhibitor leupeptin for 12 hr prior to harvest , and protein expression was analyzed by 10-plex TMT at 24 hr , 48 hr and 72 hr post-infection ( Proteomic Series 3 , Figure 8A ) . 10 . 7554/eLife . 22206 . 011Figure 8 . Multiple US12 gene family targets are degraded via the lysosomal pathway . ( A ) Workflow of Proteomic Series 3 . ( B ) Number of proteins targeted by the US12-21 block and additionally rescued >2 fold in the dataset by the △US12-21 deletion mutant or rescued >2 fold or >1 . 5 fold by leupeptin treatment for both WCL and PM . C . Comparable degree of rescue of US12-21 target proteins by US12-21 deletion and leupeptin treatment . Quantitation of a subset of these proteins is shown relative to the maximum abundance ( set to 1 ) . p-values were calculated as described in Figure 4 , for the ratios of leupeptin treatment or US12-21 deletion virus infection at each time point compared to the matched wild-type Merlin-infected control . p<0 . 05 , *p<0 . 0001 . SDC1 was only quantified by one peptide . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 011 The △US12-21 ‘block’ deletion caused substantial shifts in WCL proteins , with >80 proteins rescued >2 fold at 48 hr , compared to the parent viral infection ( Figure 8B ) . Of these , 26% were also rescued >2 fold and 51% > 1 . 5 fold with leupeptin suggesting that the US12 family targets multiple host proteins to the lysosome . Leupeptin treatment had a more limited effect on rescuing PM proteins back to the the cell surface ( Figure 8B ) , consistent with our previous observation of intracellular accumulation of MICA without protein relocalisation to the cell surface during treatment with leupeptin or other lysosomal inhibitors ( Fielding et al . , 2014 ) . Of the 21 proteins identified in Figure 4 within the key categories ‘natural killer cell-mediated cytotoxicity’ , ‘cytokine-cytokine receptor interaction’ and ‘cell adhesion molecules’ , 11 were rescued >2 fold by leupeptin , with the remainder exhibiting 1 . 3–2-fold rescue ( Figure 8C ) . MICB was clearly rescued by leupeptin treatment ( Figure 8C ) . Interestingly , control of MICB and ULBP2 levels correlated with regulation of UL16 by US13 and US12 ( Figure 9 ) . However , leupeptin rescue of ULBP2 and UL16 was less convincing than for MICB ( Figure 9 ) . Although many cellular substrates of the US12 family are clearly being targeted for lysosomal degradation , there may be additional mechanisms by which this family regulates the PM proteome . 10 . 7554/eLife . 22206 . 012Figure 9 . Regulation of UL16 levels by the US12 family . Quantitation of UL16 in proteomic series 1–3 in both PM and WCL . Relative abundance of each protein is expressed relative to the sample with the highest abundance ( set to 1 ) . For proteomic series 1 and 2 , p values were calculated as described in Figure 4: p<0 . 05 , p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 012 We used flow cytometry to validate a proportion of the various PM protein targets that are regulated by the US12 family ( Figure 10 ) . Next , we examined NKG2DL MICA and MICB ( Figure 10 ) . We found that MICA expression was rescued to a greater degree with the US12-21 block mutant than with the individual US18 or 20 deletion mutants ( Figure 10 ) . This is consistent with our previous analysis showing that a combined US18 and US20 deletion had a greater effect on cell surface MICA than single gene deletion mutants ( Fielding et al . , 2014 ) . The US13 deletion mutant caused a minor , but consistent , elevation in cell surface expression of MICB . Other US12 family genes may yet contribute to regulating MICB , as there was a further increase in the US12-21 block mutant ( Figures 10 and 4C ) . 10 . 7554/eLife . 22206 . 013Figure 10 . Validation of cell surface proteins regulated by the US12 family . Flow cytometry confirmed proteomic data for proteins representative of each category enriched in the DAVID analysis . Staining in mock/US12 family member deletion mutant infections ( blue line ) is shown relative to the parental HCMV infection ( red line ) . Flow cytometry was carried out for cell surface expression of MHC I ( W6/32 ) as a control for HCMC infection and isotype antibody staining controls ( with directly PE-conjugated IgG1 , IgG2a or IgG2b antibodies or for unconjugated antibodies mIgG and an anti-mouse-AF647 conjugated secondary antibody ) . Infected cells were assessed by the % cells with down-regulated MHC I compared to the mock-infected cells ( HCMV 94% , ΔUS12 94% , ΔUS13 94% , ΔUS14 83% , ΔUS15 95% , ΔUS16 97% , ΔUS17 85% , ΔUS18 94% , ΔUS19 96% , ΔUS20 94% , ΔUS21 94% , ΔUS12-21 96% ) . Results are representative of at least two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 013 The fact that , in addition to targeting MICA , US18 and US20 together are also implicated in targeting B7-H6 is extraordinary . B7-H6 is a known ligand for NKp30 and exogenous expression in fibroblasts increased NK degranulation ( Figure 11A–C ) . Ectopic expression of US20 , and to a lesser extent US18 , from adenovirus vectors reduced levels of exogenously expressed B7-H6 detected by immunoblotting ( Figure 11D ) . Both PM and WCL expression of B7-H6 was induced by the HCMV US18 , US20 and US12-21 deletion mutants but not by the parental virus ( Figure 4C , Figure 5A ) , which was confirmed by flow cytometry and immunoblot ( Figures 10 and 12A ) . B7-H6 is therefore induced as a ‘stress ligand’ during productive infection , but its expression is controlled by US18 and US20 , acting in concert , to target it for proteolysis ( Figure 8C ) . Being the dominant ligand for the NK activating receptor NKp30 , B7-H6 is potentially a critical target for HCMV . 10 . 7554/eLife . 22206 . 014Figure 11 . Adenovirus expressed B7-H6 regulates NK cell activation and B7-H6 levels are regulated by ectopically-expressed US18 and US20 . ( A ) HF-CARs were infected with control ( RAd-CTRL ) or B7-H6-expressing ( RAd-B7-H6 ) adenovirus vectors ( MOI 5 ) . Cells were harvested 48 h p . i . and used as targets in a CD107 degranulation assay with buffy-coat derived PBMC from 3 separate donors in duplicate or triplicate . Results ( shown as mean and SD ) were analyzed by unpaired two-tailed Student’s T-test . p<0 . 001 , **p<0 . 0001 . ( B and C ) HF-CARs were infected with control ( RAd-CTRL ) or B7-H6-expressing ( RAd-B7-H6 ) adenovirus vectors ( MOI 5 ) . Cells were harvested 48 h p . i . and B7-H6 cell surface expression analyzed by flow cytometry ( B ) or western blot ( C ) . Results are representative of at least 2 independent experiments . ( D ) HF-CARs were infected with adenovirus control ( CTRL ) , US18-expressing adenovirus , US20-expressing adenovirus or a combination of both US18 and US20 ( MOI 5 each , made up to a total MOI of 10 with RAd-CTRL ) . After 24 hr incubation , HF-CARs were infected with adenovirus control ( CTRL ) or B7-H6-expressing adenovirus ( B7–H6 ) at MOI 5 as indicated before incubation for a further 48 hr . Whole cell lysates were prepared and analysed by immunoblotting with the antibodies indicated . Results are representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 01410 . 7554/eLife . 22206 . 015Figure 12 . Differences in B7-H6 levels on HCMV US12 family deletion mutant-infected cells regulate NKp30-mediated responses . ( A ) Whole cell lysates were prepared from mock , HCMV or the series of US12 family deletion mutant infected fibroblasts and analyzed by immunoblotting with antibodies specific for B7-H6 ( CH31 ) or actin . Results are representative of two independent experiments . ( B ) An NKp30-responsive 2B4 reporter cell line containing a NFAT-GFP reporter construct ( CT299 ) was incubated for 24 hr with mock , HCMV or the series of US12 family deletion mutant infected fibroblasts in triplicate . Fixed cells were then analyzed for GFP fluorescence by flow cytometry compared to cells incubated with no target fibroblasts . Results are expressed as the % GFP positive reporter cells ( mean and SD ) and were analyzed by unpaired two-tailed Student’s T-test . Results are representative of two independent experiments . ( C ) HF-TERTs were transfected with control ( C ) or B7-H6 ( B7 ) siRNAs for 24 hr prior to infection with the parent HCMV , △US18 , △US20 or △US12-21 mutants for 72 hr . Cells were then incubated for 24 hr with CT299 NKp30 reporter cells in triplicate and GFP+ cells determined by flow cytometry , and analysed by unpaired two-tailed Student’s t-test . Results ( mean and SD ) are representative of two independent experiments . ( D ) HF-TERTs infected with HCMV , △US18 , △US20 or △US12-21 mutants were incubated for 24 hr with CT299 NKp30 reporter cells , in the presence isotype control ( C ) or B7-H6 blocking antibodies ( B7 ) in triplicate , GFP+ cells determined by flow cytometry and analyzed by unpaired two-tailed Student’s t-test . Results ( mean and SD ) are representative of two independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 015 We investigated the effect of the US12 family members upon activation of an NKp30-responsive reporter cell line , and found that the presence of US18 and US20 was required to inhibit reporter activity ( Figure 12B ) . Reporter cell fluorescence could be inhibited either by transfection of B7-H6 specific siRNA or by use of a B7-H6 specific blocking antibody ( CH31 ) ( Figure 12C and D respectively ) . The presence of both US18 and US20 was required to prevent B7-H6 activating the NKp30 reporter function . We sought to differentiate the functional impact exerted by US18 and US20 on B7-H6 and MICA . B7-H6 knockdown in cells infected with HCMV US18 and US20 deletion mutants was correspondingly able to reduce donor NK cell activation ( Figure 13A and B ) . The impact of B7-H6 knockdown on the HCMV US12-21 block mutant was more variable and consistent with the US12 gene family encoding additional NK modulating functions ( Figure 13B , Figure 1 ) . A B7-H6 blocking antibody had an even more pronounced effect , reducing NK activation in response to the US18 , US20 and US12-21 deletion mutants to levels of the parental HCMV control ( Figure 13C ) . These results are consistent with B7-H6 playing a major role in NK cell recognition of HCMV-infected cells , a function that is countered by the US12 family members US18 and US20 . 10 . 7554/eLife . 22206 . 016Figure 13 . B7-H6 regulation has a major effect on NK activation in response to HCMV infection . ( A and B ) HF-TERTs were transfected with control ( C ) or B7-H6 ( B7 ) siRNAs for 24 hr prior to infection with the parent HCMV , △US18 , △US20 or △US12-21 mutants . Cells were harvested 72 h p . i . and B7-H6 cell surface expression analyzed by flow cytometry ( E ) or used as targets in a CD107 degranulation assay with donor-derived PBMC in triplicate ( F ) . Results ( mean and SD ) were analyzed by unpaired two-tailed Student’s t-test and are representative of two independent experiments using 2 separate donors in each . Infected cells were assessed by the % cells with down-regulated MHC I compared to the mock-infected cells ( HCMV CTRL siRNA 95% , ΔUS18 CTRL siRNA 91% , ΔUS20 CTRL siRNA 93% , ΔUS12-21 CTRL siRNA 86% , HCMV B7-H6 siRNA 96% , ΔUS18 B7-H6 siRNA 84% , ΔUS20 B7-H6 siRNA 96% , ΔUS12-21 B7-H6 siRNA 79% ) . ( C ) HF-TERTs infected with HCMV , △US18 , △US20 or △US12-21 mutants were used as targets in a CD107 degranulation assay with donor-derived PBMC , in the presence isotype control ( C ) or B7-H6 ( B7 ) blocking antibodies in triplicate . Results ( mean and SD ) were analyzed by unpaired two-tailed Student’s t-test and are shown for two separate donors . p<0 . 05 , p<0 . 01 , p<0 . 001 , ***p<0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 016 The HCMV genome contains 15 gene families of various sizes that have been acquired during evolution to promote virus persistence . Clusters of US12-related genes can be detected in cytomegaloviruses of New World primates , i . e . Green Monkey and Owl Monkey CMVs , thus the capture and expansion of an ancestral precursor presumably took place over 41 million years ago ( Davison et al . , 2013 ) . Within Cynomolgus and Rhesus macaques ( Old World primates ) , chimpanzee and human CMVs , the US12 family is maintained as a well conserved contiguous tandem array of 10–11 genes located in the same relative position on each genome . Amongst circulating HCMV clinical strains , the US12 family exhibits relatively high levels of sequence conservation and genetic integrity ( Sijmons et al . , 2015 ) . The complexity of the HCMV genome , its restricted host range in vitro and protracted replication cycle has frustrated studies into HCMV gene function . However , the advent of high-resolution multiplexed proteomics is revolutionizing our understanding of how HCMV orchestrates host cell gene expression and evades host defenses ( Weekes et al . , 2014; Hsu et al . , 2015 ) . By systematically analyzing a bespoke panel of HCMV deletion mutants , we have discovered that the US12 family selectively targets a broad range of plasma membrane proteins that include , not only NK cell activating ligands , but also T-cell co-stimulatory molecules , cell adhesion molecules , and cytokine/cytokine receptors . The diversity of proteins targeted by many of the US12 family implies a broad strategy of redirecting cell surface receptors . A sizeable proportion of targeted plasma membrane proteins do not accumulate internally , but are degraded in lysosomes , as indicated by rescue with the inhibitor leupeptin . The overlap that exists between US12 gene family targets with those of the KSHV K5 viral E3 ubiquitin ligase ( MICA , MICB , B7 family , ALCAM , IFNGR1 , PTPRM , EPHA2 , CD99 , MPZL1/2 ) could represent convergent evolution and/or targeting of similar cellular pathways ( Timms et al . , 2013 ) . The US12 family has a major impact on NK cell recognition with four US12 family members US12 , US14 , US18 and US20 consistently contributing toward significant levels of NK cell protection . Although the HCMV US21 deletion mutant also scored in functional assays , this could be due to enhanced expression of pUS20 associated with this construct . The NKG2DL MICB and ULBP2 are strongly upregulated during HCMV infection , but fail to reach the cell surface as they are retained in the ER by the NK evasion function gpUL16 ( Rolle et al . , 2003 ) . Our proteomic data reveal that deletion of US12 , US13 and US20 leads to increased cell surface and intracellular expression of MICB , ULBP2 and UL16 itself ( Figure 9 ) . These observations are consistent with US12 family members acting in concert to direct UL16 , MICB and ULBP2 towards proteolytic degradation , with leupeptin treatment clearly able to rescue MICB , and this may contribute to the NK evasion properties of US12 ( Figure 1 ) . They also imply that some of the effects of the US12 family could be the result of co-operation with other HCMV genes . There is some precedent for this situation , although not previously on such a wide scale , as co-operation between US2 and UL141 was observed in the proteasomal degradation of some protein targets e . g . CD112 ( Hsu et al . , 2015 ) . None of the US14 targets are recognized NK cell ligands , suggesting that this gene product utilizes a novel NK evasion mechanism . B7-H6 is the major ligand identified for the natural cytotoxicity receptor ( NCR ) NKp30 and a tumor antigen ( Brandt et al . , 2009 ) . Previously , B7-H6 expression was shown to be induced by inflammatory stimuli , for example pro-inflammatory cytokines and bacterial-derived toll-like receptor ( TLR ) ligands ( Matta et al . , 2013 ) . We show here that B7-H6 expression is induced as a stress protein by HCMV infection ( and potentially other virus infections ) and that US18 and US20 act together to suppress cell surface expression of B7-H6 in the context of HCMV , thereby inhibiting NK cell activation . US18 and US20 were also able to regulate exogenously expressed B7-H6 when expressed individually . Therefore , they appear to target B7-H6 directly and not cellular pathways leading to the expression of B7-H6 . The control of B7-H6 is likely to be of major significance as NKp30 is expressed on γδ T-cells ( Vδ2- ) that are induced during HCMV infection post-transplantation and correlate with control of disease ( Merville et al . , 2000; Lafarge et al . , 2001; Correia et al . , 2011 ) . The US12 family clearly impacts on a broad range of cellular functions including adhesion molecules and cytokine receptors . Many of these adhesion molecules play roles in co-stimulation or immune synapse formation of T-cells e . g . ALCAM , ICOSLG , CXADR ( Hassan et al . , 2004; Wang et al . , 2000; Witherden et al . , 2010 ) . The US12 family targets pro-inflammatory mediators , such as multiple members of the tumour necrosis factor receptor ( TNFR ) superfamily ( TNFRSF8 , TNFRSF12A , NGFR , LTBR ) , gp130 ( IL6ST ) , the IL-6 receptor signaling receptor , pannexin-1 ( PANX1 ) and the TLR4 ligand high-mobility group protein B1 ( HMGB1 ) ( Croft et al . , 2013; Kanneganti et al . , 2007; Park et al . , 2004 ) . IL6ST signaling may also have antiviral effects in the context of HCMV ( Harwardt et al . , 2016 ) , which was suggested by the finding that disruption of gp130 STAT3 binding resulted in IFN-like signaling through STAT1 ( Costa-Pereira et al . , 2002 ) . The closest cellular homologues to the US12 family are the TMBIM family with recognised functions in controlling apoptosis , ER stress , ROS production , actin production , glucose metabolism and protein trafficking ( Rojas-Rivera and Hetz , 2015 ) . The US12 family may have arisen through an ‘accordion’ gene expansion from a ‘captured’ ancestor TMBIM gene . Out of the entire US12 family , US21 displays the highest level of homology with TMBIM1/4 , with the homologous region limited to the transmembrane domain and loops ( unpublished observations , [Lesniewski et al . , 2006; Holzerlandt et al . , 2002] ) . The divergence of the US12 and TMBIM families is also reflected in their different membrane topologies , as TMBIM family have cytosolic N- and C-termini with six full transmembrane spanning regions , and US20 has a cytosolic N-terminus and lumenal C-terminus ( Carrara et al . , 2012; Cavaletto et al . , 2015 ) . The functional relevance of the US12 family/TMBIM homology remains unclear , although it may represent a ‘functional scaffold’ ( Lesniewski et al . , 2006 ) , as a number of TMBIMs also target proteins for lysosomal degradation ( Lee et al . , 2012; Yamaji et al . , 2010 ) . Our data indicate that the US12 family targets multiple host immune ligands , consistent with the family having arisen , been selected and diverged in function as a consequence of immune selection . US12 family genes differ from the majority of characterised immune evasion viral genes , as they act together and do not exhibit single gene effects . The targeting of multiple proteins by a HCMV immunevasin or co-operation between HCMV gene products is not unprecedented , but was not previously observed on this scale or with this diversity of target proteins ( Tomasec et al . , 2005; Prod'homme et al . , 2010; Smith et al . , 2013; Hsu et al . , 2015 ) . The majority of US12 family targets contain an immunoglobulin-like domain , including the MHC I-related proteins , MICA and B7-H6 . Our data highlight the importance that HCMV gene families are likely to play in terms of HCMV persistence in vivo and identifies the US12 family as a critical region for regulation of the host immune response . Healthy adult volunteers provided blood for this study following written informed consent ( approved by the Cardiff University School of Medicine Ethics Committee Ref . no: 10/20 ) or buffy coats provided by the Welsh Blood Service , following informed consent . Human foreskin fibroblasts ( HFFs ) , HFFs immortalized by human telomerase ( HF-TERT ) , HF-TERTs transfected with the Coxsackie-adenovirus receptor ( HF-CAR ) , were maintained at 37°C in 5% CO2 in growth medium ( Dulbecco’s minimal essential medium ( DMEM ) supplemented with penicillin/streptomycin and 10% fetal calf serum ( Invitrogen , Paisley , UK ) ( McSharry et al . , 2001 ) . HCMV deletion mutants were generated by recombineering of the bacterial artificial chromosome ( BAC ) of HCMV strain Merlin ( GenBank accession number GU179001 . 1 ) , as described previously ( Stanton et al . , 2010 ) . Strain Merlin contains the complete genetic complement of HCMV , and is frame shifted in two genes ( RL13– , UL128– ) . Alterations to the BAC were monitored by local PCR and Sanger sequencing , and the entire genomes of the viruses were confirmed by Illumina sequencing as described previously ( Murrell et al . , 2016 ) . Details of the viruses are provided in Table 1 and a list of primers used in their construction and local sequencing of the BACs are provided in Tables 2 and 3 respectively . Primers used in the construction of the HCMV △US18 and △US20 deletion mutants were detailed previously ( Fielding et al . , 2014 ) . 10 . 7554/eLife . 22206 . 017Table 2 . Primers used in the construction of the HCMV US12 deletion mutants library . DOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 017PrimerSequence ( 5’ > 3’ ) US12 GalK ForGCGGGGGACAAAGGACAGTACGACAGATTAGGTGATAGAAACGTTTTTTTCCTGTTGACAATTAATCATCGGCAUS12 GalK RevAAACTTGCCGGGTACCTGAAGCCCCGACGACTGTTCGTCGAGCACCCG TCTCAGCACTGTCCTGCTCCTTUS12 DeleteGCGGGGGACAAAGGACAGTACGACAGATTAGGTGATAGAAACGTTTTTTTGACGGGTGCTCGACGAACAGTCGTCGGGGCTTCAGGTACCCGGCAAGTTTUS13 GalK ForCTTCAGGTACCCGGCAAGTTTTATAGAGAAAGGGGGACGATGGGTGGTGGCCTGTTGACAATTAATCATCGGCAUS13 GalK RevGAAGACTCCACCGAGACGCTCACCCGTTCACTCGGGCGCATCACCCGC CTTCAGCACTGTCCTGCTCCTTUS13 DeleteCTTCAGGTACCCGGCAAGTTTTATAGAGAAAGGGGGACGATGGGTGGTGGAGGCGGGTGATGCGCCCGAGTGAACGGGTGAGCGTCTCGGTGGAGTCTTCUS14 GalK ForGAGTGAACGGGTGAGCGTCTCGGTGGAGTCTTCTTATAAACCAGCGGG TCCCTGTTGACAATTAATCATCGGCAUS14 GalK RevCTGTAGCTTCGAGACCTTGCGGATACGCCGCCGGGCGCTGCGGTCCCG ACTCAGCACTGTCCTGCTCCTTUS14 DeleteGAGTGAACGGGTGAGCGTCTCGGTGGAGTCTTCTTATAAACCAGCGGGTCGTCGGGACCGCAGCGCCCGGCGGCGTATCCGCAAGGTCTCGAAGCTACAGUS15 GalK ForCTCCATGTCGGGACCGCAGCGCCCGGCGGCGTATCCGCAAGGTCTCGAAGCCTGTTGACAATTAATCATCGGCAUS15 GalK RevCGGAACTGGTTTTCGGACAGAGCAGCCGTTTCCAGAGAACGCAGCGCA CCTCAGCACTGTCCTGCTCCTTUS15 DeleteCTCCATGTCGGGACCGCAGCGCCCGGCGGCGTATCCGCAAGGTCTCGAAGGGTGCGCTGCGTTCTCTGGAAACGGCTGCTCTGTCCGAAAACCAGTTCCGUS16 GalK ForCGTTCTCTGGAAACGGCTGCTCTGTCCGAAAACCAGTTCCGAACGAAAATCCTGTTGACAATTAATCATCGGCAUS16 GalK RevCCCCACGGATCTCGCGTCTTAGACGCGCGGTCATATAGCCTCCGGCTG TCTCAGCACTGTCCTGCTCCTTUS16 DeleteCGTTCTCTGGAAACGGCTGCTCTGTCCGAAAACCAGTTCCGAACGAAAATGACAGCCGGAGGCTATATGACCGCGCGTCTAAGACGCGAGATCCGTGGGGUS17 GalK ForTTGGTGGAGACGGCCGGCGCGGCGGGTGGGGGAAACGACGAGTTTTTCCGCCTGTTGACAATTAATCATCGGCAUS17 GalK RevACACTCTATAAACGGTTTCTCATACGCGCCTTTTGATAGCCACCGCCG TCTCAGCACTGTCCTGCTCCTTUS17 DeleteTTGGTGGAGACGGCCGGCGCGGCGGGTGGGGGAAACGACGAGTTTTTCCGGACGGCGGTGGCTATCAAAAGGCGCGTATGAGAAACCGTTTATAGAGTGTUS19 SacB ForCAGCACCCGGTTACCGCGGATTTGATTGACGTCACGAGTGTGGTCAAACCGTGGCGGCACCCTGTATCCGACCCGTCGCCTGTGACGGAAGATCACTTCGUS19 SacB RevGCTACGCCTCTATGTCGAAAATGTGGCTTTATTCATCGGCATGTACCATCTTCTGAGGCTCTGGTTGTGGAGCCCATGACTGAGGTTCTTATGGCTCTTGUS19 DeleteACGTCACGAGTGTGGTCAAACCGTGGCGGCACCCTGTATCCGACCCGTCGGGCGACAAGCGCGGCTGCTGTGAAAACGGGCGCGGTTTTATAGGCATTAGUS21 SacB ForTGCGGCGCACCTACCCTTCTCTTATACACAAGCGAGCGAGTGGGGCACGGTGACGTGGTCACGCCGCGGACACGTCGACCTGTGACGGAAGATCACTTCGUS21 SacB RevCAGCGCCCACACTGCTCAGACGACGGTCGCTGCGACGGTCGCTGCCACAGCAGCGGCGTCGCCCCAGTTCGTCTCCTAACTGAGGTTCTTATGGCTCTTGUS21 DeleteCAAGCGAGCGAGTGGGGCACGGTGACGTGGTCACGCCGCGGACACGTCGAGGCGGCAACGCCGGCGGTTATCGCCGAGATTCGTCTAAATACACGAAGCGUS12-21 DeleteGCGGGGGACAAAGGACAGTACGACAGATTAGGTGATAGAAACGTTTTTTTGGCGGCAACGCCGGCGGTTATCGCCGAGATTCGTCTAAATACACGAAGCG10 . 7554/eLife . 22206 . 018Table 3 . Primers used in the local sequencing of the HCMV US12 deletion mutant libraryDOI: http://dx . doi . org/10 . 7554/eLife . 22206 . 018PrimerSequence ( 5’ > 3’ ) US12 Seq ForCCCTGTCTAGACTCAAAAGCTGUS12 Seq RevATCGTCCCCCTTTCTCTATAUS13 Seq ForGCCGAGTGGCTCGCCUS13 Seq RevCTGGGCACCTATCATCATTAUS14 Seq ForGGAGGGAAGCCCATTGCUS14 Seq RevTCATTACCTGTCTAGCCGUS15 Seq ForCGGACGCGGCTTCCUS15 Seq RevGTCGCTACAGCTCTTTATTAUS16 Seq ForGGGGCACGTAGATGACCGUS16 Seq RevCTCATTAGACAAACTCATCGUS17 Seq ForGTCTAAGACGCGAGATCCGUS17 Seq RevCCCAGTAGACAGACAGAACAUS19 Seq ForGGAGCGGCACGATGGTGACCUS19 Seq RevTCTGCCCACCTAACCAATGCUS21 Seq ForGCTGAAAGATGAAGATGGCGUS21 Seq RevACCCGACCAGATGGGAGACG The replication-deficient adenovirus vector ( RAd-CTRL , pAL1253 ) and adenovirus vectors expressing US18 ( RAd-US18 ) and US20 ( RAd-US20 ) from the Merlin strain of HCMV have been described previously ( Fielding et al . , 2014 ) . A B7-H6 ( NCR3LG1 ) expressing adenovirus was generated by synthesising the B7-H6 CDS ( corresponding to Accession number NM_001202439 . 2 , bases 209–1573 retaining the stop codon ) with 5’ and 3’ arms of homology to the AdZ BAC ( corresponding to the For and Rev primers used in the AdZ recombineering protocol ) ( Stanton et al . , 2008 ) and flanking EcoRI sites ( gene synthesis from Genscript , Piscataway , NJ ) . The cassette containing the B7-H6 CDS and both arms of homology was released from the pUC57 vector by EcoRI digest , gel purified and inserted into pAL1141 by recombineering to produce pAL1593 . BAC DNA of pAL1593 purified by maxiprep ( Macherey-Nagel , Dueren , Germany ) was transfected into 293 TREX to generate the adenovirus , as previously described ( Stanton et al . , 2008 ) . For proteomic analysis , 24 hr prior to each infection 1 . 5 × 107 HFFs were plated in a 150 cm2 flask . Cells were sequentially infected at multiplicity of infection 10 with HCMV strain Merlin >90% of cells were routinely infected using this approach as assessed by MHC I down-regulation . Infections were staggered such that all flasks were harvested simultaneously . For NK cell degranulation , flow cytometry and western blot analyses , cells were seeded in growth medium at appropriate cell densities ( 1 × 106 cells for a 25 cm2 flask ) . The following day , the cells were infected with virus at the required multiplicity of infection in an appropriate volume of growth medium ( 2 ml for a 25 cm2 flask ) for 2 hr on a rocker at 37°C in 5% CO2 . The innoculum was then replaced with fresh growth medium ( 7 ml for a 25 cm2 flask ) , and the cells were incubated for the required times . Fetal calf serum was omitted from the growth medium for mock and HCMV infections . For inhibitor studies , cells were treated 12 hr prior to harvesting with lysosomal inhibitors ( leupeptin 200 µM , Merck Millipore , Watford , UK; Cat . no . 108975 ) in DMEM . For siRNA experiments , cells were seeded in a 25 cm2 flask at 8 × 105 cells/flask 24 hr prior to transfection and then transfected in Optimem medium ( Invitrogen ) with 120 pmol B7-H6 ( SI04761351 , Hs_DKFZp686O24166_5 , Qiagen , Manchester , UK ) or control siRNA ( AllStar Negative Control siRNA , 1027281 , Qiagen ) using Lipofectamine RNAiMax ( Invitrogen ) for a further 24 hr before infection with HCMV ( MOI 20 ) in serum free DMEM for a further 72 hr . Adenovirus infections were carried out in HF-CAR ( MOI 5 ) . Cells were seeded in growth medium at appropriate cell densities ( 1 × 106 cells for a 25 cm2 flask ) . The following day , the cells were infected with virus at the required multiplicity of infection in an appropriate volume of growth medium ( 2 ml for a 25 cm2 flask ) for 2 hr on a rocker at 37°C in 5% CO2 . The inoculum was then replaced with fresh growth medium ( 7 ml for a 25 cm2 flask ) , and the cells were incubated for the required times . Preparation of PM and WCL protein and peptide samples was performed as described previously ( Weekes et al . , 2014 ) . For PM analysis , 100% of each tryptic peptide sample was labeled with TMT reagent , and 6 fractions generated from combined peptide samples by tip-based strong cation exchange . For WCL analysis , cells were lysed in 6 M Guanidine/50 mM HEPES pH8 . 5 then processed as described ( Weekes et al . , 2014 ) . Proteins were digested with LysC then Trypsin . Peptides were labeled with TMT reagent , and 12 fractions generated by high pH reversed phase HPLC . Mass spectrometry and data analysis were performed as described previously ( Weekes et al . , 2014 ) . Briefly , we performed mass spectrometry using an Orbitrap Fusion , and quantified TMT reporter ions from the MS3 scan ( McAlister et al . , 2012; Ting et al . , 2011 ) . Peptides were identified and quantified using a Sequest-based in-house software pipeline . A combined database was searched , consisting of: ( a ) human Uniprot , ( b ) Merlin strain HCMV Uniprot and ( c ) all additional novel HCMV ORFs ( Stern-Ginossar et al . , 2012 ) . Peptides spectral matches ( PSM ) were filtered to a 1% peptide false discovery rate ( FDR ) using linear discriminant analysis ( Huttlin et al . , 2010 ) . The resulting dataset was further collapsed to a final protein-level FDR of 1% . Protein assembly was guided by principles of parsimony . Where all PSM from a given HCMV protein could be explained either by a canonical gene or novel ORF , the canonical gene was picked in preference . Proteins were quantified by summing TMT reporter ion counts across all matching PSM after filtering based on isolation specificity ( Pease et al . , 2013 ) . Reverse and contaminant proteins were removed , and protein quantitation values were exported for normalization and further analysis in Excel . Where all PSMs from a given HCMV protein could be explained either by a canonical gene or novel ORF , the canonical gene was picked in preference . For five viral proteins that had related novel ORFs ( N-terminal extensions of the viral protein ) , some peptides could either have originated either from the canonical protein or the novel ORF . In these cases , each of the novel ORFs were quantified based only on unique peptides that could only have originated from that ORF . Peptides that could either have originated from the canonical protein or the novel ORF were assigned to the canonical protein . We estimated p values for the ratios of each mutant compared to HCMV Merlin or leupeptin-treated cells infected with HCMV Merlin to HCMV Merlin-infected cells , using Benjamini-Hochberg corrected Significance B values ( Cox and Mann , 2008 ) . Hierarchical clustering was performed using centroid linkage with Pearson correlation . The Database for Annotation , Visualization and Integrated Discovery ( DAVID , RRID:SCR_001881 ) was used to determine protein family enrichment amongst KEGG pathways ( Huang et al . , 2009 ) . A given cluster was always searched against a background of all proteins quantified within the relevant experiment . The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium ( Vizcaíno et al . , 2014 ) via the PRIDE partner repository ( Vizcaíno et al . , 2016 ) with the dataset identifier PXD005883 . Flow cytometry was performed as described previously ( Fielding et al . , 2014 ) , except HF-TERTs were harvested using HyQtase ( Thermo Fisher Scientific , Paisley , UK ) for 3 min at 37°C , instead of Trypsin/EDTA . The following phycoerytherin ( PE ) -conjugated antibodies were used at the indicated dilutions ( 200 µl per stain ) : anti-CD166/ALCAM ( Biolegend , London , UK , Clone 3A6 , Cat . no . 343904 , 1:80 , RRID:AB_2289302 ) , anti-EPHA2 ( Biolegend , Clone SHM16 , Cat . no . 356804 , 1:80 , RRID:AB_2561807 ) , anti-CD323/JAM3 ( Biolegend , Clone SHM33 , Cat . no . 356704 , 1:80 , RRID:AB_2561802 ) , anti-CD130/IL6ST ( Biolegend , Clone 2E1B02 , Cat . no . 362003 , 1:80 , RRID:AB_2563401 ) , IgG1 isotype ( Biolegend , Clone MOPC-21 , Cat . no . 400112 , 1:80 , RRID:AB_326434 ) , mouse IgG2a isotype ( Biolegend , Clone MOPC-173 , Cat . no . 400212 , 1:160 , RRID:AB_326460 ) and mouse IgG2b isotype ( Biolegend , Clone MPC-11 , Cat . no . 400314 , 1:80 , RRID:AB_326492 ) . The following unconjugated antibodies were used at the indicated dilutions ( 200 µl per stain ) : anti-CXADR/CAR ( Merck Millipore , Cat no . 05–644 , Clone RmcB , 1:500 , RRID:AB_309871 ) , anti-CD266/TWEAK R/TNFRSF12A ( Biolegend , Clone ITEM-4 , Cat . no . 314102 , 1:200 , RRID:AB_2240752 ) , anti-MICA ( BAMOMAB , Graefelfing , Germany , Cat . no . AM01 , 1:400 , RRID:AB_2636811 ) , anti-MICB ( BAMOMAB , Cat . no . BM02 , 1:400 , RRID:AB_2636812 ) , anti-B7-H6 ( Biotechne R and D Systems , Abingdon , UK , Clone 875001 , Cat . no . MAB7144 , 500 µg/ml , 1:50 , RRID:AB_2636810 ) , anti-MHC I ( BioRAd/AbD Serotec , Kidlington , UK , Clone W6/32 , MCA81EL , 1:1000 , RRID:AB_324063 ) and mouse IgG ( Santa Cruz Biotechnology , Heidelberg , Germany , Cat . no . Sc-2025 , 400 µg/ml , 1:40 , RRID:AB_737182 or Sigma Aldritch , Gillingham , UK , Cat . no . I-5381 1 mg/ml , 1:100 , RRID:AB_1163670 ) , followed by an Alexa Fluor 647 goat anti-mouse IgG secondary antibody ( Thermo Fisher Scientific , Cat . no . A21237 , 1:500 , RRID:AB_2535806 ) . Whole cell lysates were prepared in 1x Nupage gel sample buffer ( Thermo Fisher Scientific ) plus 10 mM DTT and samples were denatured at 95°C or 50°C ( US18 and US20 adenovirus experiment ) for 10 mins . SDS-PAGE and subsequent immunoblotting was carried out as previously described either using X-ray film or a G:Box Chemix-xx6 GeneSys system ( Syngene , Cambridge , UK ) to visualise the blots ( Fielding et al . , 2014 ) . Membranes were probed with antibodies directed against B7-H6 ( non-commercial CH31 monoclonal , purified at 2 µg/ml final or hybridoma supernatant 1:5 ) , anti-V5 antibody ( BioRad/AbD Serotec , Cat . no . MCA1360 , 1:2000 , RRID:AB_322378 ) and actin ( Sigma Aldritch , Cat . no A2066 , 1:5000 , RRID:AB_476693 ) , followed by HRP-conjugated goat anti-mouse or anti-rabbit antibodies ( BioRad/Ab Serotec , Cat . no . 170–6516 , RRID:AB_11125547 and 170–6515 , RRID:AB_11125142 respectively , both 1:5000 ) . NK degranulation assays were performed as described previously using anti-CD107a-FITC ( Cat . no . 555800 , BD Biosciences , Oxford , UK , RRID:AB_396134 ) or isotype control-FITC ( Cat no . 555748 , BD Biosciences , RRID:AB_396090 ) and anti-CD3-PE-Cy7 ( Cat . no . 737657 , Beckman Coulter , High Wycombe , UK RRID:AB_2636813 ) and anti-CD56-PE ( Cat . no . A07788 , Beckman Coulter , RRID:AB_2636814 ) antibodies and PBMC derived from buffy coats or donor blood , except that infected HF-TERTs were harvested using HyQtase for 3 min at 37°C , instead of Trypsin/EDTA ( Prod'homme et al . , 2007 , 2010; Fielding et al . , 2014 ) . Blocking experiments were performed as previously described except using B7-H6 blocking antibody ( CH31 ) or isotype IgG1 control antibody at 10 µg/ml ( Biolegend , Cat . no . 401404 , Clone MG1-45 , RRID:AB_345426 or Biolegend , Cat no . 401402 Clone MG1-45 , RRID:AB_345424 ) ( Fielding et al . , 2014 ) . HF-TERTs were seeded into 96-well plates and either mock-infected or infected with HCMV or US12 family deletion mutants ( MOI 10 ) for 72 hr . The CT299 reporter line ( 2B4 cells stably transfected with an NFAT-GFP reporter and NKp30 ) was washed in complete RPMI and 50 , 000 reporter cells added per well ( 5:1 ratio of reporter to target cells ) . In some experiments , the B7-H6 blocking antibody ( CH31 ) or isotype IgG1 control antibody ( Biolegend , Cat . no . 401404 , Clone MG1-45 , RRID:AB_345426 or Biolegend Cat no . 401402 Clone MG1-45 , RRID:AB_345424 ) at 10 µg/ml were also added to the wells . After 24 hr incubation , reporters were harvested , washed in FACS buffer and fixed with 2% paraformaldehyde before analysis by flow cytometry .	Cytomegalovirus ( CMV ) is one of eight herpesviruses that can infect humans . Most people will at some point become infected with CMV , yet the virus tends only to cause serious disease in people whose immune system is not working properly . Individuals living with HIV/AIDS and organ transplant recipients ( who have to take drugs that suppress their immune system to prevent the organ being rejected ) are particularly vulnerable to CMV infections . Critically , the virus can cross the placenta to infect of the foetus . CMV infection in the womb can cause miscarriage , lead to severe developmental problems in babies and is a major cause of deafness . Herpesvirus infections are for life . While the immune system cannot eliminate CMV , it does have many systems that combine to sense and control infections . Natural killer cells are known to play a critical role in detecting and destroying cells infected with CMV . The virus , in turn , has nine genes that help to protect it against natural killer cells . This includes two genes that belong to a group of similar genes called the US12 family , but it is not clear whether other members of this gene family also provide protection against natural killer cells . Fielding et al . now show that at least four members of the US12 gene family help CMV to evade natural killer cells . For example , two members work together to target a human protein called B7-H6 that acts a sensor to alert natural killer cells if a particular cell is infected . However , the impact of the US12 family goes even wider . The whole family works together to control proteins that are found on the surface of human cells , and many of these proteins appear to be involved in regulating the immune response . The findings of Fielding et al . provide an insight into how the US12 gene family works , and how CMV has evolved to escape the human immune system . New therapies to control CMV infections are urgently needed so the next challenge is to design new antiviral agents that will target CMV’s defence systems .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "immunology", "and", "inflammation" ]	2017	Control of immune ligands by members of a cytomegalovirus gene expansion suppresses natural killer cell activation
A hallmark of Spemann organizer function is its expression of Wnt antagonists that regulate axial embryonic patterning . Here we identify the tumor suppressor Protein tyrosine phosphatase receptor-type kappa ( PTPRK ) , as a Wnt inhibitor in human cancer cells and in the Spemann organizer of Xenopus embryos . We show that PTPRK acts via the transmembrane E3 ubiquitin ligase ZNRF3 , a negative regulator of Wnt signaling promoting Wnt receptor degradation , which is also expressed in the organizer . Deficiency of Xenopus Ptprk increases Wnt signaling , leading to reduced expression of Spemann organizer effector genes and inducing head and axial defects . We identify a '4Y' endocytic signal in ZNRF3 , which PTPRK maintains unphosphorylated to promote Wnt receptor depletion . Our discovery of PTPRK as a negative regulator of Wnt receptor turnover provides a rationale for its tumor suppressive function and reveals that in PTPRK-RSPO3 recurrent cancer fusions both fusion partners , in fact , encode ZNRF3 regulators . The Spemann organizer is an evolutionary conserved signaling center in early vertebrate embryos , which coordinates pattern formation along the anterior–posterior , dorsal–ventral , and left–right body axes ( Harland and Gerhart , 1997; De Robertis et al . , 2000; Niehrs , 2004 ) . In amphibian embryos , the organizer corresponds to the upper dorsal blastopore lip , constituting mostly dorsal mesendoderm . Molecularly , the Spemann organizer functions by negative regulation of BMP , Nodal , and Wnt signaling . Wnt/β-catenin signaling plays a key role in antero-posterior ( a-p ) patterning the Xenopus neural plate where a signaling gradient promotes posterior fate ( Hoppler et al . , 1996; Hoppler and Moon , 1998; Kiecker and Niehrs , 2001 ) , a role , which is evolutionary conserved ( Niehrs , 2010 ) . Various Wnt antagonists or membrane-bound Wnt inhibitors are expressed in neural-inducing dorsal mesoderm and/or the prospective neuroectoderm itself to promote organizer function , and to pattern the neural plate , including cerberus , frzb , dkk1 , shisa , tiki , notum , angptl4 , and bighead ( Bouwmeester et al . , 1996; Leyns et al . , 1997; Glinka et al . , 1998; Yamamoto et al . , 2005; Zhang et al . , 2012; Cruciat and Niehrs , 2013; Zhang et al . , 2015; Kirsch et al . , 2017; Ding et al . , 2018 ) . Thus , the Xenopus Spemann organizer has been a treasure trove for the discovery of negative Wnt regulators , informing on their function in cell and tissue homeostasis as well as in disease ( Cruciat and Niehrs , 2013 ) . With regard to the latter , activation of Wnt/β-catenin signaling is a ubiquitous feature in colorectal cancer ( Nusse and Clevers , 2017; Zhan et al . , 2017 ) and thus comprehensive understanding of Wnt regulators is a key towards developing therapeutic approaches for cancer . Wnt/β-catenin signaling operates via the transcriptional coactivator β-catenin , whose level is tightly regulated by Axin/APC/GSK3 destruction complex-mediated phosphorylation , ubiquitination , and proteasomal degradation . Binding of Wnt ligands to Frizzleds ( FZDs ) receptors and co-receptors of the LDL Receptor Related Protein ( LRP ) −5 and −6 family inhibits GSK3 and the destruction complex , hence β-catenin can accumulate and translocate to the nucleus ( Nusse and Clevers , 2017; Zhan et al . , 2017 ) . In addition , Wnt signaling is also elaborately tuned at the receptor level ( Niehrs , 2012; Kim et al . , 2013; Green et al . , 2014 ) . For example , the single transmembrane E3 ligases ZNRF3/RNF43 ubiquitylate and downregulate FZDs and LRP6 , imposing negative feedback control on Wnt signaling . R-spondin ligands sequester ZNRF3/RNF43 with LGR4/5/6 and lead to the membrane clearance of ZNRF3/RNF43 ( Carmon et al . , 2011; de Lau et al . , 2011; Glinka et al . , 2011; Hao et al . , 2012; Koo et al . , 2012 ) . Thereby , R-spondins increase the membrane abundance of Wnt receptors and potentiate Wnt signaling . Aberrant Wnt/R-spondin/ZNRF3 signaling is implicated in tumorigenesis , where 7% of colon cancer and 31% of serrated adenoma samples harbor RSPO3 gene fusions with the neighboring Protein tyrosine phosphatase receptor-type kappa ( PTPRK ) gene ( Seshagiri et al . , 2012; Sekine et al . , 2016 ) . In these gene fusions , the signal sequence of PTPRK is fused to RSPO3 , reducing PTPRK and leading to elevated RSPO3 protein levels , which in transgenic mouse models are sufficient to drive tumor initiation ( Han et al . , 2017 ) . The tumor-promoting effect of PTPRK-RSPO3 gene fusions is solely attributed to elevated R-spondin levels , while little attention has been paid to a possible role of PTPRK in this context . PTPRK belongs to R2B subfamily of Receptor type protein tyrosine phosphatases ( RPTP ) ( Jiang et al . , 1993 ) , which contain an adhesion molecule-like extracellular domain and a cytoplasmic tyrosine phosphatase domain ( Lee et al . , 2015 ) . PTPRK can be cleaved by multiple proteases to generate a soluble intracellular fragment that can translocate into the nucleus ( Anders et al . , 2006; Tonks , 2006 ) . Hence , PTPRK can have a variety of substrates from the membrane to the nucleus . PTPRK dephosphorylates and inactivates oncogenic proteins such as STAT3 , EGFR and CD133 , is frequently downregulated in human cancers , and is considered a tumor suppressor ( McArdle et al . , 2001; Flavell et al . , 2008; Tarcic et al . , 2009; Assem et al . , 2012; Scrima et al . , 2012; Mo et al . , 2013; Sun et al . , 2013; Chen et al . , 2015; Shimozato et al . , 2015 ) . Here , we report that not only RSPO3 but also its fusion partner PTPRK encodes a regulator of ZNRF3 and Wnt/β-catenin signaling . In Xenopus embryos , both ptprk and znrf3 are expressed in the Spemann organizer and are required to inhibit Wnt signaling to promote early embryonic axial patterning and head formation . PTPRK binds to ZNRF3 , causes its tyrosine-dephosphorylation at a conserved ‘4Y’ internalization signal , and enhances ZNRF3-mediated Wnt receptor turnover . Thus , PTPRK has the opposite function of RSPO3 , promoting- instead of preventing Wnt receptor removal . Our study suggests that in PTPRK-RSPO3 gene fusions , truncation of PTPRK and increased expression of RSPO3 in fact work in the same direction , impairing ZNRF3 to augment Wnt signaling . To uncover novel regulators of Wnt/β-catenin signaling , a genome-wide small interfering RNA ( siRNA ) screen using Wnt reporter assay as a readout was previously performed ( Cruciat et al . , 2010 ) and PTPRK was discovered as a potential candidate . In the H1703 human lung adenocarcinoma cell line , knockdown of PTPRK enhanced Wnt3a induced signaling in Topflash reporter assays ( Figure 1A and Figure 1—figure supplement 1A ) as well as expression of the endogenous Wnt target gene AXIN2 ( Figure 1B and Figure 1—figure supplement 1B ) . siPTPRK also increased cytosolic β-catenin levels and nuclear accumulation of β-catenin upon Wnt3a treatment ( Figure 1C–D ) . PTPRK was reported to promote membrane association of β-catenin ( Novellino et al . , 2008 ) , but we found no change in β-catenin in the membrane fraction in siPTPRK treated cells ( Figure 1—figure supplement 1C ) . Furthermore , in epistasis experiments , siPTPRK increased Topflash reporter activity only when the Wnt reporter was activated by Wnt3a but not following transfection of Wnt1/Fzd8/LRP6 , Dvl1 ( Dishevelled 1 ) , or constitutively active β-catenin ( S37A ) ( Figure 1E ) . PTPRK affected Wnt signaling only upon knockdown , but not overexpression ( Figure 1—figure supplement 1D ) . Moreover , unlike other negative Wnt regulators such as Naked , APC , or GSK3 , which act universally , Wnt inhibition by PTPRK was not observed in e . g . HEK293T cells ( Figure 1—figure supplement 1E ) , and hence PTPRK seems to act cell-type specifically . In addition , when we tested other RPTPs expressed in H1703 cells ( based on available RNAseq databases ) , siPTPRK showed the strongest effect on inducing AXIN2 expression , besides siPTPRF ( Figure 1—figure supplement 1F ) . Taken together , these results indicate that PTPRK acts at the receptor level to inhibit Wnt/β-catenin signaling in H1703 cells . We next studied the role of PTPRK in vivo in the African clawed frog Xenopus tropicalis , since the role of early Wnt signaling in the Spemann organizer of amphibian embryos is well-established ( Niehrs , 2004 ) . Analysis of Xenopus ptprk by qRT-PCR showed that it was expressed maternally and continued to be expressed at similar levels during gastrulation , increasing with organogenesis ( Figure 2—figure supplement 1A ) . By whole-mount in situ hybridization , ptprk was expressed in the animal hemisphere of blastula embryos ( Figure 2—figure supplement 1B ) . In early gastrulae , ptprk was prominently expressed in the Spemann organizer ( Figure 2A ) . While clearly enriched on the dorsal side , ptprk expression was not exclusive to the organizer but was also weakly detected in ventral cells . Interrogating a database derived from RNAseq of Xenopus genes with ranked organizer-specific expression ( Ding et al . , 2017 ) confirmed differential expression of ptprk on the dorsal side , but with lower enrichment than some other ‘organizer genes’ ( Figure 2D ) . In neurulae and tailbud embryos , ptprk was most prominently expressed in the notochord ( Figure 2B–C ) , an organizer derivative , which plays a critical role in neural patterning ( Hemmati-Brivanlou et al . , 1990; Yamada et al . , 1991; Roelink et al . , 1994; Barnett et al . , 1998; Wilson and Maden , 2005 ) . Low expression was detected in the neural plate , as well as branchial arches and dorsal lateral plate ( Figure 2B ) . We conclude that Xenopus ptprk is prominently expressed in the Spemann organizer and notochord . We depleted Ptprk by Morpholino antisense oligo ( Mo ) injection , targeting the splice site between exon 1 and intron 1 of Xenopus tropicalis ptprk , and efficiently reduced ptprk mRNA ( Figure 2—figure supplement 1C ) . Microinjection of ptprk Mo in Xenopus ( ‘morphants’ ) led to reduced head structures and shortened body axis , which was rescued by coinjection of untargeted human PTPRK mRNA ( Figure 2E and Figure 2—figure supplement 1D ) , demonstrating Mo specificity . To further confirm specificity of these defects , we carried out CRISPR/Cas9 mediated ptprk gene editing . A single guide RNA ( sgRNA ) was designed to target a sequence within ptprk exon one and the genome modification was confirmed by StuI enzyme digestion ( Figure 2—figure supplement 1F–G ) . The ptprk genome-edited embryos ( ‘crispants’ ) showed the same phenotype as ptprk morphants . Anterior and tail formation defects are characteristically observed following overactivation of zygotic Wnt signaling ( Christian and Moon , 1993 ) and expectedly Wnt8 DNA overexpression phenocopied the ptprk morphant and crispant phenotype ( Figure 2E and Figure 2—figure supplement 1D–E ) . Concordantly , depletion of Ptprk upregulated Wnt-induced Topflash activity in Xenopus embryos , both in morphants ( Figure 2F ) and crispants ( Figure 2G ) . Increased Wnt activity in ptprk morphants was restored by human wild-type PTPRK RNA but not by an intracellular domain deletion mutant ( PTPRK-ΔC ) or phosphatase-dead mutants ( PTPRK-CS , PTPRK-DA ) ( Figure 2F , H–I ) , indicating that the tyrosine phosphatase activity is essential for Wnt inhibition . The importance of PTPRK phosphatase activity in Wnt regulation was also confirmed in H1703 cells ( Figure 2—figure supplement 1H ) . We conclude that ptprk depletion upregulates Wnt signaling and phenocopies Wnt overactivation during early Xenopus tropicalis development , supporting that Ptprk is a negative regulator of Wnt signaling not only in H1703 cancer cells but also in vivo . Inhibition of zygotic Wnt signaling is required for normal organizer gene expression ( Hoppler et al . , 1996; Kirsch et al . , 2017; Ding et al . , 2018 ) . Consistently , microinjection of ptprk Mo downregulated expression of Spemann organizer effector genes , including chordin ( chd ) , goosecoid ( gsc ) and Xnot2 ( Figure 3A and Figure 3—figure supplement 1A–C ) . Zygotic Wnt signaling inhibits anterior neural gene expression , which is counteracted by Wnt antagonists . To corroborate the role of Ptprk in Wnt-mediated anterior neural patterning , we analyzed expression of the forebrain markers , bf1 and otx2 . Unilateral injection of ptprk Mo with lineage tracer downregulated bf1 and otx2 expression on the injected side , as did Wnt8 DNA overexpression ( Figure 3B and Figure 3—figure supplement 1D–E ) . Neural induction was not impaired as expression of the pan-neuronal marker sox3 was unaffected ( Figure 3B and Figure 3—figure supplement 1F ) . We carried out rescue experiments in Xenopus noggin-neuralized animal cap explants ( Lamb et al . , 1993 ) . BMP4 inhibition by noggin mRNA injection expectedly induced neural markers , and ptprk Mo reduced the expression of bf1 and otx2 , but not sox3 ( Figure 3C–E and Figure 3—figure supplement 1G ) . Importantly , knockdown of lrp6 or β-catenin using established Mos ( Heasman et al . , 2000; Hassler et al . , 2007 ) rescued the effects of ptprk Mo on bf1 or otx2 expression in a dose-dependent manner ( Figure 3C–D ) . These results confirm that the reduction of forebrain markers in ptprk morphants resulted from increased Wnt activity . We conclude that Ptprk promotes Spemann organizer function by negatively modulating Wnt/β-catenin signaling at the Lrp6 receptor level in vivo . The in vitro and in vivo data clearly indicated that PTPRK regulates Wnt signaling at the receptor level . Moreover , PTPRK depletion increased not only LRP6 phosphorylation/activation , but also total LRP6 levels in H1703 cells and Xenopus embryos ( Figure 4A–C and Figure 4—figure supplement 1A , C ) , without affecting LRP6 mRNA levels ( Figure 4—figure supplement 1B , D ) . This suggests that PTPRK directly or indirectly reduces LRP6 protein levels . The transmembrane E3 ligases ZNRF3 and its homolog RNF43 are key negative regulators of Wnt receptor levels at the plasma membrane ( Hao et al . , 2012; Koo et al . , 2012 ) . Hence , we explored if PTPRK may act through ZNRF3/RNF43 . PTPRK depletion upregulated LRP6 levels similarly to knockdown of ZNRF3 and RNF43 in H1703 cells ( Figure 4B and Figure 4—figure supplement 1A ) as well as in Xenopus embryos ( Figure 4C and Figure 4—figure supplement 1C ) . ZNRF3/RNF43 degrade not only LRP6 but also FZD receptors ( Hao et al . , 2012; Koo et al . , 2012 ) . We therefore monitored FZD levels at the plasma membrane by flow cytometry using a pan-FZD antibody ( OMP-18R5 ) ( Gurney et al . , 2012; Hao et al . , 2012 ) . Consistently , not only siZNRF3/RNF43 but also siPTPRK increased FZD cell surface levels ( Figure 4D ) . Examining their epistasis , siPTPRK and siZNRF3/RNF43 treatments both elevated LRP6 cell surface levels , but LRP6 levels were not further enhanced by their combined knockdown ( Figure 4E ) . Likewise , depletion of PTPRK or ZNRF3/RNF43 elevated Topflash activity , while the combined knockdown did not further increase it ( Figure 4F and Figure 4—figure supplement 1E ) . Since a role for ZNRF3 has not been reported in Xenopus , we characterized its expression in Xenopus tropicalis . Maternal znrf3 mRNA was detected in the animal hemisphere; in gastrulae it was prominently expressed in the organizer ( Figure 5A ) , consistent with RNAseq analysis ( Figure 2D ) . ZNRF3 is a Wnt target gene ( Hao et al . , 2012 ) and likewise in Xenopus embryos it shows a pattern that follows high Wnt activity ( Figure 5A ) ( Borday et al . , 2018 ) , including the posterior of early neurulae , and in tailbud embryos the midbrain , the dorsal neural tube and branchial arches . We knocked down znrf3 in Xenopus with two independent antisense Mos . One targets the splice site between exon 1 and intron 1 of Xenopus tropicalis znrf3 , robustly reducing znrf3 mRNA levels ( Mo1 , Figure 5—figure supplement 1A ) ; the other targets the 5’-UTR ( Mo2 ) . Depletion of Znrf3 elicited axial defects that phenocopied ptprk morphants/crispants ( Figure 5B–C ) . Xenopus znrf3 morphants were rescued by coinjection of human untargeted ZNRF3 mRNA ( Figure 5B–C ) . Expectedly , znrf3 Mo robustly induced Topflash activity in Xenopus embryos ( Figure 5—figure supplement 1B ) . Both ptprk and znrf3 show Spemann organizer expression and downregulate Wnt signaling . Accordingly , to examine whether Ptprk regulates Spemann organizer genes through Znrf3 , we coinjected ptprk Mo with or without human ZNRF3 mRNA . ZNRF3 overexpression rescued both gsc and chordin expression , which were decreased by ptprk Mo ( Figure 5D and Figure 5—figure supplement 1C ) . To test for their functional cooperation , we co-injected ptprk and znrf3 antisense Mos at sub-threshold doses , which individually hardly produced an effect . However , when combined , ptprk and znrf3 Mos synergistically enhanced Topflash activity ( Figure 5E ) . In addition , overexpression of human ZNRF3 rescued Topflash induction by ptprk Mo ( Figure 5F ) . Taken together , the results support that PTPRK is an upstream positive regulator of ZNRF3 and thereby reduces cell surface Wnt receptors , which is essential for proper Spemann organizer function and Xenopus axial patterning . We explored by co-immunoprecipitation ( CoIP ) if PTPRK and ZNRF3 physically interact . We used ZNRF3-ΔRING as it is more stable at the plasma membrane compared to wild-type ZNRF3 . In CoIP experiments , full-length and phosphatase dead ( DA ) PTPRK bound to ZNRF3-ΔRING , whereas PTPRK-ΔC did not ( Figure 6A ) . Moreover , PTPRK but not PTPRK-ΔC colocalized with ZNRF3-ΔRING in punctae at the plasma membrane ( Figure 6—figure supplement 1A ) . These results indicate that PTPRK binds to ZNRF3 via its intracellular domain . We generated a H1703 cell line harboring doxycycline ( Dox ) inducible ZNRF3-HA ( TetOn ZNRF3-HA ) to overcome both , poor transfection efficiency in this cell line and general lack of ZNRF3 antibodies . Employing this cell line , we tested if ZNRF3 is tyrosine phosphorylated and may be a substrate of PTPRK . By CoIP and Western blot detection with a phospho-Tyr-specific antibody , we observed very little phosphorylated ZNRF3 ( Figure 6B , lane 2 ) . However , inhibiting endocytic traffic and lysosomal degradation with bafilomycin induced ZNRF3 phosphorylation , and treatment with the pan-PTP inhibitor Na-pervanadate ( PV ) massively increased ZNRF3 phosphorylation ( Figure 6B , lanes 4 , 6 ) . These results suggest that i ) ZNRF3 is tyrosine-phosphorylated but becomes rapidly dephosphorylated by PTPs , ii ) that its phosphorylation status is related to vesicular traffic and lysosomal degradation . Interestingly , siPTPRK enhanced tyrosine phosphorylation of ZNRF3 both in control as well as in bafilomycin-treated cells ( Figure 6B , lane 3 , 5 ) , suggesting that ZNRF3 is a substrate of PTPRK . Concordantly , when phosphorylated ZNRF3 was bound to immobilized PTPRK , ZNRF3 could be eluted by vanadate ( Figure 6C ) , which mimics the conformation of the phosphate group at the transition state for phosphoryl transfer ( Lindquist et al . , 1973 ) , hence indicating an enzyme-substrate interaction . Moreover , siRNA knockdown of other PTPRs also increased ZNRF3 phosphorylation , notably siPTPRF ( Figure 6—figure supplement 1B ) , which also induced Wnt signaling ( AXIN2 expression; Figure 1—figure supplement 1F ) . ZNRF3 and RNF43 continuously degrade Wnt receptors by binding and recruiting them to the lysosome in an ubiquitin-dependent manner ( Koo et al . , 2012; Tsukiyama et al . , 2015; Park et al . , 2018 ) . Hence , we analyzed whether PTPRK regulates ZNRF3 plasma membrane levels using a cell surface biotinylation assay . siPTPRK robustly increased surface levels of ZNRF3 but not that of ZNRF3-ΔRING ( Figure 6D and Figure 6—figure supplement 1C–D ) , indicating that PTPRK promotes ZNRF3 internalization for which the RING domain is required . To analyze if tyrosine phosphorylation impacts the E3 ligase activity of ZNRF3 , we carried out an in vitro ubiquitination assay , monitoring autoubiquitination of ZNRF3 by using immunoprecipitated ZNRF3 and recombinant E2 ubiquitin conjugating enzyme . There was no change in ZNRF3 autoubiquitination following increased tyrosine phosphorylation upon either siPTPRK or Na-pervanadate treatment ( Figure 6—figure supplement 1E ) . This suggests that tyrosine phosphorylation does not regulate the catalytic activity of ZNRF3 . We hypothesized that increased surface ZNRF3 upon PTPRK depletion is due to reduced lysosomal traffic . Concordantly , siPTPRK reduced the colocalization of ZNRF3 with the lysosomal marker LAMP1 ( Figure 6E ) . In contrast , siPTPRK did not increase vesicular colocalization of ZNRF3 and Rab11 ( recycling endosome marker ) ( Figure 6—figure supplement 1F ) . ZNRF3 and RNF43 deplete Wnt receptors from the cell surface and target them towards lysosomal degradation ( Koo et al . , 2012; Tsukiyama et al . , 2015 ) . Consistently , in TetOn ZNRF3-HA cells , Dox treatment dose-dependently increased ZNRF3 and decreased LRP6 levels ( Figure 6F ) . siPTPRK treatment reversed the effect on LRP6 and further increased ZNRF3 levels . Similarly , transfected ZNRF3 reduced FZD5 dose-dependently , while this effect of ZNRF3 was abolished upon siPTPRK treatment ( Figure 6—figure supplement 1G ) . To confirm this result , we monitored the kinetics of LRP6 internalization and degradation using cleavable biotin . In siCo cells , internalized LRP6 was detected after 30 min ( Figure 6G , compare lanes 2 and 4 ) and decreased after 90 min , likely due to lysosomal degradation ( compare lanes 4 and 5 ) . In contrast , siPTPRK prevented degradation of internalized LRP6 ( compare lanes 6 and 7 ) . Taken together , these results support a model in which vesicular trafficking of ZNRF3 and its ability to degrade Wnt receptors is regulated by tyrosine phosphorylation: Phosphorylation maintains plasma membrane residence while dephosphorylation by PTPRK promotes lysosomal targeting and degradation ( Figure 7—figure supplement 2 ) . Tyrosine-containing motifs are known to play a critical role in regulating endocytosis of transmembrane proteins . Specifically , unphosphorylated YXXXφ , φXXY , as well as YXXφ ( φ = bulky hydrophobic amino acid ) sites can serve as internalization motifs ( Zhang and Allison , 1997; Roush et al . , 1998; Bonifacino and Traub , 2003; Royle et al . , 2005 ) . By multisequence alignment and inspection of the intracellular domain of ZNRF3 , we identified a matching cluster of four adjacent tyrosine residues , or ‘4Y’ motif ( Y465 , Y469 , Y472 and Y473 ) , which is highly conserved among vertebrates ( Figure 7A ) . Each of these four tyrosine residues conforms to the aforementioned internalization motifs , suggesting that 4Y represents a cluster of four consecutive internalization signals . To test whether the 4Y motif regulates ZNRF3 endocytosis , we designed a deletion construct ZNRF3 ( Δ4Y ) ( deletion of 9 amino acids encompassing the four tyrosines ) and monitored its subcellular localization . Indeed , ZNRF3 ( Δ4Y ) displayed enhanced membrane staining compared to wild-type ( Wt ) ZNRF3 ( Figure 7B ) . Moreover , PTPRK knockdown induced tyrosine phosphorylation of Wt ZNRF3 but not that of ZNRF3 ( Δ4Y ) ( Figure 7C; compare lanes 3 and 5 ) . This result was confirmed with a mutant ZNRF3 ( 4YF ) , wherein all four tyrosine residues are substituted by phenylalanine ( Figure 7—figure supplement 1A; compare lanes 3 and 7 ) . Note though , that Na-pervanadate ( PV ) treatment induced massive tyrosine phosphorylation of ZNRF3 regardless of its mutation status , indicating additional PTPRK-independent phosphosites . We hypothesized that reduced endocytosis of ZNRF3 ( Δ4Y ) would impair its ability to internalize Wnt receptors and render it hypoactive . Concordantly , ZNRF3 ( Δ4Y ) downregulated FZD5 less efficiently than Wt ZNRF3 ( Figure 7D ) . Moreover , ZNRF3 ( Δ4Y ) and ZNRF3 ( 4YF ) were less efficient in inhibiting Topflash reporter assays compared to Wt ZNRF3 ( Figure 7E; Figure 7—figure supplement 1B ) . Taken together , our results suggest a model ( Figure 7—figure supplement 2 ) where the 4Y motif of ZNRF3 represents an endocytic signal that promotes ZNRF3-Wnt receptor co-internalization . Phosphorylation of the 4Y motif by an unknown tyrosine kinase ( s ) prevents internalization and degradation of Wnt receptors , resulting in higher Wnt signaling . PTPRK counteracts this activity by dephosphorylating the 4Y motif , allowing efficient endocytosis of ZNRF3-Wnt receptor complexes and reducing Wnt signaling . The three main conclusions of this study are i ) that the transmembrane phosphatase PTPRK , whose gene is found in prominent cancer-related fusion events with the ZNRF3 negative regulator RSPO3 , is itself a positive regulator of ZNRF3 . Thereby , PTPRK acts as negative regulator of Wnt/β-catenin signaling , enhancing Wnt receptor turnover; ii ) that PTPRK depletes cell surface LRP6 and FZD by promoting lysosomal trafficking of ZNRF3 , which it binds and whose tyrosine dephosphorylation on a 4Y endocytic signal it promotes; iii ) that Wnt inhibition by PTPRK and ZNRF3 is essential in the Spemann organizer to regulate anterior neural development . During animal development , Wnt signaling serves as a posteriorizing signal , and the tail-to-head gradient of Wnt activity is critical for the a-p specification of the neural plate ( Petersen and Reddien , 2009; Niehrs , 2010 ) . The Spemann organizer is a rich source of negative Wnt regulators , which maintain organizer function and promote anterior development . Joining this group of proteins , Ptprk is essential to downregulate Lrp6 and Wnt signaling to promote Spemann organizer and anterior development in Xenopus . Also the zebrafish ptprk ortholog is expressed in the early dorsal axis and notochord ( van Eekelen et al . , 2010 ) . In contrast , Ptrpk null mutant mice are viable ( Skarnes et al . , 1995 ) and similarly we observed in mammalian cell-lines that the function of PTPRK is not universal but cell-line dependent , possibly reflecting redundancy with other RPTPs . Species differences in the essentiality of orthologous genes is common , even between the more closely related mouse and man , where > 20% of human essential genes have nonessential mouse orthologs ( Liao and Zhang , 2008 ) . Indeed , we found that PTPRF may also regulate ZNRF3 and Wnt signaling and be functionally redundant with PTPRK . Despite its key importance as a negative Wnt regulator , the regulation of ZNRF3 is incompletely understood ( Deng et al . , 2015; Shi et al . , 2016; Ci et al . , 2018; Qiao et al . , 2019 ) . Our results in H1703 cells and Xenopus embryos clearly indicate that PTPRK regulates Wnt signaling in a phosphatase activity-dependent manner . Concordantly , PTPRK binds to ZNRF3 via its intracellular domain and the binding is abolished by vanadate , corroborating that ZNRF3 is a PTPRK substrate . We identify a 4Y endocytic signal in ZNRF3 , which is tyrosine phosphorylated by an unknown kinase and dephosphorylated by PTPRK and whose mutation leads to plasma membrane accumulation of ZNRF3 . Tyrosine phosphorylation is known to play an important role in sorting of transmembrane proteins to endosomes and lysosomes . For example , tyrosine phosphorylation of an endocytic YXXφ signal was shown to inhibit endocytosis and lysosomal targeting of CTLA-4 by decreasing binding to the endocytic adaptor protein AP2 ( Bonifacino and Traub , 2003 ) . Our results support a model in which lysosomal trafficking of ZNRF3 regulates its ability to degrade Wnt receptors , likely by escorting them ( Figure 7—figure supplement 2 ) . However , while PTPRK promotes ZNRF3 internalization via the 4Y motif , the overexpressed ZNRF3 ( Δ4Y ) mutant is still able to deplete Wnt receptors and inhibit Wnt signaling , albeit less efficiently ( Figure 7D–E ) . Hence , the 4Y motif and PTPRK only have a modulatory role towards ZNRF3 . PTPRK belongs to the R2B RPTP subfamily , which also includes PTPRM , PTPRT and PTPRU , sharing a common protein architecture ( Craig and Brady-Kalnay , 2015 ) . Among these , PTPRK , PTPRM , and PTPRT are implicated as tumor suppressors ( Zhao et al . , 2010; Sudhir et al . , 2015 ) , raising the possibility that they may also regulate ZNRF3 in certain cell types . We found that knockdown of PTPRF ( R2A ) and PTPRH ( R3 ) also increases ZNRF3 tyrosine phosphorylation in H1703 cells . PTPRK ( R2B ) , PTPRF and PTPRS ( R2A ) were reported to dephosphorylate EGFR and attenuate EGF signaling ( Suárez Pestana et al . , 1999; Xu et al . , 2005; Wang et al . , 2015 ) , suggesting that there can be functional redundancy between R2A and R2B RPTPs . PTPRK is a candidate tumor suppressor in mouse intestinal tumorigenesis as per insertional mutagenesis ( Starr et al . , 2009; March et al . , 2011 ) , and is a gene fusion partner with the oncogene RSPO3 in colorectal cancers ( Seshagiri et al . , 2012 ) . Our results provide a rationale how PTPRK may function as a tumor suppressor in Wnt-ON tumors . ZNRF3 and RNF43 play a widespread role as negative feedback regulators in Wnt signaling ( Hao et al . , 2012 ) . They are frequently mutated in a variety of cancers and their mutation signatures have shown promise as predictive biomarkers in pre-clinical models for the efficacy of upstream Wnt inhibitors ( Hao et al . , 2016 ) . Downregulation of PTPRK and hence ZNRF3 would derepress Wnt-receptors , activate Wnt signaling , and promote tumorigenesis . However , other modes of action of PTPRK , such as dephosphorylating other signaling factors like β-catenin , EGFR , or STAT3 ( Xu et al . , 2005; Novellino et al . , 2008; Chen et al . , 2015 ) , or cell junction proteins ( Fearnley et al . , 2019 ) , may also contribute to its tumor-suppressive function . Translocations where the signal sequence or part of the extracellular domain of PTPRK is fused to RSPO3 are recurrent events in a subset of colorectal cancers ( Seshagiri et al . , 2012 ) . The tumorigenicity of these fusions has been solely attributed to upregulation of RSPO3 and hence ZNRF3/RNF43 depletion . However , our results indicate that haploinsufficiency of PTRPK could also contribute to tumorigenicity by further reducing ZNRF3 and increasing Wnt receptor levels . The close proximity of PTPRK and RSPO3 loci and occurrence in gene fusions , their common function as Wnt signaling regulators , and the fact that at least one more Wnt regulatory gene , RNF146 ( Zhang et al . , 2011 ) , is located within the 1 . 5 Mb genomic interval encompassing RSPO3 and PTPRK , is suggestive of a mini ‘Wnt-operon’ at this locus . Hence , it may be worthwhile probing the other four genes located in this interval ( ECHDC1 , SOGA3 , THEMIS , C6orf58 ) for a Wnt-regulatory function . Our results suggest that tyrosine kinases phosphorylating ZNRF3 at its 4Y endocytic signal are candidate targets for Wnt-directed tumor therapy , as their inhibition may promote ZNRF3 internalization and Wnt receptor turnover . Hence , it will be interesting in the future to characterize the 4Y kinase ( s ) . H1703 cells ( ATCC ) were maintained in RPMI with 10% FBS , supplemented with 2 mM L-glutamine , 1 mM sodium pyruvate and penicillin/streptomycin . HEK293T cells ( ATCC ) were maintained in DMEM with 10% FBS , supplemented with 2 ml L-glutamine and 1 mM penicillin/streptomycin . Cell identity was authenticated by ATCC by STR profiling . Regular mycoplasma test showed both cell lines were mycoplasma negative . V5 tagged PTPRK ( 1–1446 ) , PTPRKΔC ( 1-776 ) and FZD5 ( 1–585 ) were generated by inserting human PTPRK or FZD5 into a pCS-based vector containing the V5 epitope after the signal peptide of mouse Krm2 . Site directed mutagenesis for V5-PTPRK-CS ( C1089S; Catalytic Cys in phosphate binding site changed to Ser ) and V5-PTPRK-DA ( D1057A; Asp in WPD loop changed to Ala ) was done by two-step PCR and mutations were validated by sequencing . hZNRF3-HA , Myc-ZNRF3-ΔRING and Myc-ZNRF3-ΔC were kindly provided by F . Cong ( Hao et al . , 2012 ) . ZNRF3 ( Δ4Y ) -HA ( deletion of 465–474 ) or ZNRF3 ( 4YF ) -HA ( phenylalanine substitution of Y465 , Y469 , Y472 and Y473 ) were done by amplification of whole plasmids with 5’-phosphorylated primers followed by DpnI digestion and self-ligation . H1703 cells in 12-well plates were lysed with RNA lysis buffer containing 1% β-mercaptoethanol . RNA isolation was performed with Nucleospin RNA isolation kit following the manufacturer’s instruction ( Macherey-Nagel , Düren , Germany ) . Reverse transcription and PCR amplification were performed as described before ( Berger et al . , 2017 ) . Primers and siRNA information are listed in Supplementary file 1 . For Topflash assay in H1703 cell line , 3 . 25 × 103 cells per well were plated in 96-well plates . Where indicated , cells were transfected with 25 nM siRNAs using Dharmafect ( Dharmacon , Lafayette , CO ) . After 24 hr , cells were transfected with plasmids including 5 ng of pTK-Renilla , 25 ng of SuperTop , 2 ng of mWnt1 , 2 . 4 ng of hLRP6 , 0 . 24 ng of Mesd , 0 . 8 ng of mFzd8 , 12 ng of hDvl1 , 0 . 08 ng of human β-catenin S37A using Lipofectamine 3000 ( Invitrogen , Carlsbad , CA ) . pCS2+ vector was used to adjust total DNA amount to 100 ng per well . For Topflash assay in HEK293T cell line , 104 cells per well were plated in 96-well plates and 1 ng of pTK-Renilla , 5 ng of SuperTop , were transfected . After 48 hr of DNA transfection , luciferase activities were measured with Dual-luciferase kit ( Promega , Madison , WI ) . When necessary , Wnt3a conditioned medium was treated 24 hr before measuring the luciferase activities . For isolation of total cell lysates , cells were harvested in cold PBS and lysed with Triton lysis buffer ( 20 mM Tris-Cl , pH 7 . 5 , 1% Triton X-100 , 150 mM NaCl , 1 mM EDTA , 1 mM EGTA , 1 mM β-glycerophosphate , 2 . 5 mM sodium pyrophosphate , 1 mM Na-orthovanadate ) supplemented with complete protease inhibitor cocktail ( Roche , Basel , Switzerland ) . For membrane-enriched fractions , cells were lysed with Saponin lysis buffer ( 20 mM Tris-Cl , pH 7 . 5 , 0 . 05% Saponin , 1 mM MgCl2 , 1 mM Na-orthovanadate ) supplemented with complete protease inhibitor cocktail ( Roche , Basel , Switzerland ) . After centrifugation , the supernatant ( cytosolic fraction ) was discarded and the pellets were lysed with Triton lysis buffer . Lysates were cleared by centrifugation , and Bradford assay was performed to measure the protein concentration . For Western blot , 30 μg of lysates were mixed with NuPage LDS sample buffer containing 50 mM DTT and heated at 70°C for 10 min . For co-immunoprecipitation or pull-down assay , 300 ~ 800 μg of total cell lysates were precleared with 10 μl of A/G plus agarose ( Santacruz Biotechnologies , Santacruz , CA ) on a rotator at 4°C for 1 hr . Precleared lysates were incubated with 10 μl of anti-V5 agarose ( Sigma Aldrich , St . Louis , MO ) or with 20 μl of A/G plus agarose with anti-HA ( 1867423; Roche , Basel , Switzerland ) on a rotator at 4°C overnight . Immunoprecipitated proteins were washed with triton lysis buffer for four times and mixed with NuPage LDS sample buffer containing 50 mM DTT , followed by heated at 70°C for 10 min . Samples were subjected to SDS-PAGE , transferred to nitrocellulose membrane , and blocked with 5% BSA in TBST ( 10 mM Tris-Cl , pH 8 . 0 , 150 mM NaCl , 0 . 05% Tween-20 ) . Primary antibodies in blocking buffer were applied overnight at 4°C , and incubation of secondary antibodies was carried out at RT for 1 hr . Western blot images were taken with SuperSignal West pico ECL ( Thermo Scientific , Waltham , MA ) using LAS-3000 ( Fujifilm , Tokyo , Japan ) . Densitometry analyses were done with Multi-gauge software ( Fujifilm , Tokyo , Japan ) . Antibody information is listed in key resource table . TetOn V5-PTPRK or V5-PTPRK-DA H1703 cells ( bait ) were harvested after 48 hr of doxycycline ( 200 ng/ml ) treatment and lysed in 400 μl lysis buffer A ( 20 mM Tris-Cl , pH 7 . 5 , 100 mM NaCl , 10% Glycerol , 1% Triton ) supplemented with complete protease inhibitor cocktail ( Roche , Basel , Switzerland ) . One milligram of total cell lysate was pulled down with 300 ng anti-V5 antibody plus 10 μl protein A magnetic beads ( 88846; Thermo Scientific , Waltham , MA ) overnight . Beads were washed twice with lysis buffer A and mixed with prey ( see below ) . TetOn ZNRF3-HA H1703 cells ( prey ) were harvested after treatment of 100 μM of freshly prepared Na-pervanadate for 30 min and washed twice with cold PBS followed by lysis with 400 μl buffer B ( 50 mM Tris-Cl , pH 7 . 5 , 150 mM NaCl , 10% Glycerol , 1% Triton , 1 mM EDTA , pH 8 . 0 , 5 mM Iodoacetamide and 10 mM NaF ) . One milligram of total cell lysate was added to the immunoprecipitated bait on protein A beads with additional 500 μl lysis buffer A for 2 hr at 4°C with rotation . After four times wash with lysis buffer A , vanadate elution was done at RT for 30 min by adding 20 mM Na-orthovanadate in 30 μl of buffer A . Supernatants were separated from beads and both were boiled with LDS sample buffer with 50 mM DTT and subjected to SDS-PAGE . For the preparation of Na-pervanadate , 10 μl of 100 mM Na3VO4 was added to 50 μl of 20 mM HEPES ( pH 7 . 5 ) containing 0 . 3% H2O2 , followed by 940 μl H2O and 5 min incubation . 2 μg of catalase ( C1345; Sigma Aldrich , St . Louis , MO ) was added for 5 min to remove unreacted H2O2 . H1703 cells were harvested in cold PBS 72 hr post siRNA transfection , and lysed with hypotonic buffer ( 5 mM HEPES , 1 mM MgCl2 , 2 mM Na-orthovanadate ) containing complete protease inhibitor cocktail ( Roche , Basel , Switzerland ) . After 30 min of incubation on ice , cell membranes was disrupted by syringe pipetting with 26 gauge needles , followed by centrifugation at 800 x g for 5 min at 4°C . The supernatant was centrifuged at 5 , 000 rpm for 5 min at 4°C to remove the debris . After further centrifugation at 30 , 000 rpm for 20 min at 4°C using a Beckman TL-100 with TLA-55 rotor , the pellet contained the membrane fraction and the supernatant the cytosolic part . The membrane pellet was solubilized in Triton lysis buffer . Equal amounts of proteins were mixed with NuPAGE LDS Sample Buffer , heated at 70°C for 10 min , followed by SDS-PAGE and Western blot analysis . TetOn ZNRF3-HA H1703 cells were seeded in 10 cm dishes ( 4 . 3 × 105 cells per dish ) and transfected with the indicated siRNAs . After 24 hours , cells were treated with doxycycline ( 200 ng/ml ) to activate ZNRF3 expression . Three days post induction cells were harvested and lysed in 400 μl Triton lysis buffer ( 50 mM Tris-Cl , pH 7 . 5 , 150 mM NaCl , 10% Glycerol , 1% Triton , 1 mM EDTA , pH 8 . 0 , 5 mM Iodoacetamide , 1 mM Na-orthovanadate , 10 mM N-Ethylmaleimide and 10 mM NaF ) . After pre-clearing the lysates with A/G plus agarose for 1 hr at 4°C , they were pulled down with 150 ng anti-HA and 20 μl A/G plus agarose for 4 hr at 4°C followed by four washes with lysis buffer ( 20 mM Tris-Cl , pH 7 . 5 , 100 mM NaCl , 10% Glycerol , 1% Triton ) and once with PBS . The ZNRF3 IP-beads were resuspended in a volume of 10 μl containing reaction buffer ( 40 mM HEPES , pH 7 . 4 , 50 mM NaCl , 8 mM magnesium acetate ) , 10 μM Ubiquitin , 30 μM ATP , 50 nM UBE1 ( E1 ) , 2 μM UbcH5b ( E2 ) as indicated in the Figure . Samples were incubated for 5 hr at 37°C with gentle shaking before boiling in NuPAGE LDS Sample Buffer containing 50 mM DTT for 2 min at 95°C , followed by PAGE analysis . Cells were grown on coverslips in 6-well plates and fixed in 4% PFA for 10 min . The immunofluorescence experiments were performed as published ( Berger et al . , 2017 ) . Coverslips were mounted with Fluoromount G . Cells were harvested with Versane solution ( Lonza , Basel , Switzerland ) and washed with FACS buffer ( PBS , 1% BSA , 0 . 1% Sodium Azide ) followed by blocking with FACS buffer containing 20 μl of FcγR inhibitor ( eBioscience , San Diego , CA ) for 30 min . After blocking , samples were incubated with 2 . 5 μg/ml of pan-FZD or LRP6 antibody at 4°C for 3 hr followed by two washes with FACS buffer . Goat anti-human Alexa488 or goat anti-mouse Alexa488 with a dilution of 1:1000 was applied to the sample for 1 hr at 4°C . After two washes with FACS buffer , samples were incubated with 1 μg/ml of propidium iodide for 5 min before analysis on a FACScalibur . Ten thousand live cells per sample were acquired and analyzed with FlowJo ( Tree Star Inc , Ashland , OR ) . H1703 cells were transfected with siRNA for 72 hr and the washed three times with cold PBS . Surface proteins were biotinylated with 0 . 25 mg/ml sulfo-NHS-LC-LC-Biotin ( Thermo scientific , Waltham , MA ) at 4°C for 30 min . For non-biotinylated control , PBS was added . The reaction was quenched by 3 washes with 10 mM Monoethanolamine and cells were harvested and lysed with Triton lysis buffer . 200–300 μg of lysate was incubated with 10 μl streptavidin agarose ( Thermo scientific , Waltham , MA ) to pull-down biotinylated surface proteins , and precipitated proteins were subjected to Western blot and detected with indicated antibodies . TetOn ZNRF3-HA H1703 cells were transfected with siRNA for 24 hr and then treated with doxycycline ( 200 μg/ml ) for 48 hr . Surface proteins were biotinylated with 0 . 5 mg/ml sulfo-NHS-SS-Biotin ( Thermo Scientific , Waltham , MA ) at 4°C for 30 min . After quenching excessive biotin with 10 mM Monoethanolamine , pre-warmed culture medium was added for the indicated times at 37°C to induce internalization . At the indicated times , remaining surface-biotin was removed by 50 mM MesNa ( 2-mercaptoethanesulfonate , membrane impermeable reducing agent ) in 100 mM Tris-HCl , pH 8 . 6 , 100 mM NaCl and 2 . 5 mM CaCl2 at 4°C for 30 min and MesNa protected biotinylated proteins were analyzed . Cells were lysed with RIPA buffer ( 20 mM Tris-Cl , pH 7 . 4 , 120 mM NaCl , 1% Triton X-100 , 0 . 25% Na-deoxycholate , 0 . 05% SDS , 50 mM sodium fluoride , 5 mM EDTA , 2 mM Na-orthovanadate ) supplemented with complete protease inhibitor cocktail ( Roche , Basel , Switzerland ) . 200–300 μg lysate was incubated with 10 μl streptavidin agarose ( Thermo Scientific , Waltham , MA ) to pull down biotinylated protein , and precipitated proteins were subjected to Western blot and detected with indicated antibodies . Xenopus tropicalis frogs were obtained from Nasco , National Xenopus Resource ( NXR ) and European Xenopus Resource Centre ( EXRC ) . In vitro fertilization , embryo culture , preparation of mRNA , and microinjection were carried out as described ( Gawantka et al . , 1995 ) . For Xenopus tropicalis embryo injection , mRNA/DNA/Morpholino oligonucleotide ( Mo ) was injected animally between the 2- to 8 cell stage . Equal amounts of total mRNA/DNA or Mo were injected by adjustment with preprolactin ( PPL ) RNA/DNA or standard control Mo ( GeneTools , Philomath , OR ) . Based on Xenopus tropicalis ptprk sequence ( ENSXETG00000010633 ) , an antisense Mo was designed: 5’-TTCTTACCTGCACACTTGGTTCTTG-3’ . The sequence of the antisense Mo targeting Xenopus tropicalis znrf3 ( ENSXETG00000019942 ) was: 5’-CCACTTACCTGCACGATCTCCCCCT-3’ ( Mo1 , splice-blocking Mo ) and 5’-AACATAATTTCCCAGTCCTCAGTGG-3’ ( Mo2 , translation-blocking Mo ) . Injected amount ( per embryo ) of each Mo was as follows: 0 . 5 or 1 ng of lrp6 Mo , 1 , 2 , or 5 ng of β-catenin Mo , and 5 or 10 ng of ptprk Mo , 2 or 10 ng of znrf3 Mo1 , 40 ng of znrf3 Mo2 . The injected mRNA amounts were 1 pg Wnt3a , 500 pg PTPRK WT or mutants , and 30 pg ZNRF3 . For luciferase reporter assays , embryos were injected with Topflash and Renilla-TK plasmid DNA plus indicated Mos and synthetic mRNA . Three pools of 5 embryos each were lysed with passive lysis buffer ( Promega , Madison , WI ) and assayed for luciferase activity using the Dual luciferase system ( Promega , Madison , WI ) . All luciferase reporter assays represent the mean ± standard error of 3 independent measurements of pools ( five embryos per pool; total n = 15 per experiment shown ) . The reproducibility was confirmed by at least three independent experiments in different batches of Xenopus tropicalis embryos . Whole-mount in situ hybridizations were carried out essentially as described ( Gawantka et al . , 1995 ) . The in situ hybridization probe for Xenopus tropicalis ptprk was generated by PCR using Xenopus tropicalis ptprk cDNA ( IMAGE ID: 7708108 ) as a template , a forward primer: 5’-CCCCCCGGGGAGCCTCCAAGGCCTATTGC-3’ , and a reverse primer: 5’-CCCGAATTCGGATGGTAGTCCCTGGATGC-3’ to amplify a fragment with a size of 835 bp . The PCR product was cloned into pBluescript KS+ using SmaI and EcoRI as the upstream and downstream cloning site respectively . The in situ hybridization probe for Xenopus tropicalis znrf3 was generated by PCR using cDNA ( IMAGE ID: 7656097 ) as template , a forward primer: 5’-ATAAGAATGCGGCCGCATGCACCCACTTGGACTCTGTAAT −3’ , and a reverse primer: 5’-ACGCGTCGACGTCCTGAAGATGCATGGTCCAGT-3’ to amplify a fragment with a size of 1000 bp . The PCR product was cloned into pBluescript KS+ using NotI and SalI . For lineage tracing , embryos were injected with 10 ng of ptprk Mo or 10 pg of Wnt8 DNA plus lacZ mRNA ( 200 pg per embryo ) . Embryos were collected at embryonic stage 11 ( gastrula ) or 18 ( neurula ) and processed for in situ hybridization . β-galactosidase staining was performed as described ( Bradley et al . , 1996 ) using Rose-Gal substrate ( Genaxxon bioscience , Ulm , Germany ) . Phenotypes were scored using a stereomicroscope by comparing wild-type and Mo-injected embryo morphology and counting embryos with the indicated abnormalities . For animal cap assay , embryos were injected at 2- to 8 cell stage with 100 pg ( per embryo ) of noggin RNA and indicated Mos into the animal hemisphere . Animal cap explants were excised at stage 9 from 20 embryos and cultivated in 0 . 5x Barth solution containing Penicillin/Streptomycin . Animal cap explants were harvested at stage 18 and lysed in TRIzol ( Thermo scientific , Waltham , MA ) for RNA extraction , and qRT-PCR assays were performed to analyze the expression of indicated genes . For qRT-PCR analysis , 10 embryos at tailbud stage ( stage 25 ) or 20 animal cap explants at neurula stage equivalent ( stage 18 ) were harvested and lysed in 1 ml of TRIzol ( Thermo Scientific , Waltham , MA ) , and RNA extraction and precipitation was performed following the manufacturer’s instruction . Reverse transcription was performed with 1 μg RNA using SuperScript II reverse transcriptase and random primers ( Invitrogen ) . The obtained cDNA was subjected to PCR amplification using UPL ( Universal ProbeLibrary; Roche , Basel , Switzerland ) probes and corresponding primers , and analyzed by LightCycler 480 ( Roche , Basel , Switzerland ) . For Western blot analysis , Xenopus tropicalis embryos were injected with indicated Mos at 2- to 8 cell stage into the animal hemisphere . Embryos were harvested at stage 18 , homogenized in NP-40 lysis buffer ( 2% NP-40 , 20 mM Tris-HCl , pH 7 . 5 , 150 mM NaCl , 10 mM NaF , 10 mM Na3VO4 , 10 mM sodium pyrophosphate , 5 mM EDTA , 1 mM EGTA , 1 mM PMSF , and protease inhibitors ( Roche , Basel , Switzerland ) with a volume of 4 μl per embryo . Lysates were cleared with Freon followed by centrifugation ( 21 , 000 x g , 10 min at 4°C ) , 70°C for 10 min with NuPAGE LDS Sample Buffer , and SDS-PAGE analysis . CRISPR/Cas9-mediated mutagenesis was performed as described ( Nakayama et al . , 2014 ) . In brief , embryos were injected at one-cell in the animal hemisphere with 5 nl per embryo . After injection , embryos were cultured in 1/18 MR until stage 18 for Luciferase assays or stage 30 for phenotyping and genotyping . The putative sgRNA target site for Xenopus tropicalis ptprk and specificity check were predicted on online database CCTop - CRISPR/Cas9 target online predictor ( https://crispr . cos . uni-heidelberg . de/ ) ( Stemmer et al . , 2015 ) and CRISPRdirect ( https://crispr . dbcls . jp/ ) ( Naito et al . , 2015 ) using the exon 1 sequence of ptprk ( Transcript ID: ENSXETT00000023302 . 3 ) . The linear DNA template for ptprk sgRNA was synthesized using a PCR-based strategy . The 5’ primer was: 5’-GCAGCTAATACGACTCACTATAGTGTGGTGGTGCAATAGGCCTGTTTTAGAGCTAGAAATA-3’ , and the 3’ primer was: 5’-AAAAGCACCGACTCGGTGCCACTTTTTCAAGTTGATAACGGACTAGCCTTATTTTAACTTGCTATTTCTAGCTCTAAAAC-3’ . For genotyping using restriction enzyme digestion , individual embryo was transferred to a 0 . 2 ml PCR tubes containing 100 μl of lysis buffer ( 50 mM Tris , pH 8 . 8 , 1 mM EDTA , 0 . 5% Tween 20 , ) with freshly added proteinase K at a final concentration of 200 μg/ml . Embryos were incubated at 56°C for 2 hr to overnight , followed by 95°C for 10 min to inactivate proteinase K . Lysates were centrifuged at 17 , 000 x g for 10 min at 4°C . One microliter of lysate was used as a template for PCR to amplify the targeted genomic region using a forward primer: 5’-AGCCTCAGTCTGGCTTTTTAATTT-3’ , and a reverse primer: 5’-CTCAAGGTTAACGCTACGAAAAATC-3’ . The PCR products were digested by StuI and analyzed by agarose electrophoresis .	How human and other animals form distinct head- and tail-ends as embryos is a fundamental question in biology . The fertilized eggs of the African clawed frog ( also known as Xenopus ) become embryos and grow into tadpoles within two days . This rapid growth makes Xenopus particularly suitable as a model to study how animals with backbones form their body plans . In Xenopus embryos , a small group of cells known as the Spemann organizer plays a pivotal role in forming the body plan . It produces several enzymes known as Wnt inhibitors that repress a signal pathway known as Wnt signaling to determine the head- and tail-ends of the embryo . Chang , Kim et al . searched for new Wnt inhibitors in the Spemann organizer of Xenopus embryos . The experiments revealed that the Spemann organizer produced an enzyme known as PTPRK that was essential to permit the head-to-tail patterning of the brain . PTPRK inhibited Wnt signaling by activating another enzyme known as ZNRF3 . Previous studies have shown that defects in Wnt signaling and in the activities of PTPRK and ZNRF3 are involved in colon cancer in mammals . Thus , these findings may help to develop new approaches for treating cancer in the future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2020	The tumor suppressor PTPRK promotes ZNRF3 internalization and is required for Wnt inhibition in the Spemann organizer
Adaptive decision-making uses information gained when exploring alternative options to decide whether to update the current choice strategy . Magnocellular mediodorsal thalamus ( MDmc ) supports adaptive decision-making , but its causal contribution is not well understood . Monkeys with excitotoxic MDmc damage were tested on probabilistic three-choice decision-making tasks . They could learn and track the changing values in object-reward associations , but they were severely impaired at updating choices after reversals in reward contingencies or when there were multiple options associated with reward . These deficits were not caused by perseveration or insensitivity to negative feedback though . Instead , monkeys with MDmc lesions exhibited an inability to use reward to promote choice repetition after switching to an alternative option due to a diminished influence of recent past choices and the last outcome to guide future behavior . Together , these data suggest MDmc allows for the rapid discovery and persistence with rewarding options , particularly in uncertain or changing environments . Making adaptive decisions in complex uncertain environments often necessitates sampling the available options to determine their associated values . However , for such exploratory decisions to be of any use for future choice strategies , it is critical that the identity of selected options during 'search' choices are appropriately maintained; without this , the outcomes of such choices can not inform subsequent decisions about whether to continue to sample other alternatives or to terminate the search and instead persist with this chosen option ( Quilodran et al . , 2008 ) . Converging evidence suggests that the integrity of orbital and medial parts of prefrontal cortex supports the ability to use feedback to allow rapid regulation of choice behavior and to shift from search to persist modes of responding ( Hayden et al . , 2011; Khamassi et al . , 2013; Morrison et al . , 2011; Walton et al . , 2004; 2011 ) . However , it is not yet clear how all the relevant information is efficiently integrated across these cortical networks for this to occur . One subcortical structure interconnected to these neural networks and therefore in a prime position to help coordinate the rapid integration of choices and outcomes is the mediodorsal thalamus ( MD ) . The MD is heavily interconnected with the prefrontal cortex , and also receives inputs from the amygdala and ventral striatum ( Aggleton and Mishkin , 1984; Goldman-Rakic and Porrino , 1985; McFarland and Haber , 2002; Ray and Price , 1993; Russchen et al . , 1987; Timbie and Barbas , 2015; Xiao et al . , 2009 ) . Causal evidence from animal models indicates that MD provides a critical contribution in many reward-guided learning and decision-making tasks , particularly those requiring rapid adaptive updating of stimulus values ( Chudasama et al . , 2001; Corbit et al . , 2003; Mitchell and Dalrymple-Alford , 2005; Mitchell et al . , 2007b; Mitchell and Gaffan , 2008; Ostlund and Balleine , 2008; Parnaudeau et al . , 2013; Wolff et al . , 2015 ) . By contrast , implementation of pre-learned strategies and memory retention remains intact after selective damage to the magnocellular subdivision of MD ( MDmc ) ( Mitchell et al . , 2007a; Mitchell and Gaffan , 2008 ) . Yet the precise role of MDmc in facilitating such rapid learning and adaptive choice behavior remains to be determined . One potential clue comes from the fact that the functional dissociations occurring after MDmc damage are reminiscent of those observed following lesions to parts of orbitofrontal cortex ( OFC ) ( Walton et al . , 2010; Baxter et al . , 2007; Izquierdo et al . , 2004 ) , to which the MDmc subdivision is reciprocally connected ( Ray and Price , 1993; Timbie and Barbas , 2015 ) . Moreover , intact communication between MDmc and OFC ( as well between MDmc and amygdala ) is critical for rapid updating of reward-guided choices ( Browning et al . , 2015; Izquierdo and Murray , 2010 ) . Lesions to both MD and OFC have been shown to cause deficits on discriminative reversal learning tasks , a finding frequently accompanied by perserveration of choice to the previously rewarded stimulus ( Chudasama et al . , 2001; Clarke et al . , 2008; Floresco et al . , 1999; Hunt and Aggleton , 1998; Ouhaz et al . , 2015; Parnaudeau et al . , 2013; Chudasama and Robbins , 2003 ) . This suggests that the main role for MDmc is to promote flexibility by supporting OFC in inhibiting responding to the previously highest value stimulus and/or learning from negative feedback . However , recent functional imaging , electrophysiology and lesion studies have refined theories of OFC function , suggesting it might play an important role in contingent value assignment or in determining the state space to allow such learning to be appropriately credited ( Jocham et al . , 2016; Walton et al . , 2010; Takahashi et al . , 2011; Wilson et al . , 2014 ) . Therefore , a second possibility is that the MD plays a key role in adaptive decision making by facilitating the rapid contingent learning performed by OFC-centered networks . However , it is important to keep in mind that causal animal evidence indicates that damage to the MDmc does not simply replicate the deficits observed after selective lesions to interconnected prefrontal regions ( Baxter et al . , 2007; 2008; Mitchell et al . , 2007a; Mitchell and Gaffan , 2008 ) , suggesting that the functional role of MDmc may be distinct from any individual cortical target . Indeed , in addition to the OFC , the MDmc also has connections to several other parts of rostral , ventral and medial prefrontal cortex ( Goldman-Rakic and Porrino , 1985; McFarland and Haber , 2002; Ray and Price , 1993; Xiao et al . , 2009 ) . These regions are known to be important not just for value learning , but also for aspects of value-guided decision making such as computing the evidence for persisting with a current default option or to switch to an alternative ( Chau et al . , 2015; Boorman et al . , 2013; Noonan et al . , 2011; Kolling et al . , 2014 ) . Based on this evidence , we predicted that the role of MDmc might go beyond simply enabling OFC-dependent contingent learning and might also directly regulate decisions about when to shift from a search strategy ( sampling the alternatives to build up a representation of their long-term value ) to a persist strategy ( repeating a particular stimulus choice ) . To determine the precise role of MDmc in facilitating trial-by-trial learning and adaptive decision-making , we tested macaque monkeys before and after bilateral neurotoxic lesions to MDmc , and matched unoperated control monkeys , on a series of probabilistic , multiple option reward-guided learning tasks that are sensitive to OFC damage ( Noonan et al . , 2010; Walton et al . , 2010 ) . To perform adaptively , the monkeys had to learn about , and track , the values associated with 3 novel stimuli through trial-and-error sampling and use this information to decide whether or not to persist with that option . In some task conditions ( referred to as ‘Stable’ or ‘Variable’ schedules: see Figure 1B ) , the reward probabilities linked to each stimulus would change dynamically and the identity of the highest value would reverse half way through each session; in others , the probabilistic reward assignments remained fixed throughout the session . If the MDmc is critical for inhibiting responses to a previously rewarded stimulus , then the monkeys with MDmc damage will only be impaired post-reversal and will display perseverative patterns of response selection . If , on the other hand , the MDmc supported contingent learning , these lesioned monkeys would show impairments akin to those observed in monkeys with OFC damage ( Noonan et al . , 2010; Walton et al . , 2010 ) . That is , MDmc-lesioned monkeys would not only be slower to update their choices post-reversal or , in Fixed situations where they had to integrate across multiple trials to determine which option was best , they would also exhibit aberrant patterns of stimulus choices such that a particular reward would be assigned based on the history of all past choices rather than to its causal antecedent choice . Alternatively and finally , if the MDmc is required to regulate adaptive choice behavior , then the lesioned animals would also have a deficit post-reversal or in any Fixed schedules when multiple options are rewarding , but this would be characterized by an impairment in determining when to shift from search to persist modes of responding . 10 . 7554/eLife . 13588 . 003Figure 1 . Task design . ( A ) Schematic of a single trial . At the start of each trial , 3 stimuli were presented on the screen in one of four spatial configurations . Monkeys chose a stimulus by touching its location on the screen . Once selected , the alternative options disappeared and reward was or was not delivered according to a pre-determined schedule ( note that the red box is shown for illustration only , but was not presented during testing ) . Following an intertrial interval , the next trial would begin . ( B ) Schematic of two varying schedules , 'Stable' ( upper panel ) and 'Variable' ( lower panel ) , showing the running average probability ( across 20 trials ) during a session that selecting that option would result in reward . DOI: http://dx . doi . org/10 . 7554/eLife . 13588 . 003 In fact , while we found that MDmc lesions dramatically influenced the speed and patterns of monkeys’ choices , particularly when the identity of the highest rewarded stimulus reversed , there was no evidence either for a failure to inhibit previously rewarded choices or for a misassignment of outcomes based on choice history as had been observed after OFC lesions . Instead , the monkeys with MDmc damage were strikingly deficient at re-selecting a sampled alternative after a search choice that yielded a reward – i . e . , they were more likely to select a different stimulus to that chosen on the previous trial . Further analyses suggested this was caused by the MDmc-lesioned monkeys exhibiting less influence of associations based on their most recent stimulus choices coupled with an intact representation of longer term choice trends , which impaired their ability to update their stimulus choices rapidly in situations when they had a varied choice history . Together , these findings support a novel , key contribution of MDmc in regulating adaptive responding . Pre-operatively , monkeys in both the control group ( n=7 ) and MDmc group ( n=3 ) were able to rapidly learn to choose the highest value stimulus in either the Stable or Variable schedules ( median number of trials to reach criterion of ≥65% V1sch choices over 20 trial window: Controls: Stable 25 . 6 ± 2 . 3 , Variable 27 . 7 ± 6 . 7; MDmcs: Stable 35 . 0 ± 14 . 0 , Variable 21 . 0 ± 0 . 0; all averages are the means across animals ± S . E . M . ) and to update their stimulus choices when the values changed such as after a reversal in the reward contingencies ( median number of trials to reach criterion after reversal: Controls: Stable 82 . 4 ± 16 . 0 , Variable 79 . 6 ± 15 . 3; MDmcs: Stable 96 . 3 ± 28 . 7 , Variable 89 . 0 ± 33 . 9 ) ( Figure 3 ) . Comparison of the rates of selection of the best option , either calculated objectively based on the programmed schedules ( V1sch ) , or as subjectively defined by the monkeys’ experienced reward probabilities based on a Rescorla-Wagner learning algorithm ( V1RL ) ( Figure 3—figure supplement 1 ) , using a repeated measures ANOVA with lesion group ( control or MDmc ) as a between-subjects factor and schedule ( Stable or Variable ) as a within-subjects factor showed no overall difference between the two groups ( main effect of group: F1 , 8 < 0 . 7 , p>0 . 4 ) . The only factor that reached significance was the interaction between group and schedule for the subjectively defined values ( Objective values: F1 , 8 = 1 . 70 p=0 . 23; Subjective values: F1 , 8 = 5 . 46 , p=0 . 048 ) . Importantly , post-hoc tests showed that this effect was not driven by a significant difference between the groups on either condition ( both p>0 . 21 ) , but instead by a significant overall difference in performance in the controls between the Stable and Variable that was not present in the MDmc group . 10 . 7554/eLife . 13588 . 006Figure 3 . Choice performance and latencies on the varying schedules . ( a ) Mean proportion of choices ( ± S . E . M . ) of the V1sch in the control and MDmc groups both pre- and post-operatively . Left and center panel depict group average choices over the whole session ( Controls = blue filled line , MDmc = red dashed line ) ; ( c ) right panel depicts choices divided into the first and last 150 trials ( dots = individual monkey’s choices ) . ( b ) Proportion of trial-by-trial choice response times grouped into 100 ms bins for the controls ( blue bars ) or MDmc monkeys ( red bars ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13588 . 00610 . 7554/eLife . 13588 . 007Figure 3—figure supplement 1 . Choice performance and latencies on the varying schedules for the subjective values . Mean proportion of choices ( ± S . E . M . ) of the V1RL in the control and MDmc groups ( Controls = blue filled line , MDmc = red dashed line ) both pre- and post-operatively . Left panel depicts group average choices over the whole session prior to the MDmc surgery; right panel depicts choices post the MDmc surgery . DOI: http://dx . doi . org/10 . 7554/eLife . 13588 . 007 However , after bilateral neurotoxic damage to the MDmc , as shown in Figure 3 , there was a marked change in choice performance in the MDmc group compared to the control group . A repeated measures ANOVA , with group as a between-subjects factor and both schedule and surgery ( pre-MD surgery or post-MD surgery ) as within-subjects factors , showed a selective significant interaction of lesion group x surgery for the V1sch ( F1 , 8 = 5 . 537 , p=0 . 046 ) . The interaction of lesion group x surgery for the subjective values ( V1RL ) showed a trend for significance ( p=0 . 054 ) ( Figure 3—figure supplement 1 ) . Post-hoc tests indicated that the MDmc group showed a significant decrement in choice performance after surgery , with the lesioned monkeys choosing the highest valued stimulus less frequently than the control monkeys ( all p’s<0 . 05 ) . This change was also accompanied by a marked speeding of choice latencies on all trials . An analogous repeated measures ANOVA using the log transformed response latencies showed a significant lesion group x surgery interaction: F1 , 8 = 25 . 79 , p<0 . 01 ) ( Figure 3b ) . Previous studies in rodents suggest an important role for MD in reversal learning paradigms ( Block et al . , 2007; Chudasama et al . , 2001; Hunt and Aggleton , 1998; Ouhaz et al . , 2015; Parnaudeau et al . , 2013; 2015 ) , although in these studies the damage sustained in the MD involves all subdivisions of the nucleus . To examine whether our selective MDmc lesion caused a particular problem when needing to switch away from the initial highest valued stimulus ( V1sch ) , we separately re-analyzed choice performance during the 1st 150 trials , where the identity of V1sch is fixed , and during the 2nd 150 trials , after the reversal in reward contingencies for V1sch ( Figure 3c ) , again including schedule as a within-subjects factor . While the lesion had no consistent effect during the initial learning phase ( V1sch 1st half: lesion group x surgery: F1 , 8 = 0 . 659 , p=0 . 440 ) , there was a significant change in choices post-reversal ( V1sch 2nd half: lesion group x surgery interaction: F1 , 8 = 5 . 990 , p=0 . 040 ) . Post-hoc tests showed that after surgery the MDmc-lesioned monkeys were selectively worse at choosing the V1sch than controls ( p=0 . 031 ) . These data suggest that the monkeys with damage to the MDmc could not flexibly update their choice behavior in a comparable manner to control monkeys following the reversal in identity of the highest value stimulus . One common explanation for a deficit in behavioral flexibility is that animals with lesions inappropriately continue to choose the previously highest valued stimulus ( 'perseveration' ) , potentially because they fail to learn from negative feedback . Such an effect has been observed during reversal learning in rodents following disruption of the MD ( Floresco et al . , 1999; Hunt and Aggleton , 1998; Ouhaz et al . , 2015; Parnaudeau et al . , 2013 ) . However , despite deficits in reversal learning , our monkeys with MDmc damage displayed no evidence of either perseveration , or a failure to learn from negative feedback . First , in the 50 trials after reversal , the MDmc lesion group and the control group had a similar likelihood of choosing what had been the highest valued stimulus pre-reversal ( ex-V1sch ) ( proportion of ex-V1sch choices: Controls: Pre-surgery: Stable , 39 . 9% ± 5 . 6 , Variable , 39 . 3% ± 3 . 1 , Post-surgery: Stable , 46 . 7% ± 3 . 3 , Variable , 45 . 8% ± 6 . 0; MDmc: Pre-surgery: Stable , 35 . 3% ± 7 . 0 , Variable , 41 . 22% ± 2 . 0 , Post-surgery: Stable , 47 . 1% ± 3 . 3 , Variable , 52 . 11% ± 1 . 3; interactions between group x surgery or group x surgery x schedule: both F1 , 8 < 0 . 6 , p>0 . 45 ) . In fact , as can be observed in Figure 4a , the rate of switching actually increased in the MDmc group after surgery . An analysis of the proportion of times the monkeys explored an alternative option after just receiving a reward ( positive feedback ) or reward being omitted ( negative feedback ) showed a significant lesion group x surgery x previous outcome interaction ( F1 , 8 = 15 . 01 , p=0 . 005 ) . There was also a strong trend towards a 4-way interaction between lesion group , surgery , previous outcome and pre- or post-reversal ( F1 , 8 = 5 . 12 , p=0 . 054 ) ; switch rates selectively increased in the MDmc group after surgery and this was particularly pronounced after a reward ( average post-surgery increase in switch probability: after reward 0 . 15 ± 0 . 07 , post-hoc tests: p=0 . 051; after no reward: 0 . 08 ± 0 . 06 , p=0 . 23 ) . 10 . 7554/eLife . 13588 . 008Figure 4 . Switching behavior in the control and MDmc lesioned monkeys . Mean likelihood of switching to a different stimulus in the two groups both pre- and post-operatively ( a ) throughout each schedule ( mean ± S . E . M . ) or ( b ) divided up into switches ( SW ) following a choice leading to a reward ( CORRECT–SW ) or to no reward ( ERROR–SW ) ( dots = individual monkey’s switching probabilities ) . ( c ) Mean response latency in each animal following a repeated choice of the same stimulus ( ‘St’ ) or a switch to a different stimulus ( ‘Sw’ ) in the two groups ( dots = individual monkey’s latencies ) . Note that two MDmc monkeys had very similar latencies pre-operatively and so their data is overlapping . DOI: http://dx . doi . org/10 . 7554/eLife . 13588 . 008 To explore this increase in switching in more detail , we ran two more repeated measures ANOVAs , one focusing on the pre-reversal period and one on the post-reversal period . While there were no significant interactions with lesion group x surgery in the pre-reversal period ( all F1 , 8 < 2 . 5 , p>0 . 15 ) , there was a significant lesion group x surgery x previous outcome post-reversal ( F1 , 8 < 10 . 9 , p=0 . 011 ) . Further post-hoc tests showed that this effect was mainly driven by a selective increase for the MDmc group to show a tendency to switch to choosing a different stimulus just after having received a reward ( increase in switching probability from pre- to post-MD surgery: 0 . 21 ± 0 . 07 , p=0 . 056 ) , an effect less evident after no reward ( 0 . 06 ± 0 . 08 , p=0 . 47 ) or in the control animals after either outcome ( reward: –0 . 04 ± 0 . 06 , p=0 . 56; no reward 0 . 02 ± 0 . 05 p=0 . 74 ) . This change in switching behavior highlights that , in addition to the absence of evidence for a reduction in sensitivity to negative feedback after an MDmc lesion , there was a change in how positive feedback influenced future choices , particularly in the post-reversal period . As can be seen in Figure 4b , this change meant that in the post-reversal phase , monkeys with MDmc damage became no more likely to stay with a current choice after reward delivery than reward omission . In other words , when the identity of the highest value stimulus changed , the MDmc group , postoperatively , was severely impaired at using the receipt of reward as evidence to continue persisting with that particular previously chosen stimulus . This maladaptive pattern of switching was also reflected in the monkeys’ choice latencies . The control monkeys , and MDmc group pre-surgery , all responded slower on trials where they changed their stimulus choice ( i . e . , choice on current trial ‘n’ ≠ choice on previous trial ‘n-1’ ) compared to trials where they continued to select the same option ( choice on trial n = choice on trial n-1 ) ( Figure 4c ) . However , after surgery , the MDmc group failed to exhibit this post-exploration response slowing on exploration trials ( group x surgery x switch-stay: F1 , 8 = 5 . 77 , p=0 . 043 ) ( Figure 4c ) . MDmc has connections with all parts of the OFC , although they are particularly densest with the lateral OFC , a region that has been implicated in guiding flexible learning and choice behavior ( Murray and Wise , 2010; Wallis and Kennerley , 2010; Walton et al . , 2011 ) . Therefore , it is possible that impaired contingent value learning , observed after lesions to this region ( Walton et al . , 2010 ) , might also underlie the change in performance observed in the MDmc lesioned group . One characteristic of the lateral OFC lesion is that , while the monkeys were unable to correctly credit a reward outcome with a particular choice , they still possessed non-contingent learning mechanisms allowing them to approximate value learning based on the weighted history of all recent choices and rewards , irrespective of the precise relationship between these choices and rewards ( Noonan et al . , 2010; Walton et al . , 2010 ) . This meant that , after a long history of choosing one stimulus ( e . g . , option A ) , a new choice ( e . g . , option B ) would be less likely to be reselected on the following trial after positive than negative feedback and the previously chosen stimulus A would be more likely to be reselected . To determine whether the MDmc lesioned monkeys also approximated associations based on choice history rather than contingent choice-outcome pairs , we ran a series of analyses to establish the specificity of learning as a function of recent reinforcement and choices . In a first analysis , we looked for the effect described above: whether an outcome – reward or no reward – received for choosing any particular option ( ‘B’ ) might be mis-assigned to proximal choices of another option ( ‘A’ ) as a function of how often ‘A’ had been chosen in the recent past ( 'choice history' ) ( note , there were no changes in the ‘B’ reward likelihood as a function of lesion group , surgery or choice history: all F’s < 1 . 54 , p’s>0 . 23 ) . For this analysis , we collapsed across Stable and Variable conditions . As can be observed in Figure 5a , pre-operatively both groups were more likely to re-select the previous ‘B’ option after a reward than no reward across all choice histories . In contrast , post-operatively , the influence of the positive outcome significantly reduced in the MDmc group only ( lesion group x surgery interaction: F1 , 8 = 6 . 32 , p=0 . 036 ) . Further , the number of recent ‘A’ choices made by the MDmc lesion group did not affect the likelihood of re-selecting option ‘B’; the MDmc group post-operatively were no less likely to re-select that option after a reward whether they had a short or long choice history on option ‘A’ ( see Figure 5a ) . Moreover , it was not the case that the influence of the recent outcome was being selectively mis-assigned to option ‘A’ based on recent choices , as was observed in the lateral OFC animals ( Figure 5b ) . Instead , the MDmc group showed no overall significant change in their likelihood of reversing back to option ‘A’ again on the next trial , but rather showed a small increase in the likelihood of switching to the 3rd alternative , ‘C’ ( A choice: group x surgery interaction: F1 , 8 = 2 . 75 , p=0 . 14; C choice: group x surgery interaction: F1 , 8 = 7 . 33 , p=0 . 027 ) ( Figure 5c ) . In other words , after a recent switch , the MDmc lesion group was unable to use the past reward as evidence to reselect either their previous choice or even the option chosen most frequently over recent trials . 10 . 7554/eLife . 13588 . 009Figure 5 . Influence of recent choice history over subsequent choices . ( a–c ) Differential likelihood ( mean ± S . E . M . across monkeys ) of repeating a ‘B’ choice ( a ) , switching back to option ‘A’ ( b ) or switching away to option ‘C’ ( c ) after a ‘B’ choice made on trial n-1 either was rewarded or was not rewarded . Data are plotted in runs following a switch to ‘B’ as a function of the recent choice history: just one choice of a different ‘A’ stimulus on trial n–2 ( ‘A1B ? ’ ) , two choices of ‘A’ on trials n–2 and n–3 ( ‘A2B ? ’ ) , three choices of ‘A’ on trials n–2 to n–4 ( ‘A3B ? ’ ) or four to seven choices of ‘A’ on trials n–2 to n–5–8 ( ‘A4-7B ? ’ ) . ‘A’ and ‘B’ do not refer to particular stimulus identities but instead to arbitrary choices of one option or another . Main plots show Controls ( blue lines ) and MDmc ( red lines ) , filled lines = pre-MDmc surgery; dashed lines = post-MDmc surgery . Insets ( green lines ) depict data from lateral OFC ( LOFC ) lesioned animals taken from a previous experiment reported in Walton et al . ( 2010 ) . ( d ) Differential likelihood ( mean ± S . E . M . across monkeys ) of repeating a ‘B’ choice after a reward or no reward plotted as a function of the number of times option ‘B’ was selected in the 5 previous trials ( n–2 to n–6 ) . Controls = blue lines; MDmc = red lines; Pre-MDmc surgery = filled lines; Post-MDmc surgery = dashed lines . DOI: http://dx . doi . org/10 . 7554/eLife . 13588 . 009 As can be observed in Figure 5a–c , this pattern of choice history appears in marked contrast to monkeys with lateral OFC lesions ( Noonan et al . , 2010; Walton et al . , 2010 ) . To test this difference formally , we directly compared the groups by re-running the ANOVAs now including the lateral OFC group . The analysis of the likelihood of re-selecting ‘B’ option again revealed a lesion group x surgery interaction ( F2 , 10 = 7 . 08 , p=0 . 012 ) but importantly , also now a lesion group x surgery x choice history interaction ( F6 , 30 = 2 . 51 , p=0 . 044 ) . Post-hoc tests showed that , while both the MDmc and lateral OFC groups were on average significantly different to the controls ( both p<0 . 05 ) , the influence of the choice history on the two groups was distinct: only the lateral OFC group post-surgery , but not the MDmc group or controls , exhibited a significant reduction in repetitions of ‘B’ choices after increasing numbers of previous ‘A’ choices ( p=0 . 007 ) . Moreover , while the ‘B’ repetition likelihood was reduced in the MDmc group compared to controls post-surgery , this only occurred when the history of previous ‘A’ choices was p<0 . 05 for A1B and A3B , p=0 . 08 for A2B ) . In contrast , the lateral OFC group were only different to controls after 3 or more previous ‘A’ choices ( p<0 . 05 for A3B and A4-7B ) . Similarly , analysis of the likelihood of returning to option ‘A’ now revealed a lesion group x surgery interaction , driven by a significant overall increase in ‘A’ choices in the lateral OFC group that was not present in either the controls or MDmc lesioned animals . Therefore , unlike the lateral OFC lesion group , which displayed less precise and potentially maladaptive learning based on associating a past outcome with the history of recent choices , the behavior of the MDmc group was instead characterized by a reduced likelihood of repeating a rewarded choice after just having switched to that option . Importantly , it was not the case that the MDmc lesioned monkeys were never able to use reward to promote persistence with a chosen option . In a novel companion analysis , we again probed the influence on subsequent behavior of the past choice and outcome , but now investigated how this was influenced by the frequency with which that particular option had been chosen in the previous 5 trials ( 'choice frequency' ) ( Figure 5d ) . Before surgery , both groups of monkeys were always more likely to persist with an option after being rewarded for that choice than if not rewarded , and this was not significantly influenced by choice frequency . However , post-operatively , although the MDmc lesioned monkeys again were no more likely to re-select the previous choice after reward than after no reward when that option had a low recent choice frequency , this impairment went away in situations when the monkeys had selected that same option on the majority of recent trials . This behavior resulted in a significant surgery x group x choice history interaction ( F2 , 16 = 3 . 80 , p=0 . 045 ) . Therefore , the MDmc lesioned monkeys were just as proficient as controls at weighing the influence of positive over negative feedback in situations where they had a long choice history on the just chosen option , but not if they had seldom chosen that option in the recent past . To further investigate the influence of recent choices and outcomes on future behavior , in a third analysis we ran an identical multiple linear regression analysis used previously ( Walton et al . , 2010 ) focusing on all possible combinations of the past 5 stimulus choices and past 5 outcomes as regressors ( all 25 combinations are shown graphically in Figure 6a ) . This analysis allowed us to look not only at how recent specific choice-outcome pairs might guide future behavior ( red crosses on Figure 6a ) , characteristic of contingent learning known to depend on the lateral OFC , but also to tease out the influence of false associations between recent choices and unrelated past outcomes ( blue area / crosses , Figure 6a ) or recent outcomes and unrelated past choices ( green area / crosses , Figure 6a ) known not to require an intact lateral OFC . A set of confound regressors from combinations of choices/outcomes 6 trials in the past was also included to capture longer-term choice/reward trends , though not shown in the figures . 10 . 7554/eLife . 13588 . 010Figure 6 . Logistic regression on the influence of combinations of recent choices and recent outcomes . ( a ) Representation of the design matrix used in the logistic regression consisting of all combinations of the five previous choices ( rows ) and five previous outcomes ( columns ) . The white squares on the diagonal with red crosses represents the influence of correct contingent learning – choice x outcome combinations; the blue area represents the non-contingent influence of a past outcome spreading forwards to influence more recent choices; the green area represents the non-contingent influence of more recent outcome spreading backwards to associate with an earlier choice . ( b ) Regression weights averaged across the controls and MDmc groups for choices of each of the 3 potential stimuli pre- and post-MD surgery ( lighter shades = larger average regression weights; values have been log transformed for ease of visualization ) . ( c–e ) Regression weights ( mean ± S . E . M . across monkeys , arbitrary units ) for trials n–1 to n–5 for the contingent choice x outcome pairs ( corresponding to the red crosses in a ) ( c ) , past choice x all previous outcomes ( middle panel , blue crosses in panel a ) ( d ) , and past outcome x all previous choices ( lower panel , green crosses in panel a ) ( e ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13588 . 010 Preoperative , both MDmc and control groups exhibited a strong influence of the outcomes they received for the stimuli they had chosen on their future choices , an effect that diminished as the trials became increasingly separated from the current one ( Figure 6b , c ) . In other words , they displayed appropriate contingent value learning . Moreover , as had been observed previously , there was also evidence for non-contingent learning mechanisms as demonstrated by a positive influence of the interaction between ( i ) the most recent choice and unrelated past outcomes ( Figure 6b , d ) and ( ii ) the most recent outcome and unrelated past choices ( Figure 6b , e ) , which also was usually larger for choices / outcomes more proximal to the current trial . Postoperatively , in the MDmc group , although there was a reduction in the influence of the most recent choice-outcome association , the overall past influence of these specific pairs looking back over a 5 trial history was no different after the lesion to pre-surgery levels ( surgery x group interactions: F’s < 1 . 35 , p’s>0 . 27 ) ( Figure 6c ) . This lack of effect suggests that stimulus-outcome contingent learning mechanisms do not necessarily depend on the integrity of MDmc . Similarly , there was also no change in the non-contingent association of the previous choice ( trial n-1 ) with unrelated past outcomes ( trials n-2 to n-5 ) ( surgery x group interactions: F’s < 2 . 2 , p’s>0 . 09 ) ( Figure 6d ) . This result further demonstrates that the MDmc-lesioned animals have intact representations of past outcomes , which can become associated with subsequent choices via 'false' spread-of-effect associations ( Thorndike , 1933 ) . By contrast , there was a change in the influence of associations based on interactions between each received outcome and the recent history of choices in the MDmc lesioned monkeys ( Figure 6e ) , which resulted in a significant surgery x group x past choice interaction ( F4 , 32 = 3 . 13 , p=0 . 028 ) . To explore this effect further , we re-analyzed the data divided into either more recent choices ( stimuli chosen on trial n-1 and n-2 ) or more distant past trials ( choices n-4 and n-5 ) interacting with the current reward . This analysis revealed a surgery x group x choice recency interaction ( F1 , 8 = 9 . 83 , p=0 . 014 ) . Post-hoc pairwise comparisons demonstrated that the MDmc group , post-surgery , had a significantly diminished influence from associations made between the previous outcome and the most recent contingent and non-contingent choices ( p<0 . 05 ) although not with more distant non-contingent choices ( p>0 . 2 ) . The interaction remained significant even if the analysis was re-run with recent trials restricted to non-contingent choices on trials n-2 and n-3 to avoid the potential confound that the weight assigned to trial n-1 could result from either correct contingent learning and non-contingent spread-of-effect ( surgery x group x choice recency interaction: F1 , 8 = 5 . 99 , p=0 . 040 ) . Together this significant interaction suggests that the MDmc monkeys do not simply have a primary deficit in contingent value learning . Instead , MDmc monkeys more generally have a degraded representation of their recent – but not more distant – choices , which prevents each outcome from reinforcing the choices made in the past few trials . Without such a mechanism , monkeys will be poor at re-selecting any recently rewarded choices unless they happen to have an extended history of choosing that particular option . Analysis of overall performance on the varying reward schedules ( Stable or Variable ) highlighted a particular decision making deficit in the MDmc lesioned monkeys that was most prominent when the identity of the best option reversed . However , the problem these monkeys were displaying—a selective failure to persist with a rewarded stimulus choice after having just switched to choosing that stimulus—suggests that the MDmc group may not have a problem with reversals per se but instead in any uncertain situations where they need to use reward to determine which choices should be repeated . This deficit may be particularly pronounced in situations when ( a ) choice histories are not uniform ( e . g . , following a reversal or if the value difference between the available options is small ) and/or ( b ) all potential alternatives are associated with some level of reward . To investigate this idea , we therefore tested the groups post-operatively on additional 3-armed bandit schedules where the reward probability associated with each stimulus was fixed across a session and so there were no reversals of the reward contingencies . In these 'Fixed' schedules ( see Figure 7a ) , the reward ratio of the three options remained the same , but the absolute reward yield changed across the schedules ( the yield of the second schedule was 0 . 75 times the first schedule , the third was 0 . 5 times the first ) . For all schedules , monkeys have to sample the three stimulus options and use receipt of reward to determine which stimulus to persist with . 10 . 7554/eLife . 13588 . 011Figure 7 . Fixed schedule performance . ( a–b ) Schematic of Fixed schedules ( upper panels , a ) and average proportion of choices ( ± S . E . M . ) of the V1sch in the control and MDmc groups in each schedule ( lower panels , b ) . ( c ) Proportion of V1sch choices in the first and last 20 trials in each session for each animal , plotted along with the best-fit linear regression and 95% confidence limits for each group ( Controls or MDmc ) . ( d ) Box plots showing average proportion V1sch choices in the last 20 trials for sessions in which animals made a low number of V1sch choices in the first 20 trials ( ≤25% V1sch choices; 'EARLY LOW' ) or a high number of V1sch choices in the first 20 trials ( ≥75% V1sch choices; 'EARLY HIGH' ) . For all box plots , the central mark is the median , the edges of the box are the 25th and 75th percentiles , and the whiskers extend to the most extreme data points not treated as outliers . ( *p<0 . 05 , Independent Samples Kolmogorov-Smirnov Test , treating each session as an independent sample . ) DOI: http://dx . doi . org/10 . 7554/eLife . 13588 . 011 The control monkeys rapidly learned to find and persist with the best option in all three schedules , reaching a criterion of choosing V1sch on ≥65% of trials on average in 24 . 4 ± 5 . 5 trials ( S . E . M . ) across all schedules ( note that one control monkey was not run on these schedules ) ( Figure 7b ) . By contrast , even without a reversal , the MDmc lesions affected the rate of learning and likelihood of persisting with V1sch , with the MDmc group taking 55 . 0 ± 2 . 2 trials on average to reach the same criterion . Moreover , as can be observed in Figure 7b , the impairment appeared present not just at the start of the session when the animals were initially learning the values , but also persisted throughout the schedule . Therefore , we performed a repeated measures ANOVA comparing performance across Fixed 1–3 ( 'schedule' ) on the first and last half of the schedule ( 'start-end period' ) . This analysis again revealed a main effect of group ( F1 , 7 = 6 . 59 , p=0 . 037 ) , as the MDmc group overall made significantly fewer choices of the best option . Importantly , there was also a significant quadratic interaction between schedule x group x start-end period ( F1 , 7 = 7 . 01 , p=0 . 033 ) . Post-hoc tests showed that while the control monkeys made significantly more choices of V1sch in the second half of the period for all three schedules ( all F’s1 , 7 > 6 . 59 , p’s<0 . 05 ) , the MDmc group failed to do this on two out of the three schedules . In other words , across sessions using the three different reward schedules , the MDmc group were consistently impaired at rapidly finding and persisting with the best option in situations when all options had some probability of reward . Given the results from the varying schedules , we hypothesized that the MDmc lesion should most affect the ability of the monkeys to use reward as evidence to persist with the best option when it had a mixed choice history . Specifically , if the lesioned animals started by mainly sampling the mid and worst options during the initial trials , they should subsequently be less likely to find and persist with the best option; by contrast , if they built up a choice history on the best option in the initial trials , they should then often be just as able as controls to persist with the best option . To investigate this hypothesis , we examined average proportion of V1sch choices at the start of the session ( 1st 20 trials ) and compared that against performance at the end ( last 20 trials ) of each of the 5 sessions that the animals completed on the three Fixed schedules . As we had predicted , it was not the case that the MDmc animals never managed to find and persistently select V1sch ( defined as choosing V1sch on ≥65% of the last 20 trials ) ; this ability occurred on 40% of sessions in the MDmc group ( compared to 78% of Fixed sessions in control animals ) ( Figure 7c ) . Crucially , however , this ability almost never occurred in sessions where they had failed to choose this option on the initial trials ( Figure 7c , d ) . To quantify this difference , we contrasted performance at the end of sessions where the animals had either chosen V1sch on ≤25% ( 'EARLY LOW' ) or ≥75% ( 'EARLY HIGH' ) of the first 20 trials . As can be observed in Figure 7d , there was a marked difference between the median proportion of V1sch choices at the end of EARLY LOW sessions in the two groups ( 0 . 18 V1sch choices for the MDmc group compared to 0 . 75 for controls; p<0 . 05 , Independent Samples Kolmogorov-Smirnov Test , treating each session as an independent sample ) but was overlapping in EARLY HIGH sessions ( median V1sch choices: 0 . 95 MDmc group compared to 1 . 0 for controls; p>0 . 05 ) . Together , this result demonstrates that the MDmc is not simply required to appropriately update behavior after a reversal , but instead in any situations that require the rapid integration of a reward with a recently sampled alternative to provide evidence for which of several probabilistically rewarded options to persist with . The current study sought to determine the influence of MDmc when learning and tracking probabilistic reward associations in stochastic reward environments . In the first set of experiments assessing learning and decision-making on the varying reward schedules , we found that the integrity of the MDmc is critical to allow monkeys to update their behavior efficiently following a reversal in the identity of the highest value stimulus . Similar deficits have been previously reported in studies using rats with complete MD lesions ( Block et al . , 2007; Chudasama et al . , 2001; Parnaudeau et al . , 2013 ) , an impairment often attributed to failure to prevent perseveration to a previously rewarded option or strategy , though see ( Wolff et al . , 2015 ) . However , such an explanation cannot account for the patterns of choices observed in the current study , as the monkeys with MDmc lesions were no more likely to persevere with the previously highest rewarded option post reversal than controls . In fact , what was most markedly altered in the post-reversal period in the MDmc group was the ability to reselect an option after a rewarded choice of that option . Without this faculty , the lesioned monkeys continued to show maladaptive switching between all three alternatives throughout the post-reversal period and never learned to persist with the new best option ( Figure 4 ) . It is not the case , however , that MDmc is simply required whenever there is a need to learn from positive outcomes or to respond on the basis of stimulus identity . The lesioned monkeys were not reliably different from controls in the initial acquisition stage of the varying schedules , despite the use of stochastic reward associations and novel stimuli for each testing session ( Figure 3 ) . In other studies , similar results of no deficits during acquisition have also been observed . For example , in rodents , complete removal of MD leaves acquisition of serial 2-object visual discrimination learning or 2-choice conditional learning intact ( Chudasama et al . , 2001; Cross et al . , 2012 ) . Further , in other studies , monkeys with MD or MDmc lesions could acquire concurrent object discriminations when presented across sessions ( Aggleton and Mishkin , 1983; Browning et al . , 2015; Mitchell et al . , 2007b ) and could implement a learned decision strategy ( Mitchell et al . , 2007a ) . Equally , however , it was not that the MDmc is only required to perform appropriately when contingencies reverse . For example , during the Fixed schedules in the current study , the MDmc lesioned monkeys also had a reduced ability to find and persevere with the best option compared to the control monkeys in spite of the fact that the identity of the stimulus associated with the highest reward probability never changed in a session . In the Fixed schedule sessions , the value difference between the options is not substantial and selection of any of the three options could be rewarded , probabilistically . At first glance , the pattern of results looks very similar to those reported following lesions of the OFC in monkeys ( Walton et al . , 2010 ) , a region heavily interconnected with the MDmc . In that study , the OFC-lesioned monkeys also were initially able to learn and track the value of the best option , but were severely impaired when updating their responses after the identity of the highest value option reversed . OFC-lesioned monkeys also showed deficits on certain fixed schedules . Such a finding might be expected given that the MDmc is the part of MD with major reciprocal connections to the OFC . Indeed , recent behavioral evidence in rodents and monkeys has highlighted that the MD thalamus and cortex work as active partners in cognitive functions ( Browning et al . , 2015; Cross et al . , 2012; Parnaudeau et al . , 2013; 2015 ) . Nonetheless , our analyses suggest that the two regions play dissociable , though complementary , roles during value-guided learning and adaptive decision-making . The OFC impairment resulted from the loss of an ability to favor associations based on each choice and its contingent outcome , rather than ones based on non-contingent associations between recent history of all choices and all outcomes . This caused a paradoxical pattern of choice behavior such that the OFC-lesioned monkeys became more likely to reselect an option that had been chosen often in the past even if they had just received a reward for selecting an alternative . By contrast , there was no choice history effect in the MDmc lesioned monkeys . In fact , after a recent switch , these monkeys showed no bias towards either the just rewarded option or the alternative that had been chosen in the recent past , and instead were more likely to sample the 3rd option on the subsequent trial ( Figure 5a–c ) . This difference in patterns of responding between monkeys with OFC or MDmc damage was also evident in the logistic regression analysis looking at the conjoint influence of the past 5 choices and rewards . Monkeys without an OFC had a selective reduction in the influence of past choice-outcome pairings ( Walton et al . , 2010 ) . In contrast , the MDmc group had a particular loss in the weight assigned to the most recent past choices ( n-1 to n-3 ) and the last outcome , but no statistically reliable change across the past trials of precise paired associations between each choice and each outcome . This selective impairment meant that once the monkeys with MDmc damage had an extended choice history on one option ( for instance , as occurred on certain sessions at the start of the Fixed schedules: Figure 7c ) , they were just as able as control monkeys to use the outcomes gained from their choices to guide their future behavior . This ability could be seen when examining the monkeys’ likelihood of reselecting a stimulus as a function of the number of times that option had been chosen in the past 5 trials ( Figure 5d ) . While the controls and MDmc monkeys before surgery exhibited the expected bias to repeat a rewarded choice irrespective of recent history , post surgery the MDmc group only displayed this pattern if they had selected that option on multiple occasions within the recent past . This selective impairment in attributing reward to recent choices , accompanied by the sparing of a faculty to approximate associations based on histories of choices and rewards , is consistent with theories that emphasize the importance of MDmc ( and OFC ) in goal-directed learning , which requires acquisition of specific future reward predictions of a choice , but not habit learning that relies on longer term trends in choices and outcomes ( Ostlund and Balleine , 2008; Bradfield et al . , 2013; Parnaudeau et al . , 2015 ) . Taken together , this study implies that a primary function of MDmc is to support the representation of recent stimulus choices to facilitate rapid reward-guided learning and adaptive choice behavior . This function would play a similar role to an eligibility trace in reinforcement learning models , which is essentially a temporary record of recent events used to facilitate learning ( Lee et al . , 2012; Sutton and Barto , 1998 ) . Several studies have suggested that MD might be particularly important during rapid task acquisition rather than performance based on previously acquired associations ( Mitchell et al . , 2007a; 2007b; Mitchell and Gaffan , 2008; Mitchell , 2015; Ostlund and Balleine , 2008; Ouhaz et al . , 2015 ) . Such a role is also consistent with the electrophysiology finding that some cells in monkey MDmc , as well as in more lateral parvocellular MD , are modulated both when making cue-guided actions and when receiving feedback post-response ( Watanabe and Funahashi , 2004 ) . The ability to keep track of recent stimulus choices , and their predicted values , is of particular importance when monkeys are sampling alternatives in order to determine the values associated with different objects . At such times , an online representation , or 'hypothesis' , of what alternatives might be worth sampling would allow rapid updating if their selection leads to a beneficial outcome . Therefore , the MDmc might be described as being critical to facilitate an appropriate balance between exploration and exploitation . However , rather than computing when and what to explore in order to gain valuable new information , functions ascribed to areas such as frontopolar and anterior cingulate cortex and their projecting neuromodulators ( Boorman et al . , 2009; Donahue et al . , 2013; Frank et al . , 2009 ) , the role of the MDmc might instead be to help facilitate re-selection and persistence with a beneficial option once it has been found . In line with this idea , it was notable that one striking effect of the MDmc lesions was that , on top of a general speeding in response latencies , the lesioned monkeys also no longer exhibited a characteristic retardation in latencies on trials where they switch to an alternative compared to when they persisted with the same choice . Taken together with the MDmc lesioned monkeys’ increased tendency to sample all three options during exploration , this evidence implies that MDmc is required to exert rapid regulation of stimulus-based choices , particularly when needing to decide when to stop searching and instead persist with a recently sampled optimal option . Subjects were ten rhesus monkeys ( Macaca mulatta; all males ) aged between 4 and 10 years . After preoperative testing , three monkeys received bilateral neurotoxic ( NMDA/ibotenic acid ) injections under general anesthesia using aseptic neurosurgical conditions ( see Surgery details below ) to MDmc whereas the rest remained as unoperated controls . Four of these controls were tested alongside the lesioned monkeys . For analysis , the data from these controls were combined with data from three monkeys that were used as unoperated controls in a previously published study using comparable training and identical testing protocols ( Walton et al . , 2010 ) . When the performance of these earlier unoperated controls were compared to the four monkeys tested alongside the lesioned monkeys , they were comparable in performance on all measures , with the exception that the control monkeys from the earlier study selectively made more choices of V1sch in the second half of the Stable schedule ( see Figure 1B; testing group x condition x session period interaction: F1 , 7 = 10 . 14 , p=0 . 015 ) . Note , however , that the critical statistical tests in this study determine changes between the pre- and post-operative testing sessions . All experimental procedures were performed in compliance with the United Kingdom Animals ( Scientific Procedures ) Act of 1986 . A Home Office ( UK ) Project License ( PPL 30/2678 ) obtained after review by the University of Oxford Animal Care and Ethical Review committee licensed all procedures . The monkeys were socially housed together in same sex groups of between two and six monkeys . The housing and husbandry were in compliance with the guidelines of the European Directive ( 2010/63/EU ) for the care and use of laboratory animals . The computer-controlled test apparatus was identical to that previously described ( Mitchell et al . , 2007b ) . Briefly , monkeys sat in a transport box fixed to the front of a large touch-sensitive colormonitor that displayed the visual stimuli for all of the experiments . Monkeys reached out through the bars of the transport box to respond on the touchscreen and collect their food reward pellets from a hopper that were automatically dispensed by the computer . Monkeys were monitored remotely via closed circuit cameras and display monitors throughout the testing period . Prior to the start of the experiments reported here , all monkeys had been trained to use the touchscreens and were experienced at selecting objects on the touchscreen for rewards . On each testing session , monkeys were presented with three novel colorful stimuli , ( 650 x 650 mm ) , which they had never previously encountered , assigned to one of the three options ( A–C ) . Stimuli could be presented in one of four spatial configurations ( see Figure 1A ) and each stimulus could occupy any of the three positions specified by the configuration . Configuration and stimulus position was determined randomly on each trial meaning that monkeys were required to use stimulus identity rather than action- or spatial-based values to guide their choices . A task programme using Turbo Pascal controlled stimulus presentation , experimental contingencies , and reward delivery . Reward was delivered stochastically on each option according to predefined schedules . Data are reported from two varying schedules ( ‘Stable’ and ‘Variable’ ) and three Fixed schedules ( Figures 1b , 7a ) . The monkeys were also tested on several additional varying 3-option schedules , the data from which are not reported here . The likelihood of reward for any option , and for V1sch ( the objectively highest value stimulus available ) and V1RL ( the subjectively highest value stimulus given the monkeys’ choices as derived using a standard Rescola-Wagner learning model with a Boltzmann action selection rule ) was calculated using a moving 20 trial window ( ±10 trials ) . Whether reward was or was not delivered for selecting one option was entirely independent of the other two alternatives . Available rewards on unchosen alternatives were not held over for subsequent trials . Each animal completed five sessions under each schedule , tested on different days with novel stimuli each time . For the two varying schedules , the sessions were interleaved and data were collected both pre- and postoperatively . For the fixed conditions , the three schedules ( Figure 7a ) were run as consecutive sessions , starting with the five sessions of Fixed 1 ( Figure 7a , left panel ) , then five sessions of Fixed 2 ( Figure 7a , middle panel ) , and finally five sessions of Fixed 3 ( Figure 7a , right panel ) . In all cases for the reported Fixed schedules , only postoperative data were collected and data acquisition occurred after completion of testing on the varying schedules ( note , the animals had performed some other Fixed schedules pre-surgery so had experience of sessions without stimulus reversals ) . One control monkey was unable to be run on these fixed schedules . The varying schedules comprised of 300 trials per session and the fixed schedules of 150 trials per session . The data from the varying schedules were analyzed both as a function of V1sch and of V1RL . For the latter , a learning rate was fitted individually to each animal’s pre-surgery data using standard nonlinear minimization procedures and used for analysis of both pre- and post-operative data . Where appropriate , data from all tasks are reported using parametric repeated-measures ANOVA . The regression analyses were analogous to those described in Walton et al . ( 2010 ) . In brief , to establish the contribution of choices recently made and rewards recently received on subsequent choices , we performed three separate logistic regression analyses , one for each potential stimulus ( A , B , C ) . For each individual regression , the stimulus in question ( e . g . , ‘A’ ) would take the value of 1 whenever chosen and 0 whenever one of the other two stimuli ( e . g . , ‘B’ or ‘C’ ) was chosen . We then formed explanatory variables ( EVs ) based on all possible combinations of recent past choices and recent past rewards ( trials n-1 , n-2 , … , n-6 ) . Each EV took the value of 1 when , for the particular choice-outcome interaction , the monkey chosen A and was rewarded , –1 when the monkey chose B or C and was rewarded , and the 0 when there was no reward . We then fit a standard logistic regression with these 36 EVs ( 25 EVs of interest and 11 additional confound regressors describing combinations of choice / outcome n-6 ) . This gave us estimates of β^A and C^A . We then repeated this process for the other two stimuli . This gave us three sets of regression weights , β^A , β^B , β^C and three sets of covariances , C^A , C^B , C^C . We proceeded to combine the regression weights into a single weight vector using the variance-weighted mean:β^= ( C^A−1+C^B−1+C^C−1 ) −1 ( C^A−1β^A+C^B−1β^B+C^C−1β^C ) Neurosurgical procedures were performed in a dedicated operating theatre under aseptic conditions and aided by an operating microscope . Steroids ( methylprednisolone , 20 mg/kg ) were given the night before surgery intramuscularly ( i . m . ) , and 4 doses were given 4–6 hr apart ( intravenously [i . v . ] or i . m . ) on the day of surgery to protect against intraoperative edema and postoperative inflammation . Each monkey was sedated on the morning of surgery with both ketamine ( 10 mg/kg ) and xylazine ( 0 . 25–0 . 5 mg/kg , i . m . ) . Once sedated , the monkey was given atropine ( 0 . 05 mg/kg , i . m . ) to reduce secretion , antibiotic ( amoxicillin , 8 . 75 mg/kg ) as prophylaxis against infection , opioid ( buprenorphine 0 . 01 mg/kg , repeated twice at 4- to 6-hr intervals on the day of surgery , i . v . or i . m . ) and nonsteroidal anti-inflammatory ( meloxicam , 0 . 2 mg/kg , i . v . ) agents for analgesia , and an H2 receptor antagonist ( ranitidine , 1 mg/kg , i . v . ) to protect against gastric ulceration as a side effect of the combination of steroid and non-steroidal anti-inflammatory treatment . The head was shaved and an intravenous cannula put in place for intraoperative delivery of fluids ( warmed sterile saline drip , 5 ml/h/kg ) . The monkey was moved into the operating theatre , intubated , placed on sevoflurane anesthesia ( 1–4% , to effect , in 100% oxygen ) , and then mechanically ventilated . A hot air blower ( Bair Hugger ) allowed maintenance of normal body temperature during surgery . Heart rate , oxygen saturation of hemoglobin , mean arterial blood pressure , and tidal CO2 , body temperature , and respiration rate were monitored continuously throughout the surgery . After completion of all behavioral testing , each monkey was sedated with ketamine ( 10 mg/kg ) , deeply anesthetized with intravenous barbiturate and transcardially perfused with 0 . 9% saline followed by 10% formalin . The brains were extracted and cryoprotected in formalin-sucrose and then sectioned coronally on a freezing microtome at 50 μm thickness . A 1-in-10 series of sections was collected throughout the cerebrum that was expanded to a 1-in-5 series throughout the thalamus . All sections were mounted on gelatin-coated glass microscope slides and stained with cresyl violet .	A small structure deep inside the brain , called the mediodorsal thalamus , is a critical part of a brain network that is important for learning new information and making decisions . However , the exact role of this brain area is still not understood , and there is little evidence showing that this area is actually needed to make the best choices . To explore the role of this area further , Chakraborty et al . trained macaque monkeys to choose between three colorful objects displayed on a touchscreen that was controlled by a computer . Some of their choices resulted in the monkeys getting a tasty food pellet as a reward . However the probability of receiving a reward changed during testing , and in some cases , reversed , meaning that the highest rewarded object was no longer rewarded when chosen and vice versa . While at first the monkeys did not know which choice was the right one , they quickly learned and changed their choices during the test according to which option resulted in them receiving the most reward . Next , the mediodorsal thalamus in each monkey was damaged and the tests were repeated . Previous research had suggested that such damage might result in animals repeatedly choosing the same option , even though it is clearly the wrong choice . However , Chakraborty et al . showed that it is not as simple as that . Instead monkeys with damage to the mediodorsal thalamus could make different choices but they struggled to use information from their most recent choices to best guide their future behavior . Specifically , the pattern of the monkeys’ choices suggests that the mediodorsal thalamus helps to quickly link recent choices that resulted in a reward in order to allow an individual to choose the best option as their next choice . Further studies are now needed to understand the messages that are relayed between the mediodorsal thalamus and interconnected areas during this rapid linking of recent choices , rewards and upcoming decisions . This will help reveal how these brain areas support normal thought processes and how these processes might be altered in mental health disorders involving learning information and making decisions .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Critical role for the mediodorsal thalamus in permitting rapid reward-guided updating in stochastic reward environments
Nutrient uptake by roots often involves substrate-dependent regulated nutrient transporters . For robust uptake , the system requires a regulatory circuit within cells and a collective , coordinated behaviour across the tissue . A paradigm for such systems is boron uptake , known for its directional transport and homeostasis , as boron is essential for plant growth but toxic at high concentrations . In Arabidopsis thaliana , boron uptake occurs via diffusion facilitators ( NIPs ) and exporters ( BORs ) , each presenting distinct polarity . Intriguingly , although boron soil concentrations are homogenous and stable , both transporters manifest strikingly swift boron-dependent regulation . Through mathematical modelling , we demonstrate that slower regulation of these transporters leads to physiologically detrimental oscillatory behaviour . Cells become periodically exposed to potentially cytotoxic boron levels , and nutrient throughput to the xylem becomes hampered . We conclude that , while maintaining homeostasis , swift transporter regulation within a polarised tissue context is critical to prevent intrinsic traffic-jam like behaviour of nutrient flow . In Arabidopsis thaliana , boron is shuttled from the soil across the epidermis , cortex and endodermis into the stele , to finally be taken up into the xylem , mediated by the action of secondary active boron exporters ( BORs ) and the enhancement of permeability through NIPs . Previously , we analysed boron patterning by using a two-dimensional cross-sectional model of the entire Arabidopsis root meristem , considering all spatial nuances of transporter localisation , polarity and intensity , whilst neglecting transporter dynamics and regulation ( Shimotohno et al . , 2015-04 ) . Instead , transporter levels and activity were static , fixed to the observed steady state transporter expression under constant boron medium condition of 0 . 3 μM . While neglecting regulation of transporter expression or activity , a characteristic boron pattern emerged on the level of the root , presenting highest concentrations at the stem cell niche . This boron profile gradually decreased longitudinally up to the start of the differentiated tissues . The computationally simulated characteristic boron distribution , which we confirmed experimentally through LA-ICP-MS , strongly indicates that the mature root tissue is involved mainly in transporting boron from the soil into the xylem , while the distal meristematic tissues have a differential boron transport function , namely , to locally provide higher boron levels for fuelling local growth ( i . e . , cell wall material ) at those regions . In that study , however , we did not address the dynamics of the transporter regulation nor how nutrient homeostasis is achieved . Here , we investigate the dynamics of the transporters and their mutual feedback with the generated boron distribution . We do so by spatially focussing on the differentiated tissue region involved in xylem-uptake . At the elongation and differentiation zone of the root , transporters possess a striking polar expression; BORs are located at the inside-facing membranes of the cells , while NIPs are concentrated to the outside facets ( see Figure 1A , B ) . Although protein levels can present strong variation between cell files , the transcription of the transporters takes place throughout the entire root tissue , even for NIP5;1 ( Figure 1—figure supplement 1 ) . We therefore use the simplifying assumption that all cell files are intrinsically the same in respect to the potential of expressing the transporters , and variations arise solely as a consequence of the nutrient distributions . Our analysis focuses on the transversal nutrient flow through the root that results from boron entering and leaving the different cell files transversally . Effectively , in our simple model , we only consider a single row of cells and , for simplicity , only four cell files over which boron is transported from the soil into the xylem ( Figure 1C ) . To capture the dynamics of boron transport and transporter activities in root tissue , the model’s variables are the boron concentration in cells ( C ) and cell walls ( W ) , and transporter activities of NIPs ( N ) and BORs ( B ) for each individual cell ( n ) . The mutual dependency between these variables is described using ordinary differential equations ( ODEs ) . For each cell file of the model , the transporters are produced at an equal rate , have the same transport potential , and are regulated in the same way . The modelled cells loosely map to the outermost epidermis ( cell 1 ) , the cortex ( cell 2 ) , the endodermis ( cell 3 ) , and finally to the pericycle/vasculature ( cell 4 ) . The last cell of the row is endowed with an upward convective flow , which captures the shootwards convective flow of the xylem . For reasons of simplicity , we consider these four cells to have equal dimensions , and make the same assumption for all intermediate cell walls ( Figure 1C ) . Given that transporter regulation can be observed in response to varying the boron conditions in the medium ( Takano et al . , 2005 , Takano et al . , 2006-06Takano et al . , 2006-06 , 2010; Miwa et al . , 2013 ) , ( Figure 1—figure supplement 2 ) , boron sensing underpinning those responses could either involve measurements of the intracellular concentration or involve measurements of the intercellular concentration ( i . e . , the levels within the apoplast ) . In the case of NIP5;1 , we recently revealed that NIP5;1 mRNA degradation is triggered by boron-dependent ribosome stalling during the translation process ( Tanaka et al . , 2016 ) . The boron-dependent degradation was also observed in an in vitro system without cell wall fraction , which suggests that the sensing mechanism depends on the cytoplasmic boron concentration . For the BORs the precise subcellular location of the boron sensing has not yet been determined , but it is reasonable to assume that the regulation is also responding to cytosolic boron levels , given that the purpose of the nutrient homeostasis is to keep the nutrient levels within the cell within bounds , and to this end directly sensing the cytoplasmic boron concentration is beneficial . In short , given the current knowledge , cytosolic boron sensing is the most parsimonious assumption . Hence , our model assumes that all the boron transporters of a cell are regulated by the cytosolic boron concentration within that respective cell . Note that we loosely use the term ‘boron’ in our model to refer to both boric acid as well as borate . They can be found in a dynamic chemical equilibrium , B ( OH ) 3+H2O⟺B ( OH ) 4−+H+ , with a pK value of 9 . 24 ( Woods , 1996 ) . Given that this process does not involve any enzyme kinetics , this distinction is not explicitly considered in our model . Moreover , we solely focus on the import of soluble boron ( both boric acid and borate ) , not considering any boron that is bound to the cell wall . The temporal dynamics of the soluble boron concentration within the cytosol ( Ci ) and in the cell walls ( Wi , being the cell wall adjacent to the inward facing facet of cell i ) is determined by their inflow and outflow from and to neighbouring compartments: ( 1 ) Ci˙={ ( p+Ni ) ⁢ ( Wi-1-Ci ) -Bi⁢Ci+p⁢ ( Wi-Ci ) }⁢1l⁢c , ( 2 ) Wi˙={p ( Ci−Wi ) + ( p+Ni+1 ) ( Ci+1−Wi ) +BiCi}1lw . Here , Ni and Bi represent boron permeabilities due to , respectively , the bi-directional transport by NIPs and the directional transport by BORs within cell i . Transporter-independent boron permeability through the plasma-membrane is also taken into account , described by parameter p . To capture the tissue context , boundary conditions are set at both extremities of the transversal cell series . The outer cell wall of the first cell ( cell 1 ) is in contact with medium . Given the rapid diffusion rate of boric acid in water , we assume that the boron concentration in the outermost cell wall ( W0 ) is the same as in the medium , constant over time . The xylem transport that occurs from roots to upper tissue is represented by a removal term , with rate a , attributed to the innermost cell ( i=n ) of the cell row: ( 3 ) Cn˙= ( p+Nn ) ⁢ ( Wn-1-Cn ) ⁢1l⁢c-a⁢Cn⁢1h⁢c . Assuming that transport activities are proportional to protein concentration , it follows that the dynamical regulation of NIP and BOR permeability ( Ni and Bi ) are in direct proportion to the dynamics of their proteins ( we will later show that our main insights do not depend on the assumption that the BOR transporter does not saturate under the range of concentrations here treated ) . It was therefore not necessary to introduce explicit variables for protein concentrations . Instead we directly use permeability , representing the protein concentrations and their regulation . The protein concentration , and thus the effective permeability , is determined by production and degradation rates . As mentioned , accumulation of NIP5;1 mRNA is regulated through boron-dependent mRNA degradation . This suggest that the production of NIP5;1 protein is boron dependent . On the other hand , degradation of NIP5;1 protein is not found to be boron dependent ( Takano et al . , 2010; Tanaka et al . , 2016 ) . Thus , the levels of Ni vary due to a production term that is dependent on cytosolic boron concentration and a constant degradation term: ( 4 ) Ni˙=αN⁢kNnNkNnN+CinN-ξN⁢Ni , where αN is the production rate , ξN is degradation rate , and nN is the Hill coefficient capturing the boron-dependent inhibitory regulation . Accordingly , given that BOR protein levels are regulated through boron-dependent degradation ( Takano et al . , 2005 ) , but there is no evidence of boron dependent production , Bi varies over time due to constant production and boron-dependent degradation terms: ( 5 ) Bi˙=αB-ξB⁢ ( 1+dB⁢CinBkBnB+CinB ) ⁢Bi , where αB is the production rate , ξB the basal degradation rate , dB the maximum boron-independent degradation rate , and nB the Hill coefficient capturing the boron-dependent inhibitory regulation . We wish to use this model to understand the significance , if any , of the swift transporter regulation . To assess the effects of regulation swiftness on transport , we therefore introduce a time-scaling factor ϵ , which multiplies both the entire Ni˙ and Bi˙ terms ( Equations 6 and 7 ) . To align ourselves with the current experimental evidence that indicates swift regulation of NIP5;1 on the mRNA level and of BORs on the protein level , we modulate the production rate and degradation rate for NIP5;1 and BOR1 , respectively . However , changing production rates of NIPs through a factor ϵ without changing the degradation rate would result in changes of the overall protein concentration which might influence the system’s behaviour , not due to the swiftness of the regulation but due to the changes in equilibrium protein concentration . This also holds for the BOR regulation . Thus , to avoid such possible undesired effects , the time-scaling factor multiplies the entire right-hand side of the dynamics: ( 6 ) Ni˙=ϵN⁢ ( αN⁢kNnNkNnN+CinN-ξN⁢Ni ) , ( 7 ) Bi˙=ϵB⁢{αB-ξB⁢ ( 1+dB⁢CinBkBnB+CinB ) ⁢Bi} . We define ϵ=1 as representing the swiftness of regulation in the wild type scenario , while reducing ϵ allows us to simulate behaviours that would occur with less rapid regulatory dynamics , and increasing ϵ allows us to consider even higher swiftness of regulation than experimentally observed . Note that the regulation of the transporter activity by cytosolic boron concentrations occurs through a sigmoidal relationship , in which kB and kN are the boron concentrations at the half maximum of the Hill function , defining the sensitivity of the response to the boron concentration . We set kB≃30kN , consistent with our observations that the sensitivity of BOR expression is much less than that of NIP5:1 ( Figure 1–Figure supplement 2 ) , but will also show that our results do not require large differences between these two parameters . We have assessed what is the biologically acceptable range for each parameter , based on the available literature ( see Material and methods for details ) . Default parameter values were chosen to lie within these valid ranges , and are given in Table 1 . When regulation swiftness lies within the experimentally observed regime ( ϵB=ϵN=ϵ=1 ) , a constant flow of boron that is transported across the root ensues ( Figure 2 ) . Boron concentrations within the cells ( Cn ) remain constant over time . This steady state behaviour , however , is disrupted when the transporter regulation swiftness is decreased ten-fold ( ϵ=0 . 1 ) . Oscillations in the boron concentration arise in all cells , but most strikingly in the endodermis ( cell 3 ) and cortex ( cell 2 ) ( Figure 2A ) , as well as in the cell walls ( Figure 2B ) . Decreasing the transporters’ regulation swiftness even further ( ϵ=0 . 01 , 0 . 001 ) consistently enhances the amplitude of such oscillations , and enlarges their periods ( Figure 2A , B ) . We conclude that even when environmental boron levels ( C0 ) and xylem activity ( a ) remain constant , simply deviating from the rapid experimentally observed transporter’s regulation swiftness is sufficient to push the boron transport system into an oscillatory dynamics . The reason the cells undergo such dramatic concentration variations at lower regulation swiftness can be intuitively linked to the accompanying changes in transporter levels which also ensue ( Figure 2C , D ) . They are initially triggered by minute fluctuations in the concentrations , but then cause increasingly larger changes in cytosolic and apoplastic concentrations . It is not immediately clear , however , how it depends on either the NIP or BOR transporter regulation , why regulation swiftness exacerbates oscillation amplitudes and how their interdependencies either produce stable flows or unstable oscillations that propagate throughout the tissue . What we can already conclude though , is that within a polarised tissue rapid transport dynamics are necessary to ensure flux homeostasis and avoid instabilities in concentration levels . But to fully understand the process we will first observe how in the model the steady flow regime behaves in terms of transporter expression levels when we assume that the regulatory dynamics are as swift as experimentally observed . Given that our question is focused on the possible advantages of swift transporter regulation within a polarised tissue context , our model purposely ignored details of tissue-specific regulation , using the simplifying assumption that all cells are endowed with the same potential of expressing transporters . This assumption may initially seem at odds with the actual distinct GFP levels of the GFP-tagged proteins ( Figure 1B ) , which qualitatively indicates stronger BOR1 expression in the epidermis and endodermal files and pronounced NIP5;1 expression in the outermost cell file ( lateral root cap or epidermis ) only . It was therefore interesting to observe that our parsimonious model could already account for general features of the the observed transporters’ expression patterns ( Figure 2C , D ) solely as a consequence of the relative positioning of the cells within the cell row: in the model , NIP only becomes highly expressed in the outermost cell , the epidermis ( N1 , representing NIP5;1 in cell 1 , is by far the highest variable in the steady flow case ) . These relative levels are in qualitative agreement with the experimental findings regarding NIP5;1 . Moreover , in contrast to this strong expression in the epidermis , in the model NIPs are present at much lower amounts in all the other inner cell files ( cell 2 , cell 3 and cell 4 ) , again qualitatively similar to the experimental observations . These model results can be understood as follows . Although all in silico cells share the same intrinsic properties , as a consequence of NIPs acting as permeability facilitators rather than directional transporters , it is impossible for the cell file in contact with the medium ( cell 1 , the outermost cell file , that is , epidermal cell ) to reach higher concentrations than the level in the medium . Only the next cell inwards , cell 2 , is capable of reaching higher levels , due to the directed polar action of the BORs . The inevitable low boron levels in the epidermis give rise to the observed lack of NIP downregulation within the first cell file , and hence to its preferential expression . The expression patterns of the active exporters , BORs , are quite different from NIP5;1 , but again there are similarities between the model results ( B values ) and confocal images , with BORs distinctly expressed in the epidermis ( cell 1 ) at the highest levels , in the cortex ( cell 2 ) at high levels , whilst in the endodermis it is relatively weaker ( cell 3 ) ( Figure 1B ) . The model’s qualitative correspondence between the transporter expression patterns and those experimentally observed suggests that concentration-dependent regulation is at least sufficient for such expression patterns to arise . Corroborating this , GUS expression under the NIP5;1 promoter is indeed present in the inner root tissues , albeit the NIP5;1 protein levels in those cells are very low , supporting the notion that these files are likely responsive and capable of expressing NIP5;1 , but inhibited to do so by the boron levels that arise due to the cells’ positioning within the tissue ( Figure 1—figure supplement 1 ) . However , these results , while suggesting sufficiency of our parsimonious assumptions in relation to observed transporter levels , do not offer evidence that tissue-specific regulation is not occurring . At most , the qualitative matches justify , at first instance , that the model is kept simple , without the need of introducing ad hoc tissue-specific dependencies , when exploring the effects transporter regulation dynamics . Furthermore , these results indicate that under our experimentally observed parameter values , the system presents steady nutrient throughput and constant cellular concentrations and transporter levels over time . We next investigated more carefully the spatial-temporal characteristics of the oscillatory behaviour that arises when transporter swiftness is decreased ( Figure 3A ) . Kymographs make it visually clear that oscillations in the individual cells are in fact spatially coupled , manifested as a boron wave that propagates backwards over the tissue , that is , the pulse moves in the direction contrary to the nutrient flow itself . The initiation of the instability , as seen by the first peak in concentration , occurs close to the xylem ( i . e . , the last cell , here , cell 4 ) , followed by a strong rise in endodermal concentrations ( in cell 3 ) around 90 min later . Due to boron’s inhibitory influence on the transporters , this intracellular rise leads to a decrease in both NIPs and BORs within that cell . This transporter downregulation causes a drop in boron throughput across that cell , leading to an accumulation of boron in the adjacent , outward facing , cell wall . Background permeability rates along the plasma membrane allow this apoplastic rise in concentrations to trigger an increase boron concentrations in the next outermost neighbouring cell , in this case , the cortex cell ( cell 2 ) . Cortex concentration levels thus rise , again triggering a shutdown of transporters , causing the same process to ensue in a spatially coupled manner in the outermost cell , the epidermis . In the epidermis , however , boron levels can never rise above the soil concentration levels , as discussed previously , such that the backward travelling wave loses its amplitude and terminates . The overall process is observed to be cyclical , with levels rising again close to the xylem , until a next wave is triggered in the innermost cells ( Video 2 ) . Moreover , we found that such traffic-jam-like behaviour is robust when smaller differences between kB and kN are considered ( Figure 2—figure supplement 1A–C ) , as long as kB is sufficiently large as not to too strongly interrupt boron throughput through the tissue altogether ( Figure 2—figure supplement 1D ) . Obviously , if kB is too low , and traffic-jam-like behaviour are prevented , the plant would also not be able to take up boron , thus , this would be a parameter setting that is biologically irrelevant . Therefore , although our interpretation of the experimental results ( Figure 1—figure supplement 2 ) suggest larger differences between kB and kN , given that these are indirect estimates that rely on underlying assumptions , this robustness analysis shows that , within a biologically reasonable window , our results hold even when kB were to be smaller . We next wondered if this traffic-jam-like effect dissipates over larger tissues . Our four-cell model is a simplified representation of A . thaliana , which has less cell files between the epidermis and the xylem than most other roots of experimental plant models . However , if one distinguishes the pericycle from the vasculature , a five-cell model would be more appropriate . We show that the variables of the five-cell model present similar dynamics as our default model when BOR transporter regulation is lowered to 400 μM ( Figure 4—figure supplement 2A–D ) . Lowering kB slightly stabilises the dynamics ( as shown in Figure 2—figure supplement 1 ) , which counteracts increasing destabilisation of the constant flow regime when the number of cell files in the tissue increases . Conversely , given that this renders smaller tissues ( such as our default four-cell model ) , more robust against traffic-jam-like behaviour , our default setting should be regarded as a worse-case scenario for such dynamics to occur , rather than being a special case . To further appreciate tissue-size dependencies , and explore the generality of these effects for other plants which might have highly deviating cell file numbers , we gradually increased the in silico tissue from 4 to 10 cell files . Figure 3B–G shows that the phenomenon is conserved irrespective of the number of cells between the medium and the convective xylem flux . Larger transversal tissue segments are able to generate higher wave amplitudes , but in all cases the peak in boron propagates contrary to the nutrient flow with approximately the same speed . The same phenomenon was still observed even when the model contained 100 cell files ( Figure 3—figure supplement 1 ) , suggesting that large tissues do not prevent these effects from occurring , but rather increase its likelihood . In short , when transporter regulation swiftness is sufficiently slow , due to the downregulation of transporters under the control of increased boron concentrations , a traffic jam-like behaviour emerges: a high boron concentration peak appears , correlating with locally lower transport rates , triggering a low-transport high-boron wave that back-propagates from cell to cell in the direction contrary to the transport polarisation direction within the tissue . This occurs independently of the size of the plant tissue under consideration . We next studied the importance of the relative dynamics of BOR and NIP regulation to trigger these phenomena , that is , which of the processes have to be sufficiently slow . Under what conditions and parameter values for transporter regulation rates does the traffic jam phenomenon manifest itself ? In our previous simulations , we simultaneously decreased the swiftness of both the NIP and the BOR regulation . Oscillations in boron concentration then arose due to a change in stability of both the flow and steady state of boron concentrations . We next probed the system’s stability while varying the regulation swiftness of the NIPs and BORs independently . This was done by analysing the equilibrium of the simplified , A . thaliana inspired , four-cell model . We linearised the ODEs around the equilibrium using a Taylor expansion , and then evaluated the stability of the equilibrium in terms of the largest eigenvalue , for 40 , 000 different parameter combinations . The phase-portrait that emerges ( Figure 4A ) shows a large region in which the system becomes unstable and oscillations arise ( indicated in red ) , as well as a region in which sufficient regulation swiftness ensures that the system is stable and oscillations do not develop ( indicated in blue ) . For the stable ( blue ) parameter settings , any perturbation dampens out ( Figure 4B ) , whereas for parameter settings within the red region , oscillations dominate . The frequency of the oscillations is determined by a relative lack of swift regulation for both transporters , as shown by the colour map in Figure 4 . This phase diagram is qualitatively unaffected for the five-cell model at kB=400 μM ( Figure 4—figure supplement 2F ) . Furthermore , we also checked if saturation in the activity of the BOR transporter might obstruct or stabilise the system against the appearance of the oscillations . ( The NIPs , being a channel , should be less prone to undergo concentration-dependent saturation effects in permeability . ) Contrary to this , we found that such a saturation in permeability of the BOR transporter does not change the likelihood or transporter-response dependency to present traffic-jam-like behaviour , but rather , when they arise due to slow dynamics , the oscillations present much higher amplitudes in concentration variations , with peak boron levels becoming more than twice as high due to the transport saturation ( Figure 4—figure supplement 3 ) . From this combined analysis it becomes apparent that the parameter values we derived to capture wild type dynamics of transporter regulation ( i . e . , point 1 , 1 ) fall into the stable regime . Hence , the wild type system is robust against perturbations and does not present traffic jam-like behaviour . However , independent small reductions from these values in either NIP or BOR regulation rate can cause the tissue to experience oscillations . Not only does the analysis reveal that both transporters are required to present a rapid response , and that this is independent of the BOR-dependent fluxes becoming saturated , but their combined swiftness is synergistic: if one transporter is extremely fast , the other can be bit slower than what is otherwise needed . However , for any parameter setting , a certain speed in responsiveness needs to present to prevent oscillations . Traffic jams are commonly experienced by vehicle drivers in urban areas . In terms of human transport , traffic jams are an undesirable outcome as they reduce the throughput through the highway and increase the time to reach the destination . Traffic jams displace backwards in space ( contrary to the flow ) while giving rise to increased car densities at the specific location of the traffic jam ( Kesting and Treiber , 2013 ) . All those properties can also observed in our plant tissue model , namely diminished throughput of boron from the soil to the xylem and backwards propagating pulses of locally increased boron concentrations within cells . Moreover , again analogous to real-life traffic jams , traffic-jam-like behaviour in plant nutrient uptake has undesirable implications in the biological context as well . High intracellular boron is detrimental to plant health as it causes DNA damage and ultimately causes cell death ( Sakamoto et al . , 2011-09 ) . Plants are therefore likely to have evolved mechanisms to avoid high intracellular boron levels to limit boron-induced damage . Lower throughput across the root implies reduced xylem loading and hence a reduction in the boron absorption efficiency , critical for plant growth at low boron conditions . To better quantify the magnitude of these effects , we performed a simulation screen to evaluate the expected physiological impact of the transporter swiftness in the form of increased boron levels within the cells due to the instabilities , as well reduction in throughput ( see Materials and methods for details ) . We observe increasing maximum levels in boron concentration for slower regulation swiftness ( Figure 5A ) , as well as higher fold-changes in boron concentration ( Figure 5B ) . This suggests that sufficiently rapid transport regulation is important to reduce the risk of DNA damage when either considering absolute concentrations or the increase over the basal equilibrium boron concentrations . Moreover , under the conditions in which traffic jams and large-amplitude variations occur , we measured a considerable reduction in total throughput through the tissue ( Figure 5C ) . In our default model we assume the BOR transporter-driven flux to be linear with the cellular boron concentration . If we instead consider Michaelis-Menten saturated BOR transport ( using a biologically reasonable Michaelis constant — the boron concentration at which the flux becomes half as large due to saturation , as well as the concentration at which the flux reaches its half-maximum value — of cB=1000 μM ) the system’s capacity of uptake and throughput only marginally decreases . In stark contrast , the detrimental effects of the instabilities that arise due to slowing down the regulatory dynamics are amplified when BOR saturation is considered ( see details of saturation implementation in the caption of Figure 4—figure supplement 3 ) . The impact of the traffic jams becomes much more severe , with peak levels during the concentration fluctuations more than twice as high as the default scenario ( compare Figure 2 to Figure 4—figure supplement 3 ) . This is a direct consequence of the cell not being able to effectively efflux boron when intracellular levels spike . Taken all into consideration , we infer that selective pressures operating on the root’s capacity to absorb boron optimally and robustly could result in the system evolving to a regime of rapid regulation of transporters . Our model thus far consisted of a segment of polarised tissue , of a variable number of cell files , linking the soil to the xylem via a polar flow of nutrients . Consequently , both soil and xylem , although presenting stable characteristics over time , do constitute abrupt boundary conditions: the first sets a constant medium concentration , and the second presents a constant convective flow away from the transversal section . We thus questioned if these boundary conditions are responsible for triggering the traffic jam-like regime that arises under lower rates of transport gene regulation . During the last decades a similar discussion has been taking place regarding what triggers the analogous traffic jams in discrete macroscopic systems such as road networks ( Kesting and Treiber , 2013 ) . Although traffic jams are readily triggered by bottlenecks ( such as road obstructions , roundabouts , etc . ) , theoretical models of traffic jam behaviour predicted that the collective systems parameters alone could be sufficient to render the free flow state unstable . It depends on the car density exceeding a critical value as well as how speed relates to the local car density . To prove this theoretical insight , Sugiyama et al . , 2008 developed an experiment in which cars were confined to move on a homogeneous circular road . Above a critical car density they observed a transition from a free flow to a traffic jam state , due to the collective effect of the vehicles . They concluded that , although no bottleneck was present , the intrinsic parameters of the system rendered it unstable , such that the smallest of naturally occurring fluctuations was enhanced to disrupt the free flow . This experiment illustrates the non-intuitive notion of traffic jams being generated spontaneously under certain critical parameter regimes . Inspired by this strategy , and for us to be able to rule out that the origin of the nutrient-flow instabilities within the plant tissue is provided by the soil boron concentration or the xylem flow , we constructed an in silico polarised tissue composed of 6 cells , wrapped up into a ring , that is , the first cell is connected to the last . The ring structure avoids boundary conditions ( i . e . , bottlenecks ) and offers a scenario in which uninterrupted flows ( either stable or unstable ) can be studied . We allow the dynamics of the polarised tissue to evolve from an initial situation in which concentrations are homogeneous over the whole ring-tissue . When transporter regulation is as swift as in wild type ( Figure 6A ) , a steady flow with no oscillation in the concentrations is observed . When transporter regulation swiftness is decreased to sufficiently low values as to cause unstable flows ( as derived from the analysis shown in Figure 4 ) , oscillations arise . After an initial period in which the flows seem constant , random numerical fluctuations in the simulation bring forth a rise of boron concentration in a random cell , which , due to the mechanism of inhibition by boron of the boron transporters , results in local interruption of flows and the back propagation of the boron concentration pulse ( Figure 6C , D , Video 1 ) . The frequency of these oscillations is decreased , and the wave broadened , as the regulation speed is decreased ( Figure 6C , D ) . These simulations demonstrate that the unstable flow regime — which we now can conclusively eliminate as being triggered by boundary conditions — continuously self-propagates . Our model analysis highlights the requirement for rapid regulation of both BOR1 and NIP5;1 transporters to stabilise boron flux through root cells and minimise risks associated with transient high levels of boron in cells , both considered to be important constraints for the plant . Current experimental evidence shows rapid downregulation of the BOR1 protein ( Takano et al . , 2005; Takano et al . , 2010 ) , and NIP5;1 transcript ( Tanaka et al . , 2011; Tanaka et al . , 2016 ) , although the fine details regarding the molecular mechanisms remain to be elucidated . The modelling however predicts permanently fast regulation , not only when boron levels drop , but also when they remain constant . This is however difficult to prove with techniques that we have used previously . To experimentally explore the regulatory swiftness of the bidirectional transporter NIP5;1 at the transcript level , we used single molecule RNA FISH ( smFISH ) . In this method ~40 complementary fluorescently labeled oligonucleotide probes are used to visualise RNA at the cellular level ( Raj and Tyagi , 2010; Duncan et al . , 2017 ) . Initially , we combined probe sets to detect mRNA for both NIP5;1 and the housekeeping gene PP2A ( specifically A3 scaffolding subunit of PP2A , At1g13320 ) and compared abundance and cellular distribution ( Figure 7A ) ( Czechowski et al . , 2005; Duncan et al . , 2016 ) . Consistent with published smFISH experiments , PP2A mRNA appeared evenly diffused throughout all cells ( Figure 7B ) ( Duncan et al . , 2016 ) . In contrast , we observed many nuclear accumulations of NIP5;1 RNA that were accompanied by few cytoplasmic mRNA copies ( 125 nuclear accumulations in 149 cells ) ( Figure 7B ) . Large nuclear smFISH RNA signals have been reported for highly transcribed genes . They are typically accompanied by many copies of cytoplasmic mRNA and considered to represent bursts of transcription , where multiple Pol II associate with a locus ( Battich et al . , 2015; Duncan et al . , 2016 ) . Arabidopsis genes that contain introns must undergo pre-mRNA splicing before mature transcripts are translated ( Morello and Breviario , 2008-06 ) . Co-transcriptional splicing is common for plant genes and this ensures that introns are removed from the pre-mRNA and rapidly degraded close to the site of transcription ( Reddy et al . , 2013-10 ) . This system allows intronic smFISH RNA labelling to be used as an effective method to identify loci actively engaged in transcription ( Levesque and Raj , 2013; Duncan et al . , 2016; Rosa et al . , 2016 ) . We used this approach to further investigate NIP5;1 transcription and found 72% of cells with at least one NIP5;1 intron signal . As expected , this was lower than the 84% previously reported for the more highly expressed gene PP2A ( Duncan et al . , 2016 ) . When we combined NIP5;1 exonic and intronic RNA smFISH probe sets we observed 92 . 5% of nuclear accumulations co-localised with intron signals ( n=111 ) ( Figure 7C ) . This is consistent with the nuclear accumulations representing sites of ongoing RNA production and degradation , rather than separate nuclear storage compartments . The suggested rapid degradation of NIP5;1 mRNA after transcription was further supported by an independent approach , in which we performed qRT-PCR using probes specific to pre- or mature mRNA to measure mature mRNA/pre-mRNA ratio in root cells . The average mature mRNA/pre-mRNA ratio of NIP5;1 was less than 25% of that of PP2A ( Figure 7—figure supplement 1 ) . In accordance with our interpretation of the smFISH results , this again indicates a high degradation rate of NIP5;1 mRNA . In light of our model predictions , these observations combined point to a highly dynamic sensing system where RNA is turned over at or near the site of transcription , to limit NIP5;1 levels when boron is above a threshold level . We recently demonstrated that NIP5;1 transcript degradation occurs through ribosome stalling , triggering degradation in the cytoplasm ( Tanaka et al . , 2016 ) . We also identified an upstream sequence that promotes mRNA degradation , but does not trigger ribosome stalling . The data presented in Figure 7 provides evidence of mRNA degradation occurring close to the site of transcription . Together these results suggest that mRNA degradation could occur through two coordinated B-dependent processes; one where cytoplasmic degradation is triggered by ribosome stalling and another localised in the nucleus where RNA is turned over at the site of transcription . Relief of this repression could then ensure rapid boron uptake when required . Our model suggests that such a dynamic mRNA production/decay process could support the necessary swift boron-mediated transporter regulation critical for the evolution of controlled , stable boric acid flows through polarised tissue . These results also support the notion of a constant turnover ( both producing and breakdown the RNA ) occurring for regulating NIP5;1 transcript levels . We propose that this ‘wasteful’ process can now be understood in light of constraints of stable flows through polarised tissue . Interestingly , the regulation of the transporters is of a different nature , one predominantly on the level of protein degradation ( BOR1 ) , the other occurring on the level of transcript accumulation ( NIP5;1 ) . We conclude that the experimentally observed swiftness of boron transporters’ boron-mediated downregulation can be understood as a necessary mechanism to maintain optimal constant xylem loading rates over time , while also avoiding DNA damage due to oscillations within cells . One could argue that without a substrate-mediated downregulation of the transporters there would not be the need to avoid traffic jam-like behaviour through swift regulatory dynamics in the first place . However , were the system not to present the regulation , the plant would be unable to shield itself from toxic soil boron levels ( Figure 8A ) and be much less efficient in growing at low soil boron levels ( Figure 8B ) . Thus , some kind of inhibitory regulation necessarily has to be present , to ensure plant plasticity under different environmental conditions . As a consequence , the speed of the transporter downregulation needs to be sufficiently high . We showed that this is an intrinsic requirement , simply because boron transporters are polarly located: The requirement itself does not depend on external boron fluctuations in the medium , number of cell files composing the root tissue , possible saturable activity of the transporter itself nor the magnitude , location or strength of the xylem flow . Our study has been based on the well-characterised system of directed boron transport , owing to the richness in quantitative experimental measurements regarding BOR and NIP regulation . However , the implications of our derived insights apply to any system in which the dynamics of polar transport of a given substance relies on an inhibitory feedback between the concentration of that substance and its local transport rate . For example , the phosphate and iron transport systems present both polar transporters as well as substrate-dependent regulation , serving as candidates for such phenomena and dynamical constraint avoidance . In the case of the phosphate transport system , transporters localised to the outer side of the epidermis are responsible for uptake ( LePT1 and LePT2 , Liu et al . , 1998 ) , and they have recently been found to carry a domain possibly involved in phosphate sensing which leads to rapid degradation of phosphate transporter in a phosphate-dependent manner ( Gu et al . , 2016 ) . Similar regulation is also reported for the iron ( Fe ) transport system , where polar accumulation of IRT1 , a major Fe transporter for Fe uptake into roots , is regulated in an Fe-dependent manner ( Barberon et al . , 2014 ) . Even other transport systems which do not shuttle nutrients might display this sort of instability , such as polar auxin transport , for which stable flows might also require fast response dynamics regarding the activity of auxin importers and efflux facilitators ( Robert and Friml , 2009 ) . Indeed , for any polarised tissue in which adaptive regulation of transport levels is necessary we can expect to find intrinsic temporal constraints operating to ensure that steady state flows are maintained and large fluctuations avoided . In molecular terms , we discovered that although both transporters are predicted to work synergistically and are both required to be effectively similar in regard to rapid regulation , their regulations are biologically encoded in unique ways . We here experimentally focused on the regulatory swiftness of the bidirectional transporter NIP5;1 , and found strong evidence that it constitutes a system in which production is maintained high but degradation is rapidly controlled transcriptionally , allowing for the necessarily rapid boron-dependent response . It will be interesting to speculate what underlying reason , if any , has caused the directed exporter , BOR1 , which is similarly known and required to be swiftly regulated , to be controlled on a biologically distinct level . For this work , the analogy with motorway traffic jams was helpful to build an understanding of principles of polarised transport dynamics , albeit both systems obviously deviate in important aspects . Boron concentrations are continuous and can reach arbitrarily high levels , as opposed to ( incompressible ) cars . Yet , the discrete nature of cells and their regulated polar transporters , combined with the boron-dependent inhibition of the transporters themselves , resembles cars slowing down in response to busy regions along the motorway . Moreover , in both systems highly sensitive responses can prevent jamming ( Sugiyama et al . , 2008 ) . We therefore propose that the notion of traffic jams can be universally instructive and helpful for biologists considering physiology and regulated growth through polarised transport in plants , as well as for future efforts to enhance plant growth under diverse environmental conditions . For plant culture , MGRL medium was used ( Fujiwara et al . , 1992 ) supplemented with 1% sucrose and solidified with 1 . 5% gellan gum . For Figure 1 , A . thaliana transgenic lines carrying BOR1-GFP ( Kasai et al . , 2011 ) and GFP-NIP5;1 #8 ( Tanaka et al . , 2011 ) were germinated and pre-incubated for 5 days on normal MGRL medium plate , and then grown for 2 days on MGRL plates with 0 . 3 μM boric acid . For Figure 1—figure supplement 2 , plants were grown for seven days on MGRL medium with the indicated boric acid concentrations . Images were captured with a confocal microscope ( FV1000 , Olympus , Japan ) . GFP fluorescence was observed with 473 nm excitation and 510 nm emission . Cell wall was stained with 10 μg/mL propidium iodide aqueous solution for 3 min , and observed with 559 nm excitation and 619 nm emission . NIP5;1 promoter activity was observed using a transgenic A . thaliana strain carrying NIP5;1 promoter-GUS ( strain −2 , 180ΔUTR312 #8 in Tanaka et al . , 2011 ) . Seedlings were germinated and cultured on MGRL medium plate . Four-day-old seedlings were stained with GUS staining solution: 100 mM Na2HPO4 ( pH 7 . 0 ) , 0 . 1% Triton X-100 , 2 mM K4Fe ( CN ) 6 , 2 mM K3Fe ( CN ) 6 , 0 . 5 mg/mL X-Gluc ( 5-bromo-4-chloro-3-indolyl β-D-glucuronide cyclohexylammonium Salt , Glycosynth , UK ) in a decompressed desiccator for 20 min at room temperature . After rinsed with phosphate buffer ( pH 7 . 0 ) , stained seedlings were mounted in 45°C molten 5% agar ( gelling temperature 30–31°C , Nacalai , Japan ) aqueous solution and solidified at 4°C for 3 hr . The samples were trimmed into blocks and sliced with a vibrating microtome with thickness of 100 μm . The sliced sections were observed with microscope . Although we have constructed our model as parsimonious as possible to qualitatively explore what kind of behaviours could emerge from a polarised tissue with transporter regulation , our model requires a few important parameters to be set . We have sought to explore the possible dynamics of our system while staying well within the expected limits of what is experimentally considered plausible in terms of the boron transport system . Below we give a brief description of the data our parameter choices were based on , and , when possible , how reasonable ranges could be established . The plasmamembrane boron permeability rate ( p ) is set to a maximum rate of 8·10-2 s-1 and a minimum rate of 4 . 4·10-3 s-1 , based on the estimation of boric acid membrane permeability by Raven , 1980 and on more recent experimental measurements performed on internodal cells of the charophyte alga Chara corallina ( Stangoulis et al . , 2001 ) . Note that the plasmamembrane’s permeability rate for boric acid is relatively high compared to other nutrient elements , due to boric acid existing in a non-charged form under physiological pH . The degradation rates ξB , ξN represent the effective decay rate of the actual transporter activity , not just of the transporter protein . This includes processes such as its removal from the membrane , its inactivation due to usage , etc . As it amalgamates a range of inactivation/degradation processes , we set it higher than typical protein degradation rates , but lower than typical membrane residence times , using a half life of 9 s . The valid ranges for the production rate of the transporter activities , αB and αN , were derived from the degradation rates ξB , ξN , combined with qualitative conditions for the transporter equilibrium levels . It is as yet not possible to obtain these values directly through experiments , as several intermediate steps are involved . Instead we impose that the transporter activity steady state level does not exceed the plasmamembrane’s permeability by a thousand fold: ( 8 ) p<B* , N*<1000⁢p . Combining these conditions we could then extract the valid ranges for αB and αN . As explained in the main text , we take kB≃30kN , based on our observations of the sensitivity of BOR and NIP5;1 expression under different boron concentrations ( Figure 1—figure supplement 2 ) , further corroborated by earlier data on NIP5;1 , as displayed in Figure 1B in Tanaka et al . , 2011 . We also ran simulations to show robustness of the observed behaviours for smaller variations between kB and kN by assuming lower values of kB ( Figure 2—figure supplement 1 , Figure 4—figure supplement 2 ) . The parameter dB determines the maximum fold increase in BOR degradation at very high intracellular boron concentrations . It was set to a 50-fold increase , based on rough assessments of experiments in which plants were transferred to high boron conditions , with the caveat that assessment of the temporal change in functional , membrane-located BOR as a function of the intracellular boron concentration is very challenging . Typical cell sizes ( height , h⁢c; width , l⁢c ) and cell wall width ( l⁢w ) were estimated from confocal microscope images of A . thaliana roots . The xylem loading rate a was calculated as follows: Hosy et al . , 2003 reported that in mature Arabidopsis the rate of transpiration is 0 . 1 g water/g FW tissue/hour at 70% humidity under daytime conditions . Given an Arabidopsis FW at 18 days of around 0 . 2 g/plant , transpiration is 20 mg/hour/plant , or roughly 5·106 μm3/s . Considering a stem xylem cross-sectional surface area being roughly 107 μm2 , the convective velocity a should therefore be around 5·106 μm3/s / 107 μm2 = 0 . 5 μm/s . Numerical and analytical calculations were computed using Wolfram Mathematica 10 . 2 . To calculate time development of the model , ODEs were solved numerically using the NDSolve[] function , setting the initial conditions for each variable to zero . The parameters used in the models are shown in Table 1 . The solutions of the ODEs were visualised as a line plot or kymograph , using the Plot[] or ListDensityPlot[] functions . To evaluate the physiological impact ( maximum concentration and amplitude of cytosolic boron , and throughput towards shoots ) of transporter regulation swiftness , numerical simulations were repeated for combinations of ϵB and ϵN , which were independently varied between 0 . 01 and 2 , at intervals of 0 . 01 . For each simulation , minimum and maximum boron concentration for each cell and average throughput through the cell row between 24 to 48 h simulation time were recorded using the ‘EventLocator’ method of the NDSolve[] function . For the ring model , the boundary conditions in the original model were removed by connecting the last cell in the tissue-segment to the first one . This was done by using a tissue composed of six cells and then introducing a cell wall between cell 6 and cell 1 . The boron concentration in the novel cell wall is referred to as W6-1 . This generates a six-cell ring model . Because within a ring-model the total amount of boron is conserved , introducing an additional ODE for this cell wall component effectively means an overdetermination of variables , causing numerical issues . Hence the value of the new variable was instead defined as follows: ( 9 ) W6-1= ( Bt⁢o⁢t⁢a⁢l-l⁢c⁢h⁢c⁢∑i=16Ci-l⁢w⁢h⁢c⁢∑i=15Wi ) ⁢1l⁢w⁢h⁢c , where Bt⁢o⁢t⁢a⁢l is the total amount of boron in the whole system , which is constant over time as there is no input into nor outflow from the system . The initial boron concentration of all cells and cell walls in the simulation were set to 300 μM , and Bt⁢o⁢t⁢a⁢l was determined accordingly . The local stability analysis was performed as follows: First , all possible equilibria for the set of ODEs were found using the NSolve[] function . Next , analytical descriptions of the Jacobian matrix of the ODEs around the equilibria were derived using the D[] function . Finally , the stability of the system for varying values of ϵB and ϵN was determined according to the signs of the eigenvalues of the Jacobian matrix . ( Please note that changing ϵB and/or ϵN does not change the number of equilibria or their value , only their stability can change . ) The system is stable when the real parts of all the eigenvalues are negative , and unstable in any other case . The imaginary part of the largest eigenvalues was then used to estimate the oscillatory period in the unstable region , based on the following conversion: ( 10 ) O⁢s⁢c⁢i⁢l⁢l⁢a⁢t⁢o⁢r⁢y⁢p⁢e⁢r⁢i⁢o⁢d=2⁢πλi⁢m . To compare the predicted oscillatory period close to the unstable equilibrium with the actual oscillatory period of the system , the oscillatory period was also calculated numerically ( Figure 4—figure supplement 1 ) . For certain parameters individual variables can present rather complicated patterns , for example temporarily maximum values , followed by a slight decrease and then further rise before decreasing again . We therefore used the following algorithm to determine the oscillatory period: The ODEs were numerically solved using the NDSolve[] function . The ‘EventLocator’ method was then used to collect all extrema for three variables , C2 , C3 and N2 , for the period between 24 and 72 hr . Using the first set of extrema as a reference point , the second and potentially later sets of extrema were then compared against the first set . The first case for which all differences were within ±0 . 1% for all three variables was then labeled as being identical to the reference point , and the corresponding time interval reported as the oscillatory period . The probes were designed using the online program Stellaris™ Probe Designer version 2 . 0 and ordered from LGC Biosearch Technologies ( Petaluma , CA ) . For probe sequences see Table 2 . Col-0 seeds were stratified for 2 d at 4°C and then germinated and grown under 16/8 hr light conditions at 20°C on MGRL plates ( 30μM boric acid ) . smFISH was carried out on seedlings as described by Duncan et al . , 2017 . Briefly , 5d old seedlings were fixed for 30 min in 4% paraformaldehyde . Root squash samples were prepared on slides . Tissue permeabilisation was achieved by immersing the samples in 70% ethanol for a minimum of one hour . Two washes were carried out with wash buffer ( 10% formamide and 2x SSC ) . 100 mL of hybridisation solution ( 0 . 3 M Sodium chloride and 30 mM tri-Sodium citrate dihydrate , referred to as ’2x SSC’ , 10% dextran sulfate and 10% formamide ) containing probes ( final concentration 25 nM ) was then left to hybridise at 37°C in the dark overnight . Each sample was washed twice with wash buffer following hybridisation buffer removal with the second wash left to incubate for 30 min at 37°C . Each slide was then incubated with 100mL 4’ 6-diamidino-2-phenylindole dihydrochloride ( DAPI , Sigma , St . Luis , MO ) ( 100 ng/mL ) at 37°C for 30 min . The DAPI solution was removed before 100 μL 2x SSC was added and then removed . Samples were then incubated for 2 min at room temperature with 100 μL GLOX buffer ( 0 . 4% glucose in 10 mM Tris , 2x SSC ) before being mounted in 100 μLof GLOX buffer containing 1 μL of glucose oxidase ( #G0543 , Sigma ) and 1 μL catalase ( Sigma ) under 22 mm x 22 mm No . 1 coverslips ( Slaughter , Uppminster , UK ) and sealed by nail varnish . A Zeiss Elyra PS1 inverted microscope was used for imaging . A x100 oil-immersion objective ( 1 . 46 NA ) and cooled EM-CCD Andor iXon 897 camera ( 512×512 QE>90% ) were used to obtain all images . ( Quasar®570 probes: excitation by 561 nm laser , signals detected between 570–640 nm . Quasar®670 probes: excitation by 642 nm laser , signals detected between 655–710 nm . DAPI: excitation by 405 nm laser , signals detected between 420–480 nm . ) For all experiments a series of optical sections were set up with z-steps of 0 . 2 μm . All smFISH images were de convolved with Huygens Professional version 16 . 05 ( Scientific Volume Imaging , the Netherlands , http://svi . nl ) using the QMLE algorithm with Good signal to noise settings and 50 iterations . Maximum Z projections in Figure 7 were performed using Fiji ( an implementation of ImageJ , a public domain program by W . Rasband available from http://rsb . info . nih . gov/ij/ ) . Col-0 seedlings were germinated and grown on MGRL plates containing 30 μM boric acid for eight days . Total RNA was extracted from whole roots of about 25 seedlings with four replicates , using NucleoSpin RNA Plant ( MACHEREY-NAGEL ) . To reduce contamination of genome DNA , 500 ng of prepared RNA was subjected to DNase treatment with RQ1 RNase-Free DNase ( Promega ) , following the manufacture’s protocol . Reverse transcription was performed with Super Script III reverse transcriptase ( Thermo Fisher Scientific ) with random primers . To compare the abundance of cDNAs with different sequences , absolute quantification by realtime PCR was conducted , using PCR-synthesised DNA fragments with known concentration as standard templates . To detect mature or pre-mRNA specifically , exon-exon or intron-exon probes were designed ( Figure 7—figure supplement 1 and Table 3 ) . Realtime PCR was performed with Thermal Cycler Dice® Real Time System and SYBR Premix Ex Taq II ( Tli RNaseH Plus ) ( Takara ) , following the supplied standard shuttle 2-step PCR protocol . As standard DNA templates , DNA fragments corresponding to cDNA of mature- or pre-mRNA of NIP5;1 and PP2A were amplified by PCR with Prime Star GXL ( Takara ) , using Col-0 cDNA or genome DNA as a template . The primers used are shown in Table 3 . The PCR products were subjected to electrophoresis and the target fragment was purified from the gel slices with QIAquick Gel Extraction Kit ( Qiagen ) . The concentration of the purified DNA was estimated by NanoDrop 1000 ( Thermo Fisher Scientific ) . Tenth dilution series of 1 pg/μL~1 ag/μL template DNA were applied for standard curve . The copy number of each standard DNA solution was normalized by realtime PCR Ct values of the exon-exon proves which amplify both mature- and pre-mRNA sequences ( Figure 7—figure supplement 1 and Table 3 ) . As a negative control , in parallel , after the total RNA extraction , the same samples were processed with the same procedure except reverse transcription , confirming that realtime PCR signals from contaminated genome DNA does not affect the final results .	Every multicellular organism , including all plants and animals , faces the challenge of taking up the nutrients it needs and distributing them throughout its body . Plants absorb many nutrients including nitrogen and boron from the soil into their roots , often using tightly controlled processes that require energy to work . Plant roots contain several distinct layers of cells and the nutrients need to cross these layers to reach a channel at the centre of the root known as the xylem , which transports the nutrients to other parts of the plant . Plants need boron to grow . However , high levels of this nutrient are toxic so plants have evolved to change the rate at which they absorb boron to optimize growth in different environments . When there is little boron in the soil , certain transporter proteins move to the surface of root cells to bring boron into the root more effectively . On the other hand , when plants grow in soils with high boron , their root cells have fewer of these transporters on their surfaces to prevent too much boron entering the plant . This regulation of boron uptake appears logical , except for one detail: at any given location , the amount of boron in the soil is relatively stable and changes only very slowly . Why do plants invest energy in responding rapidly to the supply of a nutrient that changes so slowly in nature ? Sotta et al . used mathematics and experimental approaches to study boron uptake in a plant known as Arabidopsis . The work reveals that the plants ability to rapidly alter how efficiently boron moves into root cells actually serves to avoid internal “traffic jams” in boron transport . If the numbers of transporter proteins on the surface of root cells changed more slowly , individual cells would occasionally experience high levels of boron that would interfere with the movement of boron further into the root , causing a jam . Furthermore , these ‘peaks’ of boron could damage the individual cells they affect . The findings of Sotta et al . reveal that , by being able to rapidly change the numbers of certain transporter proteins on the surface of root cells , plants can ensure they receive a steady supply of boron . This work suggests that to develop artificial systems that can adapt to changing surroundings , researchers will need to engineer solutions like those found in plants in order to avoid similar traffic jams in the systems . Along with considering how plants interact with their environment , studying how they avoid internal traffic jams in nutrient uptake may help efforts to alter plants , including crops , so that they grow better in harsh environments .	[ "Abstract", "Introduction", "materials", "and", "methods" ]	[ "plant", "biology", "computational", "and", "systems", "biology" ]	2017	Rapid transporter regulation prevents substrate flow traffic jams in boron transport
The postembryonic brain exhibits experience-dependent development , in which sensory experience guides normal brain growth . This neuroplasticity is thought to occur primarily through structural and functional changes in pre-existing neurons . Whether neurogenesis also mediates the effects of experience on brain growth is unclear . Here , we characterized the importance of motor experience on postembryonic neurogenesis in larval zebrafish . We found that movement maintains an expanded pool of forebrain neural precursors by promoting progenitor self-renewal over the production of neurons . Physical cues associated with swimming ( bodily movement ) increase neurogenesis and these cues appear to be conveyed by dorsal root ganglia ( DRG ) in the zebrafish body: DRG-deficient larvae exhibit attenuated neurogenic responses to movement and targeted photoactivation of DRG in immobilized larvae expands the pallial pool of proliferative cells . Our results demonstrate the importance of movement in neurogenic brain growth and reveal a fundamental sensorimotor association that may couple early motor and brain development . During postembryonic development , the brain begins processing sensory information from the environment for the first time and continues to grow , exhibiting elevated levels of neuroplasticity compared to later stages of life . The combination of these factors makes postembryonic brain development highly susceptible to sensory experience ( Knudsen , 2004 ) . This susceptibility to experience is evident in the ‘critical’ and ‘sensitive’ periods early in life , in which sensory experiences drive permanent or near permanent changes in brain structure and function , respectively ( Knudsen , 2004 ) . Historically , neuroplastic changes associated with early sensory experience were thought to be restricted to structural and functional changes in pre-existing neurons , such as visual experience-dependent synaptic remodeling in thalamocortical projections associated with the development of ocular dominance ( Coleman et al . , 2010 ) . However , neurogenesis persists postembryonically throughout the brain , either ceasing or curtailing in adolescence or adulthood ( Lindsey and Tropepe , 2006 ) . Some neuronal populations appear to be uniquely generated during postembryonic development ( Wei et al . , 2011 ) , and this process may be regulated by sensory experience ( He et al . , 2014 ) . Thus , neurogenesis may also mediate the effects of early experience on brain development . Outside of postembryonic development , one of the best-characterized models of experience-dependent regulation of neurogenesis is the increase in cell proliferation in the subgranular zone ( SGZ ) of the dentate gyrus in the adult mammalian hippocampus ( HP ) following periods of aerobic running exercise ( van Praag et al . , 1999a ) . Since its initial discovery , studies have gone on to link exercise-induced neurogenesis to improvements in cognition , such as spatial learning ( van Praag et al . , 1999b ) and cognitive flexibility ( Anacker and Hen , 2017 ) . Studies have also incorporated exercise as a therapeutic intervention to combat neuropsychiatric disorders associated with impaired neurogenesis ( Vakhrusheva et al . , 2016; Kandola et al . , 2016 ) . However , whether physical activity affects forebrain neurogenesis during postembryonic development , when animals first gain control of their movements while exhibiting elevated levels of neurogenesis throughout the brain compared to adulthood , remains unexplored . Furthermore , such a relationship between movement and forebrain growth early in development may help explain the positive correlation between physical activity and cognitive function reported in human children ( Tomporowski et al . , 2008; Best , 2010 ) . Here , we sought to investigate the relationship between movement and neurogenesis in larval zebrafish during a developmental period in which they first begin to exhibit voluntary movements ( Buss and Drapeau , 2001 ) , have brains and peripheral nervous systems sufficiently developed to process sensory input , and continue to exhibit elevated rates of neurogenesis in many brain divisions compared to adulthood ( Lindsey and Tropepe , 2006; Feliciano et al . , 2015 ) . In addition to the well-documented advantages of larval zebrafish as models for genetic and pharmacological tractability , here we also use them for our ability to control the sensory experiences of larvae , enabling the isolation of different sensory cues associated with movement to identify the nature of sensory feedback driving neurogenic change . We first developed paradigms to both reduce and increase swimming behaviour in larvae and sample for changes in neurogenesis in the forebrain . We then sought to isolate the different sensory cues associated with movement and identify which cue drives a neurogenic response in the forebrain . Finally , we tested whether dorsal root ganglia ( DRG ) , sensory neurons that convey mechanical sensations from the zebrafish trunk , are required to mediate movement-dependent neurogenesis . We did this by both pharmacologically generating larvae deficient in DRG along the trunk and by stimulating the DRG via photoactivation of ankyrin-containing transient receptor potential channels ( TRPA1b ) in completely immobilized larvae . Altogether , we present a novel and robust relationship between movement and postembryonic forebrain neurogenesis , demonstrating that neural feedback associated with physical movement may provide a simple mechanism through which motor and brain development become coupled early in life . We first established a paradigm through which we could control the amount of swimming exhibited by zebrafish larvae noninvasively . We used movement restraint , in which larvae were confined to a smaller portion of 6-well plates by a mesh cylinder ( Figure 1A ) and tested if such restraint would reduce swimming from 3 to 9 days post fertilization ( dpf ) . Movement restraint significantly reduced the hourly distance swam by 6 and 8 dpf in both non-repeated ( Figure 1B; Treatment x Age Interaction: 1F2 , 80 = 14 . 08 , p<0 . 01 ) and repeated ( Figure 1C; Video 1; Treatment x Age Interaction: 2F2 , 34 = 14 . 16 , p<0 . 01 ) experimental designs compared to unrestrained controls . Furthermore , movement restraint also prevented the increase in the proportion of fast swims ( >10 mm/s ) first exhibited by control larvae on 6 dpf ( Figure 1D–E; Treatment x Age Interaction: 3F2 , 68 = 14 . 90 , p<0 . 01 ) . Because of the possibility that chronic movement restraint may impair larval development , we sampled body length of control and restrained larvae throughout the restraint period . We found that movement restraint did not affect larval body length by 6 dpf , but reduced body length by 9 dpf ( Figure 1—figure supplement 1A; 4F2 , 111 = 13 . 10 , p<0 . 01 ) . To test if this reduction in body length affected motor ability , we repeated our movement restraint paradigm and , on 8 dpf , we moved restraint larvae into control wells to record unrestrained swimming behaviour . Prior movement restraint did not impair either hourly distance swam ( Figure 1—figure supplement 1B; 5t20 = 0 . 98 , p=0 . 34 ) or proportion of fast swims ( Figure 1—figure supplement 1C , 6t20 = 0 . 16 , p=0 . 87 ) in unrestrained conditions . Collectively , we found that our restraint paradigm reduced motor experience in larvae by 6 dpf without impairing swimming ability . To test for changes in neurogenesis in the restrained larval zebrafish brain , we sampled the proportion of PCNA+ cells in the pallium , subpallium , olfactory bulb , and optic tectum of 6 dpf larval zebrafish . Movement restraint significantly reduced the proportion of proliferative ( PCNA+ ) cells in the forebrain of zebrafish by 6 dpf ( Figure 2A–C; 7t9 = 4 . 07 , p<0 . 01 ) sampled across consecutive coronal sections ( Figure 2—figure supplement 1 ) , without affecting forebrain size ( Table 2 ) . This difference was attributed to a reduction in the proportion of PCNA+ cells in both the subpallium ( Figure 2—figure supplement 2A; 8t5 = 3 . 77 , p=0 . 01 ) and pallium ( Figure 2—figure supplement 2B; 9t5 = 7 . 36 , p<0 . 01 ) . Conversely , movement restraint did not affect the size ( Table 2 ) or proportion of PCNA+ cells in the olfactory bulb ( Figure 2—figure supplement 2C; 10t9 = 0 . 53 , p=0 . 61 ) or optic tectum ( Figure 2—figure supplement 2D; 11t8 = 0 . 87 , p=0 . 41 ) by 6 dpf . Movement restraint reduced forebrain size by 9 dpf ( Table 2 ) ; however , after correcting for forebrain size , chronic restraint still significantly reduced the proportion of PCNA+ cells in the forebrain by 9 dpf compared to controls ( Figure 2D–F; 12U = 0 , p<0 . 01 ) . Movement restraint also reduced the proportion of tbr2+ cells , a protein marker of intermediate progenitors and newly generated neurons ( Englund et al . , 2005 ) , in the pallium by 9 dpf ( Figure 2G–I; 13t16 = 3 . 37 , p<0 . 01 ) without affecting the proportion of pallial GFAP+ radial neural stem cells in Tg ( GFAP:gfp ) embryos ( Figure 2J–L; 14t12 = 0 . 35 , p=0 . 73 ) . Thus , movement restraint reduced the size of the pool of proliferative cells , presumably neural progenitors , in the forebrain specifically , without affecting the size of the resident radial stem cell population . We then asked how movement restraint results in a reduced forebrain proliferative cell population . We reasoned that a reduction in this cell population might occur when either proliferative cells generate more differentiated cells at the expense of self-renewal , reducing the size of the proliferative population over successive divisions , or by apoptosis in the proliferative population . To sample cell differentiation in these forebrain populations , we exposed larvae to 5 mM 5-Ethynyl-2'-deoxyuridine ( EdU ) , a synthetic thymidine analog that is incorporated in dividing cells , for 24 hr starting on 5 dpf , then sampled the proportion of EdU+ cells that also express Elavl3 protein , a marker of cells with a differentiated neuronal fate ( Lindsey et al . , 2012 ) . If movement restraint biased the production of differentiating cells over progenitor self-renewal , we would predict that restrained larvae would exhibit more EdU+ cells that co-expressed Elavl3 . Movement restraint significantly increased the proportion of newly generated cells that co-label with Elavl3 in both the pallium ( Figure 2M; 15t6 = 6 . 02 , p<0 . 01 ) and subpallium ( Figure 2N–T; 16t6 = 3 . 43 , p=0 . 01 ) without affecting the absolute number of EdU+ cells produced in the pallium ( Figure 2—figure supplement 2E; 17t6 = 0 . 35 , p=0 . 74 ) or subpallium ( Figure 2—figure supplement 2F; 18t7 = 0 . 73 , p=0 . 49 ) . Conversely , movement restraint did not affect the number of cells expressing the apoptotic marker activated caspase-3 ( Casp3 ) in the forebrain ( Figure 2—figure supplement 2G; 19t7 = 1 . 06 , p=0 . 32 ) by 6 dpf . By 9 dpf , however , movement restraint significantly increased the number of activated Casp3+ cells in the forebrain ( Figure 2—figure supplement 2H; 20t13 = 3 . 56 , p<0 . 01 ) . This increase in apoptosis at 9 dpf was specific to the pallium ( Figure 2—figure supplement 2I; 21U = 1 , p<0 . 01 ) and not found in the subpallium ( Figure 2—figure supplement 2J; 22t13 = 1 . 76 , p=0 . 10 ) . Despite this increase in pallial apoptosis , cell death rates remained low by 9 dpf and Casp3+ cells were not observed along the midline or dorsal surface of the brain , where the neurogenic niche lies . Together , these findings suggest that movement restraint biased newly generated cells to differentiate into neurons , ultimately at the expense of self-renewal . Because physical restraint may restrict more than just movement ( i . e . , reducing sensory input in a smaller space ) , we tested whether increasing movement could also impact forebrain cell proliferation . We raised larvae in groups ( n = 15–20 ) housed in transparent plastic canals against different strengths of water current . Control larvae experienced no displacing current ( Figure 3A; water dripping in and out , current did not displace larvae ) and ‘exercised’ larvae experienced a strong current ( Figure 3B; water flow strong enough to displace larvae ) from 3 to 9 dpf following a daily schedule ( Figure 3C ) . In the current condition , larvae would have to swim to counteract the flow of water and maintain their position in the canal , akin to forced exercise paradigms in rodents ( Leasure and Jones , 2008 ) . On 9 dpf , larvae reared against a strong current exhibited a greater proportion of PCNA+ cells in the pallium ( Figure 3D; 25t15 = 2 . 80 , p=0 . 01 ) , but not the subpallium ( Figure 3E; 26t19 = 1 . 22 , p=0 . 24 ) . On 9 dpf , the size of both brain regions was not affected by rearing treatment ( Table 2 ) . When we sampled pallial proliferation in larvae earlier , at 6 dpf , we again found an increase in the proportion of PCNA+ cells in the pallium in larvae reared against a strong current ( Figure 3—figure supplement 1A; 23t7 = 2 . 76 , p=0 . 03 ) , while the subpallium was unaffected ( Figure 3—figure supplement 1B; 24t6 = 1 . 27 , p=0 . 25 ) . Again , the size of both brain regions was not affected by rearing in a strong current on 6 dpf ( Table 2 ) . To test if increased movement affected cell differentiation as in our restraint paradigm , we exposed larvae to EdU in a petri dish overnight for 13 hr from 8 to 9 dpf prior to being returned to their swimming canals for a final 5 hr of current-rearing . Rearing larvae against a current reduced the number of newly generated ( EdU+ ) cells that also expressed Elavl3 compared to controls ( Figure 3F; 27t11 = 2 . 39 , p=0 . 04 ) , consistent with increased movement maintaining an expanded proliferative cell population over the generation of differentiated neurons . Rearing larvae against a current from 3 to 9 dpf did not affect body length ( Figure 3G , 28U = 102 , p=0 . 5081 ) , suggesting these effects on pallial neurogenesis are not a product of overall growth . Together with our movement restraint data , the increase in pallial cell proliferation following exercise suggests that motor experience regulates forebrain neurogenesis specifically in the pallium , similar to the neurogenic effect of exercise in the mammalian SGZ . Upon establishing a link between motor experience and pallial neurogenesis , we asked if we could identify the modality of sensory feedback associated with movement driving cell proliferation . To isolate visual and physical components of movement , we restrained larvae entirely in agarose from 3 to 6 dpf , preventing locomotion . We then re-introduced visual stimulation associated with movement ( optic flow ) by exposing immobilized larvae to computer-generated visual gratings to simulate visual motion . Physical input associated with movement was re-introduced to immobilized larvae by cutting the tail of larvae free from agarose embedding , enabling swimming tail movement without bodily displacement in the environment . Control larvae were provided both visual stimulation ( gratings ) and tail movement ( Figure 4A ) , whereas treatment groups experienced only either tail movement ( Figure 4B ) or visual stimulation ( Figure 4C ) . Blocking tail movement ( complete immobilization ) significantly reduced the proportion of PCNA+ cells in the pallium by 6 dpf ( Figure 4D; 29F2 , 14 = 7 . 89 , p<0 . 01 ) . Conversely , removing just visual stimulation had no impact on the number of PCNA+ cells in the pallium . Neither removing visual stimulation nor tail movement affected the proportion of PCNA+ cells in the subpallium ( Figure 4E; 30F2 , 14 = 2 . 42 , p=0 . 13 ) . Because immobilization could impair brain growth globally , we also sampled the number of Hoechst+ cells per section as a proxy for absolute forebrain size . Immobilization did not reduce the total number of cells in the pallium or subpallium ( Table 2 ) , instead affecting the PCNA+ cell population specifically . These results suggest that physical input associated with locomotion , specifically tail movement during swimming , drives changes in pallial neuroproliferation . One source of neural feedback that could detect physical movement is the lateral line , a system of hair cells distributed along the teleost body that detects changes in water flow ( Dijkgraaf , 1963 ) . We treated 3 dpf larvae with 30 µM copper sulfate for 30 min , an ototoxin that ablates lateral line hair cells and impedes subsequent regeneration of these cells over the following days ( Mackenzie and Raible , 2012 ) . We confirmed hair cell ablation by the complete absence of beta-acetylated tubulin ( AcTub ) expression in hair cell cuppulae following treatment with copper sulfate ( Figure 5A ) . If the lateral line is involved in mediating movement-dependent neurogenesis , then removal of this feedback should affect PCNA+ cell populations in the pallium . However , copper sulfate treatment did not affect the proportion of PCNA+ cells in the 6 dpf larvae pallium ( Figure 5B; 31t12 = 0 . 51 , p=0 . 62 ) when all larvae were reared in unrestrained wells . Intact lateral line signaling does not appear to mediate movement-dependent changes in pallial neurogenesis . In vertebrates , DRG collect sensory feedback from the body and communicate these signals via ascending pathways to the CNS in the spinal cord ( Vandewauw et al . , 2013 ) . Accordingly , DRG represent another system of neural feedback that could convey physical cues associated with movement . We tested whether blocking DRG development in the trunk would reduce PCNA+ cell populations in the pallium associated with swimming . We blocked development of DRG along the larval trunk by treating embryos with the ErbB receptor antagonist AG1478 in a limited window from 8 to 30 hpf followed by a wash-out period of almost 2 days ( Honjo et al . , 2008 ) . By 3 dpf , we confirmed that earlier AG1478 treatment reduced DRG development in Tg ( isl2b:mgfp ) transgenic embryos ( Figure 6Ai–iii ) . However , AG1478 treatment affected neither the pallium size ( Table 2 ) nor proportion of pallial PCNA+ cells on 3 dpf , prior to any motor treatments ( Figure 6—figure supplement 1A; 32t7 = 0 . 04 , p=0 . 97 ) . Thus , we divided 3 dpf AG1478- and DMSO-treated larvae into control and movement restraint conditions and sampled PCNA+ cell populations as above . By 6 dpf , prior AG1478 treatment did not affect swimming compared to DMSO-treated controls ( Figure 6—figure supplement 1B; 33F3 , 17 = 16 . 16 , p<0 . 01 ) . Whereas DMSO-treated larvae exhibited a movement-dependent change in the proportion of pallial PCNA+ cells , prior AG1478 treatment blocked this effect ( Figure 6—figure supplement 1C; 34F3 , 77 = 4 . 17 , p<0 . 01 ) . Prior AG1478 treatment did not affect 6 dpf pallium size between unrestrained larvae ( Table 2 ) . Furthermore , when unrestrained larvae were exposed to EdU from 5 to 6 dpf as in our restraint paradigm , prior AG1478 treatment increased the number of newly generated ( EdU+ ) cells that also express Elavl3 ( Figure 6—figure supplement 1D; 35t8 = 2 . 53 , p=0 . 04 ) compared to DMSO-treated controls , suggesting early AG1478 treatment affects pallial neurogenesis similarly to chronic restraint , albeit at a reduced magnitude . To resolve differences in the proportion of PCNA+ cells in the pallium of AG1478- and DMSO-treated larvae , we repeated this experiment and extended control and movement restraint conditions until 9 dpf as above . Prior AG1478 treatment also did not affect swimming on 8 dpf ( Figure 6B; 36F3 , 20 = 27 . 59 , p<0 . 01 ) . By 9 dpf , AG1478-treated larvae exhibited movement-dependent differences in the proportion of pallial PCNA+ cells , however , the magnitude of this effect was significantly attenuated compared to DMSO-treated controls ( Figure 6C–G; 37F3 , 23 = 26 . 68 , p<0 . 01 ) . Prior AG1478 treatment did not affect 9 dpf pallium size between unrestrained larvae ( Table 2 ) . Together , these results suggest that neural feedback from DRG mediates , at least in part , movement-dependent forebrain neuroproliferation . If DRGs mediate neural feedback during movement to stimulate pallial cell proliferation , then direct stimulation of DRGs independent of movement should also drive pallial neurogenesis . We used an optopharmacological approach to activate DRGs by exposing larvae to a combination of light and Optovin , a small molecule that enables photoactivation of TRPA1 receptors ( Kokel et al . , 2013 ) . TRPA1 receptors are found in DRG ( Vandewauw et al . , 2013 ) , trigeminal neurons and Rohon-Beard cells in larval zebrafish ( Prober et al . , 2008 ) . In zebrafish , Optovin acts specifically on the TRPA1b paralog , which is exclusively expressed in sensory ganglia up to 5 dpf ( Prober et al . , 2008 ) . To repeatedly photoactivate DRGs using Optovin , we exposed unrestrained 5 dpf larvae isolated in a 24-well plate to either Optovin or DMSO and adjusted light exposures and intermittent darkness to achieve repeatable behavioural activation . Unrestrained larval zebrafish incubated in Optovin exhibited intense , sporadic bouts of movement during light exposure , presumably as spinal reflexes in response to intense DRG activation ( Kokel et al . , 2013 ) . 5 dpf larvae exhibited repeatable , photo-activated motor responses to 2 s of exposure to light every 5 min , whereas larvae treated with DMSO exhibited no such responses ( Figure 7A–B , Figure 7—figure supplement 1; Treatment x Timebin Interaction: 38F2 , 44 = 6 . 36 , p<0 . 01 ) . Therefore , we used 2 s of light stimulation every 5 min as a paradigm to regularly stimulate DRG in immobilized larvae . We immobilized 3 dpf larvae in agarose individually in 24-well plates . On 5 dpf , larvae were incubated with either DMSO or Optovin and exposed to either darkness or a 5 hr session of light presentations as above . Twelve hours following the end of this session , light exposures significantly increased the proportion of PCNA+ cells in the pallium of larvae exposed to Optovin ( Figure 7C; Drug x Light Treatment interaction; 39F1 , 32 = 4 . 47 , p=0 . 04 ) , whereas light treatments had no effect on the proportion of PCNA+ cells in DMSO-incubated larvae . Furthermore , 6 dpf larvae deficient in trunk DRGs ( using a transient 8–30 hpf treatment with AG1478 as above ) did not exhibit this optovin-and-light-dependent increase in the proportion of PCNA+ cells in the pallium compared to controls ( 8–30 hpf DMSO treatment; Figure 7D; 40t13 = 2 . 33 , p=0 . 04 ) . Thus , DRG activation appears sufficient to increase pallial neurogenesis in the zebrafish larvae in the absence of physical movement . We found that movement plays a critical role in determining the number of neural progenitors in the zebrafish forebrain during postembryonic development . Previous work has focused on coupling increased physical activity via aerobic exercise with increases in cell proliferation in the adult mammalian SGZ ( Fabel and Kempermann , 2008 ) . Here , we found that physical activity also modulates forebrain cell proliferation postembryonically in the larval zebrafish pallium . Whereas we found that increased physical activity in fish led to an increase in pallial cell proliferation , we also report a negative neurogenic response when movement is reduced via restraint or immobilization . In the most extreme case , restricting larval movement resulted in the near absence of a proliferative population in the pallium by 9 dpf , even though these larvae were fully capable of swimming normally thereafter . Furthermore , we found that these changes in progenitor populations had subsequent impacts on neurogenic brain growth: restrained larvae , who exhibit reduced pallial cell proliferation by 6 dpf , develop smaller forebrains by 9 dpf due to a combination of reduced neurogenesis and , to a lesser extent , pallial cell apoptosis . The mechanisms through which the neurogenic niche is affected by exercise in the adult rodent hippocampus include proposed changes in cell fate , cell cycling , and apoptosis in neural precursors ( Overall et al . , 2016 ) . Here , we found that , postembryonically , movement appears to maintain proliferative cell populations in the zebrafish pallium primarily by promoting self-renewal in neural progenitor cell populations whereas restraining movement promoted their premature differentiation . Because control and restrained larvae produced the same number of cells in the forebrain from 5 to 6 dpf , movement-dependent regulation of postembryonic forebrain cell proliferation appears to occur predominantly through regulating self-renewal and the production of differentiated cells over factors that might affect the absolute number of cells produced , such as cell cycle length . Within the neurogenic niche , movement-dependent maintenance of the progenitor pool may involve the Shh signaling pathway , which expands progenitor populations via symmetric cell division ( Lai et al . , 2003; Machold et al . , 2003; Yang et al . , 2015 ) . Collectively , our findings demonstrate the importance of movement in maintaining a source of new neurons to support forebrain growth postembryonically and present zebrafish as a novel model in which movement modulates early brain development over the course of a few days . In addition to characterizing the relationship between movement and postembryonic neurogenesis in the forebrain , we also sought to identify the nature of the feedback signal associated with movement that drives this neurogenic change . We found that physical cues associated with movement send ascending neural feedback to the brain via DRGs to drive changes in neurogenesis . Specifically , larvae deficient in DRGs along their trunk exhibited an attenuated neurogenic responses to swimming compared to controls . However , we still found significant modulation of pallial cell proliferation on 9 dpf in DRG-deficient larvae . This continued modulation of neurogenesis in the older larvae may be attributed to the nature of our treatment , which blocks the development of DRGs along the trunk , but does not affect DRG populations that are derived from neural crest cells in the head that may also signal movement ( Honjo et al . , 2008 ) . Other proposed mechanisms of movement-dependent neurogenesis , such as the circulation of growth factors proposed to mediate exercise-dependent adult neurogenesis ( Cotman et al . , 2007 ) , and other mechanosensory cell populations , such as Rohon-Beard cells , may also play a role in driving motor experience-dependent neurogenic brain development . A previous study has demonstrated that treating zebrafish embryos with AG1478 can impair proliferation during embryogenesis in the zebrafish optic tectum ( Sato et al . , 2015 ) . In that study , changes in tectal neurogenesis were observed using a near 8-fold increase in AG1478 concentration ( compared to that used here ) and neurogenesis was found to resume normally within hours following drug washout . Recognizing the possibility of a lasting effect of early AG1478 treatment , we sampled forebrain neurogenesis in 3 dpf larvae treated earlier with AG1478 or DMSO and found no effect of AG1478 treatment on pallial neurogenesis prior to movement or Optovin manipulations . We also found that earlier AG1478 treatment had no effect of pallium size in restrained or unrestrained control larvae by 6 dpf . In conjunction with our original movement and Optovin experiments , which do not include AG1478 manipulations , our results suggest that motor experience-dependent neurogenesis is mediated , in part , via peripheral neural feedback and is likely not attributed to early AG1478 treatment . However , future work would benefit by contrasting the results obtained here with larvae in which DRG development is blocked using alternative means . Furthermore , the ErbB signaling inhibitor used in our studies may have non-neural effects on skin or heart development . Although it is unlikely that these could be a factor in determining transmission of movement information to the brain to alter forebrain neurogenesis , this could be examined in the future . Using completely immobilized larvae , we were able to stimulate pallial cell proliferation by stimulating DRG along with other cells . Furthermore , this increase in pallial cell proliferation due to stimulation was not observed in larvae deficient in trunk DRGs . In conjunction with our studies reducing specific DRG populations along the trunk , our results suggest that neural feedback associated with movement is sensed predominantly by DRG and that DRG activation is sufficient to expand a progenitor pool in the forebrain . This previously undocumented role for DRG in conveying physical cues associated with movement to expand pools of forebrain progenitors may in turn provide a larger source of neurons and support more neurogenic brain growth in the most active animals . We found that physical movement of the body was the most important component of movement driving pallial neurogenesis . Accordingly , we propose that movement triggers mechanosensory input detected by DRGs that are then sent to the brain . In zebrafish larvae , mechanosensory input is most likely to come from one of three sources . The first possibility is the lateral line , which detects changes in water flow and vibration in the environment ( Dijkgraaf , 1963 ) . Here , ototoxic ablation of this system had no impact on pallial proliferation , suggesting it does not play a role in maintaining pallial progenitor populations . Second , Rohon-Beard ( RB ) cells , an early-developing population of spinal neurons , transmit mechanosensory signals ( Faucherre et al . , 2013 ) and contain TRPA1b receptors ( Prober et al . , 2008 ) that can be activated by Optovin stimulation ( Kokel et al . , 2013 ) . Originally , RB cells were thought to die off entirely by 4 dpf ( Reyes et al . , 2004 ) , but subsequent work using transgenic markers suggests they may persist up to 1–2 weeks post-fertilization ( Kucenas et al . , 2006; Palanca et al . , 2013 ) . Our studies showed that early AG1478 treatment , which specifically affected DRG development with resident RB cell populations remaining unaffected , blocked the Optovin-induced increase in proportional PCNA+ cells in the pallium following light presentations , suggesting DRG and not RB are responsible for the increase in pallial neurogenesis following Optovin and light treatment . However , our data indicate that RBs may also contribute to motor experience-dependent pallial cell proliferation , particularly in DRG-deficient larvae by 9 dpf . Finally , DRGs can transmit mechanosensory information from the trunk owing to the array of sensory channels they contain including TRPA1b ( Kokel et al . , 2013 ) . Because both the removal and activation of DRGs produced neurogenic consequences in the pallium , we propose that these sensory neurons are the primary mediators of movement-dependent postembryonic neurogenesis . TRPA1 channels exhibit deep evolutionary conservation across vertebrates ( Christensen and Corey , 2007 ) . Zebrafish have two orthologs of the TRPA1 channel: only TRPA1b , however , likely processes external signals ( Prober et al . , 2008 ) and is activated by Optovin treatment ( Kokel et al . , 2013 ) . Whereas TRPA1 function has been implicated in touch stimuli in DRG of mice ( Kwan et al . , 2006; Brierley et al . , 2011 ) and sensory neurons in C . elegans ( Kindt et al . , 2007 ) , this channel is also associated with transducing chemosensory and nociceptive input ( Prober et al . , 2008 ) . Here , we found evidence suggesting a novel function for TRPA1 in transducing physical cues associated with bodily movement during locomotion . Our results demonstrate a robust connection between motor and brain development during postembryonic development . Motor development in most vertebrates begins early in the postembryonic period , including both viviparous species , such as with fetal motor development in humans ( de Vries et al . , 1982 ) , and oviparous species , such as the larvae studied here . Therefore , if conserved across taxa , this close relationship between movement and neurogenesis may couple early motor and brain development . Furthermore , this relationship could help explain correlations between early physical and mental development , such as the long-observed comorbidity of physical and mental impairments ( Barnett et al . , 2012 ) and correlation between sedentary lifestyle and depression ( Anton et al . , 2006 ) , which has been previously associated with impaired neurogenesis ( Jacobs et al . , 2000 ) , in children . All zebrafish used in this study were of an AB genetic background . Larval strains used in this study include: Tg ( dlx5/6:gfp ) ( generously provided by Dr . Marc Ekker , University of Ottawa ) and Tg ( GFAP:gfp ) ( generously provided by Dr . Pierre Drapeau , Université de Montreal ) in motor restraint and visual vs . physical movement cue experiments; Tg ( βactin:gfp ) ( generously provided by Dr . Ashley Bruce , University of Toronto ) for copper sulfate treatments; and Tg ( isl2b:gfp ) ( generously provided by the late Dr . Chi-Bin Chien , University of Utah ) larvae for all AG1478 and Optovin treatment experiments . All adult zebrafish crossings included 2–3 male and female fish . Larvae were collected on the day of fertilization in system water and moved to a dark incubator held at 28°C . On 1 dpf , larval water was bleached for 30 s , rinsed four times with fresh system water , and larvae were dechorionated using forceps , before being returned to the incubator . From 3 dpf onward ( dpf ) , larvae were housed in a facility room held under a 14/10 light/dark cycle at 28°C ( Lights on at 08:00/Lights off at 22:00; light intensity = 300 lux ) . Larvae housed individually in well plates had half of their system water replaced twice daily ( at 08:00 and 14:00 ) . From 5 dpf onward , larvae were also fed size 0 zebrafish food ( Gemma Micro; Skretting , Tooele , Utah ) twice daily immediately following water changes . Zebralab ( ViewPoint , Montreal , Canada ) recordings were made in a separate testing room with similar environmental conditions as the fish facility . After recordings , all larvae were returned to the housing facility . In all experiments aside from those involving movement tracking ( see below ) , larvae were randomly assigned to experimental conditions from the same clutch prior to experimental manipulations . Minimum sample sizes were selected to mirror those in preliminary experiments and our initial findings reported here , in which we found that restraint influenced the number of proliferative cells in the forebrain . Following all experimental procedures , larvae were sacrificed using an overdose of tricaine prior to tissue collection and fixation ( see below ) . All animal experiments were performed with the approval of the University of Toronto Animal Care Committee in accordance with the guidelines from the Canadian Council for Animal Care ( CCAC ) . To restrict movement , we reared isolated 3 dpf larvae in wells of either unmodified 6-well plastic plates ( well diameter = 3 . 5 cm ) or modified 6-well plates in which larvae were confined to a central portion of each well within a cylinder of plastic mesh ( Figure 1A; cylinder diameter = 1 cm ) . This mesh barrier was selected to both allow water flow in and out of the confined region and the rest of the well and prevent the larvae from escaping . To track swimming of larvae throughout experiments , we used the Zebralab automated tracking system . Larvae were recorded on 4 , 6 , and 8 dpf . On each recording day , one tray of larvae was moved into the Zebrabox ( ViewPoint; held at 800 lux light intensity ) recording apparatus by 08:30 . Following a 30 min habituation period , swimming was recorded for 4 hr before larvae were moved into fresh system water in a new tray and returned into the Zebrabox recording apparatus for a 30 min habituation and then an additional 4 hr of swimming before being returned to the facility room ( at 17:00 ) . On 6 and 8 dpf , larvae were also fed prior to habituation in the morning ( at 08:00 ) and between recording sessions ( at 13:00 ) . Because only single trays of larvae were recorded in a session , restraint and control groups were reared as cohorts offset by 1 day and , accordingly , were recorded on alternating days . In all movement tracking experiments , we used 2 tanks of mixed-sex adult zebrafish ( each containing 3 males and 3 females ) from the same genetic background to generate each cohort . To control for genetic variation between families , parentage was reversed and balanced between treatment groups in subsequent cohorts . For example , the tank of adult zebrafish crossed to generate the control group in our first cohort was crossed to generate the restraint group in our second cohort . Tracking experiments were repeated to achieve the sample sizes reported here and included at least two cohorts . Zebralab tracking thresholds were set up as follows: Inactive/Small Swim Threshold = 5 mm/s; Small/Large Swim Threshold = 10 mm/s . To prevent physical movement from 3 to 6 dpf , we embedded larvae in 1 . 2% agarose dissolved in system water at 12:00 on 3 dpf . We moved larvae into plastic wells in a droplet of system water , anesthetized larvae using tricaine ( 4 g/L; Sigma-Aldrich , Oakville , Canada ) , and introduced a drop of 1 . 2% agarose to mix with the system water . Once set , additional warmed agarose was added around the embedded larvae to secure the embedded larvae in the well . For treatments requiring free tail movement , newly embedded larvae were submerged in system water and a scalpel was used to cut a block of agarose free from below the larvae’s neck to beyond the base of the tail . To simulate optic flow , we generated a visual grating stimulus using PsychoPy ( Peirce , 2008 ) in which a black-and-white striped gradient moves along a randomly selected axis ( between 0–360° ) for 30 s , remains stationary for 30 s , and then begins again along a new , randomly selected axis . Gratings were displayed on a Dell P2212Hb computer monitor mounted horizontally , with plastic trays containing immobilized larvae sitting on top of the screen . In pilot experiments , grating bandwidth and speed were adjusted such that they would drive 6 dpf free-swimming larvae in a petri dish placed on the screen to swim along the grating axis . Gratings were presented to immobilized larvae starting on 3 dpf from 15:00-19:00 , on 4–5 dpf for two 4 hr sessions ( 08:00-12:00 and 15:00-19:00 ) , and on 6 dpf for just the morning session . On 3 dpf , larvae were exposed to either system water or 30 µM copper sulfate ( Sigma-Aldrich ) added to the swimming media for 30 min starting at 09:00 . Following copper sulfate exposure , both groups of larvae were rinsed three times with fresh system water . A subset of each treatment group was kept in a petri dish for an additional 30 min before sacrifice and tissue collection to validate ototoxicity of the copper treatment . The rest of the larvae were all isolated in control 6-well plates for the remainder of the experiment . We dechorionated Tg ( isl2b:mgfp ) embryos by 6 hpf and treated them with either 4 µM 4- ( 3-chloroanilino ) −6 , 7-dimethoxyquinazoline ( AG1478 ) , an ErbB receptor antagonist ( Sigma-Aldrich ) dissolved in 0 . 4% DMSO in system water or 0 . 4% DMSO in system water from 8 to 30 hpf ( Honjo et al . , 2008 ) . At 30 hpf , larvae were rinsed three times with fresh system water and kept in an incubator until 3 dpf . On 3 dpf , AG1478 treatment was confirmed using fluorescence microscopy to count trunk DRG , which express GFP in Tg ( isl2b:mgfp ) , in treated transgenic larvae . We excluded any larvae treated with AG1478 that exhibited more than 4 DRGs in the trunk . These larvae were removed from all experiments to ensure experimental manipulations only included larvae with most or all of trunk DRGs missing . We incubated larvae in Optovin ( Hit2Lead , San Diego , California ) , a photo-activated small molecule that activates TRPA1b receptors found in zebrafish sensory neurons ( Kokel et al . , 2013 ) . To establish a paradigm involving repeated photoactivation of TRPA1b , we incubated 5 dpf unrestrained larvae housed individually in 24 well plates in either 10 µM Optovin ( in 0 . 1% DMSO in system water ) or just 0 . 1% DMSO in system water for 2 hr in the Zebrabox ( Viewpoint ) tracking apparatus with the lights off . Following incubation , larvae were exposed to 2 s of white light ( 800 lux ) alternating with 5 min of dark and movement was recorded using the automated tracking parameters outlined above . After validating the utility of Optovin in unrestrained larvae , we incubated half of the larvae in 24-well plates containing 5 dpf larvae immobilized in agarose from 3 dpf ( as above ) in either Optovin or a DMSO vehicle . One tray of larvae was kept in the Zebrabox apparatus while the other was kept in complete darkness on the counter adjacent to the Zebrabox , beneath an opaque black plastic cover . From 15:00-20:00 , larvae in the Zebrabox were exposed to 2 s of light ( 800 lux ) alternating with 5 min of darkness . At 20:00 , all larvae trays were rinsed using fresh system water three times and moved into the dark incubator overnight . For coronal section immunohistochemistry , larvae were sectioned using a freezing cryostat ( 20 μm sections ) , thaw-mounted on Superfrost Plus slides ( Sigma-Aldrich ) , and dried for 3 hr in the dark at room temperature . Tissue was rehydrated in 0 . 2% Tween20 in phosphate-buffered saline ( PBT ) for 30 min at room temperature . At this point , tissue that was labeled for Elavl3 production was refixed with 4% paraformaldehyde for 20 min at room temperature and exposed to 50 mM Tris ( pH = 8 . 0 ) for 60 min at 75–80°C before being rinsed with PBS three times and PBT once . All tissue was washed with PBT three times and blocked with 2% Normal Goat Serum ( NGS ) in PBT for at least 2 hr at room temperature . Tissue was incubated with the primary antibody in 2% NGS in PBT at 4°C overnight . Primary antibodies used in this study included mouse anti-Human Neuronal Protein HuC/HuD , also called ELAV like neuron-specific RNA binding protein three in zebrafish ( Elavl3; Life Technologies , Waltham , Massachusetts , 1:400 ) , PCNA ( Invitrogen , Carlsbad , California , 1:500 ) , rabbit anti-activated caspase 3 , a marker of apoptosis ( Cell Signaling Technology , Danvers , Massachusetts , 1:500 ) , rabbit anti-GFP ( Alexa-488 conjugated , Life Technologies , 1:1000 ) , and tbr2 ( Abcam , Cambridge , United Kingdom , 1:500 ) . On the next day , tissue was rinsed three times with PBT before being incubated in a secondary antibody in 2% NGS in PBT for 1–2 hr at room temperature . Secondary antibodies used included Cy3-conjugated Goat Anti-Mouse IgG ( Jackson ImmunoResearch , West Grove , Pennsylvania , 1:500 ) , Cy3-conjugated Goat Anti-Rabbit IgG ( Jackson ImmunoResearch , 1:500 ) , and Cy2-conjugated Goad Anti-Rabbit IgG ( Jackson ImmunoResearch , 111-225-144 , 1:500 ) . Tissue was rinsed with PBT three times . To visualize EdU , a Click-iT EdU reaction was performed as per the instructions included in the kit using the Alexa 647 azide ( Invitrogen ) . Next , tissue was rinsed with PBT , counterstained with Hoechst for 10 min at room temperature , rinsed four times with PBS and coverslipped using 90% glycerol in PBS . Coverslips were sealed with clear nail polish and stored at 4°C until imaging . Images were captured with a Leica SP8 confocal microscope using a 40x objective as image stacks throughout the focus of sections compared as z-stacks with a z sampling distance of 1 μm . The treatment identity of larvae was masked prior to image analysis , which was all performed using IMARIS ( Bitplane , Belfast , United Kingdom ) . For whole mount immunohistochemistry larvae were fixed in 4% paraformaldehyde for 2 hr at room temperature , rinsed in PBS , exposed to acetone for 7 min at −20°C , rinsed again with PBS , and a mixture of 1% bovine serum albumin , 1% DMSO , and 0 . 1% TritonX-100 ( PBDT ) before being blocked in 10% NGS in PBDT for 1 hr . Following blocking , larvae were incubated overnight in mouse anti-alpha acetylated tubulin ( Abcam , Cat No: 6-11B-1; 1:500 ) . The second day of immunohistochemistry was completed as above , except exposure to the secondary antibody was extended to 5 hr at room temperature and whole mount larvae were kept in PBS at the end of staining , not coverslipped or sectioned . Images were captured with a Leica SP8 confocal microscope using a 40x objective as image stacks throughout the focus of sections compared as z-stacks with a z sampling distance of 1 μm . Image analyses performed using IMARIS ( Bitplane , Belfast , United Kingdom ) . PCNA +cell sampling was performed using landmarks summarized in Figure 2—figure supplement 1 . Olfactory bulb sampling was only possible in single , 20 µm sections and both hemispheres were sampled together . Both pallial and subpallial PCNA+ , Hoechst+ , tbr2+ , and GFAP+ cell sampling were performed across both hemispheres in three consecutive , 20 µm sections for all histology involving 6 and 9 dpf larvae . PCNA +cell sampling in 3 dpf used only two consecutive coronal sections , as the telencephalon was not sufficiently grown to span 3 consecutive sections . Optic tectum sampling on 6 dpf was performed on the first coronal section posterior to the tectal neuropil , sampled in and averaged across both hemispheres in each brain . We define a biological replicate as an individual larvae derived from a mixed clutch borne of at least two male and two female adult zebrafish . Whereas we did not perform any technical replicates of our experiments , which we define as a complete repetition of a single experiment , we instead replicated our main findings by either repeating them in later , more elaborate experiments ( for example , control vs . restraint paradigms in both initial experiments and AG1478 experiments ) or repeating experiments over multiple time courses ( for example , identifying the effects of exercise on forebrain neuroproliferation by both 6 and 9 dpf ) . All statistical test results are preceded by a superscript numeral enabling reference to each test in our calculations of statistical power summarized in Table 1 . In experiments involving two groups , treatment groups were compared using Student’s t-test or , when parametric assumptions were not met , Mann-Whitney U tests . In experiments involving three or more groups , treatment groups were compared using either one-way ANOVA or two-way ANOVA ( with a within-groups variable for repeated data ) . All post hoc comparisons were made using Tukey’s test with a correction for multiple comparisons . When neural precursor counts ( PCNA+ , tbr2+ , and GFAP+ cells ) were reanalyzed using the absolute number of cells/section ( instead of corrected for the number of Hoechst+/cells ) , similar results were obtained as present here .	Sensory experiences early in life help the brain to grow and develop . For example , raising animals in complete darkness stops the visual areas of their brain from forming properly . Previous studies have shown that sensory input helps to strengthen the connections between already existing brain cells , but it is unclear if it affects the actual creation of new brain cells . Conditions that reduce the mobility of young children , such as muscular disease , are often accompanied by learning difficulties . This suggests that physical movement may be important for healthy brain development . Scientists have previously found a link between exercise and an increased production of new brain cells in adults . However , such a link has not been established earlier in life , when the brain is developing the most . To address this , Hall and Tropepe studied how movement affects the brain development in zebrafish larvae , at an age when many of their brain cells are created . Restraining the larvae decreased their physical movement , while making them swim against a current increased their movement . Hall et al . looked at how this affected the larvae’s number of so called progenitor cells – the cells from which brain cells are created . When the larvae moved less , the number of progenitor cells decreased . But when they moved more frequently , the amount of progenitor cells increased . The experiments also showed that some sensory cells , which detect sensations associated with movement of the body during swimming , are linked to brain cell production . Blocking the development of these sensory cells prevented the rise in progenitor cells seen with increased movement in the larvae . However , activating these sensory cells in immobilised larvae increased the number of progenitor cells , even though the larvae could not move . These findings suggest that measures to increase physical movement in young children could be used to help the brain develop normally .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2018	Movement maintains forebrain neurogenesis via peripheral neural feedback in larval zebrafish
Molecular recognition reagents are key tools for understanding biological processes and are used universally by scientists to study protein expression , localisation and interactions . Antibodies remain the most widely used of such reagents and many show excellent performance , although some are poorly characterised or have stability or batch variability issues , supporting the use of alternative binding proteins as complementary reagents for many applications . Here we report on the use of Affimer proteins as research reagents . We selected 12 diverse molecular targets for Affimer selection to exemplify their use in common molecular and cellular applications including the ( a ) selection against various target molecules; ( b ) modulation of protein function in vitro and in vivo; ( c ) labelling of tumour antigens in mouse models; and ( d ) use in affinity fluorescence and super-resolution microscopy . This work shows that Affimer proteins , as is the case for other alternative binding scaffolds , represent complementary affinity reagents to antibodies for various molecular and cell biology applications . Our understanding of biological processes at the cellular level has been underpinned by the traditional disciplines of genetics , biochemistry , and molecular biology . Over the last decade , focus has shifted towards large-scale studies of genomes and transcriptomes , the latter as surrogates for cellular proteomes . These combined with high-throughput protein interaction studies , have led to the new discipline of Systems Biology , where proteins are considered in the context of networks of biochemical and developmental pathways . In the network view of protein behaviour , each protein or protein isoform may participate in many protein-protein interactions but available tools that allow researchers to test hypotheses in the biological context are lacking . Technologies such as RNAi and CRISPR-Cas9 that lower or ablate protein expression are important tools , but may cloud the interpretation of a proposed relationship between a given gene product or protein domain and the observed cellular phenotype . The next generation of tools should have the ability to block protein-protein interactions systematically without affecting expression levels . Commonly used tools for studying protein expression and function include antibodies . Antibodies have proved to be exquisite tools in many applications but there are growing concerns about the difficulty in sourcing validated and renewable antibodies ( Bordeaux et al . , 2010; Bradbury and Plückthun , 2015; Taussig et al . , 2007 ) . While there are over 500 , 000 different antibodies on the market , it has been reported that up to 75% have either not been validated , show a low level of validation or simply do not perform adequately in certain applications ( Berglund et al . , 2008 ) . In addition , the use of antibodies to block protein function inside living cells is commonly performed , but it is limited owing to the reducing environment of the cells ( Marschall et al . , 2015 ) . Even though antibody fragments , termed intrabodies ( Marschall et al . , 2015 ) or chromobodies ( Rothbauer et al . , 2006 ) can be expressed in the cytoplasm of mammalian cells , only a fraction of the repertoire of IgGs are correctly folded in the reducing environment of the cytoplasm ( Biocca et al . , 1995; Wörn and Plückthun , 2001 ) , decreasing their efficacy in functional applications ( Marschall et al . , 2015 ) . Various consortia ( Stoevesandt and Taussig , 2012 ) have been established to address the generation and validation of antibodies and their derivatives ( Berglund et al . , 2008; Renewable Protein Binder Working Group et al . , 2011; Nilsson et al . , 2005; Uhlén et al . , 2005 ) . These consortia have generated polyclonal and monoclonal antibodies against proteins and protein domains . Whilst proving successful , in providing a large catalogue of validated antibodies , such efforts have required large multidisciplinary groups across Europe and the US ( Uhlén et al . , 2015 ) . However , the ability to rapidly and cost-effectively generate renewable binding reagents for applications such as studying protein function both in vitro and in vivo and for proteomic projects would represent a major advance . Renewable binding reagents in this context refers to reagents that are recombinantly produced from a known sequence . The development of alternative binding proteins has provided the opportunity for such advances ( Škrlec et al . , 2015; Vazquez-Lombardi et al . , 2015 ) . These include reagents such as DARPins ( Binz et al . , 2003 ) , Monobodies ( Koide et al . , 1998 ) , and Affibodies ( Nord et al . , 1995 ) and a number of others ( for a recent review see [Škrlec et al . , 2015] ) . Over the past two decades these reagents have proved to be useful tools in many antibody-like applications including detection of proteins for diagnostics ( Theurillat et al . , 2010 ) , for studying protein function ( Kummer et al . , 2012 ) , intracellular targeting of protein function ( Spencer-Smith et al . , 2017; Wojcik et al . , 2010 ) and as crystallisation chaperones ( Sennhauser and Grütter , 2008 ) . In 2010 , the University of Leeds and Leeds NHS Teaching Hospital Trust established the BioScreening Technology Group ( BSTG ) , to allow rapid identification of alternative binding proteins against biological targets , particularly those of clinical interest . We now report on some of the outcomes of the more than 350 successful screens performed by the BSTG to date , and suggest that access to this and similar facilities ( eg High Throughput Binder Selection facility at the University of Zurich ) should deliver the tools needed to complement antibodies in the dissection of biological functions of individual proteins and protein isoforms . Our work is underpinned by the development of a new , engineered protein scaffold for peptide display ( Figure 1 ) . The Adhiron scaffold is a synthetic protein originally based on a cystatin consensus sequence and displays remarkable thermal stability ( Tm = 101°C ) ( Tiede et al . , 2014 ) . It is related in structure to a previously reported scaffold engineered from human stefin A ( Stadler et al . , 2011 ) . Binding proteins derived from these two scaffolds are now referred to collectively as Affimer proteins , and we use this term subsequently . 10 . 7554/eLife . 24903 . 003Figure 1 . Ribbon diagrams of three crystal structures for Affimer ( Adhiron ) reagents . ( A ) X-ray crystal structure of Affimer scaffold ( PDB ID no . 4N6T ) at 1 . 75 A resolution . The amino acids from the loops connecting the four anti-parallel beta sheets are highlighted in pink . ( B ) Crystal structure of an Affimer against p300 ( PDB ID no . 5A0O ) ( C ) Crystal structure of an Affimer isolated against human SUMO proteins ( PDB ID no . 5ELJ ) . The variable regions in B and C are shown in pink . DOI: http://dx . doi . org/10 . 7554/eLife . 24903 . 003 We have previously demonstrated use of Affimers in a number of assays including immune-like ( affinity ) assays , in biosensors and have tested their ability to be expressed in mammalian cells to manipulate cell signalling ( Tiede et al . , 2014; Kyle et al . , 2015; Rawlings et al . , 2015; Stadler et al . , 2014; Sharma et al . , 2016 ) . Here we have screened our established Affimer phage library ( Tiede et al . , 2014 ) against a broad range of targets , including homologous protein family members , to isolate highly specific and renewable binding reagents that can be used both in vitro and in vivo . For broad applicability , and to remove the bottleneck in target protein production , we also tested our ability to generate reagents against small quantities of target protein from commercial sources . We demonstrate the generation of Affimers against various target molecules , including a small organic molecule , and we report their use in a number of widely used biochemical and cell biology assays . A challenge in cell biology is to develop highly specific tools to detect and modulate the function of one member of a family of structurally and functionally similar proteins . Biological reagents that specifically target a single protein , or a subset of a family of proteins would introduce greater selectivity to in vivo studies . To demonstrate this functionality , we isolated Affimer binders to various Src-Homology 2 ( SH2 ) domains . SH2 domains are short ( ~100 amino acid ) protein domains that bind specifically to phosphotyrosine-containing motifs in partner proteins , but not to the de-phosphorylated isoforms . They have also recently been found to bind to signalling lipids ( Park et al . , 2016 ) and are involved in mediating multiple aspects of cellular signal transduction and communication . The human genome encodes some 120 SH2 domains found in 111 proteins ( Liu et al . , 2011 ) . The ability to specifically detect and inactivate each SH2 domain is a rate limiting step in our understanding of these pathways; the use of siRNA , for example , may be used to remove an entire protein , such as the protein kinases Syk or Zap70 , from a cell but will not allow determination of which of the two SH2 domains , carried by each of these kinases , mediate which signalling event ( s ) . The ability to dissect these signalling events with highly specific binding reagents has already identified new biological function using monobodies ( Wojcik et al . , 2010; Grebien et al . , 2011; Sha et al . , 2013 ) . We have addressed whether alternative binders can target a specific SH2 domain by selecting Affimers against a range of SH2 domains . We chose five SH2 domains , some of which had previously been targeted using antibodies ( Renewable Protein Binder Working Group et al . , 2011; Pershad et al . , 2010 ) . In these previous reports , highly specific binding reagents were identified against the recombinant protein but only a limited number worked efficiently in the tested applications ( Renewable Protein Binder Working Group et al . , 2011 ) . We have previously demonstrated the ability to isolate reagents against the Grb2 SH2 domain ( Tiede et al . , 2014 ) . In the present study we adopted a different target capture strategy by producing each SH2 domain with an N-terminal biotin acceptor peptide to facilitate simple direct capture from cell lysate and presentation for phage display screening . Each target was checked for efficient biotinylation by Western blot ( Figure 2—figure supplement 1 ) and it is noteworthy that this biotinylation was achieved in a BL21 ( DE3 ) -derived strain without the need for additional biotin ligase expression . From each screen we randomly selected phagemid clones and by phage ELISA confirmed that Affimers had been selected against each of the nine SH2 domains . The proportion of clones that bound to each target , but not to the control , was between 50% and 100% with an average of 87 . 6% . Next , we assessed Affimer target specificity by phage ELISA ( Figure 2A ) . Grb proteins are growth factor receptor-bound proteins which contain SH2 domains . Initially , the Grb2 , 7 , 10 and 14 SH2 domain binding Affimers were tested for cross-reactivity against the other Grb family members , and showed specific binding to the Grb SH2 domain , with the exception of Grb14 Affimers which showed weak cross-reactivity with Grb7 and Grb10 proteins but not Grb2 . The level of pairwise sequence homology between Grb7 , 10 and 14 is between 65–72% ( Daly , 1998 ) . It is notable that Affimers were isolated that bind specifically to Grb7 and Grb10 without the need for negative panning to remove cross-reactive binders . We predict that screens that include pre-panning against similar domains would results in isolation of specific Affimers that can bind Grb14 only . 10 . 7554/eLife . 24903 . 004Figure 2 . Isolation and characterisation of SH2 domain binding Affimers . ( A ) Phage ELISA from 24 monoclonal Affimer reagents isolated against the respective Grb family member SH2 domains . Specificity was tested through extent of binding to the other SH2 family members . ( B ) Western blot showing Affimer-mediated affinity-precipitation of endogenously expressed Grb2 protein from U2OS cell lysates using five Grb2 Affimers bound to colbalt magnetic beads ( n = 2 ) . A yeast SUMO binding Affimer was used as a negative control . ( C ) Phage ELISA from 24 monoclonal Affimer reagents isolated against p85 alpha N-terminal domain family member SH2 domain . Specificity was tested through extent of binding to the other p85 SH2 family members . ( D ) Western blot of immunoprecipitation using a p110 antibody on cell lysates from cells expressing p85 SH2 domain binding Affimers ( n = 3 ) . ( E ) Western blot and quantification by densitometry of AKT phosphorylation in the presence of expressed p85 SH2 domain binding Affimers ( n = 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24903 . 00410 . 7554/eLife . 24903 . 005Figure 2—figure supplement 1 . Western blot results of Avi-Tag SH2 domain proteins using an streptavidin-HRP conjugate to detect the presence of biotin . DOI: http://dx . doi . org/10 . 7554/eLife . 24903 . 005 We then examined the ability of Affimer reagents to bind to endogenous proteins . Five of the Grb2-binding Affimers were purified and bound to cobalt-based magnetic beads and their ability to pull down endogenous Grb2 from cell lysates of the human U2OS cell line was assessed ( Figure 2B , n = 2 ) . All five reagents successfully pulled-down Grb2 whereas a yeast SUMO-binding control Affimer ( Tiede et al . , 2014 ) was unable to pull down Grb2 . To further assess the ability to isolate isoform specific Affimers we investigated phosphoinositide 3-kinase ( PI3K ) a heterodimeric protein comprising a p110 catalytic subunit and a p85/p55 regulatory subunit . We examined the specificity of Affimers , raised against the N-terminal SH2 domain of the p85α variant , for cross-reactivity with the p85β and p55γ variant N-terminal SH2 domains and against the C-terminal SH2 domains of all three isoforms ( Figure 2C ) . Despite a high degree of sequence identity ( pairwise between 83–90% ) a number of p85α-specific Affimers were isolated ( e . g . , clones 1 and 2; Figure 2C ) . Affimers that recognised the α and γ but not β domain were also isolated ( clones 3 and 23 ) . None of these Affimers bound to any of the C-terminal p85/p55 SH2 domains . These results further demonstrate the ability to isolate Affimers that show high binding specificity against related targets , even within a single protein . The p85 SH2 domain-specific Affimers were expressed in NIH 3T3 cells and their ability to bind to endogenous p85 protein was assessed by co-immunopreciptation assays , in which p85α was pulled-down . The p85α antibody also pulled down both p110α and the FLAG-tagged Affimers ( Figure 2D ) . The different levels of Affimer recovered may be due to differences in Affimer expression levels , as shown in Figure 2E , and any differences in binding affinity . It is interesting to note that Affimer 1 appears to bind to endogenous p85 with high affinity ( Figure 2D ) but has little effect on signalling ( Figure 2E ) suggesting it binds outside the key SH2 interaction region . In addition the Affimers did not disrupt the p85/p110 overall complex in which p85 interacts with p110 via three domains , with the SH2 domain regulating activity via binding p110 ( Vivanco and Sawyers , 2002 ) . These results demonstrate that the Affimer is specifically binding the SH2 domain interaction without affecting the two other p85/p110 binding domains . We therefore assessed the ability of the Affimers to block the function of the p85 N-terminal SH2 domain by examining whether the Affimers led to an increase in phosphorylated protein kinase B ( AKT ) , a downstream effector of p110 . Five of the six Affimer proteins mediated an increase in AKT phosphorylation ( Figure 2E ) demonstrating that they inhibit the interaction between the N-terminal p85 SH2 domain and p110 , but importantly do not block p85-p110 complex formation . This supports a report that siRNA inhibition of p85α alone had little effect on cells , but that p85 and p110 both had to be eliminated to produce a phenotype and an effect on AKT phosphorylation ( Kim et al . , 2005 ) . Thus our data highlights a benefit of Affimers , and potentially other alternative reagents , for studying protein-protein interactions within the cellular context . Vascular Endothelial Growth Factors ( VEGFs ) are a family of secreted proteins that regulate many aspects of vascular and lymphatic biology including vasculogenesis ( de novo formation of the vascular system ) , angiogenesis ( formation of new capillaries e . g . in response to hypoxia ) , lymphangiogenesis ( de novo formation of the lymphatic system ) and arteriogenesis ( formation of new arteries e . g . following ischemia ) . The biological effects of the VEGF family are mediated through binding to a membrane-bound vascular endothelial growth factor receptor ( VEGFR ) tyrosine kinase subfamily comprising VEGFR1 , 2 and 3 . While VEGFR1 is implicated as a negative regulator of angiogenesis , VEGFR2 is a major regulator of vasculogenesis , angiogenesis and arteriogenesis . VEGFR3 activation is implicated in specifying lymphangiogenesis but cross-talk between the different VEGFRs can modulate these different processes ( Aspelund et al . , 2016 ) . Dissecting the roles of the different VEGFRs is an important goal , particularly given the success of therapeutic agents targeting VEGF-A in diseases ranging from metastatic cancer to macular degeneration . In this context , VEGFR2 is a key molecule that regulates many aspects of vascular physiology and blood vessel formation especially angiogenesis and is associated with tumour neovascularisation ( Kofler and Simons , 2015 ) . To evaluate whether Affimer proteins that perturb VEGFR2 function could be selected we screened against VEGFR2 and then tested Affimers for their ability to bind recombinant VEGFR2 protein in vitro ( Figure 3A ) . In this case DNA sequence analysis revealed that the positive clones represented only two distinct sequences . The affinities for VEGFR2 of representative Affimer proteins , A9 and B8 , were determined by SPR to be 41 ± 17 nM and 240 ± 124 nM , respectively ( Figure 3—figure supplement 1 ) . The Affimer proteins were then labelled at the C-terminal cysteine with a single biotin moiety and used to probe various tissue types for specific staining to compare with the pattern produced by a commercially available polyclonal antibody ( Figure 3B ) . The efficiency of Affimer labelling with biotin was determined to be 80–90% by mass spectrometry ( data not shown ) . To directly compare antibody and Affimer patterns a biotinylated secondary antibody was used to detect binding of the primary anti-VEGFR2 antibody . Subsequently , both antibody and Affimer binding were detected by streptavidin-coupled horseradish peroxidase activity . The Affimer reagents showed exactly the same staining pattern as the antibodies , with VEGFR2 staining being predominantly localised in the epithelial cells and with more intense staining at the cell membrane ( Figure 3B; see arrows ) . In this case , the staining developed more quickly for the Affimer binders than for the antibody indicating a higher sensitivity of staining . 10 . 7554/eLife . 24903 . 006Figure 3 . Characterisation of VEGFR2 binding Affimers . ( A ) Phage ELISA for 32 monoclonal Affimer reagents isolated against VEGFR2 . The negative control contained just streptavidin . ( B ) Immuno- and affinity-histochemistry of a polyclonal anti-VEGFR2 antibody and of representative Affimers B8 and A9 . Staining is shown as a light brown color , haemotoxylin counter staining ( blue ) . Arrows show similar staining patterns . ( C ) Tubulogenesis assay in the presence and absence of vascular endothelial growth factor A and the two Affimers with quantification of tubule length and branch point number shown to the right . The control is in the absence of any Affimer and the control Affimer is a binder against yeast SUMO ( n = 3 ) . Statistical analysis was performed using a two-way ANOVA followed by the Bonferroni multiple comparison test using GraphPad Prism software ( La Jolla , USA ) . p values less than 0 . 05 ( * ) , 0 . 01 ( ** ) are indicated on the graphs . Error bars in graphs denote ± standard error of mean . ( D ) Western blot results showing changes in downstream signalling in HUVECs treated for 0 , 5 and 15 min in the presence of vascular endothelial growth factor A and increasing concentrations of the VEGFR2 binding Affimer B8 ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24903 . 00610 . 7554/eLife . 24903 . 007Figure 3—figure supplement 1 . SPR plots for the anti-VEGFR2 , TNC and TNT binding Affimers . DOI: http://dx . doi . org/10 . 7554/eLife . 24903 . 007 Although immunohistochemistry is a qualitative rather than quantitative technique , this is an interesting observation given the apparently modest binding affinities , the monomeric nature , and the mono-biotinylated state of the Affimer binders compared to the bivalent nature of , multiply biotinylated polyclonal antibody molecules . It demonstrates the value of Affimers as affinity histo-chemistry reagents . The differential sensitivity of staining may be due to the difference in size between antibody and Affimer with the latter better able to penetrate the fixed tissue more efficiently . The Affimer may also have a more exposed binding site compared to the antibody . One or more of these properties may allow a greater number of binding events to the target resulting in higher sensitivity of Affimer staining . Alternative scaffolds have been reported to inhibit VEGFR2 including Nanobody ( Behdani et al . , 2012 ) , Adnectin ( Tolcher et al . , 2011 ) , Affibody ( Fleetwood et al . , 2014 ) and DARPin ( Hyde et al . , 2012 ) proteins , so we questioned whether the Affimer proteins could also inhibit VEGFR2 signalling in human vascular endothelial cells ( HUVECs ) . Previous siRNA studies ( Murga et al . , 2005 ) have shown that VEGFR2 signalling is required for the formation of vascular tubules by transfected HUVECs , although this siRNA-mediated effect requires 24–48 hr following transfection . By contrast , the inhibitory effect of Affimer B8 could be measured within just 30 min of treatment and also led to a decrease in VEGF-dependent tubule length and branch point formation in a tubulogenesis assay ( Figure 3C ) . Consistent with the effects on tubulogenesis , Affimer B8 also inhibited VEGF-dependent phosphorylation of VEGFR2 and downstream signalling , with decreased activation of cell signalling mediators PLCg1 , AKT , ERK , p38 and eNOS ( n = 3; Figure 3D ) . By contrast control Affimers had no effect on signalling . Overall these observations demonstrate that Affimers represent useful research reagents that are capable of blocking the biological function of specific receptors on biologically-relevant timescales . Ion channels are involved in a number of physiological processes , and are important drug targets ( Overington et al . , 2006 ) . However , there remains a lack of reagents able to modulate ion channels with the selectivity and specificity required to prevent off-target effects ( Skerratt and West , 2015 ) . Antibodies have proven to be useful as ion channel imaging reagents and have recently shown promise as therapeutics ( Lee et al . , 2014; Sun and Li , 2013 ) . Complementing the repertoire of antibodies available , smaller biologics are increasingly being used to study ion channels , for example , by providing crystallization chaperones ( Stockbridge et al . , 2015; Zhou et al . , 2001 ) . Furthermore , the high selectivity often associated with such biologics alongside their ability to access functional crevices may provide further opportunities to modulate ion channel function . Indeed , the targeting of both ligand and voltage-gated ion channels by Nanobodies and scFv’s , respectively , has already demonstrated this potential ( Danquah et al . , 2016; Harley et al . , 2016 ) . Here , we set out to isolate Affimers capable of binding to and modulating the activation of the Transient Receptor Potential Vanilloid 1 ( TRPV1 ) ion channel by screening against a peptide derived from the outer pore domain . Thirteen unique Affimer clones were identified from 24 positive clones identified by phage ELISA of 96 randomly selected colonies from the phage library screen ( Figure 4A ) . None of the 13 binders showed cross-reactivity to a distinct peptide derived from the pore region of a voltage-gated sodium channel , Nav1 . 7 . Affinity-fluorescence studies were performed to examine the ability of the Affimer proteins as detection reagents . Only Affimer 2 stained U2-OS cells expressing full-length TRPV1 ( Figure 4B ) showing co-localisation with an anti-TRPV1 antibody ( Figure 4C ) . Affimer 2 showed no staining of TRPV1-negative U2-OS control cells . None of the other 12 binders worked in this assay . 10 . 7554/eLife . 24903 . 008Figure 4 . Characterisation of TRPV1 binding Affimers . ( A ) Phage ELISA for 96 monoclonal Affimer reagents isolated against TRPV1 peptide . The negative control contained a different hydrophobic peptide sequence . ( B ) Affinity-cytochemistry on U2-OS cells transiently transfected with TRPV1 ( TRPV1+ ) or control ( TRPV1- ) using Affimer 2 . Binding was detected using an anti-HIS antibody fluorescently labeled with FITC . Binding of the Affimer is shown as a green and DAPI ( a DNA stain ) shown as blue ( n = 3 ) , ( C ) Co-localisation of Affimer staining with an anti-TRPV1 antibody . Antibody staining is shown in red . ( D ) A Flexstation was used to measure uptake of Fluo-4 AM , a calcium binding fluorescent small molecule , to measure calcium levels in capsaicin stimulated cells in the presence of Affimer control and TPRV1-binding Affimers ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 24903 . 008 Next we investigated TRPV1 modulation by measuring the levels of intracellular calcium in response to treatment with the Affimer proteins . While no direct modulation was observed six Affimers showed significant enhancement of TRPV1 activation upon treatment with the agonist capsaicin ( Figure 4D ) compared with cells treated with capsaicin alone . Previous research has explored the use of small molecule compounds as positive modulators of TRPV1 to desensitize and reduce pain ( Kaszas et al . , 2012 ) . The compound ( MRS1477 ) was hypothesised to interact with the pore-forming region of TRPV1 , leading to a three-fold increase in capsaicin activation when applied at low micromolar concentrations – an effect similar to that reported here for some of the Affimers . Overall , this study demonstrates that Affimer proteins can be raised against a peptide surrogate to recognise and alter ion channel function by positive allosteric modulation , a suggested mechanism for the treatment of TRPV1-induced chronic pain ( Lebovitz et al . , 2012 ) and may represent a novel approach and therapeutic strategy for chronic pain relief . Tenascin C ( TNC ) is an extracellular matrix protein that is abundant during early development , is expressed at low levels in adult tissues and is frequently up-regulated in cancer tissues and associated with metastasis ( Minn et al . , 2005; Oskarsson et al . , 2011 ) and poor patient outcomes ( Lowy and Oskarsson , 2015 ) . As such , it offers potential as a tumour marker for imaging and/or therapeutic targeting in vivo ( Hicke et al . , 2006 ) . Affimer binders to TNC were isolated from the phage display library ( Figure 5A ) . One Affimer protein with high affinity for TNC ( KD = 5 . 7 ± 2 . 8 nM by SPR – sup Figure 2 ) was used in subsequent assays . To evaluate its specificity for TNC we compared the staining pattern of the Affimer to that of an anti-TNC antibody in human colorectal cancer and glioblastoma xenograft tissue sections . Staining patterns with C-terminally biotinylated TNC Affimer were similar to those obtained with the TNC antibody ( Figure 5B ) . 10 . 7554/eLife . 24903 . 009Figure 5 . Characterisation of tenascin C ( TNC ) binding Affimer by affinity-histochemistry and ex vivo imaging of xenografts . ( A ) Phage ELISA for 48 monoclonal Affimers against TNC . The two controls are tenascin X ( TNX ) and streptavidin . ( B ) Immunohistochemistry of serial sections of a mouse xenograft ( SW620 cell line ) , showing staining for TNC . Antibody/Affimer staining is shown as a light brown color with haemotoxylin counter staining ( blue ) . ( C ) and ( D ) Mice were injected via their tail vein with rhodamine labelled TNC binding Affimer or a control GFP binding Affimer . After 24 , 48 , 72 and 96 hr the xenograft and organs were removed and visualized . ( C ) Organ images at 24 hr . ( D ) Quantification of rhodamine fluorescence ( radiant efficiency in p/s/cm2/sr/μW/cm2 ) ex vivo ( n = 3 ) . Mean background fluorescence intensity was normalized to sham injected control tumors and organs . DOI: http://dx . doi . org/10 . 7554/eLife . 24903 . 009 In the clinic , tumour imaging using labelled antibodies can be limited by a high background of circulating labelled antibody until this is cleared from the body . This leads to extended hospital stays or the necessity for multiple patient visits for a single test . By contrast , the smaller size of alternative binding proteins means that the molecules that do not bind to the target , will be more quickly cleared from the circulatory system , more conveniently allowing visualisation shortly after imaging agent administration . To demonstrate this in tumour bearing mice we visualised distribution of TNC Affimer compared to a control GFP-binding Affimer , both C-terminally labelled with Rhodamine Red ( Figure 5C ) . To maximise signal detection , we imaged excised tumours and organs post-sacrifice and quantitated the signal as fold-change above background . As expected , both Affimer probes were detected in kidney indicating renal clearance . However , compared to the TNC Affimer this clearance was faster for the GFP Affimer as it showed a significant decrease ( p=0 . 04 ) in fluorescence from 24 to 48 hr post-injection ( fold change 9 . 23 ± 3 . 10 at 24 hr to 2 . 98 ± 0 . 77 at 48 hr; Figure 5D ) . The TNC Affimer signal at 24 hr post-injection was significantly higher ( p=0 . 02 ) in tumours ( 6 . 26 ± 1 . 62 ) compared with the control GFP Affimer group ( 2 . 32 ± 0 . 61 ) suggesting that the TNC Affimer accumulated in the TNC expressing tumour . The TNC probe was also detected in liver tissues either due to hepatobiliary clearance or due to the fact that TNC shows low level expression in normal liver sinusoids ( Van Eyken et al . , 1990 ) . In addition , the ratio of anti-TNC Affimer binders in tumour compared to the spleen , for example , was >6 at 24 hr ( Figure 5D ) ; in contrast , anti-TNC antibodies took 2 days to reach a tumour/spleen ratio of 5 , although this did improve to 20–30 at day 10 ( De Santis et al . , 2006 ) . Thus the more rapid clearance rate of alternative binding proteins , such as Affimers , compared to antibodies has the potential to allow more rapid imaging of tumours . Further work to enhance signal detection in vivo with Affimers is underway ( Fisher et al . , 2015 ) . Marek’s Disease , caused by Marek’s Disease Virus ( MDV-1 ) , is a globally and economically significant neoplastic disease of chickens that is currently controlled by vaccination with the related Herpes Virus of Turkeys ( HVT ) . In field samples , tests for Marek’s Disease would need to be able to distinguish between proteins from HVT and their homologues in MDV . As a proof of principle we screened the phage library against HTV-derived protein UL49 , with counter screens against host proteins as well as the related proteins MDV ( RB1B ) and DEV UL49 . Phage ELISA , affinity-fluorescence and in-cell Western confirmed that the selected Affimers were specific for HTV recombinant proteins as well as their ability to specifically stain the target protein in primary Chicken Embryonic Fibroblasts ( CEFs ) containing bacterial artificial chromosome ( BAC ) clones of MDV-1 ( RB1B ) , HVT or DEV ( strain 2085 ) ( Figure 6A , B and C ) . 10 . 7554/eLife . 24903 . 010Figure 6 . Affimer detection of HVT UL49 in infected cells by in-cell Western and affinity-fluorescence . ( A ) Phage ELISA for 24 monoclonal Affimers against HVT screened against HVT , RB1B , DEV and CEF lysates to confirm specificity for HVT . ( B ) Infection of cells was confirmed using a goat anti-GFP antibody and donkey anti-goat 680 ( green ) antibody to detect GFP which is constitutively expressed by the BAC derived viruses . Infected CEFs were screened with candidate HVT UL49 Affimers at 1 . 5 µg/ml with subsequent detection ( red ) by Streptavidin 800 conjugate ( Licor ) . ( n = 3 ) ( D ) HVT infected CEFs were screened with HVT UL49 Affimers and visualised with Streptavidin-568 ( red ) . Nuclei were stained with DAPI ( blue ) . Streptavidin only control ( no Affimer ) shows no observable labelling of infected CEFs . Bar = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 24903 . 01010 . 7554/eLife . 24903 . 011Figure 6—figure supplement 1 . Bilayer Interferometry plots for the HVT binding Affimers . DOI: http://dx . doi . org/10 . 7554/eLife . 24903 . 011 The affinity of anti HVT UL49 Affimers was in the low nM range , KD values of 1 . 5 nM to 7 . 5 nM with a mean 3 . 3 nM , for the eight clones tested ( Figure 6—figure supplement 1 ) . The high specificity and affinity should be advantageous in the development of DIVA ( differentiating infected from vaccinated animals ) tests for the discrimination of vaccine and field strain viruses . We tested the performance of anti-HVT UL49 Affimers in affinity-fluorescence ( Figure 6C ) . Cultures of primary CEFs infected with HVT GFP BAC clone were subjected to affinity-fluorescence staining using biotinylated anti HVT UL49 Affimers and were visualised with streptavidin-Alexa Fluor 568 conjugate ( Figure 6C ) . Compared with low background staining with the streptavidin only control ( SA only ) there are pronounced cytoplasmic foci detected by Affimers in infected cells . These foci are consistent with data from the related alphaherpesvirus MDV ( Denesvre et al . , 2007; Rémy et al . , 2013 ) or the model alphaherpesviruses Herpes Simplex type 1 ( HSV-1 ) ( Stylianou et al . , 2009 ) and Pseudorabies virus ( PrV ) ( del Rio et al . , 2002 ) and likely indicate the cytoplasmic sites of HVT secondary viral envelopment . This distribution is also consistent for the different Affimer clones tested and is seen only within infected cells . Thus Affimers show promise as alternatives to traditional antibodies and are likely to be particularly valuable where availability/performance of existing antibody reagents is poor . Super-resolution microscopy provides the ability to localise proteins within a cell at ca . 20 nanometer resolution . A major limitation of wide-spread exploitation of this approach is the lack of highly specific reagents that can place the fluorophore in close proximity to the endogenous target protein . Antibodies are large multi-domain proteins that are normally labelled with fluorophores at random sites that limits the achievable resolution . By contrast , the smaller Affimer proteins can be labelled in a site-specific manner providing closer spatial placement of the fluorophore to the target protein , thus facilitating use of current super-resolution techniques . This approach has recently been demonstrated using Nanobodies where super-resolution microscopy was used to image GFP-tagged proteins and nuclear pore complex ( Pleiner et al . , 2015; Ries et al . , 2012 ) . Human epidermal growth factor receptor 4 ( HER4 ) , also known as c-erbB-4 , is an oncogenic transmembrane receptor protein kinase ( Lemmon and Schlessinger , 2010 ) . Although the function of this protein is not yet fully understood , it is known to be associated with increased survival and lower proliferation in breast cancer patients ( Machleidt et al . , 2013 ) . We screened the phage display library against HER4 ( Figure 7A ) and two Affimers were recombinantly produced with a C-terminal cysteine for labelling with the fluorophores Alexa Fluor 647 or CF640R maleimide . The Affimer showing the highest signal by fluorescent imaging was used for further studies . Our results show that the HER4 Affimer can bind both to CHO cells transiently expressing HER4 and to MCF7 , a breast cancer cell line expressing lower physiological levels of HER4 ( Figure 7 ) . When HER4 is over-expressed in CHO cells the Affimer showed increasing binding at concentrations from 5 nM to 100 nM , as determined by membrane signal intensity from confocal microscopy images , while in MCF7 cells that express physiological levels of HER4 binding increases from 10 to 200 nM Affimer ( Figure 7B ) . 10 . 7554/eLife . 24903 . 012Figure 7 . Use of HER4 binding Affimers in super-resolution imaging and single molecule tracking . ( A ) Phage ELISA for HER4 binding Affimers . ( B ) Average photon counts/pixel for HER4-binding Affimer labelled with CF640R and bound to CHO cells transfected with HER4 and to MCF7 cells expressing endogenous levels of HER4 . ( C ) Wide field image of CHO cells transfected with HER4-CYT-eGFP showing localisation of HER4 via GFP fluorescence ( top ) and labelled with HER Affimer–Alexa647 ( middle ) . The corresponding dSTORM image of HER4 Affimer conjugated to Alexa647 ( bottom ) with a 25 nm localisation precision . Scale bar = 2 μm . Right plots to show the number of molecules and cluster size of clusters identified by dSTORM . ( D ) Diffusion coefficients ( left panel ) , and MSD curve ( right panel ) of HER4 Affimers labelled with CF640R and tracked on MCF7 cells expressing endogenous HER4 . DOI: http://dx . doi . org/10 . 7554/eLife . 24903 . 012 Dual colour wide field fluorescence images of HER4 receptor fused at the intracellular C-terminal end with eGFP ( HER4-CYT-eGFP ) in CHO cells ( Figure 7C - top ) and labelled in the extracellular region with Affimer-Alexa 647 ( Figure 7C middle ) shows that the HER4 Affimer can be used to specifically label membrane-associated HER4 , through co-localisation of GFP and labelled Affimer fluorescence . The corresponding direct Stochastic Optical Reconstruction Microscopy ( dSTORM ) image ( Figure 7C – bottom ) has a localisation precision of ca . 25 nm . A Bayesian cluster analysis ( Griffié et al . , 2016 ) of the dSTORM image shows that the most prevalent cluster size of HER4 oligomers is between 8 . 1 and 12 nm in radius . This corresponds with the most prevalent number of HER4 molecules in a cluster being between 4 to 8 . These data show that HER4 forms oligomers as large as those previously found in EGFR . ( Needham et al . , 2016 ) This Affimer is also suitable to image HER4 under Total Internal Reflection Fluorescence ( TIRF ) -mode to undertake single-particle tracking on live cells ( Figure 7D ) . To detect single particles , the binding affinity of the Affimer must be in the low nM range to avoid saturating the sample and to reduce non-specific binding . HER4 particles were tracked with a Bayesian tracking algorithm ( Rolfe et al . , 2011 ) and the diffusion coefficient and Minimum Square Displacement ( MSD ) were calculated from the resulting trajectories . The data show that there is an immobile , or scarcely mobile , population of HER4 receptors on MCF7 cells , associated with a tail of highly mobile molecules ( Figure 7D , left panel ) . The near straight slope of the MSD plot indicates that , unlike EGFR ( Needham et al . , 2016; Zanetti-Domingues et al . , 2012 ) , the diffusion of HER4 is not confined on the timescales investigated ( Figure 7D , right panel ) . The Affimers raised against HER4 demonstrate the ability to isolate reagents that can be used in a range of super-resolution microscopy techniques . However , as there is no direct comparison to an antibody this example does not highlight the advantage of alternative proteins over the larger antibody probes . To provide this demonstration Affimers have also been raised against polymerised microtubules ( Figure 8 ) . The Affimer we selected labels interphase microtubules in a similar way to a widely-used antibody ( Figure 8A ) . However , in mitotic cells , the Affimer labels the spindle but not astral microtubules ( Figure 8A ) likely reflecting the fact that the antibody recognises tyrosinated microtubules unlike the Affimer . Interestingly , the Affimer is able to label the central region of the cytokinetic furrow ( Figure 8A ) , in which microtubules are very densely packed . Antibodies are usually excluded from this region so analysis of this feature has been problematic ( Hu et al . , 2012 ) . This highlights one advantage of using smaller probes , such as alternative binding proteins , for super-resolution microscopy and for example in this case will allow further elucidation of the role tubulin plays in cytokinetic furrows . 10 . 7554/eLife . 24903 . 013Figure 8 . Use of tubulin binding Affimer in super-resolution microscopy . ( A ) Confocal images of microtubules in HeLa cells , stained with a rat α-tubulin antibody ( YL1/2 ) which recognises tyrosinated tubulin , and an Affimer for polymerised tubulin , conjugated to Alexa Fluor 647 . Images of an interphase and metaphase cell , together with an image of the cytokinetic furrow are shown . Arrows in the metaphase cell point to astral microtubules that are predominantly labelled with the antibody . Arrows in the cyokinetic furrow indicate the central region ( Fleming body ) . Scale bar is 10 μm . ( B ) 3D dSTORM images of microtubules in a HeLa cell , labelled with Alexa Fluor 647 conjugated to a primary antibody to rat α-tubulin ( left ) and an Affimer for polymerised tubulin ( right ) . These images are from separate cells . Localisations were aggregated into 10 nm bins and projected onto a single plane , with Gaussian smoothing . Scale bar 1 µm . ( C ) Intensity profile across the microtubule image labelled in ( B ) ( yellow box ) , averaged along 510 nm of its length . The central decrease in intensity reflects the hollow structure of the microtubule . ( D ) Comparison of the average microtubule image intensity profile with antibody staining ( dashed , mean of 6 microtubule sections ) , Affimer staining ( solid , mean of 8 microtubule sections ) and actual microtubule size ( black circle ) . The FWHM of each average profile ( as in ( C ) ) was found for a Gaussian fit and a Gaussian distribution is plotted here using the mean FWHM for each staining method . DOI: http://dx . doi . org/10 . 7554/eLife . 24903 . 013 3D dSTORM images of microtubules using the antibody and Affimer look similar ( Figure 8B ) with analysis showing that Affimer labelling has the expected central decrease in fluorescence for binding to the outside of the microtubule ( Figure 8C ) . However , averaging profiles for multiple microtubules for both Affimer and antibody shows the increased localisation accuracy with Affimers , compared with antibodies ( Figure 8D ) . While localisation density may not be fully optimised in these samples ( Huang et al . , 2008 ) the average microtubule profiles were substantially narrower with Affimer labelling ( 47 ± 11 nm ) than with primary antibody labelling ( 73 ± 10 nm ) ( FWHM , mean ± s . d . ) and should allow further elucidation of tubulin structures that have previously not been solved . Overall , Affimers , and presumably other alternative binding proteins , have an advantage over antibodies in labelling for dSTORM . The generation of effective binding reagents to low molecular mass organic compounds is technically challenging . Small molecules do not display innate immunogenicity and thus are typically conjugated to carrier proteins to elicit an effective immune response . Even so it can be a problem raising an immune response to toxic molecules and those that conjugate poorly to carrier proteins . To examine whether we could isolate Affimer reagents against a small organic molecule we used 2 , 4 , 6-trinitrotoluene ( TNT ) . Previous studies have shown that presentation of TNT as a hapten for antibody production is known to be vital for the successful isolation of TNT specific antibodies ( Ramin and Weller , 2012 ) . The TNT analogue 2 , 4 , 6-trinitobenzene sulphonic acid ( TNBS ) ( Figure 9A ) contains nitro-groups ( NO2 ) located in the same positions as TNT on the benzene ring , while the methyl ( CH3 ) group is substituted by a sulfonic acid ( SO2OH ) group . This functional group reacts with primary amines and was used to prepare both TNBS-ovalbumin and TNBS-IgG conjugates for phage display screening , with counter screens performed against ovalbumin and IgG to enrich for small molecule binding . 10 . 7554/eLife . 24903 . 014Figure 9 . Affimer selection and specificity against TNT and DNT’s . ( A ) Chemical structures of TNBS , TNT , 2 , 3-DNT , 2 , 4-DNT and 2 , 6-DNT . ( B ) Phage ELISA results from 32 monoclonal Affimer reagents isolated against TNBS bound to ovalbumin . Binding specificity was also tested against TNBS bound to IgG and to unconjugated ovalbumin and IgG . ( C ) Competition ELISA of four TNT-Affimers to check specificity against a range of molecules across a concentration profile . Error bars = standard deviation from technical repeats of a representative ELISA . DOI: http://dx . doi . org/10 . 7554/eLife . 24903 . 014 To confirm binding specificity selected clones were tested against both TNBS conjugates and unconjugated protein by phage ELISA ( Figure 9B ) . The number of clones that showed strong binding to TNBS conjugated with both ovalbumin and IgG was relatively high ( 22/32 ) with a further 8/32 showing reasonable binding . Of the 32 clones tested 14 distinct sequences were identified , some containing only variable region 1 which indicates selection from a small sub-population of the original phage display library . Within variable loop 1 it was possible to define a short consensus sequence implying a common binding mode . Four Affimer proteins were purified and tested for binding to TNT and various dinitrotoluenes ( DNT; Figure 9A ) by competition ELISA ( Figure 9C ) . All four Affimers showed binding to the original TNBS-conjugate and to TNT , but differed in their specificity for the DNTs . Affimer 4 , a VR1 only Affimer , showed a higher level of specificity for TNT than any of the DNTs . By contrast , Affimer 3 binds to TNT and also provides discrimination between 2 , 4-DNT and the other two DNT’s . This high selectivity of recognition of the nitro group on position 4 , demonstrates that Affimer reagents can show remarkable specificity for such small molecular differences . It is highly likely that altering the panning strategy to include competition steps with analogues would allow selection of specificity and sensitivity of Affimers for small organic molecule targets . The ability of Affimer proteins to bind to small molecules raises the possibility that they may be used in cells to quench the effects of molecules such as Shield or doxorubicin that are currently used to regulate protein behaviour , allowing investigators to assess the effects of switching protein interactions off with the same speed with which they are currently turned on . The ability to rapidly isolate highly specific alternative binding protein affinity reagents that perform consistently in a wide range of scientific applications is the ‘holy grail’ for producing renewable binding reagents . For a recently developed artificial binding protein scaffold , known as Affimer , we demonstrate such applicability across a range of molecular and cellular studies . We have isolated Affimer proteins against more than 350 targets , but here we have exemplified their use as molecular and cellular tools against 12 different target molecules . Typically each screening regime , consisting normally of three panning rounds of phage display and phage ELISA , was normally completed in 12 days . Thus Affimer selection , as with other alternative protein or antibody fragment selection approaches that use phage display or other in vitro selection , is much faster than antibody and nanobody production techniques that involve animal inoculation . They also allow the efficient identification of binding reagents against conformational epitopes since target proteins are screening in their folded state . Each selected Affimer coding region was sub-cloned into an E . coli expression vector and recombinant protein was purified over a further seven days . Without automation the phage display platform allows an individual to screen up to 24 targets simultaneously . The benefit of such manual screening is that greater control can be exercised to moderate individual screening regimes within a set of samples . A further advantage of Affimer proteins is the ability to express recombinant protein at high yield in E . coli . Of the 36 Affimers reported on here an average yield of 83 . 3 mg/L ( 1 . 5–188 mg/L ) culture was achieved with a purity of greater than 95% following a single immobilised metal affinity step . We have not attempted to optimise the level of protein production but typically only grow 50 mL cultures for protein purification that provides suitable quantities of protein for most applications . The key to successful isolation of Affimers for cellular studies is the use of high-quality antigens normally presented via biotin/streptavidin on plates and beads . Recombinant sources of protein are normally of high quality as time and effort is taken to purify the protein . For some recombinant proteins , particularly those being expressed in mammalian cells , this can be more challenging . We expressed the small SH2 domains with an N-terminal biotin acceptor peptide to facilitate site-specific in vivo biotinylation allowing target immobilisation onto streptavidin plates , directly from cell lysate ( Figure 2—figure supplement 1 ) . This approach should have wide applicability for recombinant protein and domain target presentation for screening protocols . For the commercially sourced protein antigens ( tenascin C , VEGFR2 , tubulin and HER4 ) it became apparent that the success of the screen is dependent upon the commercial source . For example , tenascin C was sourced from several commercial suppliers but only one allowed the selection of suitable Affimers ( data not shown ) . Fortunately the potential availability of thousands of high quality proteins from structural genomics consortia together with the ability to express in vivo biotinylated protein domains for capture , without purification , should reduce the risk of screen failure associated with target quality and availability . Two papers have described the generation of antibody and antibody fragments against SH2 domains ( Renewable Protein Binder Working Group et al . , 2011; Pershad et al . , 2010 ) . The initial paper used phage display to select binding reagents that showed exquisite specificity against recombinant SH2 domains in vitro ( Pershad et al . , 2010 ) . However , none of the reagents were used in assays to demonstrate binding to endogenous proteins from cells . Colwill et al assessed the ability to isolate binding reagents against the same family of targets using both phage display of antibodies fragment libraries and monoclonal antibodies ( Renewable Protein Binder Working Group et al . , 2011 ) . They also successfully isolated binding reagents , although only a low proportion bound endogenous protein in the assays tested . By contrast , with the same class of SH2 targets , a high proportion of the monoclonal Affimer reagents reported here were successful in pull down assays and could block protein function when expressed in cells . The differences between these outcomes may be a result of library quality rather than an inherent feature of the scaffold . However , we consider it more likely due to differences in presentation of variable loop structures between antibodies and the Affimer scaffold . The ability to express Affimers intracellularly in mammalian cells , as shown by the inhibition of the p85 SH2 domain ( Figure 2 ) , represents an exciting opportunity , with the lack of disulfide bonds in the scaffold suited to the reducing environment of the cell . This feature is similar to other artificial binding proteins that can also be expressed in the cytoplasm of mammalian cells ( Kummer et al . , 2012; Spencer-Smith et al . , 2017; Wojcik et al . , 2010 ) . This raises the intriguing possibility of generating the necessary reagents , based on Affimers and other artificial binding proteins , to target specific protein domains of the human ‘interactome’ . These would provide extremely powerful tools for understanding the function of proteins and for identifying novel drug targets in disease . The lack of disulphide bonds in the Affimer and many other artificial proteins , such as DARPins , Monobodies and Affibodies , also allows the directed introduction of cysteine residue ( s ) for site specific chemical modification , including addition of a single biotin or fluorophore . A major advantage of phage display screening is the ability to isolate highly specific reagents by performing counter-screens against very similar target molecules . Interestingly for the SH2 targets , no counter-screening was performed and yet by ELISA analysis , specific Affimers were recovered for the Grb2 , 7 and 10 and p85 SH2 domains . Further studies will determine whether this level of specificity is observed at the cellular level . These results , nonetheless , provide promise for the isolation of highly specific cellular binding reagents . This high level of specificity was also demonstrated with Affimers that bind the small organic compound TNT commonly used as a model organic compound . The Affimers revealed remarkable specificity considering the small size of the molecule ( mol . mass <300 Da ) and the limited number of panning rounds used . It would be interesting to determine molecular structures of Affimer bound to TNT and 2 , 4-DNT to understand the recognition mechanism and to explain the discrimination between different DNT molecules . The ability to rapidly select Affimers that specifically detect small molecule targets represents a useful additional approach to the generation of reagents for diagnostic and monitoring applications of chemical agents , for example , in health , security or environmental settings . There are many examples of antibodies that bind to small molecules , although the in vivo nature of raising such reagents can present challenges for some compounds , such as toxins and pharmaceuticals , together with the time frame for inoculation and isolation and the need to use animals . Phage display has also been used to isolate antibody fragments ( Dörsam et al . , 1997; Vaughan et al . , 1996 ) and lipocalins that recognise small molecules ( Beste et al . , 1999; Schlehuber et al . , 2000 ) . The selection of reagents against small molecules also raises the prospect of recognising a range of post-translational modifications , potentially within the context of a specific protein . For example , DARPins that discriminate between phosphorylated and non-phosphorylated proteins have been described ( Kummer et al . , 2012 ) . These recognise conformational changes due to the phosphorylation event rather than the phosphorylated amino acid . Only time will tell whether alternative binding reagents are capable of directly reporting on post-translational modification of proteins in a similar manner to antibodies . The binding affinities of Affimers selected in this work were typically in the low nanomolar range , although some of the VEGFR2 binders had weaker affinities . Even so , these weaker binders still worked effectively and displayed specificity in affinity-histochemical assays and inhibited receptor function in biological assays . Since our monoclonal reagents have been identified by randomly selecting clones from three panning rounds it is anticipated that inhibitors with greater affinity can be developed , either through more detailed analysis of the pool of phage using next generation sequencing , or by affinity maturation . Whilst our screening strategy identifies monoclonal reagents , these can also be combined to generate polyclonal reagents that may improve sensitivity for certain in vitro applications . In conclusion , we have demonstrated the ability to rapidly isolate Affimer reagents , that are effective tools in a range of molecular and cell biology applications ( Figure 10 ) . This highlights the potential for creating a pipeline to isolate consistent renewable binding reagents against a wide variety of target molecules . Affimers are small , thermostable and simple to engineer and provide a system that compliments rather than replaces antibodies and other alternative protein scaffolds . A major aim of our laboratory is to further explore the capabilities of Affimer reagents and to test their potential protein modulating properties for use in dissecting specific cell signalling pathways as well as in studying protein-protein interactions on a proteomic scale . Affimer technology is commercially available through Avacta Life Sciences or for academic collaborations through the University of Leeds . Alternatively , the library can be synthesised as described by Tiede et al . ( 2014 ) and screened in individual laboratories , making this technology immediately accessible to the scientific community . 10 . 7554/eLife . 24903 . 015Figure 10 . Overview scheme of a range of applications that have been tested with Affimers . DOI: http://dx . doi . org/10 . 7554/eLife . 24903 . 015 Human SH2 domain coding sequences were in kanamycin-resistant pET28 SacB AP vectors ( Open Biosystems ) , with an N-terminal histidine tag . A biotin acceptor peptide ( BAP ) sequence was cloned into the vector to give an N-terminal BAP-Histag-SH2 domain sequence and the modified vector DNA introduced into Rosetta 2 ( DE3 ) cells . Single colonies were grown in 10 ml Terrific Broth ( TB ) supplemented with 100 µg/ml kanamycin and 34 µg/ml chloramphenicol , overnight at 37°C and 2 ml was used to inoculate 400 ml TB/100 µg/ml kanamycin and cultures grown until OD600 ~2 . After cooling to 18°C , for 1 hr IPTG was added to 0 . 5 mM . Cells were collected by centrifugation ( 800 g; 20 min , 4°C ) and resuspended in 10 ml Lysis buffer 1 ( 50 mM NaH2PO4; 300 mM NaCl; 30 mM imidazole; 10% glycerol; Benzonase Nuclease ( Novagen ) ; 1% Halt Protease Inhibitor Cocktail , EDTA-free; 1% Triton-X100; 1% lysozyme ) and left rocking overnight at 18°C before target proteins were purified using Amintra Ni-NTA resin ( Expedeon ) . Proteins were eluted using elution buffer ( 50 mM NaH2PO4; 500 mM NaCl; 300 mM imidazole; 10% glycerol ) . Expression and in vivo biotinylation of targets was confirmed by western blotting . The UL49 gene of Herpesvirus of Turkeys ( HVT ) was amplified by PCR using Q5 DNA polymerase ( NEB , UK ) and cloned using the Gibson Assembly kit ( NEB ) into a modified pMT-V5/6His ( Invitrogen ) in which the V5/6His cassette was replaced with 6His-AviTag and a SmaI site that was used to generate an N-terminal fusion to the UL49 gene product . pMT HVT UL49 was co-transfected with pCoHygro ( Invitrogen ) into Drosophila S2 cells using calcium phosphate precipitation and stably transformed cells were selected with hygromycin according to the manufacturer’s instructions ( Invitrogen ) before expression testing by western blot analysis 24 hr after induction with 500 µM copper sulphate . For purification , protein was extracted from stably transformed cells 36 hr after induction with copper sulphate at 20°C using a modified lysis buffer ( 25 mM Tris ( pH8 ) , 1 . 5% Triton-X100 , 50 mM arginine ( pH8 ) , 10 mM imidazole , 7 . 5% glycerol , 300 mM KCl ) . Protein was eluted from Ni-NTA ( QIAGEN ) in elution buffer ( 25 mM Tris ( pH8 ) , 50 mM arginine ( pH8 ) , 200 mM imidazole , 7 . 5% glycerol , 300 mM KCl ) . Purified protein was subjected to in vitro biotinylation using purified BirA according to the manufacturer’s instructions ( Avidity LLC ) . A 2 , 4 , 6-trinitrobenzene protein conjugate was prepared by mixing ovalbumin ( fraction VI , Sigma ) or Rabbit Gamma Globulin ( RGG ) ( Sigma ) at a concentration of 1 mg ml−1 with 0 . 05% ( w/v ) 2 , 4 , 6-trinitrobenzene sulfonic acid ( TNBSA ) ( Thermo Scientific ) in 0 . 1 M sodium bicarbonate buffer ( pH 8 . 5 ) . The mixture was incubated at 37°C for 2 hr and the resultant complex was then buffer exchanged into PBS and concentrated using a Vivaspin six column ( MWCO: 10 kDa; Sartorius ) to eliminate unconjugated TNBSA and excess buffer . The concentrated product was quantified using a Pierce Micro Bicinchoninic Acid ( BCA ) Assay ( Thermo Scientific ) , in accordance to manufactures guidelines using bovine serum albumin as the comparative standard protein . Target biotinylation and selection of Affimers by phage display was performed as described previously ( Tiede et al . , 2014 ) with some modifications in the second and third panning round . Biotinylated targets were bound to streptavidin-coated wells ( Pierce ) for 1 hr , then 1012 cfu pre-panned phage ( phage preincubated in streptavidin-coasted wells ) were added for 2 . 5 hr with shaking . Panning wells were washed 10 times and phage eluted with 50 mM glycine–HCl ( pH 2 . 2 ) for 10 min , neutralised with 1 M Tris–HCL ( pH 9 . 1 ) , further eluted with triethylamine 100 mM for 6 min , and neutralised with 1 M Tris–HCl ( pH 7 ) . Eluted phage were used to infect ER2738 cells for 1 hr at 37°C and 90 rpm then plated onto LB agar plates with 100 µg/ml carbenicillin and grown overnight . Colonies were scraped into 5 ml of 2TY medium , inoculated in 25 ml of 2TY medium with carbenicillin ( 100 µg/ml ) and infected with ca . 1 × 109 M13K07 helper phage . After 1 hr at 90 rpm , kanamycin was added to 25 μg/ml for overnight at 25°C and 170 rpm . Phage were precipitated with 4% polyethylene glycol 8000 , 0 . 3 M NaCl and resuspended in 1 ml of 10 mM Tris , pH 8 . 0 , 1 mM EDTA ( TE buffer ) . A 2 µl aliquot of phage suspension was used for the second round of selection using streptavidin magnetic beads ( Invitrogen ) . Target labelled beads were washed and incubated with pre-panned phage for 1 hr then washed five times using a KingFisher robotic platform ( ThermoFisher ) , incubated overnight at RT in 20% glycerol in PBS-T with at least one additional wash step and eluted and amplified as above . The final pan used neutravidin high binding capacity plates ( Pierce ) , as previously described for panning round one with the addition of a final overnight incubation at RT in 20% glycerol in PBS-T , and phage eluted using 100 µl of 100 mM dithiothreitol . Phage eluates were recovered from wells containing target protein and control wells to determine the level of amplification in target wells . For counter selections appropriate cell lysates or homologous proteins were added to the phage at a concentration of at least 10 µg/ml for 1 hr at room temperature before transferring the phage to the panning beads or wells . Counter screens were performed for UL49 target against purified Marek’s Disease Virus type 1 ( MDV-1 , strain RB1B ) and Duck enteritis virus ( DEV strain 684 ) , as well as cell lysate derived from Chicken Embryonic Fibroblasts ( CEF ) . The TNBS-Ovalbumin was counter selected against ovalbumin . In the second and third pan , after washing the wells/beads post incubation with the phage library , the wells/beads were incubated over night at RT in 20% glycerol in PBS-T with at least one additional wash step prior to elution . Phage were eluted in 100 µl 50 mM glycine–HCl ( pH 2 . 2 ) for 10 min , neutralised with 15 µl 1 M Tris–HCl ( pH 9 . 1 ) , further eluted with 100 µl triethylamine 100 mM for 6 min , and neutralized with 50 µl 1 M Tris–HCl ( pH 7 ) . Phage ELISA screening was performed , as previously described , on randomly selected clones from the final pan round as a method for selecting positive clones for further evaluation ( Tiede et al . , 2014 ) . Selected Affimer coding regions were amplified by PCR amplification using one of two reverse primers that would generate proteins with or without a C-terminal cysteine . Following NheI/NotI digestion the coding regions were ligated into a pET11a-derived vector and subsequently expressed in BL21 ( DE3 ) cells as previously described ( Tiede et al . , 2014 ) . Briefly , a single colony was used to inoculate a 5 ml overnight culture in 2TY/100 µg/mL carbenicillin . Then 50 ml LB-carb media was inoculated with 1 ml of overnight culture and grown for about 2 hr at 37°C and 230 rpm to an OD600 between 0 . 6–0 . 8 , before addition of IPTG to 0 . 1 mM and further grown for 6–8 hr or overnight at 25°C at 150 rpm . Cells were harvested , lysed in 1 ml Lysis buffer . The lysate was then incubated with 300 µl of washed NiNTA slurry for 1 hr , washed ( 50 mM NaH2PO4 , 500 mM NaCl , 20 mM Imidazole , pH 7 . 4 ) and eluted in 50 mM ( NaH2PO4 , 500 mM NaCl , 300 mM Imidazole , 20% glycerol , pH 7 . 4 ) . Affimers with C-terminal cysteine were biotinylated directly after elution . For each Affimer 150 µl tris ( 2-carboxyethyl ) phosphine ( TCEP ) immobilised resin ( ThermoFisher Scientific ) was washed and incubated with 150 µl of 40 µM Affimer solution on a rocker for 1 hr . The solution was centrifuged for 1 min at 1500 g and 120 µl of the supernatant was transferred into a fresh tube containing 6 µl of 2 mM biotin-maleimide ( Sigma ) and incubated for 2 hr at room temperature . Excess biotin linker was removed by using a Zeba spin desalting column ( Thermo Scientific ) according to the manufacturer’s protocol or by dialysis . Western blotting was performed on expressed BAP-tagged SH2 domains to check in vivo biotinylation of the targets . Protein samples were re-suspended in 4 X sample buffer ( 8% ( w/v ) SDS , 0 . 2 M Tris-HCl ( pH 7 ) 20% glycerol , 1% bromophenol blue ) and heated to 95°C for 10 min . Samples were loaded onto a 15% SDS-polyacrylamide resolving gel with a 5% stacking gel . Electrophoresed samples were then transferred onto a PVDF membrane with 0 . 2 µm pore size using a Trans-Blot Turbo Transfer System ( Bio-Rad , Hercules , USA ) . Membranes were blocked in 3% BSA in Tris buffered saline ( TBS ) containing 0 . 1% Tween-20 ( TBS-T ) overnight at 4°C followed by incubation with High Sensitivity Streptavidin-HRP ( ThermoFisher Scientific ) and visualised using Luminata Forte Western HRP Substrate ( Merck Millipore ) . Cell lysates for pull down of Grb2 were isolated from U-2 OS cells ( purchased from ATCC , STR profiled and mycoplasma negative - RRID:CVCL_0042 ) . Cells were washed with ice-cold 1 X PBS and lysed in lysis buffer ( 50 mM Tris , 150 mM NaCl , 1% ( v/v ) Nonidet P-40 , 1 ml per 75 cm2 flask ) , on ice . The pull down experiment was performed using the KingFisher automated platform . In brief , Affimers were expressed in 50 ml BL21 Star ( DE3 ) IPTG induced overnight culture , the BL21 Star ( DE3 ) cells were pelleted , lysed and clarified by centrifugation . Cell lysate was incubated with cobalt-based magnetic beads ( ThermoFisher ) on Kingfisher platform for 10 min prior to a single wash , and incubated with U-2 OS cell lysate ( approximately 500 µg ) for 90 min . The beads were washed a further three times on the KingFisher prior to being added to 50 µl of elution buffer . A 15 µl aliquot of eluted proteins was mixed with loading buffer prior to being heated to 95°C for 5 min and loaded on to an SDS-PAGE gel . Western blot analysis was performed using a rabbit monoclonal anti-Grb2 antibody , ( Abcam; ab32037 ) as primary antibody and anti-rabbit-HRP antibody , goat polyclonal ( Signal Technology; 7074 ) as secondary ( n = 2 , biological replicate; the number of times the experiment was independently repeated ) . For p85 pull down experiments NIH-3T3 cells transfected with p85-nSH2 Affimers were lysed in CellLytic M Cell Lysis Reagent ( Sigma ) with protease inhibitor cocktail ( Sigma ) and lysates cleared by centrifugation . For AKT activation western blot analysis , cells were serum-starved for 30 min prior to protein harvesting . Immunoprecipitation of p85α involved incubating 400 μg of total protein with 3 μl of anti-p85α antibody ( Abcam ) at 4°C with rotation overnight followed by protein G Sepharose beads ( Sigma ) for 4 hr . The beads were washed four times with PBS and resuspended in 2 X SDS sample buffer with β-mercaptoethanol ( Sigma ) , heated to 95°C for 3 min and proteins were resolved in 7% SDS-polyacrylamide gels . For western blotting , 20 μg of total protein was denatured using 5 X Laemmli sample buffer with β-mercaptoethanol and resolved in 12% SDS-polyacrylamide gels . Proteins were transferred to polyvinylidene difluoride membranes ( Biorad ) , blocked in 5% bovine serum albumin in PBS 0 . 1% Tween , and incubated with anti-pAKT ( Ser473 ) , anti-panAKT , anti-p110α ( Cell Signaling Technology ) , anti-flag M2 ( Sigma ) , anti-tubulin alpha ( AbD Serotec ) , and anti-p85α antibodies . Bound antibody was detected using anti-mouse/rabbit horseradish peroxidase-conjugated antibody and chemiluminescence ( Luminata Forte Western HRP substrate , Millipore ) . For detection of VEGFR2 signalling HUVECs ( primary human umbilical vascular endothelial cells ) were lysed in 2% ( w/v ) SDS containing 1 mM PMSF , in PBS ) . For immunoblot analysis , lysates were incubated at 95°C for 5 min and sonicated for 3 s . 25 μg of cell lysate was re-suspended in an equal volume of 2 X SDS sample buffer ( 1 M Tris-HCl pH 6 . 8 , 20% ( v/v ) glycerol , 4% ( w/v ) SDS and 0 . 1% ( w/v ) bromophenol blue , 4% ( v/v ) mercaptoethanol ) and incubated at 95°C for 5 min . Lysates were loaded onto a 10% ( v/v ) SDS-polyacrylamide resolving gel with a 5% ( w/v ) SDS-polyacrylamide stacking gel and separated at 130 V for 90 min in SDS-running buffer ( 192 mM glycine , 25 mM Tris , 0 . 1% ( w/v ) SDS ) . Proteins subjected to SDS-PAGE were transferred onto nitrocellulose membrane ( 0 . 2 μm pore size ) ( Schleicher and Schuell ) in transfer buffer ( 106 mM glycine , 25 mM Tris , 0 . 1% ( w/v ) SDS , 20% ( v/v ) methanol ) at 300 mA for 3 hr at 4°C . Membranes were incubated in 5% ( w/v ) skimmed milk ( in TBS-T ) for 30–60 min on a rocker . Membranes were rinsed in TBS-T , incubated with primary antibodies ( goat anti-VEGFR2 ( R&D Systems ) , rabbit anti-phospho-VEGFR2 ( Y1175 ) , rabbit antibodies to native and phosphorylated PLCγ1 ( Y783 ) , c-Akt ( S473 ) , p38 MAPK ( T180/Y182 ) and eNOS ( S117 ) , rabbit anti-ERK1/2 , mouse anti-phospho-ERK1/2 ( T202 , Y204 ) ( Cell Signalling Technologies ) , mouse anti-α-tubulin ( Santa Cruz Biotechnology ) ) overnight at 4°C and washed 3 times for 10 min in TBS-T prior to incubation with HRP-conjugated secondary antibody ( PerBio Sciences , Cramlington , UK ) for 1 hr at room temperature , followed by a second round of TBS-T washes and detection using a chemiluminescent solution , EZ-ECL ( Geneflow , Nottingham , UK ) . U2-OS cells were grown in DMEM medium supplemented with 10% ( v/v ) fetal bovine serum ( ThermoFisher Scientific ) , 2 mM L-glutamine , 1% penicillin-streptomycin , 10 mM sodium pyruvate in 5% CO2 in air at 37°C . Cells were seeded on to coverslips in 24-well plates to reach a density of ~60% at the time of transfection . Cells were transfected with rat TRPV1-encoding DNA 48 hr before use in ICC experiments using FuGENE Transfection Reagent ( Promega ) according to manufacturers’ instructions . At 48 hr post-transfection , Affimers were incubated on live cells at a final concentration of 5 μg/ml in assay buffer ( 130 mM NaCl , 10 mM glucose , 5 mM KCl , 2 mM CaCl2 , 1 . 2 mM MgCl2 , 10 mM HEPES , pH 7 . 4 ) for 20 min . Cells were then washed with the same buffer three times for 5 min per wash before fixation with 4% PFA for 10 min . Fixed cells were washed three times in PBS and then permeabilised with 0 . 1% Triton X100 in PBS . Cells were blocked for 30 min in 1% BSA in PBS at room temperature . Mouse monoclonal anti-6X His antibody ( Abcam: Ab18184 ) was incubated on cells for one hour at room temperature . Cells were then washed three times with PBS and incubated for one hour in the dark with anti-mouse 488 antibody ( Life Technologies: A11001 ) for 1 hr . A further two washes in PBS-T , two washes in PBS and one wash in ddH20 was followed by mounting on to glass slides with ProLong Diamond Antifade Mountant with DAPI ( ThermoFisher Scientific ) . The next day , samples were imaged using an EVOS FL imaging system ( ThermoFisher Scientific ) . For co-localisation staining with anti-TRPV1 antibody , the same protocol was followed with an additional step in which a further 1 hr incubation was conducted with anti-TRPV1 antibody ( Abcam: ab10295 ) post-fixation . Goat anti-guinea pig 647 ( Abcam: ab150187 ) was then applied for the detection of anti-TRPV1 antibody for one hour at room temperature in the dark . U2-OS cells were cultured using the above conditions . Cells were seeded in to T75cm flasks so that confluence would reach ~60% by the time of transfection . 24 hr post-transfection with TRPV1 DNA , cells were trypsinised in to black-walled 96 well-plates ( Greiner Bio-One ) at a confluence of ~100 , 000 cells per well and incubated for a further 24 hr under previously described cell culture conditions . On the day of the modulation assay , cells were washed with assay buffer and loaded with 50 μL Fluo-4 AM ( 1 μM ) ( ThermoFisher Scientific: F14201 ) for 1 hr at 37°C . Cells were washed again with 200 μL assay buffer prior to the addition of Affimer at a concentration of 1 μM in 50 μL assay buffer . Following 30 min of incubation at room temperature , the increase in intracellular Ca2+ was measured using a Flexstation 3 ( Molecular Devices; Sunnyvale , CA , USA ) . Fluorescence was detected for 60 sec at 485 nm excitation and 525 nm emission , but the peak Ca2+ response ( approximately 5 sec after addition of the orthosteric TRPV1 agonist , capsaicin ) was used for the subsequent determination of the agonist response . Initially , the effect of Affimers on Ca2+ response was tested at a capsaicin EC20 concentration whilst subsequent experiments tested a range of capsaicin concentrations in a so-called curve shift assay . Relative peak fluorescence units were normalised to the response observed in the absence of Affimer . Data analysis - ordinary one-way ANOVA with multiple comparisons was conducted for modulation assays against an EC20 concentration of capsaicin . P values less than 0 . 05 ( * ) 0 . 01 ( ) and 0 . 001 ( * ) and 0 . 0001 ( **** ) are indicated . Immulon2 HB 96-well micro-titre plates ( Nunc ) were coated with TNBS-ovalbumin conjugate at a concentration of 10 µg ml−1 in PBS ( 100 µl per well ) and incubated overnight at 5°C . Each well was washed three times with PBS containing 0 . 05 % v/v Tween20 ( PBST ) prior to blocking using 2% ( w/v ) skimmed milk powder ( Marvel ) in PBST ( blocking buffer ) by incubating for 60 min at room temperature . Each well was then washed and free TNBS-Ovalbumin , TNT or DNT analogue was diluted in blocking buffer and added to 6 replicate wells for each Affimer at a concentration of 100 µg/ml . Each hapten was then serially diluted down the plate to a final concentration of 0 . 78 µg/ml in blocking buffer . Subsequently , 50 µl of biotinylated Affimers TNT3 , 4 and 9 ( 0 . 5 µg/ml ) and TNT15 ( 1 . 0 µg/ml ) in blocking buffer was added in replicate to each of the wells containing free hapten to allow for six technical replicate dilution curves of each free hapten molecule with each Affimer to be generated ( technical replicate; the number of times the experiment was repeated within one experiment ) . For the negative control , the Affimer was substituted for blocking buffer . Each plate was incubated at room temperature for 1 hr , washed three times with PBST and Affimer that remained bound to the TNBS-Ovalbumin conjugate was detected using a high sensitivity streptavidin-HRP conjugate ( ThermoFisher Scientific ) , diluted 1:2000 in blocking buffer . The presence of HRP was detected using hydrogen peroxide ( Sigma ) and ABTS substrate ( Sigma ) in substrate buffer ( 0 . 1 M Citric acid , 0 . 2 M Na2HPO4 at pH 4 . 37 ) with the response quantified based on readings at 414 nm in an automated plate reader ( Anthos 2001 , Anthos Labtec Instruments ) . SW620 ( mycoplasma tested - RRID:CVCL_0547 ) xenograft mice were sacrificed; tumours were harvested and embedded in paraffin wax . Xenograft tissue was then processed for tenascin C immunohistochemistry as follows . Briefly , 4 μm paraffin sections were cut and collected on poly-lysine coated slides . Sections were dewaxed in xylene solutions and rehydrated in graded alcohol followed by distilled water . Antigen retrieval was performed by pressure-cooking in 0 . 01 M citric acid buffer , pH 6 . 0 . Following antigen retrieval , tissue sections were washed in distilled water and endogenous peroxidase was blocked with Bloxall blocking reagent ( 10 min , SP-6000; Vector Laboratories Ltd , Peterborough , UK ) . After washing with TBS-T , endogenous Avidin/Biotin was blocked using Avidin/Biotin blocking kit ( SP-2001 Vector Laboratories Ltd , Peterborough , UK ) . Tissue sections were washed in TBS-T and non-specific protein binding sites were blocked using 1 X casein ( 20 min , SP-5020; Vector Laboratories Ltd , Peterborough , UK ) prepared in antibody diluent ( Sigma , Poole , UK ) . Sections were then incubated overnight ( 4°C ) in mouse monoclonal anti-TNC antibody ( 1:25 , 4F10TT , IBL , USA ) and bound antibody was detected using the mouse on mouse polymer IHC kit ( ab127055 , Abcam , Cambridge , UK ) according to the manufacturer’s instructions . Tissue sections were counterstained with haematoxylin , dehydrated , cleared and mounted in DPX . Images were captured an Axioplan Zeiss microscope and AxioVision 4 . 8 software ( Carl Zeiss Inc . Germany ) . Tissue staining using the TNC Affimer was performed in a similar mannerhowever , after blocking non-specific protein binding sites with casein , sections were incubated in biotinylated TNC-Affimer overnight at 4°C . Sections were then washed in TBS-T and bound Affimer was visualized using Streptavidin/HRP ( 1:300 , 30 min , SA-5004; , Vector Laboratories Ltd , Peterborough , UK ) with 3 , 3′-diaminobenzidine ( DAB ) as substrate ( SK41-05; ImmPACT DAB , Vector Laboratories Ltd , Peterborough , UK ) . Sections were counterstained and imaged as described previously . Wax embedded tissue sections of human pancreatic tissue were collected and processed for immunostaining in a manner similar to that described above . After dewaxing and rehydrating the tissue sections , antigen retrieval for the VEGFR-2 epitope was carried out using Tris EDTA buffer ( pH 9 . 0 ) . Endogenous peroxidase , Avidin/Biotin and protein were blocked as described and tissue sections were incubated overnight at 4°C in rabbit monoclonal anti VEGFR2 antibody ( 1:25 , clone 55B11 , Cell Signaling Technology , Danvers , USA ) or VEGFR2 Affimers ( 1–11 µg/ml ) . Bound antibody was visualized using polyclonal goat anti-rabbit biotinylated antibody ( 1:200 , 30 min , clone E0432 , DAKO , UK ) and Streptavidin/HRP with DAB as substrate . Affimers were visualized using Streptavidin/HRP with DAB as substrate . Section were counterstained and imaged as described previously ( n = 3 ) . MCF7 cells ( purchased from ECACC , STR profiled and mycoplasma negative - RRID:CVCL_0031 ) were grown in RPMI medium supplemented with 10% ( v/v ) fetal bovine serum ( ThermoFisher Scientific ) , 2 mM L-glutamine , 1% penicillin-streptomycin , 10 mM sodium pyruvate in 5% CO2 in air at 37°C . Mouse fibroblasts , NIH-3T3 , were cultured in Dulbecco's Modified Eagle's Medium ( Sigma ) with 10% FCS and 2 mM L-glutamine in a humidified atmosphere at 37°C in 5% CO2 . Affimers specific to the N-terminal SH2 domain of p85 were transfected into NIH-3T3 using TransIT 293 ( Mirus , Madison , USA ) according to the manufacturers’ instructions . Human umbilical vein endothelial cells ( HUVECs ) were isolated and cultured in endothelial cell growth medium ( ECGM ) . Human umbilical cords used for isolation of primary endothelial cells were provided by written informed consent in accordance with ethical guidelines and under ethical approval ( reference CA03/020 ) of the Leeds NHS Hospitals Local Ethics Committee ( UK ) . HUVECs were seeded into 6-well plates and cultured ( for at least 24 hr ) in ECGM until ~80% confluent , washed twice in PBS and then starved in MCDB131 plus 0 . 2% ( w/v ) BSA for 2–3 hr . HUVECS were treated with 0 , 50 , 100 or 150 μg/ml Affimer for 30 min prior to stimulation with 25 ng/ml VEGF-A ( Genentech Inc . , San Francisco , USA ) for 0 , 5 or 15 min . Chicken Embryonic Fibroblasts ( CEF ) were derived from 10 day embryos and maintained in E199 medium ( Sigma ) supplemented with 10% Tryptose phosphate broth ( BD ) and 5% foetal calf serum ( Sigma ) , at 38 . 5°C . Virus infections were generated by lipofectamine transfection of BAC clones of either MDV-1 ( RB1B strain ) , HVT or DEV ( 2085 strain ) . Briefly , 2 µg of BAC DNA was diluted into 100 µl of Opti-MEM reagent , and mixed with 10 µl lipofectamine ( Invitrogen ) diluted in 100 µl Opti-MEM ( Invitrogen ) . After complex formation for 30 min a further 800 µl of Opti-MEM was added and the 1 ml sample transferred onto a well of a six well plate containing CEFs that had been rinsed twice with Opti-MEM . DNA complexes were left on CEFs for 6 hr in normal culture conditions after which 2 ml of growth medium was added per well and returned to the incubator . Once virus replication was established infected cells were passaged onto fresh CEFs as required to maintain the infection . For immunofluorescence and in-cell Western experiments primary CEFs were infected with either MDV-1 ( strain RB1B ) , HVT , or DEV ( strain 2085 ) . In all cases virus was derived from BAC clones by transfection into CEFs , these viruses constitutively express a GFP marker under the control of the thymidine kinase promoter present within the BAC element . For in-cell Western studies , infected cells were seeded into 96 well plates and allowed to adhere overnight . Cells were then fixed with 4% paraformaldehyde in PBS , washed with PBS , permeabilsed with 0 . 1% TritonX100 in PBS and blocked with 0 . 5% BSA in PBS for 30 min at room temperature . Affimers were then added at 1 . 5 µg/ml or a goat polyclonal anti-GFP antibody ( Sicgen ) at 1:2000 dilution in blocking buffer and incubated at room temperature for 1 hr . Samples were then extensively washed with PBSa before secondary antibody in blocking buffer was added . For in-cell Western , donkey anti goat 680 was used at 1:5000 dilution , with Streptavidin 800 conjugate at 1:5000 dilution ( Licor ) , and incubated for 1 hr at room temperature , then extensively washed with PBS . In-cell Westerns were imaged using the Licor Odyssey system using both the 700 nm and 800 nm channels following the manufacturer’s recommendations . Images were exported from the manufacturer’s proprietary software and processed using Adobe Illustrator . For immunofluorescence the same approach was followed after seeding of cells to coverslips . Coverslips were incubated with Affimers as the only primary detection reagent , with subsequent labelling with streptavidin-568 conjugate ( Invitrogen ) at 1:1000 dilution . After washing off excess Streptavidin-568 , cells were stained with DAPI , followed by three deionised water washes before mounting on a Vectashield mounting medium ( Vector Laboratories ) . Immunofluorescence images were captured using a Leica SP5 system and manufacturer’s software , from the 488 nm channel , 568 nm channel and the 405 nm channel using the 63 x objective and treated as for in-cell Western images . Amine coupling chips ( sensor chip CM5 , GE Healthcare ) were primed in 0 . 1 M sodium acetate pH 5 . 6 and functionalized with EDC/NHS35 µl at 5 µL/min . Target protein ( 1 mg/mL ) was immobilized to one flow cell ( 300–600 response units ) , at a flow rate of 5 µL/min and the flow cell was capped with 1 M ethanolamine-HCl ( 35 µL at 5 µL/min ) . The non-functionalised flow cell ( acting as a blank ) was treated with EDC/NHS ( 35 µL at 5 µL/min ) and 1 M ethanolamine-HCl ( 35 µL at 5 µL/min . The system was primed in PBS supplemented with 0 . 1% TritonX 100 . Five concentrations of Affimer ( 5–500 nM ) were tested . Each concentration was flowed over both the functionalised and the non-functionalised flow cells at 40 μL/min and the association and dissociation rate constants ka and kd , respectively were calculated using the Biacore software allowing determination of the equilibrium dissociation constant , KD as below . d[AB]/−dt=ka[A][B]∙d[AB]/dt=kd[AB]KD=kd/ka KD values were also determined using the Octet Red interferometer ( Pall Fortebio ) using streptavidin coated biosensors ( AMC , 18–5019 ) as previously described ( Kumaraswamy and Tobias , 2015 ) . All experiments were carried out in HBS-EP buffer ( 10 mM HEPES ( pH 7 . 4 ) , 150 mM NaCl , 3 mM EDTA , 0 . 005% ( v/v ) Tween 20 ) . KD values were determined by binding each biotinylated Affimer to a row of AMC biosensors at a constant concentration of 50 nM . Next a 2-fold dilution of unlabelled purified protein starting at 41 nM was bound to the Affimers . Raw offset values were plotted against concentration of purified HVT UL49 protein , determined by densitometry , and modelled to one site-specific binding equation using Graphpad Prism 6 . The C-terminal cysteine residues of Affimers TNC15C and GFP32C were labeled with Rhodamine Red C2 maleimide ( Thermo Fisher Scientific ) . Samples of Affimer ( TNC15C or GFP32C ) ( 80–200 µM ) in elution buffer ( 50 mM NaH2PO4 , 500 mM NaCl , 300 mM imidazole , 10% glycerol , pH 7 . 4 ) were dialysed ( 2 X with a dilution of 1000x ) into labelling buffer ( PBS containing 20% glycerol and 0 . 05% Tween-20; pH 7 . 4 ) . The samples were then treated with TCEP in H2O at 2 . 5 mM and Rhodamine Red C2 maleimide ( 20 mM in DMSO; 5 equiv . ) and rocked for 6 hr . Upon completion assessed by mass spectrometry , the reactions were quenched with β-mercaptoethanol ( 100 equiv . ) and the mixture was spin concentrated ( 3 kDa cut-off ) . The concentrated mixture was passed through a buffer exchange column ( PD-10 , GE Healthcare ) , eluting 0 . 5 mL fractions with labelling buffer . Fractions containing protein were identified by BioRad colorimetric assay and pooled taking care not to include fractions containing free Rhodamine Red dye that elute later . The labelled Affimers were then concentrated to 250–300 μM in a spin concentrator ( 3 kDa cut-off ) . The concentrations were estimated by SDS-PAGE analysis against known amounts of BSA as standard . The identities of labelled Affimers were confirmed by mass spectrometry . Samples were used immediately or flash frozen in liquid nitrogen and stored at −80°C until required . All procedures were carried out in accordance with the Animals ( Scientific Procedures ) Act 1986 under project licence approval ( PPL 70/7965 ) . Ethical review and monitoring was undertaken by the Animal Welfare and Ethics Review Committee ( AWERC ) at the University of Leeds . Twenty-four 6–10 week old BALB/c nude female mice ( originally obtained from Charles River , UK then maintained in-house ) were injected subcutaneously in the right flank with 1 × 107 SW620 cells ( obtained from ECACC and verified by single tandem repeat analysis ) . After 10–14 days of tumour growth , animals were randomised to receive either the tenascin C Affimer or a control GFP Affimer conjugated with Rhodamine Red C2 maleimide , via tail vein injection ( mean tumour volume tenascin C Affimer group 316 . 9 ± 192 . 0 mm3 vs 360 . 8 ± 216 . 6 mm3 in control GFP Affimer group; p=0 . 64 ) . Approximately 300 µM labelled Affimer in 100 μL PBS with 20% glycerol and 0 . 05% Tween-20 was injected intravenously into each animal . Fluorescent images of harvested tissues ( tumour , liver , kidney , spleen , heart , lung and brain ) were captured ex vivo using IVIS Spectrum ( excitation 570 nm , emission 620 nm; Perkin Elmer , USA ) Fluorescence intensity ( radiant efficiency in p/s/cm2/sr/μW/cm2 ) for each tissue was determined for a region of interest of defined unit area using Living Image software ( v4 . 3 . 1 , Perkin Elmer ) . Mean background fluorescence intensity was normalized to sham injected control tumours and organs .	Many of the molecules that are essential for life are too small to be visible inside cells . So , scientists use large complex proteins called antibodies that bind to these molecules to detect whether they are present and show where they are in a cell . As well as being useful tools in experiments , these antibodies can be used to help identify and treat diseases . The body produces antibodies in response to an infection . The antibodies used in experiments are purified from animal blood , but this method of producing antibodies has flaws . For example , it can be difficult to make identical batches of antibody that always behave in the same way . So scientists have developed “alternative binding proteins” that can be made in the laboratory . These proteins are much less complicated and can be developed more quickly than antibodies , and can easily be adapted for a variety of uses . An alternative binding protein called an Affimer behaves in a similar way to an antibody by binding tightly to its target molecule , but is much more stable to acidity and high temperature . Tiede et al . have now tested how well the Affimer works in a wide range of different experiments that normally use antibodies to analyse the amount of a particular molecule inside a cell . The results of the tests show that the Affimer behaves in the same way as antibodies , and sometimes works more effectively . Tiede et al . show that an Affimer can help to reveal how a particular molecule works within a cell , to create detailed pictures of molecules in cells and tissues , and to identify a tumour . It can also be used alongside a new technique called ‘super-resolution microscopy’ that allows researchers to watch the activity of individual molecules . Future challenges are to test the Affimer in even more applications and to encourage its wider use by researchers , alongside other alternative binding proteins , as as replacements for some antibodies . This could ultimately lead to the development of faster and more efficient diagnostic , imaging and therapeutic tests .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cell", "biology", "tools", "and", "resources" ]	2017	Affimer proteins are versatile and renewable affinity reagents
Motor contagions refer to implicit effects on one's actions induced by observed actions . Motor contagions are believed to be induced simply by action observation and cause an observer's action to become similar to the action observed . In contrast , here we report a new motor contagion that is induced only when the observation is accompanied by prediction errors - differences between actions one observes and those he/she predicts or expects . In two experiments , one on whole-body baseball pitching and another on simple arm reaching , we show that the observation of the same action induces distinct motor contagions , depending on whether prediction errors are present or not . In the absence of prediction errors , as in previous reports , participants' actions changed to become similar to the observed action , while in the presence of prediction errors , their actions changed to diverge away from it , suggesting distinct effects of action observation and action prediction on human actions . Our motor behaviors are shaped not just by physical interactions ( Shergill et al . , 2003; Ganesh et al . , 2014; Takagi et al . , 2017 ) but also by a variety of perceptual ( Heyes , 2011; Chartrand and Bargh , 1999; Ikegami and Ganesh , 2014; Cook et al . , 2012; Ganesh and Ikegami , 2016 ) interactions with other individuals . Motor contagions are the result of such a perceptual interaction . They refer to implicit changes in one’s actions caused by the observation of the actions of others ( Blakemore and Frith , 2005; Becchio et al . , 2007 ) . Studies over the past two decades have isolated various motor contagions in human behaviors , from the so called automatic imitation ( Brass et al . , 2001; Heyes , 2011 ) and emulation ( Edwards et al . , 2003; Becchio et al . , 2007; Gleissner et al . , 2000 ) , to outcome mimicry ( Gray and Beilock , 2011 ) and motor mimicry ( Chartrand and Bargh , 1999; Chartrand and Baaren , 2009 ) . These motor contagions are induced simply by action observation and have a signature characteristic - they cause certain features of one’s action ( such as kinematics [Brass et al . , 2001; Heyes , 2011; Kilner et al . , 2003] , goal [Edwards et al . , 2003; Becchio et al . , 2007; Gleissner et al . , 2000] , or outcome [Gray and Beilock , 2011] ) to become similar to that of the observed action . In contrast , here we report a new motor contagion that is induced not simply by action observation , but when the observation is accompanied by prediction errors - differences between actions one observes and those he/she predicts or expects . Furthermore , this contagion may not lead to similarities between observed actions and one’s own actions . Here we report results from two experiments to show that distinct motor contagions are induced by the observation of the same actions depending on whether prediction errors are present or not . We show that these contagions are present not only in high dimensional whole body movement tasks such as baseball pitching ( Experiment-1 ) ( Ikegami et al . , 2017 ) , but also simple day-to-day movement tasks such as arm reaching ( Experiment-2 ) . Thirty varsity baseball players participated in our Experiment-1 . The sample size was determined by a power analysis ( see Materials and methods ) . The participants were randomly assigned to one of three groups ( n = 10 in each ) : No prediction error ( nPE ) group , Prediction error ( PE ) group , and Control ( CON ) group . The participant’s baseball experience was balanced across the three groups ( F ( 2 , 27 ) =1 . 431 , p=0 . 257 , ηp2=0 . 096 ) . The participants in the nPE and PE groups performed five throwing sessions ( Figure 1A ) that were interspersed with four observation sessions ( Figure 1B , C ) . The participants in the CON group performed only the throwing sessions . Instead of the observation sessions , they took a break in between the throwing sessions for a time period equivalent to the length of the observation sessions . In the throwing session , the participants in all the groups threw a baseball aiming for the center of a ‘strike-zone’ sized square target placed over the ‘home plate’ ( Figure 1A ) . They made their throws while wearing ‘occlusion’ goggles that turned opaque when the participants released the ball from their hand ( see Figure 1A ) . The participants could thus see the target for aiming , but could not see where their ball hit the target . Each throwing session included ten throws . In the observation session , the participants in the nPE and PE groups watched a video of throws made by an unknown baseball pitcher . After each throw , a numbered grid ( see Figure 1B ) appeared on the target ( in the video ) once the ball hit the target , and the participants were asked to report the grid number corresponding to where they saw the ball hit the target . The purpose of this reporting task was to ensure that the participants maintained their attention on the target in the video . To cancel out any spatial bias with respect to the observed actions , half the participants in each group ( called upper-right observing participants ) were shown throws that predominantly hit the upper-right corner of the target ( most frequent at #3 , see yellow gradient Figure 2B and Materials and methods for observed throw distribution ) . The other five participants ( called lower-left observing participants ) were shown a video of throws predominantly hitting the lower-left corner of the target ( most frequent at #7 , see Materials and methods ) . Each observation session included 20 observed throws . Different instructions were provided to the nPE and PE groups to manipulate the prediction errors induced in them . The participants in the nPE group were told that ‘the pitcher in the video is aiming for different grid numbers on the target across trials . These numbers were provided by the experimenter and we display only those trials in which he was successful in hitting the number he aimed for’ . On the other hand , participants in the PE group were instructed that ‘the pitcher in the video is aiming for the center of the target’ . The two different instructions were designed to lead to greater prediction errors in the PE group compared with the nPE group . The instruction to the nPE group prevents the participants from having any prior expectation of the outcome of an observed throw , and was thus expected to attenuate any prediction error . In contrast , the instruction to the PE group makes the participants expect the observed throws to hit near the target center . Similar to previous studies ( Ikegami and Ganesh , 2014; Ondobaka et al . , 2015 ) , this instruction was thus expected to induce a difference between the throw outcome expected by the participants and the actual outcome observed by them . Specifically , we expected the upper-right and lower-left observation participants in the PE group to experience prediction errors , directed towards the upper-right and lower-left , respectively . The participants’ task performance in the throwing session was evaluated as a change in the throw hit location . The position of the hit locations was measured along the ‘parallel’ and ‘orthogonal’ diagonals ( see green arrows in Figure 2A ) . The diagonal joining the upper-right and lower-left corners , that were the predominant locations of the pitcher’s throws in the observed video , was named the ‘parallel’ diagonal . Data from the upper-right and lower-left observing participants were analyzed together by flipping the coordinate of the data from the lower-left observing participants ( see Materials and methods ) . First , in the observation session , the accuracy ( % correct ) of the report was comparable between the two groups ( nPE: 96 . 25 ± 1 . 48 ( mean ± s . d . ) % , PE: 95 . 63 ± 3 . 32 %; two sample t-test , t ( 18 ) =0 . 516 , p=0 . 612 ) . This ensures that the level of attention to the video was similar between the two groups . The performance in the throwing session , however , dramatically differed between the two groups . The throws by the nPE group progressively drifted towards where the pitcher in the video predominantly threw the ball ( blue data in Figure 2B ) . This pattern is similar to the motor contagion reported previously as outcome mimicry ( Gray and Beilock , 2011 ) . In contrast , the throws by the PE group progressively drifted away from where the pitcher in the video predominantly threw the ball ( red data in Figure 2B ) . The hit locations along the parallel diagonal ( Figure 3 ) showed a significant interaction between the sessions and groups ( F ( 8 , 108 ) =5 . 124 , p=2 × 10−5 , ηp2=0 . 275 ) . Across the sessions , the throws in the nPE group ( blue data in Figures 2B and 3 ) significantly drifted towards the direction of observed pitcher’s throws ( F ( 4 , 36 ) =2 . 910 , p=0 . 035 , ηp2=0 . 244; first vs fifth sessions by Tukey’s test: p=0 . 034 ) while the throws in the PE group ( red data in Figures 2B and 3 ) significantly drifted away from the observed pitcher’s throws ( F ( 4 , 36 ) =5 . 170 , p=2 × 10−3 , ηp2=0 . 365; first vs fifth sessions by Tukey’s test: p=5 × 10−3 ) . The throws in the CON group ( black data in Figures 2A and 3 ) did not show such a drift ( F ( 4 , 36 ) =0 . 297 , p=0 . 878 , ηp2=0 . 032 ) along the parallel diagonal . On the other hand , the hit locations along the orthogonal diagonal in all the nPE , PE , and CON groups showed no significant differences across the throwing sessions ( F ( 4 , 108 ) =1 . 762 , p=0 . 142 , ηp2=0 . 061 ) and between the groups ( F ( 2 , 27 ) =1 . 150 , p=0 . 332 , ηp2=0 . 079 ) . This result confirms that the drifts observed in the nPE and PE groups are not the result of possible cognitive fatigue induced by the extra observation task they performed ( compared with the CON group ) , or cognitive biases induced by the validity of the instructions given to them . In either case , we would have expected their throws to also drift in directions other than along the parallel diagonal . The focused drifts along the parallel diagonal suggest that the drifts were induced by the bias in the observed pitcher’s throws ( in the nPE group ) and the prediction errors ( in the PE group ) , both of which were present specifically along the parallel diagonal . In addition , the effects of observed actions on the participants’ actions emerged as an increase in spatial bias but not in spatial variability in their action outcome . For the orthogonal diagonal , the within-participant variability in their hit locations , measured by variance in each throwing session ( see Materials and methods ) , showed no significant differences across the sessions in all the three groups ( Friedman’ test , nPE: χ2 ( 4 ) =7 . 44 , p=0 . 114; PE: χ2 ( 4 ) =0 . 32 , p=0 . 989; CON: χ2 ( 4 ) =3 . 2 , p=0 . 525 ) . For the parallel diagonal , although the PE groups showed a significant change in variance ( nPE: χ2 ( 4 ) =3 . 28 , p=0 . 512; PE: χ2 ( 4 ) =10 . 48 , p=0 . 033; CON: χ2 ( 4 ) =3 . 2 , p=0 . 525 ) , we did not observe any clear trend in the median values of the within-participant variability across the sessions ( first session: 7 . 772; second: 5 . 360; third: 7 . 092; fourth: 7 . 008; fifth: 7 . 767 × 104 mm2 , respectively ) . And , a post hoc analysis found no significant difference between any pair of sessions ( Wilcoxon signed rank test , Zs <1 . 886 , ps >0 . 059 , which was considerably above the Bonferroni corrected significance level of p=0 . 005 ) . Together , these results clearly show that the observation of a same action can lead to distinct motor contagions depending on whether the observation takes place in the presence or absence of prediction errors . Next , to check whether this prediction error induced contagion is specific to sports experts , and to verify whether it can also be observed in simple everyday movement tasks , we conducted a second follow-up experiment in which we used a similar experimental design to before but tested average adult participants in an arm reaching task ( see Materials and methods ) . Thirty right-handed averaged male participants were randomly assigned to one of three groups ( n = 10 in each ) : nPE , PE , and CON groups . The participants in the nPE and PE groups performed five reaching sessions and four observation sessions ( Figure 4A ) . The participants in the CON group performed only the reaching sessions . In the reaching sessions , the participants made right arm reaching movements toward a touch screen to touch the center line , among three vertical lines presented on the screen , with their index fingers ( Figure 4A ) . They again wore occlusion goggles which , similar to the throwing experiment , became opaque as soon as their arm movements started , preventing them from observing where their finger touched the screen . Each reaching session included ten reaches . In the observation session , the participants in the nPE and PE groups watched a video of an unknown individual ( henceforth , the actor ) reaching towards the same touch screen ( Figure 4A ) . After each reach , the participants were asked to report if the actor had touched the ‘right’ ( area between the center and right lines ) , ‘left’ ( area between the center and left lines ) , or ‘outside’ ( of the right and left lines ) . Half the participants in each group watched a video showing the actor predominantly touching the right area ( called right observing participants ) , and the other half watched a video showing the actor predominantly touching the left area ( called left observing participants ) , respectively ( see Materials and methods ) . Each observation session included 20 observed reaches . To manipulate the prediction errors , we again provided different instructions to the nPE and PE groups . For the nPE group , the right and left observing participants were told that ‘the actor in the video is aiming for the right area’ and ‘the actor in the video is aiming for the left area’ , respectively . This instruction was expected to minimize the difference between the actor’s touch location expected by the participants , and the actual location observed by them , thus leading to suppression in prediction errors . On the other hand , participants in the PE group were instructed that ‘the actor in the video is aiming for the center line’ . This instruction was expected to induce substantial prediction errors as in Experiment-1 . To evaluate the participants’ reaching performance in the reaching session , the touch locations were measured along the x-axis on the screen . Similar to the main experiment , data from the right and left observing participants were analyzed together by flipping the coordinate of the data from the left observing participants ( see Materials and methods ) . First , in the observation session , we confirmed that the accuracy ( % correct ) of the report was comparable between the two groups ( nPE: 93 . 20 ± 3 . 94 ( mean ± s . d . ) % , PE: 93 . 00 ± 5 . 10 %; two sample t-test , t ( 18 ) =0 . 098 , p=0 . 923 ) . Next , in the reaching session , we observed similar results to Experiment-1; the touch locations by the participants ( Figure 4B ) showed significant interaction between the sessions and groups ( F ( 8 , 108 ) =2 . 568 , p=0 . 013 , ηp2=0 . 160 ) . Across the sessions , the touch locations by the nPE group ( blue data in Figure 4B ) drifted towards the direction of observed actor’s touch locations ( F ( 4 , 36 ) =2 . 307 , p=0 . 077 , ηp2=0 . 204 ) , although it marginally missed statistical significance . More importantly for this study , the touch locations by the PE group ( red data in Figure 4B ) significantly drifted away from the direction of the observed actor’s touch locations ( F ( 4 , 36 ) =2 . 683 , p=0 . 047 , ηp2=0 . 230 ) . The touch locations in the CON group ( black data in Figure 4B ) did not show any substantial change ( F ( 4 , 36 ) =0 . 639 , p=0 . 638 , ηp2=0 . 066 ) . In addition , the within-participant variability in their touch locations showed no significant differences across the reaching sessions ( nPE: χ2 ( 4 ) =5 . 04 , p=0 . 283; PE χ2 ( 4 ) =4 . 80 , p=0 . 308; CON: χ2 ( 4 ) =9 . 44 , p=0 . 051 ) . Thus , Experiment-2 successfully reproduced the Experiment-1 results , which suggests that the two distinct motor contagions can affect even basic everyday movements like arm reaching . Crucially , the clear motor performance changes in the PE groups observed in the two experiments validate the presence of a new motor contagion which is induced by prediction errors during action observation . Previous motor contagion studies have extensively examined how observation of the actions of others affects action production ( Heyes , 2011; Chartrand and Bargh , 1999; Becchio et al . , 2007; Brass et al . , 2001; Edwards et al . , 2003; Gray and Beilock , 2011; Kilner et al . , 2003 ) . On the other hand , while previous action prediction studies have examined how action production ( Mulligan et al . , 2016; Hamilton et al . , 2004 ) , or the ability to produce an action ( Knoblich and Flach , 2001; Urgesi et al . , 2012; Kanakogi and Itakura , 2011 ) affects prediction of observed actions , the converse , the effect of action prediction on one’s actions , has rarely been examined . One exception is our previous study ( Ikegami and Ganesh , 2014 ) , which reported that an improvement in the ability to predict observed actions affects the observer's ability to produce the same action . The observations from this previous study are likely to be a prediction error induced motor contagion , but as the direction of the observed actions and the corresponding prediction errors were not controlled in the study , it was difficult to conclude this cause . Overall , the effects of action prediction and action observation on action production have never been compared . This study compares the two effects for the first time by modulating the prediction errors perceived by participants , when they observe throws or reaches by another individual . We show that , while in the absence of prediction errors ( nPE group ) , the action observation makes the participants’ actions ( throws and reaches ) become similar to the observed actions ( as in previous reports [Heyes , 2011; Gray and Beilock , 2011; Blakemore and Frith , 2005] ) , in the presence of prediction errors ( PE group ) , the same action observation makes the participants' actions diverge away from the observed actions ( Figures 2B , 3 and 4B ) . Our results thus suggest the presence of a distinct , prediction error induced motor contagion in human behaviors . The previously reported motor contagions are believed to be caused by the presence of ‘long-term’ sensorimotor associations ( Heyes , 2011; Cook et al . , 2014 ) , in which an observed action automatically activates an individual’s sensorimotor representations of the same action , making his/her actions similar to the observed actions . While further studies are required to understand the mechanisms underlying the prediction error induced motor contagion , it is still interesting to note the parallel between this motor contagion and behavioral characteristic of motor learning . When learning a new motor task , individuals are believed to utilize prediction errors from self-generated actions to change their sensorimotor associations , and correct their actions ( Shadmehr et al . , 2010 ) . The motor contagion in the PE group seems to cause participants to make similar action corrections to compensate for their prediction errors , strangely even though the prediction errors are from actions made by another ( observed ) individual , rather than from self-generated actions . Therefore , while the previously reported motor contagions may be explained as a facilitation/interference in action production by the sensorimotor associations ( Heyes , 2011; Cook et al . , 2014 ) activated by observed actions , the new contagion we present here is probably a result of erroneous changes in the sensorimotor associations , caused by the observed prediction errors ( Ikegami and Ganesh , 2017 ) . The prediction error induced contagions presented here seem similar to the phenomena of observational motor learning , in which the observation of an action is reported to aid subsequent motor learning ( Adams , 1986; Schmidt and Lee , 2005; Mattar and Gribble , 2005; Buckingham et al . , 2014 ) . It has been suggested that observational motor learning is possible because of an observation induced adaptation of the inverse model ( or controller ) that determines motor commands for a given task ( Brown et al . , 2010 ) . On the other hand , in a recent study ( Ikegami and Ganesh , 2017 ) , we showed that the behavioral changes resulting from prediction error induced contagions can be explained only by an influence in an individual’s forward model , that uses the motor commands to estimate one’s movement outcomes . This difference in effects could result from two factors . First , unlike observational motor learning ( Maslovat et al . , 2010; Ong et al . , 2012 ) , our previous ( Ikegami and Ganesh , 2014; Ikegami and Ganesh , 2017 ) and present studies did not require the participants to learn a new motor task , and in fact included participants who were arguably over-trained in the task . This strongly suggests that the contagion effects in these studies are implicit ( not willed by the participants ) . And second , unlike the observational motor learning studies , this study systematically manipulates the participants’ belief regarding the observed actions to control the direction of prediction errors . The lack of an explicitly required learning and the effect of belief are probably key to whether and to what extent , observed actions affect the inverse , or forward modeling process , or both . However , further studies are required to address this question concretely . Finally , further studies are also needed to examine whether the prediction error induced motor contagions play a functional role in our motor actions . If the contagions are proved to contribute to motor skill acquisition or motor adaptation to a new environment , we may need to re-define the phenomenon as a new learning process rather than a new motor contagion . In conclusion , our results clearly show that action observation and action prediction can induce distinct effects on an observer’s actions . Understanding the interactions between these distinct effects is arguably critical for the complete understanding of human skill learning . For example , it can enable a better understanding of the link between the mechanisms of motor contagions and motor learning , and help develop better procedures to improve motor performances in sports and rehabilitation . Data to generate Figures 2A , B , 3 and 4B are available from the Dryad Digital Repository: http://datadryad . org/review ? doi=doi:10 . 5061/dryad . 3563k	Watching sports sometimes causes people to unintentionally move in the same way as the athlete they are observing . This type of unconscious mimicry is called a motor contagion . Observing everyday actions can also trigger motor contagion , and plays an important role in social interactions . So far , studies have focused on understanding how observing an action leads to motor contagion . They have not factored in the fact that in everyday life individuals consciously or unconsciously predict observed actions by others . Sometimes these predictions are wrong , leading to so called ‘prediction errors’ . It was not clear whether motor contagion occurs when the viewer has made an incorrect prediction , or if prediction errors change the behavior of the viewer . Now , Ikegami , Ganesh et al . show that prediction errors influence motor contagion . In one experiment , baseball players were asked to watch a video of an actor pitching a ball toward a target and predict where on the target the ball would hit . Some of the players were given misleading information intended to increase the likelihood they would incorrectly predict where the actor would throw . The players then pitched the ball towards a target themselves . When the players had just watched the actor’s throw , their throws became similar to it . When their predictions were wrong , their throws were very different from the actor’s throw . The players were not aware of the changes to their throw in either case . Ikegami , Ganesh et al . also conducted a similar experiment in which other volunteers were asked to observe an actor reaching for a target and then reach for the target themselves . The results were similar: when the volunteers’ predictions were wrong , they reached in different ways to the actor . This may be a new type of motor contagion . Learning more about this effect could help researchers to better understand the adjustments people make to their social behaviors and give new insights about the brain mechanisms that underlie normal human actions and social interactions . Sports trainers or physical therapists might also use this information to develop better strategies for maintaining athlete performances or helping people to recover movement after an injury or illness .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Prediction error induced motor contagions in human behaviors
CD8+ T cell anergy is a critical mechanism of peripheral tolerance , poorly investigated in response to immunotherapy . Here , using a pancreatic islet allograft model and CD3 antibody therapy , we showed , by single cell gene profiling , that intragraft CD8+ lymphocytes coexpressing granzyme B and perforin were selectively depleted through the Fas/FasL pathway . This step led to long-standing anergy of the remaining CD8+ T cells marked by the absence of cytotoxic/inflammatory gene expression also confirmed by transcriptome analysis . This sustained unresponsiveness required the presence of the alloantigens . Furthermore , tissue-resident CD8+ lymphocytes produced TGFβ and expressed the inhibitory receptors PD-1 and PD-L1 . Blockade of TGFβ downregulated PD-1 and PD-L1 expression and precipitated graft rejection . Neutralizing PD-1 , PD-L1 or TGFβRII signaling in T cells also abrogated CD3 antibody-induced tolerance . These studies unravel novel mechanisms underlying CD8+ T cell anergy and reveal a cell intrinsic regulatory link between the TGFβ and the PD-1/PD-L1 pathways . Upon antigen recognition , CD8+ T lymphocytes proliferate vigorously and differentiate into effector cells characterized by their ability to produce cytokines and chemokines , to migrate to inflamed tissues and to use various cytolytic pathways to kill their targets ( Williams and Bevan , 2007 ) . CD8+ T cells play major role in transplant rejection . Through direct presentation , they get activated by donor antigen-presenting cells ( APC ) , in particular dendritic cells , which migrate from the graft to secondary lymphoid organs ( Gras et al . , 2011; Ochando et al . , 2006 ) . This response is rapid , intense and induces acute graft rejection . Recipient APC also capture donor antigens from the transplant , present them to alloreactive CD8+ T cells through self MHC I molecules ( cross-priming/indirect presentation ) hereby contributing to acute and chronic graft rejection ( Celli et al . , 2011; Valujskikh et al . , 2002 ) . Cytotoxic CD8+ T lymphocytes migrate to the graft where they destroy target cells through granzyme B/perforin- and/or Fas/FasLigand-dependent cytolytic pathways . In most tolerance promoting protocols , long-term graft survival was associated with CD8+ T cell dysfunction which mainly resulted from clonal deletion and/or anergy ( Iwakoshi et al . , 2000; Monk et al . , 2003; Qian et al . , 1997 ) . Classically defined as the functional inactivation of T cells to cognate antigens , anergy was first described in vitro when T cells recognized antigens ( signal 1 ) in absence of appropriate costimulation ( signal 2 ) , usually provided by CD28 ( Schwartz , 2003 ) . T cells were not able to produce IL-2 , entered a hyporesponsive non proliferative state that prevented further responses upon antigen re-encounter . Over the last decade , better insight was gained into the signaling events leading to anergy , highlighting in particular the role of the transcription factors NF-AT ( nuclear factor of activated T cells ) and early growth response gene 2 and 3 ( Egr-2 , Egr-3 ) ( Macian et al . , 2002; Safford et al . , 2005 ) . However , characterization of the anergic phenotype and gene signature as well as the mechanisms that drive and sustain CD8 T cell anergy in vivo in the transplant setting are incompletely understood . Recently , we reported that CD3 antibodies ( CD3 Abs ) induced tolerance in fully MHC-mismatched experimental models of pancreatic islet and cardiac transplantation ( Goto et al . , 2013; You et al . , 2012 ) . The timing of treatment was crucial; long-term graft acceptance was obtained only when CD3 Abs were administered once T cell priming had occurred , a few days post-transplant . Permanent survival of a second graft from the original donor but not from third-party donor demonstrated that antigen-specific tolerance had been induced . Purified spleen CD8+ T cells from CD3 Ab-treated recipients were unable to respond when stimulated with donor antigens while response to third-party antigens was conserved ( You et al . , 2012 ) . In this model , CD3 Abs preferentially targeted and depleted activated effector T cells , notably within the graft . In the present manuscript , we addressed the molecular mechanisms underlying CD8+ T cell tolerance induced by CD3 Ab therapy . We conducted detailed analysis of alloreactive CD8+ T cells combining single-cell gene profiling , transcriptome analysis and ex vivo/in vivo functional studies . We found that CD3 Abs selectively deleted Gzmb+Prf1+ CD8+ cytotoxic effectors within the transplant . CD8+ T cells escaping this deletion became anergic . The presence of the alloantigen was mandatory for the effect just as was TGFβ signaling to promote and sustain PD-1/PD-L1-mediated CD8+ T cell tolerance . We previously showed that CD3 Ab-induced transplant tolerance was associated with a drastic reduction of CD8+ T cell infiltrates and of peripheral donor-specific CD8+ T cell responses ( You et al . , 2012 ) . Here we measured the anti-donor reactivity of graft infiltrating T cells using a 20 hr-IFNγ Elispot assay . Pancreatic islets from BALB/c mice were isolated and grafted under the kidney capsule of diabetic C57BL/6 recipients . Tolerogenic treatment with CD3 Ab F ( ab’ ) 2 fragments was applied for 5 days ( 50 µg/day ) at day 7 after transplantation . Intragraft T cells recovered after CD3 Ab treatment , on days 14 or 100 post-transplant , did not respond to BALB/c donor antigens as opposed to graft infiltrating T cells of untreated recipients analyzed few days before rejection ( day 14 ) ( Figure 1—figure supplement 1 ) . To better dissect the impact of CD3 Ab therapy on alloreactive CD8+ T lymphocytes , we took advantage of a validated multiplex single cell PCR method established by the group of B . Rocha . This technique provides information on cell heterogeneity through the analysis of the simultaneous expression of selected inflammatory and/or cytotoxic genes by individual CD8+ T cells ( Peixoto et al . , 2007 ) . We focused our analysis on Th1 and cytotoxic genes as it has been shown that the IFNγ , perforin and Fas/FasL pathways constituted predominant mechanisms of CD8+ T cell-mediated destruction of islet allografts ( Diamond and Gill , 2000; Sleater et al . , 2007 ) . Individual CD8+ T cells were sorted from the islet allografts ( 72 cells ) or spleen ( 48 cells ) recovered from 3 individual recipients on day +14 , that is right after the last injection of CD3 Abs , or on day +100 post-transplant , once tolerance was established . On day 14 post-transplant , in untreated recipients , graft infiltrating CD8+ T cells expressed the cytolytic molecules Gzmb , Prf1 and Fasl as well as Tbx21 and Klrg1 ( Figure 1A ) . Thirty three percent of these cells co-expressed 3 or more of the 7 genes tested ( Figure 1B ) . Interestingly , Gzmb was co-expressed with either Prf1 or Fasl which rarely overlapped , suggesting the presence of two distinct subsets of graft infiltrating CD8+ lymphocytes ( Figure 1C ) . Tbx21 , Eomes and Klrg1 were preferentially associated with Prf1 rather than Fasl ( Figure 1C ) . 10 . 7554/eLife . 08133 . 003Figure 1 . Coexpression of effector genes in graft-infiltrating CD8+ T cells after CD3 antibody therapy . C57BL/6 mice were transplanted under the kidney capsule with BALB/c pancreatic islets and treated or not with CD3 Abs on day +7 post-transplant . Individual CD8+ T cells ( n = 72 ) present within the graft were FACS sorted on day +14 or day +100 post-transplant and subjected to multiplex gene expression analysis . ( A ) Proportion of CD8+ T cells among the 72 cells tested that expressed Gzma , Gzmb , Prf1 , Fasl , Tbx21 , Eomes and Klrg1 mRNA in each group . ( B ) Polyfunctionality distribution of intragraft CD8+ T cells . ( C ) Coexpression of inflammatory and cytotoxic molecules by individual graft infiltrating CD8+ T cells . Each row represents one individual cell that is numbered . Each column represents a different gene . For better visualization of coexpression patterns , individual cells were ordered by the degree of gene coexpression . ( D ) FasL expression by intragraft CD8+ T cells 14 days after transplant and CD3 Ab therapy . ( Figures 1A and 1B: χ2 test , p<0 . 03 , p<0 . 01 , p<0 . 0003 , *p<0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 00310 . 7554/eLife . 08133 . 004Figure 1—figure supplement 1 . IFNγ responses by intragraft T cells after CD3 Ab therapy . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 00410 . 7554/eLife . 08133 . 005Figure 1—figure supplement 2 . Multiplex gene expression analysis on individual splenic CD8+ T cells after CD3 Ab therapy . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 00510 . 7554/eLife . 08133 . 006Figure 1—figure supplement 3 . Expression of inflammatory , cytotoxic and apoptotic markers by CD8+ T cells . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 006 In CD3 Ab-treated recipients , on day +14 after transplantation , expression of Prf1 , Tbx21 , Klrg1 and Gzmb by intragraft CD8+ T cells was clearly reduced as compared to untreated mice ( Figure 1A ) . The frequency of cells coexpressing 3 or more genes was significantly decreased ( from 33 . 3% to 15 . 3% ) while the number of cells expressing only one gene doubled after CD3 Ab treatment ( Figure 1B ) . A dramatic decrease in Gzmb+Prf1+ CD8+ T cells was observed ( Figure 1C ) . Contrasting with these findings , Fasl expression was enhanced as compared to untreated controls ( 50% versus 30 . 6% ) . The increased FasL expression on intragraft CD8+ T cells was confirmed by flow cytometry ( Figure 1D ) . Among this Fasl+ subset , 38% co-expressed Gzmb ( versus 62% in untreated recipients ) and 33 . 3% did not express any other effector genes . On day 100 post-transplant , expression of Fasl , Prf1 , Tbx21 and Eomes was detected in less than 5% of intragraft CD8+ T cells in CD3 Ab-treated tolerant mice ( Figure 1A ) . Only 5 . 6% of CD8+ T cells coexpressed 3 or more genes and 43 . 1% of the cells did not express any of the 7 inflammatory/cytotoxic genes tested ( Figure 1B , C ) . Gzmb was detected in 26% of the cells but was not associated with Fasl or Prf1 suggesting the absence of potent cytotoxic function ( Figure 1A–C ) . In the spleen , the single cell gene profile differed from that of intragraft T cells . A large proportion ( 56% ) of CD8+ T cells from untreated mice coexpressed Gzmb , Tbx21 and Klrg1 but nor Prf1 or Fasl on day +14 post-transplant ( Figure 1—figure supplement 2 ) . Such coexpression was abrogated after CD3 Ab therapy as 52% and 37 . 5% of spleen CD8+ T cells expressed none or only one of the 7 selected genes , respectively ( Figure 1—figure supplement 2B and 2C ) . RT-qPCR on total islet allografts confirmed the decreased expression of Gzmb , Prf1 , Tbx21 and Ifng ( Figure 1—figure supplement 3A ) . Fasl expression was also reduced on day +14 post-transplant which contrasts with results from single cell PCR ( Figure 1 ) . We also analyzed expression of Fas mRNA which , interestingly , increased after administration of CD3 Abs on day +14 post-transplant and decreased hereafter . We confirmed by flow cytometry that CD3 Abs rapidly induced Fas expression on CD8+ T cells after a 24 hr in vitro stimulation and most of them co-expressed FasL at 48 hr ( Figure 1—figure supplement 3B ) . To investigate whether CD3 Ab-induced depletion of alloreactive CD8+ T cells was dependent on the Fas/FasL pathway , monoclonal antibodies neutralizing FasL were administered in C57BL/6 recipients at the time of CD3 Ab therapy . Long-term islet allograft survival was not observed in this condition ( Figure 2A ) . FasL blockade abrogated CD3 Ab depleting effect: CD8+ T cells were detected within the islet allografts at levels comparable to that found in untreated recipients ( Figure 2B ) . FasL expression was drastically reduced on infiltrating CD8+ T cells ( Figure 2C ) . 10 . 7554/eLife . 08133 . 007Figure 2 . FasL blockade reversed CD3 antibody-induced transplant tolerance . ( A ) Graft survival of BALB/c islets was measured in C57BL/6 mice treated at day 7 with CD3 antibodies ( 50 μg , 5 days ) alone ( n = 10 ) or combined with neutralizing antibodies against FasLigand injected at the dose of 500 μg i . p . on day 6 , 7 and 8 post-transplant ( n = 4 ) ( p<0 . 02 between anti-CD3 and anti-CD3+anti-FasL Ab-treated mice ) . ( B ) Additional untreated ( n = 6 ) , CD3 Ab ( n = 6 ) or CD3+FasL Ab ( n = 2 ) -treated mice were sacrificed on day +14 post-transplant and proportion of CD8+ T cells was analyzed in the spleen , renal draining lymph nodes ( dLN ) and the islet allografts ( p<0 . 003 , p<0 . 0006 ) . ( C ) Expression of FasL by graft infiltrating CD8+ T cells isolated on day +14 from untreated ( n = 6 ) , CD3 Abs ( n = 6 ) or CD3+FasL Abs ( n = 2 ) -treated recipients ( p<0 . 002 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 007 We next characterized intragraft CD8+ T cells not eliminated by CD3 Ab treatment . On day 14 post-transplant , most of them were CD69+ CD44highCD62LlowCD45RBlow ( Figure 3A and Figure 3—figure supplement 1 ) . They expressed low levels of CD122 ( the β subunit of the IL-2 and IL-15 receptors ) and CD25 as opposed to intragraft CD8+ T cells from untreated mice ( Figure 3A ) . Expression of the proliferation marker Ki67 was also strongly reduced after CD3 Ab therapy . In contrast , the inhibitory receptors PD-1 , PD-L1 and LAG3 were upregulated ( 58 . 5% , 60 . 9% and 58 . 6% versus 25% , 26 . 1% and 27% without treatment , respectively ) and were mostly co-expressed by the same cells ( Figure 3A , B ) . This phenotype was maintained over long-term since a large proportion of intragraft CD8+ T cells still expressed these inhibitory receptors on day +100 post-transplant . A similar trend was observed in the spleen and draining lymph nodes where CD8+ T cells showed an increase in PD-1 , PD-L1 and LAG-3 expression after CD3 Ab therapy ( Figure 3—figure supplement 2 ) . 10 . 7554/eLife . 08133 . 008Figure 3 . Phenotypic and functional characteristics of tolerant CD8+ T cells . Pancreatic islet allografts were recovered from C57BL/6 mice after CD3 Ab treatment administered at day +7 post-transplant . ( A ) Expression of CD44 , CD62 , CD69 , CD45RB , CD122 , Ki67 , CD25 , PD-1 , PD-L1 and LAG-3 ( 6–16/group ) was evaluated on CD8+ T cells on day +7 , day +14 and day +100 post-transplant ( p<0 . 05 , p<0 . 01 , p<0 . 001 ) . ( B ) Co-expression of PD-1/PDL-1 , PD-1/LAG-3 and PD-L1/LAG-3 on graft-infiltrating CD8+ T cells recovered from CD3 Ab-treated mice on day +14 post-transplant . ( C ) Expression of PD-L1 , CD25 and T-bet by graft infiltrating CD8+ T cells isolated on day14 from untreated , CD3 Abs or CD3FasL Abs-treated recipients ( n = 2–6/group ) ( p<0 . 005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 00810 . 7554/eLife . 08133 . 009Figure 3—figure supplement 1 . Mean fluorescence intensity of CD44 and CD62L expressed by intragraft CD8+ T cells after CD3 Ab therapy . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 00910 . 7554/eLife . 08133 . 010Figure 3—figure supplement 2 . Phenotype of peripheral CD8+ T cells after CD3 Ab therapy . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 010 Interestingly , CD3 antibody-induced upregulation of PD-L1 on intragraft CD8+ T cells was inhibited following in vivo FasL blockade ( Figure 3C ) . In addition , expression of CD25 and T-bet were significantly increased in recipients treated with the combination of CD3 and FasL Abs as compared to CD3 Abs alone , reaching levels similar to those observed in untreated mice ( Figure 3C ) . To investigate the role of the PD-1/PDL-1 pathway in CD3 Ab-induced transplant tolerance , we treated C57BL/6 mice at day 7 with CD3 Abs together with monoclonal Abs to PD-1 or PD-L1 . In both cases , recipients rejected their graft with a median survival of 33 . 7 ± 1 . 5 days or 28 . 6 ± 5 . 1 days , respectively ( Figure 4A , B ) . After administration of PD-1 Abs , most graft infiltrating CD8+ T cells were CD44highCD62LlowTbet+ and strongly proliferated ( Figure 4—figure supplement 1A ) . Anti-PD-L1 Abs were also injected in tolerant recipients , on day 100 after transplantation and CD3 Ab therapy . Blockade of PD-L1 at the time of established tolerance precipitated graft rejection ( Figure 4C ) . 10 . 7554/eLife . 08133 . 011Figure 4 . Transplant tolerance and CD8+ T cell anergy rely on the PD-1/PDL-1 pathway . Graft survival of BALB/c islets was measured in C57BL/6 mice treated at day 7 with a combination of anti-CD3 F ( ab’ ) 2 and neutralizing antibodies against PD-1 ( panel A , n = 5 ) or PD-L1 ( panel B , n = 5 ) injected at the dose of 500 μg i . p . every other day for a total of 5 injections . ( **p<0 . 0001 between anti-CD3 and anti-CD3+anti-PD-1/anti-PD-L1 Ab-treated mice ) . ( C ) C57BL/6 mice showing long-term islet graft acceptance after CD3 Ab therapy were treated on day +100 post-transplant with anti-PD-L1 antibodies or isotype control IgG2a ( n = 5 ) . Graft rejection occurred 2–3 weeks later ( p<0 . 03 ) . ( D ) CD8+ T lymphocytes were purified from the spleen of CD3 Ab-treated tolerant C57BL/6 mice and were transferred into C57BL/6 Rag-/- mice ( 3x106/recipient ) . Recipient mice were grafted with pancreatic islets from BALB/c on day 0 and graft survival was monitored . On day +100 post-transplant , anti-PD-L1 antibodies or isotype control IgG2a were injected ( n = 5 ) ( p<0 . 007 ) . ( E ) Tolerant CD8+ T cells were detected in the spleen , draining lymph nodes and islet allograft of C57BL/6 Rag-/- recipients . IFNγ production and CD44 expression were compared to the ones of CD8+ T cells recovered after treatment with anti-PD-L1 antibodies or of CD8+ T cells issued from untreated C57BL-6 mice and rejecting the islet graft . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 01110 . 7554/eLife . 08133 . 012Figure 4—figure supplement 1 . Administration of anti-PD-1 or anti-PD-L1 Abs reversed CD3 Ab-induced unresponsiveness of CD8+ T cells . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 012 To further dissect the role of this pathway in CD8+ T cell tolerance , we adoptively transferred purified spleen CD8+ T cells from CD3 Ab-treated tolerant mice into immunodeficient RAG-/- recipients transplanted on the same day with BALB/c pancreatic islets . Islet grafts survived in these recipients . Transferred CD8+ T cells were detected in significant proportion in the spleen , draining lymph nodes and within recipients’ transplants ( Figure 4D–E ) ; they stained CD44high but did not produce high amounts of IFNγ in response to PMA/ionomycin stimulation ( Figure 4E and Figure 4—figure supplement 1B ) . Administration of PD-L1 Abs on day +100 post-transplant provoked graft rejection ( Figure 4C ) and restored the capacity of CD8+ T cells to secrete IFNγ ( Figure 4D–E and Figure 4—figure supplement 1B ) . To assess the role of donor alloantigens in sustaining CD8+ T cell unresponsiveness , RAG-/- C57BL/6 recipients were grafted under the kidney capsule with BALB/c pancreatic islets either on the same day or more than two months after transfer of purified CD8+ T cells recovered from tolerant mice , to exclude any impact of homeostatic proliferation on T cell functional properties . In this later condition , all grafts were rejected contrasting with the long-term survival obtained when cell transfer and islet grafts were both performed on day 0 . ( Figure 5 ) . A 4-week interval between infusion of tolerant CD8+ T cells and islet allografts also led to graft rejection ( Figure 5 ) . Control syngeneic islet grafts were tolerated . 10 . 7554/eLife . 08133 . 013Figure 5 . CD8+ T cell anergy depends on the presence of the antigens . CD8+ T lymphocytes were purified from CD3 Ab-treated tolerant C57BL/6 mice on day +100 post-transplant and were adoptively transferred into C57BL/6 Rag-/- mice ( 3x106/recipient , day 0 ) . C57BL/6 Rag-/- recipients were transplanted with pancreatic islets from BALB/c donors either on day 0 ( D0 , n = 4 ) or 4 weeks ( 4wks , n = 4 ) or 2 months ( 2mo , n = 4 ) after cell transfer ( p<0 . 007 ) . Syngeneic islets from C57BL/6 donors were grafted 4 weeks after cell transfer and were used as controls . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 013 We next analyzed the role of TGFβ in our model . RT-qPCR results on islet grafts recovered from CD3 Ab-treated recipients showed a significant increase of Tgfb1 mRNA , as compared to untreated controls ( Figure 6A ) . In addition , the intragraft Tgfb1/Ifng ratio was increased over long-term . To further identify the source of TGFβ , we sorted individual T lymphocytes , recovered on day 14 post-transplant , either from the spleen or from the islet grafts of 3 recipient mice treated with CD3 Abs . Expression of Tgfb1 mRNA was determined by single cell qPCR on 60 CD8+ and 48 CD4+ T cells . Results showed that Tgfb1 mRNA was expressed in 52% and 65% of CD4+ and CD8+ T cells , respectively ( Figure 6B ) . In untreated recipients , only 9% of CD4+ and 6% of CD8+ intragraft T cells expressed Tgfb1 . Tgfb1 mRNA was expressed in very few T cells recovered from the spleen of the same untreated or CD3 Ab-treated mice . Administration of a neutralizing antibody to TGFβ partially , yet significantly , abrogated CD3 Ab-induced tolerance ( Figure 6C ) . To assess whether TGFβ signaling within T cells was involved in the effect observed , we used as recipients transgenic DnTGFβRII C57BL/6 mice that express a dominant negative form of the human TGFβ receptor II under the control of the mouse CD4 promoter ( Gorelik and Flavell , 2000 ) . In these mice , the dnTGFBRII-CD4 transgenic construct lacks the CD8 silencer . Thus , expression of the transgene blocks TGFβ signaling specifically in both CD4+ and CD8+ T cells . Treatment of these transgenic recipients with CD3 Ab on day +7 post-transplant did not induce long-term survival of islet allografts ( median survival of 36 . 8 ± 6 . 1 days ) ( Figure 6D ) . 10 . 7554/eLife . 08133 . 014Figure 6 . CD3 antibody-induced transplant tolerance depends on in situ TGFβ production and signaling in T cells . ( A ) Expression of Tgfb1 mRNA and evaluation of the Tgfb1/Ifng ratio in pancreatic islet allografts recovered at day 14 from untreated mice or day 14 , 21 and 100 after transplantation from CD3 Ab-treated recipients . Data are shown as mean ± SEM of 5–9 individual samples ( p<0 . 05 , p<0 . 01 ) . ( B ) Single cell PCR: individual CD4+ ( n = 48 ) and CD8+ ( n = 60 ) T cells were sorted on day +14 post-transplant from the spleen or islet allografts recovered from C57BL/6 mice treated or not with CD3 antibodies . Expression of Tgfb1 mRNA was measured in each cell . Results show the proportion of CD4+ and CD8+ T cells positive for Tgfb expression ( p<0 . 05 ) . ( C ) Graft survival of BALB/c islets in wild-type C57BL/6 mice treated at day +7 with anti-CD3 F ( ab’ ) 2 and neutralizing TGFβ antibodies ( 1 mg/injection , twice a week for 3 weeks ) ( n = 4 to 8 per group , p<0 . 05 ) . ( D ) Abrogation of tolerance in DnTGFβRII C57BL/6 mice transplanted with BALB/c pancreatic islets and treated with CD3 antibodies on day7 post-transplant ( n = 5 per group , *p = 0 . 0002 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 014 We asked whether TGFβ could influence the expression of PD-1 or PD-L1 on CD8+ T cells . T cells were stimulated in vitro for 24 hr with CD3 Abs in presence of neutralizing Abs to TGFβ or recombinant TGFβ . Co-expression of PD-1 and PD-L1 was induced after TCR stimulation ( Figure 7—figure supplement 1 ) . Blockade of TGFβ downregulated expression of PD-1 and even more drastically that of PD-L1 . In contrast , addition of recombinant TGFβ to the culture further promoted PD-1 but not PD-L1 expression . To assess the relevance of this pathway in vivo , islet allograft recipients were treated with CD3 Abs either alone or combined with TGFβ Abs . Paralleling the in vitro data , in vivo neutralization of TGFβ downregulated PD-1 and PD-L1 co-expression on graft infiltrating CD8+ T cells ( Figure 7 ) . 10 . 7554/eLife . 08133 . 015Figure 7 . Induction of PD-1 and PD-L1 expression on intragraft CD8+ T cells is regulated by TGFβ . C57BL/6 mice were transplanted with BALB/c pancreatic islets and treated at day 7 with anti-CD3 F ( ab’ ) 2 with or without neutralizing TGFβ antibodies . Mice were sacrificed on day 14 post-transplant and PD-1 and PD-L1 expressions were analyzed on graft-infiltrating CD8+ T cells . ( A ) Co-expression of PD-1 and PD-L1 on CD8+ T cells . ( B ) Histograms representing PD-1 and PD-L1 expression . ( C ) Median fluorescence intensity of PD-1 and PD-L1 ( p<0 . 02 , **p<0 . 005 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 01510 . 7554/eLife . 08133 . 016Figure 7—figure supplement 1 . TGFβ modulates PD-1/PD-L1 expression on CD8+ T cells . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 016 Using the Agilent platform , we compared intragraft CD8+ T cells from CD3 Ab-treated tolerant mice to those from untreated mice showing acute rejection . Functional grouping analysis showed that gene categories related to effective T cell responses were downregulated in CD3 Ab treated mice as compared to controls ( Figure 8A ) . Accordingly , pathways induced by IL-2 or TCR signaling were also downregulated ( Figure 8B and Figure 8—figure supplement 1 ) . Analysis of immune genes differentially expressed between the two groups revealed a decreased expression of genes involved in cell migration ( Cxcl9 , Ccr10 , Nrp1 , Cd44 ) , proliferation ( Fos , Jun , Ccnd1 ) , co-stimulation ( Cd80 , Icos , Cd200 , Ctla4 ) , inflammatory cytokine signaling ( Tnf , Il2ra , Il1r2 , Il18r1 , Tnfrsf21 ) and environmental sensing ( hypoxia-inducible factor Hif1a ) ( Figure 8C ) . 10 . 7554/eLife . 08133 . 017Figure 8 . Transcriptomic analysis of tolerant intragraft CD8+ T cells after CD3 Ab therapy . Five hundred graft-infiltrating CD8+ T lymphocytes were sorted from untreated or CD3 Ab-treated C57BL/6 mice , on day +14 and day +100 , respectively ( n = 8 per group ) . Agilent Whole Mouse Genome Microarrays were performed after amplification of RNAs . Functional grouping analysis used annotations derived from Gene Ontology ( fold-change >2 , p<0 . 05 ) . ( A ) Bar chart showing the frequency of representative categories downregulated ( left panel ) or upregulated ( right panel ) in tolerant CD8+ T cells as compared to CD8+ T cells from untreated recipient mice . Below , statistically enriched categories were indicated by their adjusted p-value ( only the top 10 categories , Fisher’s exact test with Benjamini-Hochberg correction for multiple testing ) . ( B ) Bar chart showing the frequency of target pathways induced by IL-2 ou TCR signaling . ( C ) Selection of immune genes that were downregulated ( grey bars ) or upregulated ( red bars ) in tolerant CD8+ T cells as compared to CD8+ T cells from untreated recipient mice ( fold-change >2 , p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 01710 . 7554/eLife . 08133 . 018Figure 8—source data 1 . Genes upregulated in anergic CD8+ T cells from CD3 Ab-treated recipient mice . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 01810 . 7554/eLife . 08133 . 019Figure 8—figure supplement 1 . Heatmap of genes induced by IL-2 ( A ) or TCR ( B ) that are differentially expressed ( >twofold , p<0 . 05 ) in CD8+ T cells recovered from pancreatic islet allografts of CD3 antibody-treated tolerant mice ( day +100 post-transplant ) or of untreated mice ( day +14 post-transplant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08133 . 019 In tolerant CD8+ T cells , categories related to metabolic processes ( nucleotides , energy , carbohydrates , lipids… ) as well as cell cycle arrest were enriched . In addition , these cells showed upregulation of genes inhibiting or regulating T cell functions: inhibitors of cell proliferation ( Hopx , Cdkn2c , Map2k6 , Mapk9 ) , of cytokine signaling ( Socs2 , Socs6 ) , of differentiation ( Eid2 , Id2 ) , of co-stimulation ( Btla ) ( Figure 8C and Figure 8—source data 1 ) . Eomes , encoding for the IFNγ transcription factor eomesodermin , was also upregulated in tolerant CD8 T cells . In the present study , using a pancreatic islet allograft model and CD3 Ab therapy , we provide novel insights into the immune mechanisms driving and sustaining CD8+ T cell anergy thus leading to long-term graft survival . We demonstrate that CD3 Abs selectively deplete graft-infiltrating Gzmb+Prf1+ cytotoxic CD8+ T cells by a FasL-mediated pathway . This depletion was accompanied by the onset of anergy in remaining CD8+ T cells which was dependent on the presence of the alloantigens and on an in situ crosstalk between the PD-1/PD-L1 and TGFβ/TGFβRII pathways . We previously reported that a short-term course with CD3 Abs induced long-term graft survival and antigen-specific tolerance provided the treatment was applied in a defined therapeutic window , that is at the time of effector T cells priming to the alloantigens ( Goto et al . , 2013; You et al . , 2012 ) . Mechanistic studies revealed that regulatory and effector T cells were differentially affected by the treatment . Foxp3 Tregs were relatively spared from CD3 Ab-induced depletion and could transfer antigen-specific tolerance , suggesting that they play an important role in sustaining graft survival ( Baas et al . , 2013; Goto et al . , 2013; You et al . , 2012 ) . In contrast to Tregs , CD3 Abs preferentially induced apoptosis of activated effector T cells ( You et al . , 2012 ) . Long-term graft survival correlated with an absence of donor-specific CD8+ T cell responses at the periphery ( You et al . , 2012 ) . Here , T cell unresponsiveness was confirmed at the tissue level as graft infiltrating T cells from CD3 Ab-treated recipient mice did not mount efficient IFNγ responses to donor antigens . By single cell gene profiling using a method which proved optimal to discriminate functional patterns of CD8 T cells facing an immune stimulation ( Peixoto et al . , 2007; Ribeiro-dos-Santos et al . , 2012 ) , we identified two distinct subsets of graft infiltrating CD8+ T cells , co-expressing either Gzmb/Fasl or Gzmb/Prf1 mRNAs . Interestingly , Gzmb+Prf1 cytotoxic CD8 T cells were selectively depleted after CD3 Ab therapy . The frequency of individual CD8 T cells expressing Fasl was enhanced at the end of CD3 Ab treatment but it was not associated with Gzmb coexpression in contrast to what observed in untreated recipients . Flow cytometry analysis confirmed the increase of FasL+CD8+ T cells . Previous studies showed that signaling through the TCR/CD3 complex promotes activation-induced cell death ( AICD ) in activated effector T cells through a Fas/FasL interaction , but independently of perforin and granzymes ( Brunner et al . , 1995; Sobek et al . , 2002 ) . T cells expressed both Fas and FasL after CD3 Ab treatment . Our results suggest , together with previous reports using T cell hybridomas or clones , that apoptosis has occurred by triggering of Fas by FasL-expressing neighboring T cells ( Ayroldi et al . , 1997; Brunner et al . , 1995 ) . Rejection of islet allograft following FasL blockade further confirmed the Fas-dependence of CD3 antibody-induced depletion of alloreactive T cells . Therefore , our data concur to show that CD3 Abs induced apoptosis of Gzmb+Prf1 CD8+ T cells by a Fas/FasL-mediated pathway . Apoptosis of alloreactive effectors was followed by the establishment of anergy in the remaining graft-infiltrating CD8+ T cells . Single cell gene profiling showed that expression of Gzmb , Prf1 , Tbx21 , Eomes and Klrg1 mRNAs decreased after CD3 Ab therapy as well as the simultaneous expression of 3 and more of the 7 genes tested . This downregulated gene pattern was even more obvious on day +100 post-transplant as intragraft CD8+ T cells were characterized by a complete absence of co-expression of cytotoxic and inflammatory genes ( i . e . 80% of the cells expressed no or only one of the selected genes ) , a hallmark of profound intrinsic unresponsiveness . Gzmb was detected in 25% of the cells but was not associated with Prf1 or Fasl showing that these CD8+ T cells were deprived of effective killing ability . Our findings are reminiscent of the pioneer experiments of Rocha and Von Boehmer who demonstrated the existence of anergy in vivo using TCR transgenic models . They described that female CD8 T cells specific for the male antigen HY rapidly expanded when transferred into male nude recipients and most of them died by AICD . The remaining transgenic CD8+ T cells were intrinsically unresponsive to TCR stimuli ( Rocha and von Boehmer , 1991 ) . Importantly , when parked into a second female nude recipient for 2 months and subsequently transferred into a third male nude mice , the T cells regained their ability to respond to the HY antigen showing that antigen persistence was required to maintain CD8 T cell anergy ( Rocha et al . , 1993 ) . In our model , the continuous presence of the alloantigen was essential to sustain CD8+ T cell unresponsiveness in vivo as islet allografts transplanted 2 months after infusion of CD8+ T cells from tolerant mice into RAG-/- recipients were rapidly rejected . Our data suggest that , within islet allografts , chronic antigen encounter induces a suboptimal activation that reinforces anergic signals in CD8+ T cells and therefore sustains CD8 T cell tolerance . Transcriptome analysis of graft infiltrating CD8+ T cells recovered on day +100 post-transplant allowed us to identify a gene signature for tolerant T cells . Genes encoding for negative regulators of proliferation or differentiation ( Eid2 , Cdkn2c , Hopx , Id2 ) were overexpressed . Hopx ( homeodomain-only protein ) , identified as a master regulator of the anergic state of induced Tregs , inhibits the expression of the AP-1 complex ( Hawiger et al . , 2010 ) . Accordingly , Fos and Jun that compose the AP-1 complex were downregulated in tolerant CD8+ T cells as well as their downstream target cyclin D1 ( Ccnd1 ) ( Kang et al . , 1992; Sundstedt et al . , 1996 ) . Additionally , the decreased expression of Cxcl9 , encoding an IFNγ inducible chemokine contributing to the recruitment of allograft reactive T cells ( Medoff et al . , 2006 ) , as well as that of Tnf , Il2ra , Lgals3 , Il1r2 , Il18r1 and Tnfrsf9 revealed the inhibition of Th1 inflammatory responses . Tolerant CD8+ T cells also downregulated the hypoxia-inducible factor Hif1a that has been shown to positively regulate T cell differentiation and effector functions by promoting aerobic glycolysis ( Doedens et al . , 2013; Finlay et al . , 2012 ) . Finally , we found an overexpression of Eomes in tolerant CD8+ T cells . This was unexpected as eomesodermin ( Eomes ) plays well-described roles in cytotoxic CD8+ T cell differentiation and memory formation ( Intlekofer et al . , 2005; Pearce et al . , 2003 ) . However , investigations in models of chronic infections in human and mouse ( HIV , LCMV ) have demonstrated that expression of Eomes by virus-specific CD8 T cells was associated with high expression of inhibitory receptors and impaired functions ( Buggert et al . , 2014; Doering et al . , 2012; Paley et al . , 2012 ) . In our transplant model , as detailed below , the PD-1/PD-L1 pathway is mandatory for CD8 T cell tolerance . Thus , increased expression of Eomes in conjunction with PD-1/PD-L1 may characterize anergized CD8+ T cells induced after CD3 Ab therapy . This finding highlights the issue of a molecular link between Eomes and PD-1/PD-L1 . Lastly , one important feature common to our model and the infectious setting is the continuous presence of the cognate antigens . Therefore , we may hypothesize that chronic antigenic stimulation delivers signals inducing a preferential and sustained expression of Eomes . Our study highlighted a key contribution of the PD-1/PD-L1 and the TGFβ/TGFβRII pathways and we identified a new role for TGFβ in modulating PD-1 and PD-L1 expression on CD8+ T cells . After CD3 Ab therapy , graft-infiltrating CD8+ T cells exhibited a CD44highCD62LlowCD69+CD45RBlow antigen-experienced phenotype and coexpressed the inhibitory receptors PD-1 , PDL-1 and LAG-3 . The PD-1/PD-L1 pathway played a predominant role in the induction and the maintenance phase of CD3 Ab-mediated tolerance as neutralization of either receptor completely abrogated the therapeutic effect . More precisely , we demonstrated that the PD-1/PD-L1 signaling was mandatory for CD8+ T cell unresponsiveness . Indeed , after treatment with CD3 antibodies , intragraft CD8+ T cells showed a defective ability to proliferate , to produce IFNγ and to mount donor-specific responses . In vivo administration of anti-PD-1 or anti-PD-L1 antibodies restored the effector functions of allogeneic CD8+ T cells , as shown by the up-regulation of the proliferation marker Ki67 and the IFNγ transcription factor T-bet as well as the regained ability to secrete IFNγ , and resulted in islet allograft rejection . Our findings are in accordance with the well-described role of the PD-1/PD-L1 pathway as a master regulator of immune responses notably through its ability to limit effector T cell interaction with dendritic cells and to abort their activation ( Fife et al . , 2009; Francisco et al . , 2010 ) . In various settings , PD-1 has been identified as a marker of dysfunctional CD8+ T cells , and it contributed to the establishment of tolerance ( Barber et al . , 2006; Haspot et al . , 2008; Ito et al . , 2005; Lucas et al . , 2011; Riella et al . , 2012 ) . We found that CD8+ T cells coexpressed PD-1 and PD-L1 , which may contribute to the formation of stable interactions with other CD8+ T cells as well as other cells . PD-L1 is widely expressed on hematopoietic and nonhematopoietic cells and its expression increases with activation ( Keir et al . , 2008 ) . In addition , aside from PD-1 , PD-L1 can interact with B7-1 and a recent report demonstrated that this interaction contributed to the control of allogeneic T cell responses ( Keir et al . , 2008; Yang et al . , 2011 ) . We can thus hypothesize that PD-L1 engagement on graft-infiltrating CD8+ T cells may deliver inhibitory signals and that these bidirectional interactions may promote and sustain T cell anergy and graft survival . Another novel finding points to TGFβ as a regulator of PD-1 and PD-L1 expression and as a key mediator of CD3 Ab-induced allotolerance . We have previously shown in autoimmunity that CD3 antibodies cured type 1 diabetes and restored self-tolerance in a TGFβ-dependent manner ( Belghith et al . , 2003 ) . The present results not only extend these findings to transplant tolerance but also provide new insights into the key regulatory role of this cytokine . First , the single cell PCR results showed that a large proportion of CD8+ T cells ( as well as CD4+ T cells ) present within the islet allografts produced TGFβ after CD3 antibody administration . This increased expression of TGFβ was restricted to the graft as it was not observed in the spleen . Secondly , we demonstrated that TGFβ signaling in T cells is mandatory for the induction of immune tolerance as CD3 Ab treatment failed to induce long-term islet graft survival in recipients presenting a T cell-selective mutated TGFβRII ( Gorelik and Flavell , 2000 ) . This result argues for an in situ autocrine/paracrine effect of TGFβ on allogeneic T cells . Third , we demonstrated a direct link between the TGFβ and the PD-1 pathways as TGFβ blockade , through the administration of neutralizing antibodies , downregulated PD-1 and PD-L1 expression on intragraft CD8+ T cells and abrogated the tolerogenic properties of CD3 Abs . This inhibitory effect was even more drastic on PD-L1 than PD-1 showing that CD3 Ab-induced PD-L1 expression on CD8+ T cells was highly dependent on TGFβ/TGFβR signaling . In conclusion , our study highlighted new facets of immune mechanisms driving CD8+ T cell peripheral tolerance and permanent acceptance of fully mismatched allografts after CD3 Ab therapy . Tolerance was established through elimination of highly cytotoxic CD8+ T cells followed by the induction of CD8+ T cells anergy which depended on a cross-talk between the PD-1/PD-L1 and TGFβ/TGFβRII pathways acting in an autocrine and paracrine manner in the graft environment . These mechanistic findings support the therapeutic potential of CD3 Ab which , concerning clinical translation , have shown potential value in the treatment of patients with new onset type 1 diabetes ( Herold et al . , 2002; Keymeulen et al . , 2005; Sherry et al . , 2011 ) . C57BL/6 , RAG-/- C57BL/6 , and BALB/c female mice were bred in our facility under specific pathogen-free conditions . DnTGFβRII female C57BL/6 mice were obtained from the Jackson Laboratory ( Bar Harbor , USA ) . Blood glucose was measured using ACCU-CHECK Performa glucometer ( Roche Diagnostics , Meylan , France ) . Experiments were conducted according to European Directive ( 2010/63/UE ) and were approved by the Ethical Committee of Paris Descartes University ( registered number: 14–075 ) . Pancreatic islets were separated by density gradient centrifugation ( Histopaque , Sigma-Aldrich , Lyon , France ) after in situ digestion with collagenase P ( Roche Diagnostics , Meylan , France ) and transplanted ( 300 islets ) under the kidney capsule of diabetic recipients . Diabetes was induced 3 to 4 days after a single injection of streptozotocin ( Sigma-Aldrich ) at 225 mg/kg . Diagnosis of graft rejection was made after three glucose measurements >250 mg/dl . The murine myeloma cell line SP2/0 producing the genetically engineered F ( ab’ ) 2 fragments of the hamster anti-mouse CD3ε antibody 145-2C11 ( Kostelny et al . , 1992 ) was provided by J . A . Bluestone ( UCSF , San Francisco , CA ) The antibody was purified by protein G–Sepharose affinity chromatography . CD3 F ( ab’ ) 2 were injected i . v . at the dose of 50 µg/day for 5 consecutive days , starting on day 7 post-transplant . Anti-PD-L1 hybridoma ( MIH5 ) was kindly provided by Pr . M . Azuma ( Tokyo Medical and Dental University ) . Purified anti-PD-1 ( RMP1-14 ) and anti-FasLigand antibodies ( MLF4 ) were provided by Pr . H . Yagita ( Juntendo University School of Medicine , Tokyo ) . TGFβ antibodies ( 2G . 7 ) were produced in house . For flow cytometry , all antibodies were from BD Biosciences ( Pharmingen , San Diego , CA , USA ) except Foxp3 which was from eBioscience ( San Diego , CA , USA ) . RAG-/- C57BL/6 mice were reconstituted with 3 . 106 CD8+ T cells isolated from the spleen of CD3 Ab-treated tolerant C57BL/6 . The next day , recipients were transplanted with 300 BALB/c islets . PD-L1 antibodies were administered 100 days after transplantation . In another experiment , islet allografts were performed 4 weeks or 2 months after CD8+ T cell transfer . mRNA was isolated using the mMACS isolation Kit and reverse transcribed into cDNA using the mMACS One-step cDNA Kit ( MACS Molecular , Miltenyi Biotec , Bergisch Gladbach , Germany ) . RT-qPCR was performed on an ABI 7900HT fast real-time PCR system using primers , probes and master mixes from Applied Biosystems ( Life Technologies , Carlsbad , CA , USA ) . HPRT was the housekeeping gene . Individual CD8+ T cells were FACS sorted from the spleen or the islet allografts of untreated or CD3 Ab-treated recipients . After cell lysis by heating/cooling steps , RNA was specifically retrotranscribed using MuLV Reverse Transcriptase ( Applied Biosystems ) and 3’ specific primers ( Eurofins MWG , Ebersberg , Germany ) . The resulting cDNA was next amplified ( first PCR with all primers ) . Product of this first PCR was then subjected to a second PCR using SYBR Green PCR Master Mix ( Applied Biosystems ) for each primers pairs . To ensure that each well contained a T cell , Cd3e mRNA was amplified simultaneously with the genes of interest . HPRT was the housekeeping gene . Multiplex single cell PCR was performed for the following genes: granzymes A and B ( Gzma , Gzmb ) , perforin-1 ( Prf1 ) and FasLigand ( Fasl ) , which provide killing abilities , the transcription factors T-bet ( Tbx21 ) and Eomesodermin ( Eomes ) which control IFNγ expression , and the killer-cell lectin like receptor G-1 ( Klrg1 ) which marks terminally differentiated effector T cells . For each sample , 500 graft-infiltrating CD8+ T cells from individual tolerant CD3 Ab-treated recipients ( n = 8 ) or untreated recipients ( n = 8 ) were FACS sorted , lyzed using SuperAmpTMlysis buffer ( Miltenyi Biotec ) and stored at -80°C until all samples were collected . Amplification of RNA , sample hydridization ( Agilent whole mouse genome oligo microarrays ) , scanning and data acquisition were performed by Miltenyi Biotec . Functional grouping analysis has been performed using Gene Ontology . Gene expression data have been deposited in the GEO database ( accession number GSE68208 ) . Cumulative graft survival was calculated using the Kaplan-Meier method . The statistical comparison was performed using the logrank ( Mantel-Cox ) test . When appropriate , the Student's t tests or Chi square ( χ2 ) tests were used . A p value <0 . 05 was considered significant .	The immune system is always on guard for signs of infection or cells that have become diseased . When these signs are identified , a subset of white blood cells called CD8+ T cells leap into action , multiply in number and then act to eliminate the potential threat . While this response is essential to fighting off infections and other diseases like cancer , it can backfire in people with an organ transplant . Indeed , the CD8+ T cells can target and attack the cells of the transplanted organ causing the body to reject the organ . One way to avoid transplant rejection would be to turn off CD8+ T cells that have learned to recognize cells from the transplant . In fact , studies in 2012 and 2013 showed that treating transplanted animals with an antibody that binds T cells protects a transplanted organ from attack . This treatment had to be given after the CD8+ T cells had recognized and began targeting the transplanted organ to be effective . But it was not clear exactly how this antibody treatment protected the transplant . Now , Baas , Besançon et al . – including some of the same researchers involved in the earlier studies – show that the antibodies used in the treatment selectively target and eliminate the attacking CD8+ T cells . This leaves behind only inactive CD8+ T cells that don’t harm the transplant . To do this , Baas , Besançon et al . transplanted pancreatic cells from mice into other mice with a diabetes-like disorder . Next , the experiments compared gene expression in CD8+ T cells found within the transplanted tissue in mice that were treated with the antibody and those that were not treated . The expression of many genes for toxic molecules was stopped after treatment with the antibody leaving the CD8+ T cells in an inactive state . In addition , the treated CD8+ T cells expressed more of a certain type of receptor ( called PD-1 and PD-L1 ) that acts as inhibitory checkpoint for the immune system . So , Baas , Besançon et al . treated transplanted mice with both the T cell-eliminating antibody and antibodies that block these inhibitory receptors to see what would happen . The transplanted organs were quickly attacked and rejected . This shows that the inhibitory receptors play a crucial role in helping to shut down attacking CD8+ T cells in the initial antibody treatment and allowed long-term survival of the transplanted organs . Blocking another protein called TGFβ in antibody-treated mice also caused organ rejection . The findings help explain how these antibodies protect transplanted organs and may help scientists trying to develop new anti-transplant rejection drugs in the future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2016	TGFβ-dependent expression of PD-1 and PD-L1 controls CD8+ T cell anergy in transplant tolerance
Information flow through neural circuits is determined by the nature of the synapses linking the subtypes of neurons . How neurons acquire features distinct to each synapse remains unknown . We show that the transcription factor Mafb drives the formation of auditory ribbon synapses , which are specialized for rapid transmission from hair cells to spiral ganglion neurons ( SGNs ) . Mafb acts in SGNs to drive differentiation of the large postsynaptic density ( PSD ) characteristic of the ribbon synapse . In Mafb mutant mice , SGNs fail to develop normal PSDs , leading to reduced synapse number and impaired auditory responses . Conversely , increased Mafb accelerates synaptogenesis . Moreover , Mafb is responsible for executing one branch of the SGN differentiation program orchestrated by the Gata3 transcriptional network . Remarkably , restoration of Mafb rescues the synapse defect in Gata3 mutants . Hence , Mafb is a powerful regulator of cell-type specific features of auditory synaptogenesis that offers a new entry point for treating hearing loss . Synapses are the basic building blocks for the diverse types of neural circuits that mediate perception and behavior . How information is processed and transmitted is determined both by the identity of the neurons in the circuit and the nature of their synaptic connections . Indeed , synapses exhibit a range of morphologies and functions , varying both the nature of the signal and its strength . Recent efforts to unravel the molecular basis of synaptic specificity have begun to define the mechanisms that match pre- and post-synaptic partners and thereby establish the overall wiring pattern . These studies suggest that the fate of each partner is determined by cell-type specific transcription factors , which activate expression of a unique combination of cell-surface receptors to permit adhesion only between appropriate partners ( Polleux et al . , 2007; Sanes and Yamagata , 2009 ) . In contrast , we still know very little about how each partner is subsequently instructed to develop the appropriate type of pre- or post-synaptic specialization required at that synapse . Although several broadly acting transcription factors have been shown to control synapse number , the identity of the transcription factors that might direct cell-type specific features of synaptic development remains elusive . Among the most specialized synapses in the nervous system are those of the auditory system that transmit sound information from inner hair cells ( IHCs ) to spiral ganglion neurons ( SGNs ) . As the sole link between IHCs and the brain , SGNs encode the frequency , intensity , and timing of all acoustic stimuli and communicate this information rapidly and faithfully to the brain ( Appler and Goodrich , 2011 ) . To achieve the necessary speed , SGNs receive information from IHCs via ribbon synapses that are specialized for signaling with high temporal precision ( Khimich et al . , 2005; Glowatzki et al . , 2008; Buran et al . , 2010; Safieddine et al . , 2012 ) . Unlike conventional synapses , ribbon synapses contain an electron-dense multi-protein ribbon structure , which tethers a large pool of readily releasable vesicles ( Sterling and Matthews , 2005; Goutman and Glowatzki , 2007; Frank et al . , 2010 ) , thereby enabling fast and sustained release of glutamate from the IHCs ( Glowatzki and Fuchs , 2002 ) . Likewise , SGNs respond with remarkable speed due to a large postsynaptic density ( PSD ) containing abundant clusters of AMPA-type receptors ( Matsubara et al . , 1996 ) . Since the features of all sound stimuli are captured by the pattern of synaptic transmission between IHCs and SGNs , developing the correct number and types of synapses is critical for hearing . Indeed , auditory synaptopathies underlie several forms of hearing loss ( Moser et al . , 2013 ) , and ribbon synapses are particularly vulnerable to acoustic overexposure ( Kujawa and Liberman , 2009; Lin et al . , 2011 ) . Despite the importance of ribbon synapses to the sense of hearing , we have only a rudimentary understanding of how this synapse acquires its specialized structures and properties during development . In mice , ribbon synapses begin to develop perinatally ( Sobkowicz et al . , 1982; Nemzou et al . , 2006; Sendin et al . , 2007; Marrs and Spirou , 2012 ) . Presynaptic ribbons are initially distributed throughout the IHC cytoplasm and then gradually localize to the basolateral surface to form an immature synapse , with clustered ribbons in large active zones ( Sobkowicz et al . , 1982; Sendin et al . , 2007 ) . In parallel , SGN neurites reach IHCs and begin to elaborate post-synaptic terminals , characterized by increased clustering of GluR2/R3 AMPA receptors in the nascent PSD ( Huang et al . , 2012 ) . Ribbon number peaks at the end of the first postnatal week , followed by a period of pruning and refinement . By the onset of hearing ( postnatal day 12 [P12] in mice ) , ribbon number has declined to adult levels and each ribbon is anchored opposite a specialized large , well-defined PSD in the SGN terminal ( Sobkowicz et al . , 1982; Nemzou et al . , 2006; Sendin et al . , 2007 ) . In mature animals , the size of the PSD correlates with the size and position of the ribbon ( Liberman et al . , 2011 ) , indicating that pre- and post-synaptic development must be tightly coordinated . Indeed , without SGN afferent terminals , ribbon development in IHCs is impaired ( Sobkowicz et al . , 1986 ) , and conversely , afferent terminals are not stabilized in mutants lacking normal ribbons ( Sheets et al . , 2011 ) . However , the molecular mechanisms that allow SGNs to form these unusual connections with IHCs remain a mystery . Efforts to understand how SGNs acquire their auditory-specific features have established an important role for the Gata3 transcription factor , which is expressed in SGNs as soon as they can be identified ( Lawoko-Kerali et al . , 2004 ) and then maintained throughout differentiation ( Appler et al . , 2013; Duncan and Fritzsch , 2013; Luo et al . , 2013 ) . Upon deletion of Gata3 , cochlear wiring is severely disrupted and SGNs fail to express a cohort of auditory-specific genes ( Appler et al . , 2013; Duncan and Fritzsch , 2013 ) , including additional transcription factors that may work downstream of Gata3 to direct specific features of SGN differentiation . One particularly intriguing putative downstream effector is the basic leucine-zipper transcription factor Mafb . Mafb belongs to the family of large Maf factors , which activate cell-type specific programs of gene expression essential for terminal differentiation of a wide variety of cell types , including pancreatic β cells , kidney podocytes , and macrophages ( Moriguchi et al . , 2006; Kataoka , 2007; Hang and Stein , 2011 ) . Mafb was first identified as the genetic cause of the kreisler mutation ( kr ) in mice ( Cordes and Barsh , 1994 ) . In kr mutants , rhombomeres 5 and 6 are not properly specified , and the otic vesicle , which develops adjacent to these structures , fails to acquire its normal morphology . Although its function in the nervous system remains unclear , Mafb is also required in the pre-Bötzinger complex in the hindbrain ( Blanchi et al . , 2003 ) : respiratory circuits are unable to mediate normal breathing in Mafb mutants , which die from central apnea at birth . Unfortunately , the early lethality and inner ear defects in null mutants have prevented analysis of Mafb’s role in the assembly and function of auditory circuits . In this study , through genetic analysis in mice , we demonstrate that Mafb is critical for the ability of SGNs to form the specialized contacts necessary for the sense of hearing . Mafb is enriched in SGNs during synaptogenesis . Further , deletion of Mafb from SGNs disrupts post-synaptic differentiation , leading to an overall reduction in ribbon synapse number and hence impaired auditory responses . Conversely , restoration of Mafb is sufficient to rescue the synaptic defect seen in Gata3 mutant mice . These studies establish Mafb as a lineage-specific intrinsic regulator in SGNs for post-synaptic specification of ribbon synapses and a powerful effector within a broader Gata3 network during auditory circuit assembly . The first clue that Mafb might serve as a lineage-specific regulator of SGN terminal differentiation came from microarray comparisons of SGNs and the closely related vestibular ganglion neurons ( VGNs ) , which revealed a dramatic SGN-specific increase in Mafb expression during late embryogenesis ( Lu et al . , 2011 ) . SGN neurogenesis occurs along a base to apex gradient , with most SGNs exiting the cell cycle by E12 . 5 and extending projections to reach the organ of Corti by E15 . 5 ( Matei et al . , 2005; Koundakjian et al . , 2007 ) . In situ hybridization confirmed that Mafb transcription initiates in parallel with the end of neurogenesis , with expression evident in the basal to middle regions of the cochlea by E13 . 5 ( Figure 1B ) . Protein was not yet present at this stage ( Figure 1C , C′ ) . Two days later , Mafb protein could be detected in SGNs in the basal and middle turns ( Figure 1D ) and by E16 . 5 , Mafb protein was localized to SGN nuclei along the entire length of the cochlea ( Figure 1E ) . High levels of Mafb persisted during early postnatal stages ( Figure 1F , J ) and peaked at P6 ( Figure 1G , J ) . Afterward , Mafb was gradually translocated from the nucleus to the cytoplasm , and only a few SGNs maintained strong nuclear expression by P10 ( Figure 1H , arrows ) . After the onset of hearing , Mafb was still present at low levels in the SGN cytoplasm ( Figure 1I , J ) . At all stages , Mafb expression was limited to SGNs , with no protein detected in VGNs ( data not shown ) or other cell types of the inner ear . Hence , Mafb defines the SGN subset of inner ear neurons and is expressed throughout auditory synaptogenesis , which begins around perinatal stages and lasts for the first two postnatal weeks , peaking around P6 ( Sobkowicz et al . , 1982; Huang et al . , 2012; Safieddine et al . , 2012 ) . 10 . 7554/eLife . 01341 . 003Figure 1 . Mafb is expressed in SGNs during synaptogenesis . ( A ) Schematic representation of a mouse inner ear indicates the levels of section planes in ( B–F ) . ( B ) In situ hybridization shows Mafb expression in E13 . 5 SGNs . ( C and C′ ) Mafb ( C ) and Neuronal Class III β-Tubulin ( TuJ1 ) ( C′ ) double staining at E13 . 5 shows a lack of Mafb protein in the spiral ganglion ( sg ) , which is marked by TuJ1 . ( D–F ) Anti-Mafb immunostaining of the cochlea at E15 . 5 ( D ) , E16 . 5 ( E ) and P3 ( F ) confirmed specific expression of Mafb in differentiating SGNs . Mafb protein can be detected in SGNs in the basal and middle turns at E15 . 5 ( D ) and in all SGNs at E16 . 5 ( E ) . High expression persists through early postnatal stages ( F ) . At all stages , Mafb protein is restricted to SGNs . Asterisks indicate the cochlear duct . a: apex; m: middle; b: base; sg: spiral ganglion; sv: scala vestibuli; sm: scala media; st: scala tympani . ( G–I ) Anti-Mafb immunostaining of transverse sections through the middle turn of the cochlea at P6 ( G ) , P10 ( H ) and P15 ( I ) . Mafb protein is present at high levels in SGN nuclei at P6 , then gradually translocated to the cytoplasm . Only a few SGNs maintained nuclear expression of Mafb at P10 ( arrowheads ) . Mafb is present at low levels in the cytoplasm of all SGNs at P15 . ( J ) Western blots of P3 , P6 , P10 , and P15 cochlear lysates using anti-Mafb and β-actin ( loading control ) antibodies . Mafb protein level peaks at P6 and decreases afterward . ( 12 cochleae from 6 mice per age group ) . Scale bar in F is 200 µm ( B , D–F ) or 100 µm ( C and C′ ) ; and in I is 50 µm ( G–I ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 003 A role for Mafb in differentiating SGNs has not been investigated due to the early lethality and severe inner ear malformations of null mutants ( Choo et al . , 2006 ) . To bypass these constraints , we generated a conditional Mafb allele ( Mafbflox ) ( Figure 2A , B ) . Mafbflox/flox homozygotes are fertile and viable , with normal hearing and no gross anatomical defects . However , after germline Cre-mediated recombination , homozygotes produced no Mafb protein ( Figure 2C ) and exhibited the same phenotypes as previously published null mutants ( Figure 2D–F ) , which die perinatally with defective respiratory rhythmogenesis , renal dysgenesis , and cystic inner ears ( Sadl et al . , 2002; Blanchi et al . , 2003; Choo et al . , 2006; Moriguchi et al . , 2006 ) . Hence , the Mafbflox allele provides an effective tool for conditional control of Mafb function . 10 . 7554/eLife . 01341 . 004Figure 2 . Generation and validation of floxed Mafb alleles . ( A ) A map of the Mafb locus in wild-type and floxed alleles . Two LoxP sites flank the Mafb coding region that is contained in a single exon . A neoR cassette was used for positive selection and can be removed by FLP-mediated recombination of FRT sites . B: BamHI site , S: SwaI site . cds: coding DNA sequence . Purple arrows indicate forward and reverse primers used for genotyping . Green and orange bars indicate 5D probes or 3C probes used for Southern blotting in ( B ) . Blue vertical bars demarcate the sequence used to target to the Mafb genomic locus . ( B ) Southern blotting of the 3′ ( left ) and 5′ ( right ) Mafb genomic region confirmed successful homologous recombination in two different ES cell clones . ( C ) Western blot with anti-Mafb antibodies ( top ) confirms loss of protein from E16 . 5 Mafb null mutants generated from the Mafbflox line . β-actin serves as a loading control . ( D ) Mafb null neonates take only occasional gasping breaths , show cyanosis , and die within 2 hours after birth . ( E ) E18 . 5 Mafb null mutants have dystrophic kidneys compared to control littermates . ( F ) Mafb null mutants develop a small and cystic inner ear compared to control littermates . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 004 To uncover Mafb activities in the cochlea , we used the Foxg1-Cre driver ( Hebert and McConnell , 2000 ) with the Mafbflox allele and a previously reported MafbGFP null allele ( Moriguchi et al . , 2006 ) . Foxg1 is expressed in the entire otic vesicle as early as E8 . 5 ( Hebert and McConnell , 2000 ) , long before the onset of Mafb expression . Hence , in Foxg1Cre/+;Mafbflox/GFP mutants , Mafb is never produced in SGNs ( Figure 3A , B ) . In contrast to previously described Mafb mutants ( Cordes and Barsh , 1994; Choo et al . , 2006 ) , the inner ear developed normally in Foxg1Cre/+;Mafbflox/GFP mutants ( Figure 3C , D ) . Thus , Mafb does not play a direct role in inner ear morphogenesis , consistent with previous conclusions that the inner ear phenotype is secondary to hindbrain malformation ( Choo et al . , 2006 ) . The neurite organization of SGNs was also unaffected by the loss of Mafb ( Figure 3E , F ) . Unfortunately , analysis of later functions was not possible as Foxg1Cre/+;Mafbflox/GFP mice died at birth , likely due to central apnea . 10 . 7554/eLife . 01341 . 005Figure 3 . Mafb is not required for inner ear patterning or extension of SGN peripheral projections . ( A and B ) Anti-Mafb immunostaining of transverse sections through E16 . 5 control ( A ) and Foxg1Cre/+;Mafbflox/GFP mutant ( B ) cochleae confirms Mafb is not produced in mutant SGNs ( circled area ) . ( C and D ) Light microscope images of E14 . 5 control ( C ) and Foxg1Cre/+;Mafbflox/GFP mutant ( D ) inner ears filled with paint show normal inner ear morphology in mutants . ( E and F ) Confocal stacks from E18 . 5 control ( E ) and Foxg1Cre/+;Mafbflox/GFP mutant ( F ) cochlear whole-mounts stained for Neurofilament ( NF ) to label all neuronal processes . Cochlear wiring is normal in mutant animals . ( G and H ) Anti-Mafb immunostaining of transverse sections through P5 control ( G ) and Neurog1-Cre;Mafbflox/GFP ( MafbCKO ) mutant ( H ) cochleae shows Mafb is not expressed in MafbCKO SGNs ( circled area ) . ( I ) Western blot using anti-Mafb and β-actin ( loading control ) antibodies on P5 cochlear lysate ( 20 cochleae from 10 mice per genotype ) confirms loss of protein from MafbCKO cochleae . ( J and K ) P6 cochlear whole-mounts from Neurog1-Cre;Mafbflox/+;tdTomato ( J ) and Neurog1-Cre;Mafbflox/GFP;tdTomato ( K ) mice imaged for tdTomato show afferent fibers from SGNs projecting normally in MafbCKO cochlea . ( L ) SGN density was quantified in a masked area of the 32 kHz region of control and MafbCKO cochleae at P3 and P15 . Data are presented as average SGN cell density per mask for each group . SGN cell number is similar between control and MafbCKO mice at P3 and P15 . p=0 . 51 at P3 and 0 . 97 at P5 . ns: not significant . In this and all subsequent figures , numbers in parentheses indicate the number of cochleae ( one cochlea per mouse ) used for quantification . ( M and N ) Representative light micrographs of osmium-stained plastic sections show similar cell densities of control and MafbCKO cochleae at P15 . Images corresponding to the masked region are shown . Scale bar in D is 100 µm ( A and B ) or 400 µm ( C and D ) ; in H is 50 µm ( G and H ) ; in K is 20 µm ( E , F , J , K ) ; and in N is 20 µm ( M and N ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 00510 . 7554/eLife . 01341 . 006Figure 3—figure supplement 1 . Neuorg1-Cre drives Cre-mediated recombination in SGNs but not in olivocochlear efferent neurons . Projections of confocal stacks of the cochlea , from P6 Tg ( Neurog1-Cre ) ;ROSA26tm14 ( CAG-tdTomato ) Hze;Tg ( Thy1-YFP ) triple transgenic mice imaged for tdTomato ( Red; A and A′′ ) and YFP ( green; A′ and A′′ ) . This Thy1-YFP strain ( YFP-12 ) has been demonstrated previously to show expression in the inner ear only in efferent processes ( Fu et al . , 2010 ) . Hence , expression is absent from cell bodies in the spiral ganglion but prominent in the intraganglionic spiral bundle ( IGSB ) and the inner spiral bundle ( ISB ) ( A′ , arrows in A′′ ) . In contrast , SGN cell bodies and processes are clearly labeled in Neurog1-Cre;ROSA26CAG-tdTomato cochleae , including processes in the outer spiral bundle ( OSB ) ( A and A′′ ) . tdTomato expression does not colocalize with YFP ( A′′ ) , indicating that Neurog1-Cre does not activate Cre-mediated recombination in the olivocochlear efferent system . Note the absence of expression in the ISB and IGSB , which are composed of Thy1-YFP positive efferent fibers . Scale bar is 30 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 006 To examine Mafb function later in development , we generated a second conditional knock-out strain using the Neurog1-Cre driver ( Quinones et al . , 2010 ) , which mediates recombination specifically in early SGNs ( Figure 3—figure supplement 1 ) . In Tg ( Neurog1-Cre ) ;Mafbflox/GFP mice ( hereafter referred to as MafbCKO ) , Mafb was significantly reduced at E16 . 5 ( data not shown ) and nearly undetectable by P5 ( Figure 3G–I ) . However , unlike Mafb null mutants , MafbCKO mice live through adulthood , permitting analysis of auditory circuit assembly and function . As expected , early stages of SGN development proceeded normally in MafbCKO animals , with no obvious change in the organization of SGN processes ( Figure 3J , K ) or SGN density ( Figure 3L–N ) . Taken together with results from the Foxg1-Cre driver , we conclude that Mafb is not required for the initial production or differentiation of SGNs . The absence of abnormalities in early SGN development suggested that Mafb may play a critical role at later stages , perhaps during auditory synapse development . Ultrastructural analysis revealed that mutant SGNs still contact IHCs , indicating normal targeting ( Figure 4—figure supplement 1A , B ) . We therefore asked whether these contacts were able to develop into synapses . Ribbon synapses were assessed by staining for the postsynaptic marker GluR2 , labeling a subunit of the AMPA receptor , and the presynaptic marker CtBP2 , labeling the B domain of the RIBEYE scaffolding protein ( Schmitz et al . , 2000 ) . The use of anti-GluR2 antibodies ensured that we only analyzed contacts between SGNs and IHCs , as other post-synaptic markers are also present in other types of synapses in the cochlea ( Khimich et al . , 2005; Nemzou et al . , 2006 ) . Hence , ribbon synapses were reliably identified as juxtaposed pairs of immunofluorescent puncta of postsynaptic GluR2 and presynaptic CtBP2 ( Khimich et al . , 2005; Liberman et al . , 2011 ) . Because the number of IHC ribbon synapses differs along the tonotopic axis ( Meyer et al . , 2009 ) , we were careful to compare the same region between cochleae ( around 16 kHz , see ‘Materials and Methods’ ) . The analysis of GluR2 expression revealed an obvious failure in post-synaptic differentiation in MafbCKO animals . In control animals , diffuse patches of GluR2 immunofluorescence were present near the basolateral pole of the IHCs by P6 ( Figure 4A , arrowhead ) . In contrast , GluR2 staining intensity was much lower in MafbCKO animals ( Figure 4B , K ) ( p<0 . 03 ) , indicating that GluR2 fails to accumulate postsynaptically . This loss of postsynaptic clustering of GluR2 appears to reflect an overall failure in PSD development , as ultrastructural studies revealed a clear disruption in PSD morphology at P6 . During this period of active synapse development and refinement in the mouse cochlea , IHCs contain both anchored ribbons , which are localized to the cell surface , and floating ribbons , which are initially free in the cytoplasm and can become anchored as the cochlea matures . In control mice , anchored ribbons faced a large , well-defined PSD in the apposing SGN terminal ( Figure 4C , D , Figure 4—figure supplement 1E , F ) . In contrast , MafbCKO ribbons were not paired with an obvious PSD even when in close contact with an afferent SGN terminal ( Figure 4E , F , Figure 4—figure supplement 1G , H ) . Some ribbons appeared morphologically normal ( Figure 4E ) , but others were abnormally small ( Figure 4F ) or misshapen ( Figure 4—figure supplement 1H ) . 10 . 7554/eLife . 01341 . 007Figure 4 . Loss of Mafb interferes with PSD development and reduces synapse number . ( A–B′ ) Confocal stacks from P6 control ( A and A′ ) and MafbCKO ( B and B′ ) cochlear whole-mounts double stained for GluR2 ( green ) and CtBP2 ( red ) . IHC nuclei are also weakly immunopositive for CtBP2 . Yellow dotted lines ( A and B ) outline IHCs , as determined by Myo7A staining in another channel ( not shown ) . GluR2 immunofluorescence is present near the basal pole of control ( arrowhead ) but not MafbCKO IHCs . The intensity of CtBP2 immunofluorescence is decreased in MafbCKO IHCs at P6 when compared to controls . ( C–F ) Electron micrographs of IHC ribbon synapses in the middle turn of cochleae from P6 control ( C and D ) and MafbCKO ( E and F ) mice . MafbCKO mice lack a delineated PSD in their SGN afferent terminals ( arrowheads ) . ( G–H′ ) Confocal stacks from P15 control ( G and G′ ) and MafbCKO ( H and H′ ) cochlear whole-mounts double stained for GluR2 ( green ) and CtBP2 ( red ) . The number of GluR2 and CtBP2 puncta is decreased in P15 MafbCKO mice . ( I and J ) 3D reconstructions of presynaptic ribbons ( red ) and postsynaptic aggregations of GluR2 ( yellow ) from P15 control ( I ) and MafbCKO ( J ) cochleae derived from confocal stacks using Amira image-processing software . Many MafbCKO ribbons lack corresponding GluR2 puncta ( J ) . ( K ) Quantification of GluR2 immunofluorescence intensity at P6 ( left axis ) and GluR2 puncta number per IHC at P15 and adult ( right axis ) in control and MafbCKO cochleae . GluR2 immunofluorescence intensity of P6 MafbCKO cochleae is expressed as a percentage of control intensity . ( L ) Quantification of ribbon number per IHC in control and MafbCKO cochleae at P6 , P15 and adult . Ribbon number and GluR2 intensity/number of MafbCKO cochleae are significantly decreased when compared to controls at all stages . : p<0 . 05 , *: p≤0 . 001 . Scale bar in H′ is 5 µm ( A , A′ , B , B′ , G , G′ , H , H′ ) ; in F is 200 nm ( C–F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 00710 . 7554/eLife . 01341 . 008Figure 4—figure supplement 1 . MafbCKO mutants show normal targeting of afferent terminals but impaired synapse development . ( A and B ) Electron micrographs of an IHC and its associated afferent terminals ( A ) in the middle turn of cochleae from P6 MafbCKO mice . The box in ( A ) outlines the region magnified in ( B ) . MafbCKO afferent fibers target normally to IHCs . Efferent fibers ( E ) still synapse with IHCs at this stage . a: afferent terminals; e: efferent terminals . ( C–D′ ) Projections of confocal stacks from P3 control ( C and C′ ) and MafbCKO ( D and D′ ) cochlear whole-mounts double stained for GluR2 ( green ) and CtBP2 ( red ) . IHC nuclei are also weakly immunopositive for CtBP2 . The intensity of CtBP2 immunofluorescence is decreased in MafbCKO IHCs at P3 when compared to controls . ( E–H ) Electron micrographs of IHC ribbon synapses in the middle turn of cochleae from P6 control ( E and F ) and MafbCKO ( G and H ) mice . In MafbCKO synapses , the presynaptic ribbons are often abnormally shaped and are not paired with a clear postsynaptic density . a: afferent terminals . ( I–J′ ) Projections of confocal stacks from adult control ( I and I′ ) and MafbCKO ( J and J′ ) cochlear whole-mounts double stained for GluR2 ( green ) and CtBP2 ( red ) . The number of GluR2 and CtBP2 puncta is decreased in adult MafbCKO mice . Scale bar in B is 1 . 5 µm ( A ) or 500 nm ( B ) ; in H is 100 nm ( E–H ) ; in J′ is 5 µm ( C , C′ , D , D′ , I , I′ , J , J′ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 00810 . 7554/eLife . 01341 . 009Figure 4—figure supplement 2 . Transcription of GluR2 and PSD95 is not significantly changed in MafbCKO SGNs . ( A–D ) In situ hybridization revealed no obvious difference in the expression of GluR2 ( A and B ) or PSD95 ( C and D ) in control ( A and C ) vs MafbCKO ( B and D ) SGNs at P6 . Orange boxes indicate the region shown in figure insets , corresponding to the masked areas used for semi-quantitative analysis . ( E ) Semi-quantitative analysis of in situ hybridization was performed to compare the mean signal intensity in a masked area of SGNs in control and MafbCKO cochleae . Control and MafbCKO SGNs express similar levels of GluR2 and PSD95 . p=0 . 85 for GluR2 and 0 . 77 for PSD95 . ns: not significant . Scale bar in D is 100 µm ( A–D ) or 25 µm ( insets ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 00910 . 7554/eLife . 01341 . 010Figure 4—figure supplement 3 . The olivocochlear efferent system is normal in MafbCKO mice . ( A–D′′ ) Projections of confocal stacks from P15 control ( A and C–C′′ ) and MafbCKO ( B and D–D′′ ) cochlear whole-mounts triple stained for choline acetyltransferase ( ChAT , red ) , synaptophysin ( yellow ) , and vesicular acetylcholine transporter ( VAT , green ) to label the efferent innervation in cochleae . The box in ( A ) indicates the region magnified in ( C ) . Olivocochlear efferent innervation is normal in MafbCKO cochleae . Scale bar in D′′ is 60 µm ( A and B ) or 20 µm ( C–D′′ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 010 Closer examination confirmed that defective PSD differentiation secondarily affected presynaptic development . In P3 control cochleae , ribbons were dispersed broadly in the IHC cytoplasm , with many CtBP2-positive puncta not yet localized to the basolateral surface ( Figure 4—figure supplement 1C′ ) . Over the next several days , RIBEYE assembled into larger complexes that stained more brightly and were confined to the basal pole of the IHC ( Figure 4A′ ) . In contrast , anti-CtBP2 immunofluorescence intensity was noticeably reduced in MafbCKO mutants at P6 ( Figure 4B′ ) , consistent with observation of small ribbons in electron micrographs ( Figure 4F ) . In addition , ribbon number was significantly decreased ( Figure 4L , Table 1 ) . Although GluR2 immunofluorescence was not bright enough to assess PSD differentiation before P6 ( Figure 4—figure supplement 1C , D ) , CtBP2 intensity was also diminished at P3 ( Figure 4—figure supplement 1D′ ) , suggesting that SGN post-synaptic differentiation is impaired from the earliest stages . Thus , Mafb mutant SGNs fail to elaborate the specialized PSDs that are characteristic of auditory ribbon synapses , apparently leading to defects in the development and anchoring of presynaptic ribbons in the IHCs . 10 . 7554/eLife . 01341 . 011Table 1 . Quantitative analysis of IHC ribbon synapsesDOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 011GluR2 puncta number per IHC ( mouse number ) p-valueCtBP2 puncta number per IHC ( mouse number ) p-valueP6CE controln/an/a23 . 0 ± 0 . 6 ( 6 ) 0 . 0750 ( ns ) MafbCEn/a24 . 5 ± 0 . 4 ( 6 ) CKO controln/an/a23 . 2 ± 0 . 7 ( 3 ) 0 . 0187MafbCKOn/a18 . 4 ± 1 . 0 ( 3 ) P15CE control16 . 6 ± 0 . 9 ( 3 ) 0 . 6916 ( ns ) 17 . 9 ± 0 . 5 ( 3 ) 0 . 3757 ( ns ) MafbCE17 . 1 ± 0 . 2 ( 3 ) 18 . 6 ± 0 . 4 ( 3 ) CKO control16 . 9 ± 0 . 3 ( 6 ) 0 . 000118 . 0 ± 0 . 3 ( 10 ) 7 . 6 × 10−5MafbCKO8 . 6 ± 1 . 0 ( 6 ) 12 . 5 ± 1 . 0 ( 10 ) Gata3CKO control15 . 3 ± 0 . 4 ( 7 ) 6 . 4 × 10−917 . 1 ± 0 . 6 ( 7 ) 1 . 4 × 10−7Gata3CKO5 . 6 ± 0 . 5 ( 7 ) 0 . 00038 . 2 ± 0 . 5 ( 7 ) 0 . 0001Gata3CKO;MafbCE10 . 0 ± 0 . 7 ( 7 ) 12 . 3 ± 0 . 6 ( 7 ) AdultCKO control15 . 2 ± 0 . 7 ( 6 ) 2 . 7 × 10−617 . 6 ± 0 . 5 ( 8 ) 0 . 0014MafbCKO6 . 8 ± 0 . 4 ( 6 ) 12 . 0 ± 1 . 3 ( 8 ) Means ± SEMs are shown . p-values were obtained by Student’s t-test . Quantification and statistical results of GluR2 and CtBP2 puncta from all mouse strains used in this study . These reciprocal defects in early pre- and post-synaptic development resulted in a severe disruption in the mature pattern of synaptic connectivity , with an ∼50% reduction in the number of GluR2 puncta and an ∼30% reduction in ribbon number at P15 ( Figure 4G–H′ , K , L , Table 1 ) and in adults ( Figure 4K , L , Figure 4—figure supplement 1I-J′ , Table 1 ) . The loss of GluR2 and CtBP2 puncta was observed along the entire length of the mutant cochlea ( data not shown ) . Additionally , close examination of individual pre- and post-synaptic puncta showed that many CtBP2-positive ribbons were not paired with a corresponding GluR2 spot ( Figure 4I , J ) . This stands in contrast to control animals , where an individual ribbon is reliably juxtaposed to a single GluR2-positive terminal by this stage ( Khimich et al . , 2005; Nemzou et al . , 2006; Sendin et al . , 2007 ) . Hence , the overall loss of synapses seems to originate with a defect in the PSD . This phenotype is not due to a general decrease in the transcription of PSD components , as the expression levels of GluR2 and PSD95 transcripts were similar between control and Mafb mutant SGNs ( Figure 4—figure supplement 2 ) . Although olivocochlear efferent neurons also express Mafb ( NR Druckenbrod and LVG , unpublished observation ) and are thought to influence cochlear development ( Guinan , 2010 ) , the loss of afferent synapses is unlikely to be secondary to abnormalities in the efferent system since Neurog1-Cre is not active in olivocochlear neurons ( Figure 3—figure supplement 1 ) . In addition , the efferent innervation of the MafbCKO cochlea is normal ( Figure 4—figure supplement 3 ) . Thus , we conclude that Mafb acts in SGN afferents to specify post-synaptic differentiation of ribbon synapses . Ribbon synapses transmit all acoustic information from IHCs to SGNs , playing a critical role in shaping the final perception of sound . To determine how auditory function is altered when synapse development is disrupted , we measured Auditory Brainstem Responses ( ABRs ) in MafbCKO animals . ABRs reflect the electrical responses of neurons in the cochlea and auditory brainstem to sound stimuli . Acoustic sensitivity is determined by identifying the lowest sound pressure level ( SPL ) ( the threshold ) that is able to elicit a response . ABRs were recorded in response to pure tone bursts of 5 . 6 , 8 , 11 . 3 , 16 , 22 . 6 , 32 , and 45 . 2 kHz . Control mice produced characteristic ABR waveforms in response to all stimuli , with thresholds ranging between 15 and 45 dB SPL depending on the frequency of the stimulus . In contrast , MafbCKO mice showed dampened ABR waveforms and elevated auditory thresholds , with thresholds increased ∼10 dB SPL across stimuli of all frequencies ( Figure 5A , B , D , Figure 5—figure supplement 1 ) . Thresholds of MafbCKO mice were significantly higher for all frequencies measured except the lowest and highest frequencies , since sensitivity is already reduced in these regions of control cochleae ( Figure 5B ) ( p values: 5 . 6 kHz=0 . 125 , 8 kHz=0 . 044 , 11 . 3 kHz=0 . 043 , 16 kHz=0 . 031 , 22 . 6 kHz=0 . 038 , 32 kHz=0 . 047 , 45 . 2 kHz=0 . 140; n = 16 controls and 16 mutants ) . We also investigated whether this impaired auditory response involved an OHC defect by recording distortion product otoacoustic emissions ( DPOAEs ) . DPOAE levels were not significantly different in MafbCKO vs control mice , regardless of the frequency and intensity of the stimulus used ( Figure 5C ) , indicating that hair cell function is largely normal . Hence , the ABR threshold elevation is mainly caused by the reduced activation of SGNs by sound stimuli . 10 . 7554/eLife . 01341 . 012Figure 5 . Mafb is required for normal auditory function . ( A ) Representative ABR recordings from a P41 control and MafbCKO littermate exposed to a 16 kHz pure tone stimulus at intensities ranging from 20 to 80 dB . Arrows indicate the ABR threshold . The ABR response is diminished in the MafbCKO mutant compared to its littermate control . ( B and C ) Plots of threshold values from recordings of ABRs ( B ) and DPOAEs ( C ) performed on 5- to 7-week-old control ( blue line ) and MafbCKO ( red line ) mice . Auditory responses were assessed at 7 frequencies ( 5 . 6 kHz , 8 kHz , 11 . 3 kHz , 16 kHz , 22 . 6 kHz , 32 kHz , and 45 . 2 kHz ) , and across a range of sound pressure levels from 10 to 80 decibels . : p<0 . 05 . n = 16 mice for each group . MafbCKO mice show elevated ABR but normal DPOAE thresholds when compared to controls . ( D ) 16 kHz ABR waveforms of 16 control ( blue line ) and 16 MafbCKO ( red line ) mice were averaged and overlaid . Roman numerals mark the peaks of the ABR waves . Wave I is delayed and diminished in MafbCKO mutants . ( E ) Average Wave I latencies for control ( blue , n = 16 ) and MafbCKO ( red , n = 16 ) littermates in response to a 16 kHz pure tone stimulus . MafbCKO responses are significantly delayed at 80 , 70 , and 60 dB SPL . : p<0 . 05 . ( F ) Average Wave I amplitudes for control ( blue , n = 16 ) and MafbCKO ( red , n = 16 ) littermates in response to a 16 kHz stimulus . The MafbCKO responses are significantly decreased at 80 , 70 , 60 , and 50 dB SPL . : p<0 . 005 and : p<0 . 05 . ( G ) Average Wave I amplitudes for control ( blue , n = 16 ) and MafbCKO ( red , n = 16 ) littermates at all 7 frequencies of the stimulus . Stimuli were presented at a sound pressure level of 80 dB . MafbCKO responses are significantly decreased at all frequencies . *: p<0 . 01 and : p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 01210 . 7554/eLife . 01341 . 013Figure 5—figure supplement 1 . MafbCKO mutants show diminished Wave I of ABRs in response to all frequencies . ( A–F ) ABR waveforms of 5 . 6 kHz ( A ) , 8 kHz ( B ) , 11 . 3 kHz ( C ) , 22 . 6 kHz ( D ) , 32 kHz ( E ) , and 45 . 2 kHz ( F ) from 16 control ( blue line ) and 16 MafbCKO ( red line ) mice were averaged and overlaid . Wave I is delayed and reduced in MafbCKO mutants in response to a wide range of sound frequencies . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 013 We further characterized the change in SGN responsiveness by analyzing Wave I of the ABR waveforms , which represents the summed activity of SGNs evoked by sound . The strength and speed of the electrical response were evaluated by measuring the amplitude and latency of wave I . In MafbCKO mice , Wave I was delayed , with increased latency in mutants exposed to a 16 kHz pure tone ( Figure 5E ) . More strikingly , the amplitude was significantly diminished in response to stimuli across frequencies and intensities ( Figure 5D , F , G , Figure 5—figure supplement 1 ) . For instance , the response to a 16 kHz , 80 dB stimulus was 35% smaller in mutants than in controls . Hence , loss of Mafb prevents the development of normal acoustic responsiveness , with effects along the length of the cochlea . ABR responses reflect the overall activation of SGNs , which is determined both by the strength of the synaptic signals and by the intrinsic response properties of the SGNs . We wondered whether the diminished ABR responses in mutants could be explained entirely by the loss of synapses or whether other aspects of SGN responsiveness were also altered . Electrophysiological recordings offer a highly sensitive read-out of SGN function that can reveal differences in membrane potential and firing properties . Accordingly , we performed whole-cell recordings of SGNs in organotypic explants , acutely dissected at P3 . Analysis at P3 allows us to focus on SGN properties independent of any secondary effects from loss of synaptic transmission in older animals . SGNs express several voltage-dependent ionic currents that contribute to their membrane properties ( Santos-Sacchi , 1993; Adamson et al . , 2002 ) . We recorded voltage-dependent currents in voltage-clamp mode to assess membrane properties of SGNs . Voltage steps evoked prominent inward and outward currents , indicative of voltage-dependent sodium and potassium currents ( Figure 6A ) . Maximum peak outward current amplitudes were not significantly different between MafbCKO mutants and littermate controls ( MafbCKO: 4 . 1 ± 1 . 3 nA , n = 11; control: 4 . 4 ± 0 . 8 nA , n = 10 ) and there was no significant difference in the current–voltage relationships ( Figure 6B , lower left panel ) . We also analyzed fast activating and fast inactivating inward currents , characteristic of voltage-dependent sodium currents ( INa ) ( Figure 6A ) . Again , no significant difference was found in INa ( Figure 6B , lower right panel ) ( MafbCKO:−1 . 2 ± 1 . 5 nA , n = 7; controls: −2 . 1 ± 0 . 9 nA , n = 5 ) . 10 . 7554/eLife . 01341 . 014Figure 6 . Mafb is not required for development of SGN firing properties . ( A ) Representative families of voltage-dependent outward K+ currents ( top ) and Na+ currents ( bottom ) recorded from SGNs in organotypic explants harvested from control , MafbCE , and MafbCKO animals at P3 . K+ currents were evoked by voltage steps from −124 mV to 104 mV in 10 mV increments from a holding potential of −84 mV . Na+ currents were evoked by 100 ms prepulse to −104 mV from a holding potential of −84 mV , followed by voltage steps from −64 mV to 24 mV in 10 mV increments . Some current traces were removed for clarity . ( B ) Mean ( ±1 SD ) potassium current–voltage ( I–V ) ( left ) and mean ( ±1 SD ) sodium I–V relations ( right ) . Number of SGNs for each genotype is indicated in the legend . ( C ) Representative current-clamp data recorded from P3 control , MafbCE , and MafbCKO SGNs in response to 200-ms current injections in 10 pA increments from rest . Depolarizing current injections elicit action potentials in littermate control , MafbCE , and MafbCKO SGNs and all exhibit prominent voltage sag in response to hyperpolarizing current injections . ( D ) Summary graphs of resting membrane potential ( left ) , action potential ( AP ) threshold ( middle ) , and AP latency ( right ) measured from MafbCE ( dark cyan bar ) MafbCKO ( orange bar ) and their littermate controls , denoted as CE control ( white bar ) and CKO control ( gray bar ) in the graphs , respectively . Resting membrane potential was measured as membrane voltage at I = 0 ( no current injection ) . AP threshold was measured as the minimum current required to evoke an AP . AP latency was quantified as the time from the current step to the AP peak for a 50-pA current step . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 014 To assess SGN firing properties , we recorded from MafbCKO SGNs in current-clamp mode . Depolarizing current injections above 30 pA evoked action potentials ( APs ) in MafbCKO SGNs ( Figure 6C ) , suggesting that the AP firing machinery was not compromised . Quantitative analyses of AP parameters , such as resting membrane potential , AP threshold , and AP latency , indicated no significant difference between MafbCKO and control littermates ( Figure 6D ) . In summary , our results indicate that intrinsic membrane and firing properties were unaltered in postnatal MafbCKO SGNs . We therefore propose that the elevated ABR threshold and reduced eighth nerve synchronization ( Figure 5 ) is a direct consequence of the IHC-SGN synapse defect ( Figure 4 ) . Moreover , the emergence of normal SGN firing properties seems to be independent of Mafb . The absence of electrophysiological defects in mutant SGNs indicated that Mafb may play an unusually specific role directing postsynaptic differentiation . Consistent with this idea , in wild-type mice , Mafb protein is first detected just prior to the beginning of synapse development ( Figure 1 ) , as if the onset of Mafb expression triggers formation of the PSD in SGN terminals . To test this idea , we created a strain of Mafb conditional expressor mice ( MafbCE ) that produce exogenous Mafb upon Neurog1-Cre-mediated recombination [Tg ( Neurog1-Cre ) ;Gt ( ROSA ) 26SorCAG-lsl-Mafb] ( Figure 7A ) . In these mice , Mafb protein was present in SGNs by E12 . 5 , at least 3 days earlier than the expression of endogenous Mafb protein in controls ( Figure 7B–C′ ) . Overexpression of Mafb had no gross effect on cochlear wiring ( Figure 7—figure supplement 1A , B ) . However , AMPA receptors clustered earlier in SGN terminals , with multiple patches of GluR2 staining covering the entire basal pole of the IHCs ( Figure 7D , E ) . Quantification confirmed a dramatic increase of GluR2 staining intensity in MafbCE afferent terminals ( Figure 7F ) ( p=0 . 033 ) , with similar effects observed for a second PSD marker , PSD95 ( Figure 7—figure supplement 1C , D ) . Hence , increased expression of Mafb enhanced post-synaptic differentiation in SGNs . 10 . 7554/eLife . 01341 . 015Figure 7 . Overexpression of Mafb accelerates synapse development . ( A ) MafbCE mice have a Cre-dependent CAG promoter-driven Mafb-ires-tdTomato-pA cassette inserted into the Rosa26 locus . ( B–C′ ) Mafb ( green ) and TuJ1 ( red , to mark spiral ganglia indicated by arrows ) immunostaining on transverse sections of E12 . 5 control ( B and B′ ) and MafbCE ( C and C′ ) heads . The cochlear duct ( asterisks ) , hindbrain ( hb ) and saccule ( s ) are visualized with DAPI counterstain ( blue ) . Mafb is present in E12 . 5 MafbCE SGNs but not control SGNs . ( D–E′ ) P6 control ( D and D′ ) and MafbCE ( E and E′ ) cochleae double stained for GluR2 ( green ) and CtBP2 ( red ) . Yellow dotted lines ( D and E ) outline IHCs , as determined by Myo7A staining in another channel ( not shown ) . GluR2 immunostaining shows multiple intense fluorescence patches in the basal region of MafbCE IHCs compared to a single patch in controls . ( F ) Quantification of P6 GluR2 immunofluorescence intensity ( left axis ) and P15 GluR2 puncta number per IHC ( right axis ) in control and MafbCE cochleae . GluR2 immunofluorescence intensity of P6 MafbCE cochleae is expressed as a percentage of control intensity . GluR2 intensity of P6 MafbCE cochleae is significantly increased compared to controls . P15 GluR2 puncta number is similar between control and MafbCE mice . ( G ) Quantification shows similar ribbon number per IHC in control and MafbCE cochlea at P6 and P15 . : p<0 . 05 . ns: not significant . ( H ) Illustration of how the 3D distance between two ribbons was measured . ( I ) Scatter plot of the distance to nearest neighbor of P6 and P15 control ( blue ) and MafbCE ( red ) ribbons . Middle bar represents mean . : p<0 . 001 . ns: not significant . n=407 control and 428 MafbCE ribbons in 18 IHCs from 6 mice at P6 , 217 control and 215 MafbCE ribbons in 12 IHCs from three mice at P15 . P6 MafbCE ribbons are significantly closer to their nearest neighbors compared to controls , reflecting an overall increased confinement of ribbons to the basal pole of the IHC . Scale bar in C′ is 50 µm ( B–C′ ) ; in E′ is 5 µm ( D–E′ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 01510 . 7554/eLife . 01341 . 016Figure 7—figure supplement 1 . MafbCE mice show precocious development of IHC ribbon synapses . ( A and B ) Projections of confocal stacks from P6 control ( A ) and MafbCE ( B ) cochlear whole-mounts stained for Neurofilament ( NF ) to label neuronal processes . MafbCE cochleae show no gross abnormalities in neurite organization . ( C–D′ ) Projections of confocal stacks from P6 control ( C and C′ ) and MafbCE ( D and D′ ) cochlear whole-mounts double stained for PSD-95 ( green ) and CtBP2 ( red ) . MafbCE afferent terminals show more PSD95 immunofluorescence at the basal region of IHCs when compared to control littermates . ( E–F′ ) Projections of confocal stacks from P15 control ( E and E′ ) and MafbCE ( F and F′ ) cochlear whole-mounts double stained for GluR2 ( green ) and CtBP2 ( red ) . The number of GluR2 and CtBP2 puncta is similar in control and MafbCE cochleae at P15 . Scale bar in B is 20 µm ( A and B ) ; in F′ is 5 µm ( C–F′ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 016 Since abnormal development of the PSD was accompanied by a change in presynaptic differentiation in MafbCKO mice , we asked whether these changes in the development of the PSD would have any effect on the maturation of ribbons in IHCs . At P6 , ribbons are gradually relocating from the cytoplasm to the membrane , where individual ribbons are anchored opposite an afferent terminal . In MafbCE mice , this process occurred earlier and CtBP2-positive spots were noticeably more confined to the basal pole of the IHCs ( Figure 7D′ , E′ ) . To assess the distribution of ribbons independent of the angle at which each IHC was imaged , we measured the distance from each ribbon to its nearest neighbor; this distance is predicted to be shorter when ribbons are more tightly localized ( Figure 7H ) . Indeed , the nearest-neighbor distance of ribbons in P6 MafbCE mice was significantly shorter ( 1 . 83 ± 0 . 03 µm , n = 428 ribbons ) than in controls ( 2 . 06 ± 0 . 04 µm , n = 407 ribbons ) ( p<0 . 0001 ) ( Figure 7I ) . Hence , overexpression of Mafb caused synapses to develop earlier , as evidenced by increased accumulation of PSD components in SGN terminals and precocious basolateral localization of pre-synaptic ribbons in the hair cells . Notably , this acceleration in synapse development did not cause any increase in the number of ribbons ( Figure 7D′ , E′ , G , Figure 7—figure supplement 1E′ , F′ , and Table 1 ) or GluR2-positive puncta ( Figure 7F , Figure 7—figure supplement 1E , F , and Table 1 ) at P15 . Ribbon distribution was also unchanged in P15 MafbCE mice compared to controls ( p=0 . 63 ) ( Figure 7I ) . Thus , the pruning and refinement of synapses appear to be controlled by an independent pathway that is unaffected by the overexpression of Mafb . Electrophysiological studies further confirmed that early and increased expression of Mafb does not seem to have adverse effects on SGN maturation . MafbCE SGNs exhibited normal firing properties ( Figure 6 ) , with no change in the peak outward current amplitudes ( MafbCE: 4 . 8 ± 1 . 2 nA n = 5; control: 5 . 7 ± 1 . 0 nA , n = 6 ) ( Figure 6B , upper left panel ) , current–voltage relationships , or inward currents ( MafbCE: −1 . 6 ± 0 . 5 nA , n = 5; control: −1 . 8 ± 0 . 4 n = 6 ) ( Figure 6B , upper right panel ) . All AP parameters were also normal in MafbCE SGNs ( Figure 6C , D ) . Unfortunately , the majority ( >80% ) of MafbCE mice die at ∼3 weeks of age due to unknown causes , so it was not possible to test whether hearing develops normally in these animals . Nevertheless , our findings suggest that expression of Mafb is sufficient to initiate synapse development , without disrupting synaptic pruning or other aspects of SGN function . Our studies show that Mafb promotes a key feature of the SGN identity , namely the formation of the uniquely large post-synaptic terminals that are necessary for the sense of hearing . This represents a late event in the execution of neuronal identity programs that are set in motion and coordinated by the early acting transcription factor Gata3 ( Appler et al . , 2013 ) . One of Gata3’s function appears to be the activation of additional SGN-specific transcription factors , including Mafb . In the immune system , Gata3 cooperates with the large Maf factor c-Maf to control the terminal differentiation of Th2 helper cells ( Ho et al . , 1998 ) . We therefore hypothesized that Mafb similarly mediates Gata3’s effects on SGN-specific features of development . We tested this hypothesis by investigating Mafb expression and function in Gata3 conditional knock-outs ( Gata3CKO ) . Loss of Gata3 from SGNs using the Bhlhe22-Cre driver severely disrupts cochlear innervation ( Appler et al . , 2013 ) . In Gata3CKO mutants , Mafb protein levels were strongly reduced ( Figure 8B′ ) . Consistent with this loss of Mafb , the number of GluR2 and CtBP2 puncta was significantly decreased in P15 Gata3CKO mutants ( Figure 8E , E′ , G and Table 1 ) . Synaptic loss in Gata3CKO ( Figure 8G ) was more severe than what occurs in MafbCKO animals ( Figure 4K , L ) , which fits with Gata3’s earlier and broader effects on SGN neurite outgrowth , axon guidance , and survival ( Appler et al . , 2013; Duncan and Fritzsch , 2013; Luo et al . , 2013 ) . This makes it difficult to conclude whether the synaptic loss is due to the loss of Mafb or to other aspects of the Gata3 mutant phenotype . 10 . 7554/eLife . 01341 . 017Figure 8 . Mafb acts downstream of Gata3 to control synapse development . ( A-C’ ) Transverse sections of E16 . 5 control ( A and A′ ) , Gata3CKO ( B and B′ ) and Gata3CKO;MafbCE ( C and C′ ) cochleae double immunostained with antibodies against Gata3 ( A , B , C ) and Mafb ( A′ , B′ , C′ ) . Asterisks indicate the cochlear duct . Gata3 and Mafb are normally co-expressed in SGNs at E16 . 5 ( A and A′ ) . In Gata3CKO mutants ( B and B′ ) , Gata3 protein is severely reduced in SGNs ( B , circled ) but is maintained in the cochlear duct ( B , arrow ) . Mafb expression is also diminished ( B′ ) . In contrast , Mafb expression is restored in Gata3CKO;MafbCE SGNs ( C′ ) , despite the loss of Gata3 ( C ) . ( D-F’ ) Confocal stacks from P15 control ( D and D′ ) , Gata3CKO ( E and E′ ) and Gata3CKO;MafbCE ( F and F′ ) cochlear whole-mounts double stained for GluR2 ( green ) and CtBP2 ( red ) . The number of GluR2 and CtBP2 puncta is decreased in Gata3CKO mice ( E and E’ ) compared to controls ( D and D′ ) . Pre- and post-synaptic puncta are partially recovered in Gata3CKO;MafbCE mice ( F and F′ ) . ( G ) Quantification of GluR2 puncta and ribbon number per IHC in control , Gata3CKO and Gata3CKO;MafbCE cochleae at P15 . The number of GluR2 and CtBP2 puncta is decreased in Gata3CKO cochleae but partially restored in Gata3CKO;MafbCE cochleae when compared to controls . *: p<0 . 0005 . Scale bar in C′ is 50 µm ( A–C′ ) ; in F′ is 5 µm ( D–F′ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 01710 . 7554/eLife . 01341 . 018Figure 8—figure supplement 1 . Restoring Mafb expression in Neurog1-Cre;Gata3CKO SGNs partially rescues the number of IHC ribbon synapses . ( A–C′ ) Projections of confocal stacks from P15 littermate control ( Gata3flox/tauLacZ ) ( A and A′ ) , Neurog1-Cre;Gata3CKO ( Neurog1-Cre; Gata3flox/tauLacZ ) ( B and B′ ) and Neurog1-Cre;Gata3CKO;MafbCE [Neurog1-Cre;Gata3flox/tauLacZ;Gt ( ROSA ) 26SorCAG-lsl-Mafb] ( C and C′ ) cochlear whole-mounts double stained for GluR2 ( green ) and CtBP2 ( red ) . The number of GluR2 and CtBP2 puncta is decreased in Gata3CKO mice ( B and B′ ) but partially restored in Gata3CKO;MafbCE mice ( C and C′ ) . Scale bar is 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01341 . 018 To clarify the relationship between Gata3 and Mafb in SGNs , we asked whether restoration of Mafb is sufficient to rescue synapse development in the Gata3CKO cochlea . We used the MafbCE line to provide exogenous Mafb into the Gata3CKO background . In Bhlhe22Cre/+;Gata3flox/tauLacZ;ROSA26CAG-lsl-Mafb animals ( i . e . , Gata3CKO;MafbCE ) , Mafb protein was present in SGN nuclei , despite the absence of Gata3 ( Figure 8C , C′ ) . Strikingly , this restoration of Mafb was sufficient to improve the synaptic defect seen in Gata3CKO mutants ( compare Figure 8F , F′ to Figure 8E , E′ ) , with an ∼80% increase in GluR2 puncta and an ∼50% increase in ribbon number in Gata3CKO;MafbCE mice when compared to Gata3CKO littermates ( Figure 8G , Table 1 ) . Synapses could also be rescued by resupplying Mafb to Gata3CKO mice made using the Neurog1-Cre line ( Figure 8—figure supplement 1 ) , indicating that the effect is SGN-autonomous . Taken together , these findings indicate that Gata3 and Mafb are key players in a transcriptional cascade that guides SGN development and ensures the emergence of the cell-type specific features that are critical for the sense of hearing . In this study , we showed that the emergence of cell-type specific features of synapse development can be controlled by late acting lineage-specific transcription factors , with Mafb acting downstream of Gata3 to ensure formation of the specialized ribbon synapses that mediate the sense of hearing . Our results indicate that Mafb plays a highly specific role , with dramatic effects on post-synaptic differentiation in the absence of any other obvious changes in SGN function . Indeed , Mafb was capable of restoring synapses even in the Gata3 mutant background , where SGN development is severely disrupted . These findings establish Mafb as a potent regulator of SGN terminal differentiation that functions within a broader Gata3 transcriptional network , offering a potent molecular entry point for designing new treatments for age-related and noise-induced hearing loss . Efforts to elucidate the molecular basis of neuronal differentiation have highlighted the importance of transcriptional networks that progressively specify neuronal fates and direct the proper execution of these fates during terminal differentiation ( Hobert , 2011 ) . The earliest acting transcription factors are master regulators , which induce multiple parallel changes in gene expression that reinforce cell fate and promote differentiation . The endpoint of the cascade is the activation of so-called terminal selectors , which induce batteries of genes essential for that neuron’s functional properties . Terminal selectors work combinatorially , with multiple transcription factors cooperating to regulate sets of terminal effector genes that are necessary for each neuron’s mature identity , such as neurotransmitters , channels , and synaptic adhesion molecules ( Hobert , 2011 ) . For example , the ETS-factor Ast-1 activates expression of genes essential for dopamine synthesis and transport in dopaminergic neurons ( Hobert , 2011 ) , acting together with Dlx and Pbx-family transcription factors ( Doitsidou et al . , 2013 ) . Identifying analogous terminal selectors in vertebrates has been more challenging , due to the increased complexity of neuronal phenotypes , as well as the difficulty in defining discrete regulatory elements . Hence , the only established examples are for easily recognized features of mature neurons , such as the control of serotonin production by the transcription factor Pet1 ( Liu et al . , 2010 ) . Our results place Mafb at the bottom of a transcriptional network that controls the differentiation of neurons dedicated to the sense of hearing . At the top of this hierarchy are Neurogenin1 and NeuroD1 that control the generation of all inner ear neurons . Subsequently , Gata3 expression is selectively enriched in SGNs ( Lawoko-Kerali et al . , 2004; Lu et al . , 2011 ) and acts both to promote the auditory fate and to coordinate multiple aspects of SGN differentiation ( Appler et al . , 2013; Duncan and Fritzsch , 2013; Luo et al . , 2013 ) . Our findings add Mafb as a critical player downstream to Gata3: Gata3 is required for Mafb expression and Mafb is sufficient to rescue synapses in Gata3CKO mutants . Hence , Mafb provides a new example of a transcription factor that can direct post-synaptic differentiation , apparently in a cell-type specific manner . Mafb’s restricted effects contrast with those of Gata3 , which is necessary for multiple aspects of SGN development ( Appler et al . , 2013; Duncan and Fritzsch , 2013; Luo et al . , 2013 ) . This raises the important question of what other transcription factors might act downstream of Gata3 . One attractive possibility is that combinatorial activity with other large Maf factors contributes to SGN differentiation . MafA and c-Maf are also highly expressed in SGNs ( Lu et al . , 2011 and W-MY and LVG , unpublished observation ) , and both have been shown to play important roles in the terminal differentiation of neurons in the dorsal root ganglia ( Bourane et al . , 2009; Wende et al . , 2012 ) . Thus , there may be a Maf code for SGN terminal differentiation , with each factor directing a different feature required for mature function . Mafb’s potent effects on the late stages of SGN differentiation raise the intriguing possibility that Mafb may serve as a terminal selector in the vertebrate nervous system . Terminal selectors are typically recognized by two consistent characteristics: the ability to activate cell-type appropriate gene expression during terminal differentiation and an ongoing role in the maintenance of critical genes in the mature neuron ( Hobert , 2011 ) . Consistent with this definition , Mafb is not expressed until late in differentiation , just a day before functional synapses can first be detected ( Marrs and Spirou , 2012 ) , suggesting that Mafb may activate expression of terminal effectors that drive synapse formation . Additionally , Mafb is both necessary and sufficient for the acquisition of a key feature of mature SGN function , that is the ability to receive information from IHCs . Whether Mafb plays an ongoing role in SGNs remains to be determined , but its persistent expression in adults fulfills another important criterion . Notably , the output of any one terminal selector is strongly influenced by cellular context , so individual terminal selectors can also be expressed in other subtypes of neurons and not all neurons of the same subtype depend on the same set of terminal selectors ( Doitsidou et al . , 2013 ) . Similarly , the absence of Mafb from vestibular ganglion neurons , which also form ribbon synapses , suggests that different sets of transcription factors may act in SGNs and VGNs to control post-synaptic differentiation . Additional support for a terminal selector function for Mafb awaits identification of direct target genes , studies that may ultimately reveal effects on other auditory-specific features of SGN function . Our findings raise the possibility that Mafb’s effects might extend to other neuronal populations . Indeed , Mafb is expressed by a handful of other neuronal subtypes and could play a similar role in the emergence of their cell-type specific properties , including but not limited to synaptic differentiation . For instance , in addition to being expressed in SGNs , Mafb is also expressed in cochlear nucleus bushy cells ( Saul et al . , 2008 ) , which receive synaptic input from SGNs . Similarly , Mafb defines subsets of cortical interneurons ( Wang et al . , 2010 ) and a subpopulation of rhythmogenic neurons within the pre-Bötzinger complex ( Blanchi et al . , 2003 ) . Currently , the specific function of Mafb in other neuronal populations remains obscure , due both to the lethality of null mutants and the lack of tools for detecting subtle synaptic defects in other regions of the nervous system . With the availability of the new conditional alleles reported here , it will be important to investigate Mafb’s activities in other regions of the nervous system in the future . In addition , our results suggest that similar late acting transcription factors may play analogous roles in other subtypes of neurons , functioning after synaptic targeting to ensure that each neuron elaborates the correct type of synapse necessary for proper circuit function . Mafb’s ability to direct auditory synaptogenesis even in pathological situations opens up the exciting possibility of harnessing this power to repair the damaged cochlea . One of the earliest effects of acoustic overexposure is the loss of synapses followed by withdrawal of SGN terminals . The SGNs subsequently undergo a long slow degeneration , ultimately resulting in significant hearing loss long after the original injury ( Kujawa and Liberman , 2009 ) . Our results suggest that stimulation of Mafb might be sufficient to encourage damaged neurons to re-establish connections with IHCs . Moreover , the specificity of Mafb’s effects implies that synapses could be restored without inadvertently interfering with other aspects of SGN function . Importantly , we find that Mafb is still present in the adult cochlea ( data not shown ) , largely sequestered in the cytoplasm . This observation emphasizes the need for further characterization of the post-translational modifications that affect the localization and transactivation activity of Mafb . Ultimately , drugs that enhance translocation of Mafb into the nucleus might offer a new treatment for deafness . Similarly , understanding how the Gata3–Mafb cascade is deployed will be critical for the development of stem cell-based replacement therapies . Indeed , recent studies have shown that functional dopaminergic and spinal motor neurons can be directly generated from somatic cells by ectopic expression of lineage specific transcription factors ( Caiazzo et al . , 2011; Son et al . , 2011 ) , thereby avoiding the need for surgical implantation of stem cells . With the discovery of transcription factors such as Neurogenin1 , NeuroD1 , Gata3 , and Mafb , such an approach may eventually be possible in the cochlea . A linker sequence containing a loxP sequence and a SwaI restriction site for Southern blot genotyping was inserted into the middle of a 5 . 98 kB Mafb genomic fragment ( corresponding to the sequence of 2260–8240 bp in Figure 2A ) to create the Mafb 5′ targeting arm . Another linker sequence containing a BamHI restriction site for Southern blot genotyping was ligated to the 5′ end of a 1 . 07 kB Mafb genomic fragment ( corresponding to the sequence of 8241–9312 bp in Figure 2A ) to create the Mafb 3′ targeting arm . These arms were cloned into the MfeI/NheI sites and XhoI/NotI sites of a modified pBlueScript II SK+ plasmid containing a loxP sequence and an frt-flanked Pgk-Neo cassette ( 4600C conditional targeting vector ) to generate the floxed Mafb targeting construct . This construct was electroporated into J1 ES cells ( derived from 129S4/SvJae strain ) and selected under G418 for 1 week . 395 ES cell clones were screened for correct recombination by Southern blot using external 5′ and 3′ probes ( 5D and 3C , 1212–2105 and 9930–11 , 019 bp in Figure 2A ) , and two successfully recombined clones were injected into blastocysts to make Mafbflox-neo mice . These mice were crossed to a global FLPe driver [Tg ( ACTFLPe ) , a gift from Dr Susan Dymecki , Harvard Medical School] ( Rodriguez et al . , 2000 ) to remove the Pgk-Neo cassette and generate Mafbflox mice . A 50 bp sequence upstream of Mafb translational initiation site and the Mafb coding sequence were cloned into the AscI site of a modified CTV plasmid ( Addgene , Cambridge , MA ) in which tdTomato replaces eGFP ( Ctd plasmid ) . The Ctd–Mafb targeting construct was electroporated into J1 ES cells and clones were screened for correct recombination by PCR using two primer pairs amplified the 5′ and 3′ homology regions of the ROSA26 locus ( forward-CGCCTA AAGAAGAGGCTGTG and reverse-GGGCGTACTTGGCATATGAT for a 1472 bp 5′ homology region; forward-GCCTCGACTGTG CCTTCTAG and reverse-CCATTCTCAGTGGCTCAACA for a 4886 bp 3′ homology region ) . Two successfully recombined clones were injected into blastocysts to make Gt ( ROSA ) 26SorCAG-lsl-Mafb ( MafbCE ) mice . Mafbflox mice were PCR genotyped using Mafb-loxP forward ( CCTCAACGGCTTCGGGGCTCCTC ) and reverse ( CGCTCTCCGACTCTTTGGCTCTA ) primers with 5% dimethyl sulfoxide in the reaction ( purple arrows in Figure 2A indicate the primer locations; wt , 264 bp; Mafbflox , 310 bp ) . MafbCE mice were PCR genotyped using tdTomato forward ( CTGTTCCTGTACGGCATGG ) and reverse ( GGCATTAAAGCAGCGTATCC ) primers and control forward ( AAGGGAGCTGCAGTGGAGTA ) and reverse ( CCGAAAATCTGTGGGAAGTC ) primers ( wt , 297 bp; MafbCE , 196 bp ) . The following mouse strains were used and PCR genotyped as described previously: a null GFP knock-in allele of Mafb ( MafbGFP ) ( Moriguchi et al . , 2006 ) , two floxed Gata3 alleles: Gata3fl ( Zhu et al . , 2004 ) and Gata3Ho ( Pai et al . , 2004 ) , a null tauLacZ knock-in allele of Gata3 ( Gata3tauLacZ ) ( van Doorninck et al . , 1999 ) , β-actin-Cre [Tg ( ACTB-cre ) ] ( the Jackson Laboratory , Bar Harbor , ME ) , Foxg1-Cre ( Hebert and McConnell , 2000 ) , Neurog1-Cre ( Quinones et al . , 2010 ) , Bhlhe22-Cre ( Ross et al . , 2010 ) , Thy1-YFP ( YFP-12 ) ( Fu et al . , 2010 ) , and the tdTomato reporter [Ai14; Gt ( ROSA ) 26Sortm14 ( CAG-tdTomato ) Hze] ( Allen Institute for Brain Science , Seattle , WA ) . For MafbCKO mice , all mutants contained a null and a floxed allele ( Neurog1-Cre;Mafbflox/GFP ) and all controls were littermates heterozygous for Mafb ( Neurog1-Cre;Mafbflox/+ , Neurog1-Cre;MafbGFP/+ or Mafbflox/GFP ) except for two Mafbflox/flox animals used as controls for the ABRs due to the lack of heterozygous littermates . For Gata3CKO mice , all mutants contained a null and a floxed allele ( Bhlhe22-Cre;Gata3flox/tauLacZ ) , and all controls were heterozygous littermates ( Gata3flox/tauLacZ ) . For staging , noon on the day that the vaginal plug was observed was counted as embryonic day 0 . 5 ( E0 . 5 ) . For postnatal collection , the day of birth was considered P0 . Embryos and pups of both sexes were used . Animals were maintained and handled in compliance with a protocol approved by the Institutional Animal Care and Use Committee at Harvard Medical School . E13 . 5 embryo heads were fixed in 4% paraformaldehyde ( PFA ) in PBS overnight at 4°C . P6 mice were perfused with 4% PFA in PBS , and the inner ears were dissected out and post-fixed in 4% PFA in PBS for 2 hr at 4°C . All tissue was sucrose protected and then frozen in a Neg50 frozen section medium ( Richard-Allan Scientific , Kalamazoo , MI ) . 12-µm thick cryosections from pairs of control and MafbCKO animals were collected on the same Superfrost Plus slide ( VWR International , Radnor , PA ) . Non-radioactive in situ hybridization was performed as described ( Abraira et al . , 2008 ) . A detailed protocol is available at http://goodrich . med . harvard . edu/resources/resources_protocol . htm . The template DNA used for riboprobe synthesis was PCR amplified from cDNA clones ( Mafb: BC038256; GluR2: BC048248; PSD95: BC014807 ) using the following primers: Mafb , CCGGAATTCCAGTCCGACTGAACCGAAGACC and CCCAAGCTTCTCAGGAGAGGAGGGGCTGTCG ( GenitoUrinary Development Molecular Anatomy Project ) ; GluR2 , TGTATCCTTCATCACACCAAGC and GTCATCACTTGGACAGCATCAT ( Allen Brain Atlas ) ; PSD95 , CGATTACCACTTTGTCTCCTCC and AAGAAAGGCTAGGGTACGAAGG ( Allen Brain Atlas ) . For the semi-quantitative analysis of in situ hybridization , images were converted to 8-bit gray-scale . A mask corresponding to a region of 50 × 50 µm was superimposed over SGNs in the middle turn of the cochlea . The mean signal intensity of the region with the background removed was calculated using ImageJ ( National Institutes of Health , Bethesda , MD ) . E16 . 5 whole embryos or postnatal cochleae were homogenized in 25 mM Tris , pH 7 . 5 containing 95 mM NaCl , 5 mM EDTA , 2% SDS and 1 mM Pefabloc ( Roche , Penzberg , Germany ) . Western blot analysis was performed using standard protocols . Rabbit anti-Mafb ( 1:1000 , NB600-266; Novus Biologicals , Littleton , CO ) and mouse anti-β-actin ( 1:5000 , ab8226; Abcam , Cambridge , England ) primary antibodies were used . Paintfilling of E14 . 5 mouse embryos was performed by injecting White All Purpose Correction Fluid ( Sanford Corporation , Oak Brook , IL ) into the cochlea as described previously ( Abraira et al . , 2008 ) . Araldite sections were prepared and SGNs were counted as described previously ( Kujawa and Liberman , 2009 ) with two minor modifications . Sections were cut at 20 µm instead of 40-µm thickness , and a mask corresponding to a region of 60 × 60 µm was superimposed on the P3 image instead of the 90 × 60 µm mask used with the P15 ganglion due to the smaller size of the P3 ganglion . Embryonic heads were fixed directly in 4% PFA in PBS overnight at 4°C . Postnatal mice were anesthetized with ketamine and xylazine and then intravascularly perfused with 4% PFA in PBS . Inner ears were dissected and then the round and oval windows were opened to permit flushing of 4% PFA through the scalae , followed by post-fixation in 4% PFA overnight at 4°C . For mice older than P7 , inner ears were decalcified in 120 mM EDTA in PBS at room temperature for 2 days . For tissue sections , tissues were stepped through 10 , 20 , and 30% sucrose in PBS , embedded in NEG 50 ( Richard-Allan Scientific ) , and cryosectioned at 14 µm . The sections were blocked 1 hr at room temperature in a solution containing 5% normal donkey serum and 0 . 3% Triton X-100 in PBS . The sections were then incubated overnight at 4°C in primary antibodies diluted in the blocking solution . Alexa Fluor-conjugated secondary antibodies ( Life Technologies , Carlsbad , CA ) were used for signal detection . Rabbit anti-Mafb ( 1:500 , NB600-266; Novus Biologicals ) , mouse anti-TuJ1 ( 1:2000 , MMS-435P; Covance , Dedham , MA ) , and goat anti-Gata3 ( 1:100 , AF2605; R&D Systems , Minneapolis , MN ) primary antibodies were used . For cochlear whole-mount staining , fixed cochleae were dissected into three pieces corresponding to 2 . 8–11 . 3 kHz ( apex ) , 11 . 3–32 kHz ( middle ) , and 32 kHz–64 kHz ( base ) regions in the cochlea ( Muller et al . , 2005 ) . The microdissected pieces were blocked in PBS with 1%Triton X-100 and 5% normal donkey serum for 1 hr at room temperature and then incubated in primary antibodies diluted in blocking solution at 37°C for 20 hr . Alexa Fluor-conjugated secondary antibodies were used for signal detection . Chicken anti-NF-H ( 1:3000 , AB5539; Millipore , Billerica , MA ) , rabbit anti-TuJ1 ( 1:1000 , MRB-435P; Covance ) , goat anti-Choline Acetyltransferase ( 1:200 , AB144P; Millipore ) , mouse anti-synaptophysin 1 ( 1:200 , 101 011 , Synaptic Systems , Goettingen , Germany ) , rabbit anti-Vesicular Acetylcholine Transporter ( 1:1000 , ab68984 , Abcam ) , and mouse anti-PSD95 ( 1:500 , MABN68; Millipore ) primary antibodies were used . For CtBP2 and GluR2 double staining , cochlear pieces were permeabilized in 30% sucrose for 20 min and stained as described previously ( Furman et al . , 2013 ) . Rabbit anti-Myo7A ( 1:200 , 25-6790; Proteus BioSciences , Ramona , CA ) , mouse IgG1 anti-CtBP2 ( 1:200 , 612044; BD Transduction Laboratories , San Jose , CA ) and mouse IgG2a anti-GluR2 ( 1:1000 , MAB397; Millipore ) primary antibodies were used . Cochlear lengths were measured for each microdissected piece and a cochlear frequency map was determined as described previously ( Muller et al . , 2005 ) . Confocal z-stacks from the selected cochlear region were obtained on an Olympus FluoView FV1000 ( Tokyo , Japan ) using a 40X ( NA:1 . 30 ) or a 60X ( NA:1 . 40 ) oil-immersion objectives . A 512 × 512 binary image was acquired at optimal step size in the Z axis ( 0 . 55 µm for 40X and 0 . 45 µm for 60X ) . For confocal imaging of CtBP2 and GluR2 staining , z-stacks from the 16 kHz region were obtained on an Olympus FluoView FV1000 using a 60X ( NA:1 . 40 ) oil-immersion objective and 2 . 5X digital zoom . A 512 × 512 ( pixel size = 0 . 165 µm in x and y ) binary image was acquired at 0 . 45-µm step size in the Z axis , resulting in an image containing approximately 10 IHCs . Care was taken to minimize pixel saturation in each image stack and to acquire control and mutant images using the same laser power and high voltage value of photomultiplier . For quantification of GluR2 immunofluorescence of P6 cochleae , a maximum projection of the stacks containing GluR2 immunofluorescence in a region of 10 IHCs was obtained using Olympus Fluoview software . A mask corresponding to a region of 80 × 20 µm was superimposed on the image to cover the entire synaptic poles of 10 IHCs . Fluorescence intensity was calculated minus the background using ImageJ ( National Institutes of Health ) . MafbCKO or MafbCE fluorescence intensity is expressed as a percentage of value from intensity of littermate controls . For 3D reconstruction and counting of GluR2 and CtBP2 puncta , image stacks were exported to Amira imaging-processing software ( Visualization Sciences Group , Burlington , MA ) , where three-dimensional reconstruction of IHCs , synaptic ribbons and GluR2 puncta were produced . The number of ribbons and GluR2-positive puncta was divided by the total number of IHCs , as revealed by Myo7A and the presence of CtBP2-positive nuclei ( including fractional estimates , when necessary , at each end of the image stack ) . For measurement of the three-dimensional distance to nearest neighbor , x , y , and z coordinates of all ribbons were exported from Amira to Excel ( Microsoft , Redmond , WA ) . Each ribbon was assigned manually to an IHC according to the x , y , and z coordinates of a 3D rendering of the Myo7A ‘isosurfaces’ . The inter-ribbon distance within an IHC was then calculated using each ribbon’s x , y , and z coordinates . Animals were anesthetized and then intravascularly perfused with 0 . 1M sodium cacodylate buffer pH 7 . 4 containing 2 . 5% glutaraldehyde and 1 . 5% paraformaldehyde . Inner ears were removed and the round and oval windows were opened to permit flushing of fix through the scalae followed by postfixation overnight at 4°C . The cochlear coils were dissected , separated into three pieces and postfixed in 2% osmium tetroxide at room temperature for 1 hr . The cochlear pieces were dehydrated and embedded in Epon resin . 80-nm ultrathin sections from the center region of the middle turn were counter-stained with uranyl acetate and lead citrate , and observed with a JEOL 1200EX electron microscope ( Akishima , Japan ) . For measuring of auditory brainstem recordings ( ABRs ) and distortion product otoacoustic emissions ( DPOAEs ) , recordings were performed on the right ears of 5- to 7-week-old anesthetized mice in a soundproof chamber maintained at 32°C as described previously ( Buran et al . , 2010 ) . Each MafbCKO mutants was paired with a control littermate during each round of recordings ( 16 mice in each group ) . Average ABR waveforms were plotted using a script written in MATLAB ( MathWorks , Natick , MA ) by Dr Ann E Hickox in the laboratory of Dr Charles Liberman ( Massachusetts Eye and Ear Infirmary , Boston , MA ) . Data were analyzed with Excel ( Microsoft ) and GraphPad Prism ( GraphPad , La Jolla , CA ) . Student’s t-test was used to determine if two sets of data are significantly different from each other . Difference in means were considered significant if p<0 . 05 . The results are expressed as means ± SEMs unless otherwise noted . SGNs were acutely dissected from P3 and P4 mice . Following rapid decapitation , cochleae were excised from the temporal bone , and bathed in sterile MEM with glutamax ( Gibco # 41 , 090 , Grand Island , NY ) supplemented with 10 mM HEPES ( Sigma , St . Louis , MO ) and 25 mg ampicillin ( Sigma ) , pH 7 . 4 . SGN explants were divided into halves along the apical-basal axis and mounted flat on glass cover slips . SGN organotypic explants were incubated at 37°C in a humidified incubator ( 5% CO2 ) for 1–2 hr , followed by acute electrophysiological studies . All SGN explants were used within 4–5 hr of dissection . The whole-cell , tight-seal technique was used in voltage- and current-clamp modes to record from identified SGN cell bodies . SGN organotypic explants were placed into a custom-made recording chamber and viewed under Zeiss Axioskop FS upright microscope ( Oberkochen , Germany ) equipped with a 63× water-immersion lens and differential interface contrast optics . SGN explants were bathed in standard external solution that contained ( in mM ) : 137 NaCl , 0 . 7 NaH2PO4 , 5 . 8 KCl , 1 . 3 CaCl2 , 0 . 9 MgCl2 , 5 . 6 D-glucose , 10 HEPES , amino acids ( 1:50 , 11130; Gibco ) , vitamins ( 1:100 , 11120; Gibco ) ; pH 7 . 4 ( NaOH ) , 303 mOsmol/kg . Recording pipettes ( 3–5 MΩ ) were pulled from R-6 soda lime capillaries ( King Precision Glass , Claremont , CA ) , using a two-stage vertical pipette puller ( PC-10; Narishige , Tokyo , Japan ) . Recording pipettes were filled with standard internal solution that contained ( in mM ) : 135 KCl , 2 . 5 MgCl2 , 2 . 5 K2-ATP , 5 . 0 HEPES , 5 . 0 EGTA , 0 . 1 CaCl2; pH 7 . 4 ( KOH ) , 283 mOsmol/kg . Electrophysiological data from SGNs were recorded using an Axopatch 200B amplifier ( Molecular Devices , Palo Alto , CA ) . Signals were filtered at 1 kHz with a low pass Bessel filter , and digitized at ≥20 kHz using 12-bit acquisition system Digidata 1332 ( Axon Instruments , Union City , CA ) and pClamp 9 . 0 software ( Molecular Devices ) . Currents were recorded immediately after the cell membrane was broken through at giga-ohm ( GΩ ) seal and the series resistance ( Rs ) and membrane capacitance ( Cm ) were corrected . Compensated residual Rs was below 7 MΩ on average . All electrophysiological recordings were performed at room temperature ( 22-24°C ) . Offline data analysis was performed using OriginPro 7 . 5 ( Origin Lab , Northampton , MA ) and reported as mean ± SD . Liquid junction potentials ( −4 mV ) were adjusted offline for all membrane potentials .	Different types of neurons in the nervous system communicate with each other through different types of synapses . In the auditory system , for example , it is essential for the timing of signals to be preserved as they are sent from the ear to the brain , and this places special demands on the synapses in this part of the nervous system . In particular , the ribbon synapses that are found between the inner hair cells of the ear , which convert sound waves into neural signals , and the neurons of the spiral ganglion in the cochlea , which carry information about the frequency , intensity and timing of sounds to the brain , can transmit signals with remarkable fidelity . Little is known about the mechanisms by which synapses become specialized for particular functions . Previous work has suggested that a protein called Gata3 is important for the development of the neurons and synapses in the spiral ganglion , including ribbon synapses . Gata3 is a transcription factor that controls the expression of a wide range of genes that are involved in the auditory systems , including genes that are expressed as other transcription factors . Yu et al . used transgenic mice to explore what happened when one of these transcription factors , Mafb , was missing from neurons in the spiral ganglion . The results showed that ribbon synapses did not form when Mafb was absent , which meant that they were unable to respond normally to sounds . Yu et al . also studied mice in which Gata3 was absent: normally Mafb would not be present in these mice , but when genetic techniques were used to force the expression of the gene for Mafb , ribbon synapses were formed . As well as revealing a molecular pathway by which synapses become specialized for rapid and accurate transmission of auditory information , these findings might lead to new approaches to treating hearing loss in humans .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2013	A Gata3–Mafb transcriptional network directs post-synaptic differentiation in synapses specialized for hearing
When people anticipate uncertain future outcomes , they often prefer to know their fate in advance . Inspired by an idea in behavioral economics that the anticipation of rewards is itself attractive , we hypothesized that this preference of advance information arises because reward prediction errors carried by such information can boost the level of anticipation . We designed new empirical behavioral studies to test this proposal , and confirmed that subjects preferred advance reward information more strongly when they had to wait for rewards for a longer time . We formulated our proposal in a reinforcement-learning model , and we showed that our model could account for a wide range of existing neuronal and behavioral data , without appealing to ambiguous notions such as an explicit value for information . We suggest that such boosted anticipation significantly drives risk-seeking behaviors , most pertinently in gambling . When people anticipate possible future outcomes , they often prefer their fate to be revealed in advance by a predictive cue , even when this cue does not influence outcome contingency . This is usually called ‘observing’ , or ‘information-seeking’ , behavior . Recently , Bromberg-Martin and Hikosaka ( Bromberg-Martin and Hikosaka , 2009; 2011 ) reported an influential series of studies into the neural basis of the observing behavior of macaque monkeys . They tested subjects’ preferences between three targets that were followed by cues that resolved uncertainty about the volume of an upcoming reward ( small or large ) to different degrees . Subjects strongly preferred a ‘100% info’ target , which was followed by uncertainty-resolving , definitive , cues , over a ‘50% info’ target , which was followed either by definitive cues or by a totally ambiguous cue; and preferred the latter target over a ‘0% info target’ , which was always followed by an entirely ambiguous cue . Neurons in lateral habenula responded differently to the same definitive , reward-predicting , cue depending on the target that had previously been chosen ( 100% or 50% info ) . The authors concluded that these neurons index what they called ‘information prediction errors’ along with conventional reward prediction errors ( Schultz et al . , 1997; Matsumoto and Hikosaka , 2007 ) , and that biological agents ascribe intrinsic value to information . A yet more striking finding is that animals appear willing to sacrifice reward for taking advance information . This is known for birds ( Zentall , 2016; McDevitt et al . , 2016 ) , monkeys ( Blanchard et al . , 2015a ) , and humans ( Eliaz and Schotter , 2010; Molet et al . , 2012 ) . Extensive studies on birds ( mostly pigeons ) showed that animals prefer a less rewarding target that is immediately followed by uncertainty-resolving cues , over a more rewarding target without such cues ( e . g . 20% chance over 50% chance of reward ( Stagner and Zentall , 2010; Vasconcelos et al . , 2015 ) , 50% chance over 75% chance of reward ( Gipson et al . , 2009 ) , 50% chance over certain reward ( Spetch et al . , 1990; McDevitt et al . , 1997; Pisklak et al . , 2015 ) . Crucially however , the pigeons only show this preference when the delay , TDelay , between the choice and the reward is sufficiently long . Another salient experimental observation in Spetch et al . ( 1990 ) is that after choosing a less rewarding , 50% chance , target , some of the pigeons were also seen to peck enthusiastically during the delay following the cue informing them that reward would arrive . By contrast , they were comparatively quiescent during the delay after choosing the certain reward target followed by a similar cue ( Spetch et al . , 1990 ) . It remains a challenge to account for these data on the preference of advance information . The delay-dependent preference reward predictive cues shown by Spetch et al . ( 1990 ) cannot depend on conventional Shannon information , since this is normally independent of delay . Furthermore , targets associated with less Shannon information can be more attractive ( Roper and Zentall , 1999; Zentall , 2016 ) . Equally , evidence from the activity of lateral habenula neurons ( Bromberg-Martin and Hikosaka , 2011 ) provides no support for the recent suggestion that disengagement caused by uninformative cues could cause the seeking of informative cues ( Beierholm and Dayan , 2010 ) . Here , we offer a new explanation of observing and information-seeking behavior that accounts for the effects of delays reported in the pigeon experiments and the effects of changing the probability of reward . We follow an established notion called the utility of anticipation ( Loewenstein , 1987; Berns et al . , 2006; Story et al . , 2013 ) . These investigators have shown that subjects consider the delay to a future reward as itself being appetitive ( think , for instance , of yourself waiting for an upcoming vacation trip ) , associated with a positive utility of anticipation . This is often referred to as savouring ( the anticipation of negative outcomes is called dread ) , and coexists with a more conventional effect of delay , namely temporal discounting ( Loewenstein , 1987; Loewenstein and Prelec , 1993; Schweighofer et al . , 2006; Kable et al . , 2010 ) . In this framework , we hypothesize that the level of anticipation can be boosted by the ( temporal difference ) prediction errors caused by predictive cues that resolve reward uncertainty . Pigeons’ vigorous pecking following those cues is a sign of the boost . That is , the definitive reward cue following a partial target evokes a positive ( temporal difference ) prediction error – the difference between the expected ( partial chance ) and actual outcome ( reward for sure ) is positive ( Schultz et al . , 1997 ) . We suggest that the impact of this is to increase savouring . By contrast , the certain target elicits no such prediction error and so , within this account , will not increase savouring . Put simply , our model posits that unexpected news of upcoming pleasant ( or unpleasant ) outcomes boosts the savouring ( or dread ) associated with such outcomes . The hypothesis of boosting yields a parsimonious explanation for observing and information-seeking , consistent with all existing neural and behavioral data , including a range of seemingly paradoxical findings ( Gipson et al . , 2009; Spetch et al . , 1990; Stagner and Zentall , 2010; Bromberg-Martin and Hikosaka , 2009; 2011; Zentall , 2016 ) . Here we also conducted human behavioral studies to test the delay TDelay dependence of observing and information seeking , which has so far only been subject to limited tests in animal studies ( Spetch et al . , 1990 ) . Our model relies on the established economic theory called the utility of anticipation ( Loewenstein , 1987 ) , which proposes that the anticipation of future reward has a positive subjective value which is added to the actual value of the reward . Our formulation follows prior characterizations ( Loewenstein , 1987; Berns et al . , 2006; Story et al . , 2013 ) . Consider the case in which a subject takes an action and receives a pre-reward cue at t=0 , and then a reward R at t=TDelay ( Figure 1A ) . The anticipation of the reward is worth a⁢ ( t ) =R⁢e-ν⁢ ( TDelay-t ) at time t ( Figure 1A- ( ii ) ) , where ν governs the rate of growth of this factor . A small ν means that the anticipation grows gradually in time , while a large ν means that the anticipation increases steeply near the delivery of rewards . 10 . 7554/eLife . 13747 . 003Figure 1 . The model . ( A ) The value of the cue is determined by ( ii ) the anticipation of upcoming reward in addition to ( i ) the reward itself . The two are ( iii ) linearly combined and discounted; with the weight of anticipation being ( iv ) boosted by the RPE associated with the predicting cue . ( B ) The contribution of different time points to the value of predicting cue . The horizontal axis shows the time of reward delivery . The vertical axis shows the contribution of different time points to the value of the predicting cue . ( C ) The total value of the predicting cue , which integrates the contribution along the vertical axis of panel ( B ) , shows an inverted U-shape . DOI: http://dx . doi . org/10 . 7554/eLife . 13747 . 003 The value of the pre-reward cue is determined by what follows the cue , which , under conventional temporal difference ( TD ) learning , would have been the reward itself ( Figure 1A- ( i ) ) , discounted in time with a rate γ . Here , however , in addition to the reward itself , we have the anticipation of the reward that takes place continuously over time . Thus the total value of the predictive cue , Q is the sum of the discounted reward ( blue bar in Figure 1A- ( iii ) ) , and the temporally discounted anticipation , where the latter is integrated over time from the presentation of the cue up to reward delivery ( red area in Figure 1A- ( iii ) ) . Note that the integration of anticipation suggests that the total amount of anticipation contributing to the value of the cue can increase as TDelay is increased . This can be seen in Figure 1B , in which the color code indicates the contribution of the temporally discounted anticipation at time 0<t<TDelay to a predictive cue . The horizontal axis indicates different delay conditions TDelay , and the vertical axis shows the different time points t between the cue ( t=0 ) and the time of reward delivery ( t=TDelay ) . Figure 1C shows the total values for different delay length TDelay , which are the integrals of contributions over the vertical axis in Figure 1B . As seen , the total value usually takes the maximum value at a finite TDelay , which is larger than the value R of the reward itself ( Figure 1C ) . While the actual peak is determined by the competition between the anticipation ν and discounting γ , this inverted U-shape was confirmed previously for the case of savoring , using hypothetical questionnaire studies ( Loewenstein , 1987 ) . ( This inverted U-shape holds in general in the model , unless the relative weight of anticipation η is zero , or growth and discounting are too steep relative to each other ν≪γ or ν≫γ . ) Previously , the total value of cue Q has been expressed as a sum of the value of anticipation and the reward itself: ( 1 ) Q=η V[Anticipation]+V[Reward] with the relative weight of anticipation η being treated as a constant . Here we hypothesized that reward prediction errors ( RPE ) δp⁢e in response to the predictive cue ( Schultz et al . , 1997 ) can boost anticipation ( Figure 1A- ( iv ) ) . Our proposal was inspired by findings of a dramatically enhanced excitement that follows predictive cues that resolve reward uncertainty appetitively ( Spetch et al . , 1990 ) , which will generate positive RPEs . A simple form of boosting arises from the relationship ( 2 ) η=η0+c\|δpe\| where η0 specifies base anticipation , and c determines the gain . That anticipation is boosted by the absolute value of RPE turns out to be important in applying our model to comparatively unpleasant outcomes , as confirmed in our own experiment . Note that anticipation can only be boosted by the RPE that precedes it – in this case arising from the predictive cue . Any RPE associated with the delivery of reward would have no anticipatory signal within the trial that it could boost . We ignore any subsidiary anticipation that could cross trial boundaries . Figure 2A , B , C illustrates the design and results of the experiment mentioned above in which macaque monkeys exhibit observing , or information-seeking , behavior ( Bromberg-Martin and Hikosaka , 2011 ) . Briefly , there were three targets: 1 ) a 100% info target that was always followed by a cue whose shapes indicated the upcoming reward size ( big or small ) ; 2 ) a 0% info target that was always followed by a random cue whose shapes conveyed no information about reward size; and 3 ) a 50% info target that was followed half the time by informative cues and half the time by random cues ( Figure 2A ) . Figure 2B shows the strong preference the subjects exhibited for the 100% over the 50% , and the 50% over the 0% info targets . Figure 2C shows the difference in activity of lateral habenula neurons at the time of the predictive cues depending on the preceding choice of target . 10 . 7554/eLife . 13747 . 004Figure 2 . Our model accounts for the behavioral and neural findings in Bromberg-Martin and Hikosaka ( 2011 ) . ( A ) The task in ( Bromberg-Martin and Hikosaka , 2011 ) . On each trial , monkeys viewed a fixation point , used a saccadic eye movement to choose a colored visual target , viewed a visual cue , and received a big or small water reward . The three potential targets led to informative cues with 100% , 50% or 0% probability . ( Bromberg-Martin and Hikosaka , 2011; reproduced with permission ) ( B ) Monkeys strongly preferred to choose the target that led to a higher probability of viewing informative cues ( Bromberg-Martin and Hikosaka , 2011; reproduced with permission ) . ( C ) The activity of lateral habenula neurons at the predicting cues following the 100% target ( predictable ) were different from the case where the cues followed the 50% target ( unpredicted ) ( Bromberg-Martin and Hikosaka , 2011; reproduced with permission ) . The mean difference in firing rate between unpredicted and predictable cues are shown in case of small-reward and big-reward ( the error bars indicate SEM . ) . ( D ) Our model predicts the preference for more informative targets . ( E , F ) Our model’s RPE , which includes the anticipation of rewards , can account for the neural activity . Note the activity of the lateral habenula neurons is negatively correlated with RPE . DOI: http://dx . doi . org/10 . 7554/eLife . 13747 . 00410 . 7554/eLife . 13747 . 005Figure 2—figure supplement 1 . Our model can capture the preference of info targets with a wide range of parameters . The color map ( bottom ) shows the squared errors of our model’s prediction with respect to the choice preference of one of the monkeys ( Monkey Z ) reported in Bromberg-Martin and Hikosaka ( 2011 ) , while the top two panels show model’s predictions the corresponding parameters . The parameters are fixed , not optimized , as RBig=0 . 88 , RSmall=0 . 04 , ν=0 . 5sec−1 , TDelay=2 . 25sec , γ=0 . 1sec−1 , σ=0 . 08DOI: http://dx . doi . org/10 . 7554/eLife . 13747 . 00510 . 7554/eLife . 13747 . 006Figure 2—figure supplement 2 . RPE-boosting of anticipation is necessary to capture the choice preference of monkeys reported in Bromberg-Martin and Hikosaka ( 2011 ) . The baseline anticipation is the same for three targets with different levels of advance reward information . Hence the model exhibits no preference . DOI: http://dx . doi . org/10 . 7554/eLife . 13747 . 006 Figure 2D shows that our model captured subjects’ preferences for each of the targets; and Figure 2E , F shows that it also accounted for the different sizes of RPEs to the same reward predictive cues when they followed different info targets ( noting that the responses of lateral habenula neurons are negatively correlated with the size of reward prediction errors; Matsumoto and Hikosaka , ( 2007 ) ; Bromberg-Martin and Hikosaka , ( 2011 ) ) . The difference in RPE sizes arose from the different values of the targets , which itself arose from the different magnitudes of anticipation associated with the cues . We found that these results held across a wide range of parameter settings ( Figure 2—figure supplement 1 ) . Note that RPE-boosting is necessary to capture the data , as the simple baseline anticipation model predicted the same level of anticipation following each target , leading to no preference between the targets ( Figure 2—figure supplement 2 ) . We also note that our model further predicts that the magnitude of RPE to a reward predictive cue can be larger than the magnitude of RPE to the reward itself following random cues . This is because the predictive cues include the value of anticipation . Our model also accounted for puzzling irrational gambling behaviors that have been reported in many experiments ( Spetch et al . , 1990; Gipson et al . , 2009; Eliaz and Schotter , 2010; Vasconcelos et al . , 2015 ) . These include a perplexingly greater preference for a target offering 50%chance of reward over either a target offering 75% chance of reward , or a certain target offering 100% chance of reward , at least when the delays between the predicting cues and rewards are long ( Figure 3A–C ) ( Spetch et al . , 1990; Gipson et al . , 2009 ) . 10 . 7554/eLife . 13747 . 007Figure 3 . Our model accounts for a wide range of seemingly paradoxical findings of observing and information-seeking . ( A ) Abstraction of the pigeon tasks reported in Spetch et al . ( 1990 ) ; Gipson et al . 2009 ) . On each trial , subjects chose either of two colored targets ( Red or Blue in this example ) . Given Red , cue S+ or S0 was presented , each with probability 0 . 5; was followed by a reward after time TDelay , while S0 was not followed by reward . Given Blue , a cue S* was presented , and reward possibly followed after the fixed time delay TDelay with probability pB , or otherwise nothing . In Spetch et al . ( 1990 ) , pB=1 , and in Gipson et al . ( 2009 ) pB=0 . 75 . ( B ) Results with pB=1 in Spetch et al . ( 1990 ) . Animals showed an increased preference for the less rewarding target ( Red ) as delay time TDelay was increased . The results of four animals are shown . ( Adapted from Spetch et al . , 1990 ) ( C ) Results with pB=0 . 75 in Gipson et al . ( 2009 ) . Most animals preferred the informative but less rewarding target ( Red ) . ( Adapted from Gipson et al . , 2009 ) ( D ) Our model predicted changes in the values of cues when pB=1 , accounting for ( B ) . Thanks to the contribution of anticipation of rewards , both values first increase as the delay increased . Even though choosing Red provides fewer rewards , the prediction error boosts anticipation and hence the value of Red ( solid red line ) , which eventually exceeds the value of Blue ( solid blue line ) , given a suitably long delay . Without boosting , this does not happen ( dotted red line ) . At the delay gets longer still , the values decay and the preference is reversed due to discounting . This second preference reversal is our model’s novel prediction . Note that x-axis is unit-less and scaled by γ . ( E ) The changes in the values of Red and Blue targets across different probability conditions pB . Our model predicted the reversal of preference across different probability conditions of pB . The dotted red line represents when the target values were equal . We set parameters as ν/γ=0 . 5 , R=1 , η0/γ=3 , c/γ=3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13747 . 00710 . 7554/eLife . 13747 . 008Figure 3—figure supplement 1 . Related to Figure 3 . The changes in the values of Red and Blue targets across different probability conditions pB when we assume a different function form for η:η=η0+c1tanh ( c2δpe ) ( the task described in Figure 3 ) . The model’s behavior does not change qualitatively compared to Figure 3D;E . We set parameters as ν/γ=0 . 5 , R=1 , η0/γ=3 , c1/γ=1 , c2=3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13747 . 00810 . 7554/eLife . 13747 . 009Figure 3—figure supplement 2 . Related to Figure 3 . ( A ) The task reported in Stagner and Zentall ( 2010 ) ; Vasconcelos et al . ( 2015 ) ; Zentall ( 2016 ) . Subjects ( birds ) had to choose either the 100% info target ( shown as Red here ) associated with 20% chance of reward , or the 0% info target ( shown as Blue here ) associated with 50% chance of reward . Subjects preferred less rewarding Red target . ( B ) Our model accounts for the data . Because of the boosted anticipation of rewards , the model predicted a preference of less rewarding , but informative , Red target at finite delay periods TDelay . Without boosting , the model predicts a preference of Blue over Red at any delay conditions . Model parameters were taken as the same as Figure 3: ν/γ=0 . 5 , R=1 , η0/γ=3 , c/γ=3 . DOI: http://dx . doi . org/10 . 7554/eLife . 13747 . 009 Figure 3D shows that our model correctly predicted that the value of the 50% reward target ( red line ) would be smaller than that of the 100% reward target ( blue line ) when the delay Tdelay between the cues until rewards , and hence the contribution of anticipation , was small . Thus , the modelled subjects would prefer the reliable target ( indicated by the blue background on left ) . However , when the delay Tdelay was increased , the contribution to the value coming from anticipation was boosted for the 50% reward target , but not for the certain target , because of the RPE associated with the former . This resulted in a preference for the lower probability target ( indicated by the red background in the middle ) , which is consistent with the experimental finding shown in Figure 3A , B ( Spetch et al . , 1990 ) . Note that RPE-boosting is again necessary to capture the reported suboptimal behaviors , as the expected non-boosted value of the 50% reward target would be always smaller than the value of sure reward target ( see the dotted red line in Figure 3D ) . This is because in both cases , the rewards generate conventional anticipation . Finally , as the time delay increased further , both values decayed , making the value of the certain target again greater than that of the 50% reward target ( indicated by the blue background on right ) . This is because the discounting dominated the valuation , leaving the impact of anticipation relatively small . This second reversal is predicted by our model . However , it has not been observed experimentally , other than in findings based on the use of hypothetical questionnaires in Loewenstein ( 1987 ) . ( We note that one of the four animals in Spetch et al . ( 1990 ) did appear to show such a non monotonic preference; however , individual differences also appeared to be very large . ) The model can be used to interpolate between the experiments in Spetch et al . ( 1990 ) ( Figure 3B ) and Gipson et al . ( 2009 ) ( Figure 3C ) , showing a full range of possible tradeoffs between the probability of reward and informative cueing . The phase diagram Figure 3E shows this trade-off for the task described in Figure 3A , as either the probability ( pB ) of reward associated with the 0% info ( blue ) target , or the delay Tdelay change , while the reward probability of 100% info target ( red ) is fixed at 50% . The experiment in Spetch et al . ( 1990 ) ( Figure 3B ) corresponds to pB=1 , while the experiment in Gipson et al . ( 2009 ) ( Figure 3C ) , which used a 75% chance target , corresponds to pB=0 . 75 . As we show in the case of pB≥0 . 5 , the chance of reward is higher for the 0% info ( blue ) target than the 100% info ( red ) , 50% reward target in this diagram ( except for the case pB=0 . 5 , where both targets offer 50% rewards ) . Hence the conventional reinforcement learning model would predict a preference for the 0% info ( blue ) target everywhere . As seen in Figure 3E , the model predicted a similar reversal of preference as in Figure 3D across different probability conditions of pB>0 . 5 We also confirmed that this prediction depends on neither the details of functional form by which RPE influenced anticipation , nor specific parameter values in the model ( See Materials and methods and Figure 3—figure supplement 1 ) . In these calculations , we set the value of no outcome to zero , implying a lack of dread in the no outcome condition . The reason for this was that there was no effect on the preference of pigeons when the delay between the cues signalling no-reward and the no-reward was changed ( Spetch et al . , 1990 ) , while the impact of dread should change over the delay . Moreover changing the delay between choice and cues that signalled no-reward had no impact on preference ( McDevitt et al . , 1997 ) . Note , however , that our results would still hold in case of adding a value to the no outcome . In fact , as detailed in the next section , we found in our human behavioral task that participants assigned the same magnitudes of values to reward and no-reward outcomes , and yet our model still accounted for the preference of advance information . We note the generality of our model . It can account for other various experimental results in different conditions . This includes experiments showing the relative preference for a 100% info target with 25% chance of reward over a 50% info target with a 50% chance of reward ( Stagner and Zentall , 2010; Vasconcelos et al . , 2015 ) , illustrated in Figure 3—figure supplement 2 . A consequence of our model is that the values of predictive cues will be affected by how long subjects subsequently have to wait for the reward – the dynamic changes in values across delay conditions shown in Figure 1C , Figure 3D , E . This has so far only been subject to rather limited tests in animal studies ( Spetch et al . , 1990 ) . We therefore conducted a new human behavioral experiment to test these predictions . In Experiment-1 , 14 heterosexual male human volunteers chose between a 0% info target , which was followed by no cue ( Figure 4A ) , and a 100% info target , which was immediately followed by cues that predicted the presence or absence of reward . The rewards were previously validated lascivious images of female models ( Crockett et al . , 2013 ) . Using this type of primary rewards was crucial for our task design , as was also the case in Crockett et al . ( 2013 ) . This is because other types of rewards , such as monetary rewards , cannot be utilized by participants immediately on each trial as they become available . 10 . 7554/eLife . 13747 . 010Figure 4 . Human decision-making Experiment-1 . ( A ) On each trial , subjects chose either of two colored targets ( Red or Blue in this example ) . Given Red , cue S+ ( oval ) or S0 ( triangle ) was presented , each with probability 0 . 5; S+; was followed by a reward ( an erotic picture ) after time TDelay , while S0 was not followed by reward . Given Blue , either a reward or nothing followed after the fixed time delay TDelay with probability 0 . 5 each . ( B ) Results . Human participants ( n=14 ) showed a significant modulation of choice over delay conditions [one-way ANOVA , F ( 3 , 52 ) =3 . 09 , p=0 . 035] . They showed a significant preference for the 100% info target ( Red ) for the case of long delays [20 s: t⁢ ( 13 ) =3 . 14 , p=0 . 0078 , 40 s: t⁢ ( 13 ) =2 . 60 , p=0 . 022] . The mean +/- SEM indicated by the solid line . The dotted line shows simulated data using the fitted parameters . ( C ) Mean Q-values of targets and predicting cues estimated by the model . The value of informative cue is the mean of the reward predictive cue ( oval ) , which has an inverted U-shape due to positive anticipation , and the no-reward predictive cue ( triangle ) , which has the opposite U-shape due to negative anticipation . The positive anticipation peaks at around 25 s , which is consistent with animal studies shown in Figure 3 ( B , C ) . See Table 2 for the estimated model parameters . ( D ) Model comparison based on integrated Bayesian Information Criterion ( iBIC ) scores . The lower the score , the more favorable the model . Our model of RPE-boosted anticipation with a negative value for no-outcome enjoys significantly better score than the one without a negative value , the one without RPE-boosting , the one without temporal discounting , or other conventional Q-learning models with or without discounting . DOI: http://dx . doi . org/10 . 7554/eLife . 13747 . 01010 . 7554/eLife . 13747 . 011Figure 4—figure supplement 1 . ( A ) Control experiment , where the first block and the last ( 5th ) block of the experiment had the same delay duration of 2 . 5 s . Subjects showed no difference [t⁢ ( 10 ) =1 . 04 , p=0 . 32] in the preference before and after experiencing the other delay conditions . ( B ) The large change in the delay duration affects on choice behavior . Y-axis shows the difference in choice percentage between the shortest ( 2 . 5 s ) and the longest ( 40s ) delay conditions . In our main experiment , the delay duration was gradually increased ( Left ) , while in the control experiment , the delay was abruptly increased . The difference between the two procesures was significant [2 sample t⁢ ( 23 ) =2 . 15 , p=0 . 042] . Subjects reported particularly unpleasant feeling for the long delay condition in the control experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 13747 . 01110 . 7554/eLife . 13747 . 012Figure 4—figure supplement 2 . The generated choice by the model without the negative value assigned to the no-reward outcome . The model fails to capture the short delay period ( 7 . 5 s ) . This corresponds to the time point at which the the effect of negative anticipation was the largest , according to the model with R2 . DOI: http://dx . doi . org/10 . 7554/eLife . 13747 . 012 Subjects experienced blocks of trials with fixed delays ( 2 . 5 s , 7 . 5 s , 20 s and 40 s ) , where the blocks were indicated by target colors . We set the chance of reward to pB=0 . 5 , consistent with the macaque experiments ( Bromberg-Martin and Hikosaka , 2009; 2011 ) . Subjects were not told the exact reward probabilities , merely that rewards would be ‘random’ . We confirmed our model’s central prediction . Subjects showed increased preference for informative cues as the delay increased ( the solid line in Figure 4B indicates group mean and SEM ) . Subjects were on average indifferent in the case of short delays ( 2 . 5 s and 7 . 5 s ) . However , they strongly preferred to choose the informative target in the case of longer delays ( 20 s , 40 s ) . We fitted the choices to our reinforcement-learning ( RL ) model’s trial-by-trial predictions , including the effects of learning from RPE-boosted anticipation . We used a form of hierarchical Bayesian analysis to estimate group level parameters ( Huys et al . , 2011 ) ( see Materials and methods for more details ) . We found that preferences generated from the estimated parameters were consistent with subjects’ choices , as indicated by the black dotted line with the predicted standard error in Figure 4B . Our model predicted striking U-shape changes in the values of the 100% info target and of the 0% info target with respect to the changes in delay length ( Figure 4C ) . The model enjoyed a substantial iBIC score advantage over other possible well-studied reinforcement learning models , or our anticipation RL model without RPE-boosting ( Figure 4D and Table 1 ) . 10 . 7554/eLife . 13747 . 013Table 1 . iBIC scores . Related to Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 13747 . 013ModelN of parametersParametersiBICQ-learning ( with no discounting ) 3α , R+ , R-2598Q-learning ( with discounting ) 5α , R+ , R- , γ+ , γ-2643Anticipation RL without RPE-boosting7α , R+ , R- , γ+ ( =γ- ) , ν+ , ν- , η02659Boosted anticipation RL without R−4α , R+ , γ+ , ν+2616Boosted anticipation RL with no discounting5α , R+ , R- , ν+ , ν-2595Boosted anticipation RL6α , R+ , R- , γ+ ( =γ- ) , ν+ , ν-2583 To investigate possible adaptation to delay , we also ran a control experiment on an additional 11 subjects for whom we changed the order of the delays , and also repeated the same 2 . 5 s delay in the first and last ( fifth ) blocks . Preferences did not differ ( Figure 4—figure supplement 1A ) from beginning to end , suggesting stability . However , there was a moderately significant evidence of adaptation , with the effect of the 40 s delay being much greater following the extensive experience of 2 . 5 s delay than the 20 s delay ( Figure 4—figure supplement 1B ) . To investigate the robustness of the delay dependent preference of advance information further , we conducted an additional experiment ( Experiment-2 ) on a newly recruited population of 31 participants ( Figure 5 ) . At the beginning of each trial , explicit cues provided participants with full information about the current delay condition ( either 1 s , 5 s , 10 s , 20 s or 40 s ) and the ( constant , 50% ) reward probability ( Figure 5A , B ) . The basic structure of the task was similar to Experiment-1 ( Figure 5C ) ; participants had to choose either 100% info target or 0% info target . Crucially , though , the delay condition was randomized across trials . As seen in Figure 5D participants showed significant preference of the 100% info target when delay was long , replicating the results of Experiment-1 . We also found no effect of the delay condition of a previous trial , providing further support for the absence of a block order confound in Experiment 1 . ( We computed each participant’s preference of 100% info target at a particular delay condition T1 , conditioned on a previous delay condition T2 . We tested the effect of previous delay conditions T2 over participants via one-way ANOVA for ( 1 ) the preference at T1=1 s [F ( 4 , 111 ) =2 . 36;p=0 . 06]; ( 2 ) the preference at T1=5 s [F ( 4 , 101 ) =0 . 85;p=0 . 50]; ( 3 ) the preference at T1=10 s [F ( 4 , 99 ) =1 . 45;p=0 . 22]; ( 4 ) the preference at T1=20 s [F ( 4 , 87 ) =1 . 48;p=0 . 21]; ( 5 ) the preference at T1=40 s [F ( 4 , 98 ) =0 . 75; p=0 . 56] ) . Also , Experiment-2 was designed to have an equal number of trials per delay condition , while Experiment-1 was designed to equalize the amount of time that participants spent in each condition ( see Task Procedures in Materials and methods ) . The fact that we obtained the same results in both experiments illustrates the robustness of our findings . 10 . 7554/eLife . 13747 . 014Figure 5 . Human decision-making Experiment-2 . ( A ) A screen-shot from the beginning of each trial . The meaning of targets ( 'Find out now' or 'Keep it secret' ) , the duration of Tdelay ( the number of hourglass ) , and the chance of rewards ( the hemisphere =0 . 5 ) were indicated explicitly . ( B ) The number of hourglasses indicated the duration of Tdelay until reward . One hourglass indicated 5 s of Tdelay . When Tdelay=1 s , a fraction of an hourglass was shown . This was instructed before the experiment began . The delay condition Tdelay was changed randomly across trials . ( C ) The task structure . The task structure was similar to Experiment-1 , except that the 0% info target ( Blue ) was followed by a no-info cue , and an image symbolizing the lack of reward was presented when no reward outcome was delivered . ( D ) Results . Human participants ( n=31 ) showed a significant modulation of choice over delay conditions [one-way ANOVA , F ( 4 , 150 ) =3 . 72 , p=0 . 0065] . The choice fraction was not different from 0 . 5 at short delays [1 s: t⁢ ( 30 ) =0 . 83 , p=0 . 42 5 s: t⁢ ( 30 ) =0 . 70 , p=0 . 49 , 10 s: t⁢ ( 30 ) =0 . 26 , p=0 . 80] but it was significantly different from 0 . 5 at long delays [20 s: t⁢ ( 30 ) =2 . 86 , p=0 . 0077 , 40 s: t⁢ ( 30 ) =3 . 17 , p=0 . 0035] , confirming our model’s key prediction . The mean and +/- SEM are indicated by the point and error bar . DOI: http://dx . doi . org/10 . 7554/eLife . 13747 . 014 In fitting the model , we were surprised to find that subjects assigned negative values to no-reward outcomes and its predicting cue ( the bottom dotted line in Figure 4C , see also Table 2 ) for the estimated model parameters ) . This negativity emerged through our fitting , as we did not assume the sign of the value . Forcing this outcome to be worth 0 led to a significantly worse iBIC score ( Figure 4D and Table 1 ) . We found a particular effect of dread at the delay of 7 . 5 s , which the model interpreted as implying that the time scale of savoring was longer than that of dread ( see ν+ and ν- in Table 2 ) . Omitting the negative value of the no-reward cue led to a failure to fit this effect ( Figure 4—figure supplement 2 ) . 10 . 7554/eLife . 13747 . 015Table 2 . Related to Figure 4 . The group means μ that estimated by hierarchical Bayesian analysis for our human experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 13747 . 015αc⁢R+c⁢R-γ+ ( =γ- ) ν+ν-0 . 170 . 85-0 . 840 . 041 ( s⁢e⁢c-1 ) 0 . 082 ( s⁢e⁢c-1 ) 0 . 41 ( s⁢e⁢c-1 ) Thus analysis using our model showed that the values of targets were computed via a competition between savoring and dread ( Figure 4C ) . That is , when participants chose 0% info target , they experienced a mixture of the baseline savouring of possible reward and the baseline dread of possible no-reward during the wait period . On the other hand , when they chose the 100% info target , they experienced either savoring that was boosted by the RPE from the reward predictive cue , or dread that was boosted by the RPE from the no-reward predictive cue . It is this that required us to use the absolute value of the RPE to boost the effects of both savouring and dread . Specifically , RPE from the no-reward predictive cue also boosted the impact of dread , rather than damped it . Note that the difference between the data and the RPE-boosted anticipation model without dread , seen in ( Figure 4—figure supplement 2 ) at short delay periods , implies that the increase of preference of 100% info target was lower than what the model with only savouring can predict . This is because the preference of targets reflected the competition between positive savouring and negative dread , both of which changed non-monotonically with delay ( Figure 4C ) . More precisely , for delays wherein the impact of savoring and dread were similarly strong , choice preference remained at around the chance level . This phenomenon was confirmed in our Experiment-1 and Experiment-2 at short delay conditions ( <10 s ) , where the choice probability remained around 0 . 5 ( Figure 4C , 5D and Figure 4—figure supplement 2 ) . Since the timescale of dread was smaller than that for savoring , the effect of competition was present only at the short delay conditions . By contrast , at longer delay conditions ( >20 s ) , dread was discounted and savoring became dominant . This resulted in a large increase of preference for the 100% info target . This sudden increase in choice preference was caused by the non-monotonic Q-value functions of targets ( Figure 4C ) . Our model comparison analysis also supported this conclusion ( Figure 4D ) , where the model with non-monotonic value functions ( our original model with temporal discounting with the estimated group mean of the discounting rate: γ=0 . 04s−1 ) outperformed the model with monotonic value functions ( our model without temporal discounting γ=0 ) . The difference in the iBIC score was 12 . Note that temporal discounting of savoring at extremely long delays should also make subjects indifferent between 100% and 0% info targets . Unfortunately , we failed to confirm this prediction in the current study . In fact , analysis of our model suggests that in order to confirm this effect the delays concerned would need to be more than 135 s to enable detection of a difference between the preference at the extremely long delay and the preference at TDelay=20 s ( p=0 . 05 , n=30 ) . Thus , we were not able to confirm this indifference in our task , and leave this for future studies . Nonetheless , we note that our model with temporal discounting outperformed a model without temporal discounting in terms of the iBIC scores ( Figure 4D ) , where the latter model predicts monotonic value functions . The magnitude of our discount rate ( 0 . 04 s-1 ) is comparable with other intertemporal choice studies ( e . g . 0 . 095 s-1 for monetary rewards ( Schweighofer et al . , 2008 ) , and 0 . 014 s-1 for primary ( juice ) rewards ( McClure et al . , 2007 ) . Note the latter was inferred using a double exponential model , with the faster decay being 0 . 014 s-1 and the slower decay not being significantly different from 0 s-1 . Nonetheless , we should also point out that comparing discounting rates across different experimental designs is extremely difficult . For instance , the nature of discounting of primary rewards ( e . g . juice or pleasant images ) is very likely to be different from discounting of monetary rewards , as money cannot be spent immediately . In fact , Reuben et al . ( 2010 ) reported that primary rewards ( chocolates ) were more rapidly discounted than monetary rewards . It is also known that addicts discount the addictive substance at a higher rate than money ( e . g . see ( Bickel et al . , 1999 ) for cigarettes , ( Bickel et al . , 2011 ) for cocaine ) . The characteristic timescales of these experiments were much longer ( weeks or months ) and discounting rates in these literatures are , however , very small compared to ours . This also suggests that comparisons across experiments could be very misleading , since discounting can be adaptive to experimental timescales ( Kable and Glimcher , 2010 ) . We acknowledge that in the current study we did not test our model’s prediction of preference reversal , as we designed the task so that the average amount of reward obtained from each target per trial was the same for both targets . This is a limitation of our current study , and we leave this issue to future investigations . Although reward prediction errors ( RPEs ) have historically been treated as straightforward learning signals , there is increasing evidence that they also play more direct roles , including in classical conditioning ( or Pavlovian ) approach ( McClure et al . , 2003 ) and subjective well-being ( Rutledge et al . , 2014 , 2015 ) . The latter study found that subjective well-being , or happiness , is influenced more prominently by RPE than reward itself , which has echoes with an older idea often referred to as the ‘hedonic treadmill’ ( Brickman and Campbell , 1971; Frederick and Loewenstein , 1999 ) . Here we considered a further contribution of RPEs , stemming from their ability to boost the value of the anticipation of reward . RPE-boosted anticipation provides a natural account for gambling behaviors . Indeed , our model further predicts that the tendency to be risk-seeking or risk-averse is subject to change as a function of the delay between the cues and rewards . This has important consequences for gambling , as well as the nature and measurement of risk attitudes in general . Specifically , our findings suggest that an unexpected prize will have greater motivational impact when there is a moderate delay between its revelation and its realization . Further experiments will be required to confirm this novel prediction in the context of more conventional economic gambling tasks . We also note that it has been shown in macaque monkeys that changing inter-trial-intervals can impact risk sensitivity ( McCoy and Platt , 2005; Hayden and Platt , 2007 ) . This indicates that the anticipation and discounting of future outcomes over multiple trials may also play important roles in determining risk attitudes . RPE-boosted anticipation , like many apparently Pavlovian behaviors that involve innate responses , appears evidently suboptimal . As we have seen , choice can be strikingly non-normative . Our results are consistent with notions such as curiosity/exploration bonuses , and uncertainty aversion ( Loewenstein , 1994; Caplin and Leahy , 2001; Litman , 2005; Daw et al . , 2006; Fiorillo , 2011; Friston et al . , 2013; 2015; Gottlieb et al . , 2013; Blanchard et al . , 2015a; Kidd and Hayden , 2015 ) . However , whether the behaviors reflect mechanistic constraints on neural computation ( Kakade and Dayan , 2002 ) , or a suitable adaptation to typical evolutionary environments , remains a question for further research . In our human experiments , we found that participants assigned a negative value to a no-outcome . This appears not to be the case in reported pigeon experiments . One idea is that this negative value emerges from a form of normalization of subjective values ( Tobler et al . , 2005; 2007; Louie et al . , 2013 ) consistent with the finding that human subjects can assign ‘unpleasantness’ to no-reward stimuli ( Tobler et al . , 2007 ) . The effect in our task would be that subjects would apparently experience the anticipation of both positive and negative outcomes in this task as being pleasant for the reward predictive cue ( savouring ) but unpleasant for the no-outcome predictive cue ( dread ) . This was confirmed in informal debriefing after the experiment . Note that in the monkey experiments ( Bromberg-Martin and Hikosaka , 2009; 2011 ) , the lower-value outcome still involved an actual reward , albeit of a smaller size . We showed that responses of habenula neurons to reward predictive cues , which have been proposed as an ‘information prediction error’ ( Bromberg-Martin and Hikosaka , 2009; 2011 ) , could be accounted for by our model in terms of conventional reward prediction errors . This is because our model included the value of anticipation of rewards that can be boosted by RPE . Further studies are necessary to explore the calculation and representation of the anticipation itself , for both savouring ( of positive outcomes ) and dread ( of negative ones ) . We note recent experimental findings in basal forebrain suggest seductive similarities ( Monosov and Hikosaka , 2013; Monosov et al . , 2015 ) , while other brain areas , such as ventral striatum ( Jensen et al . , 2003; Hariri et al . , 2006; Salimpoor et al . , 2011 ) , posterior insula and anterior cingulate cortex ( Berns et al . , 2006; Blanchard et al . , 2015b ) , may also contribute . Furthermore , ramping dopamine signals toward the delivery of rewards ( if they generally exist , see [Morris et al . , 2004] ) could also be related to the anticipation of rewards ( Fiorillo et al . , 2003; Howe et al . , 2013; Lloyd and Dayan , 2015; Hamid et al . , 2016; Hart et al . , 2015 ) , while dopamine neurons have also been shown to manifest a stronger phasic response to one predicting an uncertain reward than to a cue predicting a certain reward ( Fiorillo , 2011 ) . It would also be interesting to study the difference between savouring and dread , particularly given debates concerning the encoding of information about punishment Fiorillo et al . ( 2003 ) , Brischoux et al . ( 2009 ) , Lammel et al . ( 2014 ) , and about the symmetry or otherwise between the encoding of positive and negative prediction errors for reward ( Fiorillo , 2013; Hart et al . , 2014 ) One issue that merits further future study is adaptation to different delays . It has been shown that human subjects are capable of optimizing their discounting rate according to task demands ( Schweighofer et al . , 2006 ) , and also that the discounting may be computed relative to the timing of other available options , rather than absolute time ( Kable and Glimcher , 2010 ) . Our control experiment for Experiment-1 showed a signature of adaptation ( Figure 4—figure supplement 1A ) , with subjects reacting differently to a sudden large increase in delays after many trials with small delays , compared to a gradual increase in delays as in Experiment-1 . However we found no evidence for this effect in our Experiment-2 in which delay conditions were randomized on a trial-by-trial basis . It would be interesting to study this further , in relation to the uncertainty in timing of rewards . In our task there was no effect of timing uncertainty on choice , as both targets are associated with the same delay . However it would become important if a task involves a choice between targets with different delay conditions . Furthermore , if the prediction error could influence subjective time ( for instance via a known effect of dopamine on aspects of timing [Gibbon et al . , 1997] ) , then this could have complex additional effects on anticipation . In sum , we account for a well-described preference for observing behavior through a suggestion that reward prediction errors modulate the contribution to subjective value that arises from the anticipation of upcoming rewards . Our study provides a new perspective on reward-based learning and decision-making under uncertainty , and may be of special relevance to gambling and addiction . 56 heterosexual male participants ( age 18–40 ) were recruited from the UCL community . Participants were paid 10 British pounds at the end of the experiment . Participants provided informed consent for their participation in the study , which was approved by the UCL ethics committee ( UCL Research Ethics Reference: 3450/002 ) . We sought to use basic rewards that could be consumed by subjects on each trial at the time of provision . We therefore employed images of female models that had previously been rated by heterosexual male subjects ( Crockett et al . , 2013 ) . In case of reward , a random one of the top 100 highest rated images was presented to subjects without replacement . We sought to determine the distribution of model parameters 𝐡 . Thus following ( Huys et al . , 2011 ) , we conducted a hierarchical Bayesian , random effects analysis , where the ( suitably transformed ) parameters 𝐡i of individual i are treated as a random sample from a population distribution , which we assume to be Gaussian , with means and variance 𝜽={𝝁θ , 𝚺θ} . The prior group distribution 𝜽 can be set as the maximum likelihood estimate: ( 3 ) θML≈argmaxθ{p ( D\|θ ) }=argmaxθ{∏i=1N∫dhi p ( Di\|hi ) p ( hi\|θ ) } We optimized 𝜽 using an approximate Expectation-Maximization procedure . For the E-step of the k-th iteration , we employed a Laplace approximation , obtaining , ( 4 ) mik≈argmaxh{p ( Di\|h ) p ( h\|θk−1 ) } ( 5 ) p ( hik\|Di ) ≈𝒩 ( mik , Σik ) , where 𝒩⁢ ( 𝐦ik , 𝚺ik ) is the Normal distribution with the mean 𝐦ik and the covariance 𝚺ik that is obtained from the inverse Hessian around 𝐦ik . For the M step: ( 6 ) μθk+1=1N∑i=1Nmik ( 7 ) Σθk+1=1N∑i=1N ( mikmikT+Σik ) −μθk+1μθk+1T . For simplicity , we assumed that the covariance Σθk had zero off-diagonal terms , assuming that the effects were independent . Also , in order to treat different delay conditions equally , we randomly sub-sampled the trials to equalize the number used per condition in order to calculate the statistics . Additionally we obtained the same results by normalizing the posterior for each delay condition when estimating the expectation . For the model-free data analysis , we used the t-test , as the data passed the Shapiro-Wilk normality test and the paired F-test for equal variances for each and between conditions . We compared models according to their integrated Bayes Information Criterion ( iBIC ) scores , based on a flat prior over models . We analysed model log likelihood log⁡ p ( D\|M ) : ( 8 ) log⁡p ( D\|M ) =∫dθp ( D\|θ ) p ( θ\|M ) ( 9 ) ≈−12iBIC=log⁡p ( D\|θML ) −12\|M\|log⁡\|D\| , where iBIC is the integrated Baysian Information Criterion , \|M\| is the number of fitted parameters of the prior and \|D\| is the number of data points ( total number of choices made by all subjects ) . Here , log⁡p ( D\|θML ) can be computed by integrating out individual parameters: ( 10 ) log⁡p ( D\|θML ) =∑ilog⁡∫dhp ( Di\|h ) p ( h\|θML ) ( 11 ) ≈∑ilog⁡1K∑j=1Kp ( Di\|hj ) , where we approximated the integral as the average over K samples 𝐡j’s generated from the prior p ( h\|θML ) . As seen in Table 1 , our model of anticipation fit better than conventional Q-learning models with or without discounting ( the latter two models being equivalent to our model with parameters set such that there is no anticipation . ) We describe our model for the case of a simple conditioning task . Suppose that a subject takes an action and receives a reward predictive cue S+ at t=0 with a probability of q followed by a reward R at t=T ( =TDelay ) , or no-reward predictive cue S0 at t=0 with a probability of 1-q followed by no reward . Following ( Loewenstein , 1987; Berns et al . , 2006; Story et al . , 2013 ) , the anticipation of the reward at time t is worth a⁢ ( t ) =R⁢e-ν⁢ ( T-t ) , where ν governs its rate . Including R itself , and taking temporal discounting into account , the total value of the reward predictive cue , QS+ , is ( 12 ) QS+=ηV[Anticipation]+V[Reward]=η∫0Te−γt′a ( t′ ) dt′+Re−γT=ηRν−γ ( e−γT−e−νT ) +Re−γT , where η is the relative weight of anticipation and γ is the discounting rate . In previous work , η has been treated as a constant; however , here we propose that it can vary with the prediction error δp⁢e at the predicting cue . While the simplest form is the linear relationship given by Equation ( 2 ) , our model’s behavior does not depend on the details of the RPE dependence of anticipation . In fact , one can instead assume ( 13 ) η=η0+c1tanh ( c2\|δpe\| ) where c1 and c2 are constants , or ( 14 ) η=η0+cθ ( \|δpe\| ) where θ⁢ ( x ) is the step function that we define: θ⁢ ( x ) =1 for x>0 and θ⁢ ( x ) =0 for x≤0 . All the findings in this paper hold for Equation ( 2 ) , Equation ( 13 ) , and Equation ( 14 ) ( see Figure 3—figure supplement 1 ) . In fact , Equation ( 2 ) and Equation ( 14 ) can be thought as different approximations to Equation ( 13 ) ( see below ) . Note that in our model , RPE affects the total value QS+ , which also affects subsequent RPEs . One might therefore wonder whether there is a stable value for the cue . In the Results , we introduced a linear ansatz for the boosting of anticipation on RPE ( Equation ( 2 ) . In a wide range of parameter regime , this ansatz has a stable , self-consistent , solution; however , in a small parameter regime , the linear ansatz fails to provide such solutions . This is because the linear assumption allows unbounded boosting . Crudely , the RPE can usually be expressed as a linear combination of Q-values . Thus , the following equation has to have a real solution: ( 15 ) δpe=α\|δpe\|+β , where α , β are determined by the task . The solution is the intercept of two lines: y=δp⁢e and y=α⁢\|δp⁢e\|+β , which does not exist when α>1 and β>0 . In this example , the prediction error at the reward predictive cue is positive , and ( 16 ) δpe=QS+−qQS+= ( 1−q ) QS+= ( 1−q ) ( ( η0+cδpe ) V [Anticipation]+V[Reward] ) . This has a solution at δpe>0 only if ( 17 ) ( 1−q ) cV[Anticipation]<1 . This condition can be violated , for instance , if c is very large ( more precisely , if c>1−qV[Anticipation] ) . Roughly speaking , the stability condition is violated when the boosted anticipation is very large . If it indeed exists , the solution isδpe= ( 1−q ) ( η0V[Anticipation]+V[Reward] ) 1− ( 1−q ) cV[Anticipation] which gives ( 19 ) QS+=η0V[Anticipation]+V[Reward]1− ( 1−q ) cV[Anticipation] . To avoid the stability problem problem , one can instead assume Equation ( 13 ) . This is a more general form which leads to the self-consistency equation: ( 20 ) δp⁢e=α⁢tanh⁡ ( c2⁢\|δp⁢e\| ) +β , which requires y=δp⁢e and y=α⁢tanh⁡ ( c2⁢\|δp⁢e\| ) +β to intersect . This can happen for any real α , β . However , importantly , our model’s behavior does not depend on the details of the RPE dependence of anticipation ( see Figure 3—figure supplement 1 ) . Hence we did not attempt to determine the exact functional form . Note that Equation ( 2 ) can be thought of as an approximation to Equation ( 13 ) when c2⁢\|δp⁢e\| is small . In the limit of c2→∞ , on the other hand , Equation ( 13 ) becomes Equation ( 14 ) . Equation ( 14 ) can be used instead of Equation ( 13 ) when c2⁢\|δp⁢e\| is large; or also can be used as an approximation of Equation ( 2 ) when the size of RPE is roughly the same from trial to trial . To see how the model works , take the task introduced in Bromberg-Martin and Hikosaka ( 2011 ) ( Figure 2A ) . We assume that the 100% info target is followed randomly by a cue SBig that is always followed by a big reward RBig after a delay T , or by a cue SSmall that is always followed by a small reward RSmall ( <RBig ) . The 0% info target is followed by a cue SRandom , which is followed by the reward RBig or RSmall with equal probabilities . The 50% info target is followed by either of the three cues SBig , SSmall , or SRandom with a probability of 1/4 , 1/4 or 1/2 , respectively . The expected values of targets ( Q100% , Q50% , Q0% ) are ( 21 ) Q100%=QSBig+QSSmall2 ( 22 ) Q0%=QSRandom ( 23 ) Q50%=QSBig+QSSmall4+QSRandom2 where QSBig , QSSmall , QSRandom are the expected values of cues SBig , SSmall , SRandom , respectively . The RPE at the cues depends on the chosen target , implying that the average values of the cues can be expressed ( assuming that the transition probabilities are properly learned ) as ( 24 ) QSBig=VSBig , 100%+VSBig , 50%2 ( 25 ) QSSmall=VSSmall , 100%+VSSmall , 50%2 ( 26 ) QSRandom=VSRandom , 50%+VSRandom , 0%2 where VSj , X% is the mean values of the cues Sj ( j= Big , Small , or Random ) in case following the X% info target ( X=100 , 50 , or 0 ) : ( 27 ) VSBig , X% , =ηSBig , X%ABig+BBig ( 28 ) VSSmall , X%=ηSSmall , X%ASmall+BSmall ( 29 ) VSRandom , X% , =ηSRandom , X%ABig+BBig+ηSRandom , X%ASmall+BSmall2 where , under the assumption Equation ( 2 ) , ( 30 ) ηSj , X%=η0+c\|δpeSj , X%\| , where δp⁢eSj , X% is RPE at the cue Sj after choosing X% target ( 31 ) δp⁢eSj , X%=QSj-QX% and Al and Bl ( l is the reward size; in our case Big or Small ) are the anticipation and the reward itself: ( 32 ) Al=Rlν−γ ( e−γT−e−νT ) ( 33 ) Bl=Rle−γT Note that we ignored any anticipation of the cues themselves after the choice . This would not alter the qualitative predictions of the model . Recent experiments ( McDevitt et al . , 1997; 2016 ) showed that delaying the timing of reward predictive cues decreased the preference for the informative target . This is consistent with our model because delaying the cue presentation means decreasing the wait time; hence leads to a smaller impact of boosted anticipation . In our experiment , the time between choice and the cue presentation was too short to be significant . The probability of choosing X% target over Y% target , PX%-Y% , is assumed to be a sigmoid function of the difference between the target values: ( 34 ) PX%-Y%=11+e-QX%-QY%σ These equations account well for the behavioral and neuronal findings in Bromberg-Martin and Hikosaka ( 2011 ) . The results in Figure 2D , E are obtained from these equations with RBig=0 . 88 , RSmall=0 . 04 , ν=0 . 5 sec−1 , TDelay=2 . 25 sec , γ=0 . 1 sec−1 , σ=0 . 08 . Spetch et al . ( 1990 ) reported the striking finding that pigeons can prefer a target that is followed by reward with a probability of 0 . 5 over a target that is always followed by a reward under certain conditions . Here we show that our model can also account for this surprisingly ‘irrational’ behavior . A generalized version of the task is schematically shown in Figure 3A . On each trial , a subject chooses either of two colored targets ( Red or Blue in this example ) . If Red is chosen , one of the cues S+ or S0 is randomly presented , where S+ is always followed by a reward after time TDelay , while S0 is never followed by reward . If Blue is chosen , a cue S* is presented and a reward is followed after the fixed time delay TDelay with a probability of pB . The task in Spetch et al . ( 1990 ) corresponds to the case with pB=1 , and the task in Gipson et al . ( 2009 ) corresponds to the case with pB=0 . 75 . In both cases , by always choosing Blue , animals can get the maximum amount of rewards; however , it has been shown that animals can prefer to choose Red over Blue ( Spetch et al . , 1990; Gipson et al . , 2009 ) , where the preference of Red with pB=1 appeared to be heavily dependent on the length of the delay between the predicting cues and the delivery of rewards . Our model can account for the irrational behaviors . We first determine the value of choice , QRed . Under the linear ansatz ( 2 ) , the prediction error at the cue S+ is ( 35 ) δpe=QS+−QRed ( 36 ) =12QS+ ( 37 ) =12 ( ( η0+cδpe ) A+B ) , or ( 38 ) δpe=12 ( η0A+B ) 1−12cA and ( 39 ) QS+= ( η0A+B ) 1−12cA ( 40 ) Qred=QS++QS02 ( 41 ) =12 ( η0A+B ) 1−12cA , where ( 42 ) A=Rν−γ ( e−γT−e−νT ) ( 43 ) B=Re−γT with R being the size of reward . Equation ( 39 ) shows how the Q-value of reward predictive cue is boosted . When there is no boosting , c=0 , the denominator is 1 . Increasing the boosting c will decrease the denominator; hence it will increase the Q-value . Note that the denominator is assumed to be positive within the linear ansatz . The value of choice Blue is simplyQBlue=pB ( η0A+B ) Hence the difference in the values in two choices is ( 45 ) QRed−QBlue= ( η0A+B ) ( 1211−12cA−pB ) . Even in the case of pB>1/2 , this expression can become positive by changing the time delay T ( Figure 3B , C ) , which can account for the irrational observing behaviors ( Spetch et al . , 1990; Gipson et al . , 2009 ) . The diagram in Figure 3E shows the results with the ansatz of Equation ( 2 ) , while Figure 3—figure supplement 1 shows the results with the ansatz of Equation ( 13 ) . Note that the two ansatz provide qualitatively very similar results . Since there are numerous variations of this experiment , here we provide with a formula for a more general case . Suppose a subject chooses either of two colored targets 100% info ( I ) or 0% info ( N ) . If target I is chosen , one of the cues S+ or S0 is randomly presented with a probability of pI and 1-pI , where S+ is always followed by a reward with a size RI after time TDelay , while S0 is never followed by reward . If target N is chosen , a cue S* is presented and a reward with a size R2 is followed after the fixed time delay TDelay with a probability of pN⁢I . The difference in the values in two choices with the linear boosting ansatz is expressed asQInfo-QNo-Info= ( η0ν-γ⁢ ( e-γ⁢T-e-ν⁢T ) +e-γ⁢T ) ⁢ ( pI⁢RI1- ( 1-pI ) ⁢c⁢AI-pN⁢RN ) where , ( 47 ) AI=RIν-γ⁢ ( e-γ⁢T-e-ν⁢T ) . It is straightforward to apply this formula to specific experiments with a specific set of condition . For example , ( Stagner and Zentall , 2010 ) ’s experimental results can be accounted for by setting RI=RN , pI=0 . 2 , and pN=0 . 5 . As shown in Figure 3—figure supplement 2 , our model reproduced the reported sub-optimal observing behavior . Our experiment , shown schematically in Figure 4A , was designed to test key aspects of the model . Subjects choose between info and non-info targets of values Qinfo and Qno-info respectively . The info target is followed by the reward predicting cue S+ with the value of Q+ , followed by a reward R+ after a delay of T , or the no-reward prediction cue S- ( here we write S- instead of S0 for our convention ) with the value Q- , followed by a no-reward with a value of R- . The no-info target is followed by no cue ( which we write S* for our convention ) and randomly followed by a reward R+ or no-reward R- after the delay of T . Note that we needed to introduce the value for the no-reward outcome based on the behaviors and self-reports . Here , we fit the simplest form of the model to the subjects’ behavior . Since we fit the model trial by trial , we need to introduce learning . After each trial , the value of chosen target was updated asQX→QX+α⁢ ( V-QX ) where X=info or no-info , α is the learning rate and V is the reward function: ( 49 ) V=ηSiAi+Bi with ( 50 ) Ai=Riνi−γ ( e−γT−e−νiT ) ( 51 ) Bi=Rie−γT where i=+ or - , in case of a reward or a no-reward , respectively . We assumed that the anticipation rate ν+ and ν- can be different but the discounting rate is the same γ=γ+=γ- . The coefficient ηSi was assumed to be ( 52 ) ηSi=η0+c⁢θ⁢ ( δp⁢eSi ) . This assumption allows us to reduce the number of free parameters . To see it , we write the expected values of targets: ( 53 ) Qinfo= ( η0+c ) ( A++A− ) +B++B−2 ( 54 ) Qno-info=η0 ( A++A− ) +B++B−2 . Hence the expected difference is ( 55 ) ΔQ=Qinfo−Qno-info=c ( A++A− ) 2 and the probability of choosing the info target Pinfo is ( 56 ) Pinfo=11+e-⁢Δ⁢Qσ . As is common , c and σ appear together with Ri’s ( c⁢R+/σ , c⁢R-/σ ) . Thus , we set c=1 and σ=1 for the fitting . Thus the model has six free parameters that are fit: R+ , R- , ν+ , ν- , γ and α . Note that a model that can allow asymmetric dependence of boosting on prediction errors ( if it is positive or negative ) leads to a related expression: ( 57 ) ηSi=η0+c+⁢θ⁢ ( δp⁢eSi ) +c-⁢θ⁢ ( -δp⁢eSi ) , which leads to ( 58 ) Qinfo=η0 ( A++A− ) +c+A++c−A−+B++B−2 ( 59 ) Qno-info=η0 ( A++A− ) +B++B−2 . Hence the expected difference is ( 60 ) ΔQ=Qinfo−Qno-info=c+A++c−A−2 . Thus c+⁢R+/σ and c-⁢R-/σ will be fitted as independent variables , which means that the asymmetric boosting will appear in the ratio between R+ , R- . For the purpose of model comparison , we also fitted the simple Q-learning model ( γ=ν+=ν-=0 ) , the Q-learning model with discounting ( ν+=ν-=0 ) , the RPE-boosting RL model with no value for the no-outcome ( R-=0 and ν-=0 ) , the RPE-boosting RL model with no discounting ( γ=0 ) , and the anticipation RL model with no-boosting . To fit the model with no-boosting ( c=0 ) , we fitted a full model with η0 . Note that from Equation ( 60 ) , the expected difference between the values of targets is zero Δ⁢Q=0 when there is no boosting c=0 . This means that RPE-boosting is necessary to account for observing behaviors that prefer advance reward information .	People , pigeons and monkeys often want to know in advance whether they will receive a reward in the future . This behaviour is irrational when individuals pay for costly information that makes no difference to an eventual outcome . One explanation is that individuals seek information because anticipating reward has hedonic value ( it produces a feeling of pleasure ) . Consistent with this , pigeons are more likely to seek information when they have to wait longer for the potential reward . However , existing models cannot account for why this anticipation of rewards leads to irrational information-seeking . In many situations , animals are uncertain about what is going to happen . Providing new information can produce a “prediction error” that indexes a discrepancy between what is expected and what actually happens . Iigaya et al . have now developed a mathematical model of information-seeking in which anticipation is boosted by this prediction error . The model accounts for a wide range of previously unexplained data from monkeys and pigeons . It also successfully explains the behaviour of a group of human volunteers from whom Iigaya et al . elicited informational and actual decisions concerning uncertain and delayed rewards . The longer that the participants had to wait for possible rewards , the more avidly they wanted to find out about them . Further research is now needed to investigate the neural underpinnings of anticipation and its boosting by prediction errors .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	The modulation of savouring by prediction error and its effects on choice
Social cellular aggregation or multicellular organization pose increased risk of transmission of infections through the system upon infection of a single cell . The generality of the evolutionary responses to this outside of Metazoa remains unclear . We report the discovery of several thematically unified , remarkable biological conflict systems preponderantly present in multicellular prokaryotes . These combine thresholding mechanisms utilizing NTPase chaperones ( the MoxR-vWA couple ) , GTPases and proteolytic cascades with hypervariable effectors , which vary either by using a reverse transcriptase-dependent diversity-generating system or through a system of acquisition of diverse protein modules , typically in inactive form , from various cellular subsystems . Conciliant lines of evidence indicate their deployment against invasive entities , like viruses , to limit their spread in multicellular/social contexts via physical containment , dominant-negative interactions or apoptosis . These findings argue for both a similar operational ‘grammar’ and shared protein domains in the sensing and limiting of infections during the multiple emergences of multicellularity . Genetic systems are locked in multilevel conflicts which span all levels of biological organization . These include intra-genomic conflicts between genetic elements within genomes , the conflict between distinct replicons in a cell and inter-organismal conflicts of unicellular and multicellular organisms between members of same or different species ( Aravind et al . , 2012; Dawkins and Krebs , 1979; Hurst et al . , 1996; Werren , 2011; Smith and Price , 1973; Austin et al . , 2009 ) . As success in these biological conflicts is central to the survival of organisms , the genetically encoded traits associated with them are under intense selection . As a result , these adaptations are part of a continuing arms-race between the genetic entities locked in conflict , wherein each tries to outperform the others resulting in constant selection for better armaments ( Dawkins and Krebs , 1979; Thompson , 1994; Iyer et al . , 2011a; Van Melderen , 2010 ) . Consequently , biological conflicts have left their prominent marks on the physiological , morphological and behavioral adaptations of organisms and show up as clear-cut signatures in the genomes of organisms . One of the recent triumphs of comparative genomics has been the successful prediction of genomic signatures of biological conflict and counter-conflict systems . Consequently , these genomic imprints of biological conflicts have served as excellent models to study evolution on a ‘fast-track’ ( Aravind et al . , 2012 ) . Additionally , they have also provided ‘work-horse’ reagents for molecular biology such as restriction enzymes , nucleic-acid-modifying enzymes and CRISPR/Cas9 technologies ( Murray , 2000; Lavender et al . , 2018; Gaj et al . , 2013 ) . Much of the molecular weaponry in these biological conflicts takes the form of effectors that accomplish their action via attacks on the biomolecules that transmit information in ‘the central dogma’ , namely genomic DNA , the various RNAs involved in protein synthesis , the protein-synthesizing engine , that is the ribosome , or proteins of the replication , transcription and translation systems ( Werren , 2011; Smith and Price , 1973; Iyer et al . , 2011a; Leplae et al . , 2011; Proft , 2005; Walsh , 2003 ) . Less-frequent modes of action include attacks on components of the signal-transduction systems or direct rupture of cells by membrane perforation or destabilization ( Rappuoli and Montecucco , 1997; Gilbert , 2002; Burroughs and Aravind , 2016; Burroughs et al . , 2015; Lambert , 1978 ) . In terms of their chemistry , the weaponry deployed in these conflicts takes a very diverse form ranging from low-molecular-weight ‘toxins’ and ‘antibiotics’ synthesized by the products of a range of multi-gene biosynthetic operons to proteinaceous toxins that feature some of the largest-described polypeptides in their ranks ( Walsh , 2003; Iyer et al . , 2017 ) . A consequence of the intense natural selection acting on these molecules is their rapid evolution with extreme diversity in terms of structure and sequence . This often serves as a telltale marker to identify them through sequence-analysis and comparative genomics ( Dawkins and Krebs , 1979; Thompson , 1994; Iyer et al . , 2011a; Van Melderen , 2010 ) . Effector-deployment poses several challenges to the host . First and foremost , given that the effectors frequently target the central dogma systems , which tend to be universally shared , there is the need for mechanisms to specifically target rival or non-self genetic systems as opposed to self biomolecules . This is most commonly achieved by specific antitoxins ( e . g . in toxin-antitoxin ( TA ) systems ) and immunity proteins ( e . g . in polymorphic toxin and related systems ) that neutralize the effector when not required ( Leplae et al . , 2011; Anantharaman and Aravind , 2003a; Kobayashi , 2001; Daw and Falkiner , 1996; Zhang et al . , 2012; Zhang et al . , 2011 ) . However , this problem is even more acute in cases where the biological conflict is between genomes in the same cell , such as those between the host genome and invasive nucleic acids like viruses and plasmids . In the simplest cases , the solution takes the form of a mechanism to distinguish self vs . non-self , with the most well-known form being the ‘marking’ of self DNA by modification of the bases or the backbone by enzymes , which are coupled to restriction enzyme effectors that target nucleic acids based on the differences in their modification status . This is the basic principle behind the diverse restriction-modification ( R-M ) systems ( Leplae et al . , 2011; Anantharaman and Aravind , 2003a; Kobayashi , 2001; Daw and Falkiner , 1996 ) . Another mechanism , typical of the CRISPR/Cas and Piwi-dependent systems , is the direct detection of non-self nucleic acids through complementary pairing mediated by guide nucleic acids ( O'Connell et al . , 2014; Ameres et al . , 2007 ) . However , these conflicts with invasive nucleic acids often present more complex decision points . For instance , a cell might sacrifice itself via apoptosis or cell suicide if it can thereby prevent the genetic furtherance of the antagonistic or infectious entity replicating within it ( Makarova et al . , 2012 ) . While this is fitness-nullifying for the cell , the behavior might still be genetically profitable if one were to consider included fitness accrued via kin benefiting from the suicidal act ( Bourke , 2014; Hamilton , 1964 ) . However , such a suicidal action , wherein the effectors are deployed against self molecules , needs to be tightly regulated to prevent its inappropriate activation when not required . Moreover , effector-production and -deployment themselves divert resources from processes central to actual organismal proliferation . Hence , the initiation of effector actions leading to dormancy or even suicide need to be weighed against alternative house-keeping functions leading to proliferation . Given these stakes , it is becoming increasingly clear that the effector deployment step in conflict systems is often subject to threshold detection mechanisms ( Burroughs et al . , 2015 ) that limit the serious energetic or even existential consequences of unintentional effector deployment . Our recent studies and follow-up wet-lab confirmations have led to the uncovering of major thresholding mechanisms that control key biological conflict systems deployed by cells against invasive nucleic acids such as viruses ( Burroughs et al . , 2015; Severin et al . , 2018; Whiteley et al . , 2019; Makarova et al . , 2014 ) . Chiefly , these include the use of a diverse array of linear and cyclic nucleotides in a subset of the CRISPR/Cas and the SMODS systems as control mechanisms regulating deployment of the effectors of this system . Briefly , detection of the non-self or invasive entity activates one of multiple evolutionarily unrelated nucleotidyltransferases that then produce a nucleotide signal . It is the detection of this nucleotide by a sensor domain that results in actual effector-deployment ( Burroughs et al . , 2015; Severin et al . , 2018; Whiteley et al . , 2019 ) . Thus , the additional step of nucleotide production serves to set a threshold for the deployment of potentially dangerous effectors . Our studies had also revealed further complexities to the thresholding mechanism in the form of the deployment of chaperones of the AAA+ ( ATPases associated with diverse cellular activities ) ATPase superfamily of P-loop NTPases . These ATPases channel energy generated by ATP hydrolysis to drive conformational changes ( Neuwald et al . , 1999; Lupas and Martin , 2002; Hanson and Whiteheart , 2005 ) . Consequently , they have chaperone activity which helps in the ( dis ) assembly of protein complexes involved in the effector response . Thus , they can act as thresholding ‘switches’ , which regulate the conformation or state of assembly of the effector complexes . We identified such a mechanism as a further regulatory element of certain nucleotide-dependent conflict systems , which has subsequently been confirmed in some wet-lab studies ( Ye et al . , 2019; Lau et al . , 2020 ) . Here , the TRIP13/Pch2 AAA+ ATPase along with the peptide-binding co-chaperone HORMA domains ( Aravind and Koonin , 1998a ) facilitates conformational switching to set a threshold for response to invasive DNA ( Rappuoli and Montecucco , 1997; Burroughs et al . , 2015; Wojtasz et al . , 2009 ) . More generally , we also observed other examples of the use of chaperones in thresholding mechanisms: in the 3-gene toxin TA system , in addition to the toxin ADP-ribosyltransferase ( ART ) , an antitoxin which is an intrinsically disordered protein , there is the third component , the chaperone SecB that regulates the folding of the antitoxin and affects TA interactions and activity ( Aravind et al . , 2015; Sala et al . , 2013 ) . The classical NtrC-like AAA+ ATPases which regulate transcription might also be seen as part of a threshold-setting control mechanism ( Banerjee et al . , 2019; Hsieh et al . , 2018 ) . Our preliminary investigations suggested that such mechanisms might be more widespread than have been appreciated . Hence , we developed a systematic procedure to identify novel chaperone-based systems and their analogs that regulate effector activity in biological conflicts via thresholding . Consequently , we identified multiple systems , unifiable by several shared mechanistic features , which we group into three categories: ( 1 ) chaperone/co-chaperone-based systems centered on conformational changes driven by a MoxR-like AAA+ ATPase and its co-chaperone , the peptide-binding von Willebrand factor A ( vWA ) domains ( Snider and Houry , 2006; Wong and Houry , 2012 ) ; ( 2 ) analogous systems which use GTPases ( in some cases with further HSP70 and tubulin family proteins ) instead of the MoxR-like AAA+ ATPase ( Leipe et al . , 2002; Nogales et al . , 1998 ) ; ( 3 ) systems which share effectors with the first two but have distinct thresholding mechanisms , such as those dependent on proteolytic cascades . Across these systems , we observe associations to further conflict systems which can act within their framework of or in parallel with these systems , including association with a diversity-generating retroelement ( DGR ) ( Wu et al . , 2018 ) , prokaryotic Ubiquitin-like ( UBL ) conjugation systems ( Burroughs et al . , 2011; Iyer et al . , 2006 ) , and a novel class of adaptor domains analogous to the Death-like domains of animal-apoptosis systems . Importantly , we show that these conflict systems are predominantly found in bacteria with complex development and multicellular organization suggesting a link between these features and the strongly regulated effector responses to invasive entities . This is also paralleled in multicellular eukaryotes and helps uncover the general evolutionary principles in the interplay between immunity and developmental complexity . To systematically identify novel conflict systems with chaperone-based threshold-setting regulatory mechanisms or their analogs , we drew on the extensive , well-annotated collection of protein domains typical of biological conflict systems . This set includes an array of over 250 effector domains such as diverse families of RNases , DNases , protein/nucleic-acid-modifying enzymes and pore-forming toxins . This collection was assembled from previous systematic analyses of biological conflict systems , such as polymorphic toxins , CR-toxins , TA , R-M , CRISPR/Cas , and nucleotide/small-molecule-activated effectors from previous studies performed by us and others over the past two decades ( Aravind et al . , 2012; Burroughs et al . , 2015; Iyer et al . , 2017; Zhang et al . , 2012; Aravind et al . , 2015; Zhang et al . , 2016; Makarova et al . , 2019 ) . As has been previously noted , such effector domains are often shared across distinct conflict systems and thus served as potential markers for the detection of new conflict systems ( Figure 1A; Aravind et al . , 2012; Burroughs et al . , 2015; Iyer et al . , 2017; Zhang et al . , 2012; Aravind et al . , 2015; Zhang et al . , 2016; Makarova et al . , 2019 ) . We used profiles of these domains as seeds in PSI-BLAST searches ( typically run up to five iterations ) run against a curated collection of 7423 complete genomes ( coding for a total of 21646808 proteins ) to obtain an initial collection of proteins potentially involved in biological conflicts . Further , selected searches were run against the NCBI nr database ( June 21st , 2019 ) to obtain a more complete coverage of rarer components and survey the frequency of such systems among currently deposited genomic data more generally . Proteins in genuine conflict systems typically show: ( 1 ) high architectural variability with repeated displacement of effectors within an otherwise well-preserved domain architectural or operonic context ( Aravind et al . , 2012; Iyer et al . , 2011a; Van Melderen , 2010; Leplae et al . , 2011; Burroughs et al . , 2015 ) ; ( 2 ) certain parts of these molecules , usually those encompassing the effector segments that directly interact with the rival entity tend to show high variability which can be measured using Shannon entropy of the sequence alignments . These above features can be observed even between closely related organisms ( Zhang et al . , 2012; Zhang et al . , 2016; Krishnan et al . , 2018 ) ; ( 3 ) Show a high degree of lateral transfer , gene loss and a tendency for extensive difference in presence and absence between closely related organisms ( Van Melderen , 2010; Leplae et al . , 2011; Zhang et al . , 2012; Iyer et al . , 2014 ) . We leveraged this knowledge to cull a subset of promising candidates displaying these features from the initial collection ( Figure 1A ) . We then extended the previous searches to more extensive databases as appropriate ( see Materials and methods ) . We extracted gene-neighborhood information for these candidates and queried the proteins coded by conserved associated genes for chaperone domains using a panel of PSI-BLAST position-specific score matrices ( PSSMs , also called sequence profiles ) for different classes of chaperones . We also followed these up with parallel searches using hidden Markov models instead of PSSMs with JACKHMMER and HMMSEARCH programs from the HMMER3 package ( Figure 1A ) to detect domains which might have eluded the first approach . By this procedure we identified four distinct prokaryotic systems with a gene-pair respectively encoding a MoxR-type AAA+ ATPase and its co-chaperone the vWA domain , along with at least one other linked gene coding for a further component of these systems . Hence , we termed all these ‘the ternary systems’ as they included at least three core components ( Figure 1A ) . In these systems the variable effector domain ( s ) were either fused directly to the vWA domain or encoded as further components of the system . Further , we found another class of systems which combined genes coding for one or two paralogous GTPases of the TRAFAC clade with tubulin homologs and a HSP70 family chaperone protein ( Figure 1A ) . Finally , having identified these systems , we also searched for analogous and related systems by investigating if any of the core components are displaced by alternative components or have been lost ( Figure 1A ) . This led us to identify a more extensive set of GTPase-centered systems wherein the tubulin and HSP70 components had been lost . In addition , this procedure also allowed us to identify analogous systems wherein the chaperone components had been displaced by other regulatory components such as predicted peptidase cascades . A subset of each these classes of systems included several other linked genes beyond the core components , which were either additional effectors or other sensory or regulatory components . We could categorize the final set of systems into three broad categories: ( 1 ) The MoxR-vWA-centric ternary systems; ( 2 ) GTPase-centric systems with or without the tubulin and HSP70 components; ( 3 ) systems with other regulatory components such as peptidase cascades ( Figure 1A ) . Below we describe in detail the systems belonging to each of these categories along with the special ‘grammatical’ features of each of them . These systems are characterized by a core chaperone-co-chaperone pair of the AAA+ MoxR protein and a vWA domain protein . There are four distinct systems with these components: ( 1 ) the VMAP systems: these are characterized by presence of a third distinct conserved protein we termed the vWA-MoxR associated protein ( VMAP ) . These are by far the most frequent in the nr database . ( 2 ) The inactive STAND ( iSTAND ) NTPase systems: other than vWA-MoxR couple , these possess a catalytically inactive version of the STAND subfamily of AAA+ NTPases as the third component . ( 3 ) The FtsH-containing systems . These feature a further AAA+ ATPase of the classical AAA+ clade , FtsH , in addition to the core chaperone-co-chaperone pair . ( 4 ) The β-propeller-containing systems . These systems are characterized by a β-propeller domain fused to the vWA component . Beyond the ternary core a subset of each of these systems contain several additional effector and regulatory components . We first discuss the two shared components of these systems: the MoxR AAA+ and the vWA protein and then discuss the unique features of each of the four systems separately . The MoxR clade of AAA+ domains is defined by a distinctive insert in helix-2 of the core AAA+ domain ( Figure 1B ) and includes the dynein-midasin , MoxR , YifB and chelatase sub-clades . Across these sub-clades , with the apparent exception of dynein , the association with a vWA domain co-chaperone is observed ( Snider and Houry , 2006; Wong and Houry , 2012; Iyer et al . , 2004a ) . The vWA domain adopts an α/β Rossmannoid fold with six strands and has two conserved aspartates at the termini of the 1st and 4th core strands with which it typically binds a divalent metal ion like Mg2+ . This Mg2+-binding site of the vWA domain is part of the so called ‘MIDAS motif’ , which mediates the binding of an extended or unstructured peptide ( Lee et al . , 1995; Figure 1C ) . Most of vWA domains found in the ternary systems we report here have the two aspartates and are likely to bind a peptide in a comparable manner . A subset of the vWA domains from the FtsH- and β-propeller- containing systems , however , appear to lack one or both aspartates ( see Figure 1—source data 1 ) . Hence , it is possible that they have acquired an alternative mechanism of peptide recognition . The MoxR-vWA coupling has been extensively studied in several systems , such as: ( 1 ) the Midasin system involved in ribosomal protein-binding during pre-60S ribosome assembly in eukaryotes ( Chen et al . , 2018 ) ; ( 2 ) Archaeal phage-tail assembly system ( Scheele et al . , 2011 ) ; ( 3 ) the RavA-viaA system involved in the assembly of the fumarate reductase respiratory complex and the lysine decarboxylase complex in certain bacteria ( Wong et al . , 2017; El Bakkouri et al . , 2010 ) ; ( 4 ) the YebA-coupled system likely involved in peptide-degradation ( Vollmer , 2012 ) ; ( 5 ) the MoxR-MoxC/L and CoxD-CoxE systems involved in activating the insertase required for metal-cluster insertion in proteins ( Wong and Houry , 2012; Pelzmann et al . , 2009; Fuhrmann et al . , 2003; Pelzmann et al . , 2014 ) ; ( 6 ) the Mg2+-chelatase involved in insertion of Mg2+ into porphyrins ( Lundqvist et al . , 2010; Fodje et al . , 2001 ) . While these systems act in disparate biological contexts , a common biochemical denominator across all these is the capture of a peptide in the substrate by the vWA domain via its peptide-binding site , followed by ATPase action of the MoxR ATPase that transmits a conformational change via helical and flexible charged linker regions to the vWA component and the protein bound by it ( Snider and Houry , 2006 ) . This results in a conformational change in the latter with a consequent altered folding and release of the substrate: this process thereby contributes to macromolecular complex assembly ( Wong and Houry , 2012 ) . In structural terms , known structures of vWA-MoxR pairs suggests that the MoxR AAA+ domain forms a hexameric ring typical of this superfamily of ATPases ( Wong and Houry , 2012; Maisel et al . , 2012 ) . The MoxR ATPases found in the conflict systems under consideration possess the characteristic arginine finger N-terminal to the core strand-5 ( Figure 1B ) , which is associated with ring-oligomerization ( Wong and Houry , 2012 ) ; hence , they are likely to form a similar hexameric ring as other characterized MoxR proteins . While vWA co-chaperone domains can also form an oligomeric ring , their stoichiometry appears to differ with respect to the AAA+ units from system to system ( Chen et al . , 2018; Scheele et al . , 2011; Lundqvist et al . , 2010; Liu et al . , 2008 ) . Hence , it is unclear how many vWA units might associate with the MoxR ring in the systems we describe here . Based on the parallels to the known vWA-MoxR systems , we propose that even in these systems the vWA domains capture a peptide from a substrate protein and the ATP-dependent activity of the associated MoxR protein serves to release the bound substrate protein in an altered conformation . Given that these ternary systems are characterized by the further presence of at least one other protein in the complex , it is likely that this action of the vWA-MoxR pair helps restructure the complex formed with this additional component . These systems display an unusual phyletic pattern being widely present in the actinobacteria , cyanobacteria , planctomycetes , and chloroflexi , and somewhat sparsely in alphaproteobacteria , gammaproteobacteria and deltaproteobacteria ( Figure 2 ) . Strikingly , there is a highly significant association between the presence of these systems and the presence of a multicellular or colonial or aggregate form ( such as rosettes of planctomycetes and cooperating intracellular bacteroid aggregates with branching structures in rhizobia ) in the life cycle of the organisms that possess them ( Figure 2 , Figure 2—figure supplement 1 , Table 1 ) . This is particularly notable in the lineages where they are sparsely found such as deltaproteobacteria , where they are only found in the multicellular myxobacteria or gammaproteobacteria , and where they are found in Beggiatoa , Thiohalocapsa and Lamprocystis which show multicellular or aggregative features ( Lyons and Kolter , 2015; Kysela et al . , 2016; Figure 2—figure supplement 1 ) . The three core genes of this system show a strict preservation of gene order: from 5’ to 3’ , the gene coding for the VMAP is followed by that for the MoxR , in turn followed by that for the vWA component . Comparison of the phylogenetic trees of these three components show a high degree of congruence across these systems ( Figure 3A ) . We observed that 36% of the organisms with VMAP ternary systems code two or more distinct systems in their genomes each with their own set of three core components ( Figure 1—source data 1 ) . A record number of six paralogous systems are encoded by the actinobacterium Streptomyces davawensis JCM 4913 . Together , these features suggest that the three core components of this system are under a strong selective constraint to function as a ternary complex with their own cognate partners and with a particular order of subunit assembly predicated by the conserved gene order . By analyzing the mean positional Shannon entropy across the alignment of the three core components , we observed that the vWA and VMAP components are evolving significantly more rapidly even between closely related species when compared to other conserved proteins in the respective genomes ( e . g . replicative DNA polymerase β subunit ) ( Figure 1E ) . This is a characteristic feature that supports involvement of these systems in biological conflicts . The MoxR-like proteins in these systems have a unique , poorly structured region N-terminal to the AAA+ domain found in no other members of the MoxR clade: this region displays two conserved motifs with multiple highly conserved aromatic residues and one nearly absolutely conserved arginine ( respectively of the form: WxhapG and PsWRxa , where x is any amino acid , h is hydrophobic , a is aromatic , p is polar and s is small ) , and might adopt an extended conformation ( Figure 1B ) . The vWA domain is flanked on both ends by well-conserved α-helical extensions with several strongly conserved residues . The N-terminal extension is typically separated from the vWA domain by a poorly structured linker region and shows a distinctive [D/E]xxx[D/E]xxxa motif . The C-terminal extension domain shows conserved S , E and R residues ( Figure 1C ) . Beyond this C-terminal extension domain , the vWA components of these systems are marked by extraordinary variability , being fused to a succession of multiple further C-terminal globular domains . These can range in number from 1 to 11 distinct domains with an average of three domains , and they often greatly differ even between closely related organisms ( Figure 1F ) . This feature again reinforces the involvement of these proteins in biological conflicts . Consistent with this , a detailed analysis of the domains in this region revealed parallels to those observed in other conflict systems ( Aravind et al . , 2012; Burroughs and Aravind , 2016; Burroughs et al . , 2015; Iyer et al . , 2017; Zhang et al . , 2012; Anantharaman et al . , 2012 ) , suggesting that they are likely to act as effector domains with a diverse range of biochemical targets ( Figure 3A–B; see below ) . The coupling of biochemically diverse effector domains is reminiscent of the pattern observed in the recently described polyvalent proteins deployed by phages and plasmids against their hosts ( Iyer et al . , 2017 ) . The third conserved component of these systems , the VMAP , is the most distinctive and it thus far not found outside of these systems . It has a tripartite modular organization ( Figure 3A–B ) . Its most conserved part is a clearly distinguishable C-terminal domain ( hereinafter VMAP-C ) , which is predicted to adopt a secondary structure with mixed α+β elements ( a core of 7 β-strands and 6–7 α-helices ) ( Figure 1D ) that could not be unified with any previously known domains . It features several , nearly universally-conserved polar residues ( Figure 1D ) , suggesting a conserved interaction interface . While it cannot be entirely ruled out , there are no clear indications of an enzymatic function for VMAP-C . The central region of the VMAPs is highly variable and using sequence similarity-based clustering we were able to identify 29 distinct versions found in at least two or more distinct VMAPs ( named VMAP-M0-28 ) ( Figures 3A–B and 4 ) . Almost all of them are predicted to adopt an all-α-helical secondary structure , suggesting that they could all be rapidly diverging variants of a common α-helical fold ( Figure 1—source data 1 ) . They occur in widely different frequencies , with VMAP-M0 being the most common , occurring in 75% of the total proteins ( 1482 ) in our dataset , with the rest occurring much less frequently . The N-terminal region of the VMAPs features one or more of 17 distinct domains belonging to three disparate classes ( Figures 3B and 4B ) : The enzymatic domains are strongly mutually exclusive of each other in VMAP architectures , but multiple EADs can occur in the same VMAP protein ( Figure 1—source data 1 ) . As a result of the different combinations of the above domains , we found a total of 71 distinct VMAP domain architectures from across 1482 distinct ternary systems of this type ( Figure 3B ) . We found a total of 163 distinct domain-architectures of the vWA component featuring 86 distinct C-terminal domains in our collection of 1482 VMAP ternary systems ( Figures 1F and 3A–B ) . A comparison of these ternary systems reveals an overarching theme: While about 15 effector domains are repeatedly observed , the rest tend to be rare and are found in less than 3% of the architectures . Comparisons between closely-related organisms suggests that this pattern emerges from effector domains continually displacing existing ones or accreting to existing architectures ( Figure 3A ) . To better understand the action of these systems , we categorized the effector domains according to their biochemical functions . We briefly describe each class below ( see also Figure 1—source data 1 ) : We next investigated if the domain architectural patterns of the C-terminal domains associated with the vWA component might throw light on the functions of these systems . We observed that when more than one VMAP ternary system is encoded by an organism , in the majority of the cases the C-terminal domains coupled to the vWA are different in each of the paralogous systems ( Figure 1—source data 1 ) . This suggests that the systems are selected for the diversity of their potential interactions . To examine the combinations of different effector domains in greater detail , we constructed an architectural network using the vWA components from a set of 1482 distinct ternary systems ( Figure 4A ) . It revealed that combinations of effector domains belonging to different functional categories are frequently observed . For instance , the RNase PIN domain might be combined in the same polypeptide with the phosphopeptide binding FHA domain or the NACHT NTPase domain while the NaeI REase domain comes with the cNMP cyclases ( Figure 4A ) . In cyanobacteria , the FGS domain is combined with at least 10 other functionally diverse domains ( Figure 4D ) ; however , there is a strict architectural syntax with all these additional domains always occurring to the N-terminus of the FGS domain ( Figures 3B and 4A , D ) . Taken together , these observations suggest that disparate domains might be coupled together in the vWA component in order to mediate a multiplicity of distinct interactions at the same time as observed in the recently-described polyvalent toxin systems ( Iyer et al . , 2017 ) . Interestingly , a plot of the length distribution of the vWA component pointed to certain preferred lengths: it shows a multimodal distribution with peaks around 850 , 1125 , 1300 and 1450 residues ( Figure 1G ) . The first peak is dominated by proteins from cyanobacteria while the remaining three are enriched in actinobacterial proteins ( Figure 1—figure supplement 1 ) . A closer examination revealed that this pattern arises from certain persistent and over-represented architectures . For example , we found the cNMP cyclase domain to be frequently coupled to N-terminal trypsin-like and C-terminal Band seven domains . The Band seven domains in turn tend to be strongly coupled to a C-terminal HAD domain . Similarly , the S/T/Y kinase domain tends to be coupled with either a C-terminal MinD/ParA domain or a wHTH domain ( Figure 4A , Figure 1—figure supplement 2 ) . Hence , it is conceivable that in these cases , despite the domains being functionally disparate , their action might be coordinated during effector deployment . While multiple second-messenger-generating and peptidase domains are shared by the vWA component and the VMAP component , they do not necessarily co-occur in the same systems . For instance , whenever there is a trypsin or a cNMP cyclase domain associated with the vWA component , they are never or rarely found associated with the cognate VMAP component . However , when there is a caspase-like domain in the vWA component , they are nearly always accompanied by a caspase-like domain fused to the VMAP component and its associated α/β-hydrolase domain . This suggests that , at least in part , the same domain has a different functional significance depending on the component in which it is found . These form the second class of MoxR-vWA ternary systems , which are distinguished from the above-described VMAP ternary systems in having a distinct third component in lieu of the VMAP . Using sequence searches as well as profile-profile comparisons , we were able to show that the conserved core of this third component is an inactive version of the STAND NTPase domain ( Leipe et al . , 2004; Saraste et al . , 1990 ) : accordingly , we call this protein the iSTAND protein ( Figure 1—source data 1 ) . While the total number of these systems are far fewer than the VMAP systems , paralleling the latter systems they show a statistically significant presence in organisms with a multicellular state in their life cycle ( Figure 2 , Figure 2—figure supplement 1 , Table 1 ) . Notably , they are concentrated in cyanobacteria , proteobacteria and bacteroidetes while being almost entirely absent in actinobacteria ( Figure 1—source data 1 ) . Thus , they are the dominant ternary systems in multicellular forms among proteobacteria and bacteroidetes where the VMAP systems are rarer or absent . In terms of their core components , the iSTAND system MoxR lacks the specific N- and C-terminal extensions seen in the VMAP systems ( Figure 1B ) . The vWA components of these systems tend to have a 4-helical N-terminal domain that can be unified with the 4-helical N-terminal domain found in the CoxE-type vWA-proteins ( Pelzmann et al . , 2014 ) . Another distinctive feature of these systems is the presence of transmembrane I segments at the N-terminus in about 28% of examples of this system ( Figure 5A ) . Here too , the C-terminal region of the vWA component is highly variable and features one or more effector domains ( Figure 1—source data 1 ) . Paralleling the VMAP systems , when multiple distinct iSTAND ternary systems are encoded by the same organism , the vWA-associated effector domains tend to be distinct in each of them ( Figure 1—source data 1 ) . The number of effector domains found in these systems are fewer than in the VMAP systems and the most common one is the FGS domain , which is found in about 22% of the examples ( Figure 5A ) . Beyond this , the superstructure forming domains , caspase-like peptidase domains and the TIR domain are common with the VMAP systems ( Figures 3B and 4A ) . Instead of the AP-GTPases , these systems feature a GTPase effector domain of the AIG subfamily of the GIMAP-Septin-like clade ( Leipe et al . , 2002 ) , which are implicated in several immunity-related processes in eukaryotes ( Poirier et al . , 1999; Reuber and Ausubel , 1996; Figure 5A ) . Importantly , these systems differ from VMAP ternary systems in possessing several peptidoglycan-associating effector domains such the PEGA , OmpA , PASTA , and Sel1-repeat domains ( Yeats et al . , 2002; Bouveret et al . , 1999 ) that are typically preceded by a TM segment ( Figure 5A ) . Together with the N-terminal TM segments , these features imply that , unlike the VMAP systems , a subset of iSTAND systems are likely to function proximal to the membrane and in some cases interact with the cell-wall ( Figure 1—source data 1 ) . This might also relate to their phyletic patterns , that is the concentration in organisms with Gram-negative cell walls ( Figure 2 ) . While the iSTAND domain of these systems could not be unified with the VMAP-C , the iSTAND component shows clear architectural parallels to the VMAP . Like in the VMAPs , either nucleotide-derived second messenger-generating domains ( TIR or DRHyd ) or peptidase ( caspase-like and trypsin-like ) or 4 of the EAD domains found in the VMAPs are also fused to the N-termini of the iSTAND domains ( Figure 5A; see below ) . This points to a similar mechanism of action for these domains in conjunction with the other components of the system . We found a distinct α-helical domain at the N-termini of a subset of the iSTAND proteins . By means of profile-profile searches , we were able to unify this domain with domains such as Death , DED , CARD and Pyrin which display the Death-like fold and function in animal apoptotic systems ( Liu et al . , 2003; Park et al . , 2007; Aravind et al . , 1999; Figure 5A ) . This is the first time a domain with the Death-like fold has been found outside of animals; this bacterial Death-like domain ( bDLD1/EAD3 ) might function similar to the Death-like domains in animal apoptosis ( see below ) . The third distinct class of MoxR-vWA-centric ternary systems , like the previous two , show a statistically significant presence in organisms with a multicellular stage in their life cycle from deltaproteobacteria , actinobacteria and planctomycetes ( Figure 2 , Figure 2—figure supplement 1 , Table 1 ) . Unlike the iSTAND and VMAP systems , they are rare in cyanobacteria ( Figure 1—source data 1 ) . This class is distinguished by the presence of a third conserved component with two copies of the FtsH-type AAA+ ATPase domain , with the N-terminal domain being catalytically inactive with loss of the P-loop and Walker B motif and the second domain possessing all the characteristics of active FtsH-type ATPases ( Okuno and Ogura , 2013; Figure 5B ) . Further , a subset of these might have an additional N-terminal domain that is thus far seen only in these proteins . The vWA component of these systems is highly divergent and , in some cases , might even show a loss of the metal-chelating aspartates characteristic of the domain ( Figure 1C , Figure 1—source data 1 ) . These systems can further be divided into two types based on the architecture of the vWA protein which occurs in the approximate proportion of 30% to 70% in the current nr database: The last group of ternary systems are found mostly only in bacteroidetes , planctomycetes , verrucomicrobia and proteobacteria ( Figure 1—source data 1 ) . While there is a significant tendency for these systems to be concentrated in forms with a multicellular habit , these are also found in certain bacteria that are not known to exhibit such an organization , for example plant-pathogenic Pseudomonas and Xanthomonas ( Figure 2 , Figure 2—figure supplement 1 , Table 1 ) . These systems are characterized by a vWA component that strictly possesses a β-propeller domain with about four blades C-terminal to the vWA domain . Further , about 62% of the vWA components are distinguished by a long N-terminal region with an uncharacterized α-helical domain and a C-terminal region with a predominantly β-strand-containing domain ( Figure 5C ) . Remarkably , in several bacteroidetes , we find insertions of the multimerizing ribosomal protein L12/ClpS domain ( Mandava et al . , 2012 ) at different points C-terminal to the vWA domain ( Figure 5C ) . The β-propeller-containing systems are enigmatic in that they show no conventional effector modules . However , their role in biological conflicts is underscored by the presence of a third conserved component with a highly variable N-terminal region ( characteristic of conflict systems ) , which additionally typically has C-terminal TPR and β-propeller regions . This N-terminal variable region , which might function in the capacity of an effector , includes a globular domain and a membrane-spanning segment indicating that these operate close to the cell-membrane ( Figure 5C ) . Given that there are β-propellers in both this third component and at the C-terminus of the vWA component , it is conceivable that it associates with the vWA component via homotypic interactions of these regions . A subset of these systems might feature additional , conserved linked genes coding for proteins that cannot be unified to known protein superfamilies . These proteins are predicted to contain unique domains , one of which is α-helical and others with mixed α+β elements ( Figure 1—source data 1 ) . These systems share some features with the FtsH-containing systems despite lacking the FtsH component: First , both feature divergent vWA domains which might be coupled to β-propeller and other distantly-related domains which do not appear to be the effector module ( Figure 1—source data 1 ) . Second , the MoxR component of both systems often features a peculiar variant of the P-loop motif of the form GXXXTAKS ( Figure 1—source data 1 ) . This suggests that the two likely share a closer common ancestor , which diverged via acquisition of different effector modules and accessory components . Consistent with this , there is a 17% overlap in the organisms containing these two sets of systems ( Figure 1—source data 1 ) . Operons coding for conflict systems often cluster together in the same genomic regions , potentially due to selection for shared regulation and concomitant deployment of these systems ( Burroughs et al . , 2015; Anantharaman et al . , 2012; Burroughs et al . , 2014 ) . Sometimes this linkage is generic and does not persist between different species . However , in several cases , independent systems come together to form larger conserved neighborhoods that might act in a synergistic or an amplifying role ( Burroughs et al . , 2015 ) . We found three distinct systems , which otherwise independently exist elsewhere , coupled to the ternary systems described above . These systems are defined by mobile gene-neighborhoods that are widely distributed across bacteria and are also present in a few archaea; they are strikingly nearly completely absent in cyanobacteria . An organism might possess up to three paralogous versions of these systems in their genomes and 6% of the organisms with such systems have at least two of them ( Figure 1—source data 1 ) . These gene-neighborhoods are analogous to the MoxR-vWA systems in combining genes coding for two conserved regulatory components with highly variable components that bear the hallmarks of effectors ( Figure 6 ) . The two conserved components of these systems are: ( 1 ) GTPases of a previously unrecognized family of the TRAFAC clade ( Leipe et al . , 2002 ) with a conserved glutamate in their Walker B motif . In 38% of these systems they occur as two paralogous copies suggesting that they function as dimers . Accordingly , we named them double-GTPases ( DO-GTPases ) . 60% of the DO-GTPases have 1–2 N-terminal zinc ribbons ( ZnR ) and about 13% of them have N-terminal TM segments . This , together with the presence of one or more TM-containing components in 67% of the systems , suggests that a major fraction of these systems is likely to function in proximity to the cell-membrane ( Figure 6 ) . ( 2 ) A previously undescribed protein , which we call the GTPase-associated protein 1 ( GAP1 ) . GAP1 is typically encoded by a gene downstream of a DO-GTPase gene and in some cases is directly fused to it . This protein is comprised of three clearly distinguishable globular domains , which we named GAP1-N , GAP1-M and GAP1-C as per their position in the protein ( Figure 6 ) . GAP1 occurs in two distinct subtypes that are readily distinguished by versions of the GAP1-N domain ( hereinafter GAP1-N1 and GAP1-N2 ) ( Figure 1—source data 1 ) . Notably , the associated DO-GTPases also show a sub-division into two basic types mirroring the type of GAP1 they occur with . These features indicate co-evolution between the DO-GTPase and GAP1 components and suggests that the GTPase dimer interacts directly with a cognate GAP1 component . Thus , we broadly classify these systems into two types based on the GAP1-N domain ( Figure 6A–B ) . Further , these systems contain either one of two types of mutually exclusive components with a distribution pattern that correlates with the version of the GAP1-N in the system ( Figure 6A–B ) . Those systems with a GAP1-N1 domain ( Figure 6A ) contain a previously uncharacterized , all α-helical protein with a conserved domain which we name the GTPase-associated system helical ( GASH ) domain . Those with a GAP1-N2 domain ( Figure 6B ) contain a protein with rapidly evolving repeats of the fibronectin type-III ( FNIII ) domain ( Little et al . , 1994 ) . Beyond these , the systems contain one or more variable components: in some cases , these occur in the form of variable domains directly fused to the C-terminus of the GAP1 component . In other cases , they are fused to the N-terminus of the FNIII component . Additionally , these variable components might also be encoded by separate genes in the operon ( Figure 6B ) . We discuss these below in the context of the two distinct subtypes of the GTPase-centric systems . These systems are most commonly seen in proteobacteria , bacteroidetes and spirochaetes ( Figure 1—source data 1 ) . Unlike the vWA-centric systems described until now , these show no special association with a multicellular habit in the organisms possessing them ( Figure 2 , Figure 2—figure supplement 1 , Table 1 ) . In addition to genes for GAP1 and the GASH protein , these are characterized by two DO-GTPase genes , with one coding for a version with a N-terminally fused ZnR domain ( Figure 6A ) . Notably , the variable effector domains in these systems almost completely overlap with those found in the VMAP- and iSTAND- MoxR-vWA ternary systems . The most common domain is the calcineurin-superfamily phosphoesterase domain predicted to act as nucleotidyl phosphodiesterase ( Aravind and Koonin , 1998b ) . This is joined by other nucleotide-related effector domains such as the PNPase and the TIR domains ( Figure 6A ) . On several occasions , when these nucleotide-related effector domains are present these systems also contain a gene for a 2TM SLATT domain protein ( Figure 6A ) . This is notable because we had earlier predicted the SLATT domain to be a nucleotide-signal-responsive pore-forming effector in other conflict systems ( Burroughs et al . , 2015 ) . Additional effectors in this system include the peptidases of the trypsin and caspase superfamily , and an α/β-hydrolase domain related to those found in the MoxR-vWA systems . On rare occasions , they may also feature a DNA-glycosylase domain ( Figure 6A ) . The variable effector domains appear to be coupled to the core components of these systems in at least three different ways . First , they can be directly fused to the GAP1 protein ( Figure 6A ) . In other cases , the GAP1 protein contains a C-terminal EAD1 and the effector which is typically encoded by a further gene in the operon is also fused to an EAD1 suggesting that they might be brought together by a homotypic interaction of this domain ( Figure 6A , see below ) . These architectures also suggest that the effectors of these systems are primarily deployed via an interaction with GAP1 , probably mediated by the GAP1-C domain . The third and a frequent configuration combines the effector domain with a previously uncharacterized domain in the same polypeptide ( Figure 6A ) . This small domain is predicted to adopt an α-helical fold and is not found outside of these systems ( Figure 1—source data 1 ) ; hence , we predict that this domain is a likely adaptor domain that directly couples the effector to the GAP1 component . Accordingly , we named this domain the GTPase-associated adaptor domain ( GAAD ) . In contrast to GAP1-N1 gene-neighborhoods , the GAP1-N2-neighborhoods ( Figure 1—source data 1 ) , like most MoxR-vWA systems , show a highly significant association with organisms with a multicellular habit and are primarily found in actinobacteria , firmicutes , planctomycetes , verrucomicrobiae and chloroflexi ( Figure 2 , Figure 2—figure supplement 1 , Table 1 ) . These systems usually contain a GTPase component with two N-terminal ZnRs; if there are two GTPase components one additionally contains four N- and one C-terminal TM helices ( Figure 6B ) . Additionally , some of these are fused to peptidoglycan-binding domains , like the bacterial SH3 , PEGA or β-sandwich domains ( Ponting et al . , 1999; Whisstock and Lesk , 1999 ) , or other ligand-binding ( NTF2 ) and peptide-binding domains ( PDZ [Ranganathan and Ross , 1997] ) beyond the C-terminal TM segment ( Figure 6B ) . Moreover , in these systems GAP1 itself can be frequently fused to C-terminal TMs ( sometimes with further C-terminal cadherin domains ) and other variable components also frequently containing TM segments ( Figure 6B ) . These features indicate a role close to the cell-membrane for the GAP1-N2-type systems , with the extracellular domains suggesting potential interactions with the cell-wall or other extracellular ligands . Only a small number of effector domains of the GAP1-N2 systems are common to the GAP1-N1 systems and predominantly include the SLATT 2TM pore-forming effector domain , which is often fused to the C-terminus of GAP1 ( Figure 6A–B ) . In planctomycetes , we observed some systems with FGS domain effectors which are shared with multiple MoxR-vWA ternary systems , while others were coupled to a gene coding for a protein with a pair of GYF domains ( Kofler and Freund , 2006; Balaji and Aravind , 2007 ) fused to either a trypsin or a TM-rich region ( Figure 6B ) . However , the majority of these systems have their own remarkable array of variable components that are mostly unlike those found in any of the systems discussed thus far . Some of these are fused to the FNIII-repeat component with others being encoded by arrays of linked genes in the neighborhood . These linked genes show lineage-specific patterns that can be divided into three non-overlapping themes . The first , comprising about 52% of all GAP1-N2 neighborhoods , is restricted to firmicutes and actinobacteria , and is centered on genes coding for proteins with tubulin domains ( Poirier et al . , 1999; Reuber and Ausubel , 1996; Figure 6B ) . The firmicutes versions are further distinguished by having two genes encoding proteins with tubulin domains fused to a C-terminal coiled-coil domain . These neighborhoods also contain two genes coding for proteins with vWA domains , one of which is fused to an uncharacterized domain called YtkA , while the other vWA-domain encoding protein has several TM helices with vWA in an extracellular position . Their GAP1 also has a highly-polar C-terminal extension ( Figure 6B ) . The actinobacterial tubulin-containing systems display a single gene coding for a tubulin further fused to an uncharacterized C-terminal α-helical domain . These systems possess two distinct FNIII-repeat-encoding proteins , one fused to the DNA-binding HRDC-domain ( Morozov et al . , 1997 ) and the other is membrane-bound and fused to a vWA domain and an uncharacterized β-strand-rich domain ( Figure 6B ) . Strikingly , the presence of these systems in actinobacteria is mutually exclusive of the VMAP-ternary systems suggesting a possible functional equivalence . The second theme , comprised of about 14% of GAP1-N2 neighborhoods is also predominantly observed in actinobacteria and firmicutes , and displays genes coding for HSP70 and its nucleotide exchange factor GrpE ( Bracher and Verghese , 2015 ) , with a subset of these containing a second gene encoding a protein with only the HSP70 domain sans its peptide-binding domain ( Figure 6B ) . The FNIII-repeat-containing protein in these systems is typically fused to the HSP70 co-factor DnaJ domain ( Mayer and Gierasch , 2019 ) or to TPR repeats ( Figure 6B ) . The third theme , seen in 10% of GAP1-N2 neighborhoods is restricted to actinobacteria . This features an association with a catalytically active protein kinase domain often fused to a zinc ribbon and a β-strand-rich domain , or to TM helices . The gene coding for the protein kinase is usually linked to adjacent genes coding for a catalytically-active protein phosphatase domain of the PP2C family and a standalone vWA domain , respectively ( Figure 6B ) . The systems point to regulated assembly of complexes akin to cytoskeletal structures and potential peptide-interactions via the vWA domains . We first uncovered 10 distinct EADs while analyzing the VMAP-ternary systems ( see above ) , where they are typically fused to the VMAP component ( Figures 3B and 4B ) . Subsequently , we found a subset of them occurring in two of the remaining MoxR-vWA-centric ternary systems and the GAP1-N1-type GTPase-centric systems ( Figures 3A , 4B , 5A–B and 6A ) . The EADs are typified by the following distinct features: An organism typically encodes multiple proteins with the same EAD both linked to the primary systems and also elsewhere in the genome . The domains fused to them expand diversity of associated effector and signaling domains beyond those linked directly to the core systems . A record number of 22 distinct proteins with EAD1 are encoded in the genome of Frankia inefficax ( Figure 1—source data 1 ) , while the cyanobacteria Mastigocoleus testarum BC008 and Scytonema hofmannii PCC 7110 both contained 19 each . In the case of EAD2 , the next most prevalent EAD , Streptomyces davaonensis JCM 4913 encodes eight paralogs . In cyanobacteria , EAD1-linked effectors add a considerable variety of effector and signaling domains to the ternary systems beyond the FGS domain ( Figure 7A ) . The two most prevalent EADs , EAD1 and EAD2 , show clear differences in phyletic patterns and architectural associations ( Figure 1—source data 1 ) : whereas EAD1 is found widely across bacteria with the above-described of systems , EAD2 is primarily found in actinobacteria and to a lesser extent in cyanobacteria . Whereas EAD1 is primarily linked to effector domains that also occur directly fused to the vWA component in MoxR-vWA ternary systems or GAP1 in the case of the GAP1-N1 GTPases systems , EAD2 occurs fused to the signaling peptidases and nucleotide-utilizing domains fused to the VMAP component of those ternary systems ( Figure 7A ) . Thus , it appears that EADs like EAD1 ( also bDLD , EAD4 , EAD5 , EAD7 , EAD8 , EAD10 ) primarily recruit other effectors to the systems , while those like EAD2 ( also EAD6 , EAD9 ) primarily recruit the signaling components of the system . Consistent with this distinction , there are multiple cases where the same VMAP component might contain both EAD1 and EAD2 fused to it ( Figure 7A ) . Notably , the duplicate presence of EADs in both the core component and the stand-alone components , the closer relationship to lineage-specific paralogs and the fact that the same effector or signaling domains might be directly fused to the core components in other cases , leads to the proposal that EADs are coupling adaptors . Thus , they are predicted to bridge effector and signaling domains to core components of different systems via homotypic interactions between themselves ( Figure 7A ) . A precedent for such a mechanism is presented by the animal apoptotic systems where α-helical domains ( comparable to some of the EADs ) of the Death-like superfamily ( e . g . the Death domain , CARD , DED and Pyrin ) have been implicated in homotypic interactions that result in polymeric effector complexes or oligomerization for the activation of the caspase effector domains fused directly to CARD domains ( Park et al . , 2007; Kao et al . , 2015 ) . Remarkably , as noted above , we found the first bacterial cognates of the Death-like superfamily ( bDLD1/EAD3 ) occupying an equivalent position as the other EADs described here ( Figures 5A , 7A and 8A ) . The implications of these parallels are elaborated further below in context of a mechanistic interpretation of these systems . The observation that the EADs show a strong coupling , either in the same gene-neighborhood or in the same genome to diverse core systems with distantly or unrelated core components prompted us to investigate if they might lead us to other such systems . By systematically examining the domain-architectures and neighborhoods of the EAD-containing proteins we found three other potential conflict systems defined by characteristic gene-neighborhoods , which might occasionally or frequently utilize EAD1 or EAD2 . We briefly describe these systems below . These systems are mostly found in alphaproteobacteria and are again significantly overrepresented in bacteria with a multicellular habit ( Figure 2 , Figure 2—figure supplement 1 , Table 1 ) . They are characterized by two core components , namely a NucA ( Endonuclease NS ) -type endonuclease domain of the HNH superfamily ( Zhang et al . , 2012; Baslé et al . , 2018 ) and a trypsin-like peptidase , in several cases fused to TM segments . Both of these might be found fused together in the same protein or encoded in multiple copies in the same operon ( Figure 7B ) . In some cases , a trypsin domain and a further peptidase domain of the subtilisin ( S8 ) family ( Rawlings and Barrett , 1994 ) might flank the NucA endonuclease domains in the same polypeptide . In a subset of these operons , one of the trypsin paralogs might be fused to an EAD1 domain . In these cases , there are other EAD1-containing proteins either in the same operon or in the same genome and are fused to a variable group of domains that potentially act as effectors . The EAD1-associated domains in the system include a lysozyme- and a papain-like peptidoglycan peptidase domain , pointing to a potential attack on peptidoglycan by these effectors . Certain operons further encode a STAND NTPase component , which is fused to an uncharacterized domain and a trypsin domain at the N-terminus ( Figure 7B ) . Analysis of the uncharacterized domain using profile-profile searches unified it with the Death-like superfamily found in animal apoptotic proteins ( Park et al . , 2007; Wilson et al . , 2009; Martinon et al . , 2001 ) , thereby making it the second family of Death-like domains to be identified in bacteria ( bDLD2 ) ( Figures 7B and 8A–B ) . Given that the organisms possessing versions of these systems with EAD1 encode no other system in the genome with EAD1 , we propose that these represent a novel type of core system that recruits additional effectors via EAD1 . This is supported by the fact that the core components are also observed in related organisms independently of EAD1 . The trypsin-like peptidases and NucA are likely to comprise the essential core of the system that is probably activated via a proteolytic cascade to deploy the NucA endonuclease . The other components like the STAND NTPase with the Death-like domain or those brought in via the interactions of EAD1 are likely coupled effectors . This system is prevalent in actinobacteria and sporadically in proteobacteria with a phyletic distribution that again significantly comports with organisms having a multicellular habit ( Figure 2 , Figure 2—figure supplement 1 , Table 1 ) . The two genes comprising this system are ( Figure 7C ) : Strikingly , we identified up to 50 distinct alternative domains belonging to a wide range of functional categories fused to the C-terminus of the constant caspase domain of the second component: these include many shared with the above systems , such as the peptidases , nucleotide-utilizing domains , signaling enzymes , STAND NTPases , superstructure-forming repeats , and a MoxR-independent version of the vWA domain ( Figure 7C ) . However , we also observed several domains not seen in any of the above systems such as JAB- and Clp-like proteases ( Burroughs et al . , 2011; Kress et al . , 2009 ) , an OMP-decarboxylase-like , SAD modified DNA-binding ( Iyer et al . , 2011b ) , and Elongation Factor Tu domains ( Figure 7C ) . There are also instances of TM-helices with the downstream extracellular periplasmic-binding protein ( PBP ) domain ( Mowbray and Cole , 1992 ) in this position ( Figure 7C ) . The large diversity of domains in this variable region is reminiscent of the C-terminal variable domains of the vWA component in several of the ternary systems ( see above , Figure 3A–B ) and points to a parallel deployment of these domains as effectors . The TM segment of EACC1 is strongly conserved and contains an unusual patch of small ( often glycine ) residues near the center of the helix ( Figure 1—source data 1 ) . This unusual attribute suggests that rather than being a mere membrane-anchoring segment , it might play a role as a membrane-associated sensor that then activates the constant caspase domain of the second component to in turn facilitate effector deployment via auto-proteolysis . The third class of systems are predominantly found in cyanobacteria and were identified by virtue of the linkages of EAD1 to a previously unrecognized domain . In related systems this domain also occurs independently of EAD1 as a constant module linked to a highly variable set of domains such as pentapeptide repeats , a caspase-superfamily peptidase , STAND NTPase and FNIII domains . Thus , these systems exhibit a similar organizational pattern as the above systems with a constant module coupled with a variety of modules that are rapidly varying between closely-related organisms . Through transitive sequence profile and profile-profile searches , we were able to unify the uncharacterized constant domain of these proteins with the TRADD-N domain . The core of the TRADD-N superfamily has an RNA-recognition motif fold with a highly conserved serine in the turn between β−2 and β−3 ( Tsao et al . , 2000; Figure 8B ) . In the animal apoptotic system , TRADD-N serves as an adaptor domain downstream of the tumor necrosis factor receptor and mediates protein-protein interactions ( Tsao et al . , 2000; Micheau and Tschopp , 2003 ) . Thus , like many proteins of the apoptotic machinery , TRADD-N is likely to have had a provenance in bacterial systems with a role in conflict . As these systems have links to other hypervariable systems in eukaryotes , they will be discussed separately elsewhere ( unpublished GK , AMB , LA ) . Although at first sight the above systems might appear rather disparate , they are unified by certain distinctive features that help reconstruct their potential mode of action: Thus , we may infer that the formation of multimeric assemblies is the common denominator of several of these systems: this has several parallels in eukaryotic apoptotic and anti-invader systems in terms of both shared components and regulation ( Figure 8A–B ) . These eukaryotic systems are tightly regulated and kept in an ‘off’ state prior to being triggered by the direct sensing of the invasive entity or apoptotic signal . This usually happens by the direct interaction of the signal ( usually a protein ) with the C-terminal superstructure-forming repeats of STAND NTPases triggering the assembly of the multimeric complex in an NTP-dependent manner ( Danot et al . , 2009; Hofmann , 2019 ) : for instance , in the classic AP-ATPase-based APAF-1 apoptosome , the Cytochrome c released from the mitochondria is sensed by the C-terminal WD40 repeats , which leads to the stepwise assembly of the heptameric apoptosome with the utilization of ATP/dATP by the STAND domain ( Figure 8B ) . This in turn serves as a platform for the assembly of a multimeric caspase proteolytic cascade resulting in apoptosis ( Leipe et al . , 2004; Chaudhary et al . , 1998; Dorstyn et al . , 2018; Zou et al . , 1999 ) . Broadly similar assemblies of AP-ATPases and NACHT-NTPases are also central to the pathogen-triggered hypersensitivity response in plants and pyroptosis/apoptosis in animals ( Hofmann , 2019; Jones et al . , 2016 ) . In the fungal hetero-incompatibility system , NACHT-NTPase proteins with similar architectures trigger programmed cell death when incompatible hyphal types fuse and result in interactions of the variable Het regions of these proteins ( Glass and Dementhon , 2006 ) . In addition to the formation of toroidal structures by the STAND NTPase proteins , homotypic interactions between the CARD domains of the Death-like superfamily also play a key role in in apoptosome-assembly ( Aravind et al . , 1999; Wilson et al . , 2009 ) . Similar interactions are also central to the vertebrate inflammasome network , where the NACHT-family STAND NTPase NLRP3 and the Pyrin domain protein AIM2 trigger oligomerization of the Death-like superfamily CARD and Pyrin domain proteins Mav3 and ASC to result in a prion-like multimeric autoassembly of the ASC protein by Pyrin-Pyrin interactions , which finally induces apoptosis ( Cai et al . , 2014; Masumoto et al . , 1999; Figure 8B ) . In the systems under consideration similar NTP-dependent multimeric assemblies are likely generated by the STAND NTPases and their various analogs ( e . g . the tubulin and HSP70 system coupled to the DO-GTPases ) while the interactions of the Death-like domains are likely paralleled by the structurally comparable EADs and the prokaryotic homologs of the Death-like superfamily ( Figure 8B–C ) . Taken together , these aspects of the effectors from such systems imply that , whereas some are likely to act in a manner comparable to ‘conventional’ conflict systems by directly targeting invasive nucleic acids or proteins , the rest are more likely to function in less-commonly encountered mechanisms that involve: ( i ) decoy interactions to prevent associations between invasive molecules and host proteins , and ( ii ) formation of diverse macromolecular assemblages that might result in cell-death of the host or containment of the invasive entity ( Figure 8B ) . The 4 vWA-MoxR-centric systems , 1 of the two DO-GTPase-centric systems , and the NucA- and EACC1- centric systems ( 7 of the eight above-reported systems ) show a significant association with organisms displaying a multicellular or colonial habit or those with differentiated cell-types in course of their development ( Lyons and Kolter , 2015; Kysela et al . , 2016; Figure 2 , Figure 2—figure supplement 1 , Table 1 ) . The vast majority of these organisms are bacteria belonging to evolutionarily diverse clades ( Figure 1—source data 1 ) . These include: ( 1 ) Bacteria showing vegetative hyphae with multinucleate compartments and aerial hyphae that spawn sporogenic structures ( e . g . Streptomyces [Barka et al . , 2016] ) . ( 2 ) Bacteria forming filaments of multiple cells that might further be enclosed in a gelatinous sheet ( e . g . Nostoc and Beggiatoa ) ( Becerra-Absalón and Tavera , 2009 ) . Further , cyanobacteria show cellular differentiation with up to four distinct cell types in the most complex forms: vegetative photosynthetic cells , hormogonia , which are small motile dispersal filaments , ‘stem-cell’-like akinetes and nitrogen-fixing non-reproductive heterocysts ( Dawkins and Krebs , 1979; Zhang et al . , 2014; Flores , 2012 ) . ( 3 ) Organisms with complex biofilms , sociality , rosette formation ( planctomycetes ) , colony-formation or structure-dependent cell differentiation ( Azospirillum and Archangium ) ( Rodrigues et al . , 2015; Muñoz-Dorado et al . , 2016; Strohl and Larkin , 1978; Wielgoss et al . , 2019; Fuerst , 1995 ) . Notably , the few occurrences in archaea are also in those with known multicellularity ( e . g . Methanosarcina mazei ) ( Mayerhofer et al . , 1992 ) or a tendency to form tight cell-clusters ( e . g . Cuniculiplasma divulgatum [Golyshina et al . , 2016] ) . This strong association with multicellularity and sociality suggests that the distinctness of these systems is likely to be closely related to this organizational aspect of these organisms . Indeed , organisms with copies of more than one system ‘type’ are nearly-exclusively tied to multicellularity ( Figure 1—source data 1 ) . Moreover , some of these organisms , like actinobacteria , are slow-growers but are strong competitors in their environment . Thus , more generally it may be said that many of the prokaryotes with these systems have a more K-selected as opposed to r-selected life history ( Smith , 1998; van Elsas et al . , 2007 ) . The sporadic presence of these systems in the core genome of evolutionarily distant lineages , which is more correlated with multicellularity rather than phylogeny ( Figure 2 ) , suggests that they likely originated in such lineages followed by extensive dissemination by lateral transfer . The presence of multiple paralogs of the ternary and GTPase-centric systems in the same organism , which are not closely related points to accretion of multiple copies of such through lateral transfer with each likely directed at a different set of invasive entities . Further , while comprised of very disparate components , the above-noted thematic convergence is rather strong across these systems . This suggests that they have been repeatedly convergently shaped by comparable selective pressures co-evolving with the emergence of the multicellular state . Emergence of the multicellular or colonial states is predicated by assembly of clusters of cells which are kin ( Bourke , 2014; West et al . , 2006 ) . The invasion of a single cell in such a cluster by an infective invasive entity like a virus presents considerable risk for its spread to the rest of the cells . Being kin , the cells have the benefit of inclusive fitness accrued via other cells in the assemblage ( Bourke , 2014; Hamilton , 1964 ) . Accordingly , when a cell is infected by the invader it pays to ‘sacrifice itself’ for the kin because such an act could limit the infection and still allow the cell to have net positive fitness via its kin ( Bourke , 2014; Hamilton , 1964 ) . Thus , we would expect apoptotic mechanisms coupled with immunity to emerge along with multicellularity ( Lyons and Kolter , 2015; Alteri et al . , 2013 ) . However , since suicide could have fitness-depressing consequences if inappropriately triggered , we would expect that these systems are tightly regulated by threshold-sensing mechanisms which respond to the strength of infection rather than the mere detection of infection . This factor , together with the more K-selected life-histories of the organisms with such systems , are the likely selective forces driving thematic convergence among these systems . A part of this thematic convergence comes from actual sharing of effector domains . We have previously reported extensive sharing of such domains between several distinct types of biological conflict systems ( Iyer et al . , 2011a; Zhang et al . , 2012; Zhang et al . , 2016; Makarova et al . , 2019 ) . Although these systems share some nucleotide-utilizing enzymatic domains and nucleic-acid-targeting enzymatic domains with several other conflict systems ( Burroughs et al . , 2015; Brzozowski et al . , 2019 ) , they generally have their own unique types of effector domains . However , several of these distinctive effector domains , EADs and signaling peptidases and nucleotide-utilizing enzymes are shared between the different types of systems reported in this study – this implies that they constitute their own effector-sharing evolutionary network which captures effectors from each other . However , their core components like the MoxR-vWA pair , DO-GTPase-GAP1 pair or the EACC1-caspase pair are not exchanged and distinguish these systems . Thus , while the MoxR-vWA pair appears to have had its origins in the more widespread CoxD-CoxE systems of bacteria ( Pelzmann et al . , 2009; Pelzmann et al . , 2014 ) , the DO-GTPase-GAP1 or the EACC1-caspase pair have no close relatives elsewhere . The presence of four distinct MoxR-vWA systems suggest that the ancestral CoxD-CoxE system was first exapted as a threshold-setting regulatory switch for pre-existing conflict systems . These then rapidly diversified by the addition of a critical third component such as VMAP or iSTAND ( Figures 3B and 5A ) , which is likely to coordinate the sensing of the invasive entity and further processing of the effectors with the activation of the switch . This resulted in the formation of a ternary core , which then retained its distinctness with only the other remaining variable modules rapidly diversifying and being exchanged between parallel systems . There are diverse origins for the effector modules in these systems . At least two of them appear to have reused sub-systems that occur independently in other conflict systems . First , the FtsH-ternary systems have used a remarkable complete tri-ligase Ubl-system ( Burroughs et al . , 2011; Iyer et al . , 2006 ) . Here , the EAD1 is fused to the Ubl and likely recruits the effectors to proteins to which the Ubl is conjugated via the E1 , E2 and E3 ligase components of the system ( Figure 5B ) . While some of these effectors , such as the caspase-like peptidase , might directly cleave target proteins , the Ubl conjugation might also route it for proteasomal degradation . This Ubl with the fused EAD1 is most likely cleaved off from the larger polypeptide , which it is part of , by the associated JAB-peptidase domain . Ubl systems with different Ubl-ligase complements are also part of other prokaryotic conflict systems , such as a version coupled to the nucleotide-activated effectors in the SMODS systems ( Burroughs et al . , 2015 ) . We also found a Ubl-conjugation system related to those in the FtsH-ternary systems coupled with a Pup-conjugation system in certain bacteria of the Thermofonsia and Planctomycete lineage , lending support to the idea the conjugation of these Ubls might be linked to target-degradation ( Burroughs et al . , 2015 ) . In eukaryotes , too , Ub modifications of effectors and signaling components have been incorporated into different immune and apoptotic responses ( Witt and Vucic , 2017; Tokunaga and Iwai , 2012 ) . Second , the previously-described DGR system for diversifying the FGS domains , which are present in several phages and bacteria , has also been incorporated into these systems ( McMahon et al . , 2005; Le Coq and Ghosh , 2011; Figures 3B and 4A , D ) . They are the primary effectors of the VMAP-ternary system in cyanobacteria . Actinobacteria parallel the cyanobacteria in showing great diversity in the effector modules of the VMAP-ternary systems even in closely-related species ( Figure 3B ) . However , this occurs via the acquisition of a diverse panoply of effector domains rather than the diversification of a single domain through a mutagenic system . Examination of their genomic neighborhoods revealed no association with a DGR element or any other potential mutator element . However , we noticed a statistically significant preferred location for VMAP-ternary systems in the genome at approximately 33% of the length of the chromosome from either end in the linear chromosomes of Streptomycetales ( n = 141 distinct organisms; χ-squared test p=1 . 628e-07 for null hypothesis of uniform as opposed to bimodally clustered distribution across 20 chromosomal intervals ) . These two symmetric preferred locations ( Figure 8D ) bound the chromosomal ‘core’ , which is enriched in the conserved and house-keeping genes ( Hopwood , 2006 ) . As opposed to the core , the arms which lie close to the ends of the chromosome show an enrichment of genes for specialized secondary metabolism and conflict systems as well as transposons and their remnants . The arms are also hotspots of numerous non-homologous end-joining-associated rearrangements such as deletions , circularization and arm exchange ( Hoff et al . , 2018 ) . Accordingly , we posit that the preferred position of the VMAP-ternary system ( Figure 8D ) , especially in actinobacteria with large linear chromosomes , allows them to undergo frequent variations by a recombinational mechanism that affects the chromosome arms . Additionally , the multinucleate nature of hyphae of such actinobacteria ( Hopwood , 2006 ) probably allows recombination between different chromosomes to foster variability of the VMAP-ternary systems . Finally , it is notable that these threshold-regulated systems of prokaryotes find close parallels in apoptotic systems of multicellular eukaryotes , which are triggered by physical interactions with proteins of invasive entities or apoptotic stimuli ( Park et al . , 2007; Aravind et al . , 1999; Wilson et al . , 2009 ) . In terms of direct evolutionary connections , we expand the set of domains that are common to both repertoires to now include the Death-like fold and the TRADD-N domains . This reinforces the earlier proposal that eukaryotic apoptotic systems emerged from component domains acquired via lateral transfer from prokaryotes ( Koonin and Aravind , 2002; Aravind et al . , 2001; Aravind et al . , 1999; Hofmann , 2019 ) . However , it also points to a more subtle feature: in both the eukaryotic and prokaryotic cases , comparable sets of domains appear to have organized themselves into systems based on common principles , which combine invader-sensing with threshold-setting for activation of terminal effectors through multimerization , nucleotide-signals and proteolytic cascades . These are most obviously seen in the convergent domain architectures , such as those of the EADs and the DEATH-like domains , especially in terms of their coupling with various effector components ( Figure 8A ) . This implies that in addition to the basic ‘vocabulary’ in the form of several shared domains there is also a shared ‘grammar’ of the similar architectures and likely interaction networks of these domains across these systems . This in turn suggests that the spread of a relatively small set of protein domains along with the repeated emergence of this ‘grammar’ in their organization has gone hand-in-hand with the multiple emergences of multicellularity across the tree of Life . The above-described systems greatly add to the diversity of themes and effectors observed in biological conflicts . The complex thresholding systems that we reconstruct for the MoxR-vWA and DO-GTPase systems have little precedent in the described universe of biological conflict systems . Based on relatively limited genomic data , we had formerly noted that several components typical of animal , fungal and plant apoptotic systems had their origins in bacteria and are enriched in taxa with a multicellular organization ( Koonin and Aravind , 2002; Aravind et al . , 2001; Aravind et al . , 1999; Hofmann , 2019 ) . The results in this article indicate that this association holds with a much more phyletically extensive dataset and expands to include: ( 1 ) multiple highly regulated systems with similar organizational principles; ( 2 ) additional domains uniquely shared with animal apoptotic systems ( e . g . TRADD-N and Death-like domains ) ; ( 3 ) common principles regarding the coupling of immune responses against invasive entities with programmed cell-death , which is also an important feature of animal and plant immunity ( Jones et al . , 2016 ) ; ( 4 ) Effectors acting through formation of multimeric complexes and superstructures . Accordingly , we propose that the selection for multiple such systems went alongside the emergence of multicellularity and differentiation and that their appropriate regulation is of general importance for the stability of such cellular organizational states . In line with such a proposal , we observed that in these systems there are often two levels of regulation as their faulty deployment could have negative consequences . We propose that the vWA-MoxR pair or the GTPase-GAP1 pair represents the first such level of such regulation , which is coupled to direct sensing of the invasive entity ( e . g . by the VMAP-M domains of the VMAP protein or the FNIII domains of GAP1-N2 ) ( Figures 3B , 5A–C and 6B ) . However , it appears that in many of these systems a second threshold is set in the form of the activation of the associated signaling domains: e . g . the peptidases and the nucleotide-utilizing enzymes . This finally releases the effectors by proteolysis or activates them through a nucleotide-derived signal . Further , in a multicellular assemblage there is much greater need for signaling the presence of the infection and limiting its spread to other cells ( Bourke , 2014; Hamilton , 1964 ) . This explains features unique to these systems: ( 1 ) the presence of coupled systems involved in production of a processed and/or modified peptide which could act as a signal to alert non-infected cells of the presence of an impending infection thereby triggering other immune mechanisms in them ( Figures 3A–B and 5B ) . ( 2 ) The super-structure-forming and multimerizing effectors and EADs , or ‘neutralizing’ effectors like the FGS domains , which could physically interact with the invasive entity and block its spread . Likewise , the effectors with inactive versions of various enzymatic domains could act as decoys to interact with the invasive entities and block their interactions with the active versions that might be their typical targets . ( 3 ) Finally , the presence of several membrane-associated versions among these systems implies that they are geared for the interception of the invasive entities even as they enter the cell . This kind of surveillance would again be useful in protecting the multiplicity of cells that typify a multicellular or social system . We hope that this study inspires further biochemical and ecological investigation of these systems to study the precise stimuli that activate them . We also believe that these systems have potential for biotechnological applications – the diversifying FGS and VMAP-M domains could be used as analogs of antibodies , whereas the dominant negative enzymatic domains could be used to probe cellular functions both in these multicellular bacteria and conventional model systems . Iterative sequence profile searches were performed using the PSI-BLAST ( RRID:SCR_001010 ) and JACKHMMER programs . Similarity-based clustering for both classification and culling of nearly identical sequences was performed using the BLASTCLUST program ( ftp://ftp . ncbi . nih . gov/blast/documents/blastclust . html ) ( RRID:SCR_016641 ) . The length ( L ) and score ( S ) threshold parameters were variably adjusted as needed . For example , the length ( L ) and score ( S ) threshold parameters for clustering near identical proteins was L = 0 . 9 and S = 1 . 89 . HMM searches were run using either HMMsearch initiated with a HMM built from an alignment or iteratively using JACKHMMER from a single sequence . The sequence databases against which these searches were run were: ( 1 ) the non-redundant ( nr ) protein database frozen at June 21 2019 of the National Center for Biotechnology Information ( NCBI ) ( Broderick et al . , 2014; Barnett et al . , 2000 ) ; ( 2 ) this nr clustered down to 50% similarity using the MMseqs program ( Hauser et al . , 2016 ) ( RRID:SCR_008184 ) ; ( 3 ) A custom database of 7423 complete prokaryotic genomes extracted from the NCBI Refseq database . The HHpred program ( RRID:SCR_010276 ) was used for profile-profile searches ( Zimmermann et al . , 2018 ) and run against ( 1 ) HMMs derived from PDB; ( 2 ) Pfam models; ( 3 ) A custom database of alignments of diverse domains curated by our group . For previously known domains , the Pfam database ( Finn et al . , 2016 ) was used as a guide , although the profiles were corrected for their boundaries where required and augmented by addition of newly detected divergent members that were not detected by the original Pfam models . All novel alignments emerging from this study can be accessed in the Supplemental material . Multiple sequence alignments were built by the Kalign ( Lassmann et al . , 2009 ) ( RRID:SCR_011810 ) and PCMA ( Pei et al . , 2003 ) programs followed by manual adjustments on the basis of profile-profile and structural alignments . Secondary structures were predicted using the JPred program ( Cole et al . , 2008 ) ( RRID:SCR_016504 ) . Structure similarity searches were performed using the DaliLite program ( Holm , 2019; Holm et al . , 2008 ) ( RRID:SCR_003047 ) . Structural visualization and manipulations were performed using the PyMol ( http://www . pymol . org ) program ( RRID:SCR_000305 ) . Contextual information from prokaryotic gene neighborhoods was retrieved using a Perl script that extracts the upstream and downstream genes of the query gene from the GenBank genome file and uses BLASTCLUST to cluster the proteins to identify conserved gene-neighborhoods . Recognition of gene neighborhood conservation relied on several filters including: ( 1 ) nucleotide distance constraints ( generally 70 nucleotides ) ; ( 2 ) conservation of gene directionality within the neighborhood; ( 3 ) presence in more than one phylum . Phylogenetic analysis was conducted using an approximately maximum-likelihood method implemented in the FastTree 2 . 1 ( RRID:SCR_015501 ) program under default parameters ( Price et al . , 2010 ) . Perl scripts were used to run automated large-scale analysis of sequences , structures and genome context as described above . Network analysis ( igraph and circlize [Gu et al . , 2014] packages ) , data processing ( knitr ( Xie , 2014 ) and dplyr packages ) , analysis and visualization was performed using the R language . Position-wise Shannon entropy values for domains were calculated as described in Krishnan et al . ( 2018 ) . Tests for significance of association of systems with multicellularity were performed thus: ( 1 ) Each organism in the above-mentioned curated prokaryotic genome database assembled from NCBI Genbank/RefSeq was systematically assessed and assigned a multicellularity flag ( True , False , Not Applicable ) using all available information obtained from the Bergey’s Manual of Systematic Bacteriology ( Whitman , 2015 ) , and the latest publications on the individual taxon . This was done for the 6956 organisms in the database ( Supplementary file 1 ) , which accounts for all the prokaryotic organisms in it other than the candidate phyla radiation ( CPR ) for which no information exists . ( 2 ) This gave us a set background frequency of organisms with or without a multicellular habit which we could use to test the significance of the associations of the systems reported here with multicellularity using the hypergeometric distribution implemented in the phyper command of the R language . For this test , the four input values were: q = the number of organisms containing a copy of a given system which score as multicellular; m = the total number of multicellular organisms in the database; n = total number of non-multicellular organisms in the database; k = total number of organisms with the given system [drawn without replacement from the total set in the database] . The χ2 test for the preferential location of VMAP-ternary systems in the genome at approximately 33% was performed using the implementation in chisq . test command in the R language ( see above for details ) .	Bacteria are the most numerous lifeforms on the planet . Most bacteria live as single cells that grow and multiply independently within larger communities of microbes . However , some bacterial cells assemble to form more complex structures where individual cells might perform distinct roles . Such bacteria are referred to as ‘multicellular bacteria’ . For example , cells of bacteria known as Streptomyces collectively form filaments that help the bacteria collect nutrients from their food sources , and aerial structures bearing reproductive spores . Bacteria in these filaments may come into contact with many other microbes in their surroundings including other bacteria within the same filament , other species of bacteria , and viruses . These contacts often lead to conflict , for example , if the microbes compete with each other for nutrients or if a virus tries to attack the bacteria . Bacteria have evolved immune systems that detect other microbes and use antibiotics , toxins and other defense mechanisms against them . Compared to single-celled bacteria , multicellular bacteria may be more vulnerable to threats from viruses because once a virus has overcome the defenses of one cell in the multicellular assembly , it may be easier for it to kill , or spread to the other cells . However , it is not clear how these systems evolved to deal with the unique problems of multicellular bacteria . Now , Kaur , Burroughs et al . have used computational approaches to search for new immune systems in diverse multicellular bacteria . The new classes of systems they found are each made of different molecular components , but all require a large input of energy to be activated . This activation barrier prevents the bacterial cells from deploying weapons unless the signal from the enemy microbe crosses a high enough threshold . Many tools used in molecular biology , and increasingly in medicine , have been derived from the immune systems of bacteria , such as the enzymes that cut or edit DNA . The findings of Kaur , Burroughs et al . may aid the development of new tools that specifically bind to viruses or other dangerous microbes , or inhibit their ability to interact with components in cells . The next step would be to perform experiments using some of the immune systems identified in this work .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "genetics", "and", "genomics" ]	2020	Highly regulated, diversifying NTP-dependent biological conflict systems with implications for the emergence of multicellularity
Nodal is considered the key inducer of mesendoderm in vertebrate embryos and embryonic stem cells . Other TGF-beta-related signals , such as Vg1/Dvr1/Gdf3 , have also been implicated in this process but their roles have been unclear or controversial . Here we report that zebrafish embryos without maternally provided vg1 fail to form endoderm and head and trunk mesoderm , and closely resemble nodal loss-of-function mutants . Although Nodal is processed and secreted without Vg1 , it requires Vg1 for its endogenous activity . Conversely , Vg1 is unprocessed and resides in the endoplasmic reticulum without Nodal , and is only secreted , processed and active in the presence of Nodal . Co-expression of Nodal and Vg1 results in heterodimer formation and mesendoderm induction . Thus , mesendoderm induction relies on the combination of two TGF-beta-related signals: maternal and ubiquitous Vg1 , and zygotic and localized Nodal . Modeling reveals that the pool of maternal Vg1 enables rapid signaling at low concentrations of zygotic Nodal . The induction of mesoderm and endoderm ( mesendoderm ) during embryogenesis and embryonic stem cell differentiation generates the precursors of the heart , liver , gut , pancreas , kidney and other internal organs . Nodal , a ligand in the TGF-beta protein family , is the key inducer of vertebrate mesendoderm ( Schier and Shen , 2000; Schier , 2009; Shen , 2007 ) , ranging from zebrafish and mouse embryos to human embryonic stem cells . Nodal mutants fail to form mesendodermal cell lineages in zebrafish and mouse ( Conlon et al . , 1991; 1994; Feldman et al . , 1998; Zhou et al . , 1993 ) , and activation of the Nodal signaling pathway drives the in vitro differentiation of embryonic stem cells into mesendodermal progenitors ( Brandenberger et al . , 2004; Camus et al . , 2006; D'Amour et al . , 2005; Hoveizi et al . , 2014; Kubo et al . , 2004; Parisi et al . , 2003; Schier and Shen , 2000; Shen , 2007; Smith et al . , 2008; Takenaga et al . , 2007; Vallier et al . , 2004; Yasunaga et al . , 2005 ) . Following its role in mesendoderm induction , Nodal activity also patterns the left-right axis . Nodal ligands are expressed in the left lateral plate mesoderm ( Collignon et al . , 1996; Levin et al . , 1995; Long et al . , 2003; Lowe et al . , 1996; Pagán-Westphal and Tabin , 1998 ) , and mutants that lack left-sided Nodal signaling exhibit multiple left-right defects ( Brennan et al . , 2002; Kumar et al . , 2008; Long et al . , 2003; Noël et al . , 2013; Saijoh et al . , 2003; Yan et al . , 1999 ) . Nodal is not the only TGF-beta-related signal implicated in mesendoderm induction and left-right patterning . Members of the Vg1/GDF1/GDF3 TGF-beta subfamily have been assigned various roles in these processes , although there are puzzling contradictions from the level of gene expression to the loss-of-function and gain-of-function phenotypes . The role of GDF1 in left-right patterning is well established . Gdf1 mutant mice exhibit left-right asymmetry defects ( Rankin et al . , 2000 ) and morpholino studies indicate that zebrafish vg1 ( dvr1/gdf3 ) is required for left-right patterning ( Peterson et al . , 2013 ) . GDF1/Vg1 alone is unable to activate the Nodal signaling pathway , but it increases the activity and range of mouse and zebrafish Nodal ligands in Xenopus assays ( Peterson et al . , 2013; Tanaka et al . , 2007 ) and the activity of mouse Nodal in tissue culture cells ( Andersson et al . , 2007; Fuerer et al . , 2014 ) . Thus , Nodal and Vg1/GDF1 family members cooperate to pattern the left-right axis . The role of the Vg1/GDF1/GDF3 TGF-beta subfamily in mesendoderm formation is less clear . In Xenopus – where Vg1 was first discovered – vg1 mRNA is localized to a vegetal crescent in the oocyte and in the vegetal hemisphere of the early embryo ( Rebagliati et al . , 1985; Weeks and Melton , 1987 ) . By contrast , zebrafish vg1 mRNA is localized to the animal pole of late stage oocytes ( Marlow and Mullins , 2008 ) , and it is present ubiquitously in the early embryo ( Dohrmann et al . , 1996; Helde and Grunwald , 1993; Peterson et al . , 2013 ) . Gdf1 and Gdf3 , which are considered to be the mammalian Vg1 orthologs ( Andersson et al . , 2007; Chen et al . , 2006; Rankin et al . , 2000; Wall et al . , 2000 ) , are expressed in the 16-cell morula ( Gdf3 ) and epiblast prior to gastrulation ( Gdf3 and Gdf1 ) ( Chen et al . , 2006; Wall et al . , 2000 ) . Some Gdf3 mutants lack a subset of endodermal and mesodermal markers , while others grow to fertile adults ( Andersson et al . , 2007; Chen et al . , 2006 ) ; conversely Gdf1 mutants only exhibit left-right asymmetry defects ( Rankin et al . , 2000 ) . Some Gdf1;Gdf3 double mutants exhibit more severe defects in endoderm and mesoderm formation than Gdf3 single mutants ( Andersson et al . , 2007 ) , and Gdf1-/-;Nodal+/-mutants resemble hypomorphic nodal mutants ( Andersson et al . , 2006; Lowe et al . , 2001 ) , suggesting some synergy between GDF1 and Nodal functions . Experiments in the chick and mouse indicate that Vg1/GDF1/GDF3 may act upstream of Nodal ( Andersson et al . , 2007; Chen et al . , 2006; Rankin et al . , 2000; Shah et al . , 1997; Skromne and Stern , 2001; Tanaka et al . , 2007 ) . Thus , these analyses suggest that mouse Nodal , GDF1 and GDF3 may cooperate during early amniote development , but their regulatory and molecular relationships have remained unclear . Morphant studies in zebrafish suggest a function for Vg1 in left-right axis formation but not in mesendoderm induction ( Peterson et al . , 2013 ) . Antisense oligonucleotide-mediated knockdown of Xenopus vg1 leads to defects in dorsal mesoderm induction ( Birsoy et al . , 2006 ) , but most mesendodermal derivatives still form . Taken together , loss-of-function studies establish crucial roles for both Nodal and Vg1/GDF1 in left-right development and for Nodal in mesendoderm induction , but the roles of GDF1/GDF3/Vg1 in mesendoderm induction remain poorly understood . Another puzzling aspect of Vg1’s function is its apparent inability to be processed and secreted . This is in stark contrast to other members of the TGF-beta superfamily , which are generated as pro-proteins that dimerize and are cleaved to generate a secreted , mature dimer that binds receptors ( Constam , 2014; Dutko and Mullins , 2011 ) . Neither cleavage nor secretion of Vg1 has been detected in Xenopus and zebrafish , and correspondingly , overexpression does not yield a phenotype ( Dale et al . , 1993; Dale et al . , 1989; Dohrmann et al . , 1996; Tannahill and Melton , 1989; Thomsen and Melton , 1993 ) . Conflicting results have been reported for GDF1 processing in heterologous systems , ranging from cleavage but inactivity in Xenopus oocytes ( Tanaka et al . , 2007 ) to no detectable cleavage in Xenopus embryos ( Wall et al . , 2000 ) . Mouse Nodal-GDF1 heterodimers , but not zebrafish Nodal-Vg1 heterodimers , have been detected in a heterologous Xenopus system ( Peterson et al . , 2013; Tanaka et al . , 2007 ) . Upon fusion of the Vg1 , GDF1 or GDF3 mature domain to the Activin or BMP prodomain , Vg1 is processed and induces mesoderm formation ( Chen et al . , 2006; Dale et al . , 1993; Dohrmann et al . , 1996; Kessler and Melton , 1995; Thomsen and Melton , 1993; Wall et al . , 2000 ) . However , it is unclear if these constructs reveal the true nature of Vg1 , or whether the fused prodomains generate ectopic functions . Thus , it remains to be resolved how Vg1 processing , secretion , dimerization and activity are regulated . In this study we address the long-standing question of Vg1’s role in vertebrate mesendoderm induction and its relationship to Nodal , using zebrafish as a model system . Current models of zebrafish mesendoderm induction have focused entirely on the roles of the two zebrafish Nodal genes , cyclops ( cyc ) and squint ( sqt ) , with no consideration of vg1 ( Bodenstine et al . , 2016; Cartwright et al . , 2008; Chea et al . , 2005; Constam , 2009; Hirokawa et al . , 2006; Juan and Hamada , 2001; Liang and Rubinstein , 2003; Papanayotou and Collignon , 2014; Pauklin and Vallier , 2015; Quail et al . , 2013; Robertson , 2014; Schier and Shen , 2000; Shen , 2007; Signore et al . , 2016; Strizzi et al . , 2012; 2009; Tian and Meng , 2006; Wang and Tsang , 2007; Whitman , 2001 ) . cyc and sqt are zygotically-expressed at the embryonic margin and act as concentration-dependent inducers of mesendoderm ( Schier , 2009 ) . cyc;sqt double mutants ( Feldman et al . , 1998 ) and other zebrafish Nodal signaling mutants ( Dubrulle et al . , 2015; Gritsman et al . , 1999 ) fail to form endoderm and head and trunk mesoderm . Conversely , ectopic expression of cyc or sqt induces mesendoderm formation ( Bisgrove et al . , 1999; Feldman et al . , 1998; Gritsman et al . , 1999; 2000; Meno et al . , 1999; Sampath et al . , 1998 ) . These results , and the lack of a vg1 morphant mesendoderm phenotype ( Peterson et al . , 2013 ) , have been interpreted to mean that Cyc and Sqt are the sole inducers of mesendoderm , without a requirement for Vg1 or other TGF-beta family members . Contrary to these models , we now report that vg1 is absolutely essential for mesendoderm induction . Vg1 is only secreted , processed and active in the presence of Nodal , while Nodal requires Vg1 for activity . Co-expression of Nodal and Vg1 results in heterodimer formation and mesendoderm induction . To determine the function of zebrafish Vg1 , we generated vg1 mutants using CRISPR/Cas9 ( Figure 1—figure supplement 1A ) . We recovered 8 bp and 29 bp deletion alleles that cause frameshifts , truncating Vg1 from a 355 amino acid protein to predicted 18 and 11 amino acid peptides , respectively ( Figure 1—figure supplement 1B ) . Zygotic homozygous vg1 ( Zvg1 ) mutants were viable , with no strong left-right asymmetry defects ( Figure 1—figure supplement 2 ) , allowing the generation of maternal vg1 ( Mvg1 ) mutants from homozygous females crossed to wild-type males ( Figure 1A ) . Mvg1 embryos lacked the derivatives of the mesendoderm , including heart , blood , pronephros , notochord , gut and trunk somites ( Figure 1A ) . To test whether the phenotype is caused by the loss of vg1 , we performed rescue experiments by injecting 5 concentrations of vg1 mRNA , spanning a 1600-fold range . 0 . 5–100 pg of vg1 rescued the phenotype , revealing that the embryo can tolerate a large range of vg1 concentrations ( Figure 1B ) . 50 pg of a vg1 mRNA containing the 8 bp deletion found in the genetic mutant was unable to rescue the phenotype ( Figure 1C ) . In contrast to previous vg1 morpholino experiments ( Peterson et al . , 2013 ) , these results reveal that vg1 is essential for mesendoderm formation . The phenotype of Mvg1 and maternal-zygotic vg1 ( MZvg1 ) embryos closely resembles that of embryos that lack Nodal ( Feldman et al . , 1998 ) , its co-receptor Oep ( Gritsman et al . , 1999 ) , or its signal transducer Smad2 ( Dubrulle et al . , 2015 ) ( Figure 2A ) . To determine whether Mvg1 embryos are defective in Nodal signaling , we analyzed the expression of a selection of Nodal target genes . The expression of these mesendoderm genes showed the same defects in Mvg1 mutants as in Nodal signaling mutants , indicating that Nodal signaling is not functional in the absence of Vg1 ( Figure 2B ) . One way Nodal signaling might be disrupted in Mvg1 embryos is through loss of Nodal gene expression . We analyzed cyc and sqt expression in wild-type and Mvg1 embryos . cyc and sqt were initially expressed at comparable levels across both genotypes , but mRNA levels subsequently increased in wild-type embryos by autoregulation ( Meno et al . , 1999 ) while they generally remained low in Mvg1 embryos ( Figure 2C ) . These results suggest that Vg1 is required for the auto-induction but not initiation of Nodal gene expression , and that the remaining endogenous levels of Nodal are not able to induce mesendoderm in the absence of Vg1 . To test whether the Nodal ligands might be inactive in the absence of Vg1 , we overexpressed cyc or sqt in Mvg1 embryos and analyzed Nodal target gene expression . High levels ( 50 pg of mRNA ) of cyc failed to induce target gene expression in Mvg1 embryos ( Figure 2D ) , whereas sqt at low ( 0 . 2 pg ) but not high ( 2–50 pg ) levels of overexpression failed to induce target gene expression ( Figures 2D and E ) . We then co-expressed 20 pg of cyc mRNA with increasing concentrations of vg1 mRNA in Mvg1 embryos . Co-expression of cyc and 5 pg of vg1 caused an increase in induction of Nodal target gene expression compared to cyc alone ( Figure 2F ) . These results indicate that Vg1 is necessary for Cyc activity and partially needed for Sqt activity . To determine whether the Nodal ligands are processed and secreted in Mvg1 embryos , we expressed superfolderGFP ( sfGFP ) -tagged derivatives of Cyc and Sqt ( Müller et al . , 2012; Pédelacq et al . , 2006 ) . No differences in cleavage or localization of Cyc and Sqt were detected in the presence or absence of Vg1 ( Figures 2G and H ) . Taken together , these results suggest that Vg1 is necessary for the endogenous activities , but not the processing and secretion , of Cyc and Sqt . Previous studies did not detect Vg1 processing in early embryos ( Dale et al . , 1989; Dohrmann et al . , 1996; Tannahill and Melton , 1989; Thomsen and Melton , 1993 ) . To examine the relationship of Vg1 processing to presence or absence of Nodal proteins , we first inserted sfGFP downstream of the Vg1 cleavage site ( Figure 3A ) . vg1-sfGFP rescued Mvg1 mutants ( Figure 3B ) but cleavage of Vg1 protein was undetectable ( Figure 3C ) . To test whether Vg1 needs to be processed to be functional , we mutated the basic residues in the Vg1 cleavage site to non-basic residues ( Vg1-NC , ‘Non-Cleavable’ ( Figure 3E ) ) . This abolished Vg1 rescuing activity ( Figure 3D ) , suggesting that endogenous Vg1 cleavage is not detectable but is required for Vg1 function . Given that the Mvg1 phenotype resembles Nodal loss-of-function phenotypes , and Vg1 requires its cleavage site , we asked whether Nodal might induce Vg1 cleavage . We co-expressed vg1-sfGFP with cyc or sqt and discovered that Vg1-sfGFP was cleaved to its mature form in the presence of Nodal ( Figure 3E , Figure 3—figure supplement 1 ) . By contrast , Vg1-sfGFP was not cleaved upon co-expression with an alternative TGF-beta-related ligand , bmp7a ( Figure 3—figure supplement 1B ) , and non-cleavable Vg1-sfGFP ( Vg1-NC-sfGFP ) was not cleaved in the presence of Cyc or Sqt ( Figure 3E , Figure 3—figure supplement 1A ) . These data reveal that Nodal induces Vg1 processing . To examine the secretion and localization of Vg1 , we expressed vg1-sfGFP in wild-type or Mvg1 embryos for in vivo imaging . In contrast to the extracellular localization of Cyc and Sqt ( Figure 2H ) , Vg1 was only detected intracellularly , predominantly in the endoplasmic reticulum ( ER ) ( Figure 4A ) ( Fodero-Tavoletti et al . , 2005; Southall et al . , 2006; Szul and Sztul , 2011; Tu et al . , 2002 ) . To determine whether Nodal can induce not only Vg1 processing but also secretion , we co-expressed vg1-sfGFP with cyc or sqt . Notably , Vg1-sfGFP formed extracellular puncta and/or diffuse extracellular signal upon co-expression with cyc or sqt ( Figure 4B , Figure 4—figure supplement 1A , Table 1 ) . By contrast , Vg1-sfGFP was not secreted upon co-expression with bmp7a ( Figure 4B ) . To directly test whether Vg1 is secreted in the presence of Nodal , we tagged Vg1 with the pH-sensitive fluorescent protein pHluorin2 ( Mahon , 2011 ) . pHluorin2 is non-fluorescent at acidic pH , as found in intracellular vesicles , but it fluoresces in the neutral pH of the extracellular space . Vg1-pHluorin2 fluorescent puncta were only visible upon co-expression with cyc or sqt , indicating that Vg1 is secreted in the presence of Nodal ( Figure 4C , Figure 4—figure supplement 1B ) . To independently test if Vg1 is only secreted in the presence of Nodal , we expressed vg1-sfGFP in single-cell embryos and co-expressed cyc-RFP in 1 cell at the 16-cell stage . At sphere stage , Vg1-sfGFP was only secreted in the cells that also expressed Cyc-RFP ( Figure 4D ) . To determine whether Vg1 and Nodal co-localize , we co-expressed vg1-sfGFP with cyc-RFP or sqt-RFP . Vg1-sfGFP displayed extensive extracellular co-localization with Cyc-RFP and Sqt-RFP ( Figure 4E , Figure 4—figure supplement 1C ) . Taken together , these results reveal that Nodal induces the secretion of Vg1 , and that Vg1 and Nodal co-localize . The co-localization of Vg1 and Nodal suggested that these secreted ligands might form heterodimers , as detected for GDF1 and Nodal , and some other TGF-beta-related signals ( Aono et al . , 1995; Dutko and Mullins , 2011; Eimon and Harland , 2002; Fuerer et al . , 2014; Guo and Wu , 2012; Hazama et al . , 1995; Israel et al . , 1996; Little and Mullins , 2009; Nishimatsu and Thomsen , 1998; Schmid et al . , 2000; Shimmi et al . , 2005; Suzuki et al . , 1997; Tanaka et al . , 2007 ) . To test this hypothesis , we performed co-immunoprecipitation experiments by co-expressing 50 pg of vg1-Flag with 50 pg of cyc-HA or sqt-HA . Vg1-Flag co-immunoprecipitated with Cyc-HA or Sqt-HA ( Figure 5A , Figure 5—figure supplement 1A ) . To test the specificity of this interaction we used two different concentrations of sqt-HA mRNA in combination with three different concentrations of vg1-Flag or bmp7a-Flag mRNA . We detected an interaction between Sqt-HA and Vg1-Flag at all six concentrations tested , whereas an interaction between Sqt-HA and Bmp7a-Flag was only detected at the highest concentration of each mRNA ( Figure 5—figure supplement 1B ) . Thus , Vg1 specifically interacts with Nodal to form heterodimers . The heterodimerization of Vg1 and Nodal raises the possibility that Vg1 is maintained in a monomeric state in the absence of Nodal . Indeed , a previous study found that Vg1 does not form homodimers , and that endogenous Vg1 is predominantly monomeric ( Dale et al . , 1993 ) . To test the monomeric or dimeric states of Vg1 in the absence of Nodal , we performed reducing and non-reducing immunoblots of wild-type and Mvg1 embryos expressing vg1-Flag , cyc-Flag or bmp7a-Flag mRNA . TGF-beta family members are disulfide-linked dimers: under reducing conditions the disulfide bonds are broken , while under non-reducing conditions the bonds are maintained , allowing the detection of dimers . Bmp7a-Flag mature homodimers were visible under non-reducing conditions , whereas Vg1-Flag homodimers were not detected ( Figure 5B , Figure 5—figure supplement 1C ) . Using a complementary approach , we tested whether Vg1 forms homodimers by co-immunoprecipitation . While Sqt-HA and Vg1-Flag co-precipitated , Vg1-HA and Vg1-Flag did not ( Figure 5C ) . These results indicate that Vg1 does not form homodimers and might be present as monomers in the absence of Nodal . Vg1 protein is synthesized before Nodal transcription and translation begin , raising the possibility that newly synthesized Nodal monomers bind to preexisting Vg1 monomers . Alternatively , Nodal might only heterodimerize with Vg1 protein that is co-translated with Nodal . To distinguish between these possibilities , we generated a Vg1-Dendra2 photoconvertible fusion protein and injected it at the 1-cell stage . At the 64-cell stage we photoconverted Vg1-Dendra2 from green to red , and co-injected the embryos with 5 pg of cyc mRNA . Imaging revealed the production of red puncta , indicating that Vg1 protein synthesized prior to Nodal synthesis was able to heterodimerize and be secreted with Nodal ( Figure 5—figure supplement 1D ) . This data suggests that Nodal can heterodimerize with pre-existing Vg1 . To determine if co-expression of Vg1 and Nodal in the same cells is required for activity , we used transplantation assays to compare target gene induction in cells co-expressing vg1 and cyc versus neighboring cells expressing either vg1 or cyc . Nodal target gene induction only occurred when vg1 and cyc were co-expressed in the same cells ( Figure 5D and Figure 5—figure supplement 2A ) . Analogously , deposition of vg1 mRNA to the yolk syncytial layer ( YSL ) of Mvg1 mutants , where cyc and sqt are expressed endogenously , was sufficient to rescue Mvg1 mutants by morphology and gene expression ( Figure 5E and Figure 5—figure supplement 2B ) . Confocal imaging of embryos injected with vg1 mRNA and a fluorescent dextran into the YSL indicated that the majority of fluorescence was localized to the YSL , but a few cells in the margin also inherited the fluorescent dextran ( Figure 5—figure supplement 2C ) . Thus , although vg1 is ubiquitously expressed in the early embryo ( Helde and Grunwald , 1993; Peterson et al . , 2013 ) ( Figure 5—figure supplement 2D ) , its co-localization with cyc and sqt is sufficient for its role in mesendoderm formation . Taken together , these results suggest that Vg1 and Nodal are active when expressed in the same cells , where they form heterodimers . The requirement for Vg1-Nodal heterodimers for mesendoderm induction raises the question of why the embryo relies on both a ubiquitous ligand , Vg1 , and localized ligands , Cyc and Sqt . We developed a basic kinetic model to test the rate of Nodal homodimer formation versus Nodal-Vg1 heterodimer formation in the presence of a maternal Vg1 pool . Simulating these two conditions revealed that the preloading of inactive Vg1 monomers in the cell allows Nodal to immediately form heterodimers whereas dimer formation is delayed when Nodal must form homodimers ( Figure 6 ) . Thus , Vg1-Nodal heterodimers can initiate signaling more quickly than Nodal homodimers , and already at low Nodal levels . Even if Nodal homodimers were as active as Vg1-Nodal heterodimers , Nodal target gene induction would still be slower in the absence of maternal Vg1 , because the association of two Nodal monomers is less likely at low Nodal concentrations than Vg1-Nodal dimerization ( Figure 6 ) . These simulations reveal that low concentrations of zygotic Nodal can be directly transformed into pathway activation via association with maternal Vg1 . Knockdown studies in zebrafish suggested no requirement for Vg1 in mesendoderm induction ( Peterson et al . , 2013 ) , but the loss-of-function mutants reported here reveal that Vg1 is absolutely required for the induction of head and trunk mesoderm and endoderm . Strikingly , vg1 mutants strongly resemble Nodal signaling mutants , showing that zebrafish Vg1 has as essential a function as Nodal , which has been considered the sole mesendoderm inducer . The results in zebrafish warrant a re-analysis of the requirements for Vg1 orthologs in other systems . For example , mouse Gdf1;Gdf3 double mutants have incompletely penetrant mesendodermal phenotypes ( Andersson et al . , 2007 ) . A closer comparison to Nodal mutants might reveal functions of GDF1 and GDF3 that are equivalent to Vg1 . It is also possible that mouse Nodal is expressed at sufficiently high levels to be less dependent on GDF1/GDF3 , akin to the overexpression of zebrafish sqt . Similarly , zebrafish southpaw might be expressed at high enough levels to act independently of zygotic vg1 during left-right development . Knockdown studies in Xenopus have suggested that Vg1 is mainly involved in inducing notochord precursors , but not other mesendodermal progenitors ( Birsoy et al . , 2006 ) . Mutant studies in Xenopus might reveal broader roles for Vg1 , or alternatively additional TGF-beta-related signals such as Derrière ( Sun et al . , 1999 ) , which has been shown to interact with Nodal ( Eimon and Harland , 2002 ) , might have complementary or overlapping functions with Vg1 . More generally , it is conceivable that the activities of the Nodal and Vg1/GDF1/GDF3 subfamilies are co-dependent in all contexts . This idea is not only supported by the co-dependence of Nodal and Vg1 in zebrafish mesendoderm induction reported here , but also the observation that Nodal expression coincides with the expression of Vg1 family members in numerous contexts ( Agius et al . , 2000; Levin et al . , 1995; Onai et al . , 2010; Range and Lepage , 2011; Range et al . , 2007; Seleiro et al . , 1996 ) . Moreover , mouse Gdf1 mutants display very similar left-right defects as mutants with impaired Nodal signaling ( Andersson et al . , 2006; Cheng et al . , 2003; Lowe et al . , 2001; Yan et al . , 1999 ) . It is therefore tempting to speculate that wherever and whenever Nodal subfamily members are expressed and active , they are accompanied by Vg1 subfamily members . In this scenario , Vg1/GDF1/GDF3 act in parallel with Nodal even when their expression is precedent , and any apparent upstream functions of Vg1 ( Andersson et al . , 2007; Chen et al . , 2006; Rankin et al . , 2000; Shah et al . , 1997; Skromne and Stern , 2001; Tanaka et al . , 2007 ) are actually the result of Nodal autoinduction: Vg1 is required together with Nodal to fully activate Nodal gene expression but it is not needed for the initiation of Nodal expression , as shown in the Mvg1 mutants . Finally , Nodal and Vg1 act through the same co-receptors ( Andersson et al . , 2007; Cheng et al . , 2003; Fuerer et al . , 2014; Tanaka et al . , 2007 ) and are both inhibited by Lefty ligands ( Agathon et al . , 2001; Bisgrove et al . , 1999; Chen and Shen , 2004; Chen and Schier , 2002; Cheng et al . , 2004; Meno et al . , 1999; 1996; Thisse et al . , 2000; Thisse and Thisse , 1999 ) . These observations suggest that Nodal signaling should henceforth be considered Nodal/Vg1 signaling . Our study clarifies previously puzzling observations on the activity and processing of Vg1 that contrast with the properties of other TGF-beta-related signals: overexpression of wild-type vg1 does not cause a phenotype ( Dale et al . , 1993; Dohrmann et al . , 1996; Tannahill and Melton , 1989; Thomsen and Melton , 1993 ) , neither secreted ( Dale et al . , 1993; Tannahill and Melton , 1989 ) nor processed Vg1 has been reliably detected ( Dale et al . , 1989; Dohrmann et al . , 1996; Tannahill and Melton , 1989; Thomsen and Melton , 1993 ) , but fusion of the Vg1 mature domain to the Activin or BMP prodomain results in processed and active Vg1 ( Dale et al . , 1993; Dohrmann et al . , 1996; Kessler and Melton , 1995; Thomsen and Melton , 1993; Wall et al . , 2000 ) . Our study explains these conundrums by revealing that Vg1 is only processed , secreted and active in the presence of Nodal . Without Nodal , Vg1 is unprocessed and predominantly resides in the ER . Upon overexpression , Vg1 contributes to this inert pool , and only fusion to heterologous prodomains allows secretion , cleavage and activation in the absence of Nodal . Thus , the dependence of Vg1 processing , secretion and activity on Nodal accounts for many of the previously confusing observations . One conundrum remains: we and others have not been able to detect the processing ( Figure 3C ) or secretion ( Figure 4A ) of Vg1 at endogenous levels of Nodal . We speculate that endogenous Nodal is expressed at very low levels and in few cells , resulting in cleavage and secretion of only a small ( and undetectable ) fraction of the total pool of Vg1 . Only upon ectopic Nodal expression , sufficiently high levels of Vg1 are processed to become detectable . The development of more sensitive detection methods is needed to directly demonstrate the cleavage , processing and secretion of endogenous Vg1 . Our study reveals that Nodal and Vg1 form heterodimers , and that Vg1 exists in a monomeric state prior to heterodimerization with Nodal . The initial localization of Vg1 to the ER suggests that this is the site of heterodimerization , which is consistent with previous studies of other heterodimers ( Duitman et al . , 2008; Hurtley and Helenius , 1989; Jalah et al . , 2013; Lorenz et al . , 2002; Persson and Pettersson , 1991; Tu et al . , 2002 ) . For example , in the case of the uroplakin proteins UPIb and UPIII , UPIb can autonomously exit the ER and translocate to the plasma membrane . By contrast , UPIII must heterodimerize with UPIb in the ER in order to exit and move to the plasma membrane ( Tu et al . , 2002 ) . Although we currently favor a model in which monomeric Vg1 meets Nodal in the ER , more complex scenarios are conceivable . For example , it is unclear whether pre-existing Vg1 might associate with other ER-resident proteins to maintain or prepare it in a state that allows dimerization with newly synthesized Nodal . Our results suggest that Nodal-Vg1 heterodimers are more potent than Nodal alone: in the case of Cyc , such heterodimers seem to be required for Cyc to activate signaling , whereas Sqt-Vg1 heterodimers appear to be more active than Sqt alone ( Figures 2D , E and F ) . The molecular basis of the increased activity is unknown , but based on previous studies in the BMP system , heterodimers might be necessary to assemble heteromeric combinations of two types of class I or II receptors ( Little and Mullins , 2009 ) . Our results also extend and generalize the previous observation that mouse GDF1 and Nodal form heterodimers ( Fuerer et al . , 2014; Tanaka et al . , 2007 ) , although those studies did not address the requirement , localization , or processing of Vg1/GDF1/GDF3 during mesendoderm formation , and instead proposed that heterodimer formation might increase the potency and/or range of Nodal . Our results uncover the alternative or additional mechanism that heterodimer formation triggers processing and secretion of Vg1 and allows Nodal to be active at physiological concentrations . Our results demonstrate a novel mode to restrict TGF-beta-related protein activity through heterodimer formation . vg1 mRNA and protein do not need to be localized in the embryo to restrict Vg1 activity: instead , it is the absence of Nodal that blocks Vg1 processing , secretion and activity . Indeed , zebrafish Vg1 is present ubiquitously in early embryos and vg1 is expressed in broader domains than Nodal in all systems analyzed . The question therefore arises whether the exquisite vegetal localization of Xenopus vg1 is important for development ( Weeks and Melton , 1987 ) . The localized activation of Xenopus Nodal genes might be sufficient to restrict mesendoderm formation to vegetal and marginal regions , but it is also possible that localized Vg1 provides an additional safeguard to spatially restrict pathway activation . Rescue experiments similar to those reported here could address this question . Modeling of hetero- and homodimerization kinetics reveals that the maternal pool of Vg1 accelerates the onset of ligand dimerization relative to a system that relies on Nodal-Nodal dimerization alone . This could be advantageous in the embryo , where mesendoderm induction cannot initiate until after the maternal-to-zygotic transition . Although it may be counterintuitive for a spatially localized signal to rely on a ubiquitous signal for pathway activation , the preloading of Vg1 in the ER could be instrumental for ensuring Nodal signaling initiates in a rapid and temporally reliable manner . Thus , the requirement for Vg1 in the zebrafish embryo can ensure rapid and sensitive response to the low concentrations of Nodal that initiate mesendoderm induction . The finding that Vg1 – together with Nodal – is an endogenous mesoderm inducer resolves some of the historical controversies in the field . Vg1 was described in 1987 as a TGF-beta-related signal present at the right time and place to be a mesoderm inducer ( Weeks and Melton , 1987 ) , but the lack of a functional requirement raised doubts about its importance . Conversely , Activin was reported in 1990 as a TGF-beta-related signal that can induce mesoderm ( Smith et al . , 1990; van den Eijnden-Van Raaij et al . , 1990 ) , but its absence during early embryogenesis ( Dohrmann et al . , 1993; Thomsen et al . , 1990 ) , and the lack of loss-of-function phenotypes ( Hawley et al . , 1995; Kessler and Melton , 1995; Matzuk et al . , 1995; Schulte-Merker et al . , 1994; Sun et al . , 1999 ) , raised doubts about its importance ( Schier and Shen , 2000 ) . With the discovery of the essential roles of mouse Nodal ( Conlon et al . , 1994; Zhou et al . , 1993 ) and zebrafish Nodal ( Feldman et al . , 1998 ) in mesoderm induction , the field converged to the view that Nodal is the key inducer . Our study indicates instead that Nodal-Vg1 heterodimers are the essential endogenous inducers of mesendoderm , while Activin serves as a powerful reagent to induce mesendoderm from embryonic stem cells . sgRNAs targeting the vg1/dvr1/gdf3 gene were designed using CHOPCHOP ( RRID:SCR_015723 ) ( Labun et al . , 2016; Montague et al . , 2014 ) and synthesized as previously described ( Gagnon et al . , 2014 ) ( See also Figure 1—figure supplement 1 ) . vg1 sgRNAs were co-injected with ~0 . 5 nL of 50 μM Cas9 protein into TLAB wild-type embryos . Injected embryos were raised to adulthood and outcrossed to TLAB adults . Clutches of embryos with potential heterozygous individuals were used to identify founders with germline mutations in vg1 by extracting DNA from 10 embryos and genotyping by MiSeq sequencing . The offspring of confirmed founders were raised to adulthood and genotyped to identify heterozygous vg1 adults . Heterozygous vg1 mutants were intercrossed to generate zygotic homozygous ( Zvg1 ) fish . For maintaining the vg1 mutant line , homozygous Zvg1 male fish were crossed to heterozygous female fish , and the resulting progeny were genotyped to identify Zvg1 adults . To generate maternal vg1 ( Mvg1 ) mutants , TLAB wild-type male fish were crossed to homozygous Zvg1 female fish . Two deletion alleles of 8 bp and 29 bp ( vg1a164 and vg1a165 respectively ) were recovered in the first exon of vg1 from the sgRNA targeting the sequence GGGTCAGAAGACAGGCTCTGAGG . Genomic DNA was extracted using the HotSHOT method ( Meeker et al . , 2007 ) and PCR was performed using standard conditions ( see primer sequences below ) , followed by Sanger sequencing or MiSeq sequencing for the 8 bp allele ( Gagnon et al . , 2014 ) or 2% gel electrophoresis for the 29 bp allele . The vg1 CDS sequence was PCR amplified from a high-stage cDNA library and cloned into the pSC vector ( Agilent ) with a beta-globin 5’UTR and an SV40 3’UTR using Gibson assembly ( Gibson et al . , 2009 ) to generate pSC-vg1 . To generate pCS2 ( + ) -cyc and pCS2 ( + ) -sqt , the cyc and sqt CDS sequences were PCR amplified from a high-stage cDNA library and cloned into the pCS2 ( + ) vector using Gibson assembly . To generate non-cleavable forms of vg1 and vg1-sfGFP , site-directed mutagenesis was used to replace the RSRRKR cleavage site with SQNTSN using a Q5 Site-Directed Mutagenesis Kit ( NEB ) . All superfolder GFP ( sfGFP ) ( Pédelacq et al . , 2006 ) , RFP , Dendra2 and pHluorin2 ( Mahon , 2011 ) fusion constructs were generated by PCR-based methods and cloned into the pCS2 ( + ) vector using Gibson assembly . Flag ( DYKDDDDK ) and HA tag ( YPYDVPDYA ) sequences were inserted by site-directed mutagenesis of pSC-vg1 , pCS2 ( + ) -cyc and pCS2 ( + ) -sqt . For Vg1 fusions , sequences encoding the fluorescent protein or Flag tag were inserted downstream of the cleavage site ( RSRRKR ) with a GSTGTT linker separating the prodomain and fluorescent protein , and a GS linker separating the fluorescent protein and the Vg1 mature domain . For Cyc fusions , sequences encoding the fluorescent proteins or HA tag were inserted two amino acids downstream of the cleavage site ( RRGRR ) ( Müller et al . , 2012 ) . For Sqt fusions , fluorescent protein and HA tag sequences were inserted 10 amino acids downstream of the cleavage site ( RRHRR ) with a GSTGTT linker separating the prodomain and fluorescent protein , and a GS linker separating the fluorescent protein and the mature domain ( Müller et al . , 2012 ) . Vectors were linearized by digestion with NotI ( pCS2 ( + ) vectors ) or XhoI ( pSC vectors ) . Capped mRNAs were synthesized using the SP6 or T7 mMessage Machine Kits ( ThermoFisher ) , respectively . For in situ hybridization , immunoblot , imaging and qPCR experiments , embryos were dechorionated using 1 mg/ml Pronase ( Protease type XIV from Streptomyces griseus , Sigma ) prior to injection , and subsequently cultured in agarose-coated dishes . Embryos were injected at the 1-cell stage unless otherwise stated . Zebrafish embryos were grown at 28°C and staged according to ( Kimmel et al . , 1995 ) . Embryos were cultured in blue water ( 250 mg/L Instant Ocean salt , 1 mg/L methylene blue in reverse osmosis water adjusted to pH 7 with NaHCO3 ) . Embryos were analyzed for mutant phenotypes at 28–32 hpf . For imaging , embryos were anesthetized in Tricaine ( Sigma ) and mounted in 2% methylcellulose then imaged using a Zeiss SteREO Discovery . V12 microscope . Embryos were raised to sphere stage and mounted in 1% low gelling temperature agarose ( Sigma ) on glass-bottomed dishes ( MatTek ) with the animal pole facing the glass . Imaging was performed on Zeiss LSM 700 and LSM 880 inverted confocal microscopes . Embryos were injected at the 1-cell stage with 100 pg of vg1-Dendra2 mRNA then grown at 28°C to the 64-cell stage and injected with 5 pg of cyc mRNA into six locations in the embryo . Embryos were mounted in 1% low gelling temperature agarose and photoconverted with 2 min of UV light at 10x magnification on the Zeiss LSM 700 inverted confocal microscope . The embryos were incubated at 28°C for 30 min before imaging on the LSM 700 microscope over a period of 2 hours . Images were processed in FIJI/ImageJ ( Schindelin et al . , 2012 ) . Brightness , contrast and color balance was applied uniformly to images . Embryos were fixed in 4% formaldehyde overnight at room temperature ( 50% epiboly or younger ) or at 4°C ( embryos older than 50% epiboly ) . Whole mount in situ hybridizations were performed according to standard protocols ( Thisse and Thisse , 2008 ) . DIG-labeled antisense RNA probes against cmlc2 , spaw , gsc , lft1 , ntl , sox32 , cyc , sqt and vg1 were synthesized using a DIG Probe Synthesis Kit ( Roche ) . NBT/BCIP/Alkaline phosphatase-stained embryos were dehydrated in methanol and imaged in benzyl benzoate:benzyl alcohol ( BBBA ) using a Zeiss Axio Imager . Z1 microscope . For DAB staining , embryos were rehydrated in PBST after completing the in situ protocol , and blocked in 10% normal goat serum/1% DMSO before incubation in primary antibody overnight ( 1:400 rabbit anti-GFP-HRP , ThermoFisher A10260 , RRID:AB_2534022 ) . Embryos were washed multiple times in PBST , incubated in DAB solution ( KPL #71-00-48 ) , and dehydrated before imaging in BBBA . For transplantation experiments , donor embryos were injected with 50 pg of cyc mRNA and/or vg1 mRNA and 50 pg of GFP mRNA and grown to sphere stage ( 4 hpf ) . At sphere stage , cells were transplanted from donor embryos to host embryos , and host embryos were grown to shield stage before fixation for in situ hybridization . For YSL injections , 1000-cell stage embryos were injected through the chorion into the YSL with approximately 100 pg of vg1 mRNA and 500 pg of Alexa Fluor 488 dextran ( ThermoFisher ) . Embryos were injected at the 1-cell stage with 50 pg of each mRNA ( unless otherwise stated ) and grown to early gastrulation ( 50% epiboly ) . 8 embryos per sample were manually deyolked with forceps and frozen in liquid nitrogen . The samples were boiled at 95°C for 5 min with 2x SDS loading buffer ( 10 μL ) and DTT ( reducing gels only , 150 mM final concentration ) and then loaded onto Any kD protein gels ( Bio-Rad ) . Samples were transferred to polyvinylidene fluoride ( PVDF ) membranes ( GE Healthcare ) . Membranes were blocked in 5% non-fat milk ( Bio-Rad ) in TBST and incubated overnight at 4°C in primary antibodies ( 1:5000 rabbit anti-GFP , ThermoFisher A11122 , RRID:AB_221569; 1:2000 rabbit anti-Flag , Sigma F7425 , RRID:AB_439687 ) . Proteins were detected using HRP-coupled secondary antibody ( 1:15 , 000 goat anti-rabbit , Jackson ImmunoResearch Labs 111-035-144 , RRID:AB_2307391 ) . Chemiluminescence was detected using Amersham ECL reagent ( GE Healthcare ) . Dechorionated embryos were injected at the 1-cell stage with 5 , 20 or 50 pg of mRNA encoding epitope-tagged constructs and grown to 50% epiboly . 50–100 embryos were transferred to 400 μL of cold lysis buffer ( 50 mM Tris at pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 10% glycerol , 1% Triton X-100 and protease inhibitors , Sigma 11836170001 ) and crushed using a homogenizer and disposable pestles before incubation on ice for 30 min with vortexing every 5 min . Samples were spun at maximum speed at 4°C for 30 min and the supernatant was transferred to tubes containing 50 μL of anti-HA affinity matrix ( Roche 11815016001 , RRID:AB_390914 ) that was pre-washed twice in lysis buffer . Samples were placed on a rotating platform at 4°C overnight . The matrix was spun down for 2 min at 3000 rcf and washed in 600 μL of cold wash buffer ( 50 mM Tris at pH 7 . 5 , 150 mM NaCl , 1% Triton X-100 and protease inhibitors ) 5 times . 2x SDS loading buffer and DTT ( 150 mM final concentration ) was added to the matrix in 10 μL of wash buffer . Immunoblots were performed as above . Embryos were injected at the 1-cell stage with sqt , cyc and/or vg1 mRNAs and grown to 50% epiboly . For the sqt experiment ( Figure 2E ) 2 sets of 10 embryos were collected per condition; for the cyc experiment ( Figure 2F ) 2 sets of 12 embryos were collected per condition . Embryos were flash frozen in liquid nitrogen and RNA was extracted using an E . Z . N . A . Total RNA Kit ( Omega ) and reverse transcription was carried out using an iScript cDNA Synthesis Kit ( Bio-Rad ) . qPCR reactions were run on a CFX96 machine ( Bio-Rad ) using iTaq Universal SYBR Green Supermix ( Bio-Rad ) and 0 . 25 μM of primers ( see primer sequences below ) . Gene expression levels were calculated relative to a reference gene , ef1a . The mean and standard error of the mean was plotted for each condition . Two technical replicates in addition to biological replicates were used per condition . Both experiments were performed multiple times . vg1_genotype_FCCTGTGTGTGTTCTTTGCTCTGvg1_genotype_RCTGTTTAAAGATTTTCCACATCTGTGef1a_qPCR_FAGAAGGAAGCCGCTGAGATGGef1a_qPCR_RTCCGTTCTTGGAGATACCAGCClft1_qPCR_FGAGATGGCCAAGTGTGTCCAlft1_qPCR_RCTGCAGCACATTTCACGGTC In the ‘primed model’ , Vg1 is already present in excess when Nodal production begins . This model describes the dynamics of Nodal monomers ( N ) , Vg1 monomers ( V ) and Nodal-Vg1 dimers ( D ) . dNdt=λN-βNN-λDNVdVdt= -βVV-λDNVdDdt= λDNV Assumptions: constitutive production of N ( rate λN ) , first-order degradation ( component half-lives of ln2/βN and ln2/βV , respectively ) and bimolecular heterodimerization with rate λDNV . Vg1 is assumed to be maternally deposited , and is thus provided to the system via the initial conditions . In the ‘cold-start’ model , Nodal monomers accumulate and dimerize after the onset of Nodal production . This model describes the dynamics of Nodal monomers ( N ) , Vg1 monomers ( V ) and Nodal-Nodal dimers ( D ) . dNdt=λN-βNN-λDN2dDdt= λDN2 Assumptions: constitutive production of N ( rate λN ) , first-order degradation ( component half-life of ln2/βN ) and bimolecular homodimerization with rate λDN2 .	All animals begin life as just one cell – a fertilized egg . In order to make a recognizable adult , each embryo needs to make the three types of tissue that will eventually form all of the organs: endoderm , which will form the internal organs; mesoderm , which will form the muscle and bones; and ectoderm , which will generate the skin and nervous system . All vertebrates – animals with backbones like fish and humans – use the so-called Nodal signaling pathway to make the endoderm and mesoderm . Nodal is a signaling molecule that binds to receptors on the surface of cells . If Nodal binds to a receptor on a cell , it instructs that cell to become endoderm or mesoderm . As such , Nodal is critical for vertebrate life . However , there has been a 30-year debate in the field of developmental biology about whether a protein called Vg1 , which has a similar molecular structure as Nodal , plays a role in the early development of vertebrates . Zebrafish are often used to study animal development , and Montague and Schier decided to test whether these fish need the gene for Vg1 ( also known as Gdf3 ) by deleting it using a genome editing technique called CRISPR/Cas9 . It turns out that female zebrafish can survive without this gene . Yet , when the offspring of these females do not inherit the instructions to make Vg1 from their mothers , they fail to form the endoderm and mesoderm . This means that the embryos do not have hearts , blood or other internal organs , and they die within three days . Two other groups of researchers have independently reported similar results . The findings reveal that Vg1 is critical for the Nodal signaling pathway to work in zebrafish . Montague and Schier then showed that , in this pathway , Nodal does not activate its receptors on its own . Instead , Nodal must interact with Vg1 , and it is this Nodal-Vg1 complex that activates receptors , and instructs cells to become endoderm and mesoderm . Scientists currently use the Nodal signaling pathway to induce human embryonic stem cells growing in the laboratory to become mesoderm and endoderm . As such , these new findings could ultimately help researchers to grow tissues and organs for human patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2017	Vg1-Nodal heterodimers are the endogenous inducers of mesendoderm
Prairie vole breeder pairs form monogamous pair bonds , which are maintained through the expression of selective aggression toward novel conspecifics . Here , we utilize behavioral and anatomical techniques to extend the current understanding of neural mechanisms that mediate pair bond maintenance . For both sexes , we show that pair bonding up-regulates mRNA expression for genes encoding D1-like dopamine ( DA ) receptors and dynorphin as well as enhances stimulated DA release within the nucleus accumbens ( NAc ) . We next show that D1-like receptor regulation of selective aggression is mediated through downstream activation of kappa-opioid receptors ( KORs ) and that activation of these receptors mediates social avoidance . Finally , we also identified sex-specific alterations in KOR binding density within the NAc shell of paired males and demonstrate that this alteration contributes to the neuroprotective effect of pair bonding against drug reward . Together , these findings suggest motivational and valence processing systems interact to mediate the maintenance of social bonds . The ability to maintain meaningful social bonds is a critical component of human health and mental well being , yet the neural capacity to maintain such relationships is not well understood . The socially monogamous prairie vole ( Michrotus ochrogaster ) presents an ideal animal model to study the neural correlates of social bond maintenance because , unlike most mammals ( Kleiman , 1977 ) , prairie voles form selective and enduring attachments to their mating partner ( Aragona et al . , 2009 ) . In both the field and laboratory , the maintenance of these bonds is associated with the expression of selective aggression towards novel conspecifics as well as selective affiliation with the mating partner ( i . e . , mate guarding ) ( Carter and Getz , 1993 ) . Importantly , the expression of selective aggression provides a robust and reliable assay that can be utilized in a laboratory setting to deconstruct neural signaling pathways involved in the regulation of social bond maintenance . To date , laboratory studies have identified that the expression of selective aggression , and therefore pair bond maintenance , requires the activation of both D1-like dopamine ( DA ) and kappa-opioid receptors ( KORs ) within the nucleus ( NAc ) shell as blockade of either one of the receptors attenuates aggressive rejection of novel conspecifics ( Aragona et al . , 2006; Resendez et al . , 2012 ) . Thus , regulation of pair bond maintenance requires neural systems that code evaluation of salient environmental stimuli as well as those that are important for the generation of motivational states ( Resendez and Aragona , 2013 ) . Interestingly , in other animal models , these receptor systems have been shown to directly interact at the molecular level ( Gerfen et al . , 1990; Carlezon et al . , 1998 ) as well as in the transition between motivational states ( Chartoff et al . , 2016 ) . However , it is unknown if similar interactions occur in the regulation of pair bond maintenance . This study therefore endeavored to examine pair bond induced neural plasticity within the DA and dynorphin/KOR systems as well as how these systems interact to mediate the expression of selective aggression , a well established indicator of a fully established pair bond . Given that activation of KORs is associated with aversive states ( Mucha and Herz , 1985; Pfeiffer et al . , 1986; Shippenberg and Herz , 1986; Bals-Kubik et al . , 1989 ) , we first determined if activation of NAc KORs prior to pairing with a novel social stimulus is sufficient to tag a recently encountered social stimulus as aversive . Next , to assess how the establishment of a pair bond alters both motivational ( DA ) and aversive ( dynorphin/KOR ) processing systems , we conducted extensive anatomical , neurochemical , and functional comparisons within the striatum of male and female prairie voles . In total , we conducted mRNA expression analysis ( RT-qPCR ) , protein binding measurements ( receptor autoradiography ) , and measures of DA concentration ( fast-scan cyclic-voltammetry ) to identify sex-specific alterations within the DA and dynorphin/KOR systems of pair bonded voles . We next utilized site-specific behavioral pharmacology to examine interactions between NAc shell D1-like and KORs in the expression of selective aggression . Finally , in male prairie voles , we show that pair bonding , but not other social manipulations , decreases the rewarding properties of the psychostimulant amphetamine and that this attenuation requires the activation of NAc shell KORs . In total , the present study demonstrates that the development of a pair bond is underpinned by sex-specific modifications in motivational ( DA/D1 ) and valence ( dynorphin/KOR ) processing systems , that these systems interact to mediate selective aggression in both sexes , and that male specific alterations in the dynorphin/KOR system buffers against the rewarding properties of amphetamine . Activation of NAc shell KORs is required for the expression of selective aggression by pair bonded voles ( Resendez et al . , 2012 ) ; however , the psychological processes that underlie the expression of this behavior are not well understood . In other rodent species , activation of these receptors has been shown to induce aversion as well as mediate avoidance behaviors ( Land et al . , 2008; Al-Hasani et al . , 2015 ) . For example , pairing of a previously neutral stimulus with either an aversive experience that results in KOR activation , such as stress , or with direct pharmacological activation of these receptors results in the avoidance of that stimulus during subsequent encounters ( Land et al . , 2008 ) . Given the known relationship between aversive processing of environmental stimuli , avoidance behaviors ( Boren et al . , 1959; D'Amato et al . , 1967 ) , and KOR activation , we hypothesized that one mechanism in which NAc shell KORs mediate social avoidance behaviors is through the encoding of novel social stimuli as aversive . To determine if activation of NAc shell KORs during a social encounter results in social avoidance behaviors , we utilized a modified version of the partner preference paradigm ( Figure 1a ) . Specifically , we employed a social pairing condition ( 1 hr cohabitation with an opposite sex conspecific ) that is insufficient to produce a preference for the familiar partner over an unfamiliar conspecific ( the stranger ) . A lack of a preference for either social stimulus is indicated by equivalent amounts of time spent with the partner and stranger during the social choice test , suggesting that both social stimuli are of equal valences . As expected , a Wilcoxon signed rank sum test for non-parametric data demonstrated that control males treated with aCSF did not show a preference for either individual ( W ( 5 ) = 33 , z = −0 . 97 , p=0 . 33 ) ( Figure 1b , c ) . Conversely , male subjects that were administered a KOR agonist ( 1 μg U50 , 488 ) into the NAc shell immediately prior to pairing with the female partner avoided the female that had been paired with KOR activation and therefore displayed a robust preference for contact with the novel female that had not been previously paired with NAc shell KOR ( Wilcoxon signed rank sum test , W ( 6 ) = 32 . 5 , z = -2 . 56 , p=0 . 01 ) ( Figure 1b , c ) . In addition to differences in direct contact time , activation of NAc shell KORs prior to pairing with a social stimulus also resulted in differences in the duration of time spent in each stimulus chamber ( two-way ANOVA , ( F ( 2 , 36 ) = 7 . 07 , p=0 . 003 ) . Specifically , males that received administration of a KOR agonist avoided the chamber containing the partner ( Bonferroni’s post hoc test , p=0 . 02 ) and spent more time in the chamber occupied by the stranger ( p=0 . 0006 ) ( Figure 1d ) . Finally , control males and males receiving site-specific administration of the KOR agonistdid not differ in total contact time ( time spent with partner + time spent with stranger ) ( t-test , t ( 11 ) = 0 . 35 , p=0 . 73 ) , indicating that reduced contact with the partner did not result from a general decrease in motivation for social contact ( Figure 1e ) . Both groups of male subjects also did not differ in affiliative social behavior or grooming behavior during the 1 hr cohabitation ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 15325 . 003Figure 1 . NAc shell KORs encode social aversion . ( a ) Experimental design . ( b ) Histological verification of injection sites . ( c ) Control males ( aCSF ) paired with a female partner for 1 hr showed no social preference or aversion ( n = 6 ) . In contrast , activation of NAc shell KORs via site-specific administration of a KOR agonist induced a partner aversion ( n = 7 ) . ( d ) Males that received site-specific injections of the KOR agonist also spent significantly less time in the partner’s cage as well as more time in the chamber containing the stranger . ( e ) There was no difference in total contact time between the two groups . Summary data are presented as mean ± SEM . p<0 . 05 , p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 00310 . 7554/eLife . 15325 . 004Figure 1—figure supplement 1 . Social and grooming behavior during 1-hr cohabitation . Prior to partner preference testing , subjects were administered either aCSF ( control ) or a KOR agonist ( 1 μg U50 , 488 ) into the NAc shell and paired with a female partner for 1 hr . ( a , b ) Control males ( aCSF ) and males who were administered a KOR agonist into the NAc shell prior to pairing with the female partner did not differ in the duration of time spent engaging in affiliative behavior at ( a ) any 10-min time bin ( two-way ANOVA , F ( 1 , 12 ) = 0 . 55 , p = 0 . 47 ) or ( b ) over the entire duration of the 60-min pairing ( t-test , t ( 12 ) = 0 . 74 , p = 0 . 47 ) . ( c , d ) Similarly , there was no difference in the duration of time spent autogrooming at ( c ) any 10-min time bin ( two-way ANOVA , F ( 1 , 12 ) = 0 . 11 , p = 0 . 74 ) or ( d ) over the entire duration of the 60-min pairing ( t-test , t ( 12 ) = 0 . 34 , p = 0 . 74 ) ( n = 6–8/group ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 004 Together , these data suggest that activation of KORs within the NAc shell induces social avoidance behaviors , potentially through the assignment of negative valence onto a previously neutral social stimulus . Given that activation of NAc shell KORs is required for the expression of selective aggression in pair bonded voles , it is possible that KOR activation within the NAc shell mediates pair bond maintenance by assigning social stimuli other than the mating partner with a negative valence signal . We therefore conducted our next series of experiments to determine how pair bonding alters neural systems involved in the regulation of selective aggression to promote pair bond maintenance . Sexually naïve prairie voles find social novelty rewarding and will readily approach and interact with novel conspecifics . In stark contrast , a pair bonded vole will avoid and aggressively reject this same social stimulus , suggesting that they find social stimuli—other than their mating partner or offspring—to be aversive ( Resendez and Aragona , 2013 ) . We therefore hypothesized that this behavioral transformation is mediated by an up-regulation of neural systems that regulate the expression of selective aggression , such as both the D1-like receptor and dynorphin/KOR systems within the ventral region of the striatum . Thus , to determine if non-pair bonded ( sibling housed ) and pair bonded ( 2 weeks cohabitation with an opposite-sex conspecific ) voles differ in the expression level of mRNA for genes that encode proteins involved in the regulation of selective aggression , we utilized RT-qPCR to compare the level of mRNAs related to the DA and dynorphin/KOR systems . For all groups , comparisons were made within the dorsal and ventral striatum ( i . e . , NAc ) . Extensive cohabitation with a mating partner predominately altered the expression of mRNA for genes that code for proteins associated with pair bond maintenance . Specifically , within the ventral striatum , t-test results show that pair bonded males and females showed higher levels of mRNA for the gene encoding dynorphin ( Pdyn ) ( Male: t ( 24 ) = 2 . 26 , p=0 . 03; t ( 26 ) = 3 . 05 , p=0 . 005 ) , the endogenous ligand for KORs . They also showed elevated levels of mRNA expression for the gene encoding D1-like receptors ( Drd1 ) ( t-test; Male: t ( 24 ) = 2 . 86 , p=0 . 009; Female: t ( 28 ) = 3 . 42 , p=0 . 002 ) ( Figure 2a and c ) . For paired males , similar elevations in mRNA for the gene that encodes KOR ( Oprk1 ) also occurred; however , due to high levels of variability in the expression of this gene , elevations in Oprk1 mRNA did not significantly differ from non-paired males ( Table 1 ) . Only moderate elevations in Oprk1 mRNA levels occurred in females and this elevation failed to reach significance ( Table 1 ) . Finally , differences in the expression of mRNA for Drd1 and Pdyn were not identified within the dorsal striatum indicating that these changes are specific to the ventral region of the striatum ( Table 2 ) . 10 . 7554/eLife . 15325 . 005Figure 2 . Pair bonding alters mRNA expression within the NAc . ( a ) Pair bonding increased the expression of Drd1 and Pdyn mRNA within the VS of males ( n = 15/group ) . ( b ) Pair bonding decreased Drd3 mRNA expression within the DS of males ( n = 15–16/group ) . ( c ) Similar to males , pair bonding increased the expression of Drd1 and Pdyn within the VS of females ( n = 6–23/group ) . ( d ) Pair bonding significantly decreased Drd2 mRNA within the DS of paired females ( n = 16/group ) . p<0 . 05 , *p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 00510 . 7554/eLife . 15325 . 006Table 1 . Non-significant statistics for mRNA comparisons in the ventral striatum . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 006SexGeneMaleFemalePdynNANAPenkt ( 24 ) = 1 . 80 , p = 0 . 09t ( 26 ) = 1 . 92 , p = 0 . 07Oprk1t ( 24 ) = 1 . 99 , p = 0 . 06t ( 11 ) = 0 . 36 , p = 0 . 72Oprm1t ( 24 ) = 0 . 13 , p = 0 . 90t ( 26 ) = 0 . 70 , p = 0 . 49Drd1NANADrd2t ( 24 ) = 0 . 10 , p = 0 . 33t ( 37 ) = 1 . 57 , p = 0 . 13Drd3t ( 24 ) = 1 . 58 , p = 0 . 13t ( 26 ) = 0 . 75 , p = 0 . 46Oxtrt ( 24 ) = 1 . 72 , p = 0 . 10t ( 37 ) = 1 . 12 , p = 0 . 27Avpr1at ( 24 ) = 0 . 82 , p = 0 . 43t ( 37 ) = 0 . 25 , p = 0 . 81Nadht ( 23 ) = 1 . 23 , p = 0 . 23t ( 28 ) = 0 . 79 , p = 0 . 4410 . 7554/eLife . 15325 . 007Table 2 . Non-significant statistics for mRNA comparisons in the dorsal striatum . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 007SexGeneMaleFemalePdynt ( 26 ) = 0 . 80 , p = 0 . 43t ( 26 ) = 0 . 21 , p = 0 . 83Penkt ( 26 ) = 0 . 56 , p = 0 . 58t ( 26 ) = 0 . 13 , p = 0 . 90Oprk1t ( 26 ) = 0 . 17 , p = 0 . 86t ( 26 ) = 1 . 19 , p = 0 . 24Oprm1t ( 26 ) = 0 . 63 , p = 0 . 53t ( 26 ) = 0 . 05 , p = 0 . 96Drd1t ( 26 ) = 1 . 15 , p = 0 . 26t ( 26 ) = 0 . 88 , p = 0 . 39Drd2t ( 26 ) = 0 . 18 , p = 0 . 86NADrd3NAt ( 26 ) = 1 . 69 , p = 0 . 10Oxtrt ( 26 ) = 0 . 20 , p = 0 . 84t ( 26 ) = 1 . 50 , p = 0 . 15Avpr1at ( 26 ) = 0 . 56 , p = 0 . 58t ( 26 ) = 0 . 85 , p = 0 . 40Nadht ( 26 ) = 0 . 97 , p = 0 . 34t ( 26 ) = 0 . 005 , p = 0 . 10 To next determine if differences in expression differences following 2 weeks of male-female cohabitation were specific to genes that encode proteins involved in the regulation of pair bond maintenance , we also examined the expression of genes that encode proteins that regulate social behaviors associated with pair bond formation . Within the NAc , these proteins include D2-like DA receptors , mu-opioid receptors ( MORs ) , and the oxytocin receptor . Following 2 weeks of male-female cohabitation , differences in the expression of genes related to pair bond formation ( Drd2 , Penk/Oprm1 , Oxtr ) were not found within the ventral striatum of pair bonded voles indicating that the differences identified above are specific to neural systems that regulate pair bond maintenance ( Table 1 ) . Also in contrast to the above findings , sex-specific alterations in the expression of genes related to pair bond formation were identified within the dorsal striatum . Specifically , compared to non-paired subjects , pair bonded males had higher levels of Drd3 mRNA ( t-test; t ( 26 ) = 2 . 34 , p=0 . 03 ) while pair bonded females had higher levels of Drd2 mRNA ( t-test; t ( 26 ) = 2 . 12 , p=0 . 04 ) ( Figure 2b , d ) . No other differences were identified within the dorsal striatum . Overall , the above findings are consistent with the proposed mechanism that the establishment of a pair bond is associated with region specific alterations in neural systems that regulate selective aggression . However , an up-regulation in the expression of mRNA is not always indicative of an increase in protein levels . We therefore utilized receptor autoradiography to examine pair bond induced differences in KOR binding density within the striatum . We focused on KORs in the present study because it has previously been shown that pair bonding increases the expression of D1-like receptors specifically within the ventral striatum ( Aragona et al . , 2006 ) . To determine whether pair bonding alters striatal KOR density , KOR binding densities were compared between non-paired ( i . e . , same-sex sibling housed ) and pair bonded prairie voles ( i . e . , 2 weeks male-female cohabitation ) of both sexes . Compared to non-paired ( sibling housed ) males , a two-way ANOVA indicated that pair bonded males had lower levels of striatal KOR binding density ( F ( 1 , 120 ) = 17 . 51 , p=0 . 0001; Figure 3a , b ) . Further examination of pair bond induced alterations in KOR binding density within the striatum of males revealed that the decrease in KOR binding was specific to the ventral region of the NAc shell ( Bonferroni’s post hoc test , p=0 . 01; Figure 3b; Table 3 ) , the region of the striatum where KORs act to regulate selective aggression ( Resendez et al . , 2012 ) and mediate aversion ( Al-Hasani et al . , 2015 ) . 10 . 7554/eLife . 15325 . 008Figure 3 . Pair bonding alters KOR binding within the striatum of males . ( a , b ) Pair bonding decreased KOR binding in the dorso-medial and ventral NAc shell of males ( n = 11/group ) . ( c , d ) There was no significant effect on KOR binding density in females ( n = 10/group ) . Summary data are presented as mean ± SEM . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 00810 . 7554/eLife . 15325 . 009Figure 3—figure supplement 1 . Sex differences in KOR binding density before and after pair bonding . ( a ) Representative autoradiographs from sibling ( top ) and pair housed ( bottom ) male ( left ) and female ( right ) prairie voles . ( b ) Prior to pair bonding , sibling housed male and female prairie voles show a sex difference in KOR binding density within the striatum ( two-way ANOVA , F ( 1 , 114 ) = 38 . 14 , p=0 . 0001 ) . Specifically , compared to sibling housed females ( i . e . , non-paired ) , sibling housed males have higher levels of KOR binding density within ventral regions of the striatum . Bonferroni’s post hoc test revealed that within the striatum , sibling housed males had significantly higher levels of KOR binding density within the NAc core ( p=0 . 02 ) , the dorso-medial region of the NAc shell ( p=0 . 0001 ) , and the ventral NAc shell ( p=0 . 005 ) . ( c ) Interestingly , following the establishment of a pair bond , these sex differences no longer exist ( two-way ANOVA , F ( 1 , 114 ) = 0 . 36 , p=0 . 55 ) . These results are consistent with findings presented in Figure 3A , which demonstrate that pair bonding decreases KOR binding density within the ventral striatum of males while leaving the KOR binding density in females unaltered . Together , these data suggest that pair bonding in males lowers KOR binding density to the levels of females regardless of the female’s social housing condition . Summary data are presented as mean ± SEM . p<0 . 05 , p<0 . 005 , *p<0 . 0005 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 00910 . 7554/eLife . 15325 . 010Table 3 . Non-significant statistics for comparisons of KOR binding density in paired versus unpaired males . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 010Striatal sub-regionBonferonni's post hoc testDorso-medial striatump>0 . 99Dorso-lateral striatump>0 . 99NAc corep = 0 . 41Nac dorso-medial shellp = 0 . 07NAc lateral shellp = 0 . 99 In contrast to paired males , significant alterations in KOR binding density following the establishment of a pair bond were not identified in females ( F ( 1 , 108 ) = 3 . 50 , p>0 . 06; Figure 3c , d ) suggesting that pair bonding induces sex-specific alterations in KOR binding density . We therefore next compared KOR binding density between males and females before and after the establishment of a pair bond . Prior to the establishment of a pair bond , a two-way ANOVA indicated that non-paired ( sibling housed ) males have significantly higher levels of KOR binding density within the striatum compared to non-paired females ( two-way ANOVA , F ( 1 , 114 ) = 38 . 14 , p=0 . 0001 ) . Specifically , non-paired males had significantly higher levels throughout the NAc , including the NAc core ( Bonferroni’s post hoc test; p=0 . 02 ) , the dorso-medial region of the NAc shell ( p=0 . 0001 ) , and the ventral region of the NAc shell ( p=0 . 005 ) . Interestingly , these sex differences in KOR binding density were not identified in pair bonded males and females as males no longer showed higher levels in KOR binding density ( two-way ANOVA , F ( 1 , 114 ) = 0 . 36 , p=0 . 55 ) ( Figure 3—figure supplement 1 ) . Together , these data suggest that pair bonding results in a reduction in KOR binding density within the NAc of male , but not female prairie voles . Previous studies have established an essential role for the activation of NAc shell D1-like receptors in the expression of social behaviors important for pair bond maintenance ( Aragona et al . , 2009 ) . These receptors are primarily of the low-affinity sub-type ( Richfield et al . , 1989 ) and their activation requires high concentrations of DA to be released , such as that which occurs during burst firing of DA neurons ( Gonon , 1997; Cheer et al . , 2007 ) . Given that activation of D1-like receptors require high concentrations of DA release and that selective aggression is only expressed in the pair bonded state , we predicted that pair bonded voles would have greater concentrations in DA release specifically within the NAc shell . To compare DA release dynamics between non-bonded and pair bonded voles , we utilized fast-scan cyclic-voltammetry ( FSCV ) to measure real-time DA release across the striatum . However , given that striatal DA release properties are unknown in this species , we first conducted a detailed characterization of DA release dynamics within the prairie vole striatum ( Figure 4a , b ) . 10 . 7554/eLife . 15325 . 011Figure 4 . Striatal DA transmission in non-pair bonded prairie voles . ( a , b ) Representative color plots of DA transmission throughout the striatum of ( a ) male and ( b ) female prairie voles . ( c , e ) A 1-pulse depolarizing stimulation evokes the greatest magnitude of DA release within the dorsal striatum and the magnitude of this release decreases along a dorsal to ventral gradient within the striatum of ( c ) males and ( e ) females . ( d , f ) An inverse relationship is seen with burst facilitation as the greatest ratio of DA release occurs within the NAc shell , an intermediate ratio occurs within the NAc core , and the lowest ratio occurs within the dorsal striatum of ( d ) males and ( f ) females . ( g–i ) Compared to male prairie voles , a 20-pulse stimulation evokes a greater magnitude of DA release within the ( g ) dorsal striatum and the ( h ) NAc core of females . ( i ) No sex difference in DA transmission occurred within the NAc shell . Summary data are presented as mean ± SEM . p<0 . 05 , p<0 . 005 , p<0 . 0005 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 011 Consistent with other mammals ( Jones et al . , 1995; Calipari et al . , 2012 ) , the concentration of striatal DA release evoked by a single pulse stimulation ( [DA]1p ) significantly decreased along a dorsal to ventral gradient ( one-way ANOVA; Male: F ( 2 , 30 ) = 17 . 28 , p<0 . 000; Female: F ( 2 , 27 ) = 8 , 57 , p=0 . 001 ) . Post hoc Tukey comparisons revealed that both the NAc core ( Male: p=0 . 009; Female: p=0 . 04 ) and the NAc shell ( Male: p=0 . 000; Female: p=0 . 001 ) had significantly lower levels of stimulated DA release compared to the dorsal striatum ( Figure 4c , e ) . Also , consistent with other species ( Zhang et al . , 2009 ) , the magnitude of DA release following stimulation parameters that evoke burst-like firing of DA neurons , such as an extra-physiological 20-pulse stimulation ( [DA]20p ) , differed across striatal sub-regions , with the most robust impact occurring within the NAc shell , a moderate impact within the NAc core , and a minimal effect within the dorsal striatum ( one-way ANOVA , Male: F ( 2 , 28 ) = 11 . 22 , p=0 . 0003; Female: F ( 2 , 25 ) = 10 . 60 , p<0 . 000 ) . This effect is represented by the greatest ratio of evoked DA release ( [DA]20p/[DA]1p ) within the NAc shell ( Tukey post hoc test; Male: p=0 . 0002; Female: p=0 . 0004 ) , an intermediate ratio within the NAc core ( Tukey post hoc test; Male: p=0 . 02; Female: p=0 . 03 ) , and the lowest ratio within the dorsal striatum ( Figure 4d , f ) . In addition , a two-way ANOVA followed by Bonferroni’s post hoc tests identified significant sex-differences in striatal DA release following a 20-pulse stimulation within the dorsal striatum ( F ( 1 , 10 ) = 5 . 25 , p=0 . 002; p=0 . 03 , Figure 4g ) as well as the NAc core ( F ( 1 , 34 ) = 4 . 05 , p=0 . 05; p=0 . 05 , Figure 4h ) , but not the NAc shell ( F ( 1 , 32 ) = 1 . 77 , p=0 . 1 , Figure 4i ) . Similar sex differences have previously been reported in other species ( Walker et al . , 2000 ) . Overall these results suggest that general striatal DA release patterns appear to be conserved among rodents . Next , to test the hypothesis that pair bonded voles have elevated DA release specifically within the NAc shell of the striatum , electrically evoked DA release was compared across striatal sub-regions of pair bonded and non-pair bonded voles ( Figure 5a , b ) . As predicted , t-test comparisons indicated that pair bonding significantly increased peak DA release within the NAc shell of pair bonded voles ( Male: t ( 17 ) = 2 . 44 , p=0 . 03; Female: t ( 13 ) = 2 . 48 , p=0 . 03 ) , but not other regions of the striatum ( Dorsal striatum male: t ( 21 ) = 0 . 09 , p=1 . 75; Dorsal striatum female: t ( 17 ) = 1 . 26 , p=0 . 22; NAc core male: t ( 18 ) = 0 . 87 , p=0 . 40; NAc core female: t ( 15 ) = 0 . 73 , p=0 . 48 ) ( Figure 5c–h ) . Additionally , although pair bonding significantly elevated NAc shell DA release in both sexes , the average percent increase was lower in males ( 34% ) compared to females ( 99% ) ( Figure 5e , h ) . Direct comparisons of peak DA release between pair bonded males and females indicated that pair bonded females had significantly higher levels of DA release within the NAc shell compared to that of pair bonded males ( Figure 5—figure supplement 1 ) . This sex difference in pair bond induced changes in DA transmission is unlikely due to initial sex differences in NAc shell DA release as differences in DA release within the NAc shell were not identified between non-paired male and females ( Figure 4 ) . Moreover , given that the release of DA is required for the activation of D1-like receptors that mediate selective aggression and that displays of selective aggression are qualitatively larger in pair bonded males than females ( Figure 5—figure supplement 2 ) , we initially expected increases in DA transmission to be greater in males . 10 . 7554/eLife . 15325 . 012Figure 5 . Pair bonding enhances NAc shell DA release . ( a , b ) Representative color plots of stimulated DA release following a 1-pulse depolarizing stimulation in ( a ) male and ( b ) female subjects . For both sexes , top row shows representative color plots for non-paired subjects and bottom row shows representative color plots for pair bonded subjects . ( c , d ) Pair bonding had no effect on DA transmission within the ( c ) dorsal striatum ( n = 11–12/group ) or ( d ) the NAc core of males ( n = 10/group ) . ( e ) Within the NAc shell , a 1-pulse stimulation resulted in significantly greater DA release within the NAc shell of paired males compared to non-paired male controls ( n = 9–10/group ) . ( f , g ) There was no difference in peak DA release between non-paired and pair bonded females following a 1-pulse stimulation within the ( f ) dorsal striatum ( n = 8–11/group ) or ( g ) the NAc core ( n = 8–9/group ) . ( h ) Similar to males , a 1-pulse depolarizing stimulation resulted in a greater level of DA release within of the NAc shell of paired females compared to non-paired females ( n = 7–8/group ) . Summary data are presented as mean ± SEM . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 01210 . 7554/eLife . 15325 . 013Figure 5—figure supplement 1 . Sex differences in striatal dopamine release following the establishment of a pair bond . ( ac ) Representative traces showing changes in DA release following a 1-pulse depolarizing stimulation applied to either the ( a ) dorsal striatum , ( b ) the NAc core , or ( c ) the NAc shell . ( d , e ) Within both the ( d ) dorsal striatum ( t-test , t ( 18 ) = 0 . 06 , p = 0 . 95; n = 12–8 ) and the ( e ) NAc core ( t-test , t ( 16 ) = 1 . 16 , p = 0 . 26; n = 10–8 ) , pair bonding did not result in significant differences in peak DA release between paired males and females . ( f ) However , following the establishment of a pair bond , a 1-pulse depolarizing stimulation resulted in significantly greater levels of peak DA release within the NAc shell of paired females compared to paired males ( t-test , t ( 14 ) = 2 . 60 , p = 0 . 02; n = 9–7 ) . Summary data are presented as mean ± SEM . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 01310 . 7554/eLife . 15325 . 014Figure 5—figure supplement 2 . Pair bonding increases selective aggression in both male and female prairie voles . ( a , b ) Prior to pair bonding , both males and females show low levels of aggression . However , following the establishment of a pair bond , both ( a ) males ( Mann-Whitney test , U ( 1 , 22 ) = 33 , z = −2 . 26 , p = 0 . 03 ) ( n = 11–13/group ) and ( b ) females increased levels of selective aggression toward novel conspecifics ( Mann-Whitney test , U ( 1 , 15 ) = 11 , z = −2 . 34 , p = 0 . 03 ) ( n = 6–11/group ) with males showing qualitatively higher levels of aggression . Summary data are presented as mean ± SEM . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 014 One possible explanation underlying sex differences in pair bond induced alterations in DA transmission is that the influence of fecundity on pair bond strength differs between males and females ( Resendez et al . , 2012; McCracken et al . , 2015 ) . More specifically , for pair bonded males , but not females , the strength of the pair bond , as indicated by the magnitude of selective aggression displayed toward intruders , is dependent on pair fecundity . We therefore tested the hypothesis that variations in pair bond induced increases in DA release between males and females were associated with reproductive success . Prior to FSCV recordings of stimulated DA release in striatal slices , fecundity of the pair was assessed by determining the stage of pregnancy following 2 weeks of male-female cohabitation . Briefly , the stage of pregnancy was determined as previously described by measuring the average neonatal weight of the offspring , with larger neonatal weights indicating shorter delays in the onset of pregnancy ( Curtis , 2010; Resendez et al . , 2012 ) . Measures of neonatal weight were then used to classify the pairs as either optimally ( mating and fertilization occurring within 48–72 hr of pairing ) or sub-optimally ( delay in establishment of pregnancy ) pregnant ( Resendez et al . , 2012 ) . Following 2 weeks of cohabitation with an opposite sex partner , males from optimally pregnant pairs showed significantly higher levels of aggression than males from sub-optimally pregnant pairs ( t-test , t ( 9 ) = 2 . 54 , p=0 . 03 ) ( Figure 6a ) . In contrast , reproductive status had no impact on pair bond strength in paired females as females from optimally and sub-optimally pregnant pairs did not differ in levels of selective aggression ( t-test , t ( 9 ) = 0 . 24 , p=0 . 82 ) ( Figure 6b ) . Moreover , direct comparisons of aggression levels among males and females from sub-optimally and optimally pregnant pairs indicates that sex differences in the magnitude of selective aggression that is displayed toward an intruder depends on pair fecundity ( two way ANOVA , F ( 1 , 37 ) = 8 . 32 , p=0 . 007 ) . Specifically , although paired males are generally more aggressive than paired females , when aggression levels were further compared by fecundity classification , only males from optimally pregnant pairs showed significantly higher levels of selective aggression than females ( Bonferroni’s post hoc test , optimally pregnant: p=0 . 01 , sub-optimally pregnant: p>0 . 99 ) ( Figure 6—figure supplement 1 ) . Thus , pair fecundity strongly influences pair bond strength in male , but not female prairie voles and only males from optimally pregnant pairs shower higher levels of aggression than paired females . We next determined if fecundity also resulted in sex specific alterations in DA transmission . 10 . 7554/eLife . 15325 . 015Figure 6 . Relationship between striatal DA release and characteristics of pair bonding . ( a ) Pair bond induced increases in selective aggression was dependent on fecundity in males as males from optimally pregnant pairs were more aggressive than males from sub-optimally pregnant pairs ( n = 5–6/group ) . ( b ) Conversely , pregnancy optimality had no effect on attack frequency in females ( n = 4–7/group ) . ( c ) Within the NAc shell , males whose females were optimally pregnant showed significantly greater levels of DA release ( n = 4–18/group ) . ( d ) In contrast , for females , there was no difference in peak DA release within the NAc shell between non-paired females and paired females categorized by their reproductive status ( n = 5–13/group ) . ( e ) Among pair bonded males , neonatal weight ( an established indicator of gestational stage ) was positively correlated with peak DA release within the NAc shell ( n = 23 ) . ( f ) However , there was no relationship between peak DA release and reproductive status in paired females . ( g , h ) Finally , in relation to attack frequency , there was a positive correlation between peak DA release and attack frequency within the within the NAc shell of ( g ) paired males ( n = 8 ) , but no such relationship was identified among paired females ( n = 10 ) . Summary data are presented as mean ± SEM . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 01510 . 7554/eLife . 15325 . 016Figure 6—figure supplement 1 . Sex differences in selective aggression by fecundity . ( a ) When selective aggression levels were compared between paired males and females based on fecundity ( suboptimal versus optimal ) sex differences in selective aggression emerged as a function of pregnancy status ( two-way ANOVA , F ( 1 , 18 ) = 6 . 01 , p = 0 . 02 ) ( n = 6–13/group ) . Specifically , paired males and females categorized as sub-optimally pregnant did not significantly differ in levels of selective aggression ( Bonferroni’s post hoc tes , p>0 . 99 ) . In contrast , males from pairs with optimally pregnant females showed robustly higher levels of selective aggression than paired females categorized as optimally pregnant ( p = 0 . 01 ) ( n = 4–7/group ) . Summary data are presented as mean ± SEM . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 01610 . 7554/eLife . 15325 . 017Figure 6—figure supplement 2 . Impact of fecundity on dopamine transmission within the dorsal striatum and NAc core of paired male and female prairie voles . ( a , b ) Within the striatum of paired males , males categorized by their partner’s stage of pregnancy did not differ in peak DA release within the ( a ) dorsal striatum ( n = 4–22/group ) or ( b ) the NAc core ( n = 3–19/group ) . ( c , d ) Similarly , no differences were found within the ( c ) dorsal striatum ( n = 22 ) or ( d ) NAc core ( n = 19 ) of paired females . ( e ) Although categorization of males by their pregnancy status did not result in any significant differences in DA transmission within the dorsal striatum , there was a positive correlation between peak DA transmission and neonatal weight within this region ( n = 26 ) . ( f ) In contrast , there was no relationship between peak DA release and neonatal weight within the NAc core of paired males ( n = 24 ) . ( g , h ) There was no relationship between peak DA release and neonatal weight within the ( g ) dorsal striatum ( n = 22 ) or ( h ) NAc core ( n = 19 ) of paired females . ( i , j ) For paired males , there was no relationship between attack frequency and peak DA release within the ( i ) dorsal striatum ( n = 14 ) or ( j ) NAc core ( n = 14 ) . ( k , l ) Similarly , there was no relationship between attack frequency and stimulated DA release within the ( k ) dorsal striatum ( n = 13 ) or the ( l ) Nac core ( n = 12 ) of paired females . Summary data are presented as mean ± SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 017 Similar to measures of selective aggression , fecundity influenced DA transmission within the NAc shell in a sex-specific manner . Specifically , examination of DA release properties in relation to the pairs reproductive status revealed that only males from optimally pregnant pairs showed significantly greater elevations in NAc shell DA release compared to non-paired males ( one-way ANOVA , F ( 3 , 39 ) = 0 . 29 , p=0 . 05; Dunnett’s post hoc test , p=0 . 04 ) ( Figure 6c ) . In contrast to paired males , reproductive status did not influence NAc shell DA transmission dynamics of paired females ( one-way ANOVA , F ( 3 , 31 ) = 1 . 67 , p=0 . 20 ) . Rather , females of all reproductive categories ( not pregnant , sub-optimally pregnant , or optimally pregnant ) showed modest elevations in stimulated DA release compared to non-paired females ( Figure 6d ) . Thus , it is possible that sex-differences in the magnitude of change in NAc shell DA transmission dynamics results from paired females showing elevations in stimulated DA release regardless of reproductive status , whereas only males from optimally pregnant pairs ( 7 out of 22 total pairs ) had enhanced DA transmission within the NAc shell . To further explore the relationship between pair fecundity and NAc shell DA transmission dynamics , we next examined the relationship between neonatal weight and DA release . For paired males , the magnitude of stimulated DA release was positively correlated with neonatal weight , with fecundity accounting for nearly 30% of the variation ( linear regression , R2 = 0 . 29 , F ( 1 , 20 ) = 8 . 16 , p=0 . 01 ) ( Figure 6e ) . Conversely , there was no relationship between fecundity and the magnitude of stimulated DA release within the NAc shell DA of paired females ( linear regression , R2 = 0 . 32 , F ( 1 , 8 ) = 3 . 77 , p=0 . 09 ( Figure 6f ) . Together , these data suggest that reproductive status alters DA transmission within the NAc shell of paired voles in a sex-specific manner . Moreover , these effects are primarily localized to the NAc shell as paired voles categorized by their reproductive status did not differ in DA transmission dynamics within the NAc core or dorsal striatum ( Figure 6—figure supplement 2 ) . However , it should be noted that despite a lack of overall differences in DA transmission within the dorsal striatum based on the categorization of pairs by pregnancy , a comparatively modest correlation between DA transmission and pregnancy was found within the dorsal striatum of males ( Figure 6—figure supplement 2 ) . Nonetheless , these data suggest that fecundity exerts sex-specific effects on DA transmission dynamics in pair bonded prairie voles . Given the identified relationship between fecundity and selective aggression and fecundity and DA transmission , we next examined if variations in DA transmission within the NAc shell contribute to variable levels of selective aggression in paired males . Prior to measures of stimulated DA release , resident intruder test were administered to male and female subjects by placing a same-sex intruder into the test subjects home cage . Following the completion of behavioral testing , stimulated DA release was measured within the striatum and the frequency of attack behavior was quantified by an experimentally blind observer . These measures were subsequently utilized to assess the relationship between the magnitude of stimulated DA release and the degree of selective aggression displayed toward a resident intruder . For pair bonded males , the intensity of aggression directed toward a resident intruder was positively correlated with NAc shell DA release , accounting for over 40% of the variation ( linear regression , R2 = 0 . 41 , F ( 1 , 9 ) = 6 . 32 , p=0 . 03 ) ( Figure 6g ) . In contrast , no relationship between NAc shell DA release and selective aggression was identified in paired females ( linear regression , R2 = 0 . 32 , F ( 1 , 8 ) = 3 . 77 , p=0 . 09 ) ( Figure 6h ) . Additionally , no relationship between stimulated DA release and attack frequency was found within other regions of the striatum for either sex ( Figure 6—figure supplement 2 ) . Thus , the relationship between stimulated DA release and attack behavior in pair bonded prairie behaviors occurs in a sex and region specific manner . When these data are considered in combination with site-specific pharmacology data demonstrating that activation of D1-like DA receptors specifically within the NAc shell is required for the expression of selective aggression ( Aragona et al . , 2006 ) , they suggest that the degree to which DA transmission dynamics are altered within the NAc shell of paired males may underlie fecundity induced modulation of pair bond strength . In other words , enhancement of DA release would facilitate the activation of low-affinity D1-like receptors , possibly leading to the display of higher levels of aggression by males from optimally pregnant pairs . Moreover , stimulation of D1-like receptors results in the production of dynorphin ( Engber et al . , 1992 ) , the endogenous ligand for KORs ( Chavkin et al . , 1982 ) and activation of these receptors is also required for the expression of selective aggression ( Resendez et al . , 2012 ) . Therefore , we next examined the possibility of pair bond induced alterations in interactions between these systems . Activation of KORs within the NAc reduces DA release within this region ( Britt and McGehee , 2008 ) . Given that pair bonding altered NAc KOR expression pattern , we next compared KOR modulation of DA transmission within the NAc shell of non-paired and paired voles . Similar to other rodent species ( Britt and McGehee , 2008 ) , bath application of a KOR agonist ( BRL 5237 ) onto striatal slices of non-pair bonded voles reduced stimulated DA release within the NAc shell ( Figure 7a , b ) . Similar effects on DA transmission were observed in both non-paired males and females as the concentration response curves did not significantly differ between the sexes ( two-way ANOVA , F ( 1 , 5 ) = 1 . 59 , p=0 . 26 ) . Moreover , a t-test did not identify significant sex differences in the dose required to achieve a 50% reduction in DA release ( IC50: t ( 5 ) = 0 . 03 , p=0 . 98; Figure 7C ) or in the slope of the concentration response curve ( t ( 5 ) = 1 . 28 , p=0 . 26 , Figure 7D ) . However , in contrast to other species , a much higher dose of the KOR agonist was needed to achieve a 50% decrease in DA release ( IC50 ~ten fold greater compared to other rodent species; [Britt et al . , 2012] ) . 10 . 7554/eLife . 15325 . 018Figure 7 . Pair bonding increases KOR modulation of NAc shell DA release in male prairie voles . ( a , b ) Similar to other species , bath application of a KOR agonist decreases DA release in the NAc shell of male and female prairie voles . ( c , d ) Non-paired males and females did not differ in ( c ) the IC50 of BRL 5237 ( a KOR agonist ) or ( d ) in the slope of the dose response curve . ( e ) Pair bonding induced sex-specific alterations in KOR modulation of DA transmission within the NAc shell ( f , g ) Following the establishment of a pair bond , KOR mediated decrease of stimulated DA release was enhanced within the NAc shell of ( f ) males ( n = 3–5/group ) , but not ( g ) females . ( h , i ) Compared to non-paired males , pair bonding significantly decreased ( h ) the IC50 of BRL 5237 as well as ( i ) the slope of the dose response curve in paired males . ( j , k ) Pair bonding did not alter KOR mediated DA transmission in females ( n = 3–4/group ) . Summary data are presented as mean ± SEM . *p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 01810 . 7554/eLife . 15325 . 019Figure 7—figure supplement 1 . Comparison of the prairie vole KOR protein sequence to other rodent species and humans . ( a ) Cladogram showing that the prairie vole ( Michrotus ochrogaster ) KOR is distinct from that of guinea pigs ( Cavia porcellus ) , humans ( Homo sapiens ) , rats ( Rattus Norvegicus ) , and mice ( Mus musculus ) . ( b ) Protein sequence comparison of the prairie vole KOR to the above listed species . indicates completely conserved regions of the gene between all species shown here , + indicates regions of the gene that are similar between prairie voles , humans , and guinea pigs , but not rats or mice , and the # indicates portions of the KOR sequence that are unique to prairie voles . Consistent with other opioid receptors , the transmembrane regions of the gene ( black lines ) are highly conserved between species . In contrast , the N-terminus region of the KOR gene contains the most protein sequence differences between prairie voles and the other species shown . Notably , there is one portion of the dynorphin binding site , located in extracellular loop 2 , where a threonine has been substituted for a valine , which is unique to the prairie vole KOR . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 01910 . 7554/eLife . 15325 . 020Figure 7—figure supplement 2 . Sex differences KOR modulation of NAc shell DA release following the establishment of a pair bond . ( a ) Comparison of dose response curves between pair bonded males and females ( n = 4–5/group ) . ( b–c ) Although pair bonding significantly altered KOR modulation over DA release in paired males , but not paired females , these groups did not significantly differ in the IC50 of BRL5237 ( a KOR agonist ) ( t ( 7 ) = 1 . 72 , p = 0 . 13 ) or in the slope of the dose response curve ( t ( 7 ) = 2 . 14 , p = 0 . 07 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 020 The necessity to use higher doses of a KOR agonist in the present study is consistent with our previous findings showing that , compared to other rodents , prairie voles also require about a 10X higher dose of a peripherally administered KOR agonist to achieve significant alterations in KOR-mediated analgesia as well as locomotor activity ( Resendez et al . , 2012 ) . The consistent requirement for higher doses of a KOR agonist to observe either a behavioral or physiological impact in prairie voles suggests potential value in comparing the genetic sequence of the prairie vole KOR to other species that have been used to study KOR pharmacology ( e . g . , rats , mice , guinea pigs , and humans ) . Indeed , the prairie vole KOR is distinct from the above-mentioned species as its genetic sequence diverges from that of rats and mice ( whose KOR structure is quite homologous ) as well as humans and guinea pigs ( whose KORs also share substantial homologies ) ( Figure 7—figure supplement 1 ) . It is also notable to mention that the prairie vole KOR is more similar to that of humans and guinea pigs than that of rats and mice in which most pharmacological studies have been conducted . In total , there are four amino acids that are unique to prairie voles , humans , and guinea pigs ( Alanine 28 , Serine 186 , Aspartic acid 218 , Aspartic acid 374 ) , one amino acid that is unique to humans and prairie voles ( Isoleucine 232 ) , and fifteen amino acids that are unique to prairie voles , including one residue that is located in the dynorphin binding site ( Rasakham and Liu-Chen , 2011; Wu et al . , 2012 ) . It is possible that these genetic differences may partially account for species differences in KOR pharmacology and determining how ligands interact with the prairie vole KOR will be an important future area of study . Nevertheless , the above data demonstrate that activation of KORs within the prairie vole striatum produces the expected decreases in DA transmission . We next compared KOR modulation of DA transmission within the NAc shell of non-bonded ( sibling housed ) and pair bonded ( 2 weeks cohabitation with a mating partner ) prairie voles to determine if pair bond induced alterations in KOR protein binding within the NAc shell impact KOR modulation of DA transmission . Similar to anatomical changes , pair bonding robustly altered KOR modulation of DA release within the NAc shell of pair bonded males , while only producing very modest alterations in females ( Figure 7e ) . More specifically , in male prairie voles , pair bonding resulted in a leftward shift in the concentration response curve ( two-way ANOVA , F ( 1 , 6 ) = 15 . 67 , p=0 . 008 ) and significantly larger reductions in stimulated DA release at multiple concentrations of the KOR agonist ( Bonferroni’s post hoc test; 0 . 1 μM , p=0 . 03; . 3 μM , p=0 . 0007; 1 μM , p=0 . 0004; 3 μM , p=0 . 0003; 10 μM , p=0 . 002; 20 μM , p=0 . 02 ) ( Figure 7f ) . In contrast , the concentration response curve only slightly differed between paired and non-paired females ( two-way ANOVA , F ( 1 , 6 ) = 9 . 21 , p=0 . 03 ) with only one resulting in greater reduction in stimulated DA release ( Bonferroni’s post hoc test; 0 . 3 μM , p=0 . 02 ) ( Figure 7g , Table 4 ) . Overall , these data suggest that alterations in paired males were more dramatic than those that occurred in paired females . 10 . 7554/eLife . 15325 . 021Table 4 . Confidence intervals for IC50’s in KOR agonist dose response study . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 021Group95% Confidence intervalMale non-pair bonded4 . 761818 to 13 . 906848Male pair bonded−1 . 2093417 to 4 . 4570177Female non-pair bonded0 . 73751 to 18 . 11349Female pair bonded−0 . 73394 to 10 . 70294 Comparison of the IC50 between paired and non-paired voles revealed that a lower dose of the KOR agonist was needed to achieve a 50% decrease in stimulated DA release within the NAc shell of pair bonded males ( t-test; IC50; t ( 6 ) = 4 . 92 , p=0 . 002 , Figure 7h ) . The slope of the concentration response curve also significantly differed between paired and non-paired males ( t-test; t ( 6 ) = 3 . 74 , p=0 . 009 , Figure 7i ) . Given that the density of KOR binding is reduced in pair bonded males , these data suggest that pair bonding may result in mechanistic changes in the function of the KOR in males , but not females , as similar measures did not significantly differ between paired and non-paired females ( IC50: t-test; t ( 6 ) = 1 . 36 , p=0 . 22; Figure 7J and slope: t-test; t ( 6 ) = 0 . 55 , p=0 . 60; Figure 7k ) . Moreover , direct comparisons of the concentration response curves for paired males and females identified a leftward shift in the concentration curve of paired males ( two-way ANOVA , F ( 1 , 7 ) = 9 . 60 , p=0 . 02 ) and multiple doses that produced significantly greater inhibition of DA release in paired males compared to females ( Bonferroni’s post hoc test; 0 . 1 μM , p=0 . 02; 0 . 3 μM , p=0 . 004; 1 μM , p=0 . 001; 3 μM , p=0 . 009 ) ( Figure 7—figure supplement 2 ) . However , it should be noted that paired males and females did not significantly differ in IC50 or slope of the concentration response curve ( Figure 7—figure supplement 2 ) . Nonetheless , the present data suggest that although KOR binding is reduced within the NAc shell of pair bonded males , the function of these receptors may be enhanced as greater reductions in KOR induced decreases in stimulated DA release occurred within the NAc shell of pair bonded males . In summary , data from the present study reveal that both the DA and dynorphin/KOR systems within the NAc shell undergo sex-specific alterations following the establishment of a pair bond ( Figure 8a , b ) . Specifically , both sexes show increases in D1 receptor and dynorphin mRNA within the ventral striatum as well as enhanced DA transmission within the NAc shell . However , in relation to the dynorphin/KOR system , only males showed an overall reduction in membrane expression of KORs as well as dramatic reductions in DA transmission in response to a KOR agonist . Given that these systems are known to directly interact with each other ( Engber et al . , 1992; Carlezon et al . , 1998; Ebner et al . , 2010; Chartoff et al . , 2016 ) and that activation of both D1-like receptors ( Aragona et al . , 2006 ) and KORs are required for the expression of selective aggression ( Resendez et al . , 2012 ) , we next determined if these systems interact in vivo to regulate pair bond maintenance . 10 . 7554/eLife . 15325 . 022Figure 8 . Pair bonding alters DA and dynorphin/KOR systems within the ventral striatum . ( a ) Non-pair bonded prairie voles readily approach novel conspecifics and have lower levels of stimulated DA release as well as Drd1 and Pdyn mRNA expression within the ventral striatum . ( b ) Following the establishment of a pair bond , male and female prairie voles aggressively reject novel conspecifics and the ventral striatum undergoes a dramatic reorganization . Specifically , pair bonding enhances DA release within the NAc shell as well as up-regulates Drd1 as well as Pdyn within the ventral striatum of both males and females . Pair bonded males also show an additional decrease in KOR binding within the NAc shell . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 022 Previous studies have shown that the DA and the dynorphin/KOR systems function in sequence of each other , with stimulation of D1-like receptor promoting downstream activation of the dynorphin/KOR system ( Gerfen et al . , 1990; Carlezon et al . , 1998 ) . We therefore tested the hypothesis that D1-like receptor regulation of selection aggression is upstream of its regulation by KORs . Similar to the anatomical characterization studies described above , prairie voles were paired with an opposite sex conspecific for 2 weeks to allow sufficient time for a pair bond to be established . At the end of the cohabitation period , site-specific behavioral pharmacology was utilized in combination with resident intruder testing to examine the sequential nature of interactions between activation of D1-like receptors and KORs on the expression of selective aggression ( Figure 9a ) . More specifically , if KOR activation is indeed downstream of D1-like receptor activation than activation of KORs despite pharmacological blockade of D1-like receptors should still result in the expression of selective aggression . Conversely , pharmacological manipulations that would result in a reduction in KOR activation , such as administration of a D1-like antagonist in the absence of a KOR agonist or administration of a KOR antagonist in the presence of a D1-like receptor agonist , should attenuate the expression of selective aggression . 10 . 7554/eLife . 15325 . 023Figure 9 . Interactions between D1-like and KORs mediate pair bond maintenance . ( a ) Experimental Design . ( b , c ) Histological location of injection sites in ( b ) males and ( c ) females . ( d ) Compared to control pair-bonded males that received site-specific infusions of aCSF prior to resident-intruder testing ( n = 6 ) , males that received site-specific infusions of a D1-like receptor antagonist into the NAc shell showed attenuated levels of selective aggression as well as ( e ) increased attack latency toward intruders ( n = 6 ) . However , aggression levels and attack latencies were returned to normal when the antagonist for the D1-like receptor was administered in combination with a KOR agonist ( n = 7 ) suggesting that D1-mediated aggression occurs through downstream activation of KORs . This interaction was confirmed by the ability of the KOR antagonist to attenuate selective aggression even when it was administered in combination with the D1-like receptor agonist ( n = 7 ) . ( f , g ) Similar to males , blockade of D1-like receptors within the NAc shell of paired females ( n = 6 ) attenuated selective aggression compared to aCSF controls ( n = 6 ) . Aggression frequency was returned to the level of paired female controls when the D1-like receptor antagonist was administered in combination with a KOR agonist ( n = 7 ) . Finally , the attenuation of attack frequency and the increase in attack latency mediated by a KOR antagonist was maintained even in the presence a D1-like receptor agonist ( n = 6 ) . Summary data are presented as mean ± SEM . p<0 . 05 , p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 023 Compared to control subjects receiving site specific administration of aCSF , pharmacological manipulation of NAc shell D1-like and KORs ( Figure 9b , c ) significantly altered the expression of selective aggression in both pair bonded males ( Attack frequency: one-way ANOVA , F ( 3 , 25 ) = 5 . 55 , p=0 . 005 , Figure 9d; Attack latency: one-way ANOVA , F ( 3 , 25 ) = 5 . 54 , p=0 . 005 , Figure 9e ) and females ( Attack frequency: one-way ANOVA , F ( 3 , 23 ) = 4 . 59 , p=0 . 01 , Figure 9f ) . However , attenuation of selective aggression was dependent on the combination of agonists and antagonists administered . As expected , pharmacological blockade of NAc shell D1-like receptors significantly attenuated measures of selective aggression in both pair bonded males ( Attack frequency: planned contrast , post hoc: p=0 . 03 , Attack latency: planned contrast , post hoc: p=0 . 04 ) and females ( Attack frequency: planned contrast , post hoc: p=0 . 01 ) . However , blockade of NAc D1-like receptors did not significantly attenuate attack latency in females ( one-way ANOVA , F ( 3 , 23 ) = 4 . 77 , p=0 . 01 , p=0 . 47 ) . Thus , activation of NAc shell D1-like receptors is required for the expression of selective aggression in both sexes , possibly due to D1-like receptor mediated activation of the dynorphin/KOR system . To determine if KOR regulation of selective aggression is indeed downstream of the DA system , we co-administered a KOR agonist along with a D1-like receptor antagonist , resulting in KORs to be activated despite the inhibition of D1-like receptors . Activation of KORs in the presence of the D1-antagonist restored selective aggression as mean attack frequency ( Male: planned contrast , post hoc: p=0 . 92; Female: planned contrast , post hoc: p=0 . 62 ) and attack latency ( Male: planned contrast , post hoc: p=0 . 54; Female: planned contrast , post hoc: p=1 . 00 ) did not differ from paired controls in either sex , suggesting that D1-like receptors mediate selective aggression through downstream activation of the dynorphin/KOR system . In contrast , D1-like receptor activation in the presence of KOR inactivation was insufficient to restore measures of selective aggression to levels of paired controls ( Male attack frequency: planned contrast , post hoc: p=0 . 006; Female attack frequency: planned contrast , post hoc: p=0 . 008; Male attack latency: planned contrast , post hoc: p=0 . 001; Female attack latency; planned contrast , post hoc: p=0 . 005; Figure 9d–g ) , further suggesting that KOR mediation of selective aggression is downstream of D1-like receptors . Finally , these manipulations specifically altered aversively motivated behaviors as there were no differences in affiliative ( Male: one-way ANOVA , F ( 3 , 25 ) = 1 . 95 , p=0 . 15; Female: one-way ANOVA , F ( 3 , 23 ) = 1 . 58 , p=0 . 23 ) or locomotor behavior ( Male: one-way ANOVA , F ( 3 , 23 ) = 0 . 75 , p=0 . 54; Female: one-way ANOVA , F ( 3 , 23 ) = 0 . 69 , p=0 . 57 ) ( data not shown ) . Together , these data support the hypothesized mechanism that DA activation of D1-like receptors promotes downstream release of dynorphin to subsequently activate KORs within the NAc shell and generate selective aggression . In addition to regulation of selective aggression , D1-like receptors within the NAc shell have also been shown to mediate the protective effects of pair bonding against drug reward ( Liu et al . , 2011 ) . Given the identification that D1-like receptor regulation over pair bonding occurs through downstream activation of the dynorphin/KOR system , it is also possible that the protective effects of pair bonding are mediated though activation of this aversive processing system . We therefore next determined if activation of the dynorphin/KOR system is required for pair bonding to exert protective effects against the rewarding properties of amphetamine ( AMPH ) . Positive social relationships , such as the formation of strong social ties , modify the brain in such a manner that results in an attenuation of the rewarding properties of drugs of abuse ( Creswell et al . , 2015 ) . Thus , identifying overlapping neural systems that mediate both social bonding and drug reward processing may have positive therapeutic value in the treatment of addiction . We therefore first examined the impact of a rewarding regimen of AMPH ( 3 AMPH injections at 1-mg/kg across 3 days ) on KOR binding in non-pair bonded prairie voles . This dose of AMPH was chosen because it is well established to elicit a preference for AMPH in the conditioned place preference task in both male and female prairie voles ( Aragona et al . , 2007; Liu et al . , 2010; 2011 ) . Compared to control males ( 3 injections of saline across 3 days ) , male subjects exposed to a rewarding regimen of AMPH showed significantly altered patterns of striatal KOR expression ( two-way ANOVA , F ( 1 , 78 ) = 15 . 97 , p=0 . 0001 , Figure 10a , b ) . Moreover , the pattern of AMPH induced alterations in KOR expression within the striatum of males was similar to the pattern induced by pair bonding , with AMPH exposure significantly reducing KOR expression within the dorso-medial ( Bonferroni’s post hoc test , p=0 . 02 ) and ventral NAc shell ( Bonferroni’s post hoc test , p=0 . 01 ) ( Figure 10b , Table 5 ) . In contrast , AMPH exposure had no significant impact on striatal KOR expression in females ( two-way ANOVA , F ( 1 , 108 ) = 0 . 44 , p=0 . 51 , Figure 10c , d ) . Moreover , when the pattern of KOR expression binding was directly compared between males and females ( two-way ANOVA , F ( 1 , 96 ) = 39 . 80 , p=0 . 0001 ) , control subjects significantly differed in KOR binding density with control males having significantly higher levels in the dorso-medial ( Bonferroni’s post hoc test , p=0 . 03 ) and ventral NAc shell ( p=0 . 0002 ) ( Figure 10—figure supplement 1 ) . However , as with the experience of pair bonding , AMPH exposure also eliminated these sex differences ( two-way ANOVA , F ( 1 , 90 ) = 1 . 89 , p=0 . 17 ) ( Figure 10—figure supplement 1 ) . Given that AMPH altered male , but not female , striatal KOR binding density , we focused next set of experiments on male subjects . 10 . 7554/eLife . 15325 . 024Figure 10 . Amphetamine decreases KOR binding within the striatum of males . ( a , b ) AMPH decreased KOR binding within the dorso-medial and ventral NAc shell of non-pair bonded males ( n = 7–8/group ) . ( c , d ) Similar to pair bonding , AMPH did not impact striatal KOR binding in females ( n = 7–8/group ) . Summary data are presented as mean ± SEM . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 02410 . 7554/eLife . 15325 . 025Figure 10—figure supplement 1 . Sex differences in prairie vole KOR binding density following amphetamine exposure . ( a ) Representative autoradiographs from saline ( top ) and amphetamine ( 1 mg/kg ) ( bottom ) conditioned male ( left ) and female ( right ) prairie voles . For all subjects , saline ( control ) or amphetamine was administered once per day for 3 consecutive days . ( b ) KOR binding density varied as a function of sex and treatment ( two-way ANOVA , F ( 1 , 96 ) = 39 . 80 , p = 0 . 0001 ) . Specifically , control males had significantly higher levels of KOR binding within the dorso-medial NAc shell ( p = 0 . 03 ) as well as the ventral region of the NAc shell ( p = 0 . 0002 ) . ( c ) Interestingly , similar to pair bonding , repeated amphetamine exposure eliminated these sex differences in KOR binding density between sibling housed males and females ( two-way ANOVA , F ( 1 , 90 ) = 1 . 86 , p=0 . 17 ) ( n = 7–10/group ) . p<0 . 05 , 4*p<0 . 0005DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 02510 . 7554/eLife . 15325 . 026Table 5 . Non-significant statistics for comparisons of KOR binding density in saline versus amphetamine treated males . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 026Striatal sub-regionBonferonni's post hoc testDorso-medial striatump>0 . 99Dorso-lateral striatump>0 . 99NAc corep>0 . 99Dorso-lateral striatump>0 . 99 Pair bonding exerts protective effects against AMPH reward ( Liu et al . , 2011 ) ; however , the establishment of a pair bond is associated with a complex suite of socially related experiences , such as exposure to a novel social stimulus , extended periods of cohabitation , the development of social familiarity , copulation , and impregnation and it is not well understood how these individual components contribute to the neural protective effects that pair bonding exerts against drug reward ( Resendez et al . , 2013 ) . To this end , we conducted a detailed analysis of pair bond associated social experiences that may contribute to the social buffering of drug reward . Specifically , male subjects were randomly assigned to one of the following treatment groups: social familiarity ( i . e . , same-sex sibling housed ) , extended cohabitation with a novel social stimulus without mating ( i . e . , 2 weeks cohabitation with a novel male or ovariectomized female ) , or extended cohabitation with a reproductive partner ( 2 weeks cohabitation with an intact female ) . Given that not all gonadally intact male-female pairs achieved pregnancy , males housed with an intact female were further categorized by the reproductive status of the female partner at the completion of testing ( i . e . , no indication of pregnancy , sub-optimally pregnant , or optimally pregnant ) . Exposing males to these different social experiences as well as categorizing mating pairs by their reproductive status allowed us to determine the influence of each social condition on the protective effects of pair bonding ( Figure 11a ) . 10 . 7554/eLife . 15325 . 027Figure 11 . Neural protection against drug reward is specific to pair bonding in males . ( a ) Male prairie voles were housed with a familiar cage mate , a novel male , an ovariectomized ( OVX ) female , or an intact female for two weeks prior to AMPH conditioning . Compared to saline treated males , all groups except males housed with an intact female formed a preference for the AMPH paired chamber ( n = 6–33/group ) . To determine if pregnancy status influenced the rewarding properties of AMPH , males housed with an intact female were further classified by the pairs pregnancy status ( inset ) . Only males paired with a female that became pregnant ( suboptimally ( SP ) or optimally ( OP ) ) during the 2-week pairing period were protected against the rewarding properties of amphetamine as males paired with females that were not pregnant ( NP ) formed a preference for the AMPH paired chamber . ( b ) The establishment of an optimal pregnancy strongly influenced the rewarding properties of AMPH as there was a negative correlation between the duration of time spent in the AMPH paired chamber and the gestational stage of the female for optimally pregnant pairs ( n = 11 ) . ( c ) In contrast , there was no relationship between pregnancy stage and AMPH preference for sub-optimally pregnant pairs ( n = 8 ) . Summary data are presented as mean ± SEM . p<0 . 05 , p<0 . 005 , p<0 . 0005 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 027 Following exposure to one of the above described social conditions , males underwent conditioned place preference procedures to identify the specific aspects of male pair bonding that contribute to the attenuation of AMPH reward . A separate group of same-sex sibling housed males was conditioned with saline only and the duration of time spent in the AMPH paired chamber during the post-test session was compared to this treatment group . Social experiences that do not result in the establishment of a pair bond failed to protect against AMPH reward ( one-way ANOVA , F ( 4 , 69 ) = 0 . 67 , p=0 . 0001 ) as male subjects paired with a same-sex sibling ( Tukey’s post hoc test , p<0 . 0001 ) , novel male ( p=0 . 004 ) , or OVX female ( p=0 . 008 ) formed significant preferences for the AMPH paired chamber ( Figure 11a ) . In contrast , males housed under conditions that promote pair bonding did not form a preference for the AMPH paired chamber ( p=0 . 14 ) . Moreover , when these males were further classified by the pair’s pregnancy status , only males from pregnant pairs exhibited protection against AMPH reward as males from both optimally ( p>0 . 99 ) and sub-optimally pregnant pairs ( p=0 . 94 ) did not form a preference for the AMPH paired chamber , while males from non-pregnant pairs formed significant preferences ( p=0 . 006 ) ( Figure 11a inset ) . Together , these data indicate that the establishment of a fully developed pair bond , and not the other associated social experiences , mediates social buffering of drug reward . To further explore the influence of pair bonding on drug reward , we examined the relationship between pair fecundity and preference for the drug-paired chamber . While males from both sub-optimally and optimally pregnant pairs showed some degree of protection against AMPH reward , males from optimally pregnant pairs showed the strongest relationship between fecundity and the attenuation of AMPH reward . Specifically , in males from optimally pregnant pairs , the rewarding properties of AMPH were negatively correlated with the pregnancy status of the female ( linear regression , R2= 0 . 403 , F ( 1 , 9 ) = 6 . 079 , p=0 . 036 , Figure 11b ) . However , a similar relationship was not found in males from sub-optimally pregnant pairs ( linear regression , R2= 0 . 025 , F ( 1 , 6 ) = 0 . 155 , p=0 . 708 , Figure 11c ) . Thus , the reproductive status of the pair influences pair bond induced protection against AMPH reward . We next determined if pair bond induced alterations in the male prairie vole KOR system contribute to neural protective effects against AMPH reward . Similar to above , males were paired with either a male partner ( non-paired ) or an intact female ( paired ) for 2 weeks prior to AMPH conditioning . On day one of conditioning , males in both groups received either peripheral administration of saline or a KOR antagonist . AMPH preference varied by housing conditioning as well as treatment ( two-way ANOVA , F ( 1 , 28 ) = 8 . 13 , p=0 . 008 ) . Specifically , compared to non-paired males that received saline injections prior to AMPH conditioning , paired males treated with saline spent significantly less time in the AMPH paired chamber ( Bonferroni’s post hoc test , p=0 . 004 , Figure 12a ) . In contrast , global blockade of KORs in pair bonded males restored the rewarding properties of AMPH as paired males that received peripheral injections of nor-BNI prior to AMPH conditioning did not differ from non-paired males in the duration of time spent in the AMPH paired chamber ( p>0 . 99 , Figure 12a ) . Together , these data indicate that the protective effects of pair bonding are in part mediated by KORs . 10 . 7554/eLife . 15325 . 028Figure 12 . NAc shell KORs mediate the protective effects of pair bonding . ( a ) Peripheral administration of nor-BNI restored the rewarding properties of AMPH for paired males ( n = 6–14/group ) . ( b ) Histological location of injection sites . ( c ) Site-specific blockade of NAc shell KORs was sufficient to alleviate pair bond induced attenuation of AMPH reward ( n = 4–7/group ) . Summary data are presented as mean ± SEM . *p<0 . 005 . DOI: http://dx . doi . org/10 . 7554/eLife . 15325 . 028 To determine if KOR buffering of AMPH reward is mediated within the NAc shell , we next examined if blockade of KORs specifically within this region was sufficient to restore AMPH preference in pair bonded males ( Figure 12b ) . A two-way ANOVA indicated that AMPH preference varied by housing condition and treatment ( F ( 1 , 29 ) = 8 . 33 , p = 0 . 007 , Figure 12c ) . Compared to non-paired males , paired males that received site-specific injections of aCSF into the NAc prior to AMPH conditioning spent significantly less time in the AMPH paired chamber ( Bonferroni’s post hoc test , p = 0 . 003 , Figure 12c ) . However , pair bonded males that were administered nor-BNI prior to AMPH conditioning did not differ from non-paired males that also received site-specific administration of nor-BNI in the duration of time spent in the AMPH paired chamber ( p>0 . 99 , Figure 12c ) . Together , these data demonstrate that KORs in the NAc shell are indeed involved in pair bond induced neuroprotection against drug reward as activation of these receptors is required for pair bond induced attenuation of drug reward . A combination of comparative anatomical approaches and behavioral pharmacological studies has been utilized to identify neural mechanisms that underlie pair bond formation and maintenance in the socially monogamous prairie vole ( Carter et al . , 1997; Bales et al . , 2007; Young et al . , 2008; Aragona et al . , 2009 ) . In regards to pair bond formation , the initial stages of bond development are mediated in part by D2-like , oxytocin , vassporessin , and mu-opioid receptors systems located within reward processing regions of the brain such as the striatum and the ventral pallidum ( Insel et al . , 1998; Young et al . , 2008; Resendez and Aragona , 2013 ) . Interesting , following social conditioning procedures that promote the establishment of a pair bond , we did not identify alterations in the expression of mRNA for genes encoding D2-like , oxytocin , vasopressin , or mu-opioid receptors ( Carter et al . , 1997; Johnson and Young , 2015 ) . One possible explanation of why a lack of an alteration in receptor systems associated with partner preference behavior was not observed in the present study is that a preference for social proximity is required in both the bonded and non-bonded state , while the emergence of an aversion towards social novelty is specific to the establishment of a pair bond and requires alterations in both social and motivational circuitry ( Resendez and Aragona , 2013 ) . In contrast to neural systems that regulate pair bond formation , motivational and valence processing systems associated with pair bond maintenance undergo a dramatic overhaul during the transition from the naive to the pair bonded state . An important function of this alteration is to render pair bonded voles hyper aggressive toward novel conspecifics in order to achieve robust mate guarding ( Resendez and Aragona , 2013 ) . As such , it is not surprising that neuroplasticity associated with pair bond development occurs within systems that mediate the expression of social behaviors associated with pair bond maintenance , DA/D1 and dynorphin/KOR systems and that these alterations occur specifically within the NAc shell , the striatal sub-region where these receptor systems act to mediate pair bond maintenance . Moreover , behavioral pharmacological data make evident the functional significance of these alterations by demonstrating that activation of KORs within the NAc shell of prairie voles mediates the assignment of negative social valence onto novel conspecifics , while blockade of either D1-like or KORs abolishes selective aggression . Together , these data support the working hypothesis that neuroplasticity within the NAc shell stabilizes established pair bonds by enhancing the perceived negative valence of novel social stimuli and the promotion of robust mate guarding . In species where monogamous breeding systems have evolved , males will often engage in mate guarding behavior to prevent access of competitor males to the female partner , increasing the males opportunity for selective breeding as well as reducing the likelihood of uncertain paternity ( Trivers and Campbell , 1972; Kleiman , 1977 ) . While this breeding strategy is adaptive for some species , mate guarding behavior is intensely energetically costly given that a great deal of time must be spent maintaining proximity to the female and a high level of energy expended engaging in risky agonistic social encounters with competitor males ( Parker , 1974; Grafen and Ridley , 1983; Getz et al . , 1990; Crews and Moore , 1986 ) . As a result , males can usually only successfully guard one female at a time ( Brotherton et al . , 2003 ) and , to maximize reproductive fitness , it is adaptive for mate guarding behavior to emerge only after the establishment of a reproductively successful pair bond ( Resendez et al . , 2012 ) . For the socially monogamous prairie vole , an indication of pair fecundity is indeed required for males to transition to the pair bonded state ( Curtis , 2010; Resendez et al . , 2012 ) ; yet the neural mechanisms that underlie this sex difference have remained elusive . In the present study , we provided the first proximal mechanistic data to explain how fecundity induced sex differences in selective aggression might arise . Consistent with the theory that mate guarding in monogamous males serves to maximize reproductive fitness ( Trivers and Campbell , 1972 ) , the present study demonstrates that the extent to which DA transmission was enhanced within the NAc shell of paired males was positively correlated with both pair fecundity as well as measures of selective aggression . Moreover , significant enhancements in NAc shell DA transmission only occurred in males from optimally pregnant pairs and males from these pairs also showed significantly higher levels of selective aggression . Thus , neuroplastic changes that are associated with the expression of selective aggression only occur in paired males when an adaptive benefit in defending the female partner has been established ( i . e . , when the benefit of defending the female outweighs the risk associated with aggressive conflict as well as predation risks that may occur when searching for a new mate ) . In contrast to paired males , a relationship between alterations in DA transmission and pair fecundity were not identified in paired females , which is consistent with both field ( Rose and Gaines , 1976 ) and laboratory ( Resendez et al . , 2012 ) studies that have yet to identify a relationship between fecundity and mate guarding behavior in female prairie voles . One possible explanation for a lack of influence for fecundity on both neural and behavioral changes associated with pair bonding in females is a difference between males and females in the ultimate mechanisms that underlie the decision to bond with a partner . For instance , given that the reproductive status of the male is constant , whereas females require extended periods of contact with a male for sexual receptivity to be induced ( Carter et al . , 1987 ) , it may be more beneficial for a female to increase the likelihood of reproductive success by remaining in contact with a male partner than to risk predation by searching for a new mate . Moreover , unlike males , female mammals do not risk exerting unnecessary energy raising offspring that are not their own if their male partner engages in extra-pair copulations ( Trivers and Campbell , 1972 ) . Thus , the risk to reward ratio for engaging in agonistic social encounters may not be as great for females and may also partially explain why females tend to be less aggressive overall than males . Together , sex differences in the adaptive value of mate guarding as well as sex differences in temporal and environmental factors that regulate reproductive activation between males and females may partially underlie known sex differences in the intensity of selective aggression as well as behavioral and neural neuroplasticity associated with the transition to the bonded state . Increasing evidence suggests that anatomical and functional heterogeneity occurs within the shell region of the NAc ( Peciña and Berridge , 2005; Resendez and Aragona , 2013; Richard et al . , 2013 ) . More recently , it has been demonstrated that heterogeneity in the valence coding properties of the dynorphin/KOR system is anatomically segregated within the NAc shell ( Castro and Berridge , 2014; Al-Hasani et al . , 2015 ) . Specifically , the release of dynorphin into the dorso-medial region of NAc shell and the subsequent activation of KOR induces positive hedonics ( Castro and Berridge , 2014; Al-Hasani et al . , 2015 ) , while the release of dynorphin into the ventral NAc shell induces aversion ( Al-Hasani et al . , 2015 ) . Interestingly , pair bonding altered the binding density of male NAc shell KORs in a sub-region specific manner that maps onto the topographical organizational of the aversive coding properties of the dynorphin/KOR system . Pair bonding down-regulated KOR binding within the ventral region of the NAc shell , while leaving KOR binding within the dorso-medial and lateral regions of the NAc shell unaltered . Thus , pair bonding altered KOR binding only in the sub-region of the NAc shell where activation of these receptors acts to encode aversion . One possible mechanism underlying the sub-region specific influence of the dynorphin/KOR system on valence coding may be due to anatomical heterogeneity in the downstream projection targets of the dorso-medial and ventral regions of the NAc shell . In general , downstream projection targets from the afferents of cell bodies originating in the dorsomedial region of the NAc shell are more widespread than those originating from cell bodies located in the ventral region of the NAc shell . For cell bodies located in the dorso-medial NAc shell , the highest density of afferents are located in the medial region of the ventral pallidum ( VP ) , the lateral preoptic area , and the lateral hypothalamus with sparser fiber labeling occurring within the lateral septum , the bed nucleus of the stria terminalis , the anterior hypothalamus , the medial preoptic area , and rostral portion of the ventral tegmental area ( Thompson and Swanson , 2010; Zahm et al . , 2013 ) . In contrast , the ventral region of the NAc shell varies from the dorso-medial region of the NAc shell in both the number and specific brain regions it projects to . While the dorso-medial NAc shell sends dense projections to the medial region of the VP ( Thompson and Swanson , 2010 ) , the ventral region of the NAc shell projects specifically to the lateral region of the VP and also sends sparser projections to the lateral preoptic area and rostral-caudal extent of the ventral tegmental area ( Zahm et al . , 2013 ) . It is therefore possible that although the release of dynorphin and the subsequent activation of NAc shell KORs has the potential to reduce dopaminergic and glutamatergic transmission throughout the entire dorsal-ventral axis of the NAc shell ( Hjelmstad and Fields , 2001; Britt and McGehee , 2008 ) , the sum of the influence on downstream neuronal networks has the potential to vary greatly between the two sub-regions . One notable downstream projection target that is unique to the ventral NAc shell is the lateral region of the VP . The lateral VP is an important brain region for reward processing ( Cromwell and Berridge , 1993 ) and is innervated by NAc shell medium spiny neurons that primarily express Gi-coupled D2-like DA receptors ( Gerfen and Young , 1988 ) . Thus , KOR-mediated reductions in DA within the ventral NAc shell would increase the activity of GABAeregic medium spiny neurons that project to the lateral VP , resulting in the inhibition of this region ( Bonci and Carlezon , 2005 ) . Interestingly , reduced activity has been observed directly within the VP following exposure to aversive stimuli ( Itoga et al . , 2016 ) as well as an increase in activity within brain nuclei that receive input specifically from GABAergic neurons located within the lateral region of VP . Of specific interest is the relief of inhibition of the periventricular nucleus of the thalamus ( PVT ) , a brain nucleus that is downstream of the VP and has been indicated in aversive processing ( Yashoshima et al . , 2007 ) . In contrast to the ventral NAc shell , reductions in DA specifically within the dorso-medial region of the NAc shell would cause a decrease in the activity of the PVT through inhibition of glutamatergic afferents from the LH that project to the PVT ( Thompson and Swanson , 2010; Zahm et al . , 2013 ) . Thus , the ventral NAc shell → lateral VP → PVT circuit may be one mechanism in which site-specific modulation of NAc shell KORs contributes to aversive coding , while the dorso-medial NAc shell → LH → PVT circuit may contribute to positive valence coding by NAc KORs . However , more work is necessary to determine how topographical organization of NAc shell contributes to opposing modulation of valence coding and the subsequent divergent responses on motivated behavioral states , such as approach versus avoidance behaviors . Complex behaviors often require interactions between multiple neural systems . Indeed , studies of pair bond formation show that partner preferences require concurrent activation of D2-like and OT receptors within the NAc shell ( Liu and Wang , 2003 ) as well as V1a receptor activation within the VP ( Lim et al . , 2004 ) . These previous studies argued that peptide systems are necessary for social recognition , while DA transmission is important for reward processing ( Young and Wang , 2004; Johnson and Young , 2015 ) . In the present study , we expand our current understanding of neural mechanisms involved in the regulation of pair bond behavior by demonstrating that interactions between opioid peptides , such as the dynorphin/KOR system , and DA transmission within the NAc may act to couple valence processing systems within motivational circuitry . Consistent with the theory that interactions between these systems mediate pair bond induced transitions in the valence encoding of novel social stimuli , previous studies of drug reward have demonstrated that interactions between DA/D1 and dynorphin/KOR systems mediate the propensity for previously rewarding stimuli to be processed as aversive . Specifically , stimulation of D1-like receptors phosphorylates cAMP response element binding protein ( CREB ) to induce the expression and release of dynorphin ( Carlezon et al . , 1998 ) , resulting in the valence encoding of a psychostimulant to be reversed from rewarding to aversive ( Pliakas et al . , 2001 ) . Together , these data suggest that alterations in the activity and , subsequently , interactions between DA and dynorphin/KOR systems play a critical role in experience mediated plasticity in reward processing . While it is known that activity within the dynorphin/KOR system plays a critical role in reward processing , the mechanism in which this system modulates the encoding of reward is not well understood . One hypothesized mechanism in which the dynorphin/KOR is thought to negatively impact motivation and reward processing is through its ability to robustly decrease dopaminergic transmission within the NAc ( Shippenberg et al . , 1996; Carlezon and Thomas , 2009 ) . Interestingly , interactions between these systems were augmented following the establishment of a pair bond in male , but not female , prairie voles . Given that paired bonded males are the more aggressive sex and that paired males also incur greater reproductive costs if their mate engages in extra-pair copulations ( Resendez et al . , 2012 ) , it is possible that paired males show a greater aversion to social novelty and this enhanced aversion may be mediated by augmented coupling between DA and dynorphin/KORs within the NAc . While more work is necessary to determine the exact mechanism in which NAc KORs mediate the expression of sex differences in selective aggression , the present study nonetheless provides an interesting example of how the dynorphin/KOR system modulates reward and motivation to promote ethologically relevant behavioral adaptation . Addiction is a debilitating disorder that is characterized in part by chronic relapse , and , while pharmacological treatments continue to be sought for the treatment of this disorder , many have been ineffective in sustaining drug abstinence ( Fattore and Diana , 2016 ) . Interestingly , there is compelling evidence that a preventative approach , focused on neural adaptations resulting from the formation and maintenance of positive social relationships , may offer positive benefits to psychological well being ( Feldman , 2015 ) . For example , in drug-addicted humans , the presence of positive social support reduces the propensity for relapse ( Creswell et al . , 2015 ) , suggesting that positive social experience , such as bonding , may wire the brain in a manner that reduces future drug seeking behavior ( Young et al . , 2011 ) . Yet , despite the demonstrated positive benefit for social bonding on mental health , the neural mechanisms underlying the relationship between social bonding and drug taking have not been extensively studied . The scarcity of studies examining this relationship is likely related to a lack of animal models that exhibit both the propensity for social bonding and drug taking behavior . The socially monogamous prairie vole offers an excellent animal model in which to study the relationship between drug and social reward because , unlike most mammals , they form selective social attachments to their mating partner . Importantly , as demonstrated in the present study , the establishment of a pair bond , but not mere social exposure or mating , is required for social bonding to exert protective effects against AMPH reward . The specificity of the establishment of a social bond to the neural protection against drug reward is likely related to the fact that the establishment of a pair bond alters the brain in a manner in which other social experiences do not ( Liu et al . , 2011; Smith and Wang , 2014 ) . Indeed , it has previously been demonstrated that activation of neural systems that mediate pair bond maintenance ( NAc D1-like receptors ) , but not those that mediate pair bond formation ( NAc D2-like receptors ) , are required for pair bond induced protection against drug reward ( Liu et al . , 2011 ) . Here , we extended these previous findings by demonstrating that , for male prairie voles , activation of NAc shell KORs is required for pair bond induced attenuation of AMPH reward . Given that drugs that act at the dynorphin/KOR system are currently under intense investigation as potential therapeutics for the treatment of addiction ( Carlezon and Miczek , 2010 ) and that pair bonding induces neural plasticity within this system to result in an attenuation in drug reward , there may be considerable therapeutic value in understanding the neural mechanisms in which social experiences alter reward circuitry to buffer against drug reward . Finally , data presented in the present study also provide support for the consideration of socially related cognitive therapies when developing future treatment regimens for the treatment of addiction . Following the establishment of a pair bond , male prairie voles show an enhancement in KOR-agonist induced decreases in stimulated DA within the NAc shell , despite having an overall reduction in KOR binding density within this region . Given the presumption that a reduction in receptor number would also reduce the efficacy of an agonist to produce the measured physiological response , these findings appear to contradict one another . However , G protein-coupled receptors ( GPCRS ) , including KORs , are dynamic proteins that can adopt multiple conformational states , resulting in variability in ligand binding affinity as well as efficacy of the receptor to activate distinct downstream signaling cascades ( Bruchas and Chavkin , 2010 ) . For example , G proteins have been shown to pre-couple with receptors ( Nobles et al . , 2005 ) and this coupling can lead to conformational changes in the extracellular portion of the receptor that enhance affinity of the ligand for the receptor ( Yan et al . , 2008 ) . Moreover , the percent of G protein receptor coupling can vary as a function of receptor density ( Yan et al . , 2008 ) . Thus , under certain physiological conditions , it is possible for receptor coupling efficiencies to be enhanced , resulting in a fewer number of receptors to be required for an agonist to produce the maximal biological response ( Kenakin , 2002 ) . With binding of the agonist , changes in conformational state of the receptor can also be induced , to increase affinity of the G protein to the receptor . In addition , GPCRS can engage a diverse array of signaling pathways in a manner that is dependent on the ligand ( White et al . , 2014 ) , cellular milieu ( Yan et al . , 2008 ) , as well as lipid membrane properties ( Nygaard et al . , 2013 ) . Thus , there are wide variety of dynamic receptor states that can influence functional interactions between the ligand and the receptor as well as the receptor and its G protein . Future research is therefore necessary to determine if , in addition to pair bond induced changes in striatal KOR expression patterns , if KOR mediated signaling properties also vary as a function of social state as well as the potential for such alterations to influence DA transmission dynamics . Finally , given that KORs are found on the terminal regions of multiple inputs to the NAc ( glutamatergic , GABAergic , and dopaminergic ) ( Meshul and McGinty , 2000; Svingos et al . , 2001; Hjelmstad and Fields , 2003 ) more work will also be necessary to determine if KOR binding is globally decreased within the NAc of pair bonded males or if this decrease is restricted to a specific sub-population of inputs . In the present study , we provide extensive detail of the neural mechanism involved in the maintenance of enduring social bonds . Specifically , we show that neural systems involved in aversive valence processing , such as the dynoprhin/KOR system ( Bals-Kubik et al . , 1989; Land et al . , 2008; Chartoff et al . , 2009; Koob and Volkow , 2010; Schindler et al . , 2012; Al-Hasani et al . , 2015 ) , as well as those involved in the orchestration of socially motivated behavioral states , such as the D1-like receptor system , ( Balfour et al . , 2004; Champagne et al . , 2004; Aragona et al . , 2009; Hull , 2011; Chevallier et al . , 2012; Gunaydin et al . , 2014 ) interact within the NAc shell to mediate selective aggression and the maintenance of monogamous bonds . Importantly , understanding the neurobiology of social bonding has important translational implications for psychiatric disorders of a social nature as well as motivational/affective disorders ( Feldman , 2015 ) . Thus , further investigation of social reward processing in the prairie vole model has the potential to reveal how social bonding alters motivational circuitry in a manner that buffers against psychopathology . Subjects were adult prairie voles bred in a laboratory colony at the University of Michigan . Subjects were weaned at 21 days of age and initially housed in same-sex sibling pairs . Animals were housed in a 14 hr light/10 hr dark cycle ( lights on at 6 AM and off at 8 PM ) and all experiments occurred during the light phase of the animals cycle . Food and water was available ad libitum ( Resendez et al . , 2012; Resendez and Aragona , 2013 ) . For experiments that required pair bonded prairie voles , adult subjects were paired with an opposite sex partner for 14 days in a large cage that subsequently became the pair’s ‘home cage’ cage . This cohabitation time allows for nest sharing , mating , and impregnation ( Aragona et al . , 2006 ) . Pregnancy was confirmed by extracting embryos from pregnant females and subsequently categorizing pregnancy status by average neonatal weight of the embryos ( Curtis , 2010; Resendez et al . , 2012 ) Subjects were anesthetized with a mixture of ketamine ( 90 mg/kg ) and xylazine ( 10 mg/kg ) administered at 0 . 1% of total body weight . Stereotactic surgery was subsequently performed to implant a 26-gauge bilateral guide cannula ( Plastics One , Roanoke , VA ) into the NAc shell ( +1 . 7 mm A/P; ± 1 mm M/L; -4 . 5 mm D/V ) ( Resendez et al . , 2012 ) . Cannulas were secured to the skull with stainless steel screws and dental cement . Following surgery , males were given 0 . 1 mL ketoprofen and returned to their home cage to recover with either their cage mate or mating partner 3 days prior topartner preference , selective aggression , or conditioned place preference testing . Three days prior to behavioral testing , a guide cannula was implanted above the NAc shell ( Resendez and Aragona , 2013 ) . Immediately prior to pairing with an opposite-sex conspecific , male subjects received site-specific injections of either aCSF or 1 μg U50 , 488 ( KOR agonist ) ( Muschamp et al . , 2011 ) . Following injections , subjects cohabitated with a female partner for 1 hr and were next placed in a 3-chambered partner preference apparatus with their partner restricted to one chamber and a novel opposite-sex individual ( stranger ) restricted to the opposite chamber . Test subjects were free to move throughout the apparatus . The 3 hr test was recorded and later scored by an experimentally blind observer for the duration of time spent in side-by-side contact with either the partner or stranger . 1 hr prior to resident-intruder testing , subjects received site-specific infusions of one of the following treatment groups: aCSF , 10 ng SCH 23 , 390 ( D1 receptor antagonist ) , 10 ng SCH 23 , 390 and 1 μg U50 , 488 , or 0 . 4 ng SKF 38 , 393 ( D1 receptor agonist ) and 500 μg norBNI ( KOR antagonist ) ( Aragona et al . , 2006; Resendez et al . , 2012 ) . 1 hr after drug infusion , the subject was placed in its home cage ( in isolation ) and its behavior was recorded for 10 min , allowing time for acclimation to the testing environment and the assessment of locomotor activity . Locomotor activity was assessed by counting the number of cage crosses made during the 10-min habituation period . Next , a same-sex stimulus animal was introduced to the subject’s home cage and behavioral interactions were recorded for 10 min . Resident-intruder tests were scored for the frequency of aggressive behaviors ( offensive rears , lunges , bites , and chase frequency ) . The latency to engage in agonistic behavior was determined by the first time point in which an aggressive encounter occurred . If an aggressive encounter was not observed during the testing period , an attack latency of 10 min was applied . Affiliative behavior was assessed by quantifying the sum of the duration of time that the test subject spent investigating and in side-by-side contact with the intruder . A non-biased conditioned place preference assay was used to assess the rewarding properties of 1 mg/kg of amphetamine , a dose that has been shown to reliably elicit a preference for the drug paired compartment in this species ( Aragona et al . , 2007; Liu et al . , 2010 ) . A pre-test was conducted on day 1 of conditioning to determine if the test subject had a bias for either side of the chamber . Test subjects that showed a robust preference for either chamber ( greater than 67% of the pre-test ) were excluded from the study . On the following 3 days , subjects were administered saline in the preferred chamber and 1 mg/kg AMPH the non-preferred chamber and placed in the chamber for 40 min . The order of injections was counterbalanced between subjects and treatments were administered 6 hr apart . On day five of testing , a preference for the amphetamine paired chamber was determined by placing the subject in the apparatus and allowing it to freely roam either compartment for 30 min . To determine how social conditioning impacts AMPH reward , male subjects were placed in one of several housing conditions: sibling housed , housed with a novel male , house with an OVX female , or housed with an intact female . To determine if alterations in KOR binding contribute to the protective effects of pair bonding against drug reward , non-paired or paired subjects received either peripheral administration of saline or 50 mg/kg nor-BNI dissolved in sterile saline on day 1 of conditioning . To determine if the impact of KORs on drug reward processing was specific to the NAc shell , test subjects received site-specific microinfusions of either aCSF or 500 μg nor-BNI into the NAc shell on day 1 of conditioning . These doses were chosen because they have previously been shown to reduce selective aggression in male prairie voles without altering locomotor activity ( Resendez et al . , 2012 ) . For males paired with intact females , female partners were checked for successful pregnancy after male subjects completed the post-test on day 5 of testing ( Curtis , 2010; Resendez et al . , 2012 ) . Tissue punches from the dorsal and ventral striatum were processed for mRNA quantification as previously described ( Day et al . , 2013 ) . Total RNA was extracted using the RNeasy Mini kit ( Qiagen ) following the manufacturer’s instructions . mRNA was reverse transcribed using the iScript RT-PCR kit ( Bio-Rad ) . Specific intron-spanning primers were used to amplify cDNA regions for transcripts of interest ( Drd1 , Drd2 , Drd3 , Pdyn , Penk , Oprk1 , Oprm1 , Oxtr , and Avpr1a ) . Nadh was used as an internal control for normalization using the ΔΔCt method within the VS . Oprm1 was used as the control within the DS because mRNA for this gene did not differ between groups and Nadh mRNA levels were found to be affected by pregnancy in females . KOR autoradiography was used to quantify changes in receptor density across the striatum as a function of pair bonding and AMPH administration . To examine the effect of pair bonding on KOR binding density , subjects were either paired with a novel female for two weeks or remained housed with a same-sex sibling . To determine the effects of AMPH exposure on KOR binding density , striatal slices from male prairie voles receiving either control injections of saline or injections of 1 mg/kg AMPH for three days were also processed for KOR autoradiography . Follow the completion of either social or drug exposure , subjects were sacrificed via rapid decapitation 24 hrs following the last day of pairing ( for the pair bonding manipulation ) or the final injection ( for the AMPH manipulation ) . Brains were removed , immediately frozen on powdered dry ice , and stored at −80ºC until sectioning . A cryostat was used to section the brain into a 1:4 series at 20 μM . Sections were directly mounted onto glass slides and stored at −80°C until time of assay . Autoradiography for KORs was conducted with 3H U-69 , 593 radioligands ( Perkin Elmer , USA ) ( Resendez et al . , 2012 ) . On the day of the assay , slides were thawed at room temperature ( RT ) until dry and fixed in 0 . 1% paraformaldehyde ( pH 7 . 4; 2 min ) . Slides were then washed in 50 mM Tris buffer ( pH 7 . 4; 10 min; 2 washes ) and incubated for 1 hr in tracer buffer containing 50 mM Tris buffer , 10 mM MgCl2 , 0 . 1% bovine serum albumin , and 1 nM 3H U-69 , 593 for visualization of KOR binding . Next , slides were washed in 50 mM Tris containing 0 . 2% MgCl2 for 5 min at 4°C ( 4 washes ) then for 30 min at RT ( 1 wash ) . Slides were briefly dipped in nanopure H2O , allowed to dry at RT and then exposed to BAS Imaging Plates ( GE Life Sciences , Piscataway , NJ ) for 2 weeks . All plates were scanned on the BAS-5000 plate reader using BAS-5000 Image Reader software ( Version 1 . 8 ) . Binding densities were determined with region of interest analysis using ImageJ software . Following rapid , live decapitation , brains were quickly extracted , immediately submerged in cold , pre-oxygenated high sucrose aCSF consisting of 180 mM sucrose , 30 mM NaCl , 4 . 5 mM KCl , 1 mM MgCl2 , 26 mM NaHCO3 , 1 . 2 NaH2PO4 , and 10 mM D-Glucose in deionized H2O ( pH 7 . 4 ) , and sectioned into coronal slices ( 400 μm ) containing the DS , the NAc core , and the NAc shell . Following sectioning , slices were transferred to RT aCSF buffer solution consisting of 176 . 13 mM ascorbate , 180 . 16 mM glucose , 84 . 01 mM sodium bicarbonate , 58 . 44 mM NaCl , 156 mM NaH2PO4 , 74 . 56 mM KCl , 147 . 01 mM CaCl2 , and 203 . 30 mM MgCl2 in deionized H2O ( pH 7 . 4 ) and incubated for 1 hr . A buffer solution of this same composition ( minus ascorbate ) was used to perfuse the slices during recordings ( 1 ml/min ) . Both buffer solutions were continuously bubbled with 5% CO2 and 95% ( Maina and Mathews , 2010 ) . FSCV was conducted with recording electrodes fabricated from 1 . 2 mm pulled glass capillary tubes , with the carbon fiber cut to approximately 150 μm from the capillary glass seal . Using Tarheel CV ( University of North Carolina , Chapel Hill ) software written in LABVIEW ( National Instruments , Austin , TX ) , a triangular ramp sweeping from −0 . 4 V to +1 . 2 V versus a Ag/AgCl reference was applied to the carbon-fiber electrode at a rate of 10 Hz ( Robinson et al . , 2002 ) . The characteristic oxidation current , seen at +0 . 6 V during the upward ramp , and reduction current , at -0 . 2 V during the downward ramp , of DA was identified using a background-subtracted cyclic voltammogram ( Aragona et al . , 2009 ) . The peak currents for DA were converted to concentration by calibrating each electrode to a known concentration of DA ( 3 μM ) ( Sinkala et al . , 2012; Vander Weele et al . , 2014 ) . To compare differences in striatal DA release properties between non-paired and pair bonded voles , FSCV was conducted in striatal slice preparations ( Singer et al . , 2016 ) . DA release was evoked by a 1 or 20-pulse stimulation ( 350 μA ) delivered in 5-min increments at 20 Hz with a bipolar stimulating electrode placed on the surface of the striatal slice approximately 150 μm from the recording electrode ( Zhang et al . , 2009 ) . Each recording was 15 s in duration and DA release was evoked at 5 s . A total of 3 recordings at each pulse were made within each region and peak DA release was averaged for each subject . Slice stimulations occurred at regular 5-min intervals and readings were only recorded for experimental purposes once DA release was stabilized ( Calipari et al . , 2012 ) . FSCV was used to assess changes in DA/KOR interactions following the establishment of a pair bond . The KOR agonist BRL 5237 hydrochloride was bath applied to striatal slice preparations and DA release was measured ( Britt and McGehee , 2008 ) . Increasing doses ( 0 . 001 , 0 . 01 , 0 . 03 , 0 . 1 , 0 . 3 , 1 , 3 , 10 , 20 30 μM ) of BRL 52 , 537 hydrochloride were added every 30-min to the slice’s aCSF reservoir , perfused at 1 mL/min . Dose response curves were generated using non-linear regression with the bottom set equal to 0 ( Maina and Mathews , 2010 ) . Based upon our previous behavioral pharmacology experiments in voles , 5–8 subjects are needed per group to achieve a p-value of <0 . 05 with 80% power . Therefore , each group contains at least 6–8 subjects . Consistent with established standards in the literature , we used at least 6 subjects for mRNA and receptor autoradiography experiments and at least 3 subjects for slice FSCV experiments , with multiple samples taken per slice . For most experiments , comparisons were made between biological replicates , i . e . , comparisons between treatment groups receiving a pharmacological manipulation or between different social conditions . For FSCV experiments , both biological and technical replicates , i . e . , repeated measurements from the same coronal slice under identical preparations were also made . To determine whether the data were normally distributed and equivalent in variance , we examined boxplots for each group . In cases where boxplots revealed that the data were not normally distributed or there was a lack of equal variance among groups , nonparametric tests were used . Statistical significance was assessed with either a t-test , one-way ANOVA , or two-way ANOVA . An alpha level was set at p≤0 . 05 for all statistical analysis . All analysis were performed in SPSS version 21 for Windows .	The bond between parents is one of the most important social relationships that humans have . Prairie voles are one of the few other mammals whose individuals also form long-term social bonds after having offspring together , so they have frequently been used to study the brain mechanisms that underlie such bonding . However , most previous studies have focused only on how the bond between a pair of mating partners is formed: little is known about how this bond is then maintained over months and years . When a prairie vole forms a bond with a mate , it will then aggressively reject other prairie voles . This “selective aggression” only happens once a social bond between two mating prairie voles is formed , so this behavior can be used as a proxy to confirm that the social bond exists . In order to study how prairie voles maintain bonds with a mate , Resendez et al . tracked what happens in the brain of a prairie vole during selective aggression . The experiments showed that this aggressive behaviour coincides with changes in gene expression and brain chemistry that make it unpleasant for a prairie vole to be exposed to voles that are not its partner . For male prairie voles – but not females – these changes only happened if the female mating partner became pregnant during the cohabitation period . The changes that occur in the brain as a result of bonding with a partner also mean that drugs that are normally addictive are no longer pleasant and rewarding to the prairie vole . Indeed , forming a social bond between mating animals alters the brain in similar ways to the effects produced by addictive drugs . Thus , in a sense , each member of the mating pair becomes ‘addicted’ to their partner . The results presented by Resendez et al . also have implications for humans . They suggest that having a strong social support network is a powerful way of preventing casual drug use from developing into compulsive drug addiction . This may also mean that positive social relationships could help to treat people with drug addiction problems .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Dopamine and opioid systems interact within the nucleus accumbens to maintain monogamous pair bonds
Highly polymorphic major histocompatibility complex ( MHC ) molecules are at the heart of adaptive immune responses , playing crucial roles in many kinds of disease and in vaccination . We report that breadth of peptide presentation and level of cell surface expression of class I molecules are inversely correlated in both chickens and humans . This relationship correlates with protective responses against infectious pathogens including Marek's disease virus leading to lethal tumours in chickens and human immunodeficiency virus infection progressing to AIDS in humans . We propose that differences in peptide binding repertoire define two groups of MHC class I molecules strategically evolved as generalists and specialists for different modes of pathogen resistance . We suggest that differences in cell surface expression level ensure the development of optimal peripheral T cell responses . The inverse relationship of peptide repertoire and expression is evidently a fundamental property of MHC molecules , with ramifications extending beyond immunology and medicine to evolutionary biology and conservation . Highly polymorphic class I molecules encoded by the major histocompatibility complex ( MHC ) are crucial in the adaptive immune response to viruses and some intracellular bacteria , binding peptides inside the cell and presenting them on the cell surface to CD8 T lymphocytes ( Blum et al . , 2013; International HIV Controllers Study et al . , 2013; Trowsdale and Knight , 2013 ) . The impact of the MHC in response to human immunodeficiency virus ( HIV ) is well recognized , with class I alleles like HLA-B35:01 leading to rapid onset of AIDS while HLA-B57:01 and HLA-B27:05 confer long-term non-progression ( Carrington et al . , 1999; Goulder and Walker , 2012; International HIV Controllers Study et al . , 2013 ) . Various explanations for these associations have been suggested , including antigen presentation of a particularly effective peptide or of a number of peptides to cytotoxic CD8 T cells or recognition ( primarily ) independent of peptide by natural killer ( NK ) cells ( Carrington et al . , 1999; Martin et al . , 2002; Kosmrlj et al . , 2010; International HIV Controllers Study et al . , 2013 ) . Recently , the level of cell surface expression of HLA-C alleles correlated with CD8 T cell cytotoxicity has been proposed as one important basis for control ( Thomas et al . , 2009; Apps et al . , 2013 ) . Understanding how cell expression level might impact on disease resistance is complicated in humans due to the presence of three class I loci , so we began by studying a simpler animal system before examining human class I alleles . Long ago , we reported that the relative expression level of MHC class I molecules on the surface of chicken red blood cells as assessed by flow cytometry varies significantly , with cells of the MHC haplotype B21 approximately ten-fold lower than B4 , B12 , B15 , and B19 ( Kaufman et al . , 1995 ) . This finding was of interest because the level of cell surface expression is inversely correlated with the reported levels of MHC-determined resistance to Marek's disease , an economically important disease caused by the oncogenic herpesvirus , Marek's disease virus ( MDV ) . Decades of investigation identified B21 ( and other haplotypes like B2 , B6 , and B14 ) as generally conferring resistance and B19 ( and other haplotypes such as B4 , B12 , and B15 ) as generally conferring susceptibility ( reviewed in Plachy et al . , 1992 ) . On this basis , we proposed that MHC-determined resistance to Marek's disease could be due to the cell surface expression polymorphism of class I molecules ( Kaufman et al . , 1995; Kaufman and Salomonsen , 1997 ) . Compared to humans , the chicken MHC is relatively simple ( Kaufman et al . , 1999 ) , with two classical class I genes BF1 and BF2 that flank the genes for the transporter associated with antigen presentation ( TAP ) . The TAP transporter pumps peptides from the cytoplasm to the lumen of the endoplasmic reticulum for loading nascent class I molecules , and in typical mammals is functionally monomorphic , pumping a wide variety of peptides for all members of the polymorphic class I multigene family . In contrast , chicken TAP genes are highly polymorphic , with each haplotype encoding a TAP molecule with a peptide-translocation specificity matching the peptide-binding specificity of the dominantly expressed class I molecule encoded by the BF2 gene , with the BF1 gene expressed poorly or not at all ( Wallny et al . , 2006; Shaw et al . , 2007; Walker et al . , 2011 ) . In chickens , the peptide-binding specificity of the dominantly expressed class I molecule can determine resistance to infectious pathogens as well as responses to vaccines . The peptide motifs from B4 , B12 , B15 , and B19 class I molecules are if anything more fastidious than human and mouse motifs , with only one or two amino acids found at the positions of anchor residues , with the motifs explaining the immune response to infection and vaccination ( Kaufman et al . , 1995 ) . The fastidious binding of these molecules could be easily understood from wire models of the binding site , in which charged and hydrophobic residues were found in appropriate places to interact in a simple manner with the anchor residues of the bound peptides ( Wallny et al . , 2006 ) . This view was confirmed from the crystal structure of the dominantly expressed class I molecule BF20401 from the B4 haplotype , with positive-charged residues in a narrow groove allowing only certain anchor residues from the peptide to be accommodated ( Zhang et al . , 2012 ) . In contrast , much less peptide material was isolated from B21 cells , with many amino acids found in every peptide position . Crystal structures demonstrated that the dominantly expressed class I molecule BF22101 remodels the binding site to accommodate two peptides with completely different sequences , including the anchor residues ( Koch et al . , 2007 ) . Thus , the lack of a clear peptide motif could be explained as promiscuous peptide binding due to the remodelling of the peptide-binding site . In this paper , we find that two other haplotypes known to confer resistance to Marek's disease also have low cell surface expression and promiscuous peptide motifs , and examine the structural basis for promiscuous binding in three low expressing molecules . We show that the same relationship between cell surface expression and peptide-binding repertoire is found for four human class I molecules , associated with progression to AIDS . Finally , we propose how promiscuous peptide binding might confer resistance to some pathogens , what the basis for the cell surface expression polymorphism might be , and how these properties may relate to different strategies for resistance to pathogens . We reported that the relative level of class I molecules on the surface of erythrocytes is ten-fold lower for adult chickens of the B21 MHC haplotype compared to the B4 , B12 , B15 , and B19 haplotypes , as assessed by flow cytometry with two different monoclonal antibodies ( mAb ) ( Kaufman et al . , 1995 ) . Extending this work , we examined another cell type , spleen lymphocytes , with a quantitative flow cytometric assay using a mAb to chicken heavy chain . We found a similar rank order for these and other haplotypes ( Figure 1 ) , with the difference between the highest and lowest ranging up to fourfold over many experiments . The B2 , B14 , and B21 haplotypes can be considered as ‘low-expressing haplotypes’ , while B4 , B12 , B15 , and B19 haplotypes are ‘high-expressing haplotypes’ , with low-expression correlating historically with resistance to Marek's disease . 10 . 7554/eLife . 05345 . 003Figure 1 . Cell surface expression levels of class I molecules vary markedly between chicken haplotypes , as determined by a quantitative flow cytometric assay . Spleen cells from various inbred experimental chicken lines ( with MHC haplotypes indicated ) were stained with the monoclonal antibody F21-2 against chicken major histocompatibility complex ( MHC ) class I heavy chain and the specific antigen binding capacity ( SABC , which reflects number of epitopes per cell calculated in reference to specific antibody-binding calibration beads ) . Results are means of triplicate stains , with error bars indicating standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 05345 . 003 For the high-expressing haplotypes B4 , B12 , B15 , and B19 , the dominantly expressed class I molecules all have definite peptide motifs , with two or three peptide positions each bearing one or two chemically similar residues ( Kaufman et al . , 1995; Wallny et al . , 2006 ) . In contrast , much less peptide material was isolated from low expressing B21 cells , with pool sequences showing many amino acids in every peptide position ( Koch et al . , 2007 ) . Extending this work , we sequenced pools of peptides from erythrocytes and leukocytes of chickens with B2 and B14 haplotypes and found that they also showed many amino acids in each position , mostly with very different chemical characteristics ( Figure 2A ) . Sequencing of single peptides from B2 , B14 , and B21 haplotypes confirmed this sequence diversity , with no obvious anchor positions bearing one or two residues with similar chemical characteristics ( Figure 2B ) . 10 . 7554/eLife . 05345 . 004Figure 2 . Peptides isolated from class I molecules of B2 , B14 , and B21 chickens show promiscuity of peptide binding . For all panels , amino acids are in single letter code , with basic residues shown in blue , acidic in red , polar in green , hydrophobic in black . ( A ) Sequences of peptides bound to class I molecules isolated from three chicken strains determined from peptide pools showing apparent anchor , strong and weak signals . ( B ) Sequences of individual peptides , with confirmed anchor residues in bold . ( C ) Peptide anchor residues in large letters ( or question marks for unknown ) superimposed on a model of class I α1 and α2 domains with those residues of the major ( above ) and minor ( below ) class I sequences that are both polymorphic and potentially peptide contacts indicated as smaller letters; numbering based on human class I ( HLA-A2 ) sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 05345 . 004 Thus , the chicken class I molecules from high-expressing haplotypes have clear peptide motifs , at least as fastidious as such motifs described in humans and mice . In contrast , the class I molecules from low-expressing haplotypes have no obvious peptide motifs , with promiscuous binding unlike what has been described in mammals . To confirm that these properties result in different numbers of distinct peptides on the cell surface , we isolated class I molecules from equal numbers of cells from the B19 cell line 265L and the B21 cell line AVOL-1 , and analysed the bound peptides by mass spectrometry . Despite the B19 line having twice as many class I molecules on the cell surface as the B21 line as assessed by quantitative flow cytometry , there were only one third as many distinct peptides identified by mass spectrometry ( Figure 3 ) . Thus , a promiscuous class I molecule can and does bind a greater variety of peptides that appear on the surface of the cells , in comparison to a fastidious molecule . 10 . 7554/eLife . 05345 . 005Figure 3 . There is an inverse correlation between the cell surface expression levels of class I molecules and the variety of peptides isolated from class I molecules . ( A ) The B19 cell line 265L and the B21 cell line AVOL-1 were analysed by flow cytometry by staining with the mAb F21-2 to chicken class I molecules . AVOL-1 had slightly more autofluorescence , so the settings on the FACScan were adjusted so that the mean fluorescence intensity of the isotype control sample was the same as for 265L . The histogram shows the fluorescence intensity in the FL1 channel on the x-axis and the number of events on the y-axis . ( B ) In the same flow cytometry experiment , the calibration beads from the QIFIKIT were stained separately with the secondary antibody for calibration curves to calculate the SABC , which reflects the absolute numbers of epitopes on the cell surface . As a separate experiment , the class I molecules were isolated from each cell line by affinity chromatography with F21-2 and analysed by LC-MS/MS . Table shows the SABC and the number of different peptides found for each cell line . DOI: http://dx . doi . org/10 . 7554/eLife . 05345 . 005 Unlike high-expressing haplotypes ( Wallny et al . , 2006 ) , the wire models of the dominantly expressed class I of the B21 haplotype , BF22101 , gave no clue as to which peptides might bind ( Figure 2C ) . Crystal structures , based on heavy chain and β2-microglobulin ( β2m ) expressed in bacteria and refolded with synthetic peptides , showed that BF22101 remodels the binding site to accommodate two peptides with completely different sequences , including the anchor residues ( Koch et al . , 2007 ) . Small residues bordering the groove lead to a large cavity in the middle of the groove , within which Asp24 and Arg9 can move , creating different configurations , as illustrated for the 11mer peptide GHAEEYGAETL and the 10mer peptide REVDEQLLSV ( Figure 4A , B ) . The peptide position P2 His of the 11mer interacts with the Asp24 , while the Pc-2 Glu interacts with the Arg9 ( and Pc Leu fits in a hydrophobic pocket at the end of the groove ) . The 10mer remodels the binding site , so that the P2 Glu interacts with Asp24 that also interacts with Arg9 in a so-called charge transfer mechanism , creating a hydrophobic pocket that accommodates Pc-2 Leu ( and Pc Val fits in the hydrophobic pocket ) . 10 . 7554/eLife . 05345 . 006Figure 4 . Structures of BF22101 with different peptides show several modes of promiscuous binding through remodelling of the binding site . Left panels , top down view with peptide as sticks ( N-terminus to the left; carbon atoms , yellow; nitrogen atoms , blue; oxygen atoms , red ) and class I molecule as solid surface ( grey except for positions of Asp24 side chain oxygen atoms in pink and Arg9 side chain nitrogen atoms as cyan ) . Right panels , side view from α2 domain side with peptide , Asp24 and Arg9 as sticks ( hydrogen bonds , dotted lines; carbon atoms of Asp24 and Arg9 , white; all else as in left panels ) . ( A ) GHAEEYGAETL ( peptide P316; PDB 3BEV ) ; ( B ) REVDEQLLSV ( P330; 3BEW ) ; ( C ) TNPESKVFYL ( P458; 2YEZ ) ; ( D ) TAGQEDYDRL ( P394; 4D0B ) ; ( E ) TAGQSNYDRL ( P399; 4D0C ) ; ( F ) YELDEKFDRL ( P400; 4CVZ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05345 . 006 Additional peptides eluted from MHC molecules of B21 cells ( Figure 2B ) show that other amino acids at P2 and Pc-2 can be accommodated by the critical Asp24 and Arg9 within the large central cavity in the binding site . In fact , the 10mer TNPESKVFYL binds BF22101 in similar way to the 10mer REVDEQLLSV ( Table 1 , Figure 4C ) , with P2 Asn binding the Asp24 aided by charge transfer with Arg9 , and with the rearrangement of Arg9 permitting the accommodation of Pc-2 Phe , just as for Pc-2 Leu in the previous structure . 10 . 7554/eLife . 05345 . 007Table 1 . Data collection and refinement statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 05345 . 007PDBID4d0b4d0c4cvz2yez4cvx4d0d4cw1Data collection Space groupP212121P212121P212121P212121P65P212121P212121 Cell dimensions a , b , c ( Å ) 60 . 960 . 560 . 560 . 6173 . 988 . 162 . 269 . 268 . 969 . 069 . 0173 . 992 . 590 . 5895 . 494 . 893 . 794 . 887 . 5223 . 6144 . 8 α , β , γ ( ° ) 9090909090909090909090909090909090901209090 Resolution ( Å ) 56 . 03–2 . 80 ( 2 . 95–2 . 80 ) 51 . 00–2 . 82 ( 2 . 89–2 . 82 ) 50 . 84–2 . 39 ( 2 . 56–2 . 39 ) 41 . 03–2 . 45 ( 2 . 58–2 . 45 ) 75 . 66–3 . 30 ( 3 . 56–3 . 30 ) 88 . 08–3 . 13 ( 3 . 21–3 . 13 ) 72 . 38–2 . 58 ( 2 . 65–2 . 58 ) Rsym or Rmerge0 . 118 ( 0 . 30 ) 0 . 12 ( 0 . 30 ) 0 . 15 ( 0 . 91 ) 0 . 07 ( 0 . 22 ) 0 . 17 ( 0 . 56 ) 0 . 18 ( 0 . 69 ) 0 . 14 ( 0 . 59 ) I/σI8 . 1 ( 2 . 8 ) 8 . 2 ( 1 . 9 ) 11 . 9 ( 2 . 1 ) 9 . 5 ( 1 . 5 ) 11 . 1 ( 3 . 8 ) 8 . 8 ( 2 . 1 ) 8 . 8 ( 2 . 3 ) Completeness ( % ) 91 . 2 ( 89 . 8 ) 90 . 8 ( 76 . 0 ) 99 . 9 ( 100 ) 85 . 9 ( 53 . 0 ) 100 ( 99 . 9 ) 99 . 6 ( 98 . 8 ) 99 . 3 ( 98 . 4 ) Redundancy4 . 0 ( 3 . 6 ) 3 . 8 ( 1 . 9 ) 6 . 4 ( 6 . 5 ) 2 . 8 ( 1 . 6 ) 7 . 3 ( 7 . 5 ) 4 . 6 ( 4 . 4 ) 5 . 5 ( 4 . 2 ) Refinement Resolution ( Å ) 56 . 03–2 . 8050 . 92–2 . 8150 . 84–2 . 3941 . 03–2 . 4575 . 66–3 . 3088 . 08–3 . 1372 . 38–2 . 58 No . reflections9361896716 , 086891321 , 70732 , 91626 , 228 Rwork/Rfree27 . 5/29 . 6%27 . 5/28 . 2%25 . 9/26 . 9%25 . 2/27 . 3%23 . 8/26 . 2%29 . 2/29 . 9%28 . 6/29 . 1% Number of atoms Protein3061305730793091605212 , 1026044 Ligand/ion04240000 Water53142308977 B-factors Protein30 . 3034 . 2430 . 2332 . 2277 . 8647 . 0240 . 02 Ligand/ion–26 . 3246 . 25–––– Water14 . 3312 . 8824 . 388 . 6134 . 8734 . 3129 . 50 Root mean square ( r . m . s . ) deviations Bond lengths ( Å ) 0 . 0080 . 0070 . 0070 . 0070 . 0040 . 0070 . 007 Bond angles ( ° ) 0 . 890 . 860 . 850 . 830 . 7590 . 820 . 84Highest resolution shell is shown in parenthesis . However , the two structures with the 10mers TAGQEDYDRL and TAGQSNYDRL display a completely different mode of binding ( Table 1 , Figure 4D , E ) , for which the P2 Ala does not interact with the MHC molecule at all ( and therefore is not an anchor residue ) , with Arg9 interacting as a bridge between Asp24 and Pc-2 Asp . A further mode of binding is shown by the 10mer YELDEKFDRL ( Table 1 , Figure 4F ) , for which both anchor residues at positions P2 and Pc-2 are acidic . In this structure , the P2 Glu interacts with both Asp24 and Arg9 , and Pc-2 Asp interacts with Arg9 . These various modes of binding are presumably just a few out of many and illustrate the promiscuous binding of BF22101 , unlike anything seen for mammalian class I molecules . Similar to B21 , the peptides isolated from cells of the low expressing haplotypes B2 and B14 have no obvious motifs . Pool sequences show no position at which only one or two chemically similar amino acids are present as anchor residues ( Figure 2A ) . Individual peptides have no classic pattern of anchor residues , but some features are discernable ( Figure 2B ) . The B2 peptides fall into two groups . One small group of peptides has P2 Arg and mostly P1 Arg and Pc Tyr ( Figure 2B ) , much like the dominantly expressed class I molecule of the high-expressing B15 haplotype , BF21501 ( Wallny et al . , 2006 ) . These three peptides are likely to have been isolated from the poorly expressed molecule encoded by the minor gene of the B2 haplotype , BF10201 , the wire model of which ( Figure 2C ) looks very similar to BF21501 . Size exclusion chromatography ( SEC ) of heavy chain and β2m expressed in bacteria and refolded with peptide showed that BF10201 but not BF20201 binds these peptides ( Figure 5 ) . 10 . 7554/eLife . 05345 . 008Figure 5 . The dominantly expressed class I molecule BF20201 binds VIFPAKSL but not RRALREGY , while the minor class I molecule BF10201 binds RRALREGY but not VIFPAKSL . ( A and B ) Size exclusion chromatography ( SEC ) traces for BF20201 or BF10201 heavy chains expressed in bacteria refolded with or without β2-microglobulin ( β2m ) and peptide . The heavy chain BF20201 refolded with β2m and the appropriate peptide migrates as a native monomer , whereas refolded with the inappropriate or no peptide migrates in the same position as heavy chain alone . In contrast , all these conditions for the heavy chain BF10201 result in molecules that migrate roughly the same mobility . ( C through F ) Mass spectrometry ( MALDI-TOF ) analysis of the monomer peaks of heavy chain refolded with β2m and peptide shows that VIFPAKSL but not RRALREGY can be recovered from BF20201 , while RRALREGY but not VIFPAKSL can be recovered from BF10201 . Note the many peaks for BF10201 with VIFPAKSL and for BF20201 with RRALREGY , representing background contaminants detected as sensitivity was increased in the search for the synthetic peptide . Comparable results were obtained with YPYLGPNTL , RRALREGY , RRGGVKRI , and the B15 ( BF21501 ) peptide KRLIGKRY . DOI: http://dx . doi . org/10 . 7554/eLife . 05345 . 008 All of the other B2 peptides have generally smaller hydrophobic amino acids at P2 and somewhat larger hydrophobic amino acids at Pc ( Figure 2B ) . Crystal structures with two of these peptides ( Table 1 , Figure 6A , B ) , YPYLGPNTL and VIFPAKSL , show that Pro and Ile at peptide position P2 and Leu at Pc bind shallow hydrophobic pockets ( Figure 6A , B ) , similar to the way in which anchor residues bind pockets B and F in many mammalian class I molecules . This mode of binding is completely different from BF22101 that remodels the binding site , despite the fact that BF20201 shares Asp24 and Arg9 with BF22101 . In these BF20201 structures , Arg9 interacts mostly with peptide main chain atoms and Asp24 interacts with Tyr43 ( residue 45 being the equivalent position in human class I molecules ) , a residue that differs between BF20201 and BF22101 , ultimately leading to a more hydrophobic pocket B to accommodate P2 residues ( Figure 6A , B ) . 10 . 7554/eLife . 05345 . 009Figure 6 . Structures of BF20201 and BF21401 show promiscuous binding via hydrophobic binding pockets for the anchor residues at peptide position P2 and Pc , with the class I residues at positions 24 and 9 playing supporting roles , and with residues lining the pockets explaining the relative size of anchor residues . Upper panels , side view from α2 domain side with peptide as sticks ( N-terminus of peptide to the left; carbon atoms of peptide , yellow; carbon atoms of class I molecule , white; nitrogen atoms , blue; oxygen atoms , red; hydrogen bonds , dotted lines; carbon atoms of Asp24 and Arg9 , white ) . Middle ( pocket B ) and bottom ( pocket F ) panels , side view cut-away from α2 domain side with peptide and selected class I residues as sticks ( numbering based on chicken class I sequence ) and with rest of MHC molecule as solid surface . ( A ) YPYLGPNTL bound to BF20201 ( peptide P377; PDB 4CVX ) ; ( B ) VIFPAKSL bound to BF20201 ( P473; 4D0D ) ; ( C ) SWFRKPMTR bound to BF21401 ( P479; 4CW1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05345 . 009 The B14 peptides all share two features , generally a medium to large hydrophobic amino acid at P2 and one or more basic amino acids at the end of the peptide ( Figure 2B ) . A wire model of BF21401 shows small hydrophobic and polar residues in and around the beginning of the groove and reveals many acidic residues at the end of the groove , well placed to interact with the basic residues at the peptide C-terminus ( Figure 2C ) . The crystal structure of peptide SWFRKPMTR bound to BF21401 shows that Trp at position P2 interacts with hydrophobic surfaces in pocket B , while Arg at Pc interacts with Asp73 ( human position 74 ) , Asp76 ( 77 ) and Asp113 ( 116 ) in pocket F ( Table 1 , Figure 6C ) . Both these pockets are much deeper in BF21401 compared to BF20201 , which explains the larger anchor residues found in the B14 peptides compared to most B2 peptides . The differences in pocket sizes are primarily due to difference in size between Val43 ( 45 ) and Tyr43 ( 45 ) for pocket B and between Asp113 ( 116 ) and Met113 ( 116 ) for pocket F , which are supported by a network of hydrogen bonds from Thr24 and Gln9 in BF21401 and from Asp24 and Arg9 in BF20201 ( Figure 6 ) . These data show that the three class I molecules with lower expression on the cell surface all bind a wide range of peptides with no obvious peptide motif , although the mechanism by which the promiscuous binding is achieved is different for each molecule . One feature that may be important is the width of the peptide-binding groove ( Figure 7 ) , which is the narrowest in the high expressing BF20401 molecule ( Zhang et al . , 2012 ) and the widest in the low expressing BF22101 . 10 . 7554/eLife . 05345 . 010Figure 7 . Structures of chicken class I molecules show differences in the width of the peptide-binding groove , with the fastidious BF20401 having the narrowest groove and the promiscuous BF22101 being the widest in the centre of the groove . Top down view with peptide as sticks ( N-terminus to the left; carbon atoms , yellow; nitrogen atoms , blue; oxygen atoms , red ) and class I molecule as grey solid surface . ( A ) VIFPAKSL bound to BF20201 ( P473; 4D0D ) ; ( B ) IDWFEGKE bound to BF20401 ( IE8; 4G43 ) ; ( C ) SWFRKPMTR bound to BF21401 ( P479; 4CW1 ) ; ( D ) YELDEKFDRL bound to BF22101 ( P400; 4CVZ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05345 . 010 The structural analysis of BF22101 showed that this low expressing class I molecule achieves promiscuous binding by remodelling the peptide binding site in a way never described for a mammalian class I molecule . However , analysis of BF20201 and BF21401 shows promiscuous binding by chicken molecules using anchor positions into pockets B and F , much like many mammalian classical class I molecules . In order to determine whether the inverse correlation between cell surface expression level and peptide binding promiscuity is a feature just of chickens or is indicative of a more fundamental property , we examined some human class I alleles . Only a few studies have explored the extent of the peptide-binding repertoire of different human class I molecules . One of these studies reported the predicted peptide-binding repertoires for four human class I alleles , finding a rank hierarchy from the extremely fastidious HLA-B57:01 to HLA-B27:05 to HLA-B07:02 to the highly promiscuous HLA-B35:01 , that correlated directly with progression from HIV infection to AIDS ( Kosmrlj et al . , 2010 ) . The rank hierarchy of these alleles is the same as that determined by binding of peptide libraries ( Paul et al . , 2013 ) . We identified two mAbs reported ( Apps et al . , 2009 ) to react with all four HLA-B alleles ( along with HLA-C alleles , which are poorly expressed on blood cells ) , but not with certain HLA-A alleles . Volunteers with the proper combinations of homozygous HLA-A and HLA-B alleles were recruited from a large group of bone marrow donors ( Table 2 ) , and the cell surface expression levels on their blood lymphocytes and monocytes were examined by quantitative flow cytometry . The two mAb ( Tu149 and B1 . 23 . 2 ) reacted with all four HLA-B alleles and gave similar ( but not identical ) results for both lymphocytes and monocytes , with a rank hierarchy in cell surface expression ranging from HLA-B57:01 as the highest to HLA-B27:05 to HLA-B07:02 to HLA-B35:01 as the lowest ( Figure 8A–D ) . 10 . 7554/eLife . 05345 . 011Table 2 . Anonymized donors with class I alleles and inferred HLA-C expressionDOI: http://dx . doi . org/10 . 7554/eLife . 05345 . 011DonorHLA-A allelesHLA-B allelesHLA-C allelesHLA-C expression5701/0202:01 , 02:0157:01 , 57:0106:02 , 15:02high , unknown5701/0301:01 , 01:0157:01 , 57:0106:02 , 06:02high , high5701/0401:01 , 03:0157:01 , 57:0106:02 , 07:01high , low2705/103:01 , 68:0127:05 , 27:0501:02 , 02:02high , high2705/202:06 , 11:0127:05 , 27:0501:02 , 03:03high , low2705/302:01 , 02:0127:05 , 27:0501:02 , 12:03high , high2705/402:01 , 02:1227:05 , 27:05unknown0702/102:01 , 03:0107:02 , 07:0207:02 , 07:02low , low0702/203:01 , 03:0107:02 , 07:0207:02 , 07:02low , low0702/302:01 , 03:0107:02 , 07:0207:02 , 07:02low , low0702/403:01 , 11:0107:02 , 07:0207:02 , 07:02low , low3501/111:01 , 11:0135:01 , 35:0104:01 , 04:01high , high3501/211:01 , 11:0135:01 , 35:0104:01 , 04:01high , high3501/303:01 , 03:0135:01 , 35:0104:01 , 04:01high , high3501/411:01 , 11:0135:01 , 35:0104:01 , 04:01high , high10 . 7554/eLife . 05345 . 012Figure 8 . Human class I molecules show an inverse correlation between cell surface expression level and peptide binding promiscuity . Levels of specific antibody binding capacity ( SABC ) for different mAb: ( A and B ) Tu149 , ( C and D ) B1 . 2 . 23 , ( E and F ) 22E-1 for ( A , C and E ) ex vivo lymphoctyes and ( B , D and F ) ex vivo monocytes . Each point represents the sample from a particular donor ( identified with anonymous labels that correlate with haplotypes in Table 2; donor B57:01/01 failed to donate ) ; bars indicate the mean for each HLA-B allele . DOI: http://dx . doi . org/10 . 7554/eLife . 05345 . 012 One potential concern with these data was that the difference in expression level of the HLA-B alleles might actually be due to differences in HLA-C levels , which are known to vary between alleles ( Thomas et al . , 2009; Apps et al . , 2013 ) . However , inspection shows that there is no correlation of expression levels reported for the HLA-C alleles found for the individual donors and the expression level of the mAb binding ( Table 2 ) , if anything the reverse with the low expressing HLA-B35:01 donors all having HLA-C alleles reported to have high expression . Another concern was that the difference in expression level of the HLA-B alleles actually reflects differences in affinity of binding by the two mAb . The fact that both antibodies gave similar but non-identical results suggests that they do not recognize exactly the same epitopes and therefore are unlikely by chance to vary similarly in affinity . We tested whether the two antibodies bound independently and found that B1 . 23 . 2 inhibited both itself and Tu149 very well , while Tu149 inhibited both itself and B1 . 23 . 2 much less well ( Figure 9 ) . These results suggest that B1 . 23 . 2 has a much higher affinity than Tu149 and that their epitopes overlap . We also utilized a third antibody ( 22E-1 ) that reacts with the HLA-B57:01 and HLA-B27:05 but not HLA-B07:02 and HLA-B35:01 ( nor with any HLA-C allele ) , and therefore must recognize a different epitope . We found the same relationship of expression level for HLA-B57:01 and HLA-B27:05 as with the other two antibodies ( Figure 8E , F ) . 10 . 7554/eLife . 05345 . 013Figure 9 . The epitopes on HLA-B57:01 for mAb B1 . 23 . 2 and Tu149 are overlapping , with B1 . 23 . 2 having a much higher affinity . Win cells were treated with saturating amounts of unlabelled Tu149 ( blue ) , saturating amounts of unlabelled B1 . 23 . 2 ( green ) or PBS ( red ) , and then stained with either A Tu149 conjugated to APC or B B1 . 23 . 2 conjugated to Alexa Fluor 647 . Results were compared to staining with only isotype control ( grey ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05345 . 013 Thus , in humans the high expressing class I molecules have fastidious peptide binding , and the low expressing molecules have promiscuous binding . We conclude that the correlation between peptide-binding repertoire and expression level is the same in chickens and humans and is indicative of a fundamental property of class I molecules . In this article , we make three major points . First , we extend previous work showing that the chicken BF22101 achieves promiscuity by remodelling the peptide binding site in a way unknown for mammals , but we find that two other low expressing chicken class I molecules achieve promiscuity with the same pockets typically used in mammals . By extension , these latter results provide a potential molecular explanation for recent reports that human class I molecules vary in peptide repertoire and other reports that they vary in cell surface expression level . Second , we show that peptide-binding repertoire and level of cell surface expression are inversely correlated in both chickens and humans , indicating that this relationship is a fundamental property of class I molecules . Finally , we find that there is a clear association with resistance to infectious diseases , with the low expressing promiscuous class I molecules associated with resistance to Marek's disease in chickens but associated with faster progression to AIDS in humans . We discuss each of these points in turn . The first point is the structural and mechanistic basis for differences in peptide repertoire of chicken class I molecules . In this article , we confirm and extend the finding ( Koch et al . , 2007 ) that BF22101 binds 10mer and 11mer peptides with three anchor positions , with many different amino acids with different chemical characteristics , mostly charged and polar at positions P2 and Pc-2 ( and hydrophobic at Pc ) , which is unlike any mammalian molecule described thus far . In contrast , we find that BF20201 and BF21401 bind 8mer and 9mer peptides with two anchor residues ( much like typical mammalian class I molecules ) , each with a variety of amino acids at P2 and Pc . One question is why BF22101 and BF20201 bind in such different ways , given that both have Asp24 and Arg9 in their binding sites . In part , this may be due to the difference in residues lining the peptide-binding groove . In BF22101 , small residues like Gly68 ( human position 69 ) , Ser69 ( 70 ) , Ser97 ( 99 ) , and Gly152 ( 155 ) create a big cavity in the centre of the binding groove , allowing Asp24 and Arg9 to move . In BF20201 and BF21401 , the two Ser residues are replaced by the large residues Asn69 ( 70 ) and Tyr97 ( 99 ) , creating a constricted groove much more like narrow grooves found in high expressing fastidious molecules such as BF20401 . It is also possible that the length of peptides is determined at least in part by TAP specificity . Together , the narrower groove and the shorter peptides ensure that Asp24 and Arg9 in BF20201 ( as well as Thr24 and Gln9 in BF21401 ) cannot reach the P2 and Pc-2 positions , and thus they interact ( if at all ) with the main chain atoms of the peptide . Another question is how chicken class I molecules BF20201 and BF21401 compare to human molecules like HLA-A2 , all of which bind peptides with hydrophobic anchor residues . HLA-A2 has one of the widest peptide repertoires among human class I molecules , as assessed by peptide-binding studies ( Paul et al . , 2013 ) . The anchor residues for HLA-A2 fit into narrow and specific-binding pockets , which allow almost only P2 Met and Leu for pocket B and Pc Val and Leu for pocket F ( Guo et al . , 1993 ) . Although Leu and Val in particular are relatively common in proteins ( so that the number of peptides bearing these amino acids should be relatively high ) , there is a much larger number of hydrophobic amino acids that are accommodated by the shallower pockets in both BF20201 and BF21401 . Hence , it seems likely that the peptide repertoire of the most promiscuous chicken class I molecules will be greater than the most promiscuous human class I molecules . Finally , are there obvious mechanisms by which differences in peptide repertoire arise ? In chickens , the peptides presented depend on the peptide-binding specificities of the class I molecule , but also on the peptide-translocation specificities of the TAPs and perhaps also on peptide-editing by tapasin . The genes for chicken TAP and tapasin have at least as many alleles as class I genes ( Walker et al . , 2005 , 2011; van Hateren et al . , 2013 ) . On this basis , we propose that the difference in peptide repertoire ( and in cell surface expression level ) between chicken class I alleles is determined at the level of peptide loading and editing , likely to take place mostly in the peptide-loading complex . It might not be immediately apparent whether such an explanation could account for the differences in peptide repertoire and cell surface expression of human class I molecules , since human TAPs and tapasin are functionally monomorphic . However , a recent paper suggests that the assembly and cell surface expression of different HLA-B alleles is determined by interaction with the near invariant tapasin molecule , based on variation in the certain HLA-B residues ( Rizvi et al . , 2014 ) . Indeed , it is striking that the rank order of expression level that we find for the four HLA-B alleles tested is exactly the same as the rank order of tapasin dependence found in this recent paper . The second important point for discussion is why the cell surface expression would be inversely correlated with peptide-binding repertoire . A proximal explanation might be that the biochemical mechanism of peptide loading , as mentioned in the paragraph above , leads to this relationship . Another explanation , similar to the suggestion that multiple MHC molecules lead to a greater chance of autoimmunity ( Kaufman et al . , 1995 ) , is that promiscuous class I molecules confer more resistance to pathogens but also lead to greater autoimmunity compared to fastidious class I molecules . We favour a third possibility , prompted by the experimental and supporting theoretical evidence for the notion that a greater number of MHC loci would lead to greater level of negative selection in the thymus ( Vidović and Matzinger , 1988; Nowak et al . , 1992 ) . In a similar way , one might expect that class I molecules presenting a lesser variety of peptides would negatively select fewer T cell clones , but class I molecules presenting a greater variety of peptides would negatively select many more T cells and create a pauperized T cell repertoire . Therefore , we propose that the peptide-binding repertoire and the cell surface expression levels are inversely correlated precisely to allow similar numbers of T cell clones to survive negative selection and be available in the periphery . In this view , low expressing and promiscuous class I molecules might present many more peptides but each individual peptide would be present on the surface at a much lower level , so that T cells would escape the negative selection that might have occurred at a higher concentration of a particular MHC-peptide complex . In essence , we propose that peptide-binding repertoires and cell surface expression levels have evolved to optimize peripheral T cell responses . Many papers have examined naïve CD8 T cell repertoire in humans and mice ( for example , Alanio et al . , 2010; Jenkins and Moon , 2012; Lo et al . , 2014 ) ; such analyses with high and low expressing haplotypes would be informative . The third important point for discussion is the fact that the inverse correlation between peptide binding repertoire and cell surface expression level also correlates with resistance and susceptibility to certain infectious pathogens . One important question is how it might work . In chickens , these phenomena are correlated with MHC-determined resistance to historic Marek's disease . Many genetic loci contribute to resistance to Marek's disease , but historically the B locus ( containing the MHC ) was by far the most important , and under strong selection by MDV ( Plachy et al . , 1992; Vallejo et al . , 1998 ) . Here , we show that the low expressing promiscuous class I molecules are associated with resistance to Marek's disease , while the high expressing fastidious class I molecules are associated with susceptibility . A consequence of promiscuous binding is that these low expressing class I molecules present a larger repertoire of peptides which should activate a wider range of T cell clones , which may be beneficial in the immune response to certain pathogens . We propose that this breadth of antigen presentation leads to a breadth of T cell response that is the basis of MHC-determined resistance to Marek's disease . Conversely , a narrow T cell response to Marek's disease would be the basis for MHC-determined susceptibility , and there is already evidence for a limited repertoire of CD8 T cell clones infiltrating tumours in susceptible B19 chickens ( Mwangi et al . , 2011 ) . Parenthetically , if true this model suggests that the narrowing of CD8 T cell clonality characteristic of responses in humans and mice ( Yewdell and Bennink , 1999; Yewdell and Del Val , 2004; Akram and Inman , 2012 ) may not be a feature of responses involving promiscuous class I molecules in chickens . For our example in humans , however , we find the opposite disease association . We find that the low expressing promiscuous HLA-B alleles are associated with rapid progression from HIV infection to AIDS , while the high expressing fastidious HLA-B alleles are found in non-progressors ( Carrington et al . , 1999; Goulder and Walker , 2012; International HIV Controllers Study et al . , 2013 ) . The correlation with peptide binding repertoire ( as assessed by prediction ) of HLA-B alleles had already been noted ( Kosmrlj et al . , 2010 ) , but the relationship with cell surface expression level has not . The controller alleles HLA-B57:01 and HLA-B27:05 are known to bind and present particular viral peptides that are both protective and difficult for the virus to change without a loss in fitness ( Gillespie et al . , 2006; Schneidewind et al . , 2007 ) . Thus , particular class I molecules may be selected because their fastidious motif presents a particularly protective peptide . Is there an obvious evolutionary basis for differences in peptide-binding repertoire ? The fact that the fastidious HLA-B57:01 confers resistance to AIDS while the promiscuous BF22101 confers resistance to Marek's disease may be no paradox , in that breadth of peptide presentation may be an appropriate response to some pathogens , while presentation of a particular peptide may be more suitable for other pathogens . Viewed from this perspective , the work presented in this paper has begun to define two groups ( or a range between two extremes ) of class I molecules that are strategically evolved for different modes of resistance . One possible strategy is that fastidious class I molecules with pauci-clonal T cell responses might be better suited for rapidly evolving viruses with a limited scope for immune evasion , for which a particular peptide might be the most efficient way to achieve resistance . In contrast , promiscuous class I molecules might be better suited for pathogens with coding potential for many immune evasion genes and more stability over evolutionary time , such as large DNA viruses , bacteria , and even parasites . There are few reports about pathogen peptides that are presented in chickens , but the large literature on the MHC restriction of viruses , bacteria , and parasites in humans and mice ( for example , Fitzmaurice et al . , 2015; Goulder and Walker , 2012; Macnamara et al . , 2010; Moss and Khan , 2004; Steven et al . , 1997; Yim and Selvaraj , 2010 ) may be reinterpreted , as the peptide repertoire and cell surface expression levels of the class I molecules become known . Another potentially fruitful way of considering the strategies of class I molecules ( not necessarily exclusive of the first view ) might be as generalists and specialists . There is a large literature in biology that examines the generalist/specialist paradigm at the species level , typically testing the model that ‘the jack of all trades is the master of none’ ( McArthur , 1972; Futuyma and Moreno , 1988 ) . Recent examples , among many others , include breadth of diet by insect herbivore species ( Ali and Agrawal , 2012 ) , niche breadth by bird species ( Julliard et al . , 2006 ) , competition on shared hosts by aphid parasitoids ( Straub et al . , 2011 ) , and weasel predation of voles ( Sundell and Ylönen , 2008 ) , as well as mathematical models to disentangle the contributions by various factors ( Remold , 2012; Büchi and Vuilleumier , 2014 ) . Indeed , the terms generalist and specialist MHC alleles have already been used for describing correlations of class II alleles of the striped mouse in Africa with number of nematode species carried ( Froeschke and Sommer , 2012 ) . In this view , the promiscuous class I molecules would be generalists , which one might expect to suffice for protection against a wide variety of the most common pathogens . However , such generalists might not suffice for protection from a new and/or an especially virulent pathogen that suddenly appears , at which point there would be a strong selection for a specialist class I molecule that was particularly suited to deal with the new threat . The properties of fastidious class I molecules are consistent with selection as specialists for particular pathogens , perhaps including those no longer a danger in the current population . The importance of these concepts for immunology and medicine may be clear from the discussion above , but there are also ramifications for evolutionary biology , ecology , and conservation . As one example , the number of MHC alleles is considered a key measure for population diversity in estimating the risk of extinction , both as a measure of overall genome diversity and in terms of fitness , including resilience to infection ( Sommer , 2005 ) . However , a population with a few generalist MHC alleles might remain healthy compared to a population with many inappropriate specialist MHC alleles . Ex vivo chicken cells were from inbred chicken lines with known MHC haplotypes ( Shaw et al . , 2007 ) , kept at the University of Cambridge . All procedures were performed under appropriate Home Office Licenses and after review by the Ethics Committee at the University of Cambridge . Spleens were mashed through 100-μm nylon cell strainers ( Falcon ) in RPMI-1640 medium , supernatants taken after 5 min settling , and spleen cells recovered after centrifugation at 400×g for 5 min . All procedures were carried out under Home Office licenses and with ethical approval . Chicken cell lines were from the Pirbright Institute . AVOL-1 was derived from in vitro transformation of spleen cells from a line 0 ( B21 ) chicken by the reticuloendotheliosis virus REV-T ( Smith , 2004; Yao et al . , 2008; W Mwangi and V Nair , unpublished data ) . The MDCC-265L cell line was established from a liver lymphoma of a line P2a ( B19 ) chicken infected with the RB-1B virus derived from a BAC clone ( as in Yao et al . , 2009; W Mwangi and V Nair , unpublished data ) . Both lines were maintained in RPMI 1640 medium containing 10% foetal bovine serum , 10% tryptose phosphate broth and 1% sodium pyruvate , and at 38 . 5°C in 5% CO2 . Ex vivo human cells were from Anthony Nolan registrants typed as homozygous for particular HLA-A and HLA-B locus alleles , who signed written consent forms , and with all procedures carried out under Human Tissue Act licenses and with ethical approval . Blood samples were collected by general practice or Walk-in Clinic phlebotomists and were couriered to the Anthony Nolan Research Institute within 24 hr . Whole blood was diluted 1:1 upon arrival with transport media ( RPMI 1640 [Lonza , Belgium] supplemented with 0 . 6% tri-sodium citrate and 50 nM 2-mercaptoethanol ) , and most samples were rocked at room temperature overnight . In a similar manner as originally described for cord blood ( Figueroa-Tentori et al . , 2008 ) , peripheral blood mononuclear cells were isolated ( all steps at 20°C ) using a density gradient centrifugation ( Ficoll–Paque Plus 1077 , GE Healthcare ) at 840×g for 30 min with no brake , with the buffy coat washed twice with two volumes RPMI-1640 media , spun once at 680×g for 10 min and once at 540×g for 10 min . All samples were stained , fixed , and anonymized before transfer to Cambridge for analysis by flow cytometry . The human homozygous HLA-B57 typing cell line WIN ( alias IHW9095 from 10th International Workshop , gift of W Bultitude and S Marsh , Anthony Nolan Research Institute ) was maintained in RPMI1640 with 10% foetal bovine serum and 1 mM glutamine in 5% CO2 at 37°C . As described by instructions for quantitative flow cytometry from manufacturer ( QIFIKIT , Dako ) and following previous work ( Smith and Ellis , 1999 ) , 5 × 105 cells were incubated on ice in 96-well ( U-bottom for chicken , V-bottom for human ) microtiter plates ( Nunc ) with saturating primary antibody followed by washing and then by incubation with goat anti-mouse secondary antibody conjugated to fluorescein followed by washing , and data acquired using a FACscan ( Becton-Dickenson ) . Set-up beads and calibration beads were stained separately with the secondary antibody for calibration curves to calculate the specific antigen binding capacity , which reflects the absolute numbers of epitopes on the cell surface . Primary mAb include 200 µl tissue culture supernatant of mouse mAb F21-2 for chicken class I molecules ( Crone et al . , 1985 ) and three that react with certain HLA-B antigens but not certain HLA-A alleles ( Apps et al . , 2009 ) : 200 µl tissue culture supernatant of Tu149 ( Uchańska-Ziegler et al . , 1993; gift of J Trowsdale , University of Cambridge ) , 20 µl 1 mg/ml B1 . 2 . 23 ( Rebaï and Malissen , 1983; bought from eBioscience ) diluted in PBS and 20 µl 1 mg/ml 22E1 ( Tahara et al . , 1990; bought from Caltag Medsystems ) diluted in PBS . Saturation was confirmed by staining with dilutions of antibodies on each set of cells , except for human ex vivo cells , for which saturation was confirmed by staining WIN cells . For the inhibition assay , Tu149 conjugated to APC ( Invitrogen; kind gift of S Ashraf and J Trowsdale , University of Cambridge ) and purified B1 . 23 . 2 conjugated to Alexa Fluor 647 ( purified mAb from eBioscience conjugated using the Antibody Labeling kit , Molecular Probes/Life Technologies , according to manufacturer's instructions ) were used . As above , 5 × 105 WIN cells were incubated with saturating amounts of unlabelled B1 . 23 . 2 or Tu149 ( or with PBS ) for 1 hr , washed and then incubated with a directly labelled antibody for 1 hr , before washing the cells again . Data were acquired using the red laser on a Cytek FACSanalyser ( Becton Dickinson ) . Inbred chicken lines , with known MHC haplotypes ( Shaw et al . , 2007 ) , were kept at the Institute for Animal Health at Compton . All procedures were performed under appropriate Home Office Licenses and after review by the Ethics Committee at the Institute for Animal Health . As previously described ( Wallny et al . , 2006; Koch et al . , 2007 ) , erythrocytes ( or whole mashed up spleens ) were solubilized in detergent , and class I molecules were isolated by affinity chromatography using mAb F21-2 against chicken class I molecules or F21-21 against chicken β2m . Peptides were eluted using trifluoroacetic acid and separated by reverse phase high-pressure liquid chromatography with single peaks of abundant peptides as well as pools of non-abundant peptides sequenced by Edman degradation . Immunoaffinity beads were produced , with all steps at room temperature . Protein G-Sepharose beads ( Expedeon ) were washed with 50 mM borate , 50 mM KCl , pH 8 . 0 , the equivalent of 1-ml packed beads was incubated with 3 mg F21-2 ( produced by the Microbiological Media Services of the Pirbright Institute ) for 1 hr , treated with 40 mM dimethyl pimelimidate dihydrochloride ( Sigma ) in 0 . 1 M triethanolamine , pH 8 . 3 for 1 hr to cross-link the antibody to the protein G , washed with 100 mM citric acid pH 3 . 0 , and equilibrated in 50 mM Tris , pH 8 . 0 . The two cell lines AVOL1 and 265L were washed with PBS . Pellets of 109 cells were incubated with 10 ml 1% Igepal 630 , 300 mM NaCl , 100 mM Tris pH 8 . 0 for 30 min at 4°C , subcellular debris was pelleted by centrifugation at 300×g for 10 min and 15 , 000×g for 30 min at 4°C , and the cleared lysates were incubated with 1 ml immunoaffinity beads for 1 hr at 4°C . The beads were washed with 50 mM TrisCl , pH 8 . 0 buffer , first with 150 mM NaCl , then with 400 mM NaCl and finally with no salt . Bound material was eluted with 10% acetic acid . The eluted material was dried , resuspended in 3% acetonitrile , 0 . 1% formic acid in water , loaded directly onto on a 4 . 6 × 50 mm ProSwiftTM RP-1S column ( ThermoFisher ) and eluted at 500 μl/min flow rate for 10 min with a linear gradient from 2 to 35% buffer B ( 0 . 1% formic acid in acetonitrile ) in buffer A ( 0 . 1% formic acid in water ) using an Ultimate 3000 HPLC system ( ThermoFisher ) , with fractions collected from 2 to 15 min . Protein detection was performed at 280 nm absorbance , with fractions eluting before β2m pooled and dried . For liquid chromatography tandem mass spectrometry ( LC-MS/MS ) , peptides were analysed using either a Q-Exactive ( Thermo Scientific ) or a TripleTOF 5600 ( AB Sciex ) system . For the Q-Exactive system , peptides were separated on a Ultimate 3000 RSLCnano System utilizing a PepMap C18 column , 2 μm particle size , 75 μm × 50 cm ( Thermo Scientific ) with a linear gradient from 3% to 35% buffer B in buffer A ( as above ) at a flow rate of 250 nl/min ( ∼65 MPa ) for 60 min , and the 15 most intense precursors per full MS scan were selected for MS/MS analysis using HCD fragmentation . For the TripleTOF system , peptides were separated with a 15 cm × 75 µm ChromXP C18-CL ( 3 µm particle size ) using an ekspert nanoLC 400 cHiPLC system ( Eksigent ) with a linear gradient from 8% to 35% buffer B in buffer A ( as above ) at a flow rate of 300 nl/min ( ∼1600 psi ) for 60 min , and CID fragmentation was induced on the 30 most abundant ions per full MS scan . All fragmented precursor ions were actively excluded from repeated selection for 15 s . Data were analysed using Peaks 7 ( Bioinformatics Solutions ) using a database containing all 24 , 092 Uniprot entries for the organism Gallus gallus combined with protein translations ( >8 amino acids ) of either all six reading frames of gallid herpesvirus 2 ( NCBI entry NC_002229 . 3; 10 , 026 entries ) or reticuloendotheliosis virus ( NC_006934 . 1; 412 entries ) . Results were filtered using a false discovery rate of 1% that was determined by parallel searching of a randomized decoy database . As previously described ( Koch et al . , 2007 ) , the extracellular sequence of mature BF22101 heavy chain and chicken β2m cloned in pET22b ( + ) were expressed separately as inclusion bodies in BL21 ( λDE3 ) pLysS Rosetta bacterial cells , solubilized in urea and assembled together with synthetic peptide ( synthesized commercially by fluorenyl-methoxy-carbonyl [fMOC] chemistry ) by dilution in a renaturation buffer . Assembled class I molecules were isolated after SEC using a HiLoad 26/60 Superdex 200 column ( GE Healthcare , UK ) . Similar procedures were used for BF10201 , BF20201 and BF21401 , except that a codon-optimized gene was synthesized ( GenScript ) for BF2*0201 , which was then cloned into a pET28a ( + ) vector followed by a factor X site for cleavage , a BirA site for biotinylation and a His tag for purification ( Leisner et al . , 2008 ) . As previously described ( Koch et al . , 2007 ) , heavy chains and β2m were expressed in bacteria as inclusion bodies and denatured in urea; β2m was refolded and purified by SEC . For each renaturation sample , solutions containing 30 µg heavy chain plus or minus 42 µg β2m and/or 10 µg synthetic peptide were added to refold buffer and incubated at 4°C with stirring for roughly 40 hr . After centrifugation and passage through a 0 . 45-µm sterile filter , each sample was loaded on a Superdex 200 10/300 GL SEC column ( GE Healthcare , UK ) as part of an AKTA 920 with 100 mM NaCl , 25 mM TrisCl , pH 8 . 0 running at 1 ml/min at room temperature . Peak fractions were collected and concentrated first using a pre-rinsed Amicon Ultra-4 10 kDa column and then a Vivaspin 500 10 kDa column to roughly 100 µl , buffer-exchanged into 50 mM NaCl , 10 mM Tris , pH 8 . 0 , concentrated to roughly 20 µl , and then transferred into a polypropylene microfuge tube . Acetic acid was added to 5% and the sample was concentrated to 2–4 µl using a SpeedVac at 40°C before analysis by MALDI-TOF ( Protein and Nucleic Acid Chemistry services , Department of Biochemistry , University of Cambridge ) . Recombinant MHC class I complexes were crystallized using the sitting-drop method ( see Table 3 for conditions ) . Crystals were flash frozen in liquid nitrogen and native data sets for each crystal were collected at 100K [Diamond Light Source , Harwell ( beamlines I02 , I03 , I04 or I04-1 ) , or the ESRF , Grenoble ( beamline ID29 ) ] . Data were processed using either AUTOPROC ( Vonrhein et al . , 2011 ) or XIA2 ( Winter , 2010 ) with XDS ( Kabsch , 2010 ) for integration and AIMLESS ( Evans , 2006 ) or SCALA ( Evans and McCoy , 2008 ) for scaling . All structures were solved by molecular replacement , as implemented in Phaser ( McCoy et al . , 2007 ) , part of the CCP4 software package ( Winn et al . , 2011 ) . Starting molecular replacement models were generated using CHAINSAW ( Stein , 2008 ) and the atomic co-ordinates of the chicken B21 MHC class I molecule ( PDBID: 3BEV ) with the peptide removed . Model building and refinement were carried out using COOT ( Emsley et al . , 2010 ) and AUTOBUSTER ( Bricogne et al . , 2011 ) or REFMAC ( Murshudov et al . , 1997 ) , with the heavy and light chains of the MHC molecule rebuilt first before the peptide was modelled into residual electron density ( see Table 1 for refinement statistics , Table 3 for data collection and refinement information , Table 4 for Ramachandran statistics ) . 10 . 7554/eLife . 05345 . 014Table 3 . Crystallization conditions , data collection , and refinement informationDOI: http://dx . doi . org/10 . 7554/eLife . 05345 . 014Structure ( PDB ID ) Crystallization conditionsCryo-protectantBeamlineData processingRefinement program4 d0b0 . 1M MMT buffer , pH 5 . 0 , 25% PEG 150015% ethylene glycolI02 ( Diamond Light Source ) AUTOPROC SUITE with XDS & SCALAAUTOBUSTER4d0c0 . 05M KH2PO4 , 20% PEG 800020% ethylene glycolI02 ( Diamond Light Source ) XIA2 with XDS & AIMLESSAUTOBUSTER2yez0 . 1M MMT buffer , pH 4 . 0 , 25% PEG 150015% ethylene glycolI03 ( Diamond Light Source ) AUTOPROC SUITE with XDS & SCALAAUTOBUSTER4cvz0 . 1M sodium acetate , pH 5 . 0 , 1 . 5 M ammonium sulphate8M sodium formateI04-1 ( Diamond Light Source ) XIA2 with XDS & AIMLESSAUTOBUSTER4cvx0 . 1M MMT buffer , pH 7 . 0 , 25% PEG 150015% ethylene glycolI04 ( Diamond Light Source ) XIA2 with XDS & AIMLESSREFMAC4 d0d0 . 1M MMT buffer , pH 7 . 0 , 25% PEG 150015% ethylene glycolI04 ( Diamond Light Source ) XIA2 with XDS & AIMLESSAUTOBUSTER4cw10 . 1M MIB buffer , pH 5 . 0 , 25% PEG 150015% ethylene glycolID29 ( ESRF ) XIA2 with XDS & AIMLESSAUTOBUSTER10 . 7554/eLife . 05345 . 015Table 4 . Ramachandran statisticsDOI: http://dx . doi . org/10 . 7554/eLife . 05345 . 015Structure ( PDB ID ) Ramachandran outliers number ( % ) Ramachandran favoured number ( % ) 4cvz0 ( 0% ) 371 ( 98 . 15% ) 4 d0b0 ( 0% ) 356 ( 94 . 43% ) 4d0c0 ( 0% ) 361 ( 95 . 76% ) 2yez1 ( 0 . 26% ) 364 ( 96 . 04% ) 4cvx0 ( 0% ) 710 ( 95 . 69% ) 4 d0d0 ( 0% ) 1430 ( 96 . 30% ) 4cw10 ( 0% ) 730 ( 98 . 25% )	Our immune system has the remarkable ability to recognize and destroy damaged cells or invading microbes while leaving the healthy cells in the body alone . Groups of proteins called ‘MHC class I molecules’ play important roles in defence against invading microbes . If a cell becomes infected with a virus or a bacterium , these molecules can recognize and bind to fragments of foreign or unusual proteins inside the cell , and display them on the surface of the cell . This allows immune cells to identify and kill the infected cells . Cells produce many different MHC class I molecules that are able to bind to different protein fragments . Some MHC molecules can bind to a wider variety of protein fragments than others , and the number of these molecules present on the cell surface can also vary between individuals . Researchers have noticed that individuals with particular MHC molecules tend to be more or less resistant to particular diseases . For instance , individuals with certain MHC molecules tend to take longer to develop AIDS if they become infected with HIV . However , it is not clear how either the variety of protein fragments bound or the numbers of MHC class I molecules on the surface of cells could alter the immune response . Here , Chappell , Meziane et al . studied MHC class I molecules in chickens and humans . The experiments reveal that the MHC class I molecules that can bind to a larger variety of protein fragments ( so-called ‘generalists’ ) are present in lower numbers on the surface of cells than molecules that can bind to a smaller variety of fragments ( so-called ‘specialists’ ) . Furthermore , generalist MHC molecules were found to provide resistance to Marek's disease in chickens—which causes paralysis—but some specialists slowed the progression of HIV infections into AIDS in humans . Chappell , Meziane et al . propose that these two types of MHC class I molecules evolved to perform different roles in immune responses . This is a new way of looking at the role of MHC molecules in fighting disease , and the next challenge is to explore the implications for medicine and evolutionary biology .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2015	Expression levels of MHC class I molecules are inversely correlated with promiscuity of peptide binding
The presumptive altered dynamics of transient molecular interactions in vivo contributing to neurodegenerative diseases have remained elusive . Here , using single-molecule localization microscopy , we show that disease-inducing Huntingtin ( mHtt ) protein fragments display three distinct dynamic states in living cells – 1 ) fast diffusion , 2 ) dynamic clustering and 3 ) stable aggregation . Large , stable aggregates of mHtt exclude chromatin and form 'sticky' decoy traps that impede target search processes of key regulators involved in neurological disorders . Functional domain mapping based on super-resolution imaging reveals an unexpected role of aromatic amino acids in promoting protein-mHtt aggregate interactions . Genome-wide expression analysis and numerical simulation experiments suggest mHtt aggregates reduce transcription factor target site sampling frequency and impair critical gene expression programs in striatal neurons . Together , our results provide insights into how mHtt dynamically forms aggregates and disrupts the finely-balanced gene control mechanisms in neuronal cells . Poly-glutamine ( PolyQ ) expansion in proteins is associated with multiple neuro- and muscle- degenerative diseases such as Huntington’s disease ( HD ) , X-linked spinobulbar muscular atrophy ( SBMA ) and various spinocerebellar ataxias ( SCA1 , 2 , 3 , 6 , 7 and 17 ) ( reviewed in [Blum et al . , 2013; Mohan et al . , 2014] ) . In HD , a single mutant Huntingtin allele with PolyQ tracts greater than 37 glutamines leads to selective cell death in the striatum and certain regions of the cortex , causing muscle coordination and cognitive defects ( Group , 1993; Ross and Tabrizi , 2011 ) . It has been widely observed that extended PolyQ tracts facilitate the formation of protein aggregates in the cytoplasm and nucleus of diseased cells ( Bates , 2003; DiFiglia et al . , 1997; Huang et al . , 2015 ) . Previous FRAP , FCS and in vitro super-resolution imaging provides significant insights into mHtt aggregate formation ( Cheng et al . , 2013; Duim et al . , 2014; Kim et al . , 2002; Park et al . , 2012; Sahl et al . , 2012; Wustner et al . , 2012 ) . However , the dynamics of aggregate formation or how the resulting 'plaques' might influence essential molecular transactions that disrupt gene expression programs have not been investigated at the single-molecule level in living cells . Since the original discovery of mHtt aggregates in the nucleus and cytoplasm of HD cells , the relevance of these aggregates or plaques to disease pathology has been under vigorous debate ( DiFiglia et al . , 1997; Scherzinger et al . , 1997; Woerner et al . , 2016 ) . Currently , several mechanisms have been proposed to explain how mHtt aggregates might contribute to disease states . Interestingly , it was shown that the formation of PolyQ aggregates can in some instances , protect cells from apoptosis in short-term cell culture experiments ( Saudou et al . , 1998; Taylor et al . , 2003 ) . Specifically , it was proposed that soluble fragments or oligomers of mHtt are more toxic than mHtt aggregates . Stable self-aggregation of mHtt monomers was postulated to neutralize prion protein interacting surfaces and protect cells from prion induced damage ( Arrasate et al . , 2004; Saudou et al . , 1998; Slow et al . , 2005 ) . However , this model does not address the long-term effect of mhtt aggregates in striatal cells nor does it exonerate mHtt aggregates from potentially contributing to the disease state . For example , myriad studies have reported the toxicity of aggregates in vivo ( Labbadia and Morimoto , 2013; Michalik and Van Broeckhoven , 2003; Williams and Paulson , 2008; Woerner et al . , 2016 ) . Without methods to directly observe and measure biochemical reactions and molecular interactions in living cells , it is challenging to gain mechanistic insights that may help resolve these controversies . With recent advances in imaging and chemical dye development ( reviewed in [Liu et al . , 2015] ) , it has become possible to detect and track individual protein molecules in single living cells ( Abrahamsson et al . , 2013; Chen et al . , 2014a , 2014b; Elf et al . , 2007; Gebhardt et al . , 2013; Grimm et al . , 2015; Hager et al . , 2009; Izeddin et al . , 2014; Liu et al . , 2014; Mazza et al . , 2012; Mueller et al . , 2013 ) . Decoding the complex behavior of single molecules enables us to measure molecular kinetics at a fundamental level that is often obscured in ensemble experiments . Specifically , the rapidly emerging high-resolution fast image acquisition platforms provide a means for visualizing and measuring the in vivo behavior of dynamically regulated molecular binding events . It also becomes possible to generate 3D molecular interaction maps in living mammalian cells and elucidate local diffusion patterns in the highly heterogeneous sub-cellular environment ( Chen et al . , 2014a , 2014b; Izeddin et al . , 2014; Liu et al . , 2014 ) . Here , using HD as the model , we devised a molecular imaging system to quantify the formation of protein structures and measure the real-time dynamics and behavior of PolyQ-rich proteins . First , with live-cell PALM and FRAP experiments , we compared gross structures and diffusion dynamics of wild-type ( Htt-25Q ) versus disease-inducing mutant ( mHtt-94Q ) Htt protein fragments . Interestingly , soluble fractions of wild-type Htt-25Q and mutant Htt-94Q display similar rapid diffusion kinetics . Strikingly , both Htt-25Q and mHtt-94Q protein fragments also form small , diffraction-limited clusters in live cells . These clusters are highly dynamic and resolve quickly ( mean lifetime < 10~20 s ) . However , the mutant Htt-94Q protein also forms much larger highly stable aggregates ( FRAP recovery lifetime > 45 min ) . Two-color super resolution imaging reveals that mHtt aggregates exclude chromatin and selectively interact with a set of neurological disease-related factors ( Foxp2 , TBP , Sp1 and wild-type Huntingtin ) . Fine domain mapping experiments suggest that continuous PolyQ tracts contribute to but are not necessary for binding to mHtt aggregates . Notably , we found that aromatic amino acids enabled proteins with sparse glutamines to bind mHtt aggregates . Single-molecule tracking and numerical simulation experiments suggest that mHtt aggregates act as large decoy traps in the cell , efficiently slowing down target search kinetics and decreasing target site sampling frequencies . Consistent with this model , unbiased genomic screens showed that elevating Sp1 levels in HD-affected striatal cells partially restored the expression of Sp1 dependent target genes . Sp1 target site occupancies are also significantly decreased in HD-affected cells . These findings provide new insights into potential Huntington’s disease mechanisms and reveal the role of mHtt aggregates in disrupting key dynamic biological processes in living cells . To establish a cellular imaging system for HD and evaluate the effects of PolyQ expansion on Htt protein dynamics , we fused disease-inducing Huntingtin protein exon 1 fragment ( mHtt-94Q ) ( Rubinsztein , 2002; Yamamoto et al . , 2000 ) and its wild-type counterpart ( Htt-25Q ) ( Krobitsch and Lindquist , 2000 ) to a monomeric photo-switchable protein ( mEOS3 . 2 ( Zhang et al . , 2012 ) ; w/o nuclear localization signal - NLS ) . We stably expressed each fragment in mouse embryonic stem ( ES ) cells or STHdh striatal cells ( Trettel et al . , 2000 ) . Epi-fluorescence live imaging revealed that 8~15% of the mHtt-94Q cells ( 27 of 234 cells ) contain large visible sub-cellular aggregates while the majority ( 123 of 125 cells ) of Htt-25Q cells show a relatively homogeneous intracellular distribution ( Figure 1—figure supplement 1A ) . To measure Htt protein dynamics , we next performed live-cell imaging experiments by stochastically switching the mEOS3 . 2 moiety to the red-shifted state and tracking single-molecule movement of the activated molecules ( sptPALM ) ( Manley et al . , 2008 ) . This technique allowed us to generate high-density single-molecule trajectories separated by stochastic activation times . As expected , ensemble whole-cell diffusion and velocity analysis reveals that a much larger fraction of mHtt-94Q molecules show slower and more compact diffusion kinetics compared to Htt-25Q controls ( Figure 1B , Figure 1—figure supplement 1B and C ) , suggesting that mHtt-94Q is on average significantly less mobile . High-density trajectories generated with sptPALM also allows reconstruction of high-resolution sub-cellular diffusion maps ( Figure 1A , middle and bottom panels ) . Interestingly , both mean square displacement ( MSD ) and Bayesian inference-based diffusion maps show that the slow mHtt-94Q species were mainly concentrated around aggregates , while mHtt-94Q molecules in other regions remain fast moving . Consistent with these observations , fluorescence recovery after photo-bleaching ( FRAP ) experiments within non-aggregate containing regions show full recoveries within 1 s ( Figure 1—figure supplement 1E , Video 1 and 2 ) , confirming that a portion of mHtt-94Q molecules can remain in a mobile , dynamic , and soluble state . These results are consistent with previous FRAP and FCS experiments showing fast diffusing and aggregate states of mHtt protein ( Cheng et al . , 2013; Kim et al . , 2002; Park et al . , 2012; Wustner et al . , 2012 ) . 10 . 7554/eLife . 17056 . 003Figure 1 . mHtt protein displays three distinct states in living cells . ( A ) Top , localization map of live-cell PALM data set for the indicated Htt fragment . The PALM data were recorded at 20 ms per frame using imaging conditions specified in Materials and methods . Diffraction-limited Htt-25Q aggregates are indicated by square boxes ( White ) . Middle , reconstructed single-molecule trajectories reflecting sub-cellular molecular diffusion . Each trajectory is color-colored using the diffusion coefficient calculated by the linear regression of the MSD curve with a R2 > 0 . 9 . 2000 trajectories are shown for the Htt-25Q condition . Twice as many trajectories are shown for the mHtt-94Q condition to ensure relatively equal sampling and representation in regions without mHtt-94Q aggregates . Bottom , diffusion map calculated by a Bayesian inference- based method ( El Beheiry et al . , 2015 ) . Trajectories with minimal 2 connected localizations were considered in this analysis . Each pixel in the image is color-coded with the most probable diffusion coefficient for that pixel calculated by InferenceMap . Scale bars , 2 µm . ( B ) Histogram of diffusion coefficients calculated same as in the middle panel of ( A ) for the indicated condition . Htt-25Q , 7266 trajectories; mHtt-94Q , 19 , 975 trajectories . N = 6 cells . ( C ) The temporal history of localizations in each selected region: vertical line ( Black ) indicates the frame from which the localization is detected . The blue trend line represents the cumulative distribution function of the total localizations detected before a given frame number . Three representative regions are shown for each condition . The images above each plot are sliding-window localization maps from Video 3 , showing the temporal dynamics of Htt protein from each category . t indicates when the cluster/aggregate is clearly visible . Scale bars , 2 µmDOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 00310 . 7554/eLife . 17056 . 004Figure 1—figure supplement 1 . Dynamics of Htt protein in live cells . ( A ) Representative Epi-illumination Images for Htt-25Q-mEOS3 . 2-NLS and mHtt-94Q-mEOD3 . 2-NLS containing cells . ( B ) Density histogram of single-molecule translocation distances between two adjacent imaging frames ( For calculation , see details in Materials and methods ) . 60 , 702 ( Htt-25Q , yellow ) and 149 , 970 ( mHtt-94Q , blue ) translocations were considered in this analysis . N = 6 cells . ( C ) Rose histogram of three-point jumping angles for the indicated Htt fragment ( See details in Materials and methods ) . For normalization purposes , 30 , 000 randomly selected angles are binned and showed for each condition . N = 6 cells . ( D ) Left , Maximal intensity projection image for Video 4 showing final static localization maps for each type of regions indicated on the left . Right , the temporal history of Htt-25Q localizations in control regions with no visible clusters ( presented same as in Figure 1C ) . ( E ) FRAP analysis on mHtt-94Q containing cells without visible clusters or aggregates . After initial focused bleaching at the Region of Interest ( white arrow ) , confocal images were acquired at 2 fps for 40 s ( Video 1 , Left ) . Average gray-scale intensity recovery curve from 12 measurements is plotted as the function of time . Error bars represent Standard Deviation . ( F ) FRAP analysis on mHtt-94Q aggregates . After initial focused bleaching at the Region of Interest ( white arrow ) , time-lapse confocal images were acquired with 8 s intervals for 400 s ( Video 1 , Right ) . Average gray-scale intensity recovery curve from 9 measurements is plotted as the function of time . Error bars represent Standard Deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 00410 . 7554/eLife . 17056 . 005Video 1 . FRAP experiments in mHtt aggregate negative ( left ) and positive ( right ) cells . The time lapse confocal imaging interval for the aggregate negative cell is 0 . 5 s . That for the aggregate positive cell is 8 s . DOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 00510 . 7554/eLife . 17056 . 006Video 2 . FRAP experiments in mHtt aggregate negative ( left ) and positive ( right ) cells performed in the same field of view , showing that laser power is sufficient for bleaching . The fluorescence recovery in cells with soluble mHtt-94Q is rapid , while that at aggregate regions is much slower . DOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 006 Interestingly , reconstructed molecular density maps from single molecule localization experiments reveal that even Htt-25Q fragments can form small protein polymers or clusters in live cells ( Figure 1A , white box ) . These very small clusters are below the diffraction-limit ( 50~100 nm ) and cannot be adequately resolved by conventional imaging methods . Temporal localization history and sliding window analysis ( Figure 1C and Figure 1—figure supplement 1D; Videos 3 and 4 ) suggest that protein molecules in these clusters display relatively fast association/dissociation dynamics . Specifically , time-counting PALM analysis ( tcPALM ) ( Cisse et al . , 2013 ) reveal that these small clusters form in temporal bursts ( ~10–20 s ) ( Figure 1C ) , while localization detections in control regions with similar molecular densities show no such apparent bursting behaviors ( Figure 1—figure supplement 1D ) . Localization maps for mHtt-94Q cells are more complex . Both small clusters and large aggregates can be detected ( Figure 1A and Video 3 ) . Interestingly , localization detections in small mHtt-94Q clusters display similar bursting kinetics as Htt-25Q clusters ( Figure 1C and Video 4 ) . By contrast , the localizations of large mHtt-94Q aggregates distribute relatively evenly over time , suggesting a much greater structural stability ( Figure 1C; Video 3 and 4 ) . Consistent with this observation , FRAP experiments over longer time-scales found little molecular exchange occurring at large mHtt-94Q aggregates ( FRAP recovery lifetime > 45 min , See Materials and methods for calculation details ) ( Figure 1—figure supplement 1F , Video 1 and Video 2 ) . Taken together , our results suggest that there are likely at least three distinct states for Htt proteins in living cells: 1 ) rapidly diffusing , 2 ) short-lived dynamic clustering and 3 ) very stable aggregates formed only by the Gln-expanded mutant protein but not the wild-type Htt . It is likely that PolyQ expansion tilts the kinetic balance of mHtt from a dynamic clustering state to an aggregated state , akin to recently described protein dynamic polymer/phase transitions associated with ALS/FTD neurodegenerative disorders ( Kato et al . , 2012; Molliex et al . , 2015; Murakami et al . , 2015; Nott et al . , 2015; Patel et al . , 2015; Xiang et al . , 2015 ) . These results also support previous in vitro super-resolution imaging experiment showing that there are likely distinct Htt aggregate species for seeding and polymerization states ( Duim et al . , 2014; Sahl et al . , 2015; 2012 ) . 10 . 7554/eLife . 17056 . 007Video 3 . Sliding-window representation for live-cell PALM datasets of Htt-25Q ( Left ) and mHtt-94Q ( Right ) . PALM Imaging is performed with a rate of 50 Hz . In each frame , single-molecule localizations from adjacent 1000 frames were pooled and plotted . The interval for the sliding window is 0 . 5 s ( 25 frames ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 00710 . 7554/eLife . 17056 . 008Video 4 . Zoom-In view of indicated types ( Left ) of region from Video 3 , showing distinct local molecular dynamics . DOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 008 Dysregulation of transcription has been extensively implicated in the pathological processes of HD for decades ( Dunah et al . , 2002; Kumar et al . , 2014; Mohan et al . , 2014; Nucifora et al . , 2001; Paul et al . , 2014 ) . Although altered chromatin structures have been reported in diseased cells ( Labbadia et al . , 2011 ) , ultra-fine chromatin structure and genome organization have not been extensively investigated . To address this , we labeled chromatin by stably expressing H2B-HaloTag in the mHtt-94Q cells . To avoid biases introduced by individual imaging techniques , we imaged the samples with two-color 3D live-cell structured illumination imaging , Airyscan imaging and single-molecule localization microscopy . All these experiments showed consistent results that intensities of H2B staining were dramatically depleted in regions of mHtt aggregates ( Figure 2A and Figure 2—figure supplement 1A; Video 5 ) , suggesting that these 'inclusion bodies' likely exclude chromatin . We note here that the size and shape of mHtt aggregates varies significantly in the cell . Thus , to quantify exclusion or recruitment by mHtt aggregates in an unbiased manner , we used a size-normalized , averaging based analysis method . Briefly , we rescale each mHtt-aggregate containing region to have a diameter of 100 pixels ( Figure 2B ) . Then , an intensity map was generated by averaging multiple mHtt aggregate regions obtained across different cells . The center-to-peripheral radial grayscale intensity curve is then calculated by each circular pixel increment for both channels ( Figure 2C and D ) . For molecular structures ( such as H2B ) that are excluded by mHtt aggregates , the intensity curve resembles the mirror-image of the curve for for mHtt ( Figure 2D , Figure 2—figure supplement 1B and C ) . We note that the commonly-used DNA dye DAPI non-specifically stains cytoplasmic mHtt aggregates ( Figure 2—figure supplement 1D ) , suggesting that DAPI fluorescence might not be an ideal indicator for genome localization and organization in this case . 10 . 7554/eLife . 17056 . 009Figure 2 . mHtt aggregates exclude chromatin . ( A ) Top , live-cell 3D SIM imaging of chromatin ( H2B-HaloTag-JF549 , red ) and mHtt aggregates ( mHtt-94Q-CFP , Green ) . See Video 5 for 3D rendering . Bottom , Airyscan imaging of fixed cells with the same two-color labeling . ( B ) Size-normalized ( 100 pixel diameter ) fluorescence intensity maps for 17 mHtt aggregate containing regions as circled in ( A ) . The intensity maps for corresponding regions in the H2B channel are represented in the bottom panel . ( C ) Intensity maps for different regions shown in ( B ) are summed and averaged for each channel . The mean radial intensity trend line for each channel is calculated as the function of the radius from the center ( C ) to the peripheral ( P ) . ( D ) Radial intensities are plotted as the function of the radius for mHtt-94Q ( Top ) and H2B ( Bottom ) channel . N , 17 regions from 8 cells . Scale bars , 2 µmDOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 00910 . 7554/eLife . 17056 . 010Figure 2—figure supplement 1 . Map spatial relationship between chromatin and mHtt aggregates . ( A ) live-cell H2B density map generated from single-molecule localizations acquired with the imaging condition specified in Materials and methods . 145 , 976 localizations were utilized for the reconstruction . The localization map ( left ) is overlaid with the mHtt-94Q-CFP-NLS aggregate channel ( Middle ) , producing the merged image ( right ) . ( B ) Size-normalized intensity maps for different mHtt aggregate regions for live-cell 3D-SIM imaging experiments . N , 11 regions from 6 cells . ( C ) Radial intensities are plotted as the function of the radius for mHtt-94Q ( Top ) and H2B ( Bottom ) channel . ( D ) Enrichment of DAPI staining signals at cytoplasmic mHtt aggregates . Scale bars , 2 µmDOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 01010 . 7554/eLife . 17056 . 011Video 5 . Live-cell 3D SIM imaging with ES cells expressing H2B-HaloTag ( JF549 labeling , Red ) and mHtt-94Q-CFP ( Green ) . In the video , 2D slices in the axial direction are looped 3 times followed by a rotational 3D volume rendering . DOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 011 One intriguing pathological feature of HD is selective cell death in the striatum and certain regions of the cortex ( Ferrante et al . , 1985; Rikani et al . , 2014 ) . Since Htt is widely expressed in different tissues in the body , it remains largely unclear how mutant Htt causes selective cell death in striatal neuron populations ( Strong et al . , 1993 ) . Interestingly , it was shown that long PolyQ tract containing proteins such as wild-type Htt and TATA box binding protein ( TBP ) are trapped in the mHtt aggregates ( Kim et al . , 2002; Lee et al . , 2004 ) . To test whether we could recapitulate these results in our live cell imaging system , we expressed WT Htt-25Q-HaloTag exon 1 fragment in cells with mHtt aggregates . Indeed , Htt-25Q protein becomes significantly sequestered in the mHtt aggregates in both ES cells and STH striatal cells ( Figure 3A and Figure 3—figure supplement 1B ) . Next , we deleted the 25Q PolyQ tract in the wild-type Htt fragment and found no measurable recruitment of the fragment to mHtt aggregates ( Figure 3B ) , suggesting that the 25Q PolyQ tract is likely required for the interaction . As to TBP , we found that the N-terminal PolyQ tract containing domain but not the C-terminal DNA binding domain of TBP is selectively enriched in mHtt aggregates ( Figure 3A–D ) , suggesting that recruitment of TBP to mHtt aggregates also requires the PolyQ containing domain . Importantly , these results are consistent with previous studies using different cell lines ( Kim et al . , 2002; Lee et al . , 2004 ) , suggesting that mechanisms mediating these interactions are likely not cell-type specific . 10 . 7554/eLife . 17056 . 012Figure 3 . mHtt aggregates non-specifically interact with Htt-25Q , Foxp2 , and TBP via PolyQ containing domain . ( A ) Airyscan images showing enrichment of Htt-25Q-HaloTag , HaloTag-Foxp2 and HaloTag-TBP-TAD domain ( labeled with JF646 ) at mHtt aggregate regions ( mHtt-94Q-mEOS3 . 2-NLS ) . ( B ) Deletion of 25Q in Htt-25Q or PolyQ enriched TAD in Foxp1 and TBP abolishes the recruitment of these factors to mHtt aggregates . ( C ) And ( D ) Radial intensity curves as plotted in Figure 2D for mHtt-94Q ( Top ) and TBP-TAD ( C ) or DBD ( D ) ( Bottom ) channel . N , 9 regions from 5 cells . Scale bars , 2 µmDOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 01210 . 7554/eLife . 17056 . 013Figure 3—figure supplement 1 . PolyQ tracts are important for aggregate binding . ( A ) Protein sequence information for PolyQ enriched domains in wild-type Htt , TBP and Foxp2 . Q is highlighted in green . ( B ) Airyscan Images showing specific enrichment of Foxp2-TAD ( in ES cells ) and Htt-25Q ( in STHdh cells ) to mHtt aggregates . Scale bars , 2 µmDOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 013 To identify new mHtt aggregate interacting factors and to test whether the presence of a PolyQ tract is a reliable predictor for recruitment by mHtt aggregates , we next screened PolyQ containing proteins genome-wide and identified a PolyQ tract containing transcription factor , Foxp2 . ( Figure 3—figure supplement 1 ) . Most interestingly , Foxp2 is a striatum-specific transcription factor shown to be important for speech evolution , motor learning and neural system development ( Enard et al . , 2002; Groszer et al . , 2008; Lai et al . , 2001; Shu et al . , 2005; Takahashi et al . , 2003 ) . Disruption of both copies of the Foxp2 gene caused severe motor impairment in mice ( Shu et al . , 2005 ) . Intriguingly , in humanized Foxp2 mice , the medium spiny neurons in the striatum display increased dendrite lengths and increased synaptic plasticity , suggesting that Foxp2 is important for the development of cortico-basal ganglia circuits ( Enard et al . , 2009 ) . These neural circuits are specifically affected in HD ( Albin et al . , 1989; Ferrante et al . , 1985; Ross and Tabrizi , 2011; Slow et al . , 2003 ) . We found that for the full length Foxp2 , the N-terminal PolyQ containing fragment but not the C-terminal DNA binding domain strongly interacts with mHtt aggregates ( Figure 3A and B , Figure 3—figure supplement 1B ) . Taken together , these results suggest that recruitment of these factors to mHtt aggregates likely involve PolyQ containing domains and thus , cellular functions mediated by these important transcription factors are also likely affected by the presence of mHtt aggregates . Because of the critical role of Foxp2 in HD-affected brain region and function , our findings also pose an intriguing possibility that Foxp2 might be one of the essential effectors responsible for the selective effects of HD on striatal neural circuits ( See DISCUSSION for details ) . Although our recent unbiased screen for long continuous PolyQ tract containing transcription factors did not identify the sparsely Q-rich human TF Sp1 , multiple studies had previously implicated the importance of this regulatory protein in HD . Interestingly , early studies showed that over expression of Sp1 and TAF4 ( one of its co-activators ) can partly rescue mutant Htt induced gene expression defects ( Dunah et al . , 2002 ) , and polyQ mHtt fragments inhibit Sp1-mediated transcriptional activation in vitro ( Zhai et al . , 2005 ) . These results suggested that Sp1 might also interact with mHtt protein aggregates in living cells . To test this directly , we performed live-cell 3D Structured Illumination imaging experiments . Interestingly , we observed substantial Sp1 binding even to cytoplasmic mHtt aggregates ( Figure 4—figure supplement 1A ) . Live-cell single-molecule tracking revealed that Sp1 directly binds to mHtt aggregates ( Figure 4A , B and Video 6 ) and that full length Sp1 is selectively enriched in mHtt aggregate regions ( Figure 4C–E , and Figure 4—figure supplement 1B ) . These results are somewhat surprising , because although Sp1 is enriched for Q’s , there are no obvious continuous PolyQ tracts in the Sp1 protein sequence . To probe the amino acid sequences of Sp1 responsible for this interaction , we first separated the N-terminal transcription activation domain ( TAD ) ( 1–615aa ) from the C-terminal DNA binding domain ( DBD ) ( 616–781aa ) . The Sp1 N-terminal TAD showed strong interactions with mHtt aggregates while the C-terminal DBD was excluded from mHtt aggregates ( Figure 4C and D ) . To further narrow down the interaction domain , we performed a series of truncation experiments ( Figure 5—figure supplement 1 ) and identified a 50aa Sp1 fragment ( 166–215 ) with 14Q’s that retained the interaction with mHtt ( Figure 5A ) . Tethering this fragment to HaloTag allowed us to monitor its selective recruitment to mHtt aggregates ( Figure 5B and D ) . Surprisingly , a control fragment from Sp1 ( 363 - 412aa ) also enriched for glutamines ( 18Q’s ) failed to interact with mHtt aggregates ( Figure 5C and D ) . Sequence gazing revealed that the only apparent difference between these two Q-rich fragments is the presence of three aromatic amino acids interspersed among the 14Qs of Sp1 ( 166–215aa ) ( Figure 5A ) . Mutation of these three aromatic amino acids ( Y , F , F ) to A or even Q severely impaired the ability of the fragment to interact with mHtt aggregates ( Figure 5B ) . More interestingly , converting three amino acids in the non-interacting Q-rich Sp1 ( 363 - 412aa ) fragment to phenylalanine ( F ) animated this fragment to bind mHtt aggregates ( Figure 5C and D ) . These results suggest that interspersed aromatic amino acids are likely critical for Sp1 to interact with mHtt aggregates while stretches of PolyQ rich regions are not sufficient and long contiguous PolyQ tracts may not be necessary for mHtt aggregate interactions . Our results indicate that mHtt aggregates display a remarkable degree of sequence or structural specificity as many protein fragments , even some with Q-rich patches fail to interact and become trapped by the large rafts of mHtt protein , consistent with previous studies ( Rajan et al . , 2001 ) . Instead , only certain proteins with specific Q-rich properties such as the presence of interspersed aromatic moieties were found to be retained in the live cell context . Nevertheless , we expect that more complicated and perhaps even some more promiscuous protein features may govern protein:mHtt aggregate interactions . We also anticipate that a broader spectrum of proteins than we have sampled can become trapped and thus affected by mHtt aggregates in different cell-types or cellular contexts . Indeed , previous studies suggest that other transcription factors without long continuous PolyQ stretches such as CBP and TAF4 are likely also sequestered in mHtt aggregates ( Dunah et al . , 2002; Nucifora et al . , 2001 ) . 10 . 7554/eLife . 17056 . 014Figure 4 . Single-molecule imaging of Sp1 binding to mHtt aggregates . ( A ) Single-molecule imaging of cytoplasmic Sp1-mHtt aggregate interaction Top , Epi-fluorescence images of JF549-HaloTag-Sp1 ( red ) and mHtt-94Q-CFP ( green ) before single-molecule imaging Middle , 2D molecular interaction map of Sp1 binding to mHtt aggregates . Two-point translocations are color-coded according to the jumping distance shown in ( B ) . 7399 translocations are included in this analysis . See Video 6 for single molecule tracks overlaid on raw data . A total of 215 ( trajectories ) Sp1 binding events were detected on the surface of the aggregate . Bottom , Diffusion map generated by InferenceMap as shown in Figure 1A . Each pixel in the image is color-coded with the most probable diffusion coefficient for that pixel . ( B ) Histogram for jumping distances shown in the middle panel of ( A ) . ( C ) Two-color Airyscan images of ES cells expressing indicated Sp1 fragment ( red , JF646 ) and mHtt-94Q-mEOS3 . 2-NLS ( green ) . ( D ) Zoom-in view of mHtt aggregate regions boxed in ( C ) shows that Sp1-TAD is recruited to mHtt aggregates while Sp1-DBD is excluded . Radial intensity analysis of HaloTag-Sp1 , HaloTag-DBD and HaloTag-TAD ( Imaging data are shown in Figure 4—figure supplement 1B ) shows specific enrichments of Sp1 and Sp1-TAD but depletion of Sp1-DBD at mHtt aggregate regions . Scale bars , 2 µmDOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 01410 . 7554/eLife . 17056 . 015Figure 4—figure supplement 1 . Sp1 non-specifically interacts with mHtt aggregates in live cells . ( A ) Live-cell SIM imaging shows nonspecific association of Sp1 ( red ) to the mHtt aggregates ( green ) in the cytoplasm . Zoom-In view of the boxed region is on the right . ( B ) Size-normalized ( 100 pixel diameter ) fluorescence intensity maps for Sp1 , Sp1-TAD and Sp1 DBD at mHtt aggregate regions ( Imaging data for Figure 4E ) . For each condition , the intensity maps for identical mHtt aggregate regions are on the top . Average intensity maps for each channel are shown in the left . N for Sp1; 20 regions in 12 cells . N for Sp1-DBD; 18 regions in 10 cells . N for Sp1-TAD; 18 regions in 8 cells . Scale bars , 2 µmDOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 01510 . 7554/eLife . 17056 . 016Video 6 . First segment: Single-molecule trajectories superimposed on Htt-aggregate channel . Trajectories are randomly color-coded . Second segment: Diffusion map calculated by InferenceMap superimposed on Htt-aggregate channel . Third segment: Reconstructed single-molecule trajectories of Sp1 overlaid on raw imaging data with dragon presentation ( 20 step delayed ) . Presentation software: InferenceMapDOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 01610 . 7554/eLife . 17056 . 017Figure 5 . Aromatic acids are critical for Sp1 binding to mHtt aggregate . ( A ) Protein sequence information for Sp1 fragments shown in ( B ) . The number of glutamines in each sequence is in the parenthesis on the left . The amino acids mutated in the panel ( B ) are highlighted by black triangles below . ( B ) And ( C ) – upper panel , Airyscan images show that out of three fragments list in panel ( A ) , only Sp1 166-215aa ( 14Q ) fragment is specifically recruited to mHtt aggregates . Mutation of aromatic amino acids ( Y , F ) to A or even Q abolishes the interaction . ( C ) – lower panel , Converting 3 amino acids ( N , S , T ) in Sp1 363-412aa ( 18Q ) fragment to F enables this fragment to bind mHtt aggregates . ( D ) Co-localization dot plot for results shown in ( B ) and ( C ) . Each dot denotes one cell . The percentage of mHtt aggregates displaying enrichment for the indicated Sp1 fragment ( X-axis ) in single cells is calculated and plotted on the Y-axis . Scale bars , 2 µmDOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 01710 . 7554/eLife . 17056 . 018Figure 5—figure supplement 1 . Mapping atypical Sp1:mHtt aggregate interaction domain . Aryscan images showing that the Sp1:mHtt aggregate binding activities reside in the 166-215aa ( 14Q ) fragment . Truncated Sp1 fragments ( N-terminal HaloTag fusion , JF646 labeling , red ) , mHtt aggregates ( mHtt-94Q-mEOS3 . 2 , green ) . Left shows the schematics of the Sp1 fragment used . DOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 018 Complex multi-protein biochemical reactions that drive gene expression and molecular trafficking involve target search processes that are tightly regulated by modulating specific versus non-specific macromolecular interactions within the cell . Previously , we demonstrated that key transcription factor ( TF ) target search processes in the cell are dominated by non-specific highly transient transactions with chromatin ( Chen et al . , 2014b ) . One model for HD to affect gene regulation is to imagine that mHtt aggregates form large molecular traps in certain cells . Such a scenario would dramatically increase non-specific trial-and-error sampling by TFs before they reach cognate target sites . To quantitatively evaluate how various parameters ( volume , number and size ) of mHtt aggregates might affect the target search process , we next performed numerical simulation experiments ( See Materials and methods for details ) . Briefly , we inject a TF into a random position in the nucleus in silico . We assumed that TFs navigate the nucleus via 3D random walk and that the nucleus contains both bona fide target sites and mHtt aggregates of various sizes and numbers ( Figure 6—figure supplement 1F and Video 9 , See Details in Materials and methods ) . In the simulation , TF movement is slowed down within the mHtt aggregates as suggested by our SPT experiments ( Figures 4 and 6 ) . Next , we recorded the average search time for the TF to reach a target site for the first time and found that the search time inversely correlated with the volume ratio of aggregates to the nucleus ( VR ) ( Figure 6—figure supplement 1C and D ) . Strikingly , merely adding aggregates equivalent to ~5% of the total nuclear volume slows down the TF target search process by ~20 fold ( Figure 6—figure supplement 1D ) . Interestingly , with a fixed total volume , smaller aggregates more efficiently slow down the target search process , suggesting that merging small aggregates to larger ones might reduce the detrimental effects on target search ( Figure 6—figure supplement 1E ) , consistent with the previous report on the protective effects of larger aggregates in cells ( Saudou et al . , 1998; Taylor et al . , 2003 ) . To determine whether the target search process of transcription factors and wild-type Htt protein are actually influenced by mHtt aggregates in live cells , we perform SPT experiments on Sp1 , Foxp2 and Wt Htt protein . For these three factors , we were able to directly visualize the trapping effects of mHtt aggregates at the single molecule level ( Figures 4 , 6 , Figure 6—figure supplement 1 and Figure 7A; Video 6 , 7 and 8 ) . Specifically , diffusion coefficients of these factors become dramatically reduced within mHtt aggregates ( Figure 6A , B , Figure 6—figure supplement 1A and B ) , indicating binding and longer residence times in these areas . The fraction of slowly diffusing molecules of Sp1 , Foxp2 and Wt Htt all become substantially elevated in cells containing mHtt aggregates compared to controls ( Figures 6C , D and 7A ) , consistent with an increased number of non-productive binding events due to greater numbers of decoy non-specific binding sites in the cell . Thus , these direct SMT results support the model that mHtt aggregates form large sticky traps , waylaying the target search process and reducing the cognate target site sampling frequency ( Equations 1–7 ) . 10 . 7554/eLife . 17056 . 019Figure 6 . mHtt aggregates effectively slow down target search processes in living cells . ( A ) Single-molecule tracking of Foxp2 in mHtt aggregate positive ( upper ) and negative cells ( lower ) . The resulting single-molecule trajectories are color-coded using the diffusion coefficient calculated by the linear regression of the MSD curve with a R2 > 0 . 9 . 6000 trajectories are shown for the Htt aggregate positive cells ( upper ) . 2000 trajectories for the mHtt aggregate negative cell ( lower ) . The epi-fluorescence images ( red , JF646-HaloTag-Foxp2; green , mHtt-94Q-mEOS3 . 2-NLS ) before single molecule imaging are on the left . ( B ) Single-molecule tracking of Htt-25Q in mHtt aggregate containing cells . The color-coding scheme is the same as in ( A ) . 2000 trajectories are shown . The epi-fluorescence images ( red , JF549-Htt-25Q-HaloTag; green , mHtt-94Q-CFP ) before single molecule imaging are on the left . Trajectories are further divided to ‘In-aggregate’ and ‘out-of-aggregate’ fractions as shown in Figure 6—figure supplement 1A and B for downstream diffusion analysis in ( D ) . mHtt aggregate region is indicated by the box with dotted line . The nucleus of the cell is labeled with the yellow contour line . See Video 7 for reconstructed trajectories overlaid on imaging data . ( C ) and ( D ) Histograms of diffusion coefficients for the indicated condition shown in ( A ) and ( B ) . Foxp2 in aggregate positive cells , 11 , 667 trajectories ( 6 cells ) ; Foxp2 in aggregate negative cells , 19 , 975 trajectories ( 8 cells ) . Htt-25Q at aggregation regions , 5394 tracks ( 6 cells ) ; Htt-25Q at non-aggregation regions , 48 , 630 tracks ( 6 cells ) . Scale bars , 2 µmDOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 01910 . 7554/eLife . 17056 . 020Figure 6—figure supplement 1 . mHtt aggregates clog target search . ( A ) Trajectories shown in Figure 6B are computationally divided into two populations based on the physical proximity of the trajectory to the mHtt aggregate region . Trajectories contain no localization events in the mHtt aggregate region are shown on the left . Trajectories fully or partially included in the mHtt aggregate region are shown on the right with a Zoom-in view . ( B ) A second example of single-molecule tracking of Htt-25Q in mHtt aggregate containing cells . The representation is the same as in Figure 6B and in the panel ( A ) of this figure . Scale bar , 2 µm ( C ) , ( D ) and ( E ) . 3D Target search simulation in mHtt aggregate containing cells . See Video 9 for schematic illustration of the simulation . Detailed simulation parameters are in Materials and methods . Briefly , TF is injected into the cell at a random position . TF navigates through the cells via Brownian diffusion . The diffusion is slowed down 100 fold in spaces occupied by mHtt aggregate . The average first passage time in each condition is normalized to that in the cell without mHtt aggregates as the Fold of Delay . In the panel C , the diameter ( D ) of aggregates is fixed to 20 spatial units with varying Number ( N ) of aggregates . In the panel D , N is fixed to 100 with varying D . In the panel E , the total volume of aggregate in the cell is fixed to 5 . 1% volume ratio ( VR , aggregate volume/the total cell volume ) with varying N and D . ( F ) Images showing cells with different N , D and VR . DOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 02010 . 7554/eLife . 17056 . 021Figure 7 . Elevated Sp1 levels rescue a subset of Htt-111Q dose-dependent down-regulated genes in cultured striatal cells . ( A ) mHtt aggregates slow down Sp1 diffusion in the nucleus . Histogram ( Red ) of diffusion coefficients for Sp1 trajectories ( 28 , 548 ) in mHtt aggregate containing cells ( N = 12 cells ) and that for Sp1 trajectories ( 20 , 855 ) in Htt-25Q control cells ( N = 12 cells ) . In this experiment , HaloTag-Sp1 is labeled with PA-JF646 ( See Materials and methods for details ) . ( B ) mRNA-Seq differential gene-expression analysis for indicated sample pairs . Each dot in the plot represents one gene . The log2 scale values of FPKM ( Fragments Per Kilobase of transcript per Million mapped reads ) of the gene for the comparing samples are plotted . Orange dots , 1 . 5-fold up-regulation compared to wildtype; green dots , 1 . 5-fold down compared to wildtype . Orange number , total number of 1 . 5-fold up-regulated genes; Green number , total number of 1 . 5-fold down-regulated genes . ( C ) Heatmap showing Q111 dose-dependent gene expression changes . Genes that are up/down regulated in Q111 dose dependent or independent fashion are clustered . Co-regulated up/down genes: genes that are up/down regulated in both Q111/Q7 and Q111/Q111 striatal cells . The expression level of each gene is color-coded . For each group , genes are ranked by the expression levels in the wild-type control ( Q7/Q7 ) . ( D ) Percentage of Sp1 ChIP-exo peaks in core and proximal promoter ( 0–5 kb ) , distal region ( 5–50 kb ) , intergenic region ( 50–500 kb and > 500 kb ) of annotated Refseq TSS in wild type striatal cells ( Q7/Q7 ) , the distribution was calculated by GREAT ( McLean et al . , 2010 ) . ( E ) Venn diagram showing overlaps between Sp1 target genes and genes showing Q111 dose-dependent changes in Htt mutant striatal cells . Sp1 target genes are defined as genes with Sp1 peaks within 10 kb of their TSS . 124 of dose-dependent down-regulated genes have Sp1 binding events within 10 kb from their TSS ( p<2 . 07e-8 , hypergeometric test ) . ( F ) Sp1 overexpression in Htt homozygous mutant cells ( Q111Q111 ) rescues the expression levels of a fraction of its direct target genes . Left panel , western blot using anti-Sp1 antibody to show expression of endogenous and Halo-tagged Sp1 in Q111/Q111 striatal cells with no ( - ) , low ( + ) and high ( ++ ) expression level of HaloTag-Sp1 . Q111/Q111 lines with high HaloTag-Sp1 expression level were used in RNA-seq experiments to evaluate gene expression rescue . Right panel , Bar graph showing the numbers of rescued ( 1 . 5 fold up- or down- regulated compared to Q111/Q111 ) and unchanged Sp1 dose-dependent targets upon HaloTag-Sp1 overexpression . ( G ) Representative ChIP-exo and RNA-seq tracks of Q111 dose-dependent downregulated Sp1 target genes that were rescued by HaloTag-Sp1 overexpression . Left panel: Ier3 , Right panel: Phlda1 . ChIP-exo tracks show 5’-end of sequence tags on the sense ( red ) and anti-sense ( green ) strands . For RNA-seq tracks , y-axle was set to the same scale ( 0–460 for Ier3 , 0–675 for Phlda1 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 02110 . 7554/eLife . 17056 . 022Figure 7—figure supplement 1 . Over-expression of Sp1 restores gene expression defects in HD affected striatal cells . ( A ) Cystathione gamma-lyse ( Cth ) expression levels in wild-type ( Q7/Q7 ) , heterozygous ( Q7/Q111 ) and homozygous ( Q111/Q111 ) striatal cells measured by RNA-seq . Expression levels of Cth FPKM in different samples are normalized to that in wildtype striatal cells . ( B ) Sp1 antibody detects a single band in western blot . Whole cell extracts of wild type and Sp1 null ES cells were analyzed by western blot with Tubulin as the loading control . ( C ) Sp1 immunofluorescence staining of wild-type ( upper panel ) and Sp1 null ( lower panel ) ES cells . Nuclei were counterstained with DAPI . ( D ) Representative ChIP-exo and RNA-seq tracks of Q111 dose-dependent downregulated Sp1 target genes that were rescued by HaloTag-Sp1 overexpression . Upper panel: Myc , Lower panel: Rnd1 . ChIP-exo tracks show 5’-end of sequence tags on the sense ( red ) and anti-sense ( green ) strands . For RNA-seq tracks , y-axle was set to the same scale for different samples ( 0–692 for Myc , 0–125 for Rnd1 ) . ( E ) Q-PCR analysis of indicated genes in different striatal cell lines , including Q7/Q7 , Q7/Q111 , Q111/Q111 , and Q111/Q111 with Sp1 over-expression . Error bars present standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 02210 . 7554/eLife . 17056 . 023Video 7 . Reconstructed single-molecule trajectories of Htt-25Q ( HaloTag , JF549 ) overlaid on imaging data in STHdh cells with dragon presentation ( 6 step delayed ) . Magenta; Htt-25Q single-molecule imaging . Green; mHtt-94Q-CFP Epi-fluorescence image . Presentation software: Imaris . DOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 02310 . 7554/eLife . 17056 . 024Video 8 . Zoom-in view at the mHtt aggregate containing region , showing non-specific trapping of Foxp2 at the single-molecule level . Presentation software: Fiji . DOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 02410 . 7554/eLife . 17056 . 025Video 9 . Target search simulation in Htt-aggregate containing cell . Small black dots and semi-transparent spheres represent target sites and mHtt aggregates , respectively . TF molecule movement is slowed down ( red fragments ) in the aggregate regions . Presentation software: MatlabDOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 025 Reduction of target site sampling or interaction frequencies between TF and cognate DNA target sites would presumably affect downstream biochemical activities such as transcriptional activation or repression ( Chen et al . , 2014b ) . Here we use Sp1 as the model to test this hypothesis . To do so , we first performed differential gene expression ( mRNA-seq ) experiments on STHdh striatal cell lines derived from Wt ( Q7/Q7 ) , heterozygous ( Q7/Q111 ) and homozygous ( Q111/Q111 ) HD mutant mice ( Trettel et al . , 2000 ) . We identified Q111 allele dose-dependent down ( 449 ) or up-regulated ( 558 ) genes ( Figure 7B and C and Supplementary file 2 ) . To map Sp1 binding sites in the striatal cells , we generated an antigen-purified polyclonal Sp1 antibody for western blot , immuno-precipitation and cell staining ( Figure 7—figure supplement 1B and C , See Materials and methods for details of antibody generation and purification ) . This antibody detects a single band of Sp1 of the right size in whole cell extracts and specifically stained the cell nucleus ( Figure 7—figure supplement 1B and C ) . Importantly , we confirmed that the antibody is specific because both western blot and nuclear staining of Sp1 disappear when Sp1-null ES cell samples were used . We next performed ChIP-exo experiments with this highly specific anti-Sp1 to map genome-wide Sp1 binding sites in STHdh Q7/Q7 cells ( Trettel et al . , 2000 ) and identified 6 , 698 Sp1 target sites in the mouse genome . Most of these binding sites are tightly associated with transcription start sites ( TSS ) ( Figure 7D , Supplementary file 3 and Supplementary file 4 ) . Strikingly , Sp1 binding sites are significantly enriched around Q111 dose-dependent down-regulated but not up-regulated genes ( Figure 7E ) , suggesting that mHtt mostly likely influences Sp1 transcriptional activation of target genes . It’s likely that the Sp1 target site sampling frequency is reduced in HD affected cells and this could eventually lead to down regulation of the Sp1 target genes expression . Previously , we showed that tuning-up TF concentrations can increase target site sampling frequencies ( Chen et al . , 2014b ) and thus , potentially counteract the Sp1 trapping effects caused by mHtt aggregates . To test this , we over-expressed HaloTag-Sp1 in STHdh Q111/Q111 cells ( Figure 7F ) and performed mRNA-seq experiments . A modest elevation ( ~ 2x ) of Sp1 levels partly rescued 50 dose-dependent down-regulated Sp1 target genes , including Rnd1 , Phlda1 , Ier3 and Myc ( Figure 7F and G , Figure 7—figure supplement 1D and E ) . Interestingly , Rnd1 is a neuron-specific GTPase important for dendritic spine formation ( Ishikawa et al . , 2003 ) . Both Phlda1 and Ier3 genes are involved in anti-apoptotic pathways ( Neef et al . , 2002; Wu , 2003 ) . To further validate whether the Sp1 target sampling frequencies are reduced in Q111/Q111 cells , we performed Sp1 ChIP-qPCR experiments on Ier3 and Rnd1 promoter regions in Q7/Q7 and Q111/Q111 cells . Indeed , the enrichment of Sp1 ChIP-signals were dramatically reduced in Q111/Q111 cells compared to control ( Figure 8A ) . Taken together , functional genomic and ChIP studies on the striatal cell culture system further reinforced the concept that mHtt aggregates likely slow down TF target search processes and reduce the target sampling frequencies that can subsequently lead to altered gene expression programs in diseased cells ( Figure 8B–D ) . 10 . 7554/eLife . 17056 . 026Figure 8 . Model of Htt aggregates affecting TF dynamics . ( A ) Sp1 target site occupancies in Q7/Q7 and Q111/Q111 cells assayed by ChIP-qPCR . The ChIP signals are normalized to the amount of input chromatin DNA . ( B–D ) Large numbers of decoy binding sites created by Htt aggregates in the cell increase trial-and-error collision steps before a TF finds its cognate target site . So , the apparent Kon of TF binding to specific sites is reduced . This leads to reduction of target site sampling frequencies ( C ) . Presumably if the TF is a critical transcriptional activator for a gene , the target gene expression will be impeded as well ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17056 . 026 It is worth noting that consistent with a recent report identifying the Cth ( Cystathionine Gamma-Lyase ) gene as a key HD disease effector ( Paul et al . , 2014 ) , our genomic assays faithfully detected Cth as a Q111 dose-dependent down-regulated gene ( Figure 7—figure supplement 1A ) and Cth gene expression can be partially rescued by Sp1 overexpression ( Figure 7—figure supplement 1E ) . However , we did not detect Sp1 binding at the proximal enhancer region of the gene , suggesting that long-distance enhancer regulation might be involved at this locus . Using advanced live-cell , single molecule imaging , we were able to directly visualize slower diffusion and retention of key neurological disease related regulatory factors to the mHtt aggregates , effectively operating as large sticky molecular traps . With an abundance of these decoy sites snagging TFs in the nucleus and the cytoplasm , target searching by transcription factors must undergo many more trials and nonproductive collisions before reaching their designated cognate binding sites to activate the expression of essential genes . Our simulation experiments determined that these large molecular traps greatly increased target search times and reduced target site sampling . In addition , if TF molecules become permanently incorporated into mHtt aggregates , this would effectively lead to the reduction in copy number of actively searching TF molecules in the nucleus . Both mechanisms would inevitably reduce the effective concentration of TFs in the cells and thus impair molecular kinetics associated with these factors ( Equations 1–8 ) . Consistent with this model , elevating concentrations of TFs such as Sp1 counteracts the Q111 induced gene expression changes in Htt mutant striatal cells . Our SMT experiments suggest that the biochemical reactions associated with Foxp2 and wild type Htt-25Q also become impaired by the same mechanism . The observation that Foxp2 is recruited to mHtt aggregates via its TAD is particularly interesting , as Foxp2 is a critical factor for the cortical-basal ganglia circuits , which is the most affected brain region in HD ( Enard et al . , 2009 ) . Genetic studies suggest that disruption of Foxp2 function results in speech and language impairment , which is a common HD symptom ( Albin et al . , 1989; Lai et al . , 2001; Ross and Tabrizi , 2011; Takahashi et al . , 2003 ) . Interestingly , it was also shown that cerebellar defects were observed in Foxp2 null mice , with Purkinje cells particularly affected ( Shu et al . , 2005 ) . Consistent with this observation , extensive studies suggest reduced Purkinje cell density in patients with HD ( Jeste et al . , 1984; Rees et al . , 2014; Rub et al . , 2013 ) . Together , our findings strongly suggest that Foxp2 might be one of the essential effectors responsible for the selective disruption of specific neuronal cell populations in patients with HD . Wild-type Htt has been reported to be involved with axon trafficking in neurons ( Gauthier et al . , 2004; Strehlow et al . , 2007; Trushina et al . , 2004 ) . Genetic studies show that knockout of Htt induces certain HD phenotypes and overexpression of wild-type Htt rescues these phenotypes ( Cattaneo et al . , 2005; Gauthier et al . , 2004 ) , implying that dominant negative effects of mHtt on the wild type protein might contribute to the disease state ( Ho et al . , 2001; Nasir et al . , 1995; Zeitlin et al . , 1995 ) . Here , we provided direct evidence to support this model , as the wild-type Htt-25Q fragment also becomes selectively trapped in mHtt aggregates . Together , our data suggest that disrupting the finely tuned molecular kinetics of key regulatory proteins by mHtt aggregates may represent an important common mechanism underlying HD and we speculate that other protein aggregate forming diseases may share a similar disease mechanism . Early studies of classical enhancer binding proteins suggested that low-complexity peptides such as Q-rich domains are important for transcriptional activation ( Courey et al . , 1989 ) . Detailed mutagenesis experiments revealed that in addition to Qs , hydrophobic amino acids within these Q-rich sequences are also critical for transcriptional activation ( Gill et al . , 1994 ) . Remarkably here , our domain mapping and mutagenesis experiments found that the sequence motif in Sp1 responsible for mHtt aggregate binding is the exact same Sp1 TAD region A required for transcriptional activation . Consistent with our results , recent studies suggest that aromatic residues are essential for the formation of amyloid-like structures by low-complexity domains in RNPs ( Kato et al . , 2012; Molliex et al . , 2015; Nott et al . , 2015 ) . Thus , in the future , it would be interesting to study how aromatic amino acids affect protein structure of low-complexity sequences . Do aromatic residues promote formation of beta-sheets and position sparse Qs in the right orientation to interact with Poly-Q aggregates ? Another important message derived from this study is that continuous PolyQ tracts are not necessary for mHtt aggregate binding . Likely , a much broader spectrum of proteins with low-complexity sequences than we have studied are affected by the presence of mHtt aggregates in the cell . How PolyQ expansion affects the dynamic behavior of proteins with low-complexity domains in living cells has been difficult to study . Our live-cell , single-molecule PALM imaging experiments suggest that Htt proteins have at least 3 distinct states under physiological conditions of living cells . Specifically , wild-type Htt fragments display fast diffusing and dynamic clustering states , while mutant Htt fragments are also able to form more stable large aggregates in the cell . These findings are remarkably reminiscent of recent reports regarding the behavior of the low-complexity domain ( LCD ) containing RNA granule proteins ( such as TDP-43 , FUS , hnRNPA1 ) . Most interestingly , low-complexity domains of TDP-43 and FUS are able to form reversible liquid droplet/polymer-like hydrogels in vitro and in live cells ( Kato et al . , 2012; Molliex et al . , 2015; Murakami et al . , 2015; Nott et al . , 2015; Patel et al . , 2015 ) . As revealed by FRAP experiments , the protein in the droplet states is still quite dynamic ( recovery time < 1 min ) . However , ALS/FTD associated mutations make LCDs of TDP-43 and FUS prone to undergo phase transition to form irreversible hydrogels , impairing the RNP granule function ( Patel et al . , 2015; Xiang et al . , 2015 ) . Spatiotemporal information-rich live-cell PALM experiments allows us to investigate the temporal dynamics of different Htt states . The dynamic clustering phase of Htt protein has a lifetime of 10~20 s , forming clusters of sub-diffractive sizes . The clusters don’t grow further and they quickly disassemble , suggesting that the clustering phase is rapidly reversible and thus analogous to the reversible droplet/hydrogel phase of LCDs in RNPs . In contrast , mHtt aggregates are much more stable with little dynamic exchange of molecules as seen in our long-term FRAP experiments , suggesting that mHtt aggregates may be more akin to irreversible polymeric-hydrogel states . It seems likely that PolyQ expansion in certain proteins promote the formation of irreversible phase transitions to the aggregated state overcoming the reversible dynamic clustering state . Although the underlying protein sequence requirements of various LCDs such as the Q-rich ones studied here may be quite different from other LCDs , our studies revealed a remarkable resemblance in protein dynamics associated with LCDs in RNPs and PolyQ expansion proteins . Mutations in both classes of LCDs cause severe neurodegenerative diseases . Recent structural studies suggest that beta-sheets are likely structural motifs underlying the formation of liquid droplets and hydrogel polymers for LCDs in RNPs ( Xiang et al . , 2015 ) . It would be interesting in future studies to determine whether similar structural motifs form the basis for PolyQ expansion induced protein aggregation . Mouse D3 ( ATCC , Manassas , VA ) ES cells were maintained on 0 . 1% gelatin coated plates in the absence of feeder cells . The ES cell medium was prepared by supplementing knockout DMEM ( Invitrogen , Carlsbad , CA ) with 15% FBS , 1 mM glutamax , 0 . 1 mM nonessential amino acids , 1 mM sodium pyruvate , 0 . 1 mM 2-mercaptoethanol and 1000 units of LIF ( Millipore ) . Striatal STHdh Q7/Q7 , Q7Q111 , Q111/Q111 cells were cultured at 33°C in the high-glucose DMEM medium ( without phenol-red ) supplemented with 10% FBS , 1 mM glutamax and 1 mM sodium pyruvate . All cell lines were authenticated by genome-wide gene expression profiling ( mRNA-seq ) and tested as mycoplasma negative . Sp1 , TBP and H2B cDNA was amplified from ES cell cDNA libraries . Foxp2 cDNA was obtained from GE Dharmacon ( Cat . #: MMM1013-202798679 ) . Htt-94Q and Htt-25Q cDNA were obtained from Addgene ( Plasmid #23966; Htt-94Q and Plasmid #1177; Htt-25Q ) . Subsequently , full-length and truncated protein fragments were cloned into Piggybac transposon vector ( PB533A-2 , System Biosciences ) or a modified Piggybac transposon vector with PuroR using indicated primers or gBlock fragments ( IDT ) ( Supplementary file 1 ) . HaloTag ( Promega , Madison , WI ) or mEOS3 . 2 ( Addgene: Plasmid #54525 ) was further cloned to fuse with fragments at the N-terminus ( Sp1 , Sp1 fragment , TBP ) or C-terminus ( H2B , Htt-25Q and Htt-94Q ) . The primer information for cloning is in Supplementary file 1 . Stable cell lines were generated by co-transfection of Piggybac transposon vector with a helper plasmid that over-expresses Piggybac transposase ( Supper Piggybac Transposase , System Biosciences ) . 48 hr post-transfection , ES or STHdh cells were subjected to neomycin or puromycin ( Invitrogen , Carlsbad , CA ) selection . For electroporation , ES cells or STHdh cells were first dissociated by trypsin into single cells . Transfection was conducted by using the Nucleofector Kits for Mouse Embryonic Stem Cells ( Lonza , Basel , Switzerland ) or Nucleofector Kits for Mouse Neural Stem Cells ( Lonza , Basel , Switzerland ) . For the second vector , we use the same piggybac genome integration system . Instead of drug selection , fluorescence activated cell sorting was used to separate stably transfected cells . One day before the imaging experiment , ES cells were plated onto an ultra-clean cover glass pre-coated with IMatrix-511 ( Clontech , Mountain View , CA ) . Similarly , striatal cells were plated onto an ultra-clean cover glass without coating . The striatal cell imaging medium was the same as the culturing medium . ES cell imaging experiments were performed in the ES cell imaging medium , which was prepared by supplementing FluoroBrite medium ( Invitrogen ) with 10% FBS , 1 mM glutamax , 0 . 1 mM nonessential amino acids , 1 mM sodium pyruvate , 10 mM Hepes ( pH 7 . 2~7 . 5 ) , 0 . 1 mM 2-mercaptoethanol and 1000 units of LIF ( Millipore ) . For single-molecule imaging , we first tested the optimal HaloTag-JF549 and HaloTag-JF647 labeling concentrations . Briefly , chemical structures and synthesis procedures of JF549-HTL and JF646-HTL were described previously ( Grimm et al . , 2015 ) . Several concentrations of JF549-HTL and JF646-HTL ( 0 . 5 nM , 1 nM , 2 nM and 5 nM ) were used to treat cells for 15 min and then cells were washed with imaging medium for 3 times . The cover glasses were then transferred to live-cell culturing metal holders and mounted onto the microscope one by one . Proper HaloTag-JF549 or HaloTag-JF646 labeling concentrations were determined by the criterion that single-molecules can be easily detected after no or a minimal 2 ~ 5 s pre-bleaching . After fixing the labeling concentration for each cell line , we then proceeded to perform the 2D single-molecule imaging experiments . For PA-JF646 labeling of HaloTag-Sp1 , cells were incubated with PA-JF646 with final concentration of ~100 nM for one hour . PA-JF646 stands for a photo-activatable ( caged ) version of JF646 that would become fluorescent upon low-dose 405 nm light irradiation . The chemical structure and synthesis of PA-JF646 were as previously described ( Lavis et al . , 2016 ) . For 3D structured illumination imaging and Airyscan imaging , cells were incubated with the JF549-HTL or JF646-HTL with final concentrations around 50 nM for 30 min to ensure that the labeling is close to saturation . After saturated JF549-HTL labeling , Htt-94Q-CFP/H2B-HaloTag or Htt-94Q-CFP/HaloTag-Sp1 ES cells on the cover glasses were placed on a home-build 3D SIM microscope with environment control ( 37°C ) . Two color live-cell 3D SIM experiments were performed as previously described ( Fiolka et al . , 2012 ) , except that two color data were acquired on a plane-by-plane rather than a volume-by-volume basis to improve the local registration of the two imaging channels within the moving live cell . After saturated JF646-HTL or JF549-HTL labeling , cells on the cover glasses were first fixed with 4% paraformaldehyde for 10 min . The samples were treated with VECTASHIELD antifade mounting medium with DAPI and mounted on a glass slide . The imaging was performed on an inverted Carl Zeiss 880 LSM . Before Airyscan imaging , the beam position on the 32 detector array is calibrated to the center . The final image reconstruction is conducted using the manufacturer provided software . For this analysis , we rescale each mHtt-aggregate containing region to have a diameter of 100 pixels . Then , an intensity map is generated by averaging multiple mHtt aggregate regions obtained across different cells . The center-to-peripheral radial grayscale intensity curve is then calculated by each circular pixel increment for both channels . Htt-94Q-CFP-NLS ES cells were cultured on a clean 25 mm cover glass pre-coated with IMatrix-511 . The cover glass is mounted into an Attofluor Cell Chamber . The cells were imaged with an inverted Carl Zeiss 880 LSM with the environmental control system ( 37°C , 5% CO2 ) . CFP imaging and FRAP were performed using 440 nm laser with 100% laser power and slowed down scanning speed in the FRAP area . The recovering phase ( Figure 1—figure supplement 1F ) of mHtt aggregate FRAP curve is fitted by single exponential decay model to extract FRAP recovery lifetime . 2D single molecule imaging experiments were conducted on a Nikon Eclipse Ti microscope equipped with a 100X Oil-immersion Objective lens ( Nikon , N . A . = 1 . 41 ) , a lumencor light source , two filter wheels ( Lambda 10–3 , Sutter Instrument , Novato , CA ) , perfect focusing systems and EMCCD ( iXon3 , Andor , Belfast , United Kingdom ) and Tokai Hit ( Japan ) environmental control ( humidity , 37°C , 5% CO2 ) . Proper emission filters ( Semrock , Rochester , New York ) were switched in front of the cameras for GFP , JF549 or JF646 emission and a band mirror ( 405/488/561/633 BrightLine quad-band bandpass filter , Semrock ) was used to reflect the laser into the objective . For tracking the fast diffusion of JF549 labeled molecules , we used a 561-nm laser ( MPB Lasertech , Quebec , Canada ) of of excitation intensity ~ 800 W cm−2 with the acquisition time of 10 ms . For tracking fast diffusion of JF646 labeled molecules , we used a 630-nm laser ( Vortran Laser Technology , Inc . Sacramento , CA ) of excitation intensity ~ 800 W cm−2 with the acquisition time of 10 ms . For single-molecule localization microscopy experiment mapping the spatial relationship of chromatin ( H2B-HaloTag-JF549 ) and mHtt aggregates ( Htt-94Q-CFP ) , we used a SOLA light engine ( Lumencor , Beaverton , OR ) for imaging mHtt aggregates and a 561-nm laser ( MPB Lasertech , Quebec , Canada ) of excitation intensity ~1 kW cm−2 for single-molecule imaging of H2B-HaloTag-JF549 molcules with an acquisition time of 10 ms ( 561 nm ) . Htt-25 Q and Htt-94 Q –mEOS3 . 2 live-cell sptPALM experiment was performed using the 560-nm laser ( MPB Lasertech ) of excitation intensity ~ 1000 W cm−2 for single-molecule detection and a 405-nm laser ( Coherent , Santa Clara , CA ) of excitation intensity of 40 W cm−2 for photo-switching of mEOS3 . 2 moiety . It is important to note that the fluorescent proteins used in PALM may display triplet blinking in the millisecond time range , therefore most often filtered out by a detector integration times larger than 10 ms . Here we use a camera integration time of 20 ms to minimize the later counting contribution from blinking . Total ~20000 frames were recorded . ~30 k localized events were used for the final imaging reconstruction . PA-JF646 HaloTag-Sp1 live-cell single-molecule tracking experiment was performed using a 630-nm laser ( Vortran Laser Technology , Inc . Sacramento , CA ) of excitation intensity ~ 800 W cm−2 for single-molecule detection and a 405-nm laser ( Coherent , Santa Clara , CA ) of excitation intensity of 20 W cm−2 for photoactivation of PA-JF646 moiety . The acquisition time is 10 ms . Before single-molecule imaging , we took epi-fluorescence images in all labeling channels for later references . We calibrated our imaging system to achieve minimal drift during acquisition ( xy drift <10 nm per hour ) . For single molecule localization and tracking , the spot localization ( x , y ) was obtained through 2D Gaussian fitting based on MTT algorithms ( Serge et al . , 2008 ) using home-built Matlab program . The localization and tracking parameters in SPT experiments are listed in the Supplementary file 1 . MTT algorithm was used to track fast TF dynamics . The resulting tracks were inspected manually . Diffusion coefficients were calculated from tracks with at least 5 consecutive frames by the MSDanalyzer ( Tarantino et al . , 2014 ) with a minimal fitting R2 of 0 . 8 . For color-coded representation of the trajectory map ( Figures 1 , 4 , 6 and S6 ) , each trajectory is color-coded according to its diffusion coefficient with a ‘jet’ colormap . The trajectory map is reconstructed with the plot ( ) function in matlab 2015a . For track division in Figure 6 and Figure 6—figure supplement 1 , a binary mask is generated first by using the image from Htt-94Q channel . Trajectories that have partial fragments within the aggregate region are computationally separated for the diffusion analysis . The rest of trajectories are as the control . For sub-regional diffusion analysis in Figure 6B and Figure 6—figure supplement 1A , binary masks are generated according to mHtt aggregate channel . Trajectories are computationally divided into two populations based on the physical proximity of the trajectory to the mHtt aggregate region . Trajectories contain no localization events in the mHtt aggregate region and trajectories fully or partially included in the mHtt aggregate regions were separated for downstream diffusion analysis . Spatial dependence of Htt-25Q ( Figure 1A ) , Htt-94Q ( Figure 1A ) and Sp1 ( Figure 4A ) diffusion were analyzed using a Bayesian inference mapping algorithm as previously described ( El Beheiry et al . , 2015 ) . Trajectories were spatially partitioned using a hierarchical ( quad-tree ) mesh . Dimensions of the zones in this type of mesh were adapted to the characteristic size of the trajectory steps within them , hence accounting for spatially dependent heterogeneities in diffusive behavior . For each zone , the diffusion was presumed to be constant and was calculated by considering all trajectory steps within it ( the total length of the trajectory is not consequential ) . Trajectories were modeled by an overdamped Langevin equation , allowing for physical parameters governing single-molecule movement ( e . g . diffusion and interaction energies ) to be distinguished . The diffusion coefficient within each zone was calculated as the result of a maximum a posteriori estimate from a Bayesian inference calculation . The localizations from 1000 frames were pooled and used for reconstruction of a single localization density map . This window slides every 0 . 5 s ( 25 frames ) until the end of frames . The reconstructed images are temporally ordered and used for the sliding-window video reconstruction ( Video 4 ) . For time-counting analysis , the localizations from each cluster/aggregate region are selected and the temporal localization history of these localizations are used for the vertical line and the cumulative density plots ( Figure 1C and Figure 1—figure supplement 1D ) . The control regions were selected based on the criterion that no visible clusters were observed in these regions of the final reconstructed image . The mean lifetime of dynamic clusters in Htt-25Q and Htt-94Q cells are estimated by half of the duration between appearing and disappearing of the cluster in the sliding-window video . Tracks from the same condition were pooled , and a sliding window of 2 points was applied to each track . The physical distance between two points was calculated by the pdist2 ( ) function in the Matlab 2015a ( MathWorks Inc , Natick , MA ) . The program iteratively processed all tracks in each category and individual distance were pooled and binned accordingly for the translocation histogram ( Figure 1B ) . Tracks from the same condition were pooled , and a sliding window of 3 points was applied to each track . The angle between the vectors of the first two and the last two points was calculated by the acos ( ) function in the Matlab 2015a ( MathWorks Inc , Natick , MA ) . The program iteratively processed all tracks in each category and individual angles were pooled and binned accordingly for the angular Rose histogram ( Figure 6—figure supplement 1C ) . The minimal jumping distance between two points are set as 40 nm to ensure that the angle measurement is not significantly affected by the localization uncertainty . To initialize the conditions for target search simulation , a sphere with a diameter of 400 units is generated and represents the nucleus of an individual cell . A number of N mHtt aggregates represented by small spheres with diameter of D are randomly allocated within the nucleus sphere . 5 , 000 TF binding sites ( diameter of 2 units ) are next randomly allocated within the remaining space , in order to mimic the binding loci within the genome for the transcription factor . During one simulation , a transcription factor molecule performs 3D Brownian walking unit by unit within the free space of nucleus from a randomly assigned position as previously described ( Liu et al . , 2015 ) , until it arrives at the region of a binding locus . When it walks through the region of mHtt aggregates , the walking speed is reduced by 100-fold . The total number of walking steps is recorded for each simulation . The mean walking step of 2000 simulations is calculated when the values of N , D or VR ( volume ratio , total aggregates volume divided by nucleus volume ) are varied . For antibody used in Sp1 western blot , staining and ChIP experiments , rabbits were immunized with Sp1 residues 1-60aa GST fusion proteins . It is important to note that the sequence of the antigen region is specific to Sp1 but not to other SP family proteins . The antisera obtained were further affinity-purified using MBP-antigen fusion protein immobilized on Affigel 10/15 resin ( Bio-Rad , Hercules , CA ) . Whole cell extracts from ES Cells and SHTdh striatal cells were isolated using RIPA buffer that contained Complete Protease Inhibitor Cocktail ( Roche , Basel , Switzerland ) . Protein concentrations were measured using Bio-Rad Protein Assay against BSA standards ( Bio-Rad , Hercules , CA ) . Protein from each sample was resolved by SDS-PAGE . Primary antibodies used: Sp1 ( Custom-made , 1:1000 ) and beta-tubulin ( ab6046 , Abcam ) . HRP conjugated secondary antibodies ( Pierce , ThermoFisher Scientific , Waltham , MA ) were used at a dilution of 1:5000 . Western Lightning Plus–ECL ( PerkinElmer , Waltham , MA ) was used for chemiluminescent detection . ES cells were first fixed with 4% paraformaldehyde , permeabilized with PBST ( PBS plus 0 . 2% Triton X-100 ) and blocked with 10% FCS and 1% BSA in PBST . Samples were stained with Sp1 primary antibody ( 1:100 ) in blocking solution . Secondary antibodies: DyLight 549 conjugated secondary antibodies ( anti-rabbit , 1:400 , Jackson ImmunoResearch , West Grove , PA ) . Nuclei were counterstained with DAPI . Chromatin Immunoprecipitation ( ChIP ) was performed according to ( Boyer et al . , 2006 ) with minor modifications . Briefly , cross-linked EB chromatin was sheared using Covaris S2 system to a size range of 100bp ~ 400bp . Immuno-precipitation was conducted with Sp1 antibody-conjugated Protein A Sepharose beads ( GE Healthcare ) . ChIP-exo library was prepared by following the published protocol with minor modifications ( Rhee and Pugh , 2011 ) . Specifically , we adapted the SoLid sequencer adaptors/primers to make the final library compatible with the illumina Tru-seq small-RNA system . The detailed primer information is in Supplementary file 1 . We sequenced exo libraries in 50bp single-end format by using the illumina HiSeq platform . After removal of the 3′ most 14 bp which tend to have higher error rates , we mapped our sequencing data back to the mouse reference genome ( mm10 ) by Bowtie 2 ( Langmead and Salzberg , 2012 ) . After mapping , bound regions ( Supplementary file 3 ) were detected by using MACS2 ( Zhang et al . , 2008 ) . Raw sequencing data were deposited to NCBI GEO with the accession number of GSE84058 . Total RNA from STHdh Q7/Q7 , Q7/Q111 , Q111/Q111 and Sp1 overexpressed Q111/Q111 was isolated using RNeasy kit ( 74106 , Qiagen , Valencia , CA ) . mRNA was then purified using Dynabeads Oligo ( dT ) 25 ( 25–61002 , Life Technologies , Carlsbad , CA ) . RNA-seq library was prepared using ScriptSeq v2 RNA-seq Library Preparation Kit ( SSV21106 , Illumina , San Diego , CA ) , and then sequenced using an Illumina Hiseq 2000 sequencing platform . Raw sequencing data were deposited to NCBI GEO with the accession number of GSE84058 . We sequenced mRNA-seq samples in 50bp single-end format ( 1 lane HiSeq per sample ) . Reads were mapped to the mouse mm10 genome using Tophat , and Read counts were tallied for each Ensembl annotated protein-coding gene ( Ensembl 61 ) incremented by 1 and differential expression tested using Cuffdiff using all qualified samples . Cuffdiff results were further analyzed by CummeRbund ( Trapnell et al . , 2009; Trapnell et al . , 2012 ) . In order to reveal the relationship between Number of non-specific sites ( Nns ) , the search time to the specific site ( τsearch ) and the specific site sampling frequencies ( Fsampling ) . The following terms are defined during the calculation: kns: association rate to one non-specific site ks: association rate to one specific site Ns: number of specific binding sites in a cell . Nns: number of nonspecific binding sites in a cell . NTF: number of transcription factors in a cell . Psb: probability for a free particle to bind to a specific site . τns: non-specific residence time . τs: specific residence time . The probability Psb was calculated as follows: ( 1 ) Psb=ksNsksNs+knsNns In particular , 1/Psb gives the average number of trials ( Ntrials ) for a TF to reach a specific binding site . ( 2 ) Ntrials=1+knsNnsksNs The average duration for diffusion between two binding sites ( 3 ) τ3D=1ksNs+knsNns Previously , we demonstrate that the total search time for one TF to reach a specific site ( Chen et al . , 2014a ) ( 4 ) τsearch=Ntrials ( τ3D+τns ) −τns Therefore , by combining ( Equations 2–4 ) , the specific target search time τsearch can be calculated as ( 5 ) τsearch=1ksNs+τnsknsNnsksNs Based on ( Equation 5 ) , increasing non-specific binding sites ( Nns↑ ) in the cell would directly lead to longer specific search times ( τsearch↑ ) . Previously , we also show ( Chen et al . , 2014a ) that a first order approximation of specific site sampling interval can be defined as ( 6 ) Sampling Interval ( s ) , Tsampling= ( τsearch+τsNTF ) Ns Accordingly , ( 7 ) Sampling frequency , Fsampling=1Tsampling Based on Equations 5–7 , increasing Nns↑ in the cell would lead to longer specific search times ( τsearch↑ ) longer target site sampling intervals ( Tsampling↑ ) and lower target site sampling frequencies , Fsampling↓ . Finally , it is important to note that if TF molecules are permanently trapped in mHtt aggregates , this would directly lead to the reduction of concentrations ( copy number ) of searching TF molecules ( [TF]↓ ) in the nucleus . The effective association rates of TF molecules to DNA target sites ( Kon* ) is thus decreased , according to the equation below: ( 8 ) Kon*↓=Kon[TF]↓ This model is not mutually exclusive to the target search model presented above . In fact , both mechanisms might play a role .	Huntington's disease belongs to a group of human genetic disorders in which faulty proteins cause nerve cells to progressively die . All proteins are made from building blocks called amino acids , and these diseases are collectively called PolyQ expansion diseases because the faulty proteins usually have an abnormally long stretch that contains many repeats of an amino acid called glutamine . The stretches of glutamines make these mutant proteins stick to each other , which means that they aggregate in the diseased cells . Researchers have proposed several mechanisms to explain how aggregates of one such mutant protein , called mutant Huntingtin ( mHtt ) , might contribute to Huntington's disease . One possibility is that mHtt accumulates in the nucleus of the cell – which houses most of the cell’s DNA – and interferes with the proteins that are required to switch genes on or off . Preventing these gene regulatory proteins from carrying out their role could disrupt the normal pattern of gene activity . However , working out if this is the case would require researchers being able to follow how mHtt forms aggregates and interferes with normal processes in living cells . Li et al . have now used high-resolution , high-speed microscopy to directly track the movement of individual mHtt proteins in living mouse cells in real time . The analysis showed that PolyQ protein fragments of mHtt behaved in three distinct ways – they diffused , briefly clustered or stably aggregated . Large stable aggregates of mHtt created decoy-like traps that interfered with the behavior of the gene regulatory proteins . Using computer-aided simulations , Li et al . then showed that that these molecular traps make the proteins take much longer to find their true target genes . Further studies in nerve cells suggested that this phenomenon disrupted the normal pattern of gene activity . Next , Li et al . also used the live cell imaging system to look at which gene regulatory proteins were sequestered in the mHtt aggregates . This approach identified proteins that are important regulators involved in a number of neurological disorders . These proteins included some with long stretches of glutamines and unexpectedly others in which the glutamines were interspersed with other amino acids . Aggregates of mHtt therefore appear to affect a much broader range of proteins than previously thought . Together , the results provide insight into how mHtt forms aggregates and disrupts the finely balanced mechanisms that control gene activity in nerve cells . Future studies could explore the general principles that determine which proteins interact with mHtt aggregates . This could help reveal the faulty processes that underlie Huntington's disease and other neurodegenerative disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "structural", "biology", "and", "molecular", "biophysics" ]	2016	Real-time imaging of Huntingtin aggregates diverting target search and gene transcription
Activation of heterotrimeric G proteins is a key step in many signaling cascades . However , a complete mechanism for this process , which requires allosteric communication between binding sites that are ~30 Å apart , remains elusive . We construct an atomically detailed model of G protein activation by combining three powerful computational methods: metadynamics , Markov state models ( MSMs ) , and CARDS analysis of correlated motions . We uncover a mechanism that is consistent with a wide variety of structural and biochemical data . Surprisingly , the rate-limiting step for GDP release correlates with tilting rather than translation of the GPCR-binding helix 5 . β-Strands 1 – 3 and helix 1 emerge as hubs in the allosteric network that links conformational changes in the GPCR-binding site to disordering of the distal nucleotide-binding site and consequent GDP release . Our approach and insights provide foundations for understanding disease-implicated G protein mutants , illuminating slow events in allosteric networks , and examining unbinding processes with slow off-rates . Heterotrimeric G proteins are molecular switches that play pivotal roles in signaling processes from vision to olfaction and neurotransmission ( Oldham and Hamm , 2006; Oldham and Hamm , 2008; Johnston and Siderovski , 2007 ) . By default , a G protein adopts an inactive state in which guanosine diphosphate ( GDP ) binds between the Ras-like and helical domains of the α-subunit ( Gα , Figure 1 ) . A dimer consisting of the β- and γ-subunits ( Gβγ ) also binds Gα . G protein-coupled receptors ( GPCRs ) trigger G-protein activation by binding Gα and stimulating GDP release , followed by GTP binding to Gα and dissociation of Gα from Gβγ . Gα and Gβγ then interact with effectors that trigger downstream cellular responses . Gα returns to the inactive state by hydrolyzing GTP to GDP and rebinding Gβγ . Given the central role Gα plays , a common Gα numbering scheme ( CGN ) has been established to facilitate discussion of the activation mechanisms of different Gα homologs ( Flock et al . , 2015 ) . For example , the notation Lys52G . H1 . 1 indicates that Lys52 is the first residue in helix 1 ( H1 ) of the Ras-like domain ( also called the GTPase domain , or G ) . S6 refers to β-strand 6 and s6h5 refers to the loop between S6 and H5 . Strikingly , the GPCR- and nucleotide-binding sites of Gα are ~30 Å apart ( Figure 1 ) , but the allosteric mechanism coupling these sites to evoke GDP release remains incompletely understood ( Oldham and Hamm , 2008 ) . Biochemical and structural studies have elucidated some key steps , but the entire process has yet to be described in atomic detail . Early studies of Gα subunits revealed structures of the GDP- and GTP-bound states , as well as the transition state for GTP hydrolysis ( Sunahara et al . , 1997; Leipe et al . , 2002; Westfield et al . , 2011 ) . The high similarity of these structures and the binding of GDP or GTP deep in the protein’s core suggests that activation occurs by adoption of other conformational states that allow GDP release ( Lambright et al . , 1994 ) . One intermediate in G protein activation was suggested by the first crystal structure of a GPCR-bound G protein in which the Ras-like and helical domains of Gα are hinged apart and GDP has dissociated ( Rasmussen et al . , 2011 ) . Structural analysis has led to the proposal of a universal mechanism for G protein activation ( Flock et al . , 2015 ) . In this model , GPCR binding induces translation of H5 away from H1 , which increases disorder in H1 and the P-loop ( or Walker A motif [Leipe et al . , 2002] ) to facilitate GDP release . However , there is evidence that additional intermediates may be involved in Gα activation , ( Oldham and Hamm , 2008; Westfield et al . , 2011; Liang et al . , 2017; Hilger et al . , 2018 ) and the functional importance of this conformational ensemble has been previously suggested ( Furness et al . , 2016 ) . Furthermore , mutagenesis and nuclear magnetic resonance ( NMR ) studies have suggested important roles for other structural elements ( Toyama et al . , 2017; Sun et al . , 2015; Goricanec et al . , 2016 ) . Molecular dynamics simulations promise to capture the entire mechanism of G protein activation and synthesize the wealth of experimental data on this process . Methodological advances now enable simulations to capture millisecond timescale processes for proteins with less than 100 residues ( Lindorff-Larsen et al . , 2011 ) . For example , it is now possible to capture the binding or release of small molecules ( Buch et al . , 2011; Bowman and Geissler , 2012; Silva et al . , 2011; Plattner and Noé , 2015; Tiwary et al . , 2015 ) and peptides ( Plattner et al . , 2017; Zhou et al . , 2017 ) from small proteins . Impressive simulations on the ANTON supercomputer have revealed critical conformational dynamics of G proteins in their inactive and active states , elucidating the role of domain opening in GDP unbinding ( Dror et al . , 2015; Yao et al . , 2016 ) . However , even this specialized hardware could not capture the entire process of G protein activation and GDP release due to the size of the Gα subunit ( >300 residues ) and the slow kinetics of GDP dissociation ( ~10−3 min−1 ) ( Chidiac et al . , 1999; Ross , 2008; Mukhopadhyay and Ross , 1999 ) . Here , we introduce an approach to capture rare or long-timescale events , such as GDP release , and reveal the mechanism of Gα activation . As a test of this methodology , we apply it to Gαq , which has one of the slowest GDP release rates ( Chidiac et al . , 1999 ) and is frequently mutated in uveal melanoma ( Van Raamsdonk et al . , 2009; Van Raamsdonk et al . , 2010 ) . To highlight aspects of the activation mechanism that we propose are general to all G proteins , we focus our analysis on the behavior of secondary structure elements and amino acids that are conserved across Gα domains . Our approach first combines two powerful sampling methods , metadynamics ( Laio and Parrinello , 2002 ) and Markov state models ( MSMs ) , ( Bowman et al . , 2014 ) to observe GDP release and identify the rate-limiting step for this slow process . Then we use our recently developed CARDS method ( Singh and Bowman , 2017 ) , which quantifies correlations between both the structure and disorder of different regions of a protein , to identify the allosteric network connecting the GPCR- and nucleotide-binding sites . The resulting model is consistent with a wealth of experimental data and leads to a number of predictions , described below . Taken together , our results provide the most comprehensive model of G protein activation to date . Based on this success , we expect our approach to be valuable for studying other rare events , including conformational changes and unbinding processes . We reasoned that studying the mechanism of spontaneous GDP release from a truncated form of Gαq would be representative of the mechanism of GPCR-mediated activation of the heterotrimeric G protein while minimizing the computational cost of our simulations . This hypothesis was inspired by previous work demonstrating that a protein’s spontaneous fluctuations are representative of the conformational changes required for the protein to perform its function ( Boehr et al . , 2006; Fraser et al . , 2009; Changeux and Edelstein , 2011 ) . Therefore , we hypothesized that GPCRs stabilize conformational states that G proteins naturally , albeit infrequently , adopt in the absence of a receptor . We chose to focus on Gα since it forms substantial interactions with GPCRs , compared to the relatively negligible interactions between GPCRs and G protein β and γ subunits . This view is supported by the fact that GPCRs and ‘mini’ G proteins , essentially composed of just the Ras-like domain of Gα , recapitulate many features of GPCR-G protein interactions ( Carpenter et al . , 2016 ) . We also reasoned that truncating the last five residues of Gαq would facilitate the activation process . These residues contact Gα in some GDP-bound structures but not in GPCR-bound structures , ( Lambright et al . , 1996; Noel et al . , 1993 ) and removing these residues promotes GDP release due to a reduced GDP-binding affinity ( Denker et al . , 1992; Marin et al . , 2002 ) . Taken together , such evidence suggests that the last five residues of Gαq help stabilize the inactive state and that removing them would accelerate activation . In support of this hypothesis , we find that the energetic barrier to GDP release is lower in metadynamics simulations of the truncated variant than for full-length Gαq ( Figure 2—figure supplement 1 ) . These simulations , and those described hereafter , were initiated from an X-ray structure of the Gαq heterotrimer bound to GDP and an inhibitor of nucleotide exchange ( Nishimura et al . , 2010 ) ; Gβγ and the inhibitor were excluded from all simulations . To observe G-protein activation , we developed a variant of adaptive seeding ( Huang et al . , 2009 ) capable of capturing slow processes like ligand unbinding that are beyond reach of conventional simulation methods . First , we use metadynamics ( Tiwary et al . , 2015; Laio and Parrinello , 2002; Dama et al . , 2014 ) to facilitate broad sampling of conformational space by biasing simulations to sample conformations with different distances between GDP and Gαq . Doing so forces GDP release to occur but provides limited mechanistic information because adding a biasing force can distort the system’s kinetics or cause the simulations to sample high-energy conformations that are not representative of behavior at thermal equilibrium . To overcome these limitations , we chose starting conformations along release pathways observed by metadynamics as starting points for standard molecular dynamics simulations , together yielding an aggregate simulation time of 122 . 6 μs . These simulations should quickly relax away from high-energy conformations and provide more accurate kinetics . Then we use these simulations to build an MSM ( Source data 1 ) . MSMs are adept at extracting both thermodynamic and kinetic information from many standard simulations that , together , cover larger regions of conformational space than any individual simulation ( Bowman et al . , 2014 ) . Related approaches have successfully captured the dynamics of small model systems ( Biswas et al . , 2018 ) and RNA polymerase ( Zhang et al . , 2016 ) . This protocol enabled us to capture the entire mechanism of G-protein activation , including GDP release and the rate-limiting step for this process . Identifying the rate-limiting step for this process is of great value because GDP release is the rate-limiting step for G-protein activation and downstream signaling . Therefore , the key structural and dynamical changes responsible for activation should be apparent from the rate-limiting conformational transition for this dissociation process . To identify the rate-limiting step , we applied transition path theory ( Noé et al . , 2009; Weinan and Vanden-Eijnden , 2006 ) to find the highest flux paths from bound structures resembling the GDP-bound crystal structure to fully dissociated conformations . Then , we identified the least probable steps along the 10 highest flux release pathways ( Figure 2A and Figure 2—figure supplement 2 ) , which represent the rate-limiting step of release . By comparing the structures before and after the rate-limiting step , we define the bound state as all conformations where the distance from the center of mass of GDP’s phosphates to the center of mass of three residues on H1 that contact the GDP β-phosphate ( Lys52G . H1 . 1 , Ser53G . H1 . 2 , and Thr54G . H1 . 3 ) is less than 8 Å . Consistent with this definition , this distance remains less than 8 Å throughout the entirety of 35 . 3 μs of GDP-bound simulations . The conformational changes we observe during the rate-limiting step are consistent with previous structural and biochemical work . For example , we observe that the Ras-like and helical domains separate ( Figure 2C and Source data 3 ) , as observed in DEER experiments ( Van Eps et al . , 2011 ) and previous simulations ( Dror et al . , 2015 ) . This finding is consistent with the intuition that these domains must separate to sterically permit GDP release , and that this separation is driven by the disruption of multiple inter-domain interactions . For example , we note a disrupted salt bridge between K275G . s5hg . 1 and D155H . hdhe . 5 ( Figure 2—figure supplement 4 ) , previously identified in structural studies ( Flock et al . , 2015 ) . Domain opening is accompanied by disruption of a key salt bridge between Glu49G . s1h1 . 4 of the P-loop and Arg183G . hfs2 . 2 of switch 1 ( Figure 2D and Source data 3 ) , as well as an increase in the disorder of many of the surrounding residues ( Figure 2B and Figure 2—figure supplement 3A ) , consistent with the proposal that this salt bridge stabilizes the closed , GDP-bound state ( Liang et al . , 2017 ) . While domain opening is necessary for GDP release , previous simulations suggest it is insufficient for unbinding ( Dror et al . , 2015 ) . Indeed , we also see that this opening is necessary but not sufficient for GDP unbinding , as the Ras-like and helical domains often separate prior to release ( Figure 2C and Source data 3 ) . Notably , the Ras-like and helical domains only separate by ~10 Å during the rate-limiting step . In contrast , these domains separate by 56 Å in the first structure of a GPCR-G-protein complex . This result suggests that GDP release may have occurred long before a G protein adopts the sort of widely opened conformations observed in crystallographic structures ( Rasmussen et al . , 2011 ) . We also observe less expected conformational changes associated with GDP release . The most striking is tilting of H5 by about 26° ( Figure 3A , and Figure 3—figure supplement 1 ) . We find that H5 tilting correlates strongly with the distance between GDP and Gαq along the highest flux dissociation pathway ( Figure 3B and Source data 1 ) . In particular , substantial tilting of H5 is coincident with the rate-limiting step for GDP release . This tilting contrasts with X-ray structures and the universal mechanism , in which H5 is proposed to translate along its helical axis towards the GPCR , initiating the process of GDP release ( Figure 3A ) . During our simulations we also observe translation of H5 , but it is not correlated with the rate-limiting step of GDP release ( Figure 3C , Figure 3—figure supplement 2 , and Source data 3 ) . Therefore , we are not merely missing an important role for translation due to insufficient sampling . The potential importance of H5 tilting is supported by other structural data . For example , a crystal structure of rhodopsin ( Choe et al . , 2011 ) with a C-terminal fragment from H5 of Gαt shows a similar degree of tilting ( Figure 3A ) . Also , the tilt of H5 varies in crystal structures of the β2AR-Gs complex ( Rasmussen et al . , 2011 ) , two different GLP-1 receptor-Gs complexes ( Liang et al . , 2017; Zhang et al . , 2017 ) , and an A2AR-mini-Gs complex ( Carpenter et al . , 2016 ) . The potential relevance of tilting has also been acknowledged by a number of recent works ( Flock et al . , 2015; Rasmussen et al . , 2011; Oldham et al . , 2006 ) including four recently published structures of receptor-G-protein complexes across which H5 also shows a broad range of tilting and translational motion ( Koehl et al . , 2018; Draper-Joyce et al . , 2018; García-Nafría et al . , 2018; Kang et al . , 2018 ) . Interestingly , the tilting and translation we observe falls within the observed range of tilting and translational motions that H5 undergoes in available GPCR-G protein complex structures ( Rasmussen et al . , 2011; Koehl et al . , 2018; Draper-Joyce et al . , 2018; García-Nafría et al . , 2018; Kang et al . , 2018 ) , providing support that conformational selection plays an important role ( Figure 3—source data 1 ) . Finally , H5 is translated toward the GPCR in the A2AR-mini-Gs structure but GDP remains bound ( Carpenter et al . , 2016 ) . The authors of that study originally suggested that one of the mutations in mini-Gs decouples H5 translation from GDP release . However , given that we see GDP release without H5 translation in our simulations , it is also possible that amino acid substitutions required to create mini-Gs instead mitigate H5 tilting . Both of these models are consistent with the fact that some of the mutations in mini-Gs stabilize the GDP-bound state ( Sun et al . , 2015 ) . We propose that tilting of H5 is an early step in the GDP release process , which is followed by upward translation of this helix to form a GPCR-G protein complex primed to bind GTP . This hypothesis stems from our observation that tilting of H5 is coincident with the rate-limiting step for GDP release , while translation of H5 only becomes stable after GDP dissociates ( Figure 3C ) . This model is consistent with previous suggestions that G-protein activation occurs through a series of sequential interactions with a GPCR ( Oldham and Hamm , 2008; Rasmussen et al . , 2011 ) . Another possibility is that any displacement of H5 , whether tilting or translation , may be sufficient to trigger GDP release . While conformational changes of H5 are important for Gα activation , other regions could also play a role ( Hilger et al . , 2018; Sun et al . , 2015 ) . However , it is not straightforward to determine what other structural elements contribute to activation or their importance relative to H5 . Our hypothesis that spontaneous motions of a protein encode functionally relevant conformational changes suggests that the coupling between the GPCR- and nucleotide-binding sites of Gα should be present in simulations of the inactive protein; This provides a means to identify key elements of this allosteric network . To test this hypothesis , we ran 35 . 3 µs of simulation of GDP-Gαq . Then we applied the CARDS method ( Singh and Bowman , 2017 ) to detect correlations between both the structure and dynamical states of every pair of dihedral angles . Structural states are determined by assigning dihedral angles to the three dominant rotameric states for side-chains ( gauche+ , gauche- , and trans ) and the two dominant rotameric states for backbone dihedrals ( cis and trans ) . Dynamical states are determined by whether a dihedral angle remains in a single rotameric state ( ordered ) or rapidly transitions between multiple rotameric states ( disordered ) . These pairwise correlations serve as a basis for quantifying the correlation of every residue to a target site , such as the GPCR-binding site . Combining these correlations with structural and dynamical changes in our model of GDP release provides a basis for inferring how perturbations to the GPCR-binding site are transmitted to the nucleotide-binding site . We focus our analysis on the most direct routes for communication between the GPCR- and nucleotide-binding sites by following correlated motions that emanate from structural elements that directly contact GPCRs until they reach the GDP-binding site . There are correlations between many other elements of Gαq , including parts of the helical domain , that branch off of this allosteric network . Such correlations may be important for aspects of Gα function besides activation , but are beyond the scope of the present study , which focuses on how GPCR-binding impacts nucleotide release . To understand how H5 tilting triggers GDP release , we first identified structural elements with strong coupling to H5 and then worked our way outward in repeated iterations until we reached the nucleotide-binding site ( Figure 4 and Source data 2 ) . This analysis reveals that tilting of H5 directly communicates with and impacts the conformational preferences of the s6h5 loop , which contacts the nucleobase of GDP ( Figure 4 , Figure 5 , Source data 2 , and Source data 3 ) . During the rate-limiting step , these contacts are broken and there is increased disorder in the s6h5 loop , particularly Ala331 of the TCAT motif ( Figure 5 and Figure 2—figure supplement 3B ) . The importance of the TCAT motif in our model is consistent with its conservation and the fact that mutating it accelerates GDP release ( Iiri et al . , 1994; Posner et al . , 1998; Thomas et al . , 1993 ) . Based on our model , we propose these mutations accelerate release by weakening shape complementarity with GDP . We also observe an important role for communication from H5 to H1 , consistent with the universal mechanism . In particular , H1 is strongly coupled with the s6h5 loop ( Figure 4B and Source data 2 ) , which is sensitive to displacement of H5 . In the rate-limiting step , s6h5 moves away from H1 , contributing to an increase in disorder of H1 and the P-loop ( Figure 2—figure supplement 3A and Figure 2—figure supplement 3B ) . Increased disorder in a set of residues that directly contact the GDP phosphates ( Glu49G . s1h1 . 4 , Ser50G . s1h1 . 5 , Gly51G . s1h1 . 6 , Lys52G . H1 . 1 , and Ser53G . H1 . 2 ) likely contributes to a reduced affinity for GDP ( Figure 2—figure supplement 3A ) . Increased disorder in these residues also contributes to disruption of the salt bridge between Glu49G . s1h1 . 4 of the Ras-like domain and Arg183G . hfs2 . 2 of the helical domain , facilitating inter-domain separation . We further note that the s6h5 loop impacts the nucleotide-binding site through allosteric coupling with the HG helix , which also contacts GDP via Lys275G . s5hg . 1 and Asp277G . HG . 2 ( Figures 4E and 6 ) . The disorder of both of these residues increases during the rate-limiting step , consistent with observations of increased mobility in HG upon receptor-mediated activation ( Oldham and Hamm , 2008 ) . There are also correlations between the P-loop and Lys275G . s5hg . 1 on Helix G ( Figure 4E and Source data 2 ) , which result from the disruption of a key salt bridge between Lys275G . s5hg . 1 and Glu49G . s1h1 . 4 on the P-loop during the rate-limiting step ( Figure 6 and Source data 3 ) . Lys275G . s5hg . 1 is conserved across all Gα families , suggesting it plays an important role in the stability or function of the protein . However , attempts to experimentally examine the role of this residue by mutating Lys275G . s5hg . 1 have resulted in aggregation ( Sun et al . , 2015 ) . Our simulations suggest Lys275G . s5hg . 1 plays an important role in stabilizing the GDP-bound state and that breaking the salt bridge with Glu49G . s1h1 . 4 facilitates GDP release . This finding demonstrates the utility of our atomistic simulations , as we can examine the role of residues that are difficult to probe experimentally . To identify other important structural elements in the allosteric network underlying G protein activation , we followed correlated motions emanating from other sites that are known to interact directly with GPCRs , including the hNs1 loop , the h3s5 loop , and the h4s6 loop ( Oldham and Hamm , 2008 ) . We find that h3s5 and h4s6 are largely isolated , suggesting they play a role in forming a stable GPCR-G protein complex but not in the allosteric mechanism that triggers GDP release . This finding is consistent with sequence analysis suggesting these structural elements are important determinants of the specificity of GPCR-Gα interactions ( Flock et al . , 2017 ) . In contrast , the hNs1 loop has strong correlations with β-strands S1-S3 ( Figure 7 and Source data 2 ) . These strands , in turn , communicate with H1 , switch 1 , and the P-loop . Integrating our correlation analysis with structural insight from the rate-limiting step described above suggests an important role for S1-S3 in a complex allosteric network that couples the GPCR- and nucleotide-binding sites ( Figures 7 and 8 , Figure 7—figure supplement 1 , and Source data 2 ) . S2 and S3 twist relative to S1 and away from H1 ( Figures 2A and 9 , Figure 9—figure supplement 1 , and Source data 3 ) . This twisting disrupts stacking between Phe194G . S2 . 6 on S2 and His63G . H1 . 12 on H1 and increases disorder of side-chains in H1 ( Figures 2B and 9 , Figure 2—figure supplement 3C , and Source data 3 ) . Increased disorder in H1 is also a crucial component of the proposed universal mechanism , but in that model translation of H5 is the key trigger for changes in H1 . The role for the β-sheets in our model is consistent with previous work identifying interactions between S2 and H1 ( Flock et al . , 2015 ) , NMR experiments showing chemical exchange in the methyls of S1-S3 upon receptor binding ( Toyama et al . , 2017 ) , and mutational data . In particular , Flock et al . have previously noted the important interaction between residues Phe194G . S2 . 6 and His63G . H1 . 12 ( Flock et al . , 2015 ) . The importance of H1 and β-strands S1-S3 is underscored by mapping the global communication of every residue onto a structure of Gα ( Figure 8—figure supplement 1 ) . The global communication of a residue is the sum of its correlations to every other residue and is a useful metric for identifying residues that are important players in allosteric networks ( Singh and Bowman , 2017 ) . Interestingly , these β-strands and H1 have higher global communication than H5 and the s6h5 loop . This suggests that H1 and the β-sheets integrate conformational information from multiple sources , including the hNs1 loop , and not just H5 . The importance of the β sheets and H1 for allosteric communication is consistent with their conservation ( Sun et al . , 2015 ) , which may not simply reflect the role they play in protein folding and stability , as had been suggested previously ( Sun et al . , 2015; Hatley et al . , 2003 ) . We also find that switch 2 has strong correlations with the nucleotide-binding site , especially switch 1 ( Figure 7—figure supplement 1 and Source data 2 ) . Given that switch 2 is a major component of the interface between Gα and Gβ , this communication could enable GDP release to trigger dissociation of Gα from Gβγ . Examining the rate-limiting step for GDP release reveals that switch 2 shifts towards the nucleotide-binding site ( Figure 10 and Source data 3 ) and exhibits increased conformational disorder ( Figure 2B and Figure 2—figure supplement 3D ) . These findings are consistent with previous kinetic studies postulating that switch 2 dynamics are impacted prior to GDP release ( Herrmann et al . , 2004 ) . We have succeeded in simulating G protein activation , including both the allosteric coupling between the GPCR- and nucleotide-binding sites of Gαq and consequent unbinding of GDP . Our results reveal a previously unobserved intermediate that defines the rate-limiting step for GDP release and , ultimately , G protein activation . Our model synthesizes a wealth of experimental data and previous analyses . For example , we identify an important role for coupling from H5 to the s6h5 loop and H1 that is consistent with a previously proposed universal mechanism for G protein activation . However , we also find that this allosteric network incorporates the hNs1 loop , β-strands S1-S3 , and the HG helix . Strands S1-S3 and H1 serve as hubs in this network , simultaneously integrating information from both H5 and the hNs1 loop . Our observation is consistent with previous postulates that information flows from H5 and hNs1 to H1 ( Preininger et al . , 2013 ) . It is important to note that our model was extracted using simulations of Gαq , and so some correlations or changes in conformation and dynamics may apply only to Gαq . However , by focusing our analysis on secondary structure elements and residues that are shared across all Gα homologs , our model likely captures a universal ‘skeleton’ of changes involved in Gα activation , expanding upon a previously proposed universal mechanism for Gα activation ( Flock et al . , 2015 ) . The consistency of our model with a wide variety of structural and biochemical data suggests that it is a promising foundation for future efforts to understand the determinants of GPCR-Gα interaction specificity , how mutations cause aberrant signaling and disease , and how small molecule inhibitors modulate Gα activation . Our model also adds weight to the growing appreciation for the fact that a protein’s spontaneous fluctuations encode considerable information about its functional dynamics ( Koehl et al . , 2018; Draper-Joyce et al . , 2018; García-Nafría et al . , 2018; Kang et al . , 2018 ) . Construction of our model was enabled by a powerful combination of simulation methods , namely metadynamics and MSMs . In the future , we expect this methodology will prove valuable for understanding other slow conformational changes and unbinding processes . To determine how the GPCR- and GDP-binding regions communicate with one another , we applied the CARDS ( Singh and Bowman , 2017 ) methodology to simulations of the GDP-bound state of Gαq . CARDS measures communication between every pair of dihedrals via both correlated changes in structural motions and dynamical behavior . Structural states are captured by discretizing backbone Φ and ψ dihedrals into two structural states ( cis and trans ) , while side-chain χ angles are placed into three states ( gauche+ , gauche- , and trans ) . Every dihedral is also parsed into dynamical states , capturing whether the dihedral is stable in a single state ( ordered ) , or rapidly transitioning between multiple states ( disordered ) . These dynamical states are identified using two kinetic signatures of dihedral motion: the average time a dihedral persists in a structural state ( an ordered timescale ) , and the typical timescale for transitions between structural states ( a disordered timescale ) . Parsing into dynamical states utilizes a two-step process by ( i ) calculating the distribution or ordered and disordered times from the simulations and ( ii ) assigning each period of time between two consecutive transitions into ordered and disordered states based on which distribution the time between two transitions is most consistent with . From these states , a holistic communication IHX , Y is computed for every pair of dihedrals X and Y:IH ( X , Y ) =Iss ( X , Y ) ¯+Isd ( X , Y ) ¯+Ids ( X , Y ) ¯+ Idd ( X , Y ) ¯where Iss ( X , Y ) ¯ is the normalized mutual information between the structure ( i . e . , rotameric state ) of dihedral X and the structure of dihedral Y , Isd ( X , Y ) ¯ is the normalized mutual information between the structure of dihedral X and the dynamical state of dihedral Y , Ids ( X , Y ) ¯ is the normalized mutual information between the dynamical state of dihedral X and the structure of dihedral Y , and Idd ( X , Y ) ¯ is the normalized mutual information between the dynamical state of dihedral X and the dynamical state of dihedral Y . The Mutual Information ( I ) isIX , Y= ∑x∈X∑y∈Ypx , ylog⁡px , ypxpywhere x ∈ X refers to the set of possible states that dihedral X can adopt , p ( x ) is the probability that dihedral X adopts state x , and p ( x , y ) is the joint probability that dihedral X adopts state x and dihedral Y adopts state y . Normalized mutual information is computed using the maximum possible mutual information for any specific mode of communication . From the pairwise correlation for every dihedral-pair , we extracted how much each individual residue communicates with a target site of interest via bootstrapping with 10 random samples with replacement . After locating the group of residues communicating most strongly with a specific target site , we set this newly identified group as the new target site; The iteration of this process allows us to identify a pathway of communication from one region of interest to another . Here , we set the GPCR contact sites as our initial target sites . We then iteratively used this approach to identify pathways connecting these contact sites with the GDP-binding site of Gαq . We clustered Gαq conformations and GDP binding states separately and combined the assignments to build a Markov State Model using MSMbuilder ( Beauchamp et al . , 2011; Bowman et al . , 2009 ) and enspara ( Porter et al . , 2018 ) . First , we clustered protein conformations into 5040 states using a hybrid k-center/k-medoids method with 1 . 8 Å cutoff . Then we clustered the GDP-binding state into 321 states using the automatic partitioning algorithm ( APM ) ( Sheong et al . , 2015 ) with a residence time of 2 ns . By combining the assignments from protein conformations and the GDP-binding states , we obtained a total of 221 , 965 states . The implied timescales of this MSM show Markovian behavior with a lag time of 5 ns ( Swope et al . , 2004 ) . ( Figure 2—figure supplement 5 ) . Analyses of distances , angles , and dihedrals of interest were carried out using bootstrapping with ten random samples , with replacement . Results were insensitive to varying the number of bootstrapped samples between 5 and 30 . Histograms were generated using 100 bins . The disorder of every residue was measured by computing Shannon entropy ( Shannon , 1948 ) of each dihedral as they are natural degrees of freedom for describing protein dynamics . Shannon entropy ( H ) is defined asHX= -∑x∈Xpxlog⁡pxwhere x∈X refers to the set of possible states that dihedral X can adopt and px is the probability that dihedral X adopts state x . Dihedral angles were calculated using MDTraj ( McGibbon et al . , 2015 ) and assigned to discrete rotameric states as described above using CARDS . The entropy of a single residue was computed by summing up the entropies of its dihedrals , and normalized by the residue’s maximum possible Shannon entropy . This maximum possible Shannon entropy , using a flat distribution for the appropriate number of bins , is referred to as the 'channel capacity' and has been used to normalize other information-theoretic metrics ( Singh and Bowman , 2017 ) . Summing entropies within a residue establishes an upper bound on the degree of motion for a single residue , while ignoring intra-residue correlations between dihedrals . We used transition path theory ( TPT ) ( Noé et al . , 2009; Weinan and Vanden-Eijnden , 2006 ) to find the highest flux paths from the bound state to the unbound state ( Metzner et al . , 2009 ) . The bound state was defined as all clusters that satisfied two criteria: ( i ) GDP is within 6 Å of the backbone atoms of Lys52-Thr54 , and ( ii ) GDP has an RMSD <0 . 5 Å to its crystallographic conformation . The unbound state was defined as all clusters with GDP > 55 Å from the binding pocket . The rate-limiting step was identified by finding the bottleneck in the highest flux paths . To obtain this , we first calculate the flux between states i and j along any possible unbinding path usingfijBU={πiqi−Tijqj+i≠j0i=jwhere qi+ is the committor probability from the bound to the unbound state , and qi- is 1-qi+; πi is the weighted probability , and Tij is the transition matrix . The highest flux paths can be identified by maximizing the fluxes between the bound states and the unbound states usingcw=min⁡ ( filil+1+\|l=1 . . nw-1 ) where il are intermediate states . From this , the slowest step was extracted as the minimum flux step of the highest flux release pathway .	Cells communicate with each other by exchanging chemical signals , which allow them to coordinate their activities and relay important information about their environment . Often , cells secrete specific signals into their surroundings , which are then picked up by a receiving cell that has the right receptors to recognize the message . Once the signal attaches to the receptor , its shape or activity changes , which in turn triggers cascades inside the cell to convey the signal , much like a circuit would . A group of proteins called heterotrimeric G-proteins play an important role in these pathways . They act as molecular switches inside the cells to help transmit signals from the outside of the cell to the inside . The proteins are made up of three parts , one of which is G-alpha . When G-alpha receives a signal from its receptor , it becomes activated . To turn on , G-alpha needs to release a molecule called GDP – which is bound to G-alpha when turned off – and instead bind to another molecule called GTP . However , it remains unclear how exactly GDP is released when it receives a signal from its receptor . Faulty G-alphas have been linked to many diseases , including cancer and heart conditions . However , current treatments do not currently target this part of G-protein signaling . To develop new drugs in the future , we first need a better understanding about the critical steps driving G-alpha activation , such as the release of GDP . Now , Sun , Singh et al . used computer simulations and mathematical models to investigate how G-alpha is activated , and to identify the structural changes underlying the release of GDP . The simulations allow to observe how the atoms within G-alpha behave and were obtained from citizen-scientist volunteers , who ran simulations on their personal computers using the Folding@home app . Together , they generated an enormous amount of data that would normally take over 150 years to collect with one computer . Subsequent analyses identified the critical atomic motions driving the release of GDP and a network of amino acids located within G-alpha . These amino acids allow G-alpha to act like a switch and connect the part that receives the signal from the receptor to the GDP-binding site . In the future , this model could serve as a platform for developing drugs that target G-alpha and shed more light into how signals are transmitted within our cells .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "computational", "and", "systems", "biology" ]	2018	Simulation of spontaneous G protein activation reveals a new intermediate driving GDP unbinding
Animals rely on highly sensitive thermoreceptors to seek out optimal temperatures , but the molecular mechanisms of thermosensing are not well understood . The Dorsal Organ Cool Cells ( DOCCs ) of the Drosophila larva are a set of exceptionally thermosensitive neurons critical for larval cool avoidance . Here , we show that DOCC cool-sensing is mediated by Ionotropic Receptors ( IRs ) , a family of sensory receptors widely studied in invertebrate chemical sensing . We find that two IRs , IR21a and IR25a , are required to mediate DOCC responses to cooling and are required for cool avoidance behavior . Furthermore , we find that ectopic expression of IR21a can confer cool-responsiveness in an Ir25a-dependent manner , suggesting an instructive role for IR21a in thermosensing . Together , these data show that IR family receptors can function together to mediate thermosensation of exquisite sensitivity . Temperature is an omnipresent physical variable with a dramatic impact on all aspects of biochemistry and physiology ( Sengupta and Garrity , 2013 ) . To cope with the unavoidable spatial and temporal variations in temperature they encounter , animals rely on thermosensory systems of exceptional sensitivity . These systems are used to avoid harmful thermal extremes and to seek out and maintain body temperatures optimal for performance , survival and reproduction ( Barbagallo and Garrity , 2015; Flouris , 2011 ) . Among the most sensitive biological thermoreceptors described to date are the Dorsal Organ Cool Cells ( DOCCs ) , a recently discovered trio of cool-responsive neurons found in each of the two dorsal organs at the anterior of the Drosophila melanogaster larva ( Klein et al . , 2015 ) . The DOCCs robustly respond to decreases in temperature as small as a few millidegrees C per second ( Klein et al . , 2015 ) , a thermosensitivity similar to that of the rattlesnake pit organ ( Goris , 2011 ) , a structure known for its extraordinary sensitivity . A combination of laser ablation , calcium imaging and cell-specific inhibition studies was used to establish the DOCCs as critical for mediating larval avoidance of temperatures below ~20˚C , with the thermosensitivity of this avoidance behavior paralleling the thermosensitivity of DOCC physiology ( Klein et al . , 2015 ) . While the DOCCs are exceptionally responsive to temperature , the molecular mechanisms that underlie their thermosensitivity are unknown . At the molecular level , several classes of transmembrane receptors have been shown to participate in thermosensation in animals . The most extensively studied are the highly thermosensitive members of the Transient Receptor Potential ( TRP ) family of cation channels ( Palkar et al . , 2015; Vriens et al . , 2014 ) . These TRPs function as temperature-activated ion channels and mediate many aspects of thermosensing from fruit flies to humans ( Barbagallo and Garrity , 2015; Palkar et al . , 2015; Vriens et al . , 2014 ) . In addition to TRPs , other classes of channels contribute to thermosensation in vertebrates , including the thermosensitive calcium-activated chloride channel Anoctamin 1 ( Cho et al . , 2012 ) and the two pore domain potassium channel TREK-1 ( Alloui et al . , 2006 ) . Recent work in Drosophila has demonstrated that sensory receptors normally associated with other modalities , such as chemical sensing , can also make important contributions to thermotransduction . In particular , GR28B ( D ) , a member of the invertebrate gustatory receptor ( GR ) family , was shown to function as a warmth receptor to mediate warmth avoidance in adult flies exposed to a steep thermal gradient ( Ni et al . , 2013 ) . The photoreceptor Rhodopsin has also been reported to contribute to temperature responses , although its role in thermosensory neurons is unexamined ( Shen et al . , 2011 ) . Ionotropic Receptors ( IRs ) are a family of invertebrate receptors that have been widely studied in insect chemosensation , frequently serving as receptors for diverse acids and amines ( Benton et al . , 2009; Silbering et al . , 2011 ) . The IRs belong to the ionotropic glutamate receptor ( iGluR ) family , an evolutionarily conserved family of heterotetrameric cation channels that includes critical regulators of synaptic transmission like the NMDA and AMPA receptors ( Croset et al . , 2010 ) . In contrast to iGluRs , IRs have been found only in Protostomia and are implicated in sensory transduction rather than synaptic transmission ( Rytz et al . , 2013 ) . In insects , the IR family has undergone significant expansion and diversification , with the fruit fly D . melanogaster genome encoding 66 IRs ( Croset et al . , 2010 ) . While the detailed structures of IR complexes are unknown , at least some IRs are thought to form heteromeric channels in which a broadly-expressed IR 'co-receptor' ( such as IR25a , IR8a or IR76b ) partners one or more selectively-expressed 'stimulus-specific' IRs ( Abuin et al . , 2011 ) . Among insect IRs , IR25a is the most highly conserved across species ( Croset et al . , 2010 ) . In Drosophila , IR25a expression has been observed in multiple classes of chemosensory neurons with diverse chemical specificities , and IR25a has been shown to function as a 'co-receptor' that forms chemoreceptors of diverse specificities in combination with other , stimulus-specific IRs ( Abuin et al . , 2011; Rytz et al . , 2013 ) . IR21a is conserved in mosquitoes and other insects , but has not been associated with a specific chemoreceptor function ( Silbering et al . , 2011 ) , raising the possibility that it may contribute to other sensory modalities . Here , we show that the previously 'orphan' IR , Ir21a , acts together with the co-receptor IR25a to mediate thermotransduction . We show that these receptors are required for larval cool avoidance behavior as well as the physiological responsiveness of the DOCC thermosensory neurons to cooling . Furthermore , we find that ectopic expression of IR21a can confer cool responsiveness in an Ir25a-dependent manner , indicating that IR21a can influence thermotransduction in an instructive fashion . To identify potential regulators of DOCC thermosensitivity , we sought sensory receptors specifically expressed in the dorsal organ housing these thermoreceptors ( Figure 1a ) . Examining a range of potential sensory receptors in the larva , we found that regulatory sequences from the Ionotropic Receptor Ir21a drove robust gene expression ( via the Gal4/UAS system [Brand and Perrimon , 1993] ) in a subset of neurons in the dorsal organ ganglion , as well as in other locations ( Figure 1b , c , Figure 1—figure supplement 1 ) . Within each dorsal organ ganglion , Ir21a-Gal4 drove gene expression in three neurons ( Figure 1b , c ) . These neurons exhibited the characteristic morphology of the DOCCs , which have unusual sensory processes that form a characteristic 'dendritic bulb' inside the larva ( Klein et al . , 2015 ) . 10 . 7554/eLife . 13254 . 003Figure 1 . Dorsal Organ Cool Cells ( DOCCs ) express Ir21a-Gal4 . ( a ) First/second instar larval anterior . Each Dorsal Organ Ganglion ( grey ) contains three DOCCs ( blue ) . Anterior-Posterior axis denoted by double-headed arrow . ( b , c ) Ir21a-Gal4;UAS-GFP ( Ir21a>GFP ) labels larval DOCCs . Arrows denote cell bodies and arrowheads dendritic bulbs . ( d ) Temperature responses of Ir21a-Gal4;UAS-GCaMP6m-labeled DOCCs . Left panels , raw images; right panels , colors reflect fluorescence intensity . Arrows denote cell bodies . ( e ) Fluorescence quantified as percent change in fluorescence intensity compared to minimum intensity . n=22 cells ( from 6 animals ) . ( f , g ) Temperature-responses of Ir21a-Gal4;R11F02-Gal4;UAS-GCaMP6m-labeled DOCCs . n=26 ( 7 ) . Scale bars , 10 microns . Traces , average +/- SEM . Figure 1—figure supplement 1 provides an example of the 3-D imaging stacks used for calcium imaging data acquisition . DOI: http://dx . doi . org/10 . 7554/eLife . 13254 . 00310 . 7554/eLife . 13254 . 004Figure 1—figure supplement 1 . Larval-wide expression patterns of Ir21a-Gal4 and R11F02-Gal4 . ( a ) Ir21a-Gal4; UAS-GFP expression . ( b ) R11F02-Gal4; UAS-GFP expression . In addition to expression in the Dorsal Organ , both Gal4s exhibit expression in ~100 cells in the brain and ventral ganglion , neurons along the larval body wall and in the tail . R11F02-Gal4 is also expressed by sensory neurons in the Terminal Organ . DOI: http://dx . doi . org/10 . 7554/eLife . 13254 . 00410 . 7554/eLife . 13254 . 005Figure 1—figure supplement 2 . Calcium-imaging data are obtained as a three-dimensional imaging stack . ( a ) Dimensions of imaging volume . DOCCs depicted in blue . ( b ) Maximum intensity projections used for visualizing fluorescence intensity . ( c ) Representative image of maximum intensity projections of Ir21a>GCaMP6m-labeled DOCCs . DOCC cell bodies remain within imaging field throughout . DOI: http://dx . doi . org/10 . 7554/eLife . 13254 . 005 To confirm that the Ir21a-Gal4-positive neurons were indeed cool-responsive , their thermosensitivity was tested by cell-specific expression of the genetically encoded calcium indicator GCaMP6m under Ir21a-Gal4 control . Consistent with previously characterized DOCC responses ( Klein et al . , 2015 ) , when exposed to a sinusoidal temperature stimulus between ~14˚C and ~20˚C , GCaMP6m fluorescence in these neurons increased upon cooling and decreased upon warming ( Figure 1d , e and Figure 1—figure supplement 2 ) . The expression of Ir21a-Gal4 was also compared with that of R11F02-Gal4 ( Figure 1—figure supplement 1 ) , a promoter used in the initial characterization of the DOCCs ( Klein et al . , 2015 ) . As expected , GCaMP6m expressed under the combined control of Ir21a-Gal4 and R11F02-Gal4 revealed their precise overlap in three cool-responsive neurons with DOCC morphology in the dorsal organ , further confirming the identification of the Ir21a-Gal4-expressing cells as the cool-responsive DOCCs ( Figure 1f , g ) . To assess the potential importance of Ir21a in larval thermosensation , we tested the ability of animals to thermotax when Ir21a function has been eliminated . Two Ir21a alleles were generated , Ir21a123 and Ir21a∆1 . Ir21a123 deletes 23 nucleotides in the region encoding the first transmembrane domain of IR21a and creates a translational frameshift ( Figure 2a ) . Ir21a∆1 is an ~11 kb deletion removing all except the last 192 nucleotides of the Ir21a open reading frame , including all transmembrane and ion pore sequences ( Figure 2a ) . As the deletion in Ir21a∆1 could also disrupt the nearby chitin deacetylase 5 ( cda5 ) gene ( Figure 2—figure supplement 1 ) , Ir21a-specific rescue experiments were performed to confirm all defects reflected the loss of Ir21a activity ( see below ) . 10 . 7554/eLife . 13254 . 006Figure 2 . Larval cool avoidance requires Ir21a and Ir25a . ( a ) Sequence alterations in Ir21a and Ir25a alleles . Ir21a regulatory sequences present in Ir21a-Gal4 are denoted in green and regions encoding transmembrane domains ( TMs ) and pore region in red . Additional details provided in Figure 2—figure supplement 1 . ( b ) Thermotaxis is quantified as navigational bias . Cool avoidance behavior was assessed by tracking larval trajectories on a ~0 . 36˚C/cm gradient extending from ~13 . 5˚C to ~21 . 5˚C , with a midpoint of ~17 . 5˚C . ( c ) Cool avoidance requires Ir21a and Ir25a . Ir21a>Ir21a denotes a wild type Ir21a transcript expressed under Ir21a-Gal4 control . {Ir21a+} and {Ir25a+} denote wild type genomic rescue transgenes . Letters denote statistically distinct categories ( alpha=0 . 05; Tukey HSD ) . wild type , n=836 animals . Ir21a∆1 , n=74 . Ir21a∆1;Ir21a-Gal4 , n=48 . Ir21a∆1;UAS-Ir21a , n=10 . Ir21a∆1;Ir21a>Ir21a , n= 88 . Ir21a∆1/ Ir21a123 , n=71; Ir21a∆1/ Ir21a123; {Ir21a+} n=70; Ir25a2 , n =100 . Ir25a2; {Ir25a+} n= 247 . Additional mutant analyses provided in Figure 2—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13254 . 00610 . 7554/eLife . 13254 . 007Figure 2—figure supplement 1 . Structure of Ir21a locus and analysis of thermotaxis in Ir8a and Ir76b mutants . ( a ) Cda5/Ir21a genomic region , denoting positions of the FRT-containing transposon insertions used to generate Ir21a∆1 ( PBc04017 and PBc02720 ) , the sequences deleted in Ir21a∆1 , the Ir21a sequences present in the UAS-Ir21a rescue construct and the sequences present in the {Ir21+} genomic rescue construct . Untranslated regions are in gray . ( b ) Larval thermotaxis of Ir8a and Ir76b mutants quantified as navigational bias . Neither Ir8a nor Ir76b is required for cool avoidance; Ir8a mutants show enhanced cool avoidance compared to wild type . Letters denote statistically distinct categories ( alpha=0 . 05; Tukey HSD ) . wild type , n=836 animals . Ir8a , n=166; Ir76b1 , n=96 , Ir76b2 , n= 100 . DOI: http://dx . doi . org/10 . 7554/eLife . 13254 . 007 The loss of Ir21a function strongly disrupted larval thermotaxis . When exposed to a thermal gradient of ~0 . 36˚C/cm , ranging from ~13 . 5˚C to ~21 . 5˚C , Ir21a∆1 null mutants as well as Ir21a123/ Ir21a∆1 heterozygotes were unable to navigate away from cooler temperatures and toward warmer temperatures ( Figure 2b , c ) . These defects could be rescued by expression of a wild-type Ir21a transcript under Ir21a-Gal4 control and by a wild-type Ir21a genomic transgene ( Figure 2c ) . Taken together , these results are consistent with a critical role for Ir21a in larval thermotaxis . As IRs commonly act in conjunction with 'co-receptor' IRs , we examined the possibility that larval thermotaxis involved such additional IRs . Animals homozygous for loss-of-function mutations in two previously reported IR co-receptors , Ir8a and Ir76b , exhibited robust avoidance of cool temperatures , indicating that these receptors are not essential for this behavior ( Figure 2—figure supplement 1 ) . By contrast , Ir25a2 null mutants failed to avoid cool temperatures , a defect that could be rescued by the introduction of a transgene containing a wild type copy of Ir25a ( Figure 2c ) . Thus , Ir25a also participates in cool avoidance . To assess IR25a expression , larvae were stained with antisera for IR25a . Robust IR25a protein expression was detected in multiple cells in the dorsal organ ganglion , including the three Ir21a-Gal4-expressing DOCCs ( Figure 3a ) . Within DOCCs , IR25a strongly labels the 'dendritic bulbs' , consistent with a role in sensory transduction . Staining was absent in Ir25a null mutants demonstrating staining specificity ( Figure 3b ) . Thus Ir25a is required for thermotaxis and is expressed in the neurons that drive this behavior . 10 . 7554/eLife . 13254 . 008Figure 3 . DOCCs express IR25a . ( a ) Left panel , Ir21a>GFP-labeled DOCCs . Middle panel , IR25a protein expression in dorsal organ . Right panel , Ir21a>GFP-labeled DOCCs express IR25a protein . Arrows denote DOCC cell bodies and arrowheads DOCC dendritic bulbs . ( b ) IR25a immunostaining is not detected in Ir25a2 null mutants . Scale bar , 10 microns . DOI: http://dx . doi . org/10 . 7554/eLife . 13254 . 008 To assess whether Ir21a and Ir25a contribute to cool detection by the DOCCs , DOCC cool-responsiveness was examined using the genetically encoded calcium sensor GCaMP6m . Consistent with a role for Ir21a in cool responses , DOCCs exhibited strongly reduced responses to cooling in Ir21a∆1 deletion mutants , and this defect was robustly rescued by expression of an Ir21a transcript in the DOCCs using R11F02-Gal4 ( Figure 4a-e , h ) . Similarly , DOCC thermosensory responses were greatly reduced in Ir25a mutants , a defect that was rescued by a wild type Ir25a transgene ( Figure 4f–h ) . Together these data demonstrate a critical role for Ir21a and Ir25a in the detection of cooling by the DOCCs . 10 . 7554/eLife . 13254 . 009Figure 4 . DOCC cool responses require Ir21a and Ir25a . DOCC responses monitored using R11F02>GCaMP6m . DOCCs exhibit robust cool-responsive increases in fluorescence ( a , c ) , which are dramatically reduced in Ir21a ( b , d ) and Ir25a ( f ) mutants . ( e ) Ir21a transcript expression under R11F02-Gal4 control rescues the Ir21a mutant defect . ( g ) Introduction of an Ir25a genomic rescue transgene rescues the Ir25a mutant defect . ( h ) Ratio of fluorescence at 14˚C versus 20˚C depicted using a violin plot . Letters denote statistically distinct categories , p<0 . 0001 , Steel-Dwass test . Scale bars , 10 microns . Traces , average +/- SEM . wild type , n=33 cells ( from 11 animals ) . Ir21a∆1 , n= 58 ( 14 ) . Ir21a∆1; R11F02>Ir21a , n=32 ( 9 ) . Ir25a2 , n=43 ( 16 ) . Ir25a2; {Ir25a+} , n=30 ( 10 ) . Analyses of brv1 and brv2 mutants provided in Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13254 . 00910 . 7554/eLife . 13254 . 010Figure 4—figure supplement 1 . Analysis of putative null mutants of brv1 and brv2 . ( a ) brv1 but not brv2 mutants exhibit defects in larval cool avoidance . Thermotaxis quantified as navigational bias . Letters denote statistically distinct categories ( alpha=0 . 05; Tukey HSD ) . wild type , n=836 animals . brv1L653stop , n =43 . brv2W205stop , n =99 . b ) Ir21a>GCaMP6m-labelled DOCCs respond to cooling in brv1L653stop mutants . n= 35 cells ( from 6 animals ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13254 . 010 Prior work has suggested that three TRP channels , Brivido-1 , Brivido-2 and Brivido-3 , work together to mediate cool sensing in adult thermosensors ( Gallio et al . , 2011 ) . Putative null mutations are available for two of these genes , brv1 and brv2 , and we used these alleles to test the potential role of Brivido function in DOCC cool sensing ( Gallio et al . , 2011 ) . Although brv1 mutant showed defects in thermotactic behavior , DOCC responses to cooling appeared unaffected in brv1 mutants ( Figure 4—figure supplement 1 ) . brv2 nulls exhibited no detectable thermotaxis defects ( Figure 4—figure supplement 1 ) . Thus , we detect no role for these receptors in cool sensing by the DOCCs . The requirement for Ir21a and Ir25a in DOCC-mediated cool sensing raised the question of whether ectopic expression of these receptors could confer cool-responsiveness upon a cell , as might be predicted for a cool receptor . Attempts to express IR21a and IR25a together or separately in heterologous cells , including S2 cells , Xenopus oocytes and HEK cells , failed to yield detectable responses to cooling or warming , as did attempts to confer thermosensitivity upon non-thermosensitive neurons by ectopically expressing them separately or together in Drosophila , broadly throughout the larval nervous system and in adult chemosensory neurons ( G . B . , L . N . , M . K . and P . G , unpublished ) . However , ectopic expression of IR21a in one set of neurons in the adult , Hot Cell thermoreceptors in the arista that normally respond to warming rather than cooling , conferred cool-sensitivity . The adult arista contains three warmth-activated thermosensory neurons , termed Hot Cells ( or HC neurons ) ( Gallio et al . , 2011 ) . We found that forced expression of IR21a in the HC neurons could significantly alter their response to temperature . As previously reported ( Gallio et al . , 2011 ) , wild-type HC neurons respond to warming with robust increases in intracellular calcium and to cooling with decreases in intracellular calcium , as reflected in temperature-dependent changes in GCaMP6m fluorescence ( Figure 5a , c ) . In contrast , HC neurons in which IR21a is expressed under the control of a pan-neuronal promoter ( N-syb>Ir21a animals ) frequently exhibited elevations in calcium not only in response to warming , but also at the coolest temperatures ( Figure 5b , d , f , Figure 5—figure supplement 1a ) . Thus , ectopic IR21a expression causes HC neurons , which are normally inhibited by cooling , to become responsive to both cooling and warming . 10 . 7554/eLife . 13254 . 011Figure 5 . IR21a expression confers cool-sensitivity upon warmth-responsive Hot Cell neurons . ( a , b ) Temperature responses of wild type ( a ) or N-syb>Ir21a-expressing ( b ) thermoreceptors in the arista , monitored with N-syb>GCaMP6m . Cell bodies of warmth-responsive Hot Cells outlined in red and cool-responsive Cold Cells in blue . Arrows highlight Hot Cells at 14˚C . Traces of Hot Cell and Cold Cell responses shown at right . Scale bar , 10 microns . ( c-e ) Fluorescence of Hot Cells in response to sinusoidal 14˚C to 30˚C temperature stimulus , quantified as percent ∆F/Fmin . Dotted lines denote temperature minima . Traces , average +/- SEM . ( f ) Difference between ∆F/Fmin at 14˚C vs 20˚C ( average +/- SEM ) . Responses of N-syb>Ir21a cells were statistically distinct from both wild type and Ir25a2;N-syb>Ir21a ( p<0 . 01 , Steel-Dwass test; letters denote statistically distinct groups ) . wild type , n= 16 cells ( from 8 animals ) . N-syb>Ir21a , n= 16 ( 10 ) . Ir25a2; N-syb>Ir21a , n= 20 ( 10 ) . Analysis of endogenous IR25a expression in the Hot Cells and of the consequences of Hot Cell-specific misexpression of IR21a provided in Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13254 . 01110 . 7554/eLife . 13254 . 012Figure 5—figure supplement 1 . Hot Cell neurons express IR25a protein , and IR21a confers cool-sensitivity upon the Hot Cell neurons . ( a ) Difference between ∆F/Fmin at 14˚C vs 20˚C for each cell imaged in Figure 5 . ( b , c ) Left panel , HC>GFP-labeled Hot Cell neurons . Middle panel , IR25a immunostaining . Right panel , HC>GFP and IR25a co-expression . Arrows indicate Hot Cell neuron cell bodies . Specific IR25a immunostaining is absent in Ir25a null mutants ( b ) . ( d , e ) Temperature responses of wild type ( d ) , HC>Ir21a-expressing ( e ) thermoreceptors in the adult arista , monitored using HC>GCaMP6m . Dotted lines denote temperature minima . Traces , average +/- SEM . wild type , n=4 cells ( 2 animals ) . HC>IR21a n=25 ( 9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13254 . 012 As Ir21a-dependent cool detection in the DOCCs relies upon Ir25a , we examined the requirement for Ir25a in IR21a-mediated cool activation of the HC neurons . Consistent with previously reported IR25a expression in the arista ( Benton et al . , 2009 ) , we observed robust IR25a protein expression in the HC neurons ( Figure 5—figure supplement 1b , c ) . Consistent with a role for Ir25a in Ir21a-mediated cool-responsiveness , ectopic IR21a expression failed to drive significant HC neuron cool responses in Ir25a mutants ( Figure 5e , f ) . Thus , IR21a can confer cool-sensitivity upon an otherwise warmth-responsive neuron in an Ir25a-dependent fashion . Similar cool sensitivity was observed when IR21a was ectopically expressed under the control of an HC-specific promoter ( HC>Ir21a , Figure 5—figure supplement 1d , e ) . Finally , ectopic expression of IR21a in Gr28b mutant HC neurons , which lack the Gr28b ( D ) warmth receptor , yields neurons that respond only to cooling ( Figure 6 ) . Together , these data demonstrate that ectopic IR21a expression can confer cool-sensitivity in an Ir25a-dependent fashion . 10 . 7554/eLife . 13254 . 013Figure 6 . Hot Cell-specific expression of IR21a confers cool-sensitivity upon Gr28b mutant Hot Cell neurons . ( a-c ) Temperature responses of wild type ( a ) , Gr28b mutant ( b ) , and HC>Ir21a-expressing Gr28b mutant ( c ) thermoreceptors in the adult arista , monitored using HC>GCaMP6m . Dotted lines denote temperature minima . Traces , mean +/- SEM . wild type , n=11 cells ( 3 animals ) . Gr28bMi n=9 ( 3 ) . HC>IR21a; Gr28bMi n=11 ( 3 ) . ( d ) Cool responses ( ∆F/F14˚C - ∆F/F30˚C ) of HC>IR21a; Gr28bMicells were distinct from both wild type and Gr28bMi ( p<0 . 01 , Steel-Dwass test , letters denote statistically distinct groups ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13254 . 013 These data demonstrate that the ionotropic receptors IR21a and IR25a have critical roles in thermosensation in Drosophila , mediating cool detection by the larval dorsal organ cool cells ( DOCCs ) and the avoidance of cool temperatures . Combinations of IRs have been previously found to contribute to a wide range of chemosensory responses , including the detection of acids and amines ( Rytz et al . , 2013 ) . These findings extend the range of sensory stimuli mediated by these receptor combinations to cool temperatures . Interestingly , IR21a- and IR25a-dependent cool sensation appears independent of Brivido 1 and Brivido 2 , two TRP channels implicated in cool sensing in the adult ( Gallio et al . , 2011 ) . The precise nature of the molecular complexes that IRs form is not well understood . IR25a has been shown to act with other IRs in the formation of chemoreceptors , potentially as heteromers ( Rytz et al . , 2013 ) . This precedent raises the appealing possibility that IR25a might form heteromeric thermoreceptors in combination with IR21a . However , our inability to readily reconstitute temperature-responsive receptor complexes in heterologous cells suggests that the mechanism by which these receptors contribute to cool responsiveness is likely to involve additional molecular cofactors . It is interesting to note that the range of cell types in which ectopic IR21a expression confers cool-sensitivity is so far restricted to neurons that already respond to temperature . This observation suggests the existence of additional co-factors or structures in these thermosensory cells that are critical for IR21a and IR25a to mediate responses to temperature . All studies to date implicate IRs as receptors for sensory stimuli ( Rytz et al . , 2013 ) , and our misexpression studies are consistent with a similar role for Ir21a and IR25a in cool sensation . However , we cannot formally exclude the possibility that they could have indirect , and possibly separate , functions in this process , for example , in regulating the expression or function of an unidentified cool receptor . Interestingly , IR25a was recently implicated in warmth-responsive resetting of the circadian clock , and suggested to confer warmth-sensitivity on its own , without the co-expression of other IRs ( Chen et al . , 2015 ) . The ability of IR25a to serve as a warmth receptor on its own would be a surprise given both its broad expression and its established role as an IR co-receptor ( Abuin et al . , 2011 ) . As IR25a misexpression only slightly enhanced the thermosensitivity of an already warmth-responsive neuron ( Chen et al . , 2015 ) , this raises the alternative possibility that – analogous to cool-sensing – IR25a acts not on its own , but rather as a co-receptor with other IRs involved in warmth-sensing . While the present study focuses on the role of IR21a and IR25a in larval thermosensation , it is interesting to note that the expression of both IR21a and IR25a has been detected in the thermoreceptors of the adult arista ( Benton et al . , 2009 ) . Thus , related mechanisms could contribute to thermosensory responses not only in the DOCCs , but also in other cellular contexts and life stages . Moreover , the presence of orthologs of IR21a and IR25a across a range of insects ( Croset et al . , 2010 ) raises the possibility that these IRs , along other members of the IR family , constitute a family of deeply-conserved thermosensors . Ir25a2 ( Benton et al . , 2009> ) , BAC{Ir25a+} ( Chen et al . , 2015 ) , Ir8a1 ( Abuin et al . , 2011 ) , Ir76b1 ( Zhang et al . , 2013 ) , Ir76b2 ( Zhang et al . , 2013 ) , R11F02-Gal4 ( Klein et al . , 2015 ) , brv1L653stop ( Gallio et al . , 2011 ) , brv2w205stop ( Gallio et al . , 2011 ) , HC-Gal4 ( Gallio et al . , 2011 ) , Gr28bMi ( Ni et al . , 2013 ) , UAS-GCaMP6m ( P{20XUAS-IVS-GCaMP6m}attp2 and P{20XUAS-IVS-GCaMP6m}attp2attP40 [Chen et al . , 2013] ) , UAS-GFP ( p{10X UAS-IVS-Syn21-GFP-p10}attP2 [Pfeiffer et al . , 2012] ) , nSyb-Gal4 ( P{GMR57c10-Gal4}attP2 , [Pfeiffer et al . , 2012] ) , and y1 P ( act5c-cas9 , w+ ) M ( 3xP3-RFP . attP ) ZH-2A w* ( Port et al . , 2014 ) were previously described . In Ir21a-Gal4 , sequences from -606 to +978 with respect to the Ir21a translational start site ( chromosome 2L: 24 , 173 – 25757 , reverse complement ) lie upstream of Gal4 protein-coding sequences . UAS-Ir21a contains the Ir21a primary transcript including introns ( chromosome 2L: 21823–25155 , reverse complement ) placed under UAS control . The {Ir21a+} genomic rescue construct contains sequences from -1002 to +4439 with respect to the Ir21a translational start site ( chromosome 2L: 26153–20712 ) . Ir21a∆1 was generated by FLP-mediated recombination between two FRT-containing transposon insertions ( PBac{PB}c02720 and PBac{PB}c04017 ) as described ( Parks et al . , 2004 ) . Ir21a123 was generated by transgene-based CRISPR-mediated genome engineering as described ( Port et al . , 2014 ) , with an Ir21a-targeting gRNA ( 5’-CTGATTTGCGTTTACCTCGG ) expressed under U6-3 promoter control ( dU6-3:gRNA ) in the presence of act-cas9 ( Port et al . , 2014 ) . Thermotaxis of early 2nd instar larvae was assessed over a 15 min period on a temperature gradient extending from 13 . 5 to 21 . 5°C over 22 cm ( ~0 . 36˚C/cm ) as described ( Klein et al . , 2015 ) . As behavioral data appear normally distributed ( as assessed by Shapiro-Wilk test ) , statistical comparisons were performed by Tukey HSD test , which corrects for multiple comparisons . Calcium imaging was performed as previously described for larvae ( Klein et al . , 2015 ) . Pseudocolor images were created using the 16_colors lookup table in ImageJ 1 . 43r . Adult calcium imaging was performed as described for larvae ( Klein et al . , 2015 ) , with modifications to the temperature stimulus and sample preparation approach . Adult temperature stimulus ranged from 14°C to 30°C . Intact adult antennae with aristae attached were dissected and placed in fly saline ( 110 mM NaCl , 5 . 4 mM KCl , 1 . 9 mM CaCl2 , 20 mM NaHCO3 , 15 mM tris ( hydroxymethyl ) aminomethane ( Tris ) , 13 . 9 mM glucose , 73 . 7 mM sucrose , and 23 mM fructose , pH 7 . 2 , [Brotz and Borst , 1996] ) on a large cover slip ( 24 mm x 50 mm ) and then covered by a small cover slip ( 18 mm x 18 mm ) . The large cover slip was placed on top of a drop of glycerol on the temperature control stage . As quantified calcium imaging data ( Figure 4h , Figure 5f , Figure 6d ) did not conform to a normal distribution as assessed by Shapiro-Wilk test ( p<0 . 01 ) , statistical comparisons were performed by Steel-Dwass test , a non-parametric test that corrects for multiple comparisons , using JMP11 ( SAS ) . Immunostaining was performed as described ( Kang et al . , 2012 ) using rabbit anti-Ir25a ( 1:100; [Benton et al . , 2009] ) , mouse anti-GFP ( 1:200; Roche ) , goat anti-rabbit Cy3 ( 1:100; Jackson ImmunoResearch ) , donkey anti-mouse FITC ( 1:100; Jackson ImmunoResearch ) .	Animals need to be able to sense temperatures for a number of reasons . For example , this ability allows animals to avoid conditions that are either too hot or too cold , and to maintain an optimal body temperature . Most animals detect temperature via nerve cells called thermoreceptors . These sensors are often extremely sensitive and some can even detect changes in temperature of just a few thousandths of a degree per second . However , it is not clear how thermoreceptors detect temperature with such sensitivity , and many of the key molecules involved in this ability are unknown . In 2015 , researchers discovered a class of highly sensitive nerve cells that allow fruit fly larvae to navigate away from unfavorably cool temperatures . Now , Ni , Klein et al . – who include some of the researchers involved in the 2015 work – have determined that these nerves use a combination of two receptors to detect cooling . Unexpectedly , these two receptors – Ionotropic Receptors called IR21a and IR25a – had previously been implicated in the detection of chemicals rather than temperature . IR25a was well-known to combine with other related receptors to detect an array of tastes and smells , while IR21a was thought to act in a similar way but had not been associated with detecting any specific chemicals . These findings demonstrate that the combination of IR21a and IR25a detects temperature instead . Together , these findings reveal a new molecular mechanism that underlies an animal’s ability to sense temperature . These findings also raise the possibility that other “orphan” Ionotropic Receptors , which have not been shown to detect any specific chemicals , might actually contribute to sensing temperature instead . Further work will explore this possibility and attempt to uncover precisely how IR21a and IR25a work to detect cool temperatures .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	The Ionotropic Receptors IR21a and IR25a mediate cool sensing in Drosophila
Sensory deprivation during the post-natal ‘critical period’ leads to structural reorganization of the developing visual cortex . In adulthood , the visual cortex retains some flexibility and adapts to sensory deprivation . Here we show that short-term ( 2 hr ) monocular deprivation in adult humans boosts the BOLD response to the deprived eye , changing ocular dominance of V1 vertices , consistent with homeostatic plasticity . The boost is strongest in V1 , present in V2 , V3 and V4 but absent in V3a and hMT+ . Assessment of spatial frequency tuning in V1 by a population Receptive-Field technique shows that deprivation primarily boosts high spatial frequencies , consistent with a primary involvement of the parvocellular pathway . Crucially , the V1 deprivation effect correlates across participants with the perceptual increase of the deprived eye dominance assessed with binocular rivalry , suggesting a common origin . Our results demonstrate that visual cortex , particularly the ventral pathway , retains a high potential for homeostatic plasticity in the human adult . To interact efficiently with the world , our brain needs to fine-tune its structure and function , adapting to a continuously changing external environment . This key property of the brain , called neuroplasticity , is most pronounced early in life , within the so called critical period , when abnormal experience can produce structural changes at the level of the primary sensory cortex ( Berardi et al . , 2000; Hubel and Wiesel , 1970; Hubel et al . , 1977; Wiesel and Hubel , 1963 ) . During development , occluding one eye for a few days induces a dramatic and permanent reorganization of ocular dominance columns ( the V1 territory representing each eye ) in favor of the open eye ( Berardi et al . , 2000; Gordon and Stryker , 1996; Hubel and Wiesel , 1970; Hubel et al . , 1977; Wiesel and Hubel , 1963 ) , while the deprived eye becomes functionally blind or very weak . These forms of structural plasticity have been documented in animal models , including non-human primates ( Gordon and Stryker , 1996; Kiorpes et al . , 1998; Levi and Carkeet , 1993; Wiesel and Hubel , 1963 ) . A corresponding perceptual phenomenon known as amblyopia is observed in humans , and may result from exposing infants to monocular deprivation during the critical period , for example due to cataracts ( Braddick and Atkinson , 2011; Maurer et al . , 2007 ) . In infants , even a partial deprivation produced by optical defects like astigmatism and myopia leads to a permanent acuity loss that cannot be compensated in adulthood , even after correction the optical aberrations ( Freeman and Thibos , 1975 ) through Adaptive Optics ( Rossi et al . , 2007 ) . Hebbian plasticity , endorsed by Long-Term synaptic Potentiation and Depression ( LTP/LTD ) of early stage of cortical processing , underlies these changes in animal models and probably also in humans . After the closure of the critical period , structural changes of V1 resulting from Hebbian plasticity are not typically observed ( Mitchell and Sengpiel , 2009; Sato and Stryker , 2008 ) . However , there is evidence that Hebbian plasticity can be restored in adult animal models under special conditions , associated with manipulation of the excitability of the visual cortex ( Fong et al . , 2016; Frégnac et al . , 1988; He et al . , 2006; Maya Vetencourt et al . , 2008 ) . Besides Hebbian plasticity , other mechanisms can reshape primary visual cortex processing both within and outside the critical period . At the cellular level , there is evidence for homeostatic plasticity , which increases the gain of cortical responses following sensory deprivation; for example , after a brief monocular deprivation , the response gain of the deprived eye increases ( Maffei et al . , 2004 ) . This is interpreted as an homeostatic response to preserve cortical excitability in spite of the synaptic depression produced by Hebbian plasticity , suggesting a close link between these two types of plasticity ( Maffei and Turrigiano , 2008; Turrigiano , 2012 ) ( Mrsic-Flogel et al . , 2007; Turrigiano and Nelson , 2004 ) . In adult animal models and humans , there is clear evidence for both functional plasticity and for stability of the early sensory cortex ( Baseler et al . , 2002; Baseler et al . , 2011; Wandell and Smirnakis , 2009 ) . Functional changes have been observed with perceptual learning ( Dosher and Lu , 2017; Fahle and Poggio , 2002; Fiorentini and Berardi , 1980; Karni and Sagi , 1991; Karni and Sagi , 1993; Watanabe and Sasaki , 2015 ) , adaptation that , in some cases , may be very long-lasting , ( McCollough , 1965 ) , and short-term visual deprivation ( Binda and Lunghi , 2017; Kwon et al . , 2009; Lunghi et al . , 2015a; Lunghi et al . , 2011; Lunghi et al . , 2013; Mon-Williams et al . , 1998; Zhang et al . , 2009; Zhou et al . , 2013; Zhou et al . , 2014 ) . The effect of short-term deprivation in adults is paradoxical , boosting the perception of the deprived stimulus – opposite to the long-term deprivation effects during development . One of the first examples of short-term deprivation in adults is by Mon-Williams et al . , 1998 , who found that thirty minutes of simulated myopia ( optical blur achieved by wearing a +1D lens ) was followed by a transient improvement of visual acuity – opposite to the long-lasting acuity deficit produced by early onset myopia ( Rossi et al . , 2007 ) . Contrast attenuation for 4 hr leads to improved contrast discrimination thresholds and enhanced BOLD response in V1/V2 ( Kwon et al . , 2009 ) . A few hours deprivation of one cardinal orientation leads to enhanced sensitivity to the deprived orientation ( Zhang et al . , 2009 ) – opposite to the reduced sensitivity to orientations deprived during development , for example due to astigmatism . Similarly , two hours of monocular contrast deprivation is followed by a transient boost of the deprived eye ( Binda and Lunghi , 2017; Lunghi et al . , 2015a; Lunghi et al . , 2011; Lunghi et al . , 2013; Lunghi et al . , 2015b; Zhou et al . , 2013; Zhou et al . , 2014 ) and an enlargement of the deprived-eye representation at the level of V1 in non-human primates ( Begum and Tso , 2016; Tso et al . , 2017 ) – opposite to the amblyopia induced by monocular deprivation during the critical period . The mechanism supporting the perceptual boost of the deprived information could be either a form of homeostatic plasticity ( like that observed in animal models ) , and/or a release of contrast adaptation for the deprived stimulus ( Blakemore and Campbell , 1969; Boynton et al . , 1999; Gardner et al . , 2005; Maffei et al . , 1973; Movshon and Lennie , 1979 ) . Irrespective of the interpretation , the data clearly indicate that effects can be long-lasting or even permanent . For example , in patients with keratoconus ( adult-onset corneal dystrophia , often monocular ) , best corrected visual acuity is worse than in emmetropic eyes , but it is better than predicted by the corneal dystrophy ( Sabesan and Yoon , 2009; Sabesan and Yoon , 2010 ) : when corneal aberrations of the keratoconic ( KC ) eyes are simulated in the emmetropic eyes , visual acuity is worse than in the KC eyes , demonstrating a permanent perceptual boost of the deprived information . Moreover , in adult amblyopes ( Lunghi et al . , 2018 ) , short-term monocular deprivation ( of the amblyopic eye ) may lead to permanent partial recovery of acuity ( of the amblyopic eye ) . This observation resonates with the idea – introduced in the context of work at the cellular level – that homeostatic plasticity and Hebbian plasticity may be fundamentally linked ( Maffei and Turrigiano , 2008 ) and may open important new pathways for the therapy of amblyopia and , in general , for the rehabilitation of early-onset visual dysfunctions ( Legge and Chung , 2016 ) . This possibility highlights the importance of understanding the neural substrates of short-term deprivation in adult humans . So far , monocular deprivation effects have been indirectly studied with MR spectroscopy ( showing a GABA concentration change in the occipital cortex , Lunghi et al . , 2015b ) and Visual Evoked Potentials ( showing a modulation of the early visual response components , Lunghi et al . , 2015a ) . Indirect evidence also indicates that deprivation effects are not generalized but preferentially involve the parvocellular pathway – given that effects are more prominent and longer-lasting for chromatic equiluminant stimuli in humans ( Lunghi et al . , 2013 ) , and strongest in macaques when deprivation mainly affects the parvocellular activity ( Begum and Tso , 2016 ) . Here we directly measure the changes in early visual cortical areas using 7T fMRI in adult humans , before and after two hours of monocular deprivation . Assessing the BOLD change and its selectivity to spatial frequency with a newly developed approach ( conceptually similar to the population Receptive Field method , Dumoulin and Wandell , 2008 ) , we demonstrate a change of ocular drive of BOLD signals in primary visual cortex , selective for the higher spatial frequencies and strongest along the ventral pathway , consistent with a stronger plasticity potential of the parvocellular pathway in adulthood . To investigate the visual modulation of BOLD signal by short term deprivation , we performed ultra-high field ( UHF , 7T ) fMRI during the presentation of high contrast dynamic visual stimuli , delivered separately to the two eyes , before and after 2 hr of monocular contrast deprivation ( see schematic diagram in Figure 1A ) . The reliability and high signal-to-noise ratio of our system allow us to obtain significant activations with only two blocks of stimulation ( Figure 1C shows the profile of V1 BOLD response ) , thereby targeting the first 10 min after deprivation , when the perceptual effects are strongest ( Lunghi et al . , 2011; Lunghi et al . , 2013 ) . As shown in Figure 1B , the stimulation was sufficient to reliably activate most early visual areas ( dashed lines outline ROIs limited by stimulus eccentricity , as detailed in the Materials and method ) . We measured the plasticity effect by comparing activity before/after deprivation in response to stimulation in the two eyes with low- and high-spatial frequency bandpass stimuli that differentially stimulate the magno- and parvocellular pathways ( see Figure 1—figure supplement 1 panels C-D for maps of responses to stimuli in both eyes , before and after deprivation ) . Consistent with prior evidence suggesting higher susceptibility to plasticity of the parvocellular pathway ( Lunghi et al . , 2015a; Lunghi et al . , 2011; Lunghi et al . , 2015b; Lunghi and Sale , 2015 ) , we observe a strong effect of Monocular Deprivation on BOLD responses to stimuli of high spatial frequency ( peak 2 . 7 cycles per degree , high-frequency cut-off at half-height 7 . 5 cpd ) . Figure 1D shows that the V1 response to the high spatial frequency stimuli presented in the left and right eye is nearly equal before deprivation ( ‘PRE’ ) ( see Figure 1—figure supplement 1 , panels C-D and Figure 1—figure supplement 2 , panel A , mapping the difference between responses to the two eyes ) . However , after deprivation ( ‘POST’ ) , the response in the two eyes changes in opposite directions , with a boost of the BOLD response ( measured as GLM Beta values , expressed in units of % signal change ) of the deprived eye and a suppression of the non-deprived eye ( see also Figure 1—figure supplement 2 , panel B ) . This was formally tested with a two-way repeated measure ANOVA , entered with the mean BOLD responses across all vertices in the left and right V1 region , for the four conditions and each participant ( Figure 1D show averages of this values across participants ) . The result reveals a significant interaction between the factors time ( PRE , POST deprivation ) and eye ( deprived , non-deprived; interaction term F ( 1 , 18 ) = 13 . 80703 , p = 0 . 00158; the result survives a split-half reliability test: see Figure 1—figure supplement 3 ) . Fig . 1E confirms these findings with an analysis of the aggregate subject data , obtained by pooling all V1 vertices across all subjects . For each vertex , we defined an index of Ocular Dominance computed as the difference of BOLD response to the deprived and non-deprived eye . This index is not to be confused with the anatomical arrangement of vertices with different eye preference that define the ocular dominance columns ( Cheng et al . , 2001; Yacoub et al . , 2007 ) , that cannot be directly imaged with voxel size of 1 . 5 mm . However , at this low resolution , each voxel is expected to average signals from a biased sample of ocular dominance columns leading to an eye preference of that particular voxel ( the Ocular Dominance index in Figure 1E ) . Before deprivation , the Ocular Dominance index is symmetrically distributed around zero , indicating a balanced representation of the two eyes before deprivation ( yellow distribution in Figure 1E ) . After deprivation ( black distribution in Figure 1E ) , the Ocular Dominance distribution shifts to the right of 0 , indicating a preference for the deprived eye ( non-parametric Wilcoxon sign-rank test comparing the PRE and POST Ocular Dominance medians , z = 115 . 39 , p < 0 . 001 ) . In principle , the boost of responses to the deprived eye seen in Figure 1D could be produced by enhancing the response of vertices that originally preferred the deprived eye ( without shifting ocular dominance ) or by changing Ocular Dominance of vertices that originally preferred the non-deprived eye , driving them to prefer the deprived eye . The shift of the Ocular Dominance histogram in Figure 1E is more compatible with the latter case , implying a recruitment of cortical resources for the representation of the deprived eye . To investigate this further , we monitored the final POST-deprivation Ocular Dominance of individual vertices that , PRE-deprivation , preferred the deprived eye ( yellow half distribution in Figure 2B ) . The majority of vertices continue to prefer the same eye before and after deprivation . The median Ocular Dominance is significantly larger than 0 both PRE and POST ( Wilcoxon sign-rank test , z > 101 . 54 , p < 0 . 0001 in both cases ) and the correlation between Ocular Dominance indices before and after deprivation is strong and positive ( Pearson’s R ( 32236 ) = 0 . 22 [0 . 21–0 . 23] , p < 0 . 0001 ) . Note that a completely random reassignment of Ocular Dominance after deprivation would have produced a histogram centered at 0 and no correlation between Ocular Dominance indices PRE- and POST deprivation . This is not consistent with the results of Figure 2B , which thereby provide evidence that our estimates of Ocular Dominance before and after deprivation are congruent , even though they were collected in different fMRI sessions separated by 2 hr . In addition , the distribution of Ocular Dominance after deprivation is well predicted by adding only a small amount of noise to the original half distribution ( Gaussian noise with 0 . 12 standard deviation , black line ) , suggesting that these vertices were largely unaffected by monocular deprivation . This is also supported by the repeated measure ANOVA of individual subject data ( Figure 2A ) , revealing a strong main effect of eye ( F ( 1 , 18 ) = 48 . 28901 , p < 10−5 ) : the response to the deprived eye is stronger than the non-deprived eye , both before deprivation ( due the selection , t ( 18 ) = −8 . 616 , p < 10−5 ) , and after deprivation ( t ( 18 ) = −4 . 281 , p < 10−5 ) , with no effect of time and no time × eye interaction ( all F ( 1 , 18 ) = 0 . 20429 , p > 0 . 5 ) . A completely different pattern is observed for the vertices originally preferring the non-deprived ( yellow half-distribution in Figure 2D ) . Here the distribution of Ocular Dominance clearly shifts after deprivation; the median moves from significantly negative before deprivation ( Wilcoxon sign-rank test , z = −175 . 97 , p < 0 . 0001 ) to significantly positive after deprivation ( Wilcoxon sign-rank test , z = 64 . 46 , p < 0 . 0001 ) , implying a shift of dominance in favor of the deprived eye . Again , this is not consistent with a random reassignment of Ocular Dominance after deprivation , which predicts a distribution centered at 0 . Contrary to Figure 2B , the POST- Ocular Dominance distribution cannot be predicted by injecting Gaussian noise to the PRE- Ocular Dominance distribution ( black line , 0 . 12 standard deviation like for Figure 2B ) : for these vertices , there is a shift of Ocular Dominance with short term monocular deprivation . This is confirmed with the repeated measure ANOVA ( Figure 2C ) , where the time × eye interaction is significant ( F ( 1 , 18 ) = 44 . 82812 , p < 10−5 ) , implying a different modulation PRE and POST deprivation . In addition and crucially , POST-deprivation BOLD responses to the deprived eye are significantly larger than POST-deprivation responses to the non-deprived eye ( t ( 18 ) = −2 . 775 p = 0 . 012; whereas , by selection , the opposite is true before deprivation: t ( 18 ) = 12 . 034 , p < 10−5 ) . In summary , Ocular Dominance before deprivation defines two similarly sized sub-regions of V1 vertices ( 44 . 58 ± 5 . 38% and 55 . 42 ± 5 . 38% of analyzed V1 vertices; 44 . 84 ± 5 . 12% and 55 . 16 ± 5 . 12% of all V1 vertices ) with radically different behaviors that are not consistent with an artifact induced by vertex selection . The sub-region that originally represents the deprived eye does not change with deprivation; the sub-region that originally represents the non-deprived eye is rearranged with deprivation , as a large portion of vertices turn to prefer the deprived eye . If plasticity were not eye-specific and/or we failed to match our V1 vertices before/after deprivation , we would expect that splitting the distribution of V1 ocular dominance generates opposite effects in the two subpopulations: vertices preferring the deprived eye before deprivation should swap to prefer the other eye , mirroring the effect seen in the vertices preferring non-deprived eye . This is not seen , implying that we did successfully match vertices across the 2 hr of deprivation and that the selective Ocular Dominance shift , observed for about half of our vertices , is not an artifact . We also measured the perceptual effects of short-term monocular deprivation effects using Binocular Rivalry , just before the PRE- and POST-deprivation fMRI sessions . In line with previous studies ( Binda and Lunghi , 2017; Lunghi et al . , 2015a; Lunghi et al . , 2011; Lunghi et al . , 2015b; Lunghi and Sale , 2015 ) , short-term monocular contrast deprivation induced a 30% increase of phase duration for the deprived eye ( POST to PRE-deprivation ratio: 1 . 31 ± 0 . 30 ) and a 15% decrease of phase duration for the non-deprived eye ( ratio: 0 . 86 ± 0 . 30 ) , producing a significant time × eye interaction ( Figure 3A , repeated measure ANOVA on the mean phase durations for each participant , interaction: F ( 1 , 18 ) = 23 . 56957 , p = 0 . 00013 ) . This effect size is similar to that measured in recent experiments using the same paradigm , but letting subjects continue normal activity during the 2 hr of monocular deprivation ( Lunghi et al . , 2011; Lunghi et al . , 2015b; Lunghi and Sale , 2015 ) . This indicates that the prolonged high contrast stimulation delivered for retinotopic mapping to the non-deprived eye during the first ~30 min of deprivation did not modulate the deprivation effects . We defined a psychophysical index of the deprivation effect ( DIpsycho ) by using Equation . 6 in Materials and methods section , where the POST to PRE-deprivation ratio of phase durations for the deprived eye , is divided by the same ratio for the non-deprived eye . Values larger than one imply a relative increase of the deprived eye phase duration , that is the expected effect; a value less than 1 indicates the opposite effect and a value of 1 indicates no change of mean phase duration across eyes . All but two subjects have values larger than 1 , indicating a strong effect of deprivation . However , the scatter is large with values ranging from 0 . 7 to 3 , suggesting that susceptibility to visual plasticity varies largely in our pool of participants . Capitalizing on this variability , we tested whether the size of the psychophysical effect correlates with the BOLD effect across participants . Using the same Equation 6 to compute the deprivation effect on BOLD responses ( DIBOLD ) , we observed a strong correlation between the effect of monocular deprivation on psychophysics and BOLD ( shown in Figure 3B ) . Subjects who showed a strong deprivation effect at psychophysics ( DIpsycho >2 ) also showed a strong deprivation effect in BOLD responses ( DIBOLD = 1 . 85 ± 0 . 42 ) . Given that the psychophysics was measured only for central vision and at two cpd stationary grating , whereas BOLD responses were pooled across a large portion on V1 and were elicited using broadband dynamic stimuli , the correlation suggests that the psychophysical effect may be used as a reliable proxy of a general change of cortical excitability , which can be measured by fMRI . We measured the effect over the main extra-striate visual cortical areas . The selective boost of the deprived eye response to the high spatial frequency is as strong in V2 as in V1 ( Figure 1—figure supplement 1 and Figure 7E ) . The boost is present also in V3 and V4 . In V4 the boost appears to be present also for lower spatial frequencies , but again only for the deprived eye ( Figure 7A–B ) , possibly reflecting the larger spatial frequency bandwidth of V4 neurons compared to V1 . The results are very different for dorsal area V3a ( Figure 7C–D ) and hMT+ ( Figure 7E–F ) , which do not show any significant change of responses in either eye at high spatial frequencies . Although the preferred response moves to lower spatial frequencies , consistent with a stronger input of the magnocellular pathway to the dorsal visual stream ( Henriksson et al . , 2008; Singh et al . , 2000 ) , the response to the highest spatial frequency stimulus is still strong and reliable in both V3a and hMT+ . Note that the reliable BOLD estimates of Figure 7 are computed after pooling vertices within the ROI and then averaging across subjects . However , the response of hMT+ evaluated at the individual vertex do not show significant activation ( Figure 1B ) , probably reflecting more variable organization of activity within this ROI across subjects ( Smith et al . , 2006 ) . Fig 7G quantifies the effect of short-term monocular deprivation ( using the ANOVA time x eye interaction term , which measures the eye-selective modulation of BOLD response after deprivation for the highest spatial frequency ) across the main visual areas . The plasticity effect is strongest in V1 , V2 and V3; it is still strong and significant in ventral area V4 ( t ( 18 ) = 2 . 41 p = 0 . 0270 ) , but it is absent in V3a and hMT+ , where the time x eye interaction is not significantly different from 0 ( t ( 18 ) = 0 . 52 p = 0 . 6115 and t ( 18 ) = −0 . 19 p = 0 . 8513 respectively ) . The plasticity effect in ventral area V4 is significantly stronger than in dorsal areas V3a and hMT+ ( t ( 18 ) = 2 . 39 , p = 0 . 0278 and t ( 18 ) = 2 . 36 , p = 0 . 0299 for V4-V3a and V4-hMT+ respectively ) . This result suggests a preferential involvement of the parvocellular vs . magnocellular pathway , leading to the differential plasticity effect in extra-striate visual areas of the ventral and dorsal pathway . Interestingly , the plasticity effect is robust in areas where the majority of cells are binocular ( like V3 and V4 ) , indicating that the effect does not require segregated representations of the two eyes ( e . g . Ocular Dominance columns ) . We demonstrate that two hours of abnormal visual experience has a profound impact on the neural sensitivity and selectivity of V1 . BOLD activity across the V1 cortical region paradoxically increases for the eye that was deprived of contrast vision , and decreases for the eye exposed to normal visual experience . The enhanced response to the deprived eye fits well with the concept of homeostatic plasticity , first observed in rodent visual cortex , both juvenile and adult ( Maffei et al . , 2004; Mrsic-Flogel et al . , 2007; Turrigiano and Nelson , 2004 ) , which is the tendency of neural circuits to keep the average firing rates constant in spite of anomalous stimulation ( Maffei and Turrigiano , 2008; Turrigiano , 2012 ) ( Mrsic-Flogel et al . , 2007; Turrigiano and Nelson , 2004 ) . More recently , similar observations have been made in the adult macaque V1 after two hours of monocular deprivation during anesthesia ( Begum and Tso , 2016; Tso et al . , 2017 ) . The post-deprivation gain boost observed in the monkey is consistent with our observations of an increased BOLD response to the deprived eye . We also observe an antagonistic suppression of the non-deprived eye BOLD response; together , the two effects lead to a shift of ocular preference of individual vertices in favor of the deprived eye . However , this effect is only observed in those V1 vertices that responded preferentially to the non-deprived eye before deprivation . No change of ocular preference is seen in vertices that already prefer the deprived eye before deprivation , which maintain their eye-preference after deprivation . This pattern of results cannot be explained by an overall gain increase; rather , it is consistent with the idea that the representation of the deprived eye recruits cortical resources ( which may or may not correspond to cortical territory ) , normally dedicated to the other eye . A similar antagonist effect on the two eyes ( boosting the deprived eye and suppressing the non-deprived eye ) was also observed in the VEP responses after short-term monocular deprivation ( Lunghi et al . , 2015a ) , and could be implemented through a modulation of the excitatory/inhibitory circuitry . Regulation of the excitation/inhibition balance through GABAergic signaling is considered to be a key factor for cortical plasticity , including homeostatic plasticity ( Maffei and Turrigiano , 2008 ) . Interestingly , the involvement of GABAergic signaling in the effect of short-term monocular deprivation is directly supported by MR Spectroscopy data in adult humans , showing that resting GABA in a large region of the occipital cortex is specifically reduced after short-term monocular deprivation ( Lunghi et al . , 2015b ) . The functional relevance of the BOLD changes we observe is demonstrated by their correlation with our behavioral assay of plasticity , obtained through binocular rivalry . This correlates both with the BOLD ocular dominance change ( relative boost/suppression of the deprived/non-deprived eye ) , and with the BOLD acuity change for the deprived eye ( change of spatial frequency tuning , assessed with our pRF-like modeling approach ) . The correlation holds despite binocular rivalry being restricted to foveal vision , whereas the assessment of BOLD plasticity is pooled across V1 ( including the mid-periphery ) . This implies that the change of binocular rivalry dynamics is a proxy for the more general plasticity effects that involves the whole primary visual cortex . This finding has long reaching implications , as it could validate the use of binocular rivalry as a biomarker of adult cortical plasticity , based on the neural mechanisms revealed by the present 7T fMRI results . Interestingly , the binocular rivalry phenomenon originates in the primary visual cortex – probably at the earliest stages – and is an expression of the dynamics of excitatory transmission and inhibitory feedback ( Tong et al . , 2006 ) ; as such it is a measure that could reflect the overall excitation-inhibition ratio ( van Loon et al . , 2013 ) , and its modulation in plasticity ( Lunghi et al . , 2015b; Maffei and Turrigiano , 2008 ) . Our data support the notion that V1 circuitry may be optimized by perceptual experience ( Fiorentini and Berardi , 1980 ) . They are also consistent with a large perceptual learning literature suggesting that associative cortical areas retain a high degree of flexibility ( Dosher and Lu , 2017; Dosher and Lu , 1999; Fuchs and Flügge , 2014; Harris et al . , 2012; Kahnt et al . , 2011; Karni et al . , 1995; Lewis et al . , 2009; Shibata et al . , 2012; Watanabe and Sasaki , 2015 ) . Although the monocular deprivations effects observed here are more robust in V1 , they are reliable in V2 and V3 as well . However , a clear difference emerges between extra-striate visual areas in the ventral and dorsal stream . While ventral area V4 shows a strong deprivation effect , area V3a , located at a similar tier in the dorsal stream , shows no BOLD change after short-term monocular deprivation . V4 is a primary target of the parvocellular system , which is best stimulated by our highest spatial frequency stimulus; V3a and hMT+ are preferential targets of the magnocellular system , which respond more strongly to our lower spatial frequency stimuli ( see Figure 7 ) . The different plasticity response of the ventral and dorsal stream , together with the selectivity for the high spatial frequencies of the V1 plasticity , suggests that the parvocellular pathway is most strongly affected by short-term plasticity . This is consistent with the finding in non-human primates that deprivation of the stimuli that optimally drive the parvocellular system is sufficient to produce a reliable plasticity effect ( Begum and Tso , 2016 ) . It is also consistent with the finding that the effect of short-term monocular deprivation is strongest and more long-lasting for chromatic equiluminant stimuli ( Lunghi et al . , 2013 ) . Other evidence shows that short-term deprivation may affect other properties of vision . In particular , selective deprivation of orientation ( Zhang et al . , 2009 ) or spatial frequency ( Zhou et al . , 2014 ) or color ( Zhou et al . , 2017 ) or even simply phase scrambling of the image in one eye ( Bai et al . , 2017 ) may lead to a boost of the deprived signal . These effects have been interpreted as a form of release of inhibition from the adapted signal ( Zhang et al . , 2009 ) – a concept that is not distant from homeostatic plasticity , where the network aims to keep overall activity constant . The conceptual border between adaptation and plasticity is fuzzy , given that some mechanisms are shared and both effects have the same outcomes . Be it adaptation or plasticity , the monocular deprivation mechanisms are probably cortical and affect mainly the deprived eye . There is evidence that the boost of the deprived eye is also observed when the two eyes receive equally strong stimulation , but perception of one eye stimulus is suppressed experimentally ( by the continuous flash suppression technique , Kim et al . , 2017 ) ; this result dismisses the retinal or thalamic contribution to the deprivation effect . Only in rare occasions does adaptation induce effects that last over days ( McCollough , 1965 ) , yet our recent work shows that deprivation effects of short-term monocular deprivation is retained across 6 hr sleep ( Menicucci et al . , 2018 ) , consistent with plasticity reinforcement during sleep ( Raven et al . , 2018; Timofeev and Chauvette , 2017 ) . Most importantly , in adult amblyopic patients , short-term monocular deprivation is able to induce improvement of visual acuity and stereovision ( Lunghi et al . , 2018 ) for up to one year . All this evidence supports the concept that homeostatic plasticity in the human adult cortex may be linked with or may promote more stable forms of Hebbian-like plasticity . This may endorse stable changes even in the adult brain , well after the closure of the critical period . Functional changes in associative cortex in adults have been demonstrated by short-term paired TMS studies ( Chao et al . , 2015 ) . Interestingly , the decay of this functional connectivity change has a similar time-course as the monocular deprivation effect , about one hour . Also , Hebbian changes at the single cell level can be observed in V1 of adult anaesthetized cat , following activity pairing over a similar time-scale ( from minutes to a few hours ) ( Frégnac et al . , 1988 ) . All these results demonstrate that V1 retains potential for Hebbian plasticity outside the critical period – although it may need particular conditions to exploit such potential . Understanding homeostatic plasticity and its relation to Hebbian plasticity may be fundamental to open the way to new approaches to treat brain dysfunction . Particularly important is Ocular Dominance plasticity in amblyopia ( Webber and Wood , 2005 ) , a cortical deficit still without cure in adults , although recent advancements in training procedures are opening new hopes ( Levi and Li , 2009; Sengpiel , 2014 ) . Endorsing plasticity may increase the effectiveness of these treatments and preliminary data from our laboratory suggest that monocular deprivation of the amblyopic eye may indeed boost sensitivity of the deprived eye and improve its acuity ( Lunghi et al . , 2018 ) – just like an acuity change is revealed by the present BOLD measurements in normally sighted participants . Our data demonstrate that two hours of abnormally unbalanced visual experience is sufficient to induce a functional reorganization of cortical circuits , particularly of the parvocellular pathway , leading to an alteration of basic visual perceptual abilities . Monocular deprivation was achieved by patching the dominant eye for 2 hr . The operational definition of dominant eye applied to the eye showing the longer phase durations during the baseline binocular rivalry measurements . Like in previous studies ( Binda and Lunghi , 2017; Lunghi et al . , 2011; Lunghi et al . , 2013 ) , we used a translucent eye-patch made of plastic material allowing light to reach the retina ( attenuation 0 . 07 logUnits , at least 3 times smaller than the threshold for discriminating a full-field luminance decrement ( Knau , 2000 ) and more than ten times smaller than the minimum photopic luminance decrement required for shifting the spatial ( Van Nes and Bouman , 1967 ) or temporal contrast sensitivity function ( Kelly , 1961 ) . The patch prevents pattern vision , as assessed by the Fourier transform of a natural world image seen through the eye-patch . During the 2 hr of monocular deprivation , observers were either engaged in the retinotopic mapping experiment ( about 30’ , described below ) or they were free to read and use a computer . Binocular rivalry was measured in two short sessions ( each comprising two runs of 3 min each ) , immediately before the Pre- and Post-deprivation MR sessions , in a quiet space adjacent to the MR control room . Visual stimuli were created in MATLAB running on a laptop ( Dell ) using PsychToolbox ( Brainard , 1997 ) , and displayed on a 15- inch monitor ( BenQ ) . Like in ( Lunghi et al . , 2015b ) , observers viewed the visual stimuli presented on the monitor at a distance of 57 cm through anaglyph red-blue goggles ( right lens blue , left lens red ) . Responses were recorded with the computer keyboard by continuous alternate keypresses . Visual stimuli were two oblique orthogonal red and blue gratings ( orientation:±45° , size: 3° , spatial frequency: 2 cpd , contrast 50% ) , surrounded by a white smoothed circle , presented on a black uniform background in central vision . Peak luminance of the red grating was reduced to match the peak luminance of the blue one using photometric measures . All included participants had typical binocular rivalry dynamics , with low percentage of mixed percepts ( reported for 8 . 5 ± 2 . 04% of time on average ) . Only one participant experienced of mixed percepts for more than 20% of time ( exactly for 31 . 2% ) and his data are in line with the others’ . Visual stimuli were projected with an MR-compatible goggle set ( VisuaStimDigital , Resonance Technologies , Los Angeles , USA ) , connected to a computer placed in the control room . The goggles covered a visual field of approximately 32 × 24 deg , with a resolution of 800 × 600 pixels , mean luminance 25 cd/m2; the images in the two eyes were controlled independently . During all functional MRI scans participants were instructed to maintain central fixation on a red point ( 0 . 5 degrees ) that was constantly visible at the center of the screen . Bandpass noise stimuli were white noise images filtered to match the spatial frequency tuning of neurons in the visual cortex ( Morrone and Burr , 1988 ) . We generated a large white noise matrix ( 8000 × 6000 ) and filtered it with a two-dimensional circular bandpass filter Bp defined by Equation 1: ( 1 ) Bp=e -ln⁡wP22[qln2]2where P is the peak spatial frequency , q is the filter half-width at half maximum in octaves . We generated five band-pass noise stimuli , by setting q = 1 . 25 octaves and p = 0 . 1 cpd , 0 . 2 cpd , 0 . 4 cpd , 1 . 1 cpd , 2 . 7 cpd . Each stimulus was presented for a block of 3 TRs , during which the image was refreshed at 8 Hz ( randomly resampling a 800 × 600 window from the original matrix ) . Stimuli were scaled to exploit the luminance range of the display , and this yielded very similar RMS contrast values ( shown in Figure 4—figure supplement 1 ) . Stimulus blocks were separated by 4 TRs blanks , consisting of a mid-level gray screen . The five band-pass noise stimuli blocks were presented in pseudo-random order , twice per run , for a total of 70 TRs . In each run , stimuli were only presented to one eye , while the other was shown a mid-level gray screen . Each eye was tested once , before and after deprivation . Immediately upon application of the monocular patch , we performed two additional scans to perform retinotopic mapping of visual areas . Meridian and ring stimuli were presented monocularly ( to the non-patched eye ) and were defined as apertures of a mid-level gray mask that uncovered a checkerboard pattern , 1 deg at 1 deg eccentricity to 2 . 5 deg at 9 deg eccentricity , rotating and contracting at a rate of one check per second . Meridians were defined by two 45° wedges centered around 0° or around 90° . The horizontal and vertical meridian were presented interchangeably for 5 TRs each ( without blanks ) and the sequence was repeated six times for a total of 60 TRs . Rings partitioned screen space into six contiguous eccentricity bands ( 0–0 . 9 deg , 0 . 9–1 . 8 deg , 1 . 8–3 . 3 deg , 3 . 3–4 . 7 deg , 4 . 7–6 . 48 deg , 6 . 48–9 deg ) . Odd and even rings were presented in two separate runs . In each run , the three selected rings and one blank were presented in random order for 5 TRs each , and the sequence was repeated ( with different order ) 6 times for a total of 120 TRs . Scanning was performed on a Discovery MR950 7 T whole body MRI system ( GE Healthcare , Milwaukee , WI , USA ) equipped with a 2-channel transmit driven in quadrature mode , a 32-channel receive coil ( Nova Medical , Wilmington , MA , USA ) and a high-performance gradient system ( 50 mT/m maximum amplitude and 200 mT/m/ms slew rate ) . Anatomical images were acquired at 1 mm isotropic resolution using a T1-weighted magnetization-prepared fast Fast Spoiled Gradient Echo ( FSPGR ) with the following parameters: TR = 6 ms , TE = 2 . 2 ms . FA = 12 deg , rBW = 50 kHz , TI = 450 ms , ASSET = 2 . Functional images were acquired with spatial resolution 1 . 5 mm and slice thickness 1 . 4 mm with slice spacing = 0 . 1 mm , TR = 3 s , TE = 23 ms , rBW = 250 kHz , ASSET = 2 , phase encoding direction AP-PA . No resampling was performed during the reconstruction . For each EPI sequence , we acquired two additional volumes with the reversed phase encoding direction . Analyses were performed mainly with Freesurfer v6 . 0 . 0 , with some contributions of the SPM12 and BrainVoyager 20 . 6 and FSL version 5 . 0 . 10 ( Jenkinson et al . , 2012 ) packages . Anatomical images were corrected for intensity bias using SPM12 ( Friston , 2007 ) and processed by a standard procedure for segmentation implemented in Freesurfer ( recon-all: Fischl et al . , 2002 ) . In addition , each hemisphere was aligned to a left/right symmetric template hemisphere ( fsaverage_sym: Greve et al . , 2013 ) . Functional images were corrected for subject movements ( Goebel et al . , 2006 ) and undistorted using EPI images with reversed phase encoding direction ( Brain Voyager COPE plug-in Jezzard and Balaban , 1995 ) . We then exported the preprocessed images from BrainVoyager to NiFTi format . These were aligned to each participant’s anatomical image using a boundary based registration algorithm ( Freesurfer bbergister function ) and projected to the cortical surface of each hemisphere . All analyses were conducted on data in the individual subject space . In addition , for visualization purposes , we also aligned the results of timecourse analyses ( GLM and subsequent pRF and spatial frequency tuning estimates ) to the left/right symmetric template hemisphere . Averaged results across the 18 × 2 hemispheres are shown in the maps of Figure 1B , Figure 5A and Figure 1—figure supplement 1 . General Linear Model analysis was performed with in-house MATLAB software ( D’Souza et al . , 2016 ) . We assumed that fMRI timecourses result from the linear combination of N predictors: boxcar functions representing stimulus presence/absence ( one per stimulus type ) convolved by a standard hemodynamic response function ( see Equation 2 ) , plus two nuisance variables ( a linear trend and a constant ) . We modeled the hemodynamic response function as a gamma function h ( t ) : ( 2 ) ht=t-δτ ( n-1 ) e-t-δττn-1 ! with parameters n = 3 , t = 1 . 5 s , and d = 2 . 25 s ( Boynton et al . , 1996 ) . Beta weights of the stimuli predictors were taken as estimates of the BOLD response amplitude and normalized by the predictor amplitude to obtain a measure that directly corresponds to % signal change; beta weights were also scaled by an error measure to obtain t-values , following the same procedure as in ( Friston et al . , 1994 ) . Computing BOLD responses for each individual vertex of the cortical surface leads to up-sampling the functional data ( each 1 . 5 × 1 . 5 × 1 . 5 mm functional voxel projecting on an average of 3 vertices ) . We ensured that this does not affect our statistical analyses by first averaging data from all vertices within a region of interest ( e . g . V1 ) , thereby entering all ANOVAs with a single value per subject and region of interest . The pRFs of the selected voxels were estimated with custom software in MATLAB , implementing a method related to that described by Dumoulin and Wandell ( Dumoulin & Wandell , 2008 ) and provided as supplementary material . We modeled the pRF with a 1D Gaussian function defined over eccentricity , with parameters ε and σ as mean and standard deviation respectively , and representing the aggregate receptive field of all neurons imaged within the vertex area . We defined the stimulus as a binary matrix S representing the presence of visual stimulation over space ( here , eccentricity between 0 and 10 deg with 40 steps per deg ) for each of 6 ring stimuli . We used the results of our GLM analysis to estimate the vertex response to each of our 6 rings ( as t-values; using beta values yields very similar results ) . We assumed that each vertex response is the linear sum over space ( eccentricity ) of the overlap between the pRF of the voxel and the input stimulus , which is mathematically equivalent to the matrix multiplication between the stimulus and the pRF . ( 3 ) Ri=Gε , σSi where i is the index to ring number and varies between 1 and 6 . We used this equation to predict the response to our six rings for a large set of initial pRF parameters; for each vertex , we measured the correlation ( our goodness-of-fit index ) between the predicted response and the observed t-values . If the highest correlation was . 7 the vertex was discarded; otherwise , the parameters yielding the highest correlation were used to initialize a nonlinear search procedure ( MATLAB simplex algorithm ) , which manipulated εand σto maximize goodness-of-fit , with the constraint that εcould not exceed 20 deg or be smaller than 1 deg , and σcould not be smaller than . 1 deg . Successful fits were obtained for 72 . 00 ± 1 . 86% of V1 vertices , for which the initial coarse grid search gave a correlation > 0 . 7 and the nonlinear search settled within the constraints . All analyses ( on average and distribution of responses and tuning parameters ) considered the sub-region of V1 for which a successful fit was obtained . We used εto estimate the preferred eccentricity of each vertex . The main modifications of our procedure relative to that described by Dumoulin and Wandell ( Dumoulin and Wandell , 2008 ) are the following: ( a ) fMRI data were acquired in a block design with only six stimulus types ( six eccentricity bands ) rather than varying stimulus position at each TR; this allowed us to use a standard GLM approach to estimate each vertex response to the six stimuli ( assuming a standard hemodynamic response function ) and then use the pRF model to predict these six time-points – much faster than predicting the full fMRI series of 120 × 2 TRs; ( b ) our stimuli and consequently our pRFs were defined in one dimension ( eccentricity ) – whereas the standard pRF is defined in 2D , eccentricity and polar angle ( or Cartesian x and y ) ; ( c ) we maximized the correlation between the predicted and observed fMRI response time-courses rather than minimizing the root mean square error; this eliminates the need to estimate a scale factor to account for the unknown units of the BOLD signal . Using a similar logic , we also estimated the population tuning for Spatial Frequency , which represents the aggregate Spatial Frequency tuning of the population of neurons imaged within each vertex area . We modeled the population tuning using a family of functions that includes the psychophysical Contrast Sensitivity Function ( CSF ) and can be specified by the following one-parameter equation ( Difference-of-Gaussians ) : ( 4 ) pSFT=e −v2σ – e −v2σ/50 × σ Like we did for the pRF mapping , we defined a stimulus matrix S representing the Fourier spectra of our five bandpass noise stimuli , that is the energy of visual stimulation in the frequency domain ( here , between 0 . 03 cpd and 12 . 5 cpd ) for each stimulus . We used the results of our GLM analysis to estimate the vertex response to each of our five bandpass noise stimuli ( as t-values; using beta values yields very similar results ) . We assumed that each vertex response is the linear sum over frequency of the overlap between the pSFT of the voxel and the input stimulus , which is mathematically equivalent to the matrix multiplication between the stimulus and the pSFT . Like for pRFs , we estimated the best-fit σparameter of each vertex pSFT with a two-step procedure: a coarse-grid search followed by the simplex search . We used the matrix multiplication of the pSFT and the stimulus to predict the response to our five bandpass noise stimuli for a large set of initial σvalues ( between 1 and 1 , 000 in 100 logarithmic steps ) ; for each vertex , we measured the correlation ( our goodness-of-fit index ) between the predicted response and the observed t-values . If the highest correlation was < . 5 , the voxel was discarded , otherwise the parameter yielding the highest correlation were used to initialize a nonlinear search procedure ( MATLAB simplex algorithm ) , which manipulatedσ to maximize goodness-of-fit , with the constraint that σcould not be smaller than . 3 and larger than 10 , 000 . Successful fits were obtained for 88 . 84 ± 1 . 28% of V1 vertices for which we obtained a successful eccentricity fit ( 86 . 77 ± 1 . 25% of all V1 vertices ) . We express the σ parameter in terms of the high-spatial frequency cutoff of the filter ( highest spatial frequency at half maximum ) , SFco for each vertex:σ2 We computed the effects of short-term monocular deprivation on both the dynamics of binocular rivalry and our fMRI results , estimating the degree to which the two measures are correlated . In all cases , the same equation was applied to psychophysical and fMRI data . The first index , called ‘Deprivation Index’ or DIpsycho and DIBOLD is given by Equation 6 ( 6 ) DI=yDepPOST yDepPRE /yNdepPOST yNdepPRE For psychophysics , y = mean duration of Binocular Rivalry phases of the Dep or Ndep eye , during the PRE- or POST deprivation sessions; for fMRI , y = mean BOLD response across V1 vertices to stimuli in the Dep or Ndep eye , during the PRE- or POST-deprivation sessions . The second index , called ‘Deprived-eye change’ or DepCpsycho and DepCcutoff is given by Equation 7 ( 7 ) DepC=yDepPOST yDepPRE For psychophysics , y = mean duration of Binocular Rivalry phases of the Dep eye , during the PRE- or POST deprivation sessions . For fMRI , y = mean spatial frequency cut-off across V1 vertices estimated for stimuli in the Dep eye , during the PRE- or POST-deprivation sessions . Data from individual participants ( mean binocular rivalry phase durations or mean BOLD responses/pRF/pST across V1 or V2 vertices ) were analyzed with a repeated measure ANOVA approach , after checking that distributions do not systematically deviate from normality by means of the Jarque-Bera test for composite normality ( MATLAB jbtest function , p-values given in the relevant figures ) . F statistics are reported with associated degrees of freedom and p-values in the Results section , in the form: F ( df , dferr ) = value; p = value . Post-hoc paired t-tests comparing conditions follow the ANOVA results , in the form: t ( df ) = value , p = value . Associations between variables are assessed with Pearson product-moment correlation coefficient , reported in the form: r ( n ) = value , p = value . Aggregate subject data ( i . e . vertices pooled across participants and hemispheres ) were typically non-normally distributed and thereby were analysed with non-parametric tests . The Wilcoxon sign-rank test was used for comparing medians , and results are reported in the form: z = value , p = value .	The world around us changes all the time , and the brain must adapt to these changes . This process , known as neuroplasticity , peaks during development . Abnormal sensory input early in life can therefore cause lasting changes to the structure of the brain . One example of this is amblyopia or ‘lazy eye’ . Infants who receive insufficient input to one eye – for example , because of cataracts – can lose their sight in that eye , even if the cataracts are later removed . This is because the brain reorganizes itself to ignore messages from the affected eye . Does the adult visual system also show neuroplasticity ? To explore this question , Binda , Kurzawski et al . asked healthy adult volunteers to lie inside a high-resolution brain scanner with a patch covering one eye . At the start of the experiment , roughly half of the brain’s primary visual cortex responded to sensory input from each eye . But when the volunteers removed the patch two hours later , this was no longer the case . Some areas of the visual cortex that had previously responded to stimuli presented to the non-patched eye now responded to stimuli presented to the patched eye instead . The patched eye had also become more sensitive to visual stimuli . Indeed , these changes in visual sensitivity correlated with changes in brain activity in a pathway called the ventral visual stream . This pathway processes the fine details of images . Groups of neurons within this pathway that responded to stimuli presented to the patched eye were more sensitive to fine details after patching than before . Visual regions of the adult brain thus retain a high degree of neuroplasticity . They adapt rapidly to changes in the environment , in this case by increasing their activity to compensate for a lack of input . Notably , these changes are in the opposite direction to those that occur as a result of visual deprivation during development . This has important implications because lazy eye syndrome is currently considered untreatable in adulthood .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Response to short-term deprivation of the human adult visual cortex measured with 7T BOLD
The parents' phenotype , or the environment they create for their young , can have long-lasting effects on their offspring , with profound evolutionary consequences . Yet , virtually no work has considered how such parental effects might change the adaptive value of behavioural traits expressed by offspring upon reaching adulthood . To address this problem , we combined experiments on burying beetles ( Nicrophorus vespilloides ) with theoretical modelling and focussed on one adult behavioural trait in particular: the supply of parental care . We manipulated the early-life environment and measured the fitness payoffs associated with the supply of parental care when larvae reached maturity . We found that ( 1 ) adults that received low levels of care as larvae were less successful at raising larger broods and suffered greater mortality as a result: they were low-quality parents . Furthermore , ( 2 ) high-quality males that raised offspring with low-quality females subsequently suffered greater mortality than brothers of equivalent quality , which reared larvae with higher quality females . Our analyses identify three general ways in which parental effects can change the adaptive value of an adult behavioural trait: by influencing the associated fitness benefits and costs; by consequently changing the evolutionary outcome of social interactions; and by modifying the evolutionarily stable expression of behavioural traits that are themselves parental effects . Parental effects can play a major non-genetic role in determining an offspring's phenotype , either via the parents' phenotype or through the environment that parents create ( Badyaev and Uller , 2009; Wolf and Wade , 2009 ) . It is well-appreciated that parental effects can persist beyond the juvenile stage , profoundly influencing the offspring's subsequent adult phenotype , including key fitness-related traits such as lifespan ( Bateson et al . , 2004; Nussey et al . , 2007 ) and fecundity ( Wilkin and Sheldon , 2009; Emlen et al . , 2012 ) . Consequently , parental effects can change the nature and pace of ecological and evolutionary change , potentially allowing organisms to adapt quickly in a rapidly changing environment ( e . g . , Räsänen and Kruuk , 2007; Badyaev and Uller , 2009; Pfennig and Martin , 2009; Duckworth et al . , 2015 ) . Here , we consider whether parental effects can also change the adaptive value of behavioural traits , something that has been rather overlooked in previous analyses . Yet , behavioural traits are among the first aspects of the phenotype to adapt to a changed environment ( West-Eberhard , 2003; Zuk et al . , 2014 ) , and parental effects may play a key role in this process . We address this problem by focusing specifically on how conditions in early-life influence the fitness costs and benefits associated with the supply of parental care in adult life , using a combination of experiments and new theory . By choosing to focus on parental care , we are able to analyze three different ways in which parental effects might hypothetically influence the adaptive value of any behavioural trait . First , by focusing on uniparental care , we can investigate how parental effects change the relative fitness payoffs associated with a behaviour performed in adulthood . We assess these fitness payoffs experimentally by measuring the fitness gained from supplying care ( the benefit of care ) as well as the fitness simultaneously lost ( the cost of care , e . g . , Grafen , 1984 ) . The second approach takes advantage of the fact that parental care is often supplied by both parents , working as a team . Parents commonly respond to the effort put in by their partner ( e . g . , Johnstone and Hinde , 2006 ) , which means that the parental behaviour exhibited by one parent can potentially change the fitness of its partner by inducing a change in parental effort ( e . g . , Houston and Davies , 1985; Houston et al . , 2005 ) . This allows us to analyze how parental effects experienced in early life might influence the outcome of a social interaction in adulthood . For a social behaviour , such as biparental care , the relative magnitude of the fitness costs and benefits of the interaction with another individual are key to understanding whether the outcome is cooperative or exploitative ( e . g . , West et al . , 2006 ) . If the fitness of both social partners is enhanced by the interaction then the outcome is cooperative , but if one party gains fitness at the other's expense then it is selfishly exploiting its social partner . For biparental care , cooperation and selfish exploitation are each possible elements of parents working together to raise young . A cooperative element is possible when the two parents have an equal genetic stake in their shared young and so derive equal fitness benefits from the supply of care . But , a selfish exploitative element is also possible due to conflict between the parents over how to divide the fitness costs associated with the supply of care ( Arnqvist and Rowe , 2005; Houston et al . , 2005; Lessells , 2006 ) . Selection favours parents that force their partner to sustain a greater fitness cost which , in accordance with the definition of sexual conflict ( Kokko and Jennions , 2014 ) , means that the parent bearing the increased cost of care is selfishly exploited by its partner . Parental effects , experienced in early life , can potentially change the outcome of interactions during the supply of biparental care in adulthood by changing the costs and benefits associated with the supply of care ( Barta et al . , 2002; Harrison et al . , 2009; Lessells and McNamara , 2012 ) . Theoretical analyses suggest that offspring that develop in a well-resourced environment may mature into high-quality parents . In this context , the term ‘high-quality’ is precisely defined: for the same provision of care , a ‘high-quality’ parent sustains a lower fitness cost than a ‘low-quality’ parent ( Lessells and McNamara , 2012 ) . In theory , a high-quality parent is vulnerable to selfish exploitation by a low-quality partner who is incentivized to reduce its greater costs of care by offloading them onto the superior quality parent ( Lessells and McNamara , 2012 ) . Whether this ever happens unknown ( but see Monaghan et al . , 2012 ) . A third feature of parental care is that it directly influences features of the early-life environment in which the next generation develops and thus directly induces corresponding change in the offspring's phenotype: in other words , and in common with several other behavioural traits , parental care is itself a type of parental effect ( Badyaev and Uller , 2009; Wolf and Wade , 2009 ) . Thus , early-life conditions potentially change the costs and benefits of parental care in adulthood , which in turn influence the early-life conditions experienced by the next generation . The influence of these transgenerational effects on the evolution of parental care strategies is , however , unknown . We develop new theoretical models , informed by our experiments , to analyze how these long-term fitness consequences of parental effects might feed back to change optimal levels of parental care . Our experiments use the burying beetle , Nicrophorus vespilloides , as a model organism . Like other Nicrophorus beetles , this species uses the carcass of a small vertebrate for reproduction . It exhibits unusually elaborate biparental care , though one parent ( of either sex ) can raise offspring singlehandedly ( e . g . , Ward et al . , 2009 ) . A major advantage of using the burying beetle is that we can measure correlates of fitness that are associated with receiving and supplying parental care . Indeed , our study differs from most previous empirical work on parental care by measuring fitness , rather than by quantifying behaviour itself . The rationale underpinning our experimental approach is this: behavioural traits associated with care cannot straightforwardly be mapped onto fitness ( Sheldon , 2002; Harrison et al . , 2009 ) —yet , measures of fitness are key to understanding evolution of parental care ( Sheldon , 2002; Harrison et al . , 2009 ) . By measuring changes in fitness associated with the supply and receipt of parental care , rather than quantifying behavioural traits , we can therefore more easily use our experiments to draw evolutionary conclusions . In the burying beetle , all forms of care incur measurable fitness costs and benefits ( e . g . , Ward et al . , 2009; Cotter et al . , 2010 ) . We quantified the fitness benefits of care by measuring larval mass at dispersal because this is strongly correlated with survival ( e . g . , Eggert et al . , 1998; Lock et al . , 2004 ) . We used lifespan to measure the fitness costs associated with parental care , because several aspects of the burying beetle's reproductive biology mean that a longer lifespan contributes directly to greater fecundity in each sex and is therefore a good proxy for fitness . For example , in nature , N . vespilloides is an opportunistic breeder because it is reliant on locating small carrion to produce offspring ( Schwarz and Müller , 1992; Scott , 1998 ) . The longer it lives , the more likely it is to find this key resource . Males that are unsuccessful at locating carrion may attempt to attract females for mating by secreting pheromones ( Eggert and Müller , 1989a , 1989b; Eggert , 1992 ) . Females can store sperm from these matings for use in future reproduction ( Müller and Eggert , 1989 ) and males commonly gain paternity without attending the carcass upon which larvae are raised ( Müller et al . , 2007 ) . Therefore , the longer a male lives , the more likely he is to be successful in acquiring matings and increasing his reproductive success . In our previous laboratory experiments , we have shown that a longer lifespan enhances lifetime reproductive success in both sexes by affording a greater number of opportunities for reproduction ( Ward et al . , 2009; Cotter et al . , 2010 , 2011 ) . Note that there is no difference between the sexes in the lifespan of virgins ( Figure 1—figure supplement 1 ) . All the individuals used in this experiment belonged to a captive colony ( kept at a constant temperature: 21°C , with a 16-hr:8-hr light:dark cycle ) established at Cambridge University in 2005 and supplemented every summer thereafter with wild caught adults collected under licence from local field sites at Byron's Pool and Wicken Fen in Cambridgeshire . Adults were housed individually in plastic boxes ( 12 × 8 × 2 cm ) filled with moist soil and fed twice a week with ca . 1 g minced beef . For breeding , pairs of unrelated individuals were placed into larger plastic boxes ( 17 × 12 × 6 cm ) half filled with moist soil , provided with a 15–35 g freshly thawed mouse carcass and kept in the dark to simulate natural underground conditions . The larvae disperse from the carcass to pupate roughly eight days after pairing . Pupation takes approximately three weeks and sexual maturity is reached approximately two weeks after eclosion . This experiment spanned two generations . In Generation 1 , we manipulated larval developmental environment via a parental effect . We then kept these larvae to breed as adults in Generation 2 , when we measured the fitness costs and benefits associated with providing care . The data we collected and analyzed are archived in Dryad ( Kilner et al . , 2015 ) . We used linear mixed models ( LMMs ) except for the analyses of brood size at dispersal , where generalized LMMs ( GLMMs ) with a Poisson error distribution were employed because the data were not normally distributed . The R package lme4 was used ( Bates et al . , 2014 ) . The variances in parenting performance in small and large broods differed significantly ( Levene test , F1 , 77 = 14 . 41 , p < 0 . 001 ) , so we square-root transformed the data prior to analysis . The experimental data for males and females were analyzed separately , but using the same approach . All models included the family of origin of the experimental individuals and the family of the donor larvae nested within batch as random intercept effects . Parental care duration ( 0 hr , 24 hr ) and brood size ( small = 5 larvae , large = 20 larvae ) treatment effects were tested by two-level fixed factors . When present , carcass mass , brood mass at dispersal , and adult size were entered as covariates . Longevity data were analyzed using a Cox's proportional hazard model with the parent's family nested within batch using the R package coxme ( Therneau , 2012 ) . All analyses were run with R 3 . 1 . 2 ( R Development Core Team , 2013 ) . This experiment also spanned two generations . In Generation 1 , we manipulated larval developmental environment via a parental effect . We then kept only female larvae to breed as adults in Generation 2 . Females were paired with males that had received full-time care as larvae and the pair was allowed to raise offspring together . After reproduction , we measured the fitness costs and benefits associated with providing care for both sexes . The data we collected and analyzed are archived in Dryad ( Kilner et al . , 2015 ) . For data that were not normally distributed , we used GLMMs assuming a Poisson error distribution , and using the R package lme4 ( Bates et al . , 2014 ) . A LMM was used for analyzing average offspring mass , because these data were normally distributed . All models included batch number as a random factor and both the female's and male's family of origin nested within batch as random intercept effects . Maternal care duration ( 0 hr , 8 hr , 24 hr , 192 hr/full care ) was entered as a four-level fixed factor . Where relevant , carcass mass , female and male adult size were entered as covariates . In the models of reproductive output , we analyzed data collected over the two experimental breeding bouts . Analyses of brood size used the total number of offspring produced over both breeding attempts . For average larval mass , we calculated this value separately for each breeding attempt and then derived the mean . ( The results for each measure of offspring performance are qualitatively similar when data from the first breeding bout alone are analyzed . ) Longevity data were analyzed using a Cox's proportional hazards model with the family of the experimental individual nested within batch using the R package coxme ( Therneau , 2012 ) . Post hoc analyses were performed adopting the Bonferroni correction for multiple testing . All analyses were performed with R 3 . 1 . 2 ( R Development Core Team , 2013 ) . To provide further insight into our empirical results , we constructed a mathematical model of the evolution of parental care in an idealized population of burying beetles ( Figure 1 ) . Our model explores how individual variation in parental quality affects the amount of care that parents evolve to provide . We assume that the care offspring receive during development affects their own future quality as parents , a type of parental effect . To explore the impact of sexual conflict on care decisions , we compare predicted levels of care when coparents are unrelated to each other and when they are genetically identical ( i . e . , r = 1 between coparents ) . The latter scenario entirely removes sexual conflict over care decisions . We first provide a general outline of the model before considering each piece in more detail . 10 . 7554/eLife . 07340 . 003Figure 1 . Overview of the model structure , including mortality and survival ( A ) and reproduction ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07340 . 00310 . 7554/eLife . 07340 . 004Figure 1—figure supplement 1 . Longevity of adult virgin males ( n = 43 ) and females ( n = 50 ) ( A ) shown as a cumulative survival plot and ( B ) comparing mean ± S . E . M lifespan for each sex . The rate at which the beetles died was not significantly different , nor was there a difference in their mean lifespan ( X2 = 0 . 69 , d . f . = 1 , p = 0 . 41 ) . These data were collected from an experiment in which individuals were removed at eclosion from 13 different families at random , with each family contributing between 2 and 13 experimental subjects . The experimental subjects were weighed and kept under standard conditions ( see ‘Materials and methods’ ) until they died . The population was censused twice a week . In the statistical analyses reported above , body mass at eclosion and sex were fixed terms and family of origin was a random term , using the R package coxme ( see ‘Materials and methods’ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07340 . 00410 . 7554/eLife . 07340 . 005Figure 1—figure supplement 2 . Frequency distribution of the size of field-caught N . vespilloides ( shown in blue ) and the experimental N . vespilloides described in this study ( shown in red ) . The size of the beetle is given by its pronotum width . There is no sexual size dimorphism , so the data for males and females have been pooled . Our experimental treatments generated beetles that were well within the range of size of beetles that occur naturally . DOI: http://dx . doi . org/10 . 7554/eLife . 07340 . 005 We assume a large , well-mixed population that breeds continuously . Adult beetles alternate their time between searching for suitable corpses , waiting at corpses for a mate to arrive , and breeding . When a searching individual locates an unoccupied corpse , or one that is occupied only by an opposite-sex adult , it remains there to breed . Corpses that are occupied by a same-sex competitor are avoided . This means that the first female and the first male to encounter a corpse form a monogamous breeding pair . They care jointly for their offspring , and once their brood reaches maturity , the pair separates and each beetle resumes its search for corpses . Although breeding is continuous in our model , we do not specify how much time beetles spend engaged in each activity . Rather , we model trade-offs between care provision , fecundity , and mortality that do not depend explicitly on time . Adult beetles vary in their quality , with higher quality individuals suffering lower mortality costs after providing any given level of parental care . For simplicity , we divide the adult population into ‘good’ and ‘poor’ parents . Parental quality is determined during development and persists throughout an adult's life . We assume that opportunities to develop into a good parent are limited at the population level by competition for resources . The overall proportion of larvae that develop into good parents is consequently fixed , regardless of the average level of care in the population . This is analogous to density-dependent effects on survival . Care has two effects on offspring . First , the total number of larvae surviving from a brood ( ‘brood productivity’ ) increases with the sum of care provided by the two parents . Second , larvae from broods receiving high levels of care are more likely to develop into good parents than those receiving low levels of care . The level of care an individual provides can vary according to both its own parental quality ( good or poor ) and that of its mate . However , individuals do not respond plastically to the immediate level of care that their mate provides . This is equivalent to a ‘sealed bid’ model of parental care ( Houston and Davies , 1985 ) , though in our case , the bid is multidimensional ( quality-dependent ) . We also assume that there are no sex differences in the fitness costs and benefits of caring , so that males and females provide the same levels of care . We write cAB for the amount of care provided by a parent of type A that is paired with a parent of type B . Each of the subscripts can equal either G ( for good parents ) or P ( for poor parents ) to give us a total of four possible care levels: cGG , cGP , cPG , and cPP . For simplicity , we assume a haploid genetic system , where each care level is coded for by a single allele . We allow each care level to evolve independently of the others . We begin by calculating the reproductive value of good and poor parents in a population where every individual behaves in the same way . We then calculate reproductive values of mutants whose care behaviour differs from the population norm . This allows us to define selection gradients on the amount of care provided in different circumstances , which we then use to find evolutionarily stable levels of care . We repeat this analysis for populations in which coparents are genetically identical in order to explore the effects of sexual conflict over parental care . For simplicity , we assume that mortality of adult beetles occurs during the search phase between bouts of reproduction . An individual's probability of mortality increases with the level of care it provided in the previous bout . Mortality also depends on an individual's quality as a parent , with higher quality individuals suffering from lower mortality after providing any given level of care . We write mG ( c ) and mP ( c ) for the mortality risk of a good parent and a poor parent , respectively , after providing care of c . We assume that mortality risk increases as a convex function of the amount of care a parent provides ( see Appendix 1 ) . By definition , good parents suffer lower mortality than poor parents after providing the same level of care , so we assume mG ( c ) < mP ( c ) for all c . The probability that a parent of type A dies between reproductive bouts after raising offspring with a parent of type B can be written as ( 1 ) mAB=mA ( cAB ) . Now , let g1 be the ( assumed fixed ) proportion of newly mature adults that are good parents and let g2 be the proportion of all mating adults that are good parents . These two quantities might not be equal due to differences in mortality between good and poor parents . Since we assume continuous breeding , g2 is constant with time . When a good parent reproduces , the probability that its mate is also a good parent is g2 , in which case , the focal parent faces a subsequent mortality risk of mGG . Similarly , the probability that the focal parent's mate is a poor parent is 1 − g2 , in which case , the mortality risk is mGP . The overall probability that a good parent dies between reproductive bouts is thus ( 2 ) mG=g2mGG+ ( 1−g2 ) mGP . For poor parents , the average mortality risk is similarly ( 3 ) mP=g2mPG ( 1−g2 ) mPP . Good parents make up g1 of newly mature adults and mate on average 1/mG times in their lives , while poor parents make up 1 − g1 of new adults and mate 1/mP times . The proportion of mating adults that are good parents is therefore ( 4 ) g2=g1/mGg1/mG+ ( 1−g1 ) /mP . Equation 2 through Equation 4 can be solved simultaneously to yield values for g2 , mG , and mP in terms of g1 . The average number of offspring surviving from a brood depends on the total care provided by the two parents . For a mating between parents of types A and B , we write the expected brood productivity as ( 5 ) bAB=b ( cAB+cBA ) . We assume that brood productivity is a concave function of the amount of care received ( see Appendix 1 ) . In order to simplify calculations , we normalize brood productivity so that an average adult has lifetime reproductive success of one . This normalization is for mathematical convenience and does not affect the generality of our results . We calculate pre-normalized average lifetime reproductive success as follows . First , each good parent mates on average 1/mG times during its life . A proportion g2 of these matings are with good parents and result in brood productivity of bGG . The remaining proportion 1 − g2 are with poor parents and result in brood productivity of bGP . The average lifetime reproductive success of a good parent is thus ( 1/mG ) [g2bGG + ( 1 − g2 ) bGP] . Similarly , for poor parents , the average lifetime reproductive success is given by ( 1/mP ) [g2bGP + ( 1 − g2 ) bPP] . Since good parents make up a proportion g1 of newly mature adults , overall average lifetime reproductive success in the population is ( 6 ) b¯= ( g1mG ) [g2bGG+ ( 1−g2 ) bGP]+ ( 1−g1mP ) [g2bGP+ ( 1−g2 ) bPP] . Normalized brood productivity can then be calculated as ( 7 ) bAB′=bABb¯ . These calculations also lead to simple expressions for the proportions of newly mature offspring that were raised by parents of each possible combination of parental quality ( i . e . , both good , one good and one poor , or both poor ) . These are ( 8 ) qGG= ( g1mG ) g2bGG′ , ( 9 ) qGP=[ ( g1mG ) ( 1−g2 ) + ( 1−g1mP ) g2]bGP′ , ( 10 ) qPP= ( 1−g1mP ) ( 1−g2 ) bPP′ . Note that qGG + qGP + qPP = 1 . We now determine the proportion of larvae raised by each type of pair that develops into good parents . We assume that opportunities to develop into a good parent are limited at the population level by access to resources and that surviving larvae compete for a fixed number of such opportunities . An individual's relative competitiveness is an increasing function of the total amount of care it received during development . We write the relative competitiveness of individuals raised by parents of types A and B as ( 11 ) fAB=f ( cAB+cBA ) . Like brood productivity , we take competitiveness to be a concave function of the total care a brood receives ( see Appendix 1 ) . Suppose that opportunities to develop into a good parent arise continuously at a constant rate . We assume that each newly mature adult competes for such opportunities for a fixed time T , during which its probability of gaining any particular opportunity is proportional to its competitiveness with some constant of proportionality k > 0 . This means that for an individual with parents of types A and B , the probability of success can be written as ( 12 ) gAB=1−e−kfABT . By manipulating the above equation for gGG , gGP , and gPP , we obtain the relationships ( 13 ) ( 1−gGG ) 1/fGG= ( 1−gGP ) 1/fGP= ( 1−gPP ) 1/fPP . Further , since the final overall proportion of good parents is fixed at g1 , we must also have ( 14 ) gGGqGG+gGPqGP+gPPqPP=g1 . We can solve Equation 13 and Equation 14 simultaneously using numerical methods to yield values for gGG , gGP , and gPP . Note that the solution does not depend on the choice of k or T . We now have all we need to calculate the reproductive value of good and poor parents in a population where care behaviour is uniform . We write vG for the reproductive value of a good parent and vP for a poor parent . When a good parent mates with another good parent , their normalized brood productivity is bGG′ Of this brood , a proportion gGG of offspring will develop into good parents themselves , while 1 − gGG will develop into poor parents . The reproductive value that a good parent obtains from mating with another good parent is thus 12bGG′[gGGvG+ ( 1−gGG ) vP] . The factor of one half here accounts for the relatedness between an individual and its offspring . Similarly , good parents obtain reproductive value of 12bGP′[gGPvG+ ( 1−gGP ) vP] from mating with a poor parent . Since good parents mate on average 1/mG times , and a proportion g2 of these matings are with other good parents , the total reproductive value of a good parent is given by ( 15 ) vG=12mG{g2bGG′[gGGvG+ ( 1−gGG ) vP]+ ( 1−g2 ) bGP′[gGPvG+ ( 1−gGP ) vP]} . Similarly , the reproductive value of a poor parent is ( 16 ) vP=12mP{g2bGP′[gGPvG+ ( 1−gGP ) vP]+ ( 1−g2 ) bPP′[gPPvG+ ( 1−gPP ) vP]} . Equation 15 and Equation 16 can be rewritten as a matrix equation ( 17 ) [vGvP]=M[vGvP] , where the entries of M are given by ( 18 ) M11 =12mG[g2bGG′gGG+ ( 1−g2 ) bGP′gGP ]M12=12mG[g2bGG′ ( 1−gGG ) + ( 1−g2 ) bGP′ ( 1−gGP ) ]M21=12mP[g2bGP′gGP+ ( 1−g2 ) bPP′gPP]M22=12mP[g2bGP′ ( 1−gGP ) + ( 1−g2 ) bPP′ ( 1−gPP ) ] . The reproductive value of good and poor parents can then be found by taking the right eigenvector of M that corresponds to the eigenvalue 1 . We now have general expressions for the reproductive value of typical individuals in the population . We can similarly derive the reproductive values of mutants that differ in their care behaviour from population norms ( see Appendix 1 ) . We can use these to calculate selection gradients on the level of care provided in each of the four possible pairings of parental quality . We write c = ( cGG , cGP , cPG , cPP ) for the vector of population values for each care level and we write c^ for the corresponding vector for mutants . The selection gradient on c is then given by ( 19 ) S ( c ) = ( ∂v^G∂c^GG , ∂v^G∂c^GP , ∂v^P∂c^PG , ∂v^P∂c^PP ) \|c^=c , where v^G and v^P are the reproductive values of good and poor quality mutants , respectively . We located evolutionarily stable levels of care by starting with arbitrary levels of care c = c0 and then following the selection trajectories defined by S until care levels converged to an equilibrium . This was done by iterating the equation ( 20 ) ct+1=ct+ΔS ( ct ) , with Δ a small positive constant ( we found Δ = 0 . 01 suitable ) . We checked informally for multiple equilibria by running the model with many widely spaced starting vectors c0 to confirm that these starting vectors converged to the same equilibrium . Multiple equilibria were not found for any the parameter values we considered . To assess the effect of sexual conflict on parental care , we compared equilibrium care levels under the above model to a hypothetical scenario that removes all sexual conflict by assuming that coparents are genetically identical ( see Appendix 1 ) . We found that parents that received no post-hatching care as larvae matured into low-quality mothers and fathers ( using Lessells and McNamara's ( 2012 ) definition of parental quality ) . The key finding of this experiment was that males suffered greater fitness costs when raising offspring with lower quality females ( Χ2 = 15 . 05 , d . f . = 3 , p = 0 . 002; Figure 4; Table 2 ) . Males paired with females that had received no post-hatching care as larvae had significantly shorter lives than those whose partners received either 24-hr care ( Table 2b ) or 192-hr care as larvae ( Table 2b ) . Independent of their partner's quality , males also had a shorter life if the mass of their brood was greater at dispersal ( Table 2b ) and if they were relatively small in size ( Table 2b ) . By contrast , all the females in our experiment lived similarly long lives , irrespective of the conditions they experienced in early life ( Χ2 = 5 . 06 , d . f . = 3 , p = 0 . 17; Figure 4; Table 2a ) , or their size ( Table 2a ) , or the total mass of their broods at dispersal ( Table 2a ) . 10 . 7554/eLife . 07340 . 009Figure 4 . The effect of the female's early-life environment ( i . e . , the duration of care she received as a larva ) on her lifespan after reproduction ( white bars ) , and on the lifespan of the male with whom she raised offspring ( black bars ) . All males developed in a high-quality environment . The greater the difference within the pair in the environment they each experienced during development , the greater the difference in their subsequent lifespan . Low-quality mothers thus exploit high-quality fathers . Mean values with standard error bars are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 07340 . 00910 . 7554/eLife . 07340 . 010Table 2 . Results from Experiment 2: the influence of parental effects on the outcome of a social interactionDOI: http://dx . doi . org/10 . 7554/eLife . 07340 . 010NB parental effect experienced by femaleCoefficientStandard errorz valuep valuea . Female lifespan Parental effect: 8 hr vs 0 hr0 . 4270 . 2281 . 880 . 061 Parental effect: 24 hr vs 0 hr0 . 2160 . 2360 . 920 . 360 Parental effect: 192 hr vs 0 hr−0 . 0730 . 254−0 . 290 . 770 Total carcass mass0 . 0170 . 0161 . 060 . 290 Total brood mass0 . 0040 . 0290 . 160 . 870 Female pronotum0 . 5220 . 3221 . 620 . 110NB parental effect experienced by male's partnerCoefficientStandard errorz valuep valueb . Male lifespan Parental effect: 8 hr vs 0 hr−0 . 1090 . 235−0 . 470 . 640 Parental effect: 24 hr vs 0 hr−0 . 6640 . 252−2 . 630 . 008 Parental effect: 192 hr vs 0 hr−1 . 0330 . 291−3 . 550 . 0003 Total carcass mass−0 . 0270 . 023−1 . 170 . 240 Total brood mass−0 . 0940 . 032−2 . 870 . 004 Male pronotum0 . 9150 . 3662 . 500 . 012NB parental effect experienced by brood's motherEstimateStandard errorz valuep valuec . Brood size Intercept1 . 357760 . 392863 . 4560 . 0005 Parental effect: 8 hr vs 0 hr0 . 023500 . 032950 . 7130 . 476 Parental effect: 24 hr vs 0 hr0 . 162610 . 034534 . 710<0 . 0001 Parental effect: 192 hr vs 0 hr0 . 146410 . 036763 . 983<0 . 0001 Total carcass mass0 . 007650 . 003082 . 4850 . 013 Female pronotum0 . 311900 . 051336 . 076<0 . 0001 Male pronotum0 . 129730 . 057292 . 2650 . 024Parental effects were created experimentally by exposing females to 0 hr , 8 hr , 24 hr , or 192 hr of post-hatching care as larvae . They were then kept until adulthood and allowed to breed twice with a male who had received 192 hr of care as larva . The two parents raised offspring together . Each parent's lifespan thereafter was recorded , as was the mass of their brood at dispersal . Further details are given in the ‘Materials methods’ . This result cannot be attributed to males paired with low-quality females raising more offspring . Pairs where the female received no post-hatching care as a larva produced fewer offspring than those reared by females that received either 24-hr care ( Table 2c ) or 192 hr/full care ( Table 2c ) . Our model predicts that parents will provide more care when this increases the chances that their offspring develop into high-quality parents ( Figure 5 ) . Stronger parental effects selected for greater care regardless of whether sexual conflict was present or artificially removed from the model . When conflict was removed , however , parents both cared more in absolute terms and also increased their care more steeply in response to increasing parental effects ( Figure 5 ) . This is because sexual conflict dilutes the fitness benefits of investing in the current brood , whether these benefits arise from the quantity or the quality of offspring . 10 . 7554/eLife . 07340 . 011Figure 5 . The relationship between the average care received by a brood and the strength of parental effects αF ( i . e . , the extent to which care received affects the future parental quality of offspring ) . Average care levels increase with the strength of parental effects both when sexual conflict is present ( circles ) and when it is absent ( squares ) . When there is no sexual conflict , parents provide more care in absolute terms and also increase their care more steeply with increasing parental effects . Shown with g1 = 0 . 5 , bmin = fmin = 1 , αB = 5 , βG = 5 , βP = 5 , and mmin = 0 . 25 ( see Appendix 1 for details of function shapes ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07340 . 011 Good quality parents always invested more in care than poor quality parents in our model . The difference in care was large enough that good parents had higher overall mortality ( mG > mP ) . The reproductive value of good parents was nonetheless always higher , in contrast to the predictions of a previous model ( Lessells and McNamara , 2012 ) . Poor quality individuals invested relatively less in care when partnered with good than with poor partners ( i . e . , cPG < cPP ) and good quality individuals compensated for their partner's inferior efforts by increasing their own care ( cGP > cGG ) . Interestingly , this was true both with and without sexual conflict ( note that it is possible to remove conflict while retaining individual variation in parental quality , see Appendix 1 ) . A reduction of care by individuals when paired with a high-quality partner consequently does not provide evidence for sexual conflict per se . However , when coparents are unrelated ( as in the experiments and in the model with sexual conflict ) , a reduction in care by one individual will reduce the fitness of its partner and can consequently be considered exploitative . It has long been appreciated that parents can influence the development of diverse behavioural traits in their young simply by varying aspects of the environment in which their offspring grow and develop ( reviewed by e . g . , Champagne and Meaney , 2007; Daskalakis et al . , 2013; Burton and Metcalfe , 2014 ) . Furthermore , it is well-understood that parents can change their offspring's fitness simply by influencing their offspring's size ( e . g . , Steiger , 2013 ) . The work we present here provides further evidence in support of each of these well-established research findings . The novel contribution of our study is to identify three different ways in which parental effects can change the adaptive value of the offspring's behaviour in later life: by influencing the associated fitness benefits and costs , by changing the evolutionary outcome of social interactions , and by modifying the evolutionarily stable expression of behavioural traits that are themselves parental effects . Our first experiments , using single parents , showed that the extent of parental care received during development affects the fitness costs and benefits associated with the supply of care , when larvae mature into parents themselves . Whether we examined mothers or fathers , we found that individuals that received no post-hatching care as larvae were less effective at raising a large brood as parents and sustained a greater associated fitness cost . By Lessells and McNamara's ( 2012 ) definition , these were low-quality parents , deriving relatively low fitness benefits ( Figures 2 , 3 ) from supplying care whilst paying a relatively high fitness cost ( Figures 2 , 3 ) . With our second experiment , using biparental care , we showed that these long-term developmental effects on parental quality can potentially change the outcome of social interactions during biparental care in adulthood . Just as predicted by theory ( Lessells and McNamara , 2012 ) , we found that the costs of care were divided most unevenly between the sexes when the difference within the pair in parental quality was at its greatest ( Figure 4 ) . High-quality males had lower residual fitness after raising offspring with low-quality females than their brothers did after raising young with high-quality female partners , losing up to 8 days from their lifespan . Under these conditions , and according to the fitness-based definition of selfish exploitation we are using here ( see ‘Introduction’ ) , males were exploited by their lower quality partners . Based on previous behavioural observations of biparental care in N . vespilloides , the most likely explanation is that males with low-quality partners put more effort into parental duties to compensate for the shortcomings of their mate and paid a higher fitness cost accordingly ( Walling et al . , 2008; Ward et al . , 2009; Cotter et al . , 2010 ) . How these differences in lifespan translate into reproductive success in nature is unknown , but since males might mate with multiple females in 8 days ( Eggert , 1992 ) , the effects are unlikely to be trival . The biparental care experiment yielded an unexpected result . Although lower quality females paid a higher cost for raising offspring as single mothers ( Figure 2 ) , all females paid a similar cost under biparental care , irrespective of their quality ( Figure 4 ) . We suggest two explanations for these findings . One possibility is that they result from slight methodological differences between our first and second experiments . In the first experiment , we created standard broods of 5 and 20 larvae for our experimental subjects to raise , whereas in the second experiment , parents raised a brood size of their choosing . The former approach may have been more effective at exposing costs of care than the latter ( see Lessells , 1991 ) . An alternative possibility is that the results reveal strategic differences in the rules for raising offspring that depend on whether females raise offspring alone or with a partner . Perhaps when raising offspring with a partner , the female's contributions to care are calibrated more in relation to the costs associated with care , than in relation to the benefits gained . Conversely , when rearing young alone , maternal investment is perhaps determined more by the benefits to be gained rather than the costs that are borne . Which of these explanations better accounts for the results remains to be determined with further experiments . A key discovery of the biparental care experiment is that it reveals a novel potential social cost associated with receiving high levels of care during development , which is not borne until adulthood when individuals engage in biparental care themselves . The cost arises when a high-quality parent is paired with a low-quality partner and it is apparently caused by the lower quality parent forcing the higher quality parent to bear a greater share of the costs of parental investment . These findings raise two further questions . First , how likely is it that this social cost will ever show itself in nature , in any species that exhibits biparental care ? Many species with biparental care also exhibit assortative pairing ( e . g . , Jiang et al . , 2013 ) , and this will prevent the sort of mismatches in quality that we generated experimentally , and hence reduce the potential for exploitation of higher quality parents through sexual conflict . Our experimental results thus suggest a novel function of assortative pairing: to minimize differences in partner quality , and so reduce the social cost imposed by sexual conflict that we have uncovered here . In burying beetles , though , assortative pairing by mate choice is unlikely to be the norm because there is little evidence that individuals ever reject a potential mate ( Scott , 1998 ) . It might be argued that assortative pairing could instead arise as a by-product of fights for carcass ownership , because there can be intense competition within each sex over a carcass , which the largest individuals usually win ( e . g . , Hopwood et al . , 2013 ) . The dominant individuals of each sex then pair up and prepare the carcass together . However , it is unclear how frequently carcass ownership is contested in natural populations . In nature , adult burying beetles fly to locate carrion , which is a key resource for reproduction ( Scott , 1998 ) , often covering long distances in their search ( Attisano and Kilner , 2015 ) . Therefore , individuals are distributed patchily and at relatively low densities , according to the distribution of carrion . A field study in which carrion was put out for burying beetles found that ownership of the carcass was contested within both sexes in 22 out of 42 breeding events ( Müller et al . , 2007 ) . Whether the dominant pair were closely matched in body size , and more closely matched than pairs at uncontested carcasses , is not known . One interpretation of these different strands of evidence is that assortative pairing by quality is unlikely at least half of the time in N . vespilloides , and on these occasions therefore , high-quality burying beetles parents are vulnerable to exploitation by inferior quality partners . However , it is hard to draw strong inferences from the field study because it probably increased the size of the local burying beetle population by drawing individuals in to carcasses presented at a relatively high density , and so might have changed the extent of competition for a carcass . The second question raised by our discovery of this social cost in the next generation is how does it feed back to influence evolutionarily stable levels of parental care in the current generation ? Previous theoretical work has focused on contemporary negotiations between partners to deduce the best strategy of care and has rather neglected transgenerational costs of the sort we have identified here . Therefore to address this question , we needed to develop new theory . Our theoretical work showed that the threat of exploitation during biparental care in the next generation changes current interactions between parents and their young . It limits the potential benefits associated with the provision of care and so reduces the evolutionarily stable level of care currently supplied to offspring ( Figure 5 ) . Thus , our model identifies a third way in which parental effects can influence the adaptive value of behavioural traits . It shows that when behavioural traits are themselves parental effects , and can change the adaptive value of a behavioural trait in the next generation , then evolutionarily stable levels of behaviour in the current generation will change accordingly as well . Could these transgenerational effects still persist , even under assortative mating ? Although our model assumes random mating with respect to parental quality , we expect that our results would remain qualitatively unchanged under assortative mating . Assortative mating would increase the benefits of investing in good quality parents-to-be , because good parents would on average find better partners . We would consequently expect the average level of care provided to be higher when there is assortative mating , rather than random mating , and to increase more steeply with the strength of parental effects . Nonetheless , even when mating is highly assortative , there is still sexual conflict over the level of care provided by each parent , and this should reduce the total care provided . The evolution of adaptive behaviour depends both on a mechanism for inheritance of behavioural traits from generation to generation and on the net fitness gained by performing that behaviour . We have shown here that parental effects not only influence the inheritance of a behavioural trait ( because well-cared for larvae mature into high-quality parents ) but also alter the fitness associated with performing that behaviour—and in three contrasting , yet interconnected ways . Thus , our experiments show how parental effects can potentially provide a mechanism for the rapid evolution of new adaptive social behaviour in the face of environmental change ( see also Badyaev and Uller , 2009 ) . Whether parental effects similarly change the fitness costs and benefits associated with other social behaviours that might also function as parental effects , such as cooperative breeding , an individual's position in a social network , or even interspecific mutualisms , remains an exciting challenge for future work .	The burying beetle is an unusual insect in that both the father and the mother take care of their young larvae . They do this by providing food in the form of a small dead animal , such as a mouse , from which they diligently remove any fur or feathers , and by defending both the food and the larvae from rivals . These actions reduce the fitness of the parents , which can be estimated by measuring by how long they survive after caring for their brood . They also increase the health of the larvae , as measured by how large the larvae are when they move away from the carcass to pupate . Kilner et al . wanted to know how the parenting received by larvae affects their behaviour when they grow up and have their own offspring . Larvae were given varying amounts of care , ranging from none at all to five days ( which is the typical length of the larval stage for burying beetles ) . Larvae that received little or no care grew up to become low-quality parents , whereas those that received lots of care became high-quality parents . A low-quality parent is , by definition , a parent that becomes less fit as a result of rearing offspring; a high-quality parent providing the same amount of care would not suffer such a large reduction in its fitness . Each of the female beetles from this first experiment was then mated with a high-quality male and together they took care of their offspring . Kilner et al . observed that the fathers lived longer when they were paired with high-quality mothers than they did when they were paired with lower quality mothers . This happened because the lower quality mothers effectively exploited the fathers , forcing them to do more of the parenting . Although the males gained by raising healthy larvae , they paid a price by dying at a younger age . Results from these insect experiments are not directly linked to human behaviour , but they might tell us why animals of other species are generally so careful to choose a mate that matches them in quality . In this way , they can avoid being exploited when the pair work together to raise young . In future , Kilner et al . will investigate how beetles adjust their parenting effort in response to the effort put in by their partner: can they estimate parental quality directly , or do they simply observe how much care the other partner is providing ?	[ "Abstract", "Introduction", "Materials", "and", "methods", "Results", "Discussion" ]	[ "ecology" ]	2015	Parental effects alter the adaptive value of an adult behavioural trait
A major puzzle in biology is how mammalian sperm maintain the correct swimming direction during various phases of the sexual reproduction process . Whilst chemotaxis may dominate near the ovum , it is unclear which cues guide spermatozoa on their long journey towards the egg . Hypothesized mechanisms range from peristaltic pumping to temperature sensing and response to fluid flow variations ( rheotaxis ) , but little is known quantitatively about them . We report the first quantitative study of mammalian sperm rheotaxis , using microfluidic devices to investigate systematically swimming of human and bull sperm over a range of physiologically relevant shear rates and viscosities . Our measurements show that the interplay of fluid shear , steric surface-interactions , and chirality of the flagellar beat leads to stable upstream spiralling motion of sperm cells , thus providing a generic and robust rectification mechanism to support mammalian fertilisation . A minimal mathematical model is presented that accounts quantitatively for the experimental observations . During their journey from ejaculation to fertilisation , human spermatozoa have to find and maintain the right swimming direction over distances that may exceed their head-to-tail length ( ∼100 μm ) by a 1000-fold . On their path to the egg cell , mammalian sperm encounter varied physiological environments and are exposed to a variety of chemical gradients and must overcome counterflows . Whilst chemotactic sensing ( Kaupp et al . , 2008 ) is assumed to provide important guidance in the immediate vicinity of the ovum ( Spehr et al . , 2003 ) , it is not known which biochemical ( Brenker et al . , 2012 ) or physical mechanisms ( Winet et al . , 1984 ) keep the sperm cells on track as they pass through the rugged landscapes of cervix , uterus , and oviduct ( Katz et al . , 2005; Eisenbach and Giojalas , 2006; Suarez and Pacey , 2006 ) . The complexity of the mammalian reproduction process and not least the lack of quantitative data make it very difficult to assess the relative importance of the various proposed long-distance navigation mechanisms ( Eisenbach and Giojalas , 2006; Fauci and Dillon , 2006 ) , ranging from cervix contractions ( Fauci and Dillon , 2006; Suarez and Pacey , 2006 ) and chemotaxis ( Kaupp et al . , 2008 ) to thermotaxis ( Bahat et al . , 2003 ) and rheotaxis ( Miki and Clapham , 2013 ) . Aiming to understand not only qualitatively but also quantitatively how fluid-mechanical effects may help steer mammalian spermatozoa over large distances , we report here a combined experimental and theoretical study of sperm swimming in microfluidic channels , probing a wide range of physiologically relevant conditions of shear and viscosity . For both human and bull spermatozoa , we find that their physical response to shear flow , combined with an effective shape-regulated surface attraction ( Kantsler et al . , 2013 ) and head–tail counter-precession , favours an upstream spiralling motion along channel walls . The robustness of this fluid-mechanical rectification mechanism suggests that it is likely to play a key role in the long-distance navigation of mammalian sperm cells . Thus , the detailed analysis reported below not only yields new quantitative insights into the role of biophysical processes during mammalian reproduction but could also lead to new diagnostic tools and improved artificial insemination techniques ( Merviel et al . , 2010 ) . Recent experiments on red abalone ( Riffell and Zimmer , 2007; Zimmer and Riffell , 2011 ) , a large marine snail that fertilises externally , showed that weak fluid flows can be beneficial to the reproduction of these organisms , suggesting that shear flows could have acted as a selective pressure in gamete evolution . In higher organisms , which typically fertilise internally , sperm transport is much more complex and the importance of shear flows relative to chemotaxis , peristaltic pumping , or thermotaxis still poses an open problem as it is difficult to perform well-controlled in vivo studies . The complex uterine and oviduct topography ( Suarez and Pacey , 2006 ) and large travelling distances render it unlikely that local chemotactic gradients steer sperm cells during the initial and intermediate stages of the sexual reproduction process . Experimental evidence ( Kunz et al . , 1996 ) suggests that rapid sperm transport right after insemination is supported by peristaltic pumping driven by muscular contractions of the uterus , but it is not known how sperm navigate in the oviduct . Thermotaxis , the directed response of sperm to local temperature differences , was proposed as a possible long-range rectification mechanism of sperm swimming in rabbits ( Bahat et al . , 2003 ) , but recently questioned ( Miki and Clapham , 2013 ) as it is likely to be inhibited by convective currents that form in the presence of temperature gradients . On the other hand , it has long been known that , similar to bacteria ( Marcos et al . , 2012 ) and algae ( Chengala et al . , 2013 ) , mammalian sperm ( Adolphi , 1905; Roberts , 1970 ) are capable of performing rheotaxis , by aligning against a surrounding flow ( Marcos et al . , 2012; Miki and Clapham , 2013 ) , but this effect has yet to be systematically quantified in experiments ( Fauci and Dillon , 2006; Suarez and Pacey , 2006 ) . Specifically , it is not known at present how sperm cells respond to variations in shear rate and viscosity , and how long they need to adapt to temporal changes in the flow direction . Answering these questions is essential for understanding which physical effects may be important at different stages of the mammalian fertilisation process . To quantify the swimming strategies of sperm cells under well-controlled flow conditions , we performed a series of microfluidic experiments in cylindrical and planar channels ( Figure 1 ) , varying systematically shear rates γ˙ and viscosities μ through the physiologically and rheotactically relevant regime , up to μ = 20 mPa·s which is roughly 10× the viscosity of natural seminal fluid ( Owen and Katz , 2005 ) . These measurements revealed the interesting result that both human and bull spermatozoa do not simply align against the flow , but instead swim upstream on spiral-shaped trajectories along the walls of a cylindrical channel ( Figure 1A; and Video 1 ) . The previously unrecognised transversal velocity component can be attributed to the chirality of the flagellar beat . The resulting helical swimming patterns enable the spermatozoa to explore collectively the full surface of a cylindrical channel , suggesting that rheotaxis can help sperm to navigate their way through the oviduct and find the egg cell ( Miki and Clapham , 2013 ) . Using high-speed imaging , we also determined the dynamical response of human and bull spermatozoa to flow reversal at different viscosities , which is essential for understanding how active swimming , rheotaxis , and uterine peristalsis can combine to facilitate optimal sperm transport . To rationalise the experimental observations , we identify below a simple mathematical model that reproduces the main results of our measurements . 10 . 7554/eLife . 02403 . 003Figure 1 . Sperm swim on upstream spirals against shear flow . ( A ) Background-subtracted micrograph showing the track of a bull sperm in a cylindrical channel ( viscosity μ = 3 mPa·s shear rate γ˙=2 . 1s−1 ) , channel boundary false-coloured with black , see Video 1 for raw data . ( B ) Schematic representation not drawn to scale . The conical envelope of the flagellar beat holds the sperm close to the surface ( Kantsler et al . , 2013 ) . The vertical flow gradient exerts a torque that turns the sperm against the flow , but is counteracted by a torque from the chirality of the flagellar wave , resulting in a mean diagonal upstream motion . ( C ) Tracks of bull sperm near a flat channel surface . ( D ) Upstream and transverse mean velocities 〈vy , x〉 vs shear flow speed u20 at 20 μm from the surface for different viscosities . All velocities are normalised by the sample mean speed v0μ at γ˙=0 . For human sperm , in order of increasing viscosity v0μ = 53 . 5 ± 3 . 0 , 46 . 8 ± 3 . 7 , 36 . 8 ± 3 . 3 , 29 . 7 ± 3 . 9 μm/s , and for bull sperm v0μ = 70 . 4 ± 11 . 8 , 45 . 6 ± 4 . 7 , 32 . 4 ± 4 . 8 , 29 . 6 ± 4 . 1 μm/s , where uncertainties are standard deviations of mean values from different experiments . Each data point is an average over >1000 sperms . ( E ) Histograms for selected points in ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02403 . 00310 . 7554/eLife . 02403 . 004Video 1 . Human sperm cell swimming on a spiral trajectory ( green ) against a shear flow in a cylindrical channel ( fluid viscosity 3 mPa·s; channel diameter 300 μm; channel boundaries marked in red ) . Scale bar 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02403 . 004 In the experiments , samples of human and bull spermatozoa were injected into microfluidic channels of spherical or rectangular cross-section ( ‘Materials and methods’ ) . The cells were then exposed to well-defined Poiseuille shear flows , corresponding to parabolic flow profiles ( Figure 1B ) . Even in the absence of flow , sperm cells tend to accummulate at surfaces ( Rothschild , 1963; Denissenko et al . , 2012 ) due to a combination of steric repulsion ( Kantsler et al . , 2013 ) and hydrodynamic forces ( Fauci and McDonald , 1995; Elgeti et al . , 2010; Friedrich et al . , 2010; Gaffney et al . , 2011; Montenegro-Johnson et al . , 2012 ) . This can be explained by the fact that , in essence , the flagellar beat traces out a cone which , upon collision , aligns with a solid surface , so that the sperm's propulsion vector points into the boundary and the cells become effectively trapped at the surface ( Kantsler et al . , 2013 ) . In the presence of a Poiseuille shear flow , cells close to the channel boundaries experience an approximately linear vertical flow profile , whose slope is given by the shear rate γ˙ ( Figure 1B ) . To quantify the effects of shear rate and viscosity on sperm swimming , we tracked a large number of individual cells ( typically N >10 , 000 ) in planar microfluidic channels ( Figure 1C ) at different shear rates γ˙ , ranging from 0 . 2 s−1 to 9 s−1 , and different dynamic viscosities μ , ranging from 1 mPa·s ( that of water ) to 20 mPa·s ( ‘Materials and methods’ ) . The cell tracks were then used to reconstruct the velocities of sperm swimming close to the boundary . Mean values and histograms of the upstream and transverse velocity components from those measurements are summarised in Figure 1D , E . Since sperm motility depends on viscosity and may vary among different samples , it is advisable to normalise sperm velocities v = ( vx , vy ) , that have been measured at different values of γ˙ and μ , by the mean sample speed v0μ=〈\|v\|〉μ at zero shear γ˙=0 , and also to rescale the flow velocity accordingly . Figure 1D shows the thus-normalised mean upstream and mean transverse swimming velocities 〈vy , x〉μ/v0μ for bull and human spermatozoa as a function of the dimensionless rescaled shear flow speed u20=γ˙ ( 2A/v0μ ) , where A = 10 μm is the approximate amplitude of a typical flagellar beat ( i . e . , the maximum distance of the flagellar tip from the surface during a beat is 2A ) . The results for the upstream velocity reveal that both human and bull sperm exhibit optimal upstream swimming at rescaled flow speeds u20 ∼ 1 , implying that there is an optimal shear regime for the rectification of sperm swimming . Remarkably , however , we also find that , at low viscosities ( μ ≪10 mPa·s ) , human spermatozoa exhibit a substantial shear-induced transverse velocity component that becomes suppressed at very high viscosities . By contrast , for bull spermatozoa , the mean transverse component is generally weaker and less sensitive to viscosity variations . These statements are also corroborated by the corresponding velocity histograms in Figure 1E . Qualitatively , the above observations for stationary shear flows can be explained as follows . Once a sperm cell has become trapped at a surface , its tail explores , on average , regions of higher flow velocity than the head , resulting in a net torque that turns the head against the flow ( Figure 1B ) . This shear-induced rectification is counter-acted by variability in the cells swimming direction . If the shear velocity is too low the orientational ‘noise’ , which is caused by a combination of intrinsic fluctuations in the cells' swimming apparatus , thermal fluctuations , and elastohydrodynamic effects , inhibits upstream swimming , whereas if the shear velocity becomes too large the sperm will simply be advected downstream by the flow , implying that there exists an optimal intermediate shear rate for upstream swimming . Interestingly , we find that the maximum of the upstream velocity decreases more strongly with viscosity for bull sperm than for human sperm ( Figure 1D ) . This could be due to differences in cell morphology , as previous numerical studies ( Smith et al . , 2011 ) for bacterial cells suggest that differences in head shape can substantially alter swimming behavior . Bull sperm have a flatter head than human sperm , which likely suppresses the rotational motion of the cell at high viscosities thus leading to an effectively smaller vertical beat amplitude A . This could explain why , at high values of μ , the tail beat of bull sperm becomes essentially two-dimensional and constricted to the vicinity of the surface , so that alignment against the flow becomes less efficient . To understand the unexpectedly strong transverse velocity component of human sperm at μ ≲5 mPa·s , as typical of the seminal fluid ( Owen and Katz , 2005 ) , it is important to recall that sperm of invertebrae and mammals are known to exhibit different chiral beat patterns depending on environmental conditions ( Gibbons , 1982; Ishijima and Hamaguchi , 1993; Woolley and Vernon , 2001; Smith et al . , 2009 ) , and that shear flows are capable of separating particles along the transverse direction according to their chirality ( Marcos et al . , 2009; Talkner et al . , 2012 ) . Human sperm exhibit a strong helical beat component at low-to-moderate values of μ ( Video 2 ) , but this chirality becomes suppressed at high viscosities ( Video 3 ) resulting in more planar wave forms ( Smith et al . , 2009 ) . For comparison , the beat of a bull sperm flagellum is more similar to a rigidly rotating planar wave even at low viscosities ( Video 4 ) , thus exhibiting a weaker chirality and leading to smaller transverse velocities ( Figure 1D ) . Since the flagellar beating pattern can be controlled not only by viscosity but also by changes in calcium concentration ( Ishijima and Hamaguchi , 1993 ) , higher organisms appear to possess several means for tuning transverse and upstream swimming of sperm . 10 . 7554/eLife . 02403 . 005Video 2 . Human sperm cells swimming in a low-viscosity fluid ( 3 mPa·s ) near the wall of a planar channel . The video shows that , at low viscosity , the flagellar beat of a human sperm cell typically exhibits a considerable chiral component . This follows from the fact that the flagellum never appears as a straight line ( in contrast to bull sperms at same viscosity , compare Video 4 ) . Scale bar 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02403 . 00510 . 7554/eLife . 02403 . 006Video 3 . Human sperm cells swimming in a high-viscosity fluid ( 20 mPa·s ) near the wall of a planar channel . The video shows that , at very high viscosity , the chiral beat component becomes considerably weaker for there now exist instances where the flagellum appears as an almost straight line , indicating that the beat pattern approaches the shape of a planar rotating wave . Scale bar 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02403 . 00610 . 7554/eLife . 02403 . 007Video 4 . Bull sperm cells swimming in a low-viscosity fluid ( 3 mPa·s ) near the wall of a planar channel . The video shows that , even at low viscosity , the flagellar beat of bull sperm is approximately planar . This follows from the fact that at certain instances the flagellum appears as a line ( in contrast to human sperms at same viscosity , compare Video 2 ) . Scale bar 20 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 02403 . 007 In addition to typically outward directed mucus flow in the oviduct , sperm cells are also exposed to temporally varying flows driven by uterine contractions ( Fauci and Dillon , 2006; Suarez and Pacey , 2006 ) . To probe the dynamical response of sperm to changes in the flow direction , we performed additional experiments , where we tracked the motion of bull and human spermatozoa after a sudden flow reversal at two different viscosities ( Figure 2; Videos 5 , 6 ) . In those experiment , sperm were first given time to align against a stationary shear flow , then the flow direction was reversed , uy→−uy , with a switching time <1 s . Upon flow reversal , a sperm cell typically performs a U-turn ( Figure 2A , B ) . The characteristic radius of curvature of the trajectory and the typical turning time τ were found to increase strongly with viscosity . At low viscosity , μ ∼ 1 mPa·s , sperm realign rapidly against the new flow direction with a typical response time of τ ∼ 5 s to 10 s , and the curvature radius is of the order of one or two sperm lengths ℓ ∼ 60 μm ( Video 5 ) . By contrast , at a larger viscosity of μ ∼ 12 mPa·s , which is roughly 4× higher than the natural viscosity of the ejaculate , both curvature radius and response time increase by approximately a factor of 5 ( Video 6 ) . Interestingly , these response times are of the order of typical cervical contractions ( Kunz et al . , 1996 ) , suggesting a possible fine-tuning between muscular activity of the uterus and turning behavior of sperm cells . In particular , immediately after the flow reversal , sperm orientation and flow direction point for a short period of time in approximately the same direction , leading to a momentarily increased transport velocity ( see velocity peaks in Figure 2C ) . Thus , by switching flow directions back and forth at an optimal rate , the transport efficiency of an initially rectified sperm population can be enhanced . 10 . 7554/eLife . 02403 . 008Figure 2 . Temporal response of sperm cells to a reversal of the flow direction depends sensitively on viscosity . ( A ) At low viscosity , sperm perform sharp U-turns , see also Video 2 . ( B ) At high viscosity , the typical radius of the U-turns increases substantially ( Video 3 ) . White/black arrows show orientations of several cells before/after turning . ( C ) Flow velocity at distance 5 μm from the channel surface ( blue , ‘Flow’ ) , mean upstream velocity 〈vy〉 ( red , ‘Up’ ) and mean transverse velocity 〈vx〉 ( green , ‘Trans’ ) as function of time . The typical response time of sperm cells after flow reversal increases with viscosity . Peaks reflect a short period when mean swimming direction and flow direction are aligned . The time series for human sperm also signal a suppression of the beat chirality at high viscosity , consistent with Figure 1D . DOI: http://dx . doi . org/10 . 7554/eLife . 02403 . 00810 . 7554/eLife . 02403 . 009Video 5 . Reorientation of a human sperm cell swimming in a low-viscosity fluid ( 1 mPa·s ) in a planar channel , after a sudden reversal of the flow direction at time t = 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 02403 . 00910 . 7554/eLife . 02403 . 010Video 6 . Reorientation of two human sperm cells , swimming in a high-viscosity fluid in a planar channel , after a sudden reversal of the flow direction at time t = 0 . DOI: http://dx . doi . org/10 . 7554/eLife . 02403 . 010 To test whether our understanding of the experimental observations is correct and to provide a basis for future theoretical studies , we used resistive force theory to infer a minimal mathematical model that incorporates the main physical mechanisms discussed above ( details are provided in the Supplementary file 1 ) . The model assumes that the effective two-dimensional motion of a sperm cell , that swims close to a surface in the presence of a shear flow can be described in terms of its position vector R ( t ) = ( X ( t ) , Y ( t ) ) and its orientation unit vector N ( t ) = ( Nx ( t ) , Ny ( t ) ) . Focussing on an effective description of the main physical effects and assuming that the flow is in y-direction ( Figure 1B ) , the equations of motions for R and N read ( 1 ) R˙=VN+σU¯ey , ( 2 ) N˙=σγ˙α ( NxNyNy2−1 ) +σγ˙χβ ( Nx2−1NxNy ) + ( 2D ) 1/2 ( I−NN ) ·ξ ( t ) . Equation 1 states that the net in-plane velocity R˙ ( t ) of a cell arises from two main contributions: self-swimming at typical speed V in the direction of the cell orientation N , and advection by the flow , where σ = ±1 defines the flow direction and U¯>0 the mean flow speed experienced by the cell . As explained in detail in the Supplementary file 1 , the nonlinear structure of Equation 2 ensures that the length of the orientation vector N remains conserved , assuming that the change in orientation , N˙ ( t ) , is caused by three effects: shear-induced alignment against the flow with rate γ˙α , where α >0 is numerical factor that encodes geometry of the flagellar beat , shear-and-chirality-induced turning at rate γ˙β with χ∈{−1 , 0 , +1} and β >0 encoding chirality and shape of the flagellar beat , and variability ( Su et al . , 2012 ) in the swimming direction , modeled as a Stratonovich-type two-dimensional Gaussian white noise ξ with amplitude D ( Han et al . , 2006 ) . Equations 1 and 2 were obtained by approximating the flagellum by a rigid conical helix , with the polar geometry of the enveloping cone dictating the mathematical structure of the deterministic turning terms ( see Supplementary file 1 for details of the calculation ) . The simplifying assumptions underlying Equations 1 and 2 imply that this minimal model does not accurately capture the dynamics of individual cells at zero shear , as the deterministic terms in Equation 2 neglect the intrinsic curvature of cell trajectories . However , when analyzing the in-plane curvature for a large number of human sperm trajectories ( >100 , 000 sample points from more than 1200 cells ) at zero shear , we found a broad distribution of curvatures with a small positive mean curvature of ( 5 . 6 ± 1 . 3 ) ·10−4 μm at low viscosity ( 1 mPa·s ) and a small negative mean curvature ( −1 . 9 ± 0 . 1 ) ·10−3 μm at high viscosity ( 12 mPa·s ) , where the different signs are consistent with the observed change in the transverse velocity for human sperm at high viscosity ( Figure 1D ) . To account at least partially for these curvature variations , we include in Equation 2 the Gaussian white noise term . Compared with more accurate models that resolve the details of the flagellar dynamics ( Elgeti et al . , 2010; Gaffney et al . , 2011 ) , Equations 1 and 2 provide a strongly reduced description which , however , turns out be sufficient for rationalising our experimental observations ( Figure 3 ) . Values for V and U¯ can be directly estimated from experiments , and sign conventions in Equation 2 have been chosen such that χ = +1 for human sperm at low viscosity ( for weakly chiral bull sperm one can use χ = 0 in a first approximation ) . The model parameters ( α , β , D ) can be inferred from the experimental data ( Supplementary file 1 ) . By performing systematic parameter scans , we found that values α∈[0 . 2 , 0 . 4] , β∈[0 . 05 , 0 . 1] and D∈[0 . 2 , 0 . 3] rad/s yield good quantitative agreement with the experimental results for both stationary flow ( Figure 3A ) and flow reversal ( Figure 3B ) , suggesting that the coupling between shear flow and beat chirality dominates the transverse velocity dynamics . We may therefore conclude that , despite some strong simplifications , the effects included in the model capture indeed the main physical mechanisms relevant for understanding sperm motion in shear flow near a surface . 10 . 7554/eLife . 02403 . 011Figure 3 . Model simulations reproduce main experimental observations . ( A ) Upstream and transverse velocity for different values of the variability ( effective noise ) parameter D in units rad/s and dimensionless shape factors ( α , β ) . ( B ) Time response of a chiral swimmer with χ = +1 ( ‘Human’ ) and a non-chiral swimmer with χ = 0 ( ‘Bull’ ) to a reversal of the flow direction at time t = 0 . Blue dashed line shows fluid flow uy at 5 μm from the boundary . Simulation parameters ( N = 1000 trajectories , A = 10 μm , ℓ = 60 μm , V = 50 μm/s ) were chosen to match approximately those for viscosity 1 mPa·s in Figure 2C . DOI: http://dx . doi . org/10 . 7554/eLife . 02403 . 011 In conclusion , we have reported detailed quantitative measurements of sperm motion in shear flow . Our experimental results show that upstream swimming of mammalian sperm due to rheotaxis is more complex than previously thought . Human sperm cells were found to exhibit a significant transverse velocity component that could be of relevance in the fertilisation process , as the ensuing spiralling motion enables spermatozoa to explore collectively a larger surface area of the oviducts , thereby increasing the probability of locating egg cells . Our theoretical analysis implies that the transverse velocity component arises from a preferred handedness in the flagella beat in the presence of shear flow , in contrast to recent findings for male microgametes of the malaria parasite Plasmodium berghei ( Wilson et al . , 2013 ) . Due to the large sample size , our data provide substantial statistical evidence for the hypothesis that mammalian sperm have evolved to achieve optimal upstream swimming near surfaces , possibly exploiting the enhanced fluid production in the female reproductive system during the fertile phase ( Eschenbach et al . , 2000 ) and after intercourse ( Miki and Clapham , 2013 ) . The improved quantitative knowledge derived from this data may help to design more efficient artificial insemination strategies , for example , by optimising the viscosity and chemical composition of fertilisation media and adjusting injection techniques to maximise upstream swimming of sperm cells . Combined with recent measurements ( Kantsler et al . , 2013 ) , which clarified the importance of flagella-mediated contact interactions for the accumulation of sperm cells at surfaces , the results presented here yield a cohesive picture of the mechanistic and fluid-mechanical ( Friedrich et al . , 2010 ) aspects of long-distance sperm navigation . Future work should focus on merging these insights with quantitative studies of chemotaxis ( Spehr et al . , 2003; Kaupp et al . , 2008; Zimmer and Riffell , 2011; Brenker et al . , 2012 ) to obtain a differentiated understanding of the interplay between physical and chemical factors during various stages of the mammalian reproduction process . Cryogenically frozen bull spermatozoa were purchased from Genus Breeding . For each experiment , a bull sperm sample of 250 μl was thawed in a water bath at 37°C for 15 s . Human samples from healthy undisclosed normozoospermic donors were obtained from Bourn Hall Clinic . Donors provided informed consent in accordance with the regulations of The University of Cambridge Human Biology Research Ethics Committee . For each experiment , bull and human samples were washed three times by centrifugation at 500 rcf for 5 min with the appropriate medium . The medium for bull spermatozoa contained 72 mM KCl , 160 mM sucrose , 2 mM Na-pyruvate , and 2 mM Na-phosphate buffer at pH 7 . 4 ( Woolley and Vernon , 2001 ) . Human sperm medium was based on a standard Earle's Balanced Salt Solution , containing 66 . 4 mM NaCl , 5 . 4 mM KCl , 1 . 6 mM CaCl2 , 0 . 8 mM MgSO4 , N2H2PO4 1 mM , NaHCO3 26 mM , D-Glucose 5 . 5 mM supplemented with 2 . 5 mM Na pyruvate and 19 mM Na-lactate pH adjusted to 7 . 2 by bubbling the medium with CO2 . Viscosity of the medium was modified by adding methylcellulose ( M0512; Sigma-Aldrich; St . Louis , MO; approximate molecular weight 88 , 000 ) at concentrations 0% , 0 . 2% , 0 . 4% , 0 . 5% wt/vol . The absence of circular trajectories at zero-shear implies that the sperm are capacitated ( Miki and Clapham , 2013 ) . Microfluidic channels were manufactured using standard soft-lithography techniques . The master mould was produced from SU8 2075 ( MicroChem Corp . ; Newton , MA ) spun to a 340 microns thickness layer and exposed to UV light through a high resolution mask to obtain the desired structures . The microfluidic chip containing the channels cast from PDMS ( Sylgard 184; Dow Corning; Midland , MI ) and bonded to covered glass . The channel has rectangular cross-section of 0 . 34 × 3 mm . We treated PDMS surfaces of the channels prior the experiment with 10% ( wt/vol ) Polyethylene glycol ( m . w . 8000; Sigma ) solution in water for 30 min to avoid adhesion of sperm cells to the walls . Sperm suspension was introduced through inlets with a micro-syringe pump ( Harvard Apparatus; Kent , UK ) at controlled flow rates of 0 . 1–40 μl/min . The concentration of the sperm cells in the experiments was kept below 1% volume fraction . To identify the swimming characteristics of individual sperm cells , the trajectories were reconstructed by applying a custom-made particle-tracking-velocimetry ( PTV ) algorithm to image data taken with a Zeiss Axio Observer inverted microscope ( 20x or 10x objective , 25 fps ) . The flagella dynamics was captured with a Fastcam SA-3 Photron camera ( San Diego , CA; 125 fps , 40x/NA 0 . 6 objective ) . Calibration of the velocity profile in the channel was performed by measuring trajectories of fluorescent beads for different distances from the coverslip via PTV . The measured velocity profile is found identical to the calculated values from solving the Stokes equations for the given geometry . Values of the shear rate γ˙ in the different experiments were reconstructed from the flow velocity at distance 20 μm from the wall ( see below ) . Effects of viscosity variation and shear-rate variation were studied in experiments that were performed in a rectangular channel with a cross-section 0 . 34 × 3 mm , by observing sperm motion at lower and upper channel walls . The field of view ( normally 800 × 800 μm ) was chosen at the middle of the channel ( in x-direction ) , where the in-plane velocity gradient is negligible due to the high aspect ratio of the channel ( Figure 4B ) . The vy-velocity profiles , measured along the z-coordinate , were found to be in perfect agreement with the theoretically expected parabolic flow profile for this geometry ( Figure 4B ) . The shear rate γ˙ at a given flow rate was determined from the flow velocity at distance 20 μm from the wall . The depth of field of the objective was <5 μm to ensure that we only observed cells that swam close to the surface . Trajectories of individual sperm cells were analysed in MATLAB . The sample size in a single experiment exceeds 100 , 000 velocity vectors , each measurement for a given viscosity and a shear rate was repeated a few times with different sperm samples . Supplemental data tables that summarise the statistical information for each experiment are given in Supplementary file 2 . 10 . 7554/eLife . 02403 . 012Figure 4 . ( A ) Schematic of the microfluidic channel and field of view ( turquoise region ) in the sperm motility measurements . ( B ) Velocity profile at the center of the channel . Red symbols are values of the vertical velocity profile vy ( z ) measured by PTV for the flow rate 0 . 1 μl/s . The solid line shows the theoretically calculated flow profile for the same flow rate . In motility experiments , values for the velocity gradient near the boundary ( pink region ) were obtained by measuring the flow velocity at 20 μm from the boundary . ( C ) Theoretical 2D flow speed profile in ( x , z ) -plane at flow rate 0 . 1 μl/s . DOI: http://dx . doi . org/10 . 7554/eLife . 02403 . 012	A sperm cell must complete a long and taxing journey to stand a chance of fertilising an egg cell . This quest covers a distance that is thousands of times longer than the length of a sperm cell . It also passes through the diverse environments of the cervix , the uterus and , finally , the oviduct , where there might be an egg to fertilise . How the sperm cells manage to stay on course over this distance is a mystery , although it has been suggested that many different factors , including chemical signals and fluid flow , are involved . The fluids that the sperm cells travel through are not static . Evidence suggests that contractions of the cervix and uterus help to pump sperm cells along the first part of their journey . However , mucus flows out of the oviduct in the opposite direction to way the sperm cells need to go . Sperm cells mostly move along the walls of the cervix , uterus , and oviduct . This means that sperm cells must contend with two properties of the fluids they travel through—the viscosity ( or ‘thickness’ ) of the fluid , and the fact that different parts of the fluid will flow at different speeds , depending on how close it is to the wall ( ‘shear flow’ ) . Kantsler et al . have now used a technique called microfluidics—which involves forcing tiny amounts of liquid to flow through very narrow channels—to study how the movement of human and bull sperm cells along a surface is affected by the viscosity and flow rate of the fluid they are swimming through . The sperm cells were found to swim upstream , moving along the walls of the channels in a spiral movement . This is likely to help the sperm cells to find the egg , because spiralling around the oviduct will increase the chances of meeting the egg . Kantsler et al . also built a mathematical model that describes how the sperm cells move . Although further work is needed to better understand the role played by chemical signals , understanding how fluid flow and viscosity influence sperm cells could lead to more effective artificial insemination techniques .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems" ]	2014	Rheotaxis facilitates upstream navigation of mammalian sperm cells
Antigen Ki-67 is a nuclear protein expressed in proliferating mammalian cells . It is widely used in cancer histopathology but its functions remain unclear . Here , we show that Ki-67 controls heterochromatin organisation . Altering Ki-67 expression levels did not significantly affect cell proliferation in vivo . Ki-67 mutant mice developed normally and cells lacking Ki-67 proliferated efficiently . Conversely , upregulation of Ki-67 expression in differentiated tissues did not prevent cell cycle arrest . Ki-67 interactors included proteins involved in nucleolar processes and chromatin regulators . Ki-67 depletion disrupted nucleologenesis but did not inhibit pre-rRNA processing . In contrast , it altered gene expression . Ki-67 silencing also had wide-ranging effects on chromatin organisation , disrupting heterochromatin compaction and long-range genomic interactions . Trimethylation of histone H3K9 and H4K20 was relocalised within the nucleus . Finally , overexpression of human or Xenopus Ki-67 induced ectopic heterochromatin formation . Altogether , our results suggest that Ki-67 expression in proliferating cells spatially organises heterochromatin , thereby controlling gene expression . The cell proliferation antigen Ki-67 ( Ki-67 or Ki67 ) is constitutively expressed in cycling mammalian cells ( Gerdes et al . , 1983 ) . It is therefore widely used as a cell proliferation marker to grade tumours . Ki-67 is a nuclear DNA-binding protein ( MacCallum and Hall , 2000 ) with two human isoforms that have predicted molecular weights of 320kDa and 359kDa ( Gerdes et al . , 1991 ) . The domain structure of Ki-67 is represented in Figure 1 . All homologues contain an N-terminal Forkhead-associated ( FHA ) domain , which can bind both to DNA and to phosphorylated epitopes . The most characteristic feature of Ki-67 is the presence of multiple tandem repeats ( 14 in mice , 16 in human ) containing a conserved motif of unknown function , the 'Ki-67 domain' . Two other conserved motifs include a Protein Phosphatase 1 ( PP1 ) -binding motif ( Booth et al . , 2014 ) and a 31 amino acid conserved domain ( CD ) of unknown function , 100% identical between human and mouse , that includes a 22 amino acid motif conserved in all homologues . Ki-67 homologues also have a weakly conserved leucine/arginine rich C-terminus which can bind to DNA and , when overexpressed , promotes chromatin compaction ( Scholzen et al . , 2002; Takagi et al . , 1999 ) . 10 . 7554/eLife . 13722 . 003Figure 1 . Comparison of human and mouse Ki-67 structural elements . ( A ) Top: cartoon of human ( long form ) and mouse Ki-67 protein highlighting conserved elements and functional motifs . Domains are indicated by boxes ( FHA , forkhead-associated domain; PP1 , PP1-binding domain; CD , conserved domain; D-box: APC/C targeting destruction box motifs; KEN: APC/C-Cdh1 targeting KEN box motifs ) . Highly conserved regions are indicated by dotted line with percent of identical amino acids . Bottom: alignment of mouse Ki-67 repeats . ( B ) APC/C targeting motifs identified in human ( both isoforms ) and mouse Ki-67 . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 003 Ki-67 protein levels and localisation vary through the cell cycle . Its maximum expression is found in G2 phase or during mitosis ( Endl and Gerdes , 2000b ) . In interphase , Ki-67 forms fibre-like structures in fibrillarin-deficient regions surrounding nucleoli ( Verheijen et al . , 1989b; Kill , 1996; Cheutin et al . , 2003 ) . Ki-67 also colocalises with satellite DNA ( Bridger et al . , 1998 ) and is found in protein complexes that bind to satellite DNA ( Saksouk et al . , 2014 ) . It remains associated with nucleolar organiser regions of acrocentric chromosomes throughout interphase ( Bridger et al . , 1998 ) . Ki-67 is a direct substrate of the cyclin-dependent kinase CDK1 ( Blethrow et al . , 2008 ) and is hyperphosphorylated in mitosis . This may regulate its expression and / or localisation ( Endl and Gerdes , 2000a ) . In HeLa cells , Ki-67 binds tightly to chromatin in interphase , whereas this binding is weakened in mitosis ( Saiwaki et al . , 2005 ) when it associates with condensed chromosomes before relocating to the chromosome periphery ( Verheijen et al . , 1989a ) . Ki-67 is generally assumed to be required for cell proliferation . An early study found that injection of antisense oligonucleotides that block Ki-67 expression inhibited cell proliferation ( Schluter et al . , 1993 ) . Subsequent studies have shown that unperturbed Ki-67 expression levels are required for normal proliferation rates in various cell lines ( Kausch et al . , 2003; Rahmanzadeh et al . , 2007; Zheng et al . , 2006; 2009 ) . However , no complete Ki-67 loss of function studies have been reported . Therefore , it remains unclear what its functions are and whether it is essential for cell proliferation . The dynamic localisation of Ki-67 has led to suggestions that it could coordinate nucleolar disassembly and reassembly at either side of mitosis ( Schmidt et al . , 2003 ) . Indeed , Ki-67 is required to localise nucleolar granular components to mitotic chromosomes , thereby potentially playing a role in nucleolar segregation between daughter cells ( Booth et al . , 2014 ) . It has also been reported that Ki-67 has roles in ribosome biogenesis ( Rahmanzadeh et al . , 2007 ) , consistent with its nucleolar localisation and apparent role in cell proliferation . In this work , we characterise the cellular roles of Ki-67 using knockdown and genetic approaches . We find that mutant mice with disrupted Ki-67 expression are viable and fertile . Preventing Ki-67 downregulation upon cell cycle exit in vivo does not impede differentiation . Thus , Ki-67 expression can be uncoupled from cell proliferation . Instead , we show that Ki-67 is an essential mediator of heterochromatin organisation and long-range chromatin interactions , controlling gene expression . As it is expressed at high levels only in proliferating cells , our results suggest that Ki-67 links heterochromatin organisation to cell proliferation . Given the tight correlation between Ki-67 expression and cell proliferation , it is often assumed that Ki-67 is required for cell proliferation and that its downregulation might promote cell cycle exit . We tested these hypotheses genetically . Ki-67 protein expression is regulated during the cell cycle , and we speculated that it might be a target for the APC/C-Cdh1 ubiquitin ligase complex . This complex is active in late mitosis and G1 , and triggers degradation of substrates containing D-boxes and KEN boxes . Human Ki-67 isoforms contain two or three KEN boxes , whereas mouse Ki-67 contains two D-boxes ( Figure 1A , B ) . Mouse Ki-67 contains an additional sequence , AQRKQPSR at 2680–2687 , which is highly similar to the A-box , a third APC/C-Cdh1 recognition motif ( Littlepage and Ruderman , 2002 ) . To see whether Cdh1 regulates Ki-67 , we analysed mouse embryo fibroblasts ( MEFs ) lacking the Fzr1 gene that encodes Cdh1 ( Garcia-Higuera et al . , 2008 ) . Asynchronous Fzr1 heterozygous MEFs , that are at different stages of the cell cycle , had variable Ki-67 levels , whereas in the Fzr1 knockout MEFs Ki-67 was upregulated and more homogeneously expressed ( Figure 2A ) . To see whether sustained Ki-67 expression in quiescent cells would have a negative impact on cell cycle arrest in vivo , we analysed Fzr1-knockout mice . Here , Ki-67 was overexpressed and uncoupled from cells that incorporate BrdU in all tissues examined ( Figure 2B , Figure 2—figure supplement 1 ) . Thus , Ki-67 expression is regulated by APC/C-Cdh1 in mice and its downregulation is not a prerequisite for cell cycle exit . 10 . 7554/eLife . 13722 . 004Figure 2 . Maintenance of Ki-67 expression in quiescent cells in vivo by Cdh1 mutation . ( A ) Immunofluorescence analysis of Ki-67 in MEF cells isolated from embryo ( E13 . 5 ) of Fzr1 ( +/Δ ) ;Sox2-Cre and Fzr1 ( -/Δ ) ;Sox2-Cre mice . Scale bar , 25 µm . ( B ) IHC staining of Ki-67 and BrdU in sagittal sections of embryo cerebellum ( E18 . 5 ) of Fzr1 ( +/Δ ) ;Sox2-Cre and Fzr1 ( -/Δ ) ;Sox2-Cre mice . Bars , 200 µm . Bars in zoom , 50 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 00410 . 7554/eLife . 13722 . 005Figure 2—figure supplement 1 . Ki-67 expression is restricted to proliferating cells by APC/C-Cdh1 . Top , IHC staining of Ki-67 and BrdU in sagittal section of embryo heart ( E18 . 5 ) of Fzr1 ( +/Δ ) ;Sox2-Cre and Fzr1 ( -/Δ ) ;Sox2-Cre mice . Scale bar , 500 µm . Bottom , IHC staining of Ki-67 and BrdU in sagittal section of embryo lung ( E18 . 5 ) of Fzr1 ( +/Δ ) ;Sox2-Cre and Fzr1 ( -/Δ ) ;Sox2-Cre mice . Scale bar , 200 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 005 We next investigated the functional consequences of Ki-67 downregulation for normal tissue development and homeostasis . To disrupt the gene encoding Ki-67 , Mki67 , in the mouse germline , we used a TALEN pair targeting the unique ATG start codon . This is predicted to generate null alleles ( Figure 3—figure supplement 1A ) . After cytoplasmic injection of these TALEN-encoding mRNAs into zygotes , 10 out of 54 mice had mutations disrupting the coding sequence . We crossed founder mutant mice , and four gave germline transmission of the mutation . Due to mosaicism , this resulted in six lines: two lines with mutations eliminating the initiation codon and four lines with deletions which cause frameshifts immediately downstream of the ATG ( Figure 3A ) . From these , we selected a 2-nucleotide deletion ( 2nt∆ ) mutant that retains the ATG initiation codon but has a frameshift in the next codon ( Figure 3—figure supplement 1B ) , and a 21-nucleotide deletion ( 21nt∆ ) that eliminates the ATG ( Figure 3—figure supplement 1C ) . We crossed these mice and , unexpectedly , obtained homozygous mutants at the expected Mendelian frequency that were indistinguishable from heterozygous or wild-type ( WT ) littermates ( Figure 3B , Figure 3—figure supplement 1D ) . Both deletion mutant lines showed normal growth and were fertile . Sagittal sections from Mki672nt∆/2nt∆ mice did not reveal any obvious defects in tissue morphology ( Figure 3—figure supplement 2 ) . Since the intestinal epithelium is the most highly proliferative adult mouse tissue , we compared its morphology between WT and mutant mice . In WT animals , the proliferative crypt compartment was strongly stained for Ki-67 by immunohistochemistry ( IHC ) , while only minimal levels of Ki-67 were observed in the differentiated cells on the villus ( Figure 3C , top ) , as expected . In contrast , in the mutants , proliferating crypt cells showed only residual levels of Ki-67 staining by IHC ( Figure 3C , bottom ) or immunofluorescence ( Figure 3—figure supplement 3 ) . Immunoblotting of intestinal epithelium preparations could detect a weak band of similar size to WT Ki-67 ( Figure 3D; Figure 3—figure supplement 4 ) . The signal was , however , reduced by at least 90% in both mutants compared to WT tissue . Three different Ki-67 antibodies gave similar results . These are all extremely sensitive as they recognise the highly repeated Ki-67 domain . They should also detect N-terminally truncated Ki-67 that would result from translation from the ATG at position 433 . qRT-PCR analysis showed that Ki-67 mRNA level was , unexpectedly , increased rather than reduced in the intestinal tissue ( Figure 3—figure supplement 5 ) . In the intestinal epithelium , analysis of Wnt signalling and differentiation of goblet and tuft cells showed no differences between WT and Mki6721nt∆/21nt∆ mice ( Figure 3—figure supplement 6 ) . These results show that high Ki-67 levels and an intact Ki-67 gene are not required for development or differentiation in vivo . 10 . 7554/eLife . 13722 . 006Figure 3 . Mouse development with a mutated Ki-67 gene . ( A ) Table describing Ki-67 mutant mouse lines resulting from germline transmission of mutations generated by cytoplasmic injection of TALEN-encoding mRNA into zygotes . ( B ) Macroscopic appearance of littermate female mice at 10 weeks of age . Genotypes are specified . ( C ) IHC staining of Ki-67 in sagittal section of intestine from Mki67WT/WT , Mki67WT/2nt∆ and Mki672nt∆/2nt∆ mice . ( D ) Western blots of Ki-67 and cyclin A expression from intestine isolated from Mki67WT/WT , Mki67WT/2nt∆ and Mki672nt∆/2nt∆ mice . LC , loading control . ( E ) Western blot of Ki-67 in MEFs from WT , Mki67WT/2nt∆ and Mki672nt∆/2nt∆ mice . LC , loading control . ( F ) Flow cytometry profiles in WT , Mki67WT/2nt∆ and Mki672nt∆/2nt∆ MEFs showing EdU incorporation upon a 1 hr pulse and DNA content . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 00610 . 7554/eLife . 13722 . 007Figure 3—figure supplement 1 . Ki-67 mutant mice develop normally . ( A ) Pair of TALE-nucleases designed to target the initiator ATG of mouse Mki67 gene . ( B ) Sequencing traces of initiator ATG ( underlined ) of Mki67 gene in WT Mki67+/+ ( WT/WT ) , heterozygous Mki67+/2nt∆ ( WT/2nt∆ ) and homozygous Mki672nt∆/2nt∆ ( 2nt∆/2nt∆ ) mice . ( C ) Sequencing traces of initiator ATG ( underlined ) of Mki67 gene in WT Mki67+/+ ( WT/WT ) , heterozygous Mki67+/21nt∆ ( WT/21nt∆ ) and homozygous Mki6721nt∆/21nt∆ ( 21nt∆/21nt∆ ) mice . ( D ) WT Mki67+/+ ( +/+ ) , heterozygous Mki67+/21nt∆ ( +/21nt∆ ) and homozygous Mki6721nt∆/21nt∆ ( 21nt∆/21nt∆ ) mice . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 00710 . 7554/eLife . 13722 . 008Figure 3—figure supplement 2 . Ki-67 mutant mice develop normally . Sagittal sections of the whole mouse , intestine , liver and heart from WT Mki67+/+ ( WT/WT ) , heterozygous Mki67+/2nt∆ ( WT/2nt∆ ) and homozygous Mki672nt∆/2nt∆ ( 2nt∆/2nt∆ ) mice . Mouse sections , bar 5 mm; intestine sections , bar 500 µm; liver sections , bar 1 mm; heart sections , bar 500 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 00810 . 7554/eLife . 13722 . 009Figure 3—figure supplement 3 . Background Ki-67 levels in Ki-67 mutant mice . Immunofluorescence of Ki-67 and β-catenin on sagittal sections of the mouse intestinal epithelium from Mki67+/+ , heterozygous Mki67+/21nt∆and homozygous Mki6721nt∆/21nt∆mice . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 00910 . 7554/eLife . 13722 . 010Figure 3—figure supplement 4 . Background Ki-67 levels in Ki-67 mutant mice . Immunoblotting of Ki-67 and cyclin A on protein preparations of intestinal epithelium from Mki67+/+ ( +/+ ) , heterozygous Mki67+/21nt∆ ( +/21nt∆ ) and homozygous Mki6721nt∆/21nt∆ ( 21nt∆/21nt∆ ) mice . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 01010 . 7554/eLife . 13722 . 011Figure 3—figure supplement 5 . Ki-67 mRNA levels are not reduced in Ki-67 mutant mice . qRT-PCR of relative Ki-67 mRNA expression , normalised to GAPDH , on preparations of intestinal epithelium from Mki67+/+ ( WT/WT ) , heterozygous Mki67+/21nt∆ ( WT/21nt∆ ) and homozygous Mki6721nt∆/21nt∆ ( 21nt∆/21nt∆ ) mice . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 01110 . 7554/eLife . 13722 . 012Figure 3—figure supplement 6 . Normal proliferation and differentiation in Ki-67 mutant mice . Sagittal sections of the intestine from WT Mki67+/+ and homozygous Mki6721nt∆/21nt∆mice sacrificed after a 2 hr BrdU pulse , immunostained for BrdU , β-catenin , DCLK1 ( a marker for tuft cells ) or PAS staining of mucopolysaccharides ( a marker for goblet cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 01210 . 7554/eLife . 13722 . 013Figure 3—figure supplement 7 . Ki-67 mRNA levels are not reduced in MEFs from Ki-67 mutant mice . qRT-PCR of relative Ki-67 mRNA expression , normalised to GAPDH , in MEFs isolated from Mki67+/+ ( WT/WT ) , heterozygous Mki67+/2nt∆ ( WT/2nt∆ ) and homozygous Mki672nt∆/2nt∆ ( 2nt∆/2nt∆ ) mice , compared to NIH3T3 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 01310 . 7554/eLife . 13722 . 014Figure 3—figure supplement 8 . Generation of Ki-67-mutant NIH-3T3 cells . ( A ) Top and left , schematic representation of TALEN-mediated generation of NIH-3T3 biallelic Ki-67 mutant cells . Top right , immunofluorescence of Ki-67 and eGFP staining in NIH 3T3 cells transfected with two plasmids encoding TALEN pair and plasmid pEGFP . White arrows show Ki-67-negative pEGFP cotransfected cells . Scale bar 25 µm . ( B ) PCR analysis of Mki67 initiator ATG surrounding sequence in genomic DNA prepared from three WT clones and nine Ki-67 immunofluorescence-negative clones selected for further analysis . ( C ) sequencing of Mki67 initiator ATG area from selected clones ( 14 , 19 , 21 , 33 , 38 ) . ( D ) , PCR analysis targeted to the initiator ATG in Mki67 gene of genomic DNA purified from NIH 3T3 WT clone W4 and Ki-67 KO clones 14 , 19 , 21 , 33 , 38 . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 01410 . 7554/eLife . 13722 . 015Figure 3—figure supplement 9 . Ki-67-mutant NIH-3T3 cells proliferate normally . ( A ) Left , analysis of the indicated protein levels by Western blotting in NIH 3T3 WT clone 4 ( W4 ) and Ki-67 mutant clones 14 , 21 and 38 . LC , loading control . ( B ) Left , growth curves of NIH 3T3 WT clone W4 and Ki-67 mutant clones 14 , 21 and 38 . NIH 3T3 WT and mutant cells were counted every day for 4 days . Right , cell cycle distribution of the WT clone W4 and 14 , 21 , 38 Ki67 mutant clones as analysed by FACS . ( C ) qRT-PCR analysis of Ki-67 mRNA in NIH 3T3 WT clone W4 and NIH Ki-67 mutant clones 14 , 21 and 38 . Normalized by mRNA expression of B2m ( beta-2-microglobulin ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 01510 . 7554/eLife . 13722 . 016Figure 3—figure supplement 10 . Basal translation of Ki-67 with a mutated ATG . Left: quantification of association of Ki-67 mRNA with ribosome fractions in biallelic single TALEN Ki-67 mutant NIH-3T3 cells by qRT-PCR , normalised to GAPDH , b2-microglobulin and actin mRNA and to WT levels . Right: RNA quantification of fractions by spectrophotometric absorbance at 254 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 01610 . 7554/eLife . 13722 . 017Figure 3—figure supplement 11 . SILAC proteomics analysis of Ki-67 expression with a mutated ATG . ( A ) Scatterplot showing H/L m/z peak ratios ( x-axis ) against intensity levels ( y-axis ) from band 1 of SDS-PAGE purified chromatin , where H-labelled samples were Ki-67 mutant NIH-3T3 clone 14 , L samples were W4 wild-type control . Most proteins have unaltered expression between the two cell lines ( H/L ratio ≈ 1 ) . Average H/L ratio of putative Ki-67 peptides ( predicted by Maxquant on the basis of isotope profiles with the expected difference in m/z ratio from Ki-67 L-peptides positively identified by MS/MS ) is shown in red . ( B ) Isotope profile of a light Ki-67 peptide from control cell chromatin identified by MS/MS , and the corresponding heavy peaks used for quantification . Data is provided in Figure 3—figure supplement 11—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 01710 . 7554/eLife . 13722 . 018Figure 3—figure supplement 11—source data 1 . SILAC quantitation of Ki-67 peptides from WT and Mki67-mutant NIH-3T3 cells . Table showing identity and H/L ratio of Ki-67 peptides purified from chromatin of NIH-3T3 Wt , Mki67 mutant clones 14 and 21 , recovered in excised bands 1 ( >250 kDa ) or 2 ( 130-250 kDa ) . Data have been uploaded into the massIVE repository with accession number MSV000079492 . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 018 To see if cells from Ki-67 mutant mice had normal proliferation capacity we isolated embryonic fibroblasts ( MEFs ) from day-13 embryos . Homozygous Mki672nt∆/2nt∆MEFs had at least 90% lower Ki-67 levels ( Figure 3E ) . We could not confirm by immunoblotting whether or not the protein was full-length or truncated since SDS-PAGE cannot resolve 15kDa differences between proteins of nearly 400 kDa , and no antibodies are available against the N-terminus of Ki-67 . As with intestinal tissue , the loss of Ki-67 expression was not due to mRNA degradation , as shown by qRT-PCR ( Figure 3—figure supplement 7 ) . Indeed , in the homozygous mutant , the Ki-67 mRNA level was increased to a level comparable with that of proliferating NIH-3T3 cells . Mutant MEF proliferated comparably to controls , and flow cytometric assessment of EdU incorporation after a 1 hr EdU pulse showed similar numbers of replicating cells in Ki-67 WT and mutant cells ( Figure 3F ) . The low level residual Ki-67 expression in homozygous Ki-67 mutants suggests that TALEN or the conceptually-related CRISPR approaches may not lead to complete loss of expression , even when the translation initiation codon has been mutated . To further investigate whether Ki-67 translation can occur with a mutated initiation codon , we used the same TALEN pair to generate monoclonal Ki-67 mutant mouse NIH-3T3 cell lines , allowing analyses of translation that are technically impossible using animal tissues . We also performed the same procedure in the absence of TALENs to isolate wild-type clones . We obtained nine mutants with very low Ki-67 expression . Cloning and sequencing showed that five had biallelic mutations around the ATG codon ( Figure 3—figure supplement 8 ) . As in mice , even though Ki-67 was visible by immunofluorescence , Ki-67 was barely detectable by Western blot in all clones analysed ( Figure 3—figure supplement 9A ) . All clones proliferated efficiently ( Figure 3—figure supplement 9B ) . qRT-PCR showed that mutants did not have decreased Ki-67 mRNA levels compared to WT NIH-3T3 cells; indeed , like mutant MEFs , clone 14 had a higher level ( Figure 3—figure supplement 9C ) . We selected two clones for further analysis of Ki-67 translation . Clone 14 had lost the ATG codon in one allele , but had acquired an insertion of 4 nucleotides after the ATG in the second allele , generating the same shifted reading frame as the Mki672nt∆/2nt∆ mice . Clone 21 had lost the ATG codon in both alleles , thus mimicking the situation with Mki6721nt∆/21nt∆ mice . qRT-PCR quantification of Ki-67 mRNA from ribosome purifications shows that in clones 14 and 21 there was no decrease in ribosome association with Ki-67 mRNA; indeed , in clone 14 it increased , probably due to the higher Ki-67 mRNA level ( Figure 3—figure supplement 10 ) . To definitively determine levels of Ki-67 translation in the mutants , we performed SILAC quantitative mass spectrometry from exponentially growing WT or mutant NIH-3T3 cells cultured in light ( L ) ( WT ) or heavy-labelled ( H ) medium ( clones 14 and 21 ) . Chromatin was purified and run on SDS-PAGE . Peptides were purified from two gel slices , one around the predicted size of full length Ki-67 ( >250 kDa; band 1 ) and one at a smaller size ( 130k Da-250 kDa , band 2 ) . Ki-67 could not be positively identified in clones 14 and 21 ( Figure 3—figure supplement 11 ) . In peptides from WT cells ( L ) , 44 peptides derived from Ki-67 were identified by MS/MS . In contrast , in peptides from mutant cell lines ( H ) , no MS/MS spectra for Ki-67 could be identified in either band . Selecting the ‘re-quantify’ option in MaxQuant ( that forces quantitation of identified light peaks against any peaks that have the expected difference in m/z ratio ) , the ratios H/L observed for putative Ki-67 peaks were in the range of most typical contaminants that are only found unlabelled in a SILAC experiment . In band 1 , the median normalised H/L intensity ratio of the 5 corresponding m/z peaks in clone 14 was 0 . 095 ( mean , 0 . 100 , SD , 0 . 02 ) , and in clone 21 ( 7 peptides ) was 0 . 191 ( mean 0 . 203 , SD 0 . 05 ) . In band 2 , which would result from truncated or degraded Ki-67 , there were only 3 corresponding peaks , with median H/L ratios for clones 14 and 21 of 0 . 358 ( mean , 0 . 402 , SD , 0 . 30 ) and 0 . 500 ( mean , 0 . 454 , SD 0 . 33 ) respectively . Taken together , these results show that if Ki67 is translated in the mutant cell lines , it is at trace levels that are not positively identifiable by state-of-the-art mass spectrometry . At best , translation can occur from the mutated Ki-67 gene with an estimated 10% efficiency . The product is most likely an N-terminally truncated protein lacking the conserved FHA-domain , arising from a downstream in-frame ATG . Given the above results , it remained possible that very low levels of Ki-67 remain after Mki67 gene mutation and that they might suffice to sustain cell proliferation . To rule out this possibility we devised a 'double TALEN' strategy to completely eliminate Ki-67 expression . We designed and synthesised an additional TALEN pair targeting a sequence downstream of the translation stop codon . We co-transfected the ATG TALEN pair and the Stop TALEN pair with a GFP knock-in construct containing homology arms ( Figure 4A ) . We thus isolated several monoclonal cell lines in which Ki-67 mRNA was essentially eliminated ( Figure 4B ) , indicating efficient nonsense-mediated decay ( NMD ) . These cell lines had no residual Ki-67 protein expression ( Figure 4C ) , confirming that the basal immunostaining seen in clones 14 and 21 indeed reflected trace level Ki-67 expression . We used Southern blotting of genomic DNA to characterise these alleles . The 3’ end of the Mki67 gene remained intact in all clones , while the 5’ end of mutants showed a rearrangement consistent with a tandem insertion of multiple copies of the knockin construct upstream of the Mki67 ORF ( Figure 4—figure supplement 1 ) . Thus , the Mki67 gene was severely disrupted but not deleted . These Ki-67-negative cells proliferated normally . Growth curves and DNA content profiles were indistinguishable from controls ( Figure 4D ) . Time-lapse videomicroscopy showed that although individual clones had slightly different cell cycle lengths , cell division time was not significantly different between WT and mutant clones ( Figure 4E ) . We noticed that mitotic cells of one of the clones lacking Ki-67 had an altered chromosomal periphery ( Figure 4F ) . Such a phenotype has previously been reported in Ki-67 knockdown HeLa cells ( Booth et al . , 2014 ) . Nevertheless , these cells could divide efficiently . 10 . 7554/eLife . 13722 . 019Figure 4 . Cell proliferation without Ki-67 . ( A ) Schematic representation of strategy for TALEN-mediated generation of Mki67 null allele . ( B ) qRT-PCR analysis of Ki-67 mRNA levels in NIH-3T3 WT clone W4 and Ki-67-negative 60 , 65 , 99 clones . ( C ) Top: Western blot of Ki-67 and Cyclin A in NIH-3T3 WT clone W4 and Ki-67-negative mutant clones 60 , 65 , 99; LC , loading control; below , Ki-67 immunofluorescence; bar , 10 µm . ( D ) Left , growth curves of WT and Ki-67 null cell lines 60 , 65 and 99; right , cell cycle distribution analysed by flow cytometry . ( E ) Cell cycle length of WT clone W4 and Ki-67 null clones 60 and 65 as determined by time-lapse videomicroscopy . ( F ) Cells of clone 65 show altered chromosomal periphery in mitosis . The Ki-67 staining is deliberately overexposed to demonstrate absence of detectable Ki-67 in clone 65 , even in metaphase . Bar , 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 01910 . 7554/eLife . 13722 . 020Figure 4—figure supplement 1 . Generation of NIH-3T3 cells lacking Ki-67 . Top , schematic representation of the wild type ( WT ) , knock-out or eGFP knock-in Mki67 locus and the predicted insertion of tandem repeats of the eGFP insert upstream of Mki67 locus . Bottom , Southern-blot of two NIH-3T3 WT clones ( WT , W4 ) and six NIH-3T3 Ki-67 mutant clones ( 60 , 63 , 65 , 82 , 99 ) . Clones were digested with PstI or SphI and probed with 5’ or 3’ probe , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 020 Next , we tested whether Mki67 mutant clones had altered kinetics of cell cycle exit or entry . To do this , we quantified cells that could replicate by measuring 5-ethynyl-2-deoxyuridine ( EdU ) incorporation into DNA . We found that 42% of WT Mki67 cells could still incorporate some level of EdU even after 72 hr with 0 . 1% serum , but only 13% or 28% of mutant clones 60 or 65 , respectively , could do so ( Figure 5A ) . This suggested that Mki67 disruption rendered cells slightly more sensitive to serum starvation . Upon addition of serum to quiescent cells , WT and Mki67 mutants entered the cell cycle with similar kinetics ( Figure 5A ) . Ki-67 remained completely undetectable in mutants ( Figure 5B ) . 10 . 7554/eLife . 13722 . 021Figure 5 . Cells lacking Ki-67 enter the cell cycle efficiently . ( A ) Top , re-entry of cell cycle in NIH-3T3 WT clone W4 and Ki-67-negative mutant clones 60 and 65 after serum starvation-induced cell cycle arrest . Progression of cell cycle entry analysed by FACS using EdU staining . Bottom , quantification of cell cycle phases in this experiment . ( B ) Western blot analysis of Ki-67 ( upper panel ) and cyclin A2 ( lower panel ) upon cell cycle entry . LC , loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 02110 . 7554/eLife . 13722 . 022Figure 5—figure supplement 1 . Cells lacking Ki-67 proliferate efficiently . ( A ) Schematic presentation of generation of Ki-67 shRNA knockdown BJ-hTERT . ( B ) Left , Western blot for indicated proteins upon cell cycle re-entry in serum starved hTERT-transformed BJ fibroblasts ( BJ-hTERT ) after induction of shRNA against Ki-67 ( + ) or GAPDH control ( - ) . LC , loading control . Right , DNA synthesis analysed by flow cytometry after EdU pulse . ( C ) Left , asynchronous BJ-hTERT with doxycyclin-induced control or Ki-67 shRNA-expression were additionally transfected with control or Ki-67 siRNA for 48 hr . Protein levels were analysed by Western blotting . LC , loading control . Right , cell cycle distribution by flow cytometry . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 02210 . 7554/eLife . 13722 . 023Figure 5—figure supplement 2 . Cells lacking Ki-67 proliferate efficiently . ( A ) Schematic presentation of generation of Ki-67 shRNA knockdown cell lines . ( B ) Top: Western blot analysis of the indicated proteins in asynchronously growing U2OS and HeLa cells stably expressing non-targeting ( CTRL ) or Ki-67 shRNA . Lanes separated by lines were from a single exposure of a single SDS-PAGE gel and Western blot . LC , loading control . Middle: Cell cycle distribution of these cells . Bottom: qRT-PCR analysis of the indicated mRNA levels , normalized by mRNA expression of beta-2-microglobulin ( B2m ) . ( C ) Immunofluorescence of Ki-67 and EdU in asynchronous U2OS cells stably expressing control or Ki-67 shRNA , incubated with 5-ethynyl-2’-deoxyuridine ( EdU ) for 3 hr . Bar , 10 µm . Right: Ratio of EdU-positive cells to the total cell number for each incubation time; ns: not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 023 As Ki-67 is frequently used to assess proliferation in human cancer cells , we tested whether human cells lacking Ki-67 can proliferate . We generated stable human cell lines with inducible or constitutive expression of shRNA that silenced Ki-67 or a non-silencing control ( Figure 5—figure supplement 1A , 2A ) . We used the non-transformed human fibroblast cell line BJ-hTERT , and two commonly used cancer cell lines , HeLa and U2OS , which are of epithelial and mesenchymal origin , respectively . In BJ-hTERT , inducing shRNA expression largely prevented Ki-67 expression but had no detectable effect on the kinetics of entry into the cell cycle , as judged by expression of cell cycle regulators and EdU incorporation ( Figure 5—figure supplement 1B ) . Further reducing residual Ki-67 levels with siRNA also did not affect cell proliferation ( Figure 5—figure supplement 1C ) . Similarly , constitutive knockdown of Ki-67 in stable shRNA-expressing HeLa or U2OS cells had no effect on cell cycle distribution nor on the expression of cell cycle regulators ( Figure 5—figure supplement 2B ) . Analysing single cells by immunofluorescence showed that knockdown U2OS cells with undetectable Ki-67 expression incorporated EdU in a similar manner to control cells , demonstrating efficient DNA synthesis ( Figure 5—figure supplement 2C ) . Taken together , these results show that although Ki-67 elimination might have minor effects on cell cycle exit and mitosis , mammalian cells can nevertheless proliferate efficiently in the absence of detectable Ki-67 . To investigate possible molecular functions of Ki-67 , we identified interacting proteins . To do this , we expressed FLAG-tagged versions of full-length human Ki-67 or an unrelated protein ( TRIM39 ) in U2OS cells , and pulled down proteins from nuclear extracts with anti-FLAG antibody ( Figure 6—figure supplement 1 ) . These were analysed by label-free mass spectrometry . This approach identified 406 proteins specific to the Ki-67 pulldown ( Figure 6—figure supplement 2 ) . These included known Ki-67 partners: CDK1 , an established Ki-67 kinase ( Blethrow et al . , 2008 ) , nucleolar protein NIFK ( Takagi et al . , 2001 ) , protein phosphatase 1 ( Booth et al . , 2014 ) , and five subunits ( HCFC1 , HSPA8 , MATRIN3 , RBBP5 and WDR5 ) of a histone methylase complex that interacts with the nuclear receptor coregulator NRC ( also known as NCOA6; Garapaty et al . , 2009 ) . Gene ontology and STRING analysis classified the Ki-67 interactors as being enriched in two general processes: chromatin regulation and ribosomal biogenesis ( Figure 6 , Figure 6—figure supplement 2 ) . Specifically , interactors could be subdivided into groups involved in chromatin modification and transcription , ribosomal subunit biogenesis , pre-rRNA processing , protein translation , and splicing . These interactions suggested that Ki-67 might be involved not only in ribosomal biogenesis , as previously suggested ( Rahmanzadeh et al . , 2007 ) , but also in regulating chromatin . Among chromatin regulators interacting with Ki-67 , we found the NRC-interacting methylase complex; KMT2D , ASH2L and SUZ12 proteins which are components of MLL and PRC2/EED-EZH1 complexes which regulate histone H3 methylation ( Patel et al . , 2009; Pasini et al . , 2004 ) ; SETD1A , HCFC1 , HDAC2 , YY1 and RCOR1 , which are components of H3K4 demethylase complexes and co-repressors; and UHRF1 , which binds to H3K9me3-modified chromatin and is involved in both maintaining DNA methylation and heterochromatin formation ( Bostick et al . , 2007; Guetg et al . , 2010; Rottach et al . , 2010 ) . Ki-67 also interacts with TIP5 , the major component of the nucleolar remodelling complex ( NoRC ) which is required to establish and maintain perinucleolar heterochromatic rDNA , as well as NoRC-interacting proteins TTF1 and DNM3 . 10 . 7554/eLife . 13722 . 024Figure 6 . Ki-67 interacts with proteins involved in nucleolar processes and chromatin . Simplified STRING analysis reveals network interactions between proteins associating with Ki-67 . The full network is shown in Figure 6—figure supplement 2; data is provided in Figure 6—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 02410 . 7554/eLife . 13722 . 025Figure 6—source data 1 . Ki-67 interacting proteome . Table showing proteins identified by mass spectrometry eluted from immunoprecipitates of cells transfected with FLAG-tagged Ki-67 , an unrelated protein ( FLAG-tagged TRIM39 ) or empty vector . Proteins specifically interacting only with Ki-67 are presented on the first worksheet . Data are available via ProteomeXchange with identifier PXD003551 . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 02510 . 7554/eLife . 13722 . 026Figure 6—figure supplement 1 . The Ki-67 interactome . Western blot of Ki-67 in nuclear extracts from cells expressing FLAG-tagged versions of Ki-67 or a control unrelated protein , or FLAG alone . IP: immunoprecipitation; S/N: supernatant; P: pellet . Bottom , loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 02610 . 7554/eLife . 13722 . 027Figure 6—figure supplement 2 . Ki-67 interacts with proteins involved in nucleolar processes and chromatin . STRING analysis of network interactions between proteins specifically associating with Ki-67 as identified by immunoprecipitation and mass spectrometry from U2OS cells expressing FLAG-tagged full length human Ki-67 . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 027 We first focused on a possible role of Ki-67 in ribosome biogenesis , a process linked to nucleolar assembly and structure , and which is required for cell proliferation ( Hernandez-Verdun et al . , 2010 ) . One of the candidate Ki-67 interactors involved in rRNA biogenesis was the Pescadillo homologue PES1 , which participates in pre-rRNA processing and is localised in the nucleolar granular components ( GC ) ( Rohrmoser et al . , 2007; Tafforeau et al . , 2013 ) . During interphase , we found that Ki-67 localised at the cortical periphery of the GC , visualised using PES1 . Ki-67 formed a boundary between the perinucleolar heterochromatin ( clearly visible as a 'ring' in the DAPI staining ) and the GC ( Figure 7A , Figure 7—figure supplement 1 ) . Whereas nucleolar disruption using Actinomycin D or the CDK inhibitors DRB or Roscovitine caused nuclear relocalisation of Ki-67 and GC proteins ( Figure 7—figure supplement 2 ) , depletion of Ki-67 did not affect the gross structure of the nucleolus , as determined by PES1 staining ( Figure 7A , Figure 7—figure supplement 3 ) . 10 . 7554/eLife . 13722 . 028Figure 7 . Ki-67 localises PES1 to mitotic chromosomes . ( A ) Analysis of the interphase localisation of PES1 and Ki-67 proteins by immunofluorescence in HeLa cells 72 hr after transfection with control siRNA ( scramble; Scr ) or Ki-67 RNAi . Right , line scans showing the distribution of fluorescence signals within indicated nucleoli ( dashed line ) . Images were captured in confocal mode with a spinning-disk microscope . Bar , 5 μm . ( B ) Analysis of the mitotic localisation of PES1 and Ki-67 proteins by immunofluorescence in HeLa cells 72 hr after transfection with control siRNA ( scramble; Scr ) or Ki-67 RNAi . Bar , 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 02810 . 7554/eLife . 13722 . 029Figure 7—figure supplement 1 . Ki-67 is a nucleolar protein localizing in the cortical side of the GC . ( A ) Localisation of Ki-67 protein by immunofluorescence in HeLa and U2OS cells . Fibrillarin ( DFC ) or Pes1 ( GC ) were used as nucleolar markers . Images were captured in confocal mode with a spinning-disk microscope . Objective 100 x . Scale bar: 5 μm . ( B ) Line scans showing the distribution of fluorescence signals within specific nucleoli ( see dotted lines in panel A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 02910 . 7554/eLife . 13722 . 030Figure 7—figure supplement 2 . Ki-67 follows GC components upon drug-induced nucleolar disruption . Actinomycin D or the kinase inhibitors DRB and Roscovitine were added to cells during 90 min and Ki-67 was located within these cells by immunofluorescence . Fibrillarin ( A ) and PES1 ( B ) proteins were also localised by immunofluorescence . Objective 100 x . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 03010 . 7554/eLife . 13722 . 031Figure 7—figure supplement 3 . Depletion of Ki-67 does not affect overall nucleolar structure . ( A ) Histograms showing the level of Ki-67 mRNA remaining in the cell lines expressing constitutively the shRNA against Ki-67 . These levels were assessed by RT-qPCR using two different pairs of primers . Right , the Western-blot against Ki-67 shows a disappearance of a bands in the cell lines constitutively expressing the shRNA against Ki-67 . ( B ) Immuno-fluorescence against Ki-67 in control HeLa and U20S cell lines in the absence ( control ) or in the presence ( Ki-67 ) of the shRNA targeting Ki-67 mRNA . The PES1 signal shows that the nucleolar structure is maintained . Objective 100 x . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 031 During mitosis , the nucleolus undergoes a dramatic cycle of disassembly and reassembly ( Hernandez-Verdun et al . , 2010 ) . Briefly , soon after the onset of mitosis , when transcription is shut down , the nucleolus is rapidly disassembled; it then slowly reforms through the formation of intermediary organelles that undergo consecutive transformations , identifying three distinct organelle stages , and the process is complete by telophase . The first of these three intermediary organelles is a sheath of nucleolar proteins that forms around the surface of the mitotic chromosomes , the so-called 'perichromosomal region' or PR . To date , not much is known about the trans-acting factors involved in PR formation . Remarkably , we found that Ki-67 depletion totally disrupted PR formation and PES1 no longer associated with the chromosome surface ( Figure 7B ) . A similar finding , using other nucleolar proteins than PES1 as PR markers , was recently reported ( Booth et al . , 2014 ) . Having established that Ki-67 controls nucleolar assembly during mitosis , we wondered whether Ki-67 is required for pre-rRNA processing . We found that Ki-67 knocked down U2OS cells , that have essentially undetectable Ki-67 , could still incorporate normal levels of 5-ethynyl uridine ( EU ) in nucleolar RNA . This suggests that rRNA transcription is not altered ( Figure 8—figure supplement 1 ) . We next looked at pre-rRNA processing pathways by Northern blotting of precursor rRNAs or intermediates ( Figure 8—figure supplement 2 ) . This showed that silencing Ki-67 expression by shRNA or siRNA had no significant effect on pre-rRNA processing in four different cancer cell lines ( Figure 8A ) . We did , however , notice a marginal but reproducible increase in the level of the 47S precursor rRNA , indicating a mild delay in the early nucleolar pre-rRNA cleavage steps ( Figure 8B ) . The tumour suppressor TP53 ( p53 ) is a sensor of nucleolar stress resulting from defective ribosome biogenesis , and it represses ribosomal gene transcription ( Bursac et al . , 2014 ) . Impairment of early pre-rRNA cleavage steps upon depletion of Ki-67 was independent of p53 ( Figure 8B ) . Taken together , these results demonstrate that while Ki-67 is dispensable for efficient pre-rRNA processing , it is essential for the formation of the perichromosomal layer during nucleologenesis . 10 . 7554/eLife . 13722 . 032Figure 8 . Ki-67 is not required for rRNA biogenesis but controls gene transcription . ( A ) Northern-blot analysis of total RNA extracted from HeLa and U2OS cells constitutively expressing shRNA against Ki-67; and HeLa , U2OS , HCT-116 and HCT-116 TP53 ( -/- ) depleted of Ki-67 by siRNA for 72 hr in two biological replicates ( #1 and #2 ) or with scrambled siRNA control ( Scr ) . Pre-rRNA intermediates were analysed by probing with different primers located in the different spacers of the 47S sequence ( 5’ETS-green; ITS1-blue; ITS2-purple ) . ( B ) Quantification of 47S rRNA precursor in HeLa , HCT-116 and HCT-116 TP53 ( -/- ) depleted of Ki-67 by siRNA for 72 hr in three biological replicates ( n=1–3 ) . ( C ) U2OS cells ( left ) or HeLa cells ( right ) show transcriptome profile differences ( fold change >1 . 5; corrected p-value <0 . 02 ) between asynchronous cells constitutively expressing control ( CTRL ) or Ki-67 shRNA . Heat-maps present the expression levels of differentially expressed genes between biological replicates ( 1 , 2 , 3 ) and technical replicates ( 3 , 3* ) . Data is provided in Figure 8—source data 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 03210 . 7554/eLife . 13722 . 033Figure 8—source data 1 . Ki-67-dependent transcriptome in U2OS cells . Table showing statistically significant ( corrected p value < 0 . 02 , Fold-change >1 . 5 ) changes of transcript abundance from Agilent Gene chip analysis of cDNA from control U2OS ( pGIPZ-shRNA non silencing control ) and U2OS stably silenced ( pGIPZ-Ki-67 shRNA ) for Ki-67 . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 03310 . 7554/eLife . 13722 . 034Figure 8—source data 2 . Ki-67-dependent transcriptome in HeLa cells . Table showing statistically significant ( corrected p value <0 . 02 , Fold-change >1 . 5 ) changes of transcript abundance from Agilent Gene chip analysis of cDNA from control HeLa ( pGIPZ-shRNA non silencing control ) and HeLa stably silenced ( pGIPZ-Ki-67 shRNA ) for Ki-67 . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 03410 . 7554/eLife . 13722 . 035Figure 8—figure supplement 1 . Ki-67 depletion does not hinder rRNA transcription . Newly synthesised RNA in asynchronous U2OS cells expressing control or Ki-67 shRNA visualised by 5-ethynyl uridine ( EU ) incorporation and immunofluorescence . Bar , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 03510 . 7554/eLife . 13722 . 036Figure 8—figure supplement 2 . Human pre-rRNA processing pathway involves two major pathways . The two pathways ( A and B ) are characterized by specific cleavage kinetics . In pathway A , cleavages at the site A0 and 1 occur before cleavage at the site 2 , whereas cleavage at site 2 occurs before cleavages at sites A0 and 1 in the pathway B . Both pathways will end by the production of the mature 28S , 18S and 5 . 8S . The 5S rRNA is synthesized by the RNA pol III in the nucleoplasm and will join the other rRNAs during ribosome assembly . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 036 These results showed that in spite of its role in early steps of nucleologenesis , Ki-67 is not essential for expression of rRNA genes . We next asked whether it is involved in control of mRNA expression . We performed genome-wide transcriptome analysis from U2OS and HeLa cells expressing non-silencing control or Ki-67 shRNA , using Agilent gene-microarrays . Ki-67 knockdown led to downregulation ( corrected p value <0 . 02 , Fold-change >1 . 5 ) of over 200 genes ( Figure 8C ) . Expression of cell cycle regulatory genes was not affected . Additionally , Ki-67 silencing caused upregulation ( corrected p value <0 . 02 , Fold-change >1 . 5 ) of a wide variety of genes involved in neural , testis and cardiovascular system development and differentiation ( Figure 8C ) . These were strikingly enriched in genes encoding zinc-finger proteins and olfactory receptors , two gene families that are highly enriched in nucleolar associated-domains ( NADs ) of perinucleolar heterochromatin ( PNHC ) ( Németh et al . , 2010 ) . This suggested that effects of Ki-67 downregulation on gene expression might be due to an altered chromatin state , in particular of PNHC . In support of a potential role for Ki-67 in heterochromatin organisation , we found that Ki-67 knockdown in HeLa and U2OS cells caused a marked reduction in perinucleolar DAPI staining ( Figure 7A , 9A ) . We thus hypothesised that Ki-67 might be required for heterochromatin compaction . To directly assess chromatin compaction in living cells , we used a Förster Resonance Energy Transfer-Fluorescence Lifetime Imaging Microscopy ( FRET-FLIM ) -based assay . In this system , HeLa cells stably co-express versions of histone H2B labelled with eGFP and mCherry . Inter-nucleosomal interactions between H2B-eGFP and H2B-mCherry generates FRET , whose efficiency depends on the distance between nucleosomes ( Llères et al . , 2009 ) . We depleted Ki-67 by siRNA and studied the effects on FRET efficiency in interphase cells ( Figure 9—figure supplement 1 ) . As expected ( Llères et al . , 2009 ) , a heterogeneous FRET efficiency map was observed throughout control siRNA nuclei , reflecting different chromatin compaction states ( Figure 9B ) . Upon Ki-67 depletion , the mean FRET percentage decreased , reflecting a reduction in total chromatin compaction ( Figure 9B ) . The highest FRET populations , including heterochromatin regions at the nuclear periphery and around nucleoli , were largely eliminated upon Ki-67 depletion , with a few remaining condensed foci predominantly located at the nuclear periphery ( Figure 9C , D ) . In contrast , a population of less compact chromatin increased . 10 . 7554/eLife . 13722 . 037Figure 9 . Ki-67 promotes heterochromatin interactions . ( A ) DAPI staining in control and stable Ki-67-knockdown U2OS cells . ( B ) HeLa cells stably expressing GFP-H2B and mCherry-H2B , depleted using Ki-67 or non-targeting ( CTRL ) siRNA . Left , FRET efficiency ( cross shows mean value ) ** Different , p<0 . 01 . FRET efficiency and spatial distribution shown by a pseudocolour scale of FRET ( % ) values from 0 to 25% . Bars , 10 μm . ( C ) Left , representative HeLaH2B-2FP nuclei showing spatial distribution of FRET efficiency . Arrowheads show different chromatin compaction states ( high FRET , red; intermediate , green; low , blue ) , Bars , 2 μm . Right , mean FRET distribution curves from siRNA control ( blue curve , n=8 ) and siRNA Ki-67 ( red curve , n=11 ) nuclei . ( D ) Relative fraction of the three FRET efficiency populations ( blue ( low ) , FRET efficiency ≤ 8%; green ( medium ) , 8–15%; and red ( high ) , 15–25% ) in siRNA control and siRNA Ki-67 nuclei . Error bars indicate SD . ( E ) Immunofluorescence of CENP-A localisation in control and stable Ki-67 knockdown HeLa ( left ) and U2OS ( right ) cells . Nucleolar localisation ( white arrows ) . Nucleolin was used as nucleolar marker . Bar , 10 µm . Below: quantification in different cells of numbers of CENP-A spots not associated with the nucleolus . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 03710 . 7554/eLife . 13722 . 038Figure 9—figure supplement 1 . Knockdown of Ki-67 in H2B FRET cell line . Western blot analysis of the indicated proteins in asynchronously growing HeLa H2B FRET cells transiently transfected with control siRNA ( Ctrl ) or Ki-67 RNAi for 72 hr . LC , loading controls of the high ( h ) and low ( lo ) MW parts of the SDS-PAGE gel . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 038 Reduced compaction of heterochromatin implies disruption of short-range interactions of chromatin . To determine whether Ki-67 knockdown also affects long range interactions , we assessed interactions between perinucleolar and pericentromeric heterochromatin . We stained for the centromeric histone variant CENP-A to determine the localisation of centromeric DNA . In HeLa and U2OS cells , CENP-A showed a non-random nuclear localisation and clustered around nucleoli ( Figure 9E , arrowheads ) , confirming that CENP-A can be used as a surrogate marker for adjacent pericentromeric DNA . Consistent with our hypothesis , this interaction was disrupted upon Ki-67 knockdown , and the CENP-A signal was no longer grouped around nucleoli but dispersed throughout the nucleus ( Figure 9E ) . These results imply that Ki-67 mediates interaction between different regions in the genome that are normally packaged into heterochromatin , and could potentially maintain silencing of genes by recruiting them to constitutive heterochromatin . Constitutive heterochromatin , including PNHC , is characterised by histone post-translational modifications H3K9me3 and H4K20me3 . We asked whether they were affected by downregulation of Ki-67 . Stable shRNA-mediated Ki-67 knockdown in HeLa , U2OS and inducible knockdown in BJ-hTERT fibroblasts caused a visible reduction in nucleolar staining of H3K9me3 and H4K20me3 ( Figure 10—figure supplements 1–3 ) . This mark was relocalised either to foci in proximity to the nucleolus or a punctate pattern dispersed throughout the nucleus . In mouse cells , where pericentromeric heterochromatin is prominent , H3K9me3 staining colocalised with DAPI-dense chromocentres , but TALEN-ablation of Mki67 resulted in general nuclear punctate H3K9me3 that was excluded from nucleoli ( Figure 10A , B ) . Western blotting revealed that Ki-67 depletion did not affect the overall levels of these chromatin modifications ( Figure 10—figure supplement 4 ) . We also analysed the localisation of heterochromatin protein 1 ( HP1 ) , which binds to chromatin containing H3K9me3 . Immunofluoresence showed that , surprisingly , despite the loss of the intense H3K9me3 staining regions in the Mki67 mutant cells , all three HP1 isoforms maintained their localisation at DAPI-dense regions ( Figure 10—figure supplement 5 ) . Next , we determined whether heterochromatic histone marks were reduced on specific DNA sequences , or whether they were retained but the sequences themselves were delocalised . To do this we examined co-localisation of H3K9me3 with mouse major satellite DNA , by combining immunofluorescence with fluorescent in situ hybridisation ( FISH ) . In control cells , major satellite DNA , DAPI-dense regions and H3K9me3 largely colocalised ( Figure 10C ) . In cells lacking Ki-67 , major satellite DNA was still present at regions of compacted DNA , despite the loss of H3K9me3 at these regions ( Figure 10C ) . Taken together , these results suggest that Ki-67 is required for maintaining heterochromatic histone marks at genomic regions that are organised into heterochromatin . 10 . 7554/eLife . 13722 . 039Figure 10 . Ki-67 controls heterochromatin organisation . ( A ) Top , immunofluorescence analysis of H3K9me3 in mouse NIH-3T3 WT and Mki67 TALEN mutant clones 60 and 65 . Bars , 5 µm . Below: graphs showing quantification of pixel intensity scans for H3K9me3 . ( B ) Quantification of H3K9Me3 patterns in WT and Ki-67 mutant clones . ( C ) Immunofluorescence of H3K9Me3 , FISH of major satellite DNA and DAPI staining in WT W4 and Ki-67 mutant clone 60 . Bar , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 03910 . 7554/eLife . 13722 . 040Figure 10—figure supplement 1 . Heterochromatic histone mark localisation requires Ki-67 . Immunofluorescence analysis of H3K9me3 ( left ) and H4K20me3 ( right ) localisation in stable control and Ki-67 knockdown U2OS cells . Right: Fire look up table ( F-LUT ) pseudocolouring of immunofluorescence staining intensity , generated using Fiji software ( Schindelin et al . , 2012 ) . Dotted white lines denote nucleolus , while numbers 1–4 identify cells for insets as well as staining patterns within nucleolus ( 1 , 2 ) or outside the nucleolus ( 3 , 4 ) . Histograms below show the percentage of cells counted showing each pattern . The 2D Fire-LUT surface plot was generated using Fiji software ( 1 ) . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 04010 . 7554/eLife . 13722 . 041Figure 10—figure supplement 2 . Heterochromatic histone mark localisation requires Ki-67 . Immunofluorescence analysis of H3K9me3 ( left ) and H4K20me3 ( right ) localisation in stable control and Ki-67 knockdown HeLa cells . Right: Fire look up table ( F-LUT ) pseudocolouring of immunofluorescence staining intensity . Dotted white lines denote nucleolus , while numbers 1–4 identify cells for insets as well as staining patterns within nucleolus ( 1 , 2 ) or outside the nucleolus ( 3 , 4 ) . Histograms below show the percentage of cells counted showing each pattern . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 04110 . 7554/eLife . 13722 . 042Figure 10—figure supplement 3 . Heterochromatic histone mark localisation requires Ki-67 . Immunofluorescence analysis of H3K9me3 and H4K20me3 localisation in stable control and Ki-67 knockdown hTERT immortalised human fibroblasts ( HDF BJ hTERT ) cells . Scale bars , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 04210 . 7554/eLife . 13722 . 043Figure 10—figure supplement 4 . Overall heterochromatic histone mark levels do not change upon Ki-67 knockdown . Left: western blot of total H3K9me3 and H4K20me3 level in control and stable Ki-67 knockdown BJ-hTERT , U2OS and HeLa cells . LC , loading control . Right: western blot of total H3K9me3 and H4K20me3 level in NIH-3T3 WT clone W4 and Ki-67-negative TALEN clones 60 , 65 . LC , loading controlDOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 04310 . 7554/eLife . 13722 . 044Figure 10—figure supplement 5 . HP1 localises normally to chromocentres in Ki-67 mutant cells . Immunofluorescence analysis of localisation of HP1α , HP1β and HP1γ in WT and Ki-67-negative NIH-3T3 mutant clones 60 and 65 , as compared with DAPI staining . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 044 These results suggest that Ki-67 is required for heterochromatin organisation . To see whether Ki-67 is sufficient to promote heterochromatin formation , we cloned a full-length cDNA encoding human Ki-67 , that we fused to the eGFP gene , and transfected it into U2OS cells . There was a strong correlation between cells with higher levels of exogenous Ki-67 and appearance of DAPI-dense foci resembling ectopic heterochromatin , marked by H3K9me3 and HP1 ( Figure 11A ) . Cells showing this phenotype were negative for cyclin A staining , suggesting that they were unable to enter S-phase , whereas lower expression did not prevent cyclin A accumulation ( Figure 11B ) . They were also negative for histone H3 Ser-10 phosphorylation , a mitotic marker ( Figure 11—figure supplement 1 ) . If controlling heterochromatin organisation is a major function of Ki-67 , it is likely to be conserved in more distantly related Ki-67 homologues . In a proteomics-based screen for proteins associated with replicating chromatin in egg extracts , we identified a putative Xenopus Ki-67 homologue ( Figure 11—figure supplement 2 ) . To assess whether Xenopus and human Ki-67 are functionally conserved , we cloned and HA-tagged full-length Xenopus Ki-67 and expressed it in U2OS cells . Whereas in interphase , exogenous Xenopus Ki-67 is present ubiquitously on chromatin , it colocalises with endogenous Ki-67 in mitosis at the perichromosomal region ( Figure 11C ) . Overexpression of Xenopus Ki-67 , like human Ki-67 , caused extreme chromatin compaction ( Figure 11C ) . We conclude that controlling heterochromatin is a conserved essential function of Ki-67 . 10 . 7554/eLife . 13722 . 045Figure 11 . Overexpression of Ki-67 induces ectopic heterochromatin . ( A ) Overexpression of full length Ki-67 N-terminal fusion with eGFP in U2OS cells induces ectopic heterochromatin , as visualised by DAPI staining ( middle ) and immunofluorescence of H3K9Me3 ( left ) or HP1β ( right ) . Eight representative cells that have different levels of Ki-67 expression , as determined by eGFP fluorescence intensity , are shown . Bar , 10 µm . ( B ) U2OS cells expressing high levels of exogenous eGFP-Ki-67 and showing ectopic chromatin condensation are negative for cyclin A staining by immunofluorescence . ( C ) Left: Immunofluorescence analysis of the localisation of endogenous human and ectopically expressed Xenopus Ki-67 in U2OS cells , showing colocalisation in metaphase at the perichromosomal region . Right: DNA condensation caused by high overexpression of Xenopus Ki-67 in U2OS cells . Bars , 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 04510 . 7554/eLife . 13722 . 046Figure 11—figure supplement 1 . Overexpression of Ki-67 induces ectopic heterochromatin . Overexpression of full length Ki-67 N-terminal fusion with eGFP ( GFP , left panels ) in U2OS cells induces ectopic heterochromatin , as visualised by DAPI staining ( DNA , right panels ) or immunofluorescence of HP1α ( centre , middle ) , whereas cells with lower Ki-67 expression levels have normal chromatin . Immunofluorescence of Ki-67 ( top , middle ) shows colocalisation of fusion protein with overall Ki-67 pattern . Immunofluorescence of phospho-histone H3S10 shows expected staining of mitotic metaphase ( bottom , middle panel ) but no staining in a cell with ectopic heterochromatin ( top right cell , same panel ) due to Ki-67 overexpression ( GFP , bottom left panel ) . Bar , 10 µmDOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 04610 . 7554/eLife . 13722 . 047Figure 11—figure supplement 2 . A Xenopus Ki-67 homologue ( A ) Chromatin proteomics in replicating Xenopus egg extracts . Scheme of the experiment and electrophoresis separation of sample processed and analysed by mass spectrometry . Table shows peptide matches . M , molecular weight marker . ( B ) Schematic comparison of Xenopus Ki-67 homologue with human Ki-67 ( long form ) . Domains are indicated by boxes ( FHA , forkhead-associated domain; PP1 , PP1-binding domain; CD , conserved domain ) . Highly conserved regions are indicated by dotted line with percentage of identical amino acids . TP , threonine-proline phosphorylation site . DOI: http://dx . doi . org/10 . 7554/eLife . 13722 . 047 Ki-67 has long been assumed to be essential for cell proliferation ( Schluter et al . , 1993; Kausch et al . , 2003; Starborg et al . , 1996; Rahmanzadeh et al . , 2007; Zheng et al . , 2006; 2009 ) . Using various genetic and knockdown approaches , we found no evidence for this in any cell type we tested: HeLa , U2OS , BJ-hTERT and NIH-3T3 fibroblasts . Our data show that Ki-67 expression can be uncoupled from cell proliferation in both directions . Indeed , not only can cells lacking Ki-67 proliferate efficiently , conversely , interfering with Ki-67 downregulation by disrupting the Cdh1 gene did not prevent cell cycle exit in vivo . It remains possible that certain cell lines are more sensitive to inhibition of Ki-67 expression , eg cancer cell lines of bladder ( Kausch et al . , 2003 ) or renal ( Zheng et al . , 2006; 2009 ) origins . Alternatively , off-target effects of previous antisense or RNAi approaches might have contributed to the cell proliferation defects observed in previous studies , as none of them employed restoration controls using silencing-insensitive constructs . Such rescue experiments are virtually unfeasible given the large size of Ki-67 , the targeting of silencing oligonucleotides to the repeated domains , and the fact that Ki-67 overexpression induces ectopic heterochromatin . Several other studies used different approaches . In one , microinjection of an anti-Ki-67 antibody in 3T3 cells did not cause an abolition of cell division ( Starborg et al . , 1996 ) . Instead , there was a modest reduction , from 80% to 64% , of dividing cells over a 36-hr period . Another study used chromophore-mediated light inactivation of Ki-67 after injecting chromophore-labelled Ki-67 antibodies , and found an inhibition of ribosomal RNA synthesis ( Rahmanzadeh et al . , 2007 ) . However , such an approach might cause non-specific collateral damage to nucleolar processes where Ki-67 is localised . We found that rather than promoting cell proliferation , the role of Ki-67 is to organise heterochromatin . We showed that Ki-67 is required for the maintenance of a high level of compaction typical of heterochromatin , and mediates long-range interactions between different regions of the genome that are packaged into heterochromatin . We speculate that heterochromatin compaction relies on local interactions that depend on Ki-67 . Cells lacking Ki-67 show altered gene expression profiles upon long-term Ki-67 silencing , with a striking correlation between upregulation of genes that normally are physically associated with perinucleolar heterochromatin ( Németh et al . , 2010 ) . To determine possible mechanisms of action of Ki-67 , we comprehensively identified its interacting partners . We thus found at least seventeen proteins that are involved in histone methylation complexes or are interactors of methylated chromatin required for heterochromatin maintenance . This suggests that Ki67 might target these proteins to their genomic sites to promote heterochromatin formation . Consistent with this hypothesis , Ki-67 downregulation led to reduction of H3K9me3 and H4K20me3 at heterochromatin , while Ki-67 overexpression caused appearance of ectopic heterochromatic foci enriched in these methylation marks . Unexpectedly , Ki-67 downregulation did not prevent association of HP1 isoforms with heterochromatin . Possibly , low levels of H3K9me3 or H4K20me3 persist and are sufficient for recruitment of HP1 , or alternative mechanisms exist to localise HP1 to chromatin . The former hypothesis would be consistent with the observation that loss of the Suv39H methyltransferases that are responsible for H3K9me3 and H4K20me3 abrogates HP1 recruitment to heterochromatin in mice , whose late embryonic growth and survival is impaired ( Peters et al . , 2001 ) . Nevertheless , evidence for the latter possibility has been provided by a study in C . elegans , in which genome-wide distribution of HP1 binding , as assessed by ChIP-seq , was conserved in animals lacking H3K9 di- and trimethylation ( Garrigues et al . , 2015 ) . Given the requirement for Ki-67 expression in organising heterochromatin , it is perhaps surprising that mouse development is not affected by Ki-67 downregulation . This once again highlights the robustness of biological systems . However , to determine whether mouse development can occur normally in the complete absence of Ki-67 will require a gene deletion rather than a gene disruption mediated by genome-editing , as we found that even deletion of the translation initiation ATG using TALENs did not completely abolish Ki-67 expression . The eight subsequent ATG codons are out of frame , and the next in-frame ATG is 433 bp downstream . Translation from any frameshifted ATG will lead to a premature stop codon within 65 nucleotides . Although NMD of the mRNA did not occur , translation was strongly reduced . This is probably due to the presence of many out-of-frame ATG codons before the next in-frame ATG codon , as well as the distance from the 5’ end of the mRNA . It is likely that the residual Ki-67 in proliferating cells occurred from the next in-frame ATG , thus eliminating the most highly conserved domain of Ki-67 , the Forkhead-associated ( FHA ) domain . However , the unexpected translation in the mutants suggests that care should be taken to examine possible low level expression of proteins after mutating start sites using genome editing approaches . Since Ki-67 is degraded upon cell cycle exit , we speculate that this may alter chromatin structure . For example , Ki-67 degradation might be involved in heterochromatin rearrangements observed during senescence onset . Facultative heterochromatic foci , that characterise some senescent cells ( Narita et al . , 2003; 2006 ) are not a consistent feature of senescence in all cell types . In contrast , large-scale satellite heterochromatin decondensation is an early step in senescence in all cells studied and it precedes loss of H3K9me3 ( Swanson et al . , 2013 ) . As Ki-67 is required for heterochromatin compaction , its degradation may be involved in the heterochromatin decompaction occurring upon senescence onset . Heterochromatin reorganisation caused by Ki-67 downregulation does not interfere with cell cycle progression or cell proliferation , but likely contributes to remodelling of gene expression . Heterochromatin is also less compact in highly proliferative pluripotent stem cells , suggesting that heterochromatin organisation is critical for determining transcriptional responses ( Fussner et al . , 2011 ) . Ki-67 overexpression , which led to pronounced chromatin condensation , appeared to arrest the cell cycle in G1 , implying that controlled Ki-67 degradation is required to allow unperturbed progression through the cell cycle . The nucleolus is a potent cancer biomarker ( Derenzini et al . , 2009 ) , and a recently demonstrated target in cancer therapy . Inhibitors of nucleolar functions have indeed been shown to selectively kill cancer cells , leaving non-cancerous cells intact ( Bywater et al . , 2012; Peltonen et al . , 2014 ) . It is therefore critical to understand how the nucleolus forms during mitosis . An important step in nucleolar assembly is the formation of a sheath of nucleolar proteins around the chromosome surface on the metaphase plate . This so-called perichromosomal layer has been suggested to play roles in chromosome protection , in the faithful partitioning of nucleolar proteins between daughter cells , and in the segregation of opportunistic passenger proteins . Whether the PR performs any of these functions or has other , unidentified , roles will be a promising field for future studies . In this study , we have identified Ki-67 as one of the first trans-acting factors involved in PR formation during nucleologenesis , corroborating a recent report ( Booth et al . , 2014 ) . In conclusion , our data reveal a novel concept whereby heterochromatin organisation is linked to cell proliferation by Ki-67 . As heterochromatin organisation is often compromised in cancer cells ( Carone and Lawrence , 2013 ) and Ki-67 expression is widely used in clinical assessments in cancer , these data provide a rationale for further investigation of the functional consequences of Ki-67 expression in tumour samples . Importantly , our data suggest that Ki-67 is likely to modulate transcription in cancer cells . All animal experiments were performed in accordance with international ethics standards and were subjected to approval by the Animal Experimentation Ethics Committee of Languedoc Roussillon . Normal human diploid foreskin fibroblasts ( HDF ) were provided by Jacques Piette ( CRBM , Montpellier ) , the hTERT-immortalized foreskin fibroblast cell line ( BJ hTERT ) was provided by Jean Marc Lemaitre ( IRB , Montpellier ) . U2OS , HeLa , NIH3T3 mouse fibroblasts were obtained from the American Type Culture Collection . They were not authenticated but were mycoplasma-free ( tested weekly ) . U2OS , HeLa and NIH 3T3 were grown in Dulbecco modified Eagle medium ( DMEM - high glucose , pyruvate , GlutaMAX – LifeTechnologies , ThermoFisher Scientific , Paris , France ) supplemented with 10% foetal bovine serum ( Sigma-Aldrich , Lyon , France or HyClone , GE Healthcare , Paris , France ) . BJ hTERT were grown in DMEM supplemented with 10% foetal calf serum ( Sigma-Aldrich ) and 2 mM L-glutamine . Apart from murine embryo fibroblasts ( MEFs ) , cells were grown under standard conditions at 37°C in a humidified incubator containing 5% CO2 . MEFs were grown in DMEM supplemented with 10% fetal bovine serum and 1% Penicillin/Streptomycin at 37°C in an incubator containing 3% O2 and 5% CO2 . The lentiviral constitutive and inducible knockdown systems were packaged into non-replicating lentivirus HIV-1 using II generation packaging system – psPAX by PVM platform ( IGF , Montpelier ) . Immortalized BJ-hTERT and U2OS cells were infected at MOI 10 with lentivirus armed pTRIPZ GAPDH positive CTRL and Ki-67 inducible shRNA , and immortalized BJ hTERT , U2OS and HeLa were infected at MOI 10 with lentivirus armed pGIPZ shRNA non-silencing and Ki-67 constitutive system . Lentiviral transduction was performed according to the manufacturer’s protocol ( Thermo Scientific ) . 2 days after infection cells were selected with 10 µg/ml puromycin for 4 days . Cells were treated with progressively increased puromycin concentrations up to 60 µg/ml , to select the most highly transduced population . The lentiviral constitutive knockdown vectors containing shRNAs were purchased from ThermoFisher Scientific . 1 . pGIPZ shRNAmir Ki-67 ( clone ID: V2LHS-151787 ) AGGCTACAAACTCGTAAGGAAATAGTGAAGCCACAGATGTATTTCCTTACGAGTTTGTAGCCG 2 . pGIPZ shRNAmir CTRL non-targeting ( RHS4346 ) ACCTCCACCCTCACTCTGCCATTAGTGAAGCCACAGATGTAATGGCAGAGTGAGGGTGGAGGG The lentiviral doxycycline-inducible knockdown positive control vector containing shRNA GAPDH was purchased from Thermo Scientific . 3 . pTRIPZ shRNAmir GAPDH CCCTCATTTCCTGGTATGACAATAGTGAAGCCACAGATGTATTGTCATACCAGGAAATGAGGT pTRIPZ shRNAmir Ki-67 vector was obtained by sub-cloning to replace the shRNAmir GAPDH in pTripZ with the shRNAmir sequence from pGIPZ shRNAmir Ki-67 . To induce expression of shRNA , cells were treated with 2 µg/ml doxycycline hyclate ( Sigma-Aldrich ) for minimum 24 hr and then during the period of designed experiments . Downregulation of mRNA of shRNA target genes was analysed 24 hr post induction . The SMARTpool: ON-TARGETplus siRNAs were purchased from GE Dharmacon ( Lafayette , CO , USA ) . Cells were transfected with SMARTpool: ON-TARGETplus siRNA non-targeting ( D-001810-10 ) , MKI67 ( L-003280-00 ) and FZR1 ( L-015377-00 ) at 100 nM by Calcium Phosphate transfection method . Frozen pellets ( harvested by trypsinization , washed with cold PBS ) were lysed directly in Laemmli buffer at 95°C ( without β-mercaptoethanol and bromophenol blue ) and sonicated in a chilled bath for 10 min in 30 s/30 s ON/OFF intervals . Protein concentrations were determined by BCA protein assay ( ThermoFisher ) . Equivalently loaded proteins were separated by SDS-polyacrylamide gel electrophoresis ( SDS-PAGE ) ( usually on 12 cm x 14 . 5 cm 7 . 5% and 12 . 5% gels ) at 35 mA in TGS buffer ( 25 mM Tris , 200 mM glycine , 0 . 1% SDS ) . The proteins were then transferred to Immobilon membranes ( Milipore ) at 1 . 15 mA/cm2 for 120 min with a semidry blotting apparatus containing transfer buffer ( 25 mM Tris , 200 mM Glycine , 0 . 2% SDS , 20% EtOH ) . Membranes were blocked in TBST pH 7 . 6 ( 20 mM Tris , 140 mM NaCl , 0 . 1% Tween-20 ) containing non-fat dry milk ( 5% ) , incubated with antibody for 2 hr at RT with agitation in TBST containing non-fat dry milk ( 1 , 25% ) , washed several times with TBST for a total of 45 min , incubated with secondary antibody at 1/5000 dilution in TBST containing non-fat dry milk ( 1 . 25% ) for 1 hr at RT with agitation and washed several times for 1 hr in TBST . Secondary antibodies were either goat antibodies to mouse IgG-HRP ( immunoglobulin G – horseradish peroxidase ) ( DACO ) or donkey antibodies to rabbit IgG-HRP ( immunoglobulin G – horseradish peroxidase ) ( GE Healthcare ) . The detection system used was Western Lightning Plus-ECL ( PerkinElmer , Paris , France ) and Amersham Hyperfilm ( GE Healthcare ) . FLIM-FRET experiments were carried out on a HeLaH2B-2FPs cell line stably expressing GFP and mCherry tagged histone H2B as previously described ( Llères et al . , 2009 ) . Fluorescence Lifetime Imaging Microscopy ( FLIM ) was performed using an inverted laser scanning multiphoton microscope LSM780 ( Zeiss ) equipped with temperature-controlled environmental black walls chamber . Measurements were acquired at 37°C , with a 63× oil immersion lens NA 1 . 4 Plan-Apochromat objective from Zeiss . Two-photon excitation was achieved using a Chameleon Ultra II tunable ( 680–1080 nm ) laser ( Coherent ) to pump a mode-locked frequency-doubled Ti:Sapphire laser that provided sub-150-femtosecond pulses at a 80-Mhz repetition rate with an output power of 3 . 3 W at the peak of the tuning curve ( 800 nm ) . Enhanced detection of the emitted photons was afforded by the use of the HPM-100 module ( Hamamatsu R10467-40 GaAsP hybrid PMT tube ) . The fluorescence lifetime imaging capability was provided by TCSPC electronics ( SPC- 830; Becker & Hickl GmbH ) . TCSPC measures the time elapsed between laser pulses and the fluorescence photons . EGFP and mCherry fluorophores were used as a FRET pair . The optimal two-photon excitation wavelength to excite the donor ( EGFP ) was determined to be 890 nm ( Llères et al . , 2007 ) . Laser power was adjusted to give a mean photon count rate of the order 4 . 104–105 photons/s . For imaging live cells by FLIM , the standard growth medium was replaced with phenol red-free DMEM supplemented with 10% FBS . Fluorescence lifetime measurements were acquired over 90 s and fluorescence lifetimes were calculated for all pixels in the field of view ( 256×256 pixels ) and then a particular region of interest ( e . g . , nucleus ) was selected using SPCImage software ( Becker & Hickl , GmbH ) . The analysis of the FLIM measurements was performed by using SPCImage software . Because FRET interactions cause a decrease in the fluorescence lifetime of the donor molecules ( EGFP ) , the FRET efficiency can be calculated by comparing the FLIM values obtained for the EGFP donor fluorophores in the presence and absence of the mCherry acceptor fluorophores . Mean FRET efficiency images were calculated such as the FRET efficiency , EFRET = 1- ( τDA/τD ) , where τDA is the mean fluorescence lifetime of the donor ( H2B-EGFP ) in the presence of the acceptor ( mCherry-H2B ) expressed in the HeLaH2B-2FPs and τD is the mean fluorescence lifetime of the donor ( H2B-EGFP ) expressed in HeLaH2B-GFP in the absence of acceptor . In the non-FRET conditions , the mean fluorescence lifetime value of the donor in the absence of the acceptor was calculated from a mean of the τD by applying an exponential decay model to fit the fluorescence lifetime decays . In the FRET conditions , we applied a biexponential fluorescence decay model to fit the experimental decay curves f ( t ) =a e-t/τDA + b e-t/τD . By fixing the noninteracting proteins lifetime tD using data from control experiments ( in the absence of FRET ) , the value of tDA was estimated . Then , the FRET efficiency ( EFRET ) was derived by applying the following equation: EFRET=1- ( τDA/τD ) at each pixel in a selected ROI using SPCImage software . The FRET distribution curves from these ROIs were displayed from the extracted associated matrix using SPCImage and then normalized and graphically represented using Excel and GraphPad Prism software . Cells were seeded on 12 mm diameter coverslips #1 . 5 coated with 1% gelatine . Before fixation coverslips were washed once with PBS . Then , cells were fixed either in 3 . 7% formaldehyde for 15 min at RT or in cold 100% MeOH ( 10 min , -20°C ) . Formaldehyde fixed cells were immediately washed twice with PBS and permabilized in 0 . 2% TRITON X-100 for 15 min at RT , while MeOH fixed cells on coverslips were transferred on tissue paper and kept at RT to dry . Next , cells were blocked in blocking solution ( 5% FBS; 0 . 1% Tween-20 in PBS ) for 30 min at RT , incubated overnight with primary antibodies diluted in blocking solution at 4°C , washed 3 times 5 min with PBS-Tween ( 0 . 1% Tween-20 in PBS ) , incubated with secondary antibody at RT for 1 hr , and washed 4 times 5 min with PBS-Tween . Secondary antibodies were diluted 1:1000 for fluorophores Alexa488; 555; 568 and 1:500 for fluorophore Alexa647 . Coverslips were washed in distilled water prior mounting on slide with ProLong Gold Antifade Reagent with DAPI . HeLa or U2OS cell lines were cultured in 96-well plates . For siRNA-mediated Ki-67 depletion , a transfection reagent ( mix of 0 . 125 μl of Interferin and 20 μl of Optimem ) was added to each well of the plate and left for 10 min at RT° . SiRNA ( 10 μl of 100 nM ) were added and left for another 30 min at RT° . Cells ( 70 μl of 300 , 000 cells/ml dilution ) were then added to each well and the plates were incubated for 3 days at 37°C with 5% CO2 . Nucleolar structure disruption was performed by treatment of the cells with 0 . 2 μg/ml of Actinomycin D , 40 μM roscovitine or 60 μM DRB for 90 min . For immunofluorescence , cells were fixed in 2% formaldehyde , washed in PBS and blocked in PBS supplemented with 5% BSA and 0 . 3% Triton X-100 during 1 hr at RT° . Anti-Pes 1 antibody ( 1:1 , 000; courtesy from E . Kremmer ) , anti-Ki-67 antibody ( 1:500 , Cell Signaling ) and/or anti-Fibrillarin antibody ( 1:250 , antibodies online ) were diluted in PBS supplemented with 1% BSA , 0 . 3% Triton X-100 and incubated with the cells O/N at 4°C . Cells were washed in PBS and incubated with the secondary antibody coupled to AlexaFluor 488 or 594 ( 1:1 , 000; Invitrogen ) in PBS , 1% BSA , 0 . 3% Triton X-100 for 1 hr at RT° . Cells were washed in PBS and treated with DAPI . Microscopy was performed on a Zeiss Axio Observer . Z1 microscope driven by MetaMorph ( MDS Analytical Technologies , Canada ) . Images were captured in the confocal mode using a Yokogawa spindisk head and the HQ2 camera with a laser illuminator from Roper ( 405 nm 100 mW Vortran , 491 nm 50 mW Cobolt Calypso , and 561 nm 50 mW Cobolt Jive ) and 40x or 100x objectives ( Zeiss ) . Line scans and images were constructed using Image J . The CellProfiler software was used to quantify the DAPI intensity at the peri-nucleolar region of about 100 individual cells and classical statistical t-test was applied to the data to compare the intensity distributions . Freshly dissected small intestines were flushed and fixed for 4 hr in neutral buffered formalin before paraffin embedding . Briefly , 5-µm-thick sections were dewaxed in xylene and rehydrated in graded alcohol baths . Antigen retrieval was performed by boiling slides for 20 min in 10 mM sodium citrate buffer , pH 6 . 0 . Nonspecific binding sites were blocked in blocking buffer ( TBS , pH 7 . 4 , 5% dried milk , and 0 . 5% Triton X-100 ) for 60 min at RT . Sections were then incubated with primary antibodies diluted in blocking buffer overnight at 4°C . Primary antibodies used were as follows: anti Ki-67 ( Ab16667 ) and anti DCLK1 ( Ab31704 ) were from Abcam , Cambridge , UK . Anti beta-catenin ( BD610154 ) was from BD-Bioscience , Oxford , UK . Anti BrdU ( G3G4 ) was form the Developmental Studies Hybridoma Bank . Slides were then washed two times with 0 . 1% PBS-Tween ( Sigma-Aldrich ) before incubation with fluorescent secondary antibodies conjugated with either Alexa 488 , Cyanin-3 , or Cyanin-5 ( Jackson ImmunoResearch Laboratories , Inc . ) and Hoechst at 2 µg/ml ( Sigma-Aldrich ) in PBS–Triton X-100 0 . 1% ( Sigma-Aldrich ) . Stained slides were then washed two extra times in PBS before mounting with Fluoromount ( Sigma-Aldrich ) . Methods used for bright-field immunohistochemistry were identical , except that slides were incubated with 1 . 5% H2O2 in methanol for 20 min and washed in PBS to quench endogenous peroxydase activity before antigen retrieval . Envision+ ( Dako ) was used as a secondary reagent . Signals were developed with DAB ( Sigma-Aldrich ) and a hematoxylin counterstain ( DiaPath ) was used . After dehydration , sections were mounted in Pertex ( Histolab ) . Goblet cells staining was achieved with a periodic acid/Schiff's reaction ( Sigma-Aldrich ) . " Fish of major satellite DNA in combination with immunofluorescence was performed in formaldehyde fixed cells , as in ( Saksouk et al . , 2014 ) . Cells were pre-treated 5 min with 20 μg/mL of emetine , before being collected , washed and resuspended in ice cold homogenization buffer ( 50 mM Tris-HCl ph7 . 5 , 5 mM MgCl2 , 25 mM KCl , 0 . 2M Sucrose , 0 . 5% NP-40 , EDTA-free protease inhibitors ( Roche ) , 10 U/ML RNAse Out ( Invitrogen ) , DEPC water ) . We then lyzed cells using Lysing Matrix D beads and FastPrep sample preparation system ( MP Bio ) . The cleared lysate was layered on 15–50% sucrose gradient in the same buffer ( homogenization buffer minus NP-40 ) . Following centrifugation at 35 , 000 rpm ( Beckman , SW41 . Ti ) for 2 . 5 hr at 4°C , gradients were fractionated ( density gradient fractionator , Teledyne Isco ) with absorbance measured continuously at 254 nm . We isolated RNA from fractions with TRIzol ( Thermo Fisher Scientific ) following the manufacturer’s instructions . We then reverse transcribed purified RNA into cDNA following RT-PCR method . We analysed Ki-67 mRNA level in polysome fractions using two Mki67 primer pairs ( 5’-AATCCAACTCAAGTAAACGGGG-3’ , 5’-TTGGCTTGCTTCCATCCTCA-3’ and 5’-CATCAGCCCATGATTTTGCAAC-3’ , 5’-CTGCGAAGAGAGCATCCATC-3’ ) normalizing to housekeeping genes ( Gapdh: 5’-AAATGGTGAAGGTCGGTGTG-3’ , 5’-AATCTCCACTTTGCCACTGC-3’; B2m: 5’-GGTCTTTCTGGTGCTTGTCT-3’ , 5’-GCAGTTCAGTATGTTCGGCTT-3’; Actb: 5’-TCCTGGCCTCACTGTCCAC-3’ , 5’-GTCCGCCTAGAAGCACTTGC-3’; Hprt: 5’-AAGCCTAAGATGAGCGCAAG-3’ , 5’-TTACTAGGCAGATGGCCACA-3’ ) . 1 . 5x107 U2OS cells were transfected with pcDNA5 plasmid expressing 3xFLAG-Ki-67 . As a control , an equal number of cells were transfected with pcDNA3 plasmid expressing 3xFLAG-TRIM39 or 3xFLAG . 24 hr after transfection , cells were harvested and the nuclear extracts prepared . 100 µg of nuclear protein extract were combined with 40 µl anti-FLAG M2– agarose beads ( Sigma-Aldrich ) , and incubated for 1 hr at 4°C with rotation . Beads were washed 5 times for 5 min at 4°C with rotation with washing buffer ( 20 mM HEPES pH 7 . 9; 1 mM EDTA; 1 mM EGTA; 150 mM NaCl; 25% glycerol + freshly added 0 . 2 mM Na3VO4; Complete-Protease inhibitor cocktail ) and the precipitates were eluted by 50 μL of SDS denaturation buffer , and heating at 95°C for 5 min . Eluted proteins were reduced , alkylated , analysed in a 4–20% gradient gel ( BioRad ) and entire lanes were sliced . Tryptic peptides were prepared for mass spectrometry , essentially as described , and then concentrated with a pre-column ( Thermo Scientific , C18 PepMap100 , 300 μm × 5 mm , 5 μm , 100 A ) at a flow rate of 20 µL/min using 0 . 1% formic acid . Samples were separated with a C18 reversed-phase capillary column ( Thermo Scientific , C18 PepMap100 , 75 μm × 250 mm , 3 μm , 100 A ) at a flow rate of 0 . 3 µL/min using the following gradient: 8–28% acetonitrile in 40 min and then from 28–42% in 10 min . The HPLC system was coupled online to a Q-TOF Maxis Impact ( Bruker Daltonik GmbH , Bremen , Germany ) mass spectrometer . Up to 30 data-dependent MS/MS spectra were acquired in positive ion mode . MS/MS raw data were analysed using Data Analysis software ( Bruker Daltonik GmbH , Bremen , Germany ) to generate the peak lists . The Homo sapiens Complete Proteome database ( downloaded on Uniprokb 20131108 , contains 88 , 266 sequences ) was queried locally using the Mascot search engine v . 2 . 2 . 07 ( Matrix Science , http://www . matrixscience . com ) and with the following parameters: 2 missed cleavages , carbamidomethylation of Cysteine as fixed modification and oxidation of Methionine , phosphorylation of Threonine and Serine as variable modifications . MS tolerance was set to 20ppm for parent ions and 0 . 05 Da for fragment ions . For SILAC , samples were prepared as described ( Skorupa et al . , 2013 ) . Peptides were analysed online by nano-flow HPLC–nanoelectrospray ionization using an Q Exactive mass spectrometer ( Thermo Fisher Scientific ) coupled to an Ultimate 3000 RSLC ( Dionex , Thermo Fisher Scientific ) . Desalting and pre-concentration of samples were performed on-line on a Pepmap pre-column ( 0 . 3 mm × 10 mm , Dionex ) . A gradient consisting of 0–40% B in A for 60 min , followed by 80% B/20% A for 15 min ( A = 0 . 1% formic acid , 2% acetonitrile in water; B = 0 . 1% formic acid in acetonitrile ) at 300 nL/min was used to elute peptides from the capillary reverse-phase column ( 0 . 075 mm × 150 mm , Pepmap , Dionex ) . Raw data analysis was performed using the MaxQuant software ( v . 1 . 5 . 0 . 0 ) ( Cox and Mann , 2008 ) using standard parameters except Requantity option set as TRUE or FALSE . Peak lists were searched against the UniProt Mouse database ( release 2015_11; http://www . uniprot . org ) , 255 frequently observed contaminants as well as reversed sequences of all entries . Graphical representations were generated using perseus ( 1 . 5 . 3 . 2 ) . qPCR was performed using LightCycler 480 SYBR Green I Master ( Roche , Grenoble , France ) and LightCycler 480 qPCR machine . The reaction contained 5 ng of cDNA , 2 µL of 1 μM qPCR primer pair ( final concentration of each primer was 200 nM in reaction mixture ) , 5 µL 2x Master Mix , and final volume made up to 10 µL with DNase free water . qPCR was conducted at 95°C for 10 min , and then 40 cycles of 95°C for 20 s , 58°C for 20 s and 72°C for 20 s . The specificity of the reaction was verified by melt curve analysis . Each sample was performed in three replicates . TargetForwardReversehuman MKI67 QiagenQiagen QuantiTect Hs_MKI67_1_SGhuman MKI67TGACCCTGATGAGAAAGCTCAACCCTGAGCAACACTGTCTTTThuman CCNA2AGGAAAACTTCAGCTTGTGGGCACAAACTCTGCTACTTCTGGGhuman CCNE1CCGGTATATGGCGACACAAGACATACGCAAACTGGTGCAAhuman CCNB1TGTGTCAGGCTTTCTCTGATGTTGGTCTGACTGCTTGCTCThuman CDC6TTGCTCAGGAGATTTGTCAGGGCTGTCCAGTTGATCCATCTChuman FZR1TCTCAGTGGAAGGGGACTCACAACATGGACAGCTTCTTCCChuman B2M ( norm ) GCGCTACTCTCTCTTTCTGGAGAAAGACCAGTCCTTGCTGAhuman RPL19 ( norm ) ATGCCGGAAAAACACCTTGGGTGACCTTCTCTGGCATTCGmouse Mki67AATCCAACTCAAGTAAACGGGGTTGGCTTGCTTCCATCCTCAmouse B2M ( norm . ) GGTCTTTCTGGTGCTTGTCTGCAGTTCAGTATGTTCGGCTT For analysis of high-molecular-weight species , 5 μg of total RNA were resolved on agarose denaturing gels ( 6% formaldehyde/1 . 2% agarose in HEPES-EDTA buffer ) . For the analysis of the low-molecular-weight RNA species 5 μg of total RNA were separated on denaturing acrylamide gels ( 8% acrylamide-bisacrylamide 19:1/8 M urea in Tris-borate-EDTA buffer [TBE] ) for 4 hr at 350 V . Agarose gels were transferred by capillarity overnight in 10× saline sodium citrate ( SSC ) and acrylamide gels by electrotransfer in 0 . 5× TBE on nylon membranes ( GE Healthcare ) . Membranes were prehybridized for 1 hr at 65°C in 50% formamide , 5× SSPE , 5× Denhardt’s solution , 1% w/v SDS , 200 μg/ml fish sperm DNA solution ( Roche ) . The 32P-labeled oligonucleotide probe was added and incubated for 1 hr at 65°C and then overnight at 37°C . Oligo probe nameSequenceLD1827 ( ITS1 ) CCTCGCCCTCCGGGCTCCGGGCTCCGTTAATGATCLD1828 ( ITS2 ) CTGCGAGGGAACCCCCAGCCGCGCALD1829GCGCGACGGCGGACGACACCGCGGCGTCLD1844 ( 5'-ETS ) CGGAGGCCCAACCTCTCCGACGACAGGTCGCCAGAGGACAGCGTGTCAGCLD2079 ( 5'-ITS2 ) GGGGCGATTGATCGGCAAGCGACGCTCLD2132 ( 5 . 8S mat ) CAATGTGTCCTGCAATTCACLD2133 ( 7SL ) GCTCCGTTTCCGACCTGGGCCLD2655GGAGCGGAGTCCGCGGTG Plasmids encoding two TALEN pairs were purchased from Cellectis ( Paris , France ) . They were designed to bind the following sequence of Mki67 gene: 5’ TCCCGACGGCCGGGCGGACCATGGCGTCCTC GGCTCACCTGGTCACCA 3 . Underlined sequences are recognised by the left or the right TALEN , respectively . Plasmids were linearized by PacI digestion and used as a template forin vitro transcription to produce TALEN-encoding mRNAs using T7 RiboMAX Express System ( Promega ) . Transcripts were purified using MEGAclear Transcription Clean-Up Kit ( Ambion , ThermoFisher ) . Quality and quantity of transcribed mRNAs were verified by BioAnalyzer ( Agilent , Paris , France ) . Next , 32 ng or 8 ng of each TALEN-encoding mRNAs were injected into zygotes and implanted into 36 or 18 mice , respectively . 7 chimeric mutant mice were obtained with deletions ranging between 1nt and 38 nt by injection of 32 ng of each mRNAs ( 22% NHEJ ) and 1 chimeric mouse with deletion of 24 nt by injection of 8 ng of each mRNAs . Founder mice were crossed and we obtained four mice for germline transmission ( 1nt , 2nt , 3nt , 24nt deletion ) . Genomic DNA was purified from mouse-tail piece using KAPA Express Extract Kit ( Kapa Biosystems , London , UK ) . PCR was conducted using the primers 5’GGCCAGAGCTAACTTGCGCTGACTG 3 ‘and 5’AAACAGGCAGGAGCTGAGGCTCAGC 3 ‘and Pfu DNA Polymerase ( Promega ) . Product size 203 bp . Then , PCR product was cleaned up using ExoSAP-IT ( Affymetrix , High Wycombe , UK ) and sequenced using 5’GGCCAGAGCTAACTTGCGCTGACTG 3 ‘primer . Genotyped mice pups were fixed in 4% paraformaldehyde for 48 hr and formol 10% 3 days prior to after longitudinal section in 2 parts . Embryos were decalcified in EDTA 10% - Formol 2 , 5% before paraffin embedding . Tissue was dehydrated through a series of graded ethanol baths and then infiltrated with wax . Infiltrated tissues were then embedded into wax blocks . From these blocks , 5-μm-thick sections were cut and then stained with hematoxylin . MEF were isolated from E13 . 5 embryos of the corresponding genotype . The female was killed by cervical dislocation . The uterine horns were dissected and placed into a petri dish containing PBS . Each embryo was separated from its placenta and surrounding membranes . The brain , all dark red organs and the intestine were cut away and blood was removed as much as possible . The remaining parts of the embryo were transferred into a dish containing 1 ml of 1x Trypsin-EDTA 0 . 25% . They were finely minced with a razor blade and incubated at 37°C for 1 hr in a 5% CO2 incubator . Trypsin was inactivated with 4 ml of DMEM supplemented with 10% fetal bovine serum and 1% Penicillin/Streptomycin and the carcass was homogenized by several passages up and down using a pipet . Finally , 6 ml of DMEM media were added and cells were incubated at 37°C in an incubator containing 3% O2 and 5% CO2 . Mouse intestinal epithelium was processed for immunohistochemistry as described ( Gerbe et al . , 2011 ) . Cdh1 knockout mice were analysed by immunohistochemistry as described ( Eguren et al . , 2013 ) . NIH-3T3 cells were plated at a density of 3x104/cm2 in the afternoon the day before transfection . Cells were transfected with plasmids encoding: TALEN pair targeted to initial ATG described in section TALEN-mediated Ki-67 KO mouse and pEGFP , or pEGFP by itself . Next day , eGFP positive cells were sorted by FACS ( BD FACSAria ) and around 240 cells from each condition were plated in five 96-well plates ( 480 wells ) . Two weeks later , we obtained around 50 clones from each condition . Then , TALEN-mediated mutants were screened for Ki-67 expression by immunofluorescence . Nine selected clones were then screened by PCR and sequencing . Genomic DNA were purified from harvested cells using KAPA Express Extract Kit ( Kapa Biosystems ) . PCR product was amplified using the primers 5’ AGAGCTAACTTGCGCTGACT 3’ and 5’ TCGCGTCTACCGAGTGTAAAA 3’ and Pfu DNA Polymerase ( Promega ) . Product size 364 bp . For sequencing one additional amplification cycle was performed using Taq Polymerase to add a 3' dA overhang on the end of PCR fragment . Next , PCR product was ligated with pGEM-T Easy Vector ( Promega ) . Competent bacteria were transformed with ligation reaction and plated on agar plates with ampicillin , IPTG and X-Gal . Next day ten white colonies were selected from each individual ligation reaction to perform plasmid preparation . Purified plasmids were sequenced using T7 and SP6 RNA Polymerase transcription initiation site primers . TALENs were designed using TAL Effector Nucleotide Targeter 2 . 0 ( Cornell University ) software to bind following sequence of Mki67 gene: 5’ TACCAGAAAAGTGAAACTATGTAGCAAAGACATTTAAGAAGGAAAAGT 3’ and assembled using The Golden Gate TALEN kit ( AddGene ) . Underlined sequences are recognised by the left TALEN or the right TALEN , respectively . NIH-3T3 cells were plated at density of 3x104/cm2 in the afternoon the day before transfection . Cells were transfected with plasmids encoding: TALEN pair targeted to initial ATG described in section TALEN-mediated Ki-67 KO mouse; TALEN pair targeted to site of STOP codon MKI67 gene; reporter system ( Kim et al . , 2013 ) containing STOP codon area as a target sequence; linearized construct containing Mki67 locus replaced by eGFP gene . Two days after transfection , hygromycin selection was performed by culturing the cells in the presence of 2 mg/ml of hygromycin B for two days at 37°C . For clonal analysis , around 500 hygromycin-selected cells were plated in ten 96-well plates ( 960 wells ) . Two weeks after , around 100 clones ( 10% ) were screened by immunofluorescence for Ki-67 expression . Analysis of DNA replication progress in cells was achieved by treatment with 10 μM 5-ethynyl-2'-deoxyuridine ( EdU ) ( ThermoFisher ) before fixation . Replicating cells were visualized following the protocol from Click-iT EdU Alexa Fluor 488 Imaging Kit ( ThermoFisher ) . Analysis of newly synthesised RNA in cells was achieved by treatment with 2 mM 5-ethynyl uridine ( EU ) ( ThermoFisher ) for 20 min before fixation . Replicating cells were visualized following the protocol from Click-iT EU Alexa Fluor 488 Imaging Kit ( ThermoFisher ) . RNA was prepared using RNeasy Mini Kit ( Qiagen ) following the manufacturer’s instructions from U2OS shRNA non-targeting CTRL , U2OS shRNA Ki-67 , HeLa shRNA non-targeting control CTRL , HeLa shRNA Ki-67 grown 3 months after initial infection with lentivirus armed pGIPZ shRNA . RNA was purified from three shRNA CTRL tumour xenografts or three shRNA Ki-67 tumour xenografts isolated from mouse C1-SM ( 33 days after injection ) , C1-OD ( 41 days after injection ) and C2-2OR ( 46 days after injection ) using TRIzol reagent ( Life Technologies ) following the manufacturer’s instructions . Cy3-labelled cRNA was amplified and hybridized on the Agilent SurePrint G3 Human GE 8x60k Microarray according to the procedures by Imaxio company ( Lyon , France ) . Raw data were preprocessed using GeneSpring GX software ( Agilent Technologies ) to define differently expressing genes and present data by clustered heat-maps . Significant differences between experimental groups were determined using an unpaired two-tailed Student t test in Prism 5 ( GraphPad ) . For all analyses , p values <0 , 05 ( * ) , p values < 0 , 01 ( ) and p values <0 , 001 ( * ) were considered statistically significant . Transcripts that ( i ) demonstrated at least a 1 , 5-fold change in expression , ( ii ) had a greater-than-background signal intensity value and were determined to be 'Present' by Affymetrix algorithms , and ( iii ) had a value that was significant by Student's t test and FDR ( Benjamini Hochberg ) correction ( p<0 . 02 ( U2OS , HeLa ) ; p<0 . 2 ( Xenografts ) ) were considered differentially expressed .	Living cells divide in two to produce new cells . In mammals , cell division is strictly controlled so that only certain groups of cells in the body are actively dividing at any time . However , some cells may escape these controls so that they divide rapidly and form tumors . A protein called Ki-67 is only produced in actively dividing cells , where it is located in the nucleus – the structure that contains most of the cell’s DNA . Researchers often use Ki-67 as a marker to identify which cells are actively dividing in tissue samples from cancer patients , and previous studies indicated that Ki-67 is needed for cells to divide . However , the exact role of this protein was not clear . Before cells can divide they need to make large amounts of new proteins using molecular machines called ribosomes and it has been suggested that Ki-67 helps to produce ribosomes . Now , Sobecki et al . used genetic techniques to study the role of Ki-67 in mice . The experiments show that Ki-67 is not required for cells to divide in the laboratory or to make ribosomes . Instead , Ki-67 alters the way that DNA is packaged in the nucleus . Loss of Ki-67 from mice cells resulted in DNA becoming less compact , which in turn altered the activity of genes in those cells . Sobecki et al . also identified many other proteins that interact with Ki-67 , so the next step following on from this research is to understand how Ki-67 alters DNA packaging at the molecular level . Another future challenge will be to find out if inhibiting the activity of Ki-67 can hinder the growth of cancer cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2016	The cell proliferation antigen Ki-67 organises heterochromatin
Severe falciparum malaria has substantially affected human evolution . Genetic association studies of patients with clinically defined severe malaria and matched population controls have helped characterise human genetic susceptibility to severe malaria , but phenotypic imprecision compromises discovered associations . In areas of high malaria transmission , the diagnosis of severe malaria in young children and , in particular , the distinction from bacterial sepsis are imprecise . We developed a probabilistic diagnostic model of severe malaria using platelet and white count data . Under this model , we re-analysed clinical and genetic data from 2220 Kenyan children with clinically defined severe malaria and 3940 population controls , adjusting for phenotype mis-labelling . Our model , validated by the distribution of sickle trait , estimated that approximately one-third of cases did not have severe malaria . We propose a data-tilting approach for case-control studies with phenotype mis-labelling and show that this reduces false discovery rates and improves statistical power in genome-wide association studies . Severe malaria caused by the parasite Plasmodium falciparum kills nearly half a million children each year , mostly in sub-Saharan Africa ( World Health Organization , 2020 ) . By causing death in children before they reach their reproductive age , P . falciparum has exerted a substantial selective evolutionary pressure on the human genome ( Carter and Mendis , 2002; Kariuki and Williams , 2020 ) . Recent advances in whole-genome sequencing and haplotype imputation ( Teo et al . , 2010 ) , combined with data gathered prospectively from large patient cohorts , have improved our understanding of genetic susceptibility to P . falciparum infection and severe disease ( Malaria Genomic Epidemiology Network et al . , 2013; Malaria Genomic Epidemiology Network , 2014; Band et al . , 2019; Malaria Genomic Epidemiology Network et al . , 2017 ) , but many questions remain unanswered ( Kariuki and Williams , 2020 ) . A major limitation of genetic association studies in severe malaria is that the diagnosis of severe falciparum malaria in children is imprecise ( White et al . , 2013; Taylor et al . , 2004; Bejon et al . , 2007 ) . This imprecision increases with transmission intensity because of the low positive predictive value of a ‘positive blood film’ or rapid diagnostic test ( RDT ) in areas where the background prevalence of microscopy detectable parasitaemia in apparently healthy young children is high ( often around 30% , Rodriguez-Barraquer et al . , 2018 , but can exceed 90% , Smith et al . , 1994 ) . Severe falciparum malaria has been defined by experts convened by the World Health Organization ( WHO ) as clinical or laboratory evidence of vital organ dysfunction in the presence of circulating asexual P . falciparum parasitaemia ( World Health Organisation , 2014 ) . The WHO definition of severe malaria is aimed primarily at clinicians and health care workers managing patients with malaria who appear severely ill . This appropriately prioritises sensitivity over specificity ( Anstey and Price , 2007 ) . An inclusive clinical definition ensures that cases are not missed and patients receive the best treatment . In contrast , genetic association studies require high specificity ( Zondervan and Cardon , 2007 ) . For a given sample size , their statistical power , false discovery rates ( FDRs ) and the validity of their interpretation are weakened by phenotypic inaccuracy . Specificity in the diagnosis of severe malaria depends in part on the prevalence of malaria parasitaemia . This reflects background transmission intensity . In areas of low or seasonal transmission ( e . g . most of endemic Asia and the Americas ) , clinical and laboratory signs of severity accompanied by a positive blood film for P . falciparum are highly specific for severe malaria , which predominantly affects young adults . In contrast in high transmission areas in sub-Saharan Africa and in lowland areas of the island of New Guinea , where severe malaria is largely a disease of young children , the diagnostic criteria for defining severe malaria are less specific because of the high background prevalence of asymptomatic parasitaemia and the lower specificity of the clinical manifestations . Standard case definitions of severe malaria will therefore inevitably include both patients with non-malarial severe illness with concomitant parasitaemia and with concomitant non-severe malaria . Our goal was to develop a biomarker-based model that can differentiate probabilistically between ‘true severe malaria’ and severe illness not caused primarily by malaria , but with concomitant parasitaemia . We define ‘true severe malaria’ conceptually as a febrile illness caused by malaria parasites , with organ dysfunction , that can result in death whereby mortality is attributable directly to the malaria parasites . This attributable mortality can be given a formal causal definition by using a conceptual ( albeit unethical ) randomised experiment of delayed versus prompt antimalarial therapy . In a theoretical patient population with true severe malaria , delay in administration of an effective antimalarial would result in increased mortality ( Warrell et al . , 1982; Gomes et al . , 2009 ) whereas in a population with severe illness not caused by malaria ( ‘not severe malaria’ ) there would not be a corresponding increase in mortality . We developed a probabilistic diagnostic model of severe malaria based on haematological biomarkers using data from 1704 adults and children mainly from low transmission settings whose diagnosis of severe malaria is considered to be highly specific . We used this model to demonstrate low phenotypic specificity in a cohort of 2220 Kenyan children who were diagnosed clinically with severe malaria . We validated the predictions using a natural experiment , the distribution of sickle cell trait ( HbAS ) , the genetic polymorphism with the strongest known protective effect against all forms of clinical malaria ( Malaria Genomic Epidemiology Network , 2014 ) . Building on work on ‘data-tilting’ ( Nie et al . , 2013 ) , we suggest a new method for testing genetic associations in the context of case-control studies in which cases are re-weighted by the probability that the severe malaria diagnosis is correct under the model . As proof of concept , we ran a genome-wide association study across 9 . 6 million imputed biallelic variants using the subset of cases with genome-wide genotype data ( n = 1297 ) and population controls ( n = 1614 ) . Adjusting for case mis-classification decreased genome-wide FDRs ( Storey , 2002 ) and increased effect sizes in three of the top regions of the human genome most strongly associated with protection from severe malaria in East Africa ( HBB , ABO and FREM3 , Band et al . , 2019 ) . A re-analysis of 120 directly typed polymorphisms in 70 candidate malaria-protective genes in the 2220 Kenyan cases and 3940 population controls , examining differential effects between correctly and incorrectly classified cases , suggests that the protective effect of glucose-6-phosphate dehydrogenase ( G6PD ) deficiency has been obscured in this population by case mis-classification . Our results show that adding full blood count metadata – routinely measured in most hospitals in sub-Saharan Africa – to severe malaria cohorts would lead to more accurate quantitative analyses in case-control studies and increased statistical power . We used the joint distribution of platelet counts and white blood cell counts ( both on a logarithmic scale ) to develop a simple biomarker-based reference model of severe malaria . To fit the reference model ( i . e . P[Data \| Severe malaria] ) , we used platelet and white count data from ( i ) severe malaria patient cohorts enrolled in low transmission areas where severe disease accompanied by a positive blood stage parasitaemia has a high positive predictive value for severe malaria ( 930 adults from Vietnam [Hien et al . , 1996; Phu et al . , 2010] and 653 adults and children from Thailand and Bangladesh ) ; and ( ii ) severely ill African children with plasma PfHRP2 concentrations >1000 ng/mL and >1000 parasites per μL of blood ( 121 children from Uganda , Maitland et al . , 2011 ) . Severe illness accompanied by a high plasma PfHRP2 concentration makes the diagnosis of severe falciparum malaria highly specific ( Hendriksen et al . , 2012 ) . The joint distribution of platelet and white blood cell counts in severe malaria was modelled as a bivariate t-distribution with both blood count variables on the log10 scale . Figure 1A shows the reference data ( green triangles: patients with a highly specific diagnosis of severe malaria , summarised in Table 1 ) alongside data from a large Kenyan cohort of hospitalised children diagnosed with severe malaria , whose diagnosis had unknown specificity ( pink squares ) . The median platelet count in the reference data was 57 , 000 per μL , and the median total white blood cell count was 8400 per μL . In contrast , the median platelet count in the Kenyan children was 120 , 000 per μL , and the median total white blood cell count was 13 , 000 per μL . Direct comparisons of white counts across these two datasets are confounded by geography and age . Total white blood cell counts are known to be age-dependent and vary across genetic backgrounds , in particular lower neutrophil counts are associated with mutations in the ACKR1 gene that results in the Duffy negative phenotype prevalent in African populations ( Reich et al . , 2009 ) . However , after adjustment for age ( see Materials and methods ) , the marginal distributions of total white counts were comparable between Asian adults and children with severe malaria and African children with high plasma PfHRP2 ( Appendix 1 ) . Platelet counts are not age-dependent and do not vary substantially across genetic backgrounds . The marginal distributions of platelet counts were comparable between Asian adults and children with severe malaria and African children with high plasma PfHRP2 ( Appendix 2 ) . A low platelet count ( thrombocytopenia ) is a universal feature of severe malaria ( see evidence collated in Materials and methods ) . To illustrate this important point , in a cohort of 566 severely ill Ugandan children enrolled in the Fluid Expansion as Supportive Therapy ( FEAST ) trial ( Maitland et al . , 2011 ) , a trial including all severe illness not restricted to severe malaria , low platelet counts were highly predictive of blood stage parasitaemia and elevated PfHRP2 ( p=10-16 for a spline term on the log10 platelet count in a generalised additive logistic regression model predicting PfHRP2 >1000 ng/mL , Appendix 2 ) . Children enrolled in the FEAST trial who had significant thrombocytopenia ( <100 , 000 platelets per μL ) had comparable PfHRP2 concentrations to Asian adults diagnosed with severe falciparum malaria ( Figure 1B ) . We can consider the hospitalised Kenyan children in this series as a mixture of two latent sub-populations , ‘severe malaria’ and ‘not severe malaria’ ( i . e an alternative aetiology for severe illness ) . To estimate the proportion of each , we use the distribution of HbAS , the human polymorphism most protective against all forms of clinical falciparum malaria . HbAS provides at least 90% protection against severe malaria ( Taylor et al . , 2012; Malaria Genomic Epidemiology Network , 2014 ) . The causal SNP rs334 was genotyped in 2213 of the Kenyan children , of whom 57 were HbAS . The causal pathways ( a ) or ( b ) in Figure 2 ( note all children have been selected into the study on the basis of clinical symptoms consistent with severe malaria ) show how the distribution of HbAS can be used to infer the marginal probability P ( Severe malaria ) in the Kenyan cohort as the prevalence of HbAS is expected to differ in the two latent sub-populations . We assumed that cases with the highest likelihood values P ( Data \| Severe malaria ) under the reference model ( a bivariate t-distribution fit to the severe malaria reference data ) had a diagnosis of severe malaria that was 100% specific ( top 40% of cases , a sensitivity analysis varied this threshold ) . The cases with lower likelihood values were assumed to be drawn from a mixture of the two latent populations with an unknown mixing proportion; the prevalence of HbAS in the ‘not severe malaria’ subgroup was estimated from a cohort of hospitalised children enrolled in the same hospital and who were malaria blood slide positive but were clinically diagnosed as not having severe malaria ( n = 6748 of whom 364 were HbAS; Uyoga et al . , 2019 ) . We assumed that this diagnosis of ‘not severe malaria’ was 100% specific . Under these assumptions , we estimated that P ( Severe malaria ) = 0 . 64 ( 95% credible interval [C . I . ] 0 . 46–0 . 8 ) , implying that approximately one-third of the 2200 cases are from the ‘not severe malaria’ sub-population ( they have malaria parasitaemia in addition to another severe illness – likely to be bacterial sepsis – Figure 2 ) . We then estimated P ( Severe malaria \| Data ) for each Kenyan case by fitting a mixture model to the reference data and to the Kenyan data jointly . The model assumed that the platelet and white count data for the Kenyan children were drawn from a mixture of P ( Data \| Severe malaria ) and P ( Data \| Not severe malaria ) . The reference data ( Asian adults and children with severe malaria and African children with PfHRP2 >1000 ng/mL ) were assumed to be drawn only from P ( Data \| Severe malaria ) . P ( Data \| Not severe malaria ) was modelled itself as a mixture of bivariate t-distributions . We used an informative prior on the mixture proportion ( ‘severe malaria’ versus ‘not severe malaria’ ) in the Kenyan cases , a beta distribution approximating the posterior estimate from the analysis of HbAS prevalence . Figure 3A shows the bimodal distribution of the posterior individual estimates of P ( Severe malaria \| Data ) . As expected , the individual posterior probabilities of severe malaria were highly predictive of HbAS ( p=10-6 from a generalised additive logistic regression model fit , Figure 3C ) . The individual probabilities were also predictive of in-hospital mortality ( p=10-9 from a generalised additive model fit; Figure 3D ) and admission peripheral blood parasite density ( p=10-25 from a generalised additive model fit; Figure 3E ) . In the top quintile of patients with the highest estimated P ( Severe malaria \| Data ) , the prevalence of HbAS was 0 . 7% ( 3 out of 446 ) . In contrast , for patients in the lowest quintile of estimated P ( Severe malaria \| Data ) , the prevalence of HbAS was 4 . 8% ( 21 out of 444 ) . The patients with a low probability of severe malaria had a substantially higher case fatality ratio ( 18 . 8% mortality for patients in the bottom quintile of P[Severe malaria \| Data] versus 6 . 1% mortality for the top quintile of P[Severe malaria \| Data] ) . This may be explained by the higher case-specific mortality of severe bacterial sepsis ( the most likely alternative cause of severe illness ) . The admission parasite densities in patients with a probability of severe malaria close to 1 were approximately fivefold higher than in patients with a probability of severe malaria close to 0 . The blood culture positive rate was 2 . 1% in the top quintile of P ( Severe malaria \| Data ) and 4 . 4% in the lowest quintile of P ( Severe malaria \| Data ) , and the individual probabilities were predictive of blood culture results ( p=0 . 004 under a generalised additive logistic regression model fit ) . ‘False-positive’ cases reduce statistical power and dilute effect size estimates in case-control studies . We propose a novel approach for case-control studies with phenotypic imprecision based on data-tilting ( Nie et al . , 2013 ) . The idea is to ‘tilt’ the cases towards a pseudo-population with higher specificity for severe malaria . We can do this by re-weighting the data by the probabilities P ( Severe malaria \| Data ) , that is , re-weighting the contribution to the log-likelihood in an association model . We applied this approach as proof of concept to a genome-wide association study using the subset of Kenyan children who had clinical and genome-wide data available ( after quality control checks n = 1297 cases ) and a set of matched population controls ( n = 1614 ) , across 9 . 6 million biallelic variants on the autosomal chromosomes ( Band et al . , 2019 ) . We compared the data-tilting method to the standard non-weighted approach by estimating local FDRs ( Storey , 2002 ) . Compared to the standard non-weighted GWAS , data-tilting substantially increased the number of significant associations for local FDRs in the range of 1–5% ( Figure 4 ) . For example , at an FDR of 2% , the number of significant hits is more than doubled with the additional hits all around known loci associated with protection from severe malaria . We note that if the data weights were not predictive of the true latent phenotype , we would expect fewer significant hits for a given FDR because of the reduction in effective sample size . This is demonstrated by permuting the data weights ( for the cases only ) , which results in 50–75% reduction in the number of significant hits at FDRs < 5% ( Appendix 3 ) . Examining three major genetic regions strongly associated with protection from severe malaria in East Africa ( HBB: HbAS; ABO: O blood group; FREM3: in close linkage with the GYPA/B/E structural variants that encode the Dantu blood group; Band et al . , 2019 ) , the data-tilting approach estimated larger effect sizes compared to the non-weighted model in all three regions ( effect size increases: 30% around HBB , 9% around ABO and 5% around FREM3 ) . This resulted in larger –log10 p-values for HBB and ABO , but slightly smaller for FREM3 ( Figure 5 ) . We note that there was no signal of association at ATP2B4 in this subset , most likely due to limited power ( ATP2B4 had the third largest Bayes factor for association in the largest multicentre GWAS to date , Band et al . , 2019 ) . We re-analysed case-control associations for 120 polymorphisms on 70 candidate malaria-protective genes which were typed directly in the 2220 Kenyan children along with 3940 population controls . In this case-control cohort , 14 polymorphisms had previously been identified as associated with protection or increased risk in severe malaria ( MalariaGEN Consortium et al . , 2018 ) . A re-analysis of these 14 variants using the same models of association as previously published and down-weighting the likely mis-classified cases replicated the majority of associations , with increased effect sizes and increased –log10 p-values ( Appendix 4 ) . For the three major genes ( HBB , ABO , FREM3 ) , effect sizes were increased by 10–30% and associations all had higher significance levels on the –log10 scale ( 0 . 25–1 . 7 ) . The allele frequencies of all three polymorphisms were directly associated with the probability weights , showing increased protection in individuals more likely to have severe malaria ( Appendix 5 ) . Two polymorphisms on the genes ARL14 and LOC727982 , reported previously as associated with protection in severe malaria ( neither of which are related to red cells ) , showed decreased effect sizes and –log10 p-values and are thus potentially spurious hits . We explored whether there was evidence of differential effects in the Kenyan cases using P[Severe malaria \| Data] to assign probabilistically each case to the ‘severe malaria’ versus ‘not severe malaria’ sub-populations . We fitted a categorical logistic regression model predicting the latent sub-population label versus control , where the latent case label was estimated from the weights shown in Figure 3A . This resulted in approximately 1279 cases in the ‘severe malaria’ sub-population and 941 cases in the ‘not severe malaria’ sub-population . Differential effects were tested by comparing the estimated log-odds for the two sub-populations . After accounting for multiple testing , two polymorphisms showed significant differential effects: rs334 ( derived allele encodes haemoglobin S , p=10-6 ) and rs1050828 ( derived allele encodes G6PD + 202T , p=10-3 in the model fit to females only ) , see Figure 6 . As expected , rs334 was associated with protection in both sub-populations ( Scott et al . , 2011; Uyoga et al . , 2019 ) but the effect was almost eight times larger on the log-odds scale in the ‘severe malaria’ sub-population relative to the ‘not severe malaria’ sub-population ( odds ratio of 0 . 029 [95% C . I . 0 . 0088–0 . 094] in the ‘severe malaria’ population versus 0 . 63 [95% C . I . 0 . 48–0 . 83] in the ‘not severe malaria’ population ) . For rs1050828 ( G6PD + 202T allele ) , approximately the same absolute log-odds were estimated for both sub-populations but they had opposite signs . Under an additive model in females , the rs1050828 T allele was associated with protection in the ‘severe malaria’ sub-population ( odds ratio of 0 . 71 [95% C . I . 0 . 57–0 . 88] ) but with increased risk in the ‘not severe malaria’ sub-population ( odds ratio of 1 . 30 [95% C . I . 1 . 00–1 . 70] ) . The additive model including both males and females was consistent with these opposing effects but significant only at a nominal threshold ( p=0 . 02 ) . Opposing effects across the two sub-populations are consistent with the hypothesis that G6PD deficiency leads to a greater risk of being erroneously classified as severe malaria as under the severe anaemia criterion ( Watson et al . , 2019 ) , shown in more detail in Appendix 5 . Investigation of haemoglobin concentrations as a function of P ( Severe malaria \| Data ) indicates that the mis-classified group is very heterogeneous , but with a larger proportion of severe anaemia ( <5 g/dL ) relative to the correctly classified sub-population ( Appendix 6 ) . The clinical diagnosis of severe falciparum malaria in African children is imprecise ( Taylor et al . , 2004; Bejon et al . , 2007; White et al . , 2013 ) . Even with quantitation of parasite densities , specificity is still imperfect ( Bejon et al . , 2007 ) . In children with cerebral malaria ( unrouseable coma with malaria parasitaemia ) , the most specific of the severe malaria clinical syndromes , postmortem examination revealed another diagnosis in a quarter of cases studied in Blantyre , Malawi ( Taylor et al . , 2004 ) . Diagnostic specificity can be improved by visualisation of the obstructed microcirculation in vivo ( e . g . through indirect ophthalmoscopy ) or from parasite biomass indicators ( quantitation and staging of malaria parasites on thin blood films , counting of neutrophil-ingested malaria pigment , measurement of plasma concentrations of PfHRP2 or parasite DNA ) , but these are still largely research procedures and have not been widely adopted or measured at scale for genetic association studies . Our results suggest that imprecision in clinical phenotyping is more substantial than thought previously . In this cohort of 2220 Kenyan children diagnosed with severe malaria from an area of moderate transmission , a probabilistic assessment suggests that around one-third may not have had severe malaria ( although malaria may have contributed to their illness; Small et al . , 2017 ) . This supports our previous conclusion that differences in treatment effects between Asian adults and African children ( i . e the benefits of artesunate over quinine in severe malaria estimated from randomised trials; Dondorp et al . , 2005; Dondorp et al . , 2010 ) are predominantly driven by differences in diagnostic specificity ( Hendriksen et al . , 2012; White et al . , 2013 ) . Mortality was higher in the severe ‘not malaria’ patients , probably because the main illness was bacterial sepsis . This strongly supports current recommendations to give broad-spectrum antibiotics to all children in endemic areas with suspected severe malaria ( World Health Organisation , 2014 ) . Using HbAS as a natural experiment to validate the biomarker model , we show that the joint distribution of platelet and white blood cell counts is a diagnostic predictor of severe malaria . Complete blood counts are inexpensive and increasingly available in low-resource setting hospitals . Application of an upper threshold of 200 , 000 platelets per μL would have substantially decreased mis-classification in this large cohort of Kenyan children diagnosed with severe malaria . This re-analysis using rich clinical data provides additional evidence for the three major genetic polymorphisms protective against severe malaria present in East Africa . After probabilistic down-weighting of the likely mis-classified cases , substantial increases in effect sizes were found . Dilution of effect sizes resulting from mis-classification could partially explain the large heterogeneity in effects noted in the largest severe malaria GWAS to date ( Band et al . , 2019 ) . For haemoglobin S ( rs334 ) , there was a fourfold variation in estimated odds ratios across participating sites . Some of this heterogeneity can be attributed to variations in linkage disequilibrium affecting imputation accuracy ( Malaria Genomic Epidemiology Network et al . , 2013 ) , but our analysis shows an additional substantial source of heterogeneity which results from diagnostic imprecision . This can be adjusted for if detailed clinical data are available . For example , in the case of rs334 ( directly typed ) , the data-tilting approach results in a 25% increase in effect size on the log-odds scale , corresponding to 35% decrease in estimated odds ratios ( 0 . 1 versus 0 . 16 ) . As for the interpretation of genetic effects , one of the most interesting results concerns the G6PD gene . G6PD deficiency is the most common enzymopathy of humans . Its potential role in protecting against falciparum malaria has been controversial ( MalariaGEN Consortium et al . , 2017; Watson et al . , 2019 ) . A very large multi-country genetic association study with over 11 , 000 severe malaria cases and 17 , 000 population controls found no overall protective effect of the G6PD + 202T allele ( the most common mutation in sub-Saharan Africa causing G6PD deficiency ) , under an additive model ( Malaria Genomic Epidemiology Network , 2014 ) . The same pattern is observed in this Kenyan cohort ( which is a subset of the larger study ) . In the Kenyan cohort overall , a previous analysis found no clear evidence of protection for male homozygotes but substantial evidence of protection for female heterozygotes ( MalariaGEN Consortium et al . , 2015 ) . This would suggest a heterogyzote advantage leading to a balancing polymorphism . However , when the Kenyan cases are modelled as two distinct sub-populations , there is evidence of differential effects between the ‘severe malaria’ and ‘not severe malaria’ sub-populations . Hemi- and homozygous G6PD deficiency was associated with an increased risk of mis-classification ( reflecting an increased risk of severe anaemia ) , but it is unclear whether or not hemi/homozygous G6PD deficiency was protective in the 'true severe malaria' sub-population ( Figure 6C ) . On the other hand , heterozygote deficiency was very clearly protective in the true severe malaria subgroup , consistent with previous findings , and did not appear to lead to an increased risk of mis-classification ( consistent with a lower risk of extensive haemolysis and thus false classification in heterozygotes who have both normal and G6PD-deficient erythrocytes in their circulation ) . When examining the ‘severe malaria’ sub-population only , the sample size in this study is too small to discriminate between the heterozygote and additive models of association . In our view , the relationship between G6PD deficiency and severe falciparum malaria remains unanswered . A biomarker-driven approach should be applied to other case-control cohorts for a definitive understanding of the role of this major human polymorphism . The limitations of our diagnostic model can be summarised as follows . First , the validity and interpretation of the individual probabilities of severe malaria is heavily dependent on the reference model and thus the reference data . Our reference data were primarily from Asian adults in whom diagnostic specificity for severe malaria is thought to be very high . Diagnostic checks suggested that the marginal distributions of platelet counts were similar between adults and children , and we made age corrections to the white blood cell count , but small deviations could reduce the discriminatory value ( e . g . lower white counts associated with the Duffy negative phenotype; Reich et al . , 2009 ) . Second , it is possible that rare genetic conditions exist in which the probabilities of severe malaria under this model might be biased . One example is sickle cell disease ( HbSS , <0 . 5% in the Kenyan cases ) , which results in chronic inflammation with high white counts and low platelet counts relative to the normal population ( Sadarangani et al . , 2009 ) . The 11 children with HbSS in this cohort were all assigned low probabilities of severe malaria , but this should be interpreted with caution . Whether HbSS is protective against severe malaria or increases the risk of severe malaria remains unclear ( Williams and Obaro , 2011 ) . For these patients , other biomarkers such as plasma PfHRP2 may be more appropriate . Third , it is possible that the joint distribution of the complete blood count variables used to fit the reference model could be dependent on the severe malaria sub-phenotype . For example , if the reference data were biased towards cerebral malaria , and the joint distribution of platelet and white cell counts in cerebral malaria differed from those in the other severe malaria syndromes , then the predicted outliers could represent other forms of severe malaria instead of ‘not severe’ malaria . However , there are no known biological reasons why this would be the case . The strong correlation between platelet counts and PfHRP2 ( Figure 1B ) suggests that low platelet counts are a universal feature of severe malaria . In summary , under a probabilistic model based on routine blood count data , we have shown that it is possible to estimate mis-classification rates in diagnosed severe childhood malaria in a malaria endemic area of East Africa and compute probabilistic weights that can downweight the contribution of likely mis-classified cases . The well-established protective effect of HbAS provided an independent validation of the model . Relative to predicted mis-classified cases , patients predicted to have ‘true severe malaria’ had a substantially lower prevalence of HbAS , higher parasite densities , lower rates of positive blood cultures and lower mortality . These data strongly support the current guideline to give broad-spectrum antibiotics to all children with suspected severe malaria and suggest that normal range platelet counts ( >200 , 000 per μL ) could be used as a simple exclusion criterion in studies of severe malaria . Based on this analysis , we recommend that future studies in severe malaria collect and record complete blood count data . Further studies of platelet and white blood cell counts from a diverse cohort of children with severe falciparum malaria , confirmed using high-specificity diagnostic techniques such as visualisation of the microcirculation , and measurement of plasma PfHRP2 , or plasma P . falciparum DNA concentrations , should be conducted to validate this approach . In the Kenyan severe malaria cohort ( n = 2220 ) , data on platelet counts were missing in 18% , white blood counts were missing in 0 . 2% and parasite density was missing in 1 . 6% . In-hospital outcome ( died/survived ) was missing for 13 patients . rs334 genotype was missing for 7; α+-thalassaemia genotype was missing for 101 patients . In the Vietnamese adults , platelet counts were missing in 4% , white counts in 2% and parasitaemia in 0% . We did multiple imputation using random forests for all available clinical variables using the R package missForest ( targeted genotyping data was not included for imputation ) . Appendix 7 shows the missing data pattern in the studies in Vietnamese adults and in the Kenyan severe malaria cases . Ten datasets were imputed for each dataset independently and were used for the subsequent analyses . Analyses using directly typed genetic polymorphisms or the within-hospital outcome as the dependent variables used only the data where these outcomes were recorded , assuming that they were missing at random . We assume that the Kenyan cases are a latent mixture of two sub-populations: P0 is the population ‘severe malaria’ and P1 is the population ‘not severe malaria’ ( mis-classified ) . For a set of diagnostic biomarkers X , this implies that X∼G=π⁢f0+ ( 1-π ) ⁢f1 , where f0 , f1 are the sampling distributions ( likelihoods ) of each sub-population , respectively . We can infer the value of π ( proportion correctly classified as severe malaria ) without making parametric assumptions about f1 by using the distribution of HbAS ( motivated by the causal pathways shown in Figure 2 ) . This is done as follows: we first estimate f^0 by fitting a bivariate t-distribution to the reference data – this approximates the sampling distribution for P0 . We then make three assumptions: Under these assumptions , we can fit a Bayesian binomial mixture model to these data with three parameters: {π′ , p0 , p1} . The likelihood is given byN0sickle∼Binomial ( p0 , N0 ) NGsickle∼Binomial ( π′p0+ ( 1−π′ ) p1 , NG ) N1sickle∼Binomial ( p1 , N1 ) The priors used were p1∼Beta⁢ ( 5 , 95 ) ( i . e . 5% prior probability with 100 pseudo observations ) ; p0∼Beta⁢ ( 1 , 99 ) ( 1% prior probability with 100 pseudo observations ) . A sensitivity analysis with flat beta priors ( Beta[1 , 1] ) did not qualitatively change the result ( by one percentage point for the final estimate of π ) . To check the validity of the use of the external population from Uyoga et al . , 2019 , we did a sensitivity analysis using the lowest quintile of the likelihood ratio distribution as a population drawn entirely from P1 ( instead of the external data from Uyoga et al . , 2019 ) . Denote the platelet and white count data from the FEAST trial as {XiFEAST}i=1121; the data from the Vietnamese adults and children as {XiAsia}i=11583; the data from the Kenyan children as {XiKenya}i=12220 . We fit the following joint model to the reference biomarker data and the Kenyan biomarker data . XiFEAST∼Student ( μSM1 , ΣSM1 , 7 ) XiAsia∼Student ( μSM2 , ΣSM2 , 7 ) XiKenya∼πf0+ ( 1−π ) f1f0=p Student ( μSM1 , ΣSM1 , 7 ) + ( 1−p ) Student ( μSM2 , ΣSM2 , 7 ) f1=∑j=1Kαj Student ( μnotSMj , ΣnotSMj , 7 ) with the following prior distributions and hyperparameters , where α={α1 , . . , αK} such that ∑j=1Kαj=1 :π∼Beta ( 40 . 3 , 24 . 7 ) p∼Beta ( 2 , 2 ) μSM1 , 2∼Normal ( {1 . 8 , 0 . 95} , 0 . 12 ) μnotSM1 . . K∼Normal ( {2 . 5 , 1 . 5} , 0 . 252 ) α∼Dirichlet ( 1/K , . . . , 1/K ) The covariance matrices ΣS⁢M1 , 2 and ΣSM1 . 6 were parameterised as their Cholesky LKJ decomposition , where the L correlation matrices had a uniform prior ( i . e . hyperparameter ν = 1 ) . The model was implemented in rstan . This models the biomarker data in ‘not severe malaria’ as a mixture of K t-distributions . We chose K=6 as the default choice ( sensitivity analysis increasing this has no impact ) . The Dirichlet prior with hyperparameter 1/K forces sparsity in this mixture model ( most of the prior weight is on the vertices of the K-dimensional simplex ) ; see , for example , Frühwirth-Schnatter and Malsiner-Walli , 2019 . This is a very general and flexible way of modelling the ‘not severe malaria’ distribution: we are not trying to make inferences about this distribution , we just want the mixture model to be flexible enough to describe it . The model also allows for differences in the joint distribution of platelet counts and white counts between the reference datasets ( FEAST trial and the Asian studies ) . The Kenyan cases drawn from the ‘severe malaria’ sub-population are then modelled as a mix of these two reference models . For each {XiKenya}i=12220 , we estimate the posterior probability of being drawn from the sampling distribution f0 . The mean posterior probability then defines a precision weight wi which can be used in a standard generalised linear model ( glm ) with the same interpretation as inverse probability weights . The weighted glm is equivalent to computing the maximum likelihood estimate where the log-likelihood is weighted by wi . In our case-control analyses , all the controls are given weight 1 . Nie et al . , 2013 give a proof of correctness for this re-weighted log-likelihood ( equivalent to ‘tilting’ the dataset towards the desired distribution f^0⁢ ( X ) ) . The log-odds ratio computed from the weighted logistic regression can be interpreted as the causal effect of the polymorphism on ‘true severe malaria’ relative to the controls , where ‘true severe malaria’ is defined by the sampling distribution f0 . Appendix 12 shows the results of a simulation study demonstrating how the effect estimates and standard error estimates vary as a function of the proportion of mis-classified cases ( as given by the probability weights ) . Anonymised whole-genome data from the Illumina Omni 2 . 5M platform for 1944 severe malaria cases and 1738 population controls were downloaded from the European Genome-Phenome Archive ( dataset accession ID: EGAD00010001742 , release date March 2019; Band et al . , 2019 ) . This contained sequencing data on 2 , 383 , 648 variants . We used the quality control metadata provided with the 2019 data release to select SNPs and individuals with high-quality data . We first excluded 386 individuals ( due to relatedness: 155; missing data or low intensity: 226; gender: 5 ) . We then removed 616 , 426 SNPs that did not pass quality control , leaving a total of 1 , 767 , 222 SNPs . We used plink2 to prune the SNPs ( options: –maf 0 . 01 –indep-pairwise 50 2 0 . 2 ) down to a set of 462 , 120 SNPs in approximate linkage equilibrium . These SNPs were then used to calculated the first five principal components ( Appendix 13 ) , which we subsequently used to control for population structure in the genome-wide association study . We used the Michigan imputation server with the 1000 Genomes Phase 3 ( version 5 ) as the reference panel to impute 28 . 6 million polymorphisms across the 22 autosomal chromosomes . This is a web-based service that runs imputation pipelines ( phasing is done with Eagle2 , imputation with Minimac4 ) . Encrypted results are returned with a one-time password . Of the remaining 3682 individuals ( 1681 cases and 1615 controls ) , we had clinical data available for 1297 cases . We only used the subset of individuals with clinical data available in order for a fair comparison between the weighted and non-weighted genome-wide association studies . We ran subsequent genome-wide association studies on all biallelic sites with a minor allele frequency ≥5% ( 9 , 615 , 446 sites in total ) assuming an additive model of association . We used the R function glm with a binomial link for all tests of association ( genetic data were encoded as the number of ancestral alleles ) . The supplementary appendix gives the R code for weighted logistic regression . The point estimates from the weighted model estimated by glm are correct but it is necessary to transform the standard errors in order to take into account the reduction in effective sample size ( see code ) . We fit a categorical ( multinomial ) logistic regression model to the case-control status as a function of the directly typed polymorphisms ( 120 after discarding those that are monomorphic in this population; see MalariaGEN Consortium et al . , 2018 for additional details ) . We modelled the severe malaria cases as two separate sub-populations with a latent variable: ‘severe malaria’ versus ‘not severe malaria’ , resulting in three possible labels ( controls , ‘severe malaria’ , ‘not severe malaria’ ) . The models adjusted for self-reported ethnicity and sex . The model was coded in stan ( Stan Development Team , 2020 ) using the log-sum-exp trick to marginalise out the likelihood over the latent variables ( see code ) . Normal ( 0 , 5 ) priors were set on all parameters , and parameter estimates and standard errors were estimated from the maximum a posteriori value ( function optimizing in rstan ) . Code , along with a minimal clinical dataset for reproducibility of the diagnostic phenotyping model , is available via a GitHub repository: https://github . com/jwatowatson/Kenyan_phenotypic_accuracy ( Watson , 2021; copy archived at swh:1:rev:03a2de285d38b85a769aa25de46b7960487efc62 ) . A curated minimal clinical dataset is currently available alongside the code on the GitHub repository . This will also be made available at publication via the KEMRI-Wellcome Harvard Dataverse ( https://dataverse . harvard . edu/dataverse/kwtrp ) . This paper used genome-wide genotyping data generated by Band et al . , 2019 , available on request from the European Genome-Phenome Archive ( dataset accession ID: EGAD00010001742 ) . Requests for access to appropriately anonymised clinical data and directly typed genetic variants ( Malaria Genomic Epidemiology Network , 2014 ) for the Kenyan severe malaria cohort can be made by application to the data access committee at the KEMRI-Wellcome Trust Research Programme by email to mmunene@kemri-wellcome . org . The FEAST trial datasets are available from the principal investigator on reasonable request ( k . maitland@imperial . ac . uk ) . Requests for access to appropriately anonymised clinical data from the AQ and AAV Vietnam study and the Asian paediatric cohort can be made via the Mahidol Oxford Tropical Medicine Research Unit data access committee by emailing the corresponding author JAW ( jwatowatson@gmail . com ) or Rita Chanviriyavuth ( rita@tropmedres . ac ) .	In areas of sub-Saharan Africa where malaria is common , most people are frequently exposed to the bites of mosquitoes carrying malaria parasites , so they often have malaria parasites in their blood . Young children , who have not yet built up strong immunity against malaria , often fall ill with severe malaria , a life-threatening disease . It is unclear why some children develop severe malaria and die , while other children with high numbers of parasites in their blood do not develop any apparent symptoms . Genetic susceptibility studies are designed to uncover why such differences exist by comparing individuals with severe malaria ( referred to as ‘cases’ ) with individuals drawn from the general population ( known as ‘controls’ ) . But severe malaria can be a challenge to diagnose . Since high numbers of malaria parasites can be found in healthy children , it is sometimes difficult to determine whether the parasites are making a child ill , or whether they are a coincidental finding . Consequently , some of the ‘cases’ recruited into these studies may actually have a different disease , such as bacterial sepsis . This ultimately affects how the studies are interpreted , and introduces error and inaccuracy into the data . Watson , Ndila et al . investigated whether measuring blood biomarkers in patients ( derived from the complete blood count , including platelet counts and white blood cell counts ) could improve the accuracy with which malaria is diagnosed . They developed a new mathematical model that incorporates platelet and white blood cell counts . This model estimates that in a large cohort of 2 , 220 Kenyan children diagnosed with severe malaria , around one third of enrolled children did not actually have this disease . Further analysis suggests that patients with severe malaria are highly unlikely to have platelet counts higher than 200 , 000 per microlitre . This defines a cut-off that researchers can use to avoid recruiting patients who do not have severe malaria in future studies . Additionally , the ability to diagnose severe malaria more accurately can make it easier to detect and treat other diseases with similar symptoms in children with high numbers of malaria parasites in their blood . Watson , Ndila et al . ’s findings support the recommendation that all children with suspected malaria be given broad spectrum antibiotics , as many misdiagnosed children will likely have bacterial sepsis . It also suggests that using complete blood counts , which are cheap to obtain and increasingly available in low-resource settings , could improve diagnostic accuracy in future clinical studies of severe malaria . This could ultimately improve the ability of these studies to find new treatments for this life-threatening disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health", "genetics", "and", "genomics" ]	2021	Improving statistical power in severe malaria genetic association studies by augmenting phenotypic precision
The diversity of life on Earth is a result of continual innovations in molecular networks influencing morphology and physiology . Plant specialized metabolism produces hundreds of thousands of compounds , offering striking examples of these innovations . To understand how this novelty is generated , we investigated the evolution of the Solanaceae family-specific , trichome-localized acylsugar biosynthetic pathway using a combination of mass spectrometry , RNA-seq , enzyme assays , RNAi and phylogenomics in different non-model species . Our results reveal hundreds of acylsugars produced across the Solanaceae family and even within a single plant , built on simple sugar cores . The relatively short biosynthetic pathway experienced repeated cycles of innovation over the last 100 million years that include gene duplication and divergence , gene loss , evolution of substrate preference and promiscuity . This study provides mechanistic insights into the emergence of plant chemical novelty , and offers a template for investigating the ~300 , 000 non-model plant species that remain underexplored . Since the first proto-life forms arose on our planet some four billion years ago , the forces of mutation , selection and drift have generated a world of rich biological complexity . This complexity , evident at all levels of biological organization , has intrigued humans for millennia ( Tipton , 2008; Mayr , 1985 ) . Plant metabolism , estimated to produce hundreds of thousands of products with diverse structures across the plant kingdom ( Fiehn , 2002; Afendi et al . , 2012 ) , provides striking examples of this complexity . Plant metabolism is traditionally divided into primary and secondary/specialized , the former referring to production of compounds essential for plant development and the latter encompassing metabolites documented as important for plant survival in nature and metabolites of as yet unknown functional significance ( Pichersky and Lewinsohn , 2011; Moghe and Last , 2015 ) . While primary metabolism generally consists of highly conserved pathways and enzymes , specialized metabolic pathways are in a state of continuous innovation ( Milo and Last , 2012 ) . This dynamism has produced numerous lineage-specific metabolite classes such as steroidal glycoalkaloids in Solanaceae ( Wink , 2003 ) , benzoxazinoid alkaloids in Poaceae ( Dutartre et al . , 2012 ) , betalains in Caryophyllales ( Brockington et al . , 2015 ) and glucosinolates in Brassicales ( Halkier and Gershenzon , 2006 ) . The structural diversity produced by lineage-specific pathways makes them exemplary systems for understanding the evolution of novelty in the living world . Previous studies investigating the emergence of lineage-specific metabolite classes uncovered the central role of gene duplication and diversification in this process: for example in biosynthesis of glucosinolates ( Benderoth et al . , 2006; Hofberger et al . , 2013; Edger et al . , 2015 ) , acylsugars ( Ning et al . , 2015; Schilmiller et al . , 2015 ) , the saponin avenacin ( Qi et al . , 2004 ) and various alkaloid types such as benzoxazinoid ( Frey et al . , 1997; Dutartre et al . , 2012 ) , acridone ( Bohlmann et al . , 1996 ) and pyrrolizidine ( Ober and Hartmann , 2000; Kaltenegger et al . , 2013 ) . Duplications of members of enzyme families ( e . g . cytochromes P450 , glycosyltransferases , methyltransferases , BAHD acyltransferases ) also play major roles in generating chemical novelty , with biosynthesis of >40 , 000 structurally diverse terpenoids — produced partly due to genomic clustering of terpene synthases and cytochrome P450 enzymes ( Boutanaev et al . , 2015 ) — as an extreme example . These duplicate genes , if retained , may experience sub- or neo-functionalization via transcriptional divergence and evolution of protein-protein interactions as well as via changes in substrate preference , reaction mechanism and allosteric regulation to produce chemical novelty ( Ohno , 1970; Force et al . , 1999; Moghe and Last , 2015; Leong and Last , 2017 ) . In this study , we sought to understand the molecular processes by which a novel class of plant-defense related metabolites — acylsugars — emerged and diversified in the Solanaceae family . Acylsugars are lineage-specific plant specialized metabolites detected in multiple genera of the Solanaceae family including Solanum ( King et al . , 1990; Schilmiller et al . , 2010; Ghosh et al . , 2014 ) , Petunia ( Kroumova and Wagner , 2003; Liu et al . , 2017 ) , Datura ( Forkner and Hare , 2000 ) and Nicotiana ( Kroumova and Wagner , 2003; Kroumova et al . , 2016 ) . These compounds , produced in the tip cell of trichomes on leaf and stem surfaces ( Schilmiller et al . , 2012; 2015; Ning et al . , 2015; Fan et al . , 2016a ) , typically consist of a sucrose or glucose core esterified to groups derived from fatty acid or branched chain amino acid metabolism ( Figure 1A ) . Despite these simple building blocks , the combinations of the sugar cores and the different acyl chains can generate diverse structures . For example , ~81 acylsugars were detected across just two accessions of the wild tomato Solanum habrochaites ( Ghosh et al . , 2014 ) . Multiple studies performed under controlled lab settings ( Puterka et al . , 2003; Simmons et al . , 2004; Leckie et al . , 2016; Luu et al . , 2017 ) or in the wild ( Weinhold and Baldwin , 2011 ) demonstrated that acylsugars mediate plant-insect and plant-fungus interactions , and hence acylsugar production has been a target for tomato breeding efforts ( Rodríguez-López et al . , 2012; Smeda et al . , 2016 ) . Under lab conditions , the diversity of acyl chains and the sugar cores has been shown to be functionally important in deterring insects such as spider mites , thrips and whiteflies ( Puterka et al . , 2003; Leckie et al . , 2016 ) . In ecological settings , ants living near Datura wrightii plants that produced hexanoic acid-containing acylsugars were significantly more attracted to the smell of hexanoic acid , compared to ants growing near Nicotiana attenuata whose acylsugars lacked hexanoic acid ( Weinhold and Baldwin , 2011 ) . This suggests that acyl chain diversity may be of functional consequence in the wild . However , the contribution of the large number of acylsugar structural variants in plant-insect and plant-microbe interactions is still an open question . Acylsugars , compared to other specialized metabolic classes such as alkaloids , phenylpropanoids and glucosinolates , are biosynthetically rather simple , allowing reconstruction of the pathway in vitro . Previous research from our lab showed that cultivated tomato ( Solanum lycopersicum ) and its wild relatives produce these compounds in the tip cell of the long glandular secreting trichomes using a set of enzymes called acylsugar acyltransferases ( ASATs ) . These enzymes catalyze sequential addition of specific acyl chains to the sucrose molecule using acyl CoA donors ( Figure 1A ) ( Schilmiller et al . , 2012; 2015; Fan et al . , 2016a ) . ASATs are members of Clade III of the large and functionally diverse BAHD enzyme family ( St Pierre and Luca , 2000; D'Auria , 2006 ) ( Figure 1—figure supplement 1 ) . Despite their evolutionary relatedness , S . lycopersicum ASATs ( SlASATs ) are only ~40% identical to each other at the amino acid level . ASATs exhibit different activities across wild tomatoes due to ortholog divergence , gene duplication and neo-functionalization , leading to divergence in acceptor and donor substrate repertoire of ASATs between wild tomato species ( Schilmiller et al . , 2015; Fan et al . , 2016a ) . For example , a recent study showed that the acyl CoA preference of the ASAT2 enzyme orthologs in closely related Solanum species is influenced by a single amino acid change , resulting in accumulation of different acylsugar products ( Fan et al . , 2016a ) . Similarly , duplication of the ASAT3 enzyme followed by retention and duplicate gene divergence led to the emergence of different acylsugar chemotypes across different accessions of S . habrochaites ( Schilmiller et al . , 2015 ) . In addition , loss of the ASAT4 enzyme activity in northern accessions of S . habrochaites leads to accumulation of acylsugars lacking R2 position acetylation ( Kim et al . , 2012 ) . Changes that alter acylsugar profiles can also occur upstream of the biosynthetic pathway — duplication of the isopropylmalate synthase enzyme involved in amino acid biosynthesis and divergence of the duplicate was shown to alter the acyl chain composition in different accessions of S . pennellii , presumably due to changes in the acyl CoA pools in trichomes ( Ning et al . , 2015 ) . These evolutionary mechanisms that contribute to emergence of novel acylsugar phenotypes were studied in closely related wild tomato species and demonstrate the plasticity of the acylsugar biosynthesis in this clade of Solanum . Acylsugar biosynthesis and diversity , although studied very recently in Petunia axillaris ( Petunia ) ( Liu et al . , 2017 ) , is relatively underexplored in the broader Solanaceae family , prompting the question of how the pathway evolved over a much longer time period . In this study , we sought to understand the timeline for emergence of the ancestral ASAT activities and to explore the evolution of the acylsugar biosynthetic pathway since the origin of the Solanaceae . Typically , significant hits obtained using BLAST searches are analyzed in a phylogenetic context to understand enzyme origins ( Frey et al . , 1997; Ober and Hartmann , 2000; Qi et al . , 2004; Benderoth et al . , 2006 ) . However , ASAT orthologs can experience functional diversification due to single amino acid changes and/or duplication ( Schilmiller et al . , 2015; Fan et al . , 2016a ) , precluding functional assignment based on sequence similarity . Thus , we inferred the origins and evolution of the pathway with a bottom-up approach; starting by assessing the diversity of acylsugar phenotypes across the family using mass spectrometry . Our findings not only catalogue the diversity of acylsugars in different plants of the family but also illustrate the varied mechanisms by which the specialized metabolic pathway evolved . These results have broader implications for the study of chemical novelty in the plant kingdom . While the Solanaceae family comprises 98 genera and >2700 species ( Olmstead and Bohs , 2007 ) , there are extensive descriptions of acylsugar diversity reported for only a handful of species ( Severson et al . , 1985; King et al . , 1990; Shinozaki et al . , 1991; Ghosh et al . , 2014 ) . In this study , we sampled vegetative tissue surface metabolites from single plants of 35 Solanaceae and four Convolvulaceae species . These species were sampled at the New York Botanical Gardens and Michigan State University ( Figure 1B; Figure 1—source data 1A ) , and acylsugar profiles were obtained using liquid chromatography-mass spectrometry ( LC/MS ) with collision-induced dissociation ( CID; see Materials and methods ) ( Schilmiller et al . , 2010; Ghosh et al . , 2014; Fan et al . , 2016b ) . Molecular and substructure ( fragment ) masses obtained by LC/MS-CID were used to annotate acylsugars in Solanum nigrum , Solanum quitoense , Physalis alkekengi , Physalis viscosa , Iochroma cyaneum , Atropa belladonna , Nicotiana alata , Hyoscyamus niger and Salpiglossis sinuata ( Salpiglossis ) ( Figure 1B , C; Figure 1—figure supplement 2; Figure 1—source data 2 ) . Plant extracts without detectable acylsugars generally lacked glandular trichomes ( Fisher Exact Test p=2 . 3e-6 ) ( Figure 1—source data 1B ) . In addition , the acylsugar phenotype is quite dynamic and can be affected by factors such as developmental stage , environmental conditions and the specific accession sampled ( Kim et al . , 2012; Ning et al . , 2015; Schilmiller et al . , 2015 ) . These factors may also influence the detection of acylsugars in some species . The suite of detected acylsugars exhibited substantial diversity , both in molecular and fragment ion masses . Most species accumulated acylsugars with mass spectra consistent with disaccharide cores — most likely sucrose — esterified with short- to medium-chain aliphatic acyl groups , similar to previously characterized acylsugars in cultivated and wild tomatoes ( Schilmiller et al . , 2010; Ghosh et al . , 2014 ) . However , S . nigrum acylsugar data suggested exclusive accumulation of acylhexoses ( Figure 1C ) , with fragmentation patterns similar to previously analyzed S . pennellii acylglucoses ( Schilmiller et al . , 2012 ) . Mass spectra of S . quitoense acylsugars also revealed acylsugars with features distinct from any known acylsugars ( Figure 1—figure supplement 2 ) , and structures of these will be described in detail in a separate report . In total , more than 100 acylsucroses and at least 20 acylsugars of other forms were annotated with number and lengths of acyl groups based on pseudomolecular and fragment ion masses ( Figure 1—figure supplement 2 ) . Several hundred additional low-abundance isomers and novel acylsugars were also detected ( Figure 1—figure supplement 3 ) . For example , in Salpiglossis alone , >300 chromatographic peaks had m/z ratios and mass defects consistent with acylsugars ( Figure 1F ) . This acylsugar diversity is notable when compared to ~33 detected acylsucroses in cultivated tomato ( Ghosh et al . , 2014 ) . Mass spectra also revealed substantial diversity in the number and lengths of acyl chains ( Figure 1C–E , Figure 1—figure supplement 2 ) . Based on negative ion mode data , the number of acyl chains on the sugar cores ranged from two to six ( Figure 1—figure supplement 2 ) , with chain lengths from 2 to 12 carbons . Across the Solanaceae , we found species that incorporate at least one common aliphatic acyl chain in all their major acylsugars: for example chains of length C5 in A . belladonna , C4 in P . viscosa , and C8 in P . alkekengi ( Figure 1D , E; Figure 1—figure supplement 2 ) . Longer C10 and C12 chain-containing acylsugars were found in multiple species ( P . viscosa , A . belladonna , H . niger , S . nigrum and S . quitoense ) ( Figure 1C–E , Figure 1—figure supplement 2 ) . Mass spectra consistent with acylsugars containing novel acyl chains were also detected . While we could not differentiate between acyl chain isomers ( e . g . iso-C5 [iC5] vs . anteiso-C5 [aiC5] ) based on CID fragmentation patterns , our data reveal large acyl chain diversity in acylsugars across the Solanaceae . Such diversity between species can result from differences in the intracellular concentrations of acyl CoA pools or divergent substrate specificities of individual ASATs . A previous study from our lab identified allelic variation in the enzyme isopropylmalate synthase 3 , which contributes to differences in abundances of iC5 or iC4 chains in acylsugars of S . lycopersicum and some S . pennellii accessions ( Ning et al . , 2015 ) . ASAT gene duplication ( Schilmiller et al . , 2015 ) , gene loss ( Kim et al . , 2012 ) and single residue changes ( Fan et al . , 2016a ) also influence chain diversity in acylsugars , illustrating the various ways by which acylsugar phenotypes may be generated in the Solanaceae . Published data from S . pennellii and S . habrochaites revealed differences in furanose ring acylation on acylsucroses ( Schilmiller et al . , 2015 ) . Ring-specific acylation patterns can be evaluated using positive mode mass spectrometry and CID , which generates fragment ion masses from cleavage of the glycosidic linkage . We found furanose ring acylation in almost all tested species in Solanaceae ( Figure 1—figure supplement 4A–E ) ; however , the lengths of acyl chains on the ring varied . All tri-acylsucroses analyzed using positive mode CID data bore all acyl chains on one ring , likely the pyranose ring — as evidenced by neutral loss of 197 Da ( hexose plus NH3 ) from the [M+NH4]+ ion — unlike S . lycopersicum , which bears one acyl chain on the furanose ring . However , the substituents varied among tetra- and penta-acylsugars of different species , with some ( e . g . : N . alata ) , showing up to four chains on the same ring ( Figure 1—figure supplement 4A ) , and others ( e . g . : H . niger , Salpiglossis ) revealing spectra consistent with acylation on both pyranose and furanose rings ( Figure 1—figure supplement 4B , C ) . These findings illustrated that species across the family show very diverse acylsugar profiles , prompting us to quantify the overall surface metabolite diversity using Shannon Entropy ( see Materials and methods ) . We found that acylsugar producing species had significantly higher entropies compared to non-producing species [Kolmogorov-Smirnov ( KS ) test p=2e-18] , indicative of the qualitative and quantitative variation in acylsugar profiles in the producers ( Figure 1F ) . We also found that different tissues sampled from the same species shared a substantial proportion of peaks , however , peaks were generally unique between species — a finding supported by manual annotation of acylsugars ( Figure 1G; Figure 1—figure supplement 2 ) . Acylsugar producing species , although sharing a smaller proportion of peaks , shared relatively more peaks with other acylsugar producing species than with non-producers ( Figure 1—source data 5 ) . Inferences derived from additional parameters , namely peak specificity and tissue specialization ( Figure 1—figure supplement 5 ) , were consistent with these observations suggesting a high degree of specificity of surface metabolites in Solanaceae species . These results demonstrate the substantial acylsugar diversity across the Solanaceae family , most of which is unique to any given species . To identify the enzymes that contribute to this diversity , we performed RNA-seq in four phylogenetically-spaced species with interesting acylsugar profiles , namely S . nigrum , S . quitoense , H . niger and Salpiglossis . Our previous studies in cultivated tomato ( Schilmiller et al . , 2012; 2015; Ning et al . , 2015; Fan et al . , 2016a ) demonstrated that identifying genes with expression enriched in stem/petiole trichomes compared to shaved stem/petiole without trichomes is a productive way to find acylsugar biosynthetic enzymes . We sampled polyA RNA from these tissues from four species and performed de novo read assembly ( Table 1 ) . These assemblies were used to find transcripts preferentially expressed in the trichomes ( referred to as ‘trichome-high transcripts’ ) and to develop hypotheses regarding their functions based on homology . Overall , 1888–3547 trichome-high transcripts ( 22–37% of all differentially expressed transcripts ) were identified across all four species ( False Discovery Rate adjusted p<0 . 05 , fold change ≥2 ) ( Table 1 ) . These transcripts were subjected to a detailed analysis including coding sequence prediction , binning into 25 , 838 orthologous groups , assignment of putative functions based on tomato gene annotation and Gene Ontology enrichment analysis ( see Materials and methods , Table 1 ) . Analysis of the enriched categories ( Fisher exact test corrected p<0 . 05 ) revealed that only 20 of 70 well-supported categories ( ≥10 transcripts ) were enriched in at least three species ( Supplementary file 1 ) , suggesting existence of diverse transcriptional programs in the trichomes at the time of their sampling . Almost all enriched categories were related to metabolism , protein modification or transport , with metabolism-related categories being dominant ( Supplementary file 1 ) . These results support the notion of trichomes as ‘chemical factories’ ( Schilmiller et al . , 2008 ) and point to the metabolic diversity that might exist in trichomes across the Solanaceae . A major goal of this study was to define the organization of the acylsugar biosynthetic pathway at the origin of the Solanaceae , prompting us to focus on Salpiglossis , whose phylogenetic position is of special interest in inferring the ancestral state of the biosynthetic pathway . We first validated the plant under study as Salpiglossis using a phylogeny based on ndhF and trnLF sequences ( Figure 2—figure supplement 1A , B ) . A previously published maximum likelihood tree of 1075 Solanaceae species suggested Salpiglossis as an extant species of the earliest diverging lineage in Solanaceae ( Särkinen et al . , 2013 ) . However , some tree reconstruction approaches show Duckeodendron and Schwenckia as emerging from the earliest diverging lineages , and Salpiglossis and Petunioideae closely related to each other ( Olmstead et al . , 2008; Särkinen et al . , 2013 ) . Thus , our further interpretations are restricted to the last common ancestor of Salpiglossis-Petunia-Tomato ( hereafter referred to as the Last Common Ancestor [LCA] ) that existed ~22–28 mya . To infer the ancestral state of the acylsugar biosynthetic pathway in the LCA , we characterized the pathway in Salpiglossis using in vitro and in planta approaches . The acylsugar structural diversity and phylogenetic position of Salpiglossis led us to characterize the biosynthetic pathway of this species . NMR analysis of Salpiglossis acylsugars revealed acylation at the R2 , R3 , R4 positions on the pyranose ring and R1′ , R3′ , R6′ positions on the furanose ring ( Figure 2A; Figure 2—source data 1 ) . The acylation positions are reminiscent of Petunia axillaris ( Pa ) acylsucroses where PaASAT1 , PaASAT2 , PaASAT3 and PaASAT4 acylate with aliphatic precursors at R2 , R4 , R3 and R6 on the six-carbon pyranose ring , respectively ( Nadakuduti et al . , 2017 ) . Thus , we tested the hypothesis that PaASAT1 , 2 , 3 orthologs in Salpiglossis function as SsASAT1 , 2 , 3 respectively . Thirteen Salpiglossis trichome-high BAHD family members were found ( Figure 2—figure supplement 1C ) , with nine expressed in Escherichia coli ( Figure 2—figure supplement 2; Supplementary file 2 ) . Activities of the purified enzymes were tested using sucrose or partially acylated sucroses ( Fan et al . , 2016a; Fan et al . , 2016b ) as acceptor substrates . Donor C2 , aiC5 and aiC6 acyl CoA substrates were tested based on the common occurrence of these ester groups in a set of 16 Salpiglossis acylsucroses purified for NMR . Representative NMR structures that illustrate the SsASAT positional selectivity described in the results below are shown in Figure 2A . Four of the tested candidates catalyzed ASAT reactions ( Figure 2B–E ) . In the following description , we name the enzymes based on their order of acylation in the Salpiglossis acylsugar biosynthetic pathway . A description is provided in Figure 2—figure supplement 4 to assist in understanding the chromatograms . Salpiglossis sinuata ASAT1 ( SsASAT1 ) generated mono-acylsugars from sucrose using multiple acyl CoAs ( Figure 2B , Figure 2—figure supplements 4 and 5 ) , similar to the donor substrate diversity of SlASAT1 ( Fan et al . , 2016a ) ( Figure 2—figure supplement 4 ) . We infer that SsASAT1 primarily acylates the R2 position on the pyranose ring . This is based on ( a ) the S1:6 ( 6 ) negative mode CID fragmentation patterns ( Figure 2—figure supplement 5A ) and ( b ) comparisons of chromatographic migration of the mono-acylsucroses produced by SsASAT1 , PaASAT1 , which acylates sucrose at R2 and matches the major SsASAT1 product , and SlASAT1 , which acylates sucrose at R4 ( Figure 2—figure supplement 5A , B ) ( Nadakuduti et al . , 2017 ) . The Salpiglossis SsASAT1 activity is similar to P . axillaris PaASAT1 and unlike S . lycopersicum SlASAT1 . SsASAT2 was identified by testing the ability of each of the other eight cloned enzymes to acylate the mono-acylated product of SsASAT1 ( S1:5 or S1:6 ) as acyl acceptor and using aiC5 CoA as acyl donor . SsASAT2 catalyzed the formation of di-acylated sugars , which co-eluted with the PaASAT2 product but not with the SlASAT2 product ( Figure 2B; Figure 2—figure supplement 5C , D ) . Positive mode fragmentation suggested that the acylation occurs on the same ring as SsASAT1 acylation ( Figure 2—figure supplement 5C ) . This enzyme failed to acylate sucrose ( Figure 2—figure supplement 6 ) , supporting its assignment as SsASAT2 . SsASAT3 ( Figure 2B , red chromatogram ) added aiC6 to the pyranose ring of the di-acylated sugar acceptor . This aiC6 acylation reaction is in concordance with the observed in planta acylation pattern at the R3 position — aiC6 is present at this position in the majority of S . sinuata acylsugars . Surprisingly , the enzyme did not use aiC5 CoA , despite the identification of S3:15 ( 5 , 5 , 5 ) and likely its acylated derivates [S4:20 ( 5 , 5 , 5 , 5 ) , S4:17 ( 2 , 5 , 5 , 5 ) , S5:22 ( 2 , 5 , 5 , 5 , 5 ) and S5:19 ( 2 , 2 , 5 , 5 , 5 ) ] from S . sinuata extracts . SsASAT3-dependent R3 position acylation was further confirmed by testing the tri-acylated product with PaASAT4 , which acylates at the R6 position of the pyranose ring ( Figure 2—figure supplement 7 ) ( Nadakuduti et al . , 2017 ) . The successful R6 acylation by PaASAT4 is consistent with the hypothesis that SsASAT3 acylates the R3 position . Taken together , these results suggest that the first three enzymes generate acylsugars with aiC5/aiC6 at the R2 position ( SsASAT1 ) , aiC6 at the R3 position ( SsASAT3 ) and aiC5/aiC6 at the R4 position ( SsASAT2 ) . We could not identify the SsASAT4 enzyme ( s ) that performs aiC5 and C2 acylations on the R1′ and R3′ positions of tri-acylsucroses , respectively . However , we identified another enzyme , which we designate SsASAT5 , that showed three activities acetylating tri- , tetra- and penta-acylsucroses ( see Materials and methods ) . SsASAT5 can perform furanose ring acetylation on tri- ( Figure 2—figure supplement 8A , B ) and tetra-acylsucroses ( Figure 2C , D ) , and both furanose as well as pyranose ring acetylation on penta-acylsucroses ( Figure 2—figure supplement 8C–F ) . All of the products produced by SsASAT5 in vitro co-migrate with acylsugars found in plant extracts , suggesting the SsASAT5 acceptor promiscuity also occurs in planta . Our observation that SsASAT5 can perform pyranose ring acetylation — albeit weakly ( Figure 2—figure supplement 8D , F ) — is at odds with NMR-characterized structures of a set of 16 purified Salpiglossis acylsugars , which show all acetyl groups on the furanose ring . However , a previous study described one pyranose R6-acylated penta-acyl sugar S5:22 ( 2 , 2 , 6 , 6 , 6 ) in Salpiglossis ( Castillo et al . , 1989 ) , suggesting the presence of accession-specific variation in enzyme function . Despite showing SsASAT4- , SsASAT5- and SsASAT6-like activities , we designate this enzyme SsASAT5 because its products have both acylation patterns and co-migration characteristics consistent with the most abundant penta-acylsugars from the plant ( Figure 2C , D ) . Overall , in vitro analysis revealed four enzymes that could catalyze ASAT reactions and produce compounds also detected in plant extracts ( Figure 2E ) . We further verified that these enzymes are involved in acylsugar biosynthesis by testing the effects of perturbing their transcript levels using Virus Induced Gene Silencing ( VIGS ) . To test the role of the in vitro identified ASATs in planta , we adapted a previously described tobacco rattle virus-based VIGS procedure ( Dong et al . , 2007; Velásquez et al . , 2009 ) for Salpiglossis . We designed ~300 bp long gene-specific silencing constructs for transient silencing of SsASAT1 , SsASAT2 , SsASAT3 and SsASAT5 ( Supplementary files 2 , 3 ) , choosing regions predicted to have a low chance of reducing expression of non-target genes ( see Materials and methods ) . The Salpiglossis ortholog of the tomato phytoene desaturase ( PDS ) carotenoid biosynthetic enzyme was used as positive control ( Figure 3A ) , with transcript level decreases confirmed for each candidate using qRT-PCR in one of the VIGS replicates ( Figure 3B ) . As no standard growth or VIGS protocol was available for Salpiglossis , we tested a variety of conditions for agro-infiltration and plant growth , and generated at least two biological replicate experiments for each construct , run at different times ( Supplementary file 4 ) . ASAT knockdown phenotypes were consistent , regardless of variation in environmental conditions . SsASAT1 VIGS revealed statistically significant reductions in acylsugar levels in at least one construct across two experimental replicates ( Kolmogorov-Smirnov [KS] test , p-value=0 . 05 ) ( Figure 3C , D; Figure 3—figure supplement 1A ) , consistent with its predicted role in catalyzing the first step in acylsugar biosynthesis . SsASAT2 expression reduction also produced plants with an overall decrease in acylsugar levels ( Figure 3E , F; Figure 3—figure supplement 1B ) ( KS test , p=0 . 03 ) . However , these plants also showed some additional unexpected acylsugar phenotypes , namely increases in lower molecular weight acylsugars ( m/z ratio: 651–700 , 701–750; KS test p<0 . 05 ) , decreases in levels of higher molecular weight products ( m/z ratio: 751–800 , 801–850 ) ( Figure 3E , F ) as well as changes in levels of some individual acylsugars as described in Figure 3—figure supplement 2 . These results provide in planta support for the involvement of SsASAT2 in acylsugar biosynthesis , and suggest the possibility of discovering additional enzymatic activities in the future . Silencing the SsASAT3 transcript also led to an unexpected result - significantly higher accumulation of the normally very low abundance tri-acylsugars [S3:13 ( 2 , 5 , 6 ) ; S3:14 ( 2 , 6 , 6 ) ; S3:15 ( 5 , 5 , 5 ) ; S3:16 ( 5 , 5 , 6 ) ; S3:18 ( 6 , 6 , 6 ) ] and their acetylated tetra-acylsugar derivatives ( KS test p<0 . 05 ) , compared to empty vector control infiltrated plants ( Figure 4A , B; Figure 4—figure supplement 1 ) . The tetra-acylsugars contained C2 or C5 acylation on the furanose ring ( Figure 4—figure supplement 2 ) . These observations are consistent with the hypothesis that di-acylated sugars accumulate upon SsASAT3 knockdown and then serve as substrate for one or more other enzyme ( Figure 4—figure supplement 3 ) . Our working hypothesis is that this inferred activity is the as-yet-unidentified SsASAT4 activity; this is based on comparisons of in vitro enzyme assay products and in vivo purified acylsugars from Salpiglossis plants ( Figure 2A ) . We propose that the hypothesized SsASAT4 may promiscuously acylate di-acylated sugars in addition to performing C2/C5 additions on tri-acylsugars . SsASAT5 , based on in vitro analysis , was proposed to catalyze acetylation of tetra- to penta-acylsugars . As expected , its knockdown led to accumulation of tri- and tetra-acylsugars ( Figure 4C , D; Figure 4—figure supplement 3B ) . The accumulating tetra-acylsugars were C2/C5 furanose ring-acylated derivatives of the tri-acylsugars , suggesting presence of a functional SsASAT4 enzyme and further validating the annotation of the knocked down enzyme as SsASAT5 ( Figure 4—figure supplement 3B ) . Thus , in summary , we identified four ASAT enzymes and validated their impact on acylsugar biosynthesis in Salpiglossis trichomes using VIGS . Taken together , the Salpiglossis metabolites produced in vivo , combined with in vitro and RNAi results , lead to the model of the Salpiglossis acylsugar biosynthetic network shown in Figure 5 . SsASAT1 – the first enzyme in the network – adds aiC5 or aiC6 to the sucrose R2 position . SsASAT2 then converts this mono-acylated sucrose to a di-acylated product , via addition of aiC5 or aiC6 at the R4 position . Four possible products are thus generated by the first two enzymes alone . Next , the SsASAT3 activity adds aiC6 at the R3 position , followed by one or more uncharacterized enzyme ( s ) that adds either C2 or aiC5 at the furanose ring R1′ or R3′ positions , respectively . SsASAT5 next performs acetylation at the R6′ position to produce penta- acylsugars , which can then be further converted by an uncharacterized SsASAT6 to hexa-acylsugars via C2 addition at the R3′ position . Our results are consistent with the existence of at least two additional activities – SsASAT4 and SsASAT6 . These enzymes may be included in the five BAHD family candidates highly expressed in both the trichome and stem ( average number of reads >500 ) , and thus not selected for our study because they did not meet the differential expression criterion . Also , the fact that there are >300 detectable acylsugar-like peaks in the Salpiglossis trichome extracts suggests the existence of additional ASAT activities , either promiscuous activities of characterized ASATs or of other uncharacterized enzymes . Nonetheless , identification of the four primary ASAT activities can help us to investigate the origins and evolution of the acylsugar biosynthetic pathway over time . We used our analysis of SsASAT1 , SsASAT2 , SsASAT3 and SsASAT5 activities , with information about ASATs in Petunia and tomato species ( Schilmiller et al . , 2012; 2015; Fan et al . , 2016a; Nadakuduti et al . , 2017 ) , to infer the origins of the acylsugar biosynthetic pathway . Based on BLAST searches across multiple plant genomes , ASAT-like sequences are very narrowly distributed in the plant phylogeny ( Figure 6—figure supplement 1 ) . This led us to restrict our BLAST searches , which used SlASATs and SsASATs as query sequences , to species in the orders Solanales , Lamiales , Boraginales and Gentianales , which are all in the Lamiidae clade ( Refulio-Rodriguez and Olmstead , 2014 ) . Phylogenetic reconstruction was performed with the protein sequences of the most informative hits obtained in these searches to obtain a ‘gene tree’ . Reconciliation of this gene tree with the phylogenetic relationships between the sampled species ( Figure 6B ) allowed inference of the acylsugar biosynthetic pathway before the emergence of the Solanaceae ( Figure 6C; Figure 6—figure supplements 2A–C and 3A–C ) . Three major subclades in the gene tree – highlighted in blue , red and pink – are relevant to understanding the origins of the ASATs ( Figure 6A ) . A majority of characterized ASATs ( blue squares in the blue subclade , Figure 6A ) are clustered with Capsicum PUN1 — an enzyme involved in biosynthesis of the alkaloid capsaicin — in a monophyletic group with high bootstrap support ( Group #2 , red and blue subclades Figure 6A ) . Two of the most closely related non-Solanales enzymes in the tree — Catharanthus roseus minovincinine-19-O-acetyltransferase ( MAT ) and deacetylvindoline-4-O-acetyltransferase ( DAT ) — are also involved in alkaloid biosynthesis ( Magnotta et al . , 2007 ) . This suggests that the blue ASAT subclade emerged from an alkaloid biosynthetic enzyme ancestor . A second insight from the gene tree involves Salpiglossis SsASAT5 and tomato SlASAT4 , which reside outside of the blue subclade . Both enzymes catalyze C2 addition on acylated sugar substrates in downstream reactions of their respective networks . Multiple enzymes in this region of the phylogenetic tree ( Figure 2—figure supplement 1C; light blue clade ) are involved in O-acetylation of diverse substrates for example indole alkaloid 16-epivellosimine ( Bayer et al . , 2004 ) , the phenylpropanoid benzyl alcohol ( D'Auria et al . , 2002 ) and the terpene geraniol ( Shalit et al . , 2003 ) . This observation is consistent with the hypothesis that O-acetylation activity was present in ancestral enzymes within this region of the phylogenetic tree . Gene tree reconciliation with known relationships between plant families and orders ( Figure 6B ) was used to infer acylsugar pathway evolution in the context of plant evolution . We used historical dates as described by Särkinen and co-workers ( Särkinen et al . , 2013 ) in our interpretations , as opposed to a recent study that described a much earlier origination time for the Solanaceae ( Wilf et al . , 2017 ) . Based on known relationships , Convolvulaceae is the closest sister family to Solanaceae; however , we found no putative ASAT orthologs in any searched Convolvulaceae species . The closest Convolvulaceae homologs were found in the Ipomoea trifida genome ( Hirakawa et al . , 2015 ) in the red subclade . This suggests that the blue and red subclades arose via a duplication event before the Solanaceae-Convolvulaceae split , estimated to be ~50–65 mya ( Särkinen et al . , 2013 ) . Thus , this duplication event predates the whole genome triplication ( WGT ) event ancestral to all Solanaceae that occurred after the Solanaceae-Convolvulaceae divergence ( Bombarely et al . , 2016 ) . This inference is also consistent with our findings based on synonymous substitution rate ( dS ) distributions of homologs between cultivated tomato and Petunia . Specifically , we identified all orthologs and paralogs in the two species and obtained a distribution of all dS values ( black histogram/red curve , Figure 6D ) , as in a previously published study ( Bombarely et al . , 2016 ) . We then specifically identified the syntenic paralogs in tomato — likely derived from the polyploidization events — and overlaid their dS values on the previous distribution ( yellow curve , Figure 6D ) . This analysis differentiated the paralogs derived via ancestral WGD events from those derived via the Solanaceae-specific WGT event ( Bombarely et al . , 2016 ) . Two BAHDs — SlASAT1 ( Solyc12g006330 ) and Solyc07g043670 — in the blue subclade ( bold , Figure 6A ) were found to be whole genome duplicates of each other . These genes lie in a syntenic block that spans multiple gene pairs on chromosomes 12 and 7 , respectively , a majority of which are derived from the Solanaceae-specific WGT event ( blue curve , Figure 6D; Figure 6—source data 1 ) . These observations were useful in annotating the duplication node separating SlASAT1 and Solyc07g043670 in the BAHD gene tree as the Solanaceae-specific WGD node ( #5 , Figure 6A ) . We can then also infer that the duplications giving rise to the different ASAT1 , 2 , 3 clades occurred prior to this Solanaceae-specific WGT event . The lack of ASAT orthologs in Convolvulaceae could indicate that Convolvulaceae species with an orthologous acylsugar biosynthetic pathway were not sampled in our analysis . Alternatively , the absence of orthologs of the blue subclade enzymes in the Convolvulaceae , coupled with the lack of acylsugars in the sampled Convolvulaceae species ( Figure 1B ) , leads us to hypothesize that ASAT orthologs were lost early in Convolvulaceae evolution . The known phylogenetic relationships between species can help assign a maximum age for the duplication event leading to the blue and red subclades . Database searches identified homologs from species in the Boraginales ( Ehretia , Heliotropium ) and Gentianales ( Coffea , Catharanthus ) orders that clustered outside Group #2 , in a monophyletic group with high bootstrap support ( Group #1 , pink branches , Figure 6A ) . Boraginales is sister to the Lamiales order ( Figure 6B ) ( Refulio-Rodriguez and Olmstead , 2014 ) , and although Boraginales putative orthologs were identified through BLAST searches , no orthologs from Lamiales species were detected . The closest Lamiales ( Sesamum , Lavandula , Mimulus ) homologs were more closely related to SsASAT5 than to the ASATs in the blue subclade ( purple squares , Figure 6A ) . The most parsimonious explanation for these observations is that the duplication event that gave rise to the blue and red subclades occurred before the Solanaceae-Convolvulaceae split 50–65 mya but after the Solanales-Boraginales/Lamiales orders diverged 75–100 mya ( Hedges et al . , 2006 ) ( Figure 6B , C ) . Consistent with this model , we did not find any acylsugar-like peaks in leaf surface extracts of eleven species in these orders ( Figure 6—figure supplement 4 ) . These findings , which provide insights into the origin of acylsugar biosynthesis , can be interpreted under the three-step model of evolution of biological innovation , involving potentiation , actualization and refinement ( Blount et al . , 2012 ) ( Figure 8 ) . We propose that the ASAT enzymes catalyzing sucrose acylation first emerged ( or actualized ) between 30–80 mya . Events prior to the emergence of this activity — including the existence of alkaloid biosynthetic BAHDs and the duplication event 60–80 mya that gave rise to the blue and the red subclades — potentiated the emergence of the ASAT activities . Presumably , acylsugars provided a fitness advantage to the ancestral plants , leading to the refinement of ASAT activities over the next few million years . For acylsugar production to emerge , these steps in enzyme evolution would have been complemented by other innovations in ASAT transcriptional regulation leading to gland cell expression as well as production of precursors ( i . e . , acyl CoA donors and sucrose ) in Type I/IV trichomes . Overall , these observations support a view of the origins of the ASAT1 , 2 , 3 blue subclade from an alkaloid biosynthetic ancestor via a single duplication event >50 mya , prior to the establishment of the Solanaceae . This duplicate underwent further rounds of duplication prior to and after the Solanaceae-specific WGT event to generate the multiple ASATs found in the blue subclade . Thus , a logical next question is ‘what was the structure of the acylsugar biosynthetic network early in the Solanaceae family evolution ? ’ To address this , we focused on ASAT evolution across the Solanaceae family . ASAT enzymes in the blue subclade represent the first three steps in the acylsugar biosynthetic pathway . The likely status of these three steps in the Salpiglossis-Petunia-Tomato last common ancestor ( LCA ) was investigated using the gene tree displayed in Figure 7A . Mapping the pathway enzymes on the gene tree suggests that the LCA likely had at least three enzymatic activities , which we refer to as ancestral ASAT1 ( aASAT1 ) , ASAT2 ( aASAT2 ) and ASAT3 ( aASAT3 ) ; these are shown in Figure 7A as red , dark blue and yellow squares , respectively . We further traced the evolution of the aASAT1 , 2 , 3 orthologs in the Solanaceae using existing functional data and BLAST-based searches . These results reveal that the aASAT1 ortholog ( red squares , Figure 7A ) was present until the Capsicum-Solanum split ~17 mya ( Särkinen et al . , 2013 ) and was lost in the lineage leading to Solanum . This loss is evident both in similarity searches and in comparisons of syntenic regions between genomes of Petunia , Capsicum and tomato ( Figure 7B ) . On the other hand , the aASAT2 ( dark blue squares , Figure 7A ) and aASAT3 orthologs ( yellow squares , Figure 7A ) have been present in the Solanaceae species genomes at least since the last common ancestor of Salpiglossis-Petunia-Tomato~25 mya , and perhaps even in the last common ancestor of the Solanaceae family . One inference from this analysis is that aASAT2 orthologs switched their activity from ASAT2-like acylation of mono-acylsucroses in Salpiglossis/Petunia to ASAT1-like acylation of unsubstituted sucrose in cultivated tomato ( Figure 7A ) . Interestingly , despite the switch , both aASAT2 and aASAT3 orthologs in tomato continue to acylate the same pyranose-ring R4 and R3 positions , respectively ( Figure 7C; Figure 8 ) . In addition , cultivated tomato has two ‘new’ enzymes — SlASAT3 and SlASAT4 — which were not described in Petunia or Salpiglossis acylsugar biosynthesis . The functional transitions of aASAT2 and aASAT3 could have occurred via ( i ) functional divergence between orthologs or ( ii ) duplication , neo-functionalization and loss of the ancestral enzyme . Counter to the second hypothesis , we found no evidence of aASAT2 ortholog duplication in the genomic datasets; however , we cannot exclude the possibility of recent polyploidy or tandem duplication in extant species producing duplicate genes with divergent functions . We explored the more parsimonious hypothesis that functional divergence between orthologs led to the aASAT2 functional switch . Functional divergence between aASAT2 orthologs may have occurred by one of two mechanisms . One , it is possible that aASAT2 orthologs had some sucrose acylation activity prior to aASAT1 loss . Alternatively , sucrose acylation activity arose completely anew after the loss of aASAT1 . We sought evidence for acceptor substrate promiscuity in extant species by characterizing the activity of additional orthologous aASAT2 enzymes from H . niger and S . nigrum ( Figure 7C ) using aiC6 and nC12 as acyl CoA donors . HnASAT2 — like SsASAT2 — only performed the ASAT2 reaction ( acylation of mono-acylsucrose ) , without any evidence for sucrose acylation under standard testing conditions ( Figure 7C , Figure 7—figure supplement 1A–D ) . However , both SnASAT1 and SlASAT1 could produce S1:6 ( 6 ) and S1:12 ( 12 ) from sucrose . These findings suggest that until the Atropina-Solaneae common ancestor , the aASAT2 ortholog still primarily conducted the ASAT2 activity ( Figure 7C ) . However , at some point , the aASAT2 ortholog moved in reaction space towards being ASAT1 , and this activity shift was likely complete by the time the S . nigrum-S . lycopersicum lineage diverged ( Figure 7C ) , given that aASAT2 orthologs in both species utilize sucrose . These results point to a second round of major innovation in the acylsugar biosynthetic pathway . This round involved the following three steps: ( i ) Potentiation: the similarity in activities of aASAT1 , 2 and 3 , allowing changes in substrate preferences with relatively few amino acid changes . In fact , aASAT1 and aASAT2 appear to be clustered together in the gene tree , perhaps making the aASAT2 activity switch more likely . ( ii ) Actualization: this refers to the first instance of the emergence of the ASAT1 activity in aASAT2 orthologs . It is unclear whether this switch occurred prior to aASAT1 loss or after . ( iii ) Refinement: after the actualization of the ASAT1 activity , this activity was likely refined over time to the SlASAT1 activity we see today . To understand the specific order of the events in aASAT2 evolution in the Solanaceae , ASAT1 and ASAT2 enzymes from additional species between H . niger and Solanum will need to be tested . We performed experimental and computational analysis of BAHD enzymes from several lineages spanning ~100 million years , and characterized the emergence and evolution of the acylsugar metabolic network . We identified four biosynthetic enzymes in Salpiglossis sinuata , the extant species of an early emerging Solanaceae lineage , characterized in vitro activities , validated in planta roles of these ASATs and studied their emergence over 100 million years of plant evolution . These results demonstrate the value of leveraging genomics , phylogenetics , analytical chemistry and enzymology with a mix of model and non-model organisms to understand the evolution of biological complexity . We uncovered a large diversity of acylsugars across the family , all of which is based on two simple types of components - a sugar core ( mostly sucrose ) and acyl chains ( C2-C12 ) . Previous studies indicate that acylation is possible on all eight hydroxyls on the sucrose core ( Ghosh et al . , 2014; Schilmiller et al . , 2015; Fan et al . , 2016a ) . A simple calculation ( see Materials and methods ) considering only 12 aliphatic CoA donors typically found in Solanaceae acylsugars suggests >6000 theoretical possible structures for tetra-acylsugars alone . This estimate does not include estimates of tri- , penta- , hexa-acylsugars nor does it consider non-aliphatic CoAs such as malonyl CoA , esters of other sugars such as glucose or positional isomers . Although the theoretical possibilities are restricted by availability of CoAs in trichome tip cells and existing ASAT activities , we still observe hundreds of acylsugars across the Solanaceae , with >300 detectable acylsugar-like chromatographic peaks in single plants of Salpiglossis and N . alata , ( Figure 1F ) including very low abundance hepta-acylsugars and acylsugars containing phenylacetyl and tigloyl chains . The acylsugar composition is typically similar between trichomes on different plant tissues of the same individual ( Figure 1—figure supplement 2 ) . However , it varies at multiple taxonomic levels - that is between populations of a single species ( Kim et al . , 2012 ) , between closely related species ( Schilmiller et al . , 2015; Fan et al . , 2016a ) and between species across the Solanaceae ( this study ) . This structural diversity is reminiscent of other diverse lineage-specific metabolite classes such as glucosinolates in Brassicales ( Olsen et al . , 2016 ) , betalains in Caryophyllales ( Khan and Giridhar , 2015 ) , acridone alkaloids in Rutaceae ( Roberts et al . , 2010 ) , resin glycosides in Convolvulaceae ( Pereda-Miranda et al . , 2010 ) and pyrrolizidine alkaloids in Boraginaceae ( El-Shazly and Wink , 2014 ) . Does structural diversity within the same metabolite class in a single individual provide a fitness advantage ? The neutral theory of evolution serves as a null hypothesis , predicting that diversity does not confer a fitness advantage and is merely a result of drift . Indeed , the structural diversity in a single individual could be a reflection of promiscuous enzyme activities , given large enzyme families such as BAHDs , cytochrome P450s , glycosyltransferases play a role in the biosynthesis of several specialized metabolite classes . It is also possible that only a few specific metabolites belonging to the metabolite class — rather than the entire repertoire of the class — are consequential in imparting fitness advantage to an individual . Alternatively , such structural diversity may generate a meta-property , such as stickiness — in the case of acylsugars — or synergistic antimicrobial or insect protective action that may provide a fitness advantage . A related possibility is that metabolite diversity in a single individual and , subsequently , between populations of the same species , provides the standing variation that may be acted upon by natural selection in the future , as the environment of the species changes . This was postulated to be the case for the existence of aliphatic glucosinolate diversity in Arabidopsis ( Kerwin et al . , 2015 ) . A previous study also showed that accession-level glucosinolate diversity is maintained in European populations of Arabidopsis by the differences in relative abundance of two aphid species in different parts of the range ( Züst et al . , 2012 ) . In the case of acylsugar diversity , we have no evidence for one or more of these hypotheses , because most studies on the importance of acylsugar structural diversity were performed in lab settings . Thus , a significant opportunity exists to characterize the importance of the structural diversity in acylsugar and other plant specialized metabolite classes in ecologically relevant settings . Our results revealed the involvement of gene duplication and divergence , enzyme promiscuity , gene loss and functional divergence between orthologs in the evolution of the relatively compact acylsugar biosynthetic network . The process of gene loss in ASATs and other BAHDs may also have been influenced by the fractionation process occurring in the Solanaceae genomes after the Solanaceae-specific polyploidization event . Our results also point to factors such as acyl CoA pools , trichome developmental programs and genotype by environment interactions that may affect the evolution of metabolic pathways . Previous results from closely related Solanum species also revealed regulatory divergence of ASATs ( Kim et al . , 2012 ) and neo-functionalization of duplicates ( Ning et al . , 2015 ) in influencing the phenotypes produced by this pathway . These findings highlight the striking plasticity of the acylsugar biosynthetic network and illustrate the complexity that underlies the evolution of novel chemical phenotypes . Our results also reveal the important role played by tandem and whole genome duplications as well as gene loss in evolution of specialized metabolic pathways . The discovery of multiple routes to biochemical innovation in our study was facilitated by the use of a highly integrative approach . Recent studies have utilized similar integrative approaches in deciphering the molecular evolution of plant-butterfly interactions in the Brassicales ( Edger et al . , 2015 ) and evolution of herbivore-induced defense signaling in Nicotiana ( Zhou et al . , 2016 ) . Deployment of such strategies on a large scale is increasingly possible due to improvements in small molecule mass spectrometry and nucleic acid sequencing , establishment of transient knockdown protocols such as VIGS that do not require stable plant transformation , establishment of new gene editing techniques , broadly representative genome and transcriptome sequence databases and significantly better data storage and analysis abilities . Approaches such as VIGS are even more important for deciphering metabolic pathways in non-model species because of the lack of stable transformation protocols and involvement of enzyme families in specialized metabolism , which frequently requires in planta validation . Overall , such broad studies can provide a multi-dimensional view of the evolution of metabolite diversity because of the ability to integrate insights derived from different approaches into a single framework . In addition , phylogeny-guided studies that integrate comparative metabolomics and transcriptomics , such as co-expression analysis ( Zhou et al . , 2016; Wisecaver et al . , 2017 ) , can also be used for pathway discovery in non-model organisms , thus creating a robust platform for studying different aspects of biological evolution , for example co-evolutionary studies ( Edger et al . , 2015 ) . Discovery of novel enzymatic activities , coupled with recent advances in genome editing technologies , can also benefit synthetic biology of natural products , for example as described by Ignea and co-workers ( Ignea et al . , 2015 ) . In this study , we primarily focused on Salpiglossis because of a unique acylsugar profile and its phylogenetic position . S . lycopersicum is the flagship species of the Solanaceae family , and we were able to apply the insights derived with this crop as a foundation to discover novel enzymes in related species . Our investigations in Salpiglossis were also aided by the availability of ASATs and NMR structures of their products in Petunia ( Liu et al . , 2017; Nadakuduti et al . , 2017 ) , which is more closely related to Salpiglossis than tomato . ‘Anchor species’ such as Petunia , which are phylogenetically distant from flagship/model species , can enable study of a different region of the phylogenetic tree of the clade of interest . Development of limited genomic and functional resources in such anchor species , coupled with integrative , comparative approaches can offer more efficient routes for the exploration of biochemical complexity in the ~300 , 000 plant species estimated to exist on our planet ( Mora et al . , 2011 ) . Acylsugar extractions were carried out from plants at the New York Botanical Gardens and from other sources ( Figure 1—source data 1 ) . The plants sampled were at different developmental stages and were growing in different environments . The extractions were carried out using acetonitrile:isopropanol:water in a 3:3:2 proportion similar to previous descriptions ( Schilmiller et al . , 2010; Ghosh et al . , 2014; Fan et al . , 2016a ) with the exception of gently shaking the tubes by hand for 1–2 min . All extracts were analyzed on LC/MS ( Waters Corporation , USA ) using 7 min , 22 min or 110 min LC gradients on Supelco Ascentis Express C18 ( Sigma Aldrich , USA ) or Waters BEH amide ( Waters Corporation , USA ) columns ( Figure 1—figure supplement 6 ) , as described previously ( Schilmiller et al . , 2010; Ghosh et al . , 2014; Fan et al . , 2016a ) . While the 110 min method was used to minimize chromatographic overlap in support of metabolite annotation in samples with diverse mixtures of acylsugars , targeted extracted ion chromatogram peak area quantification was performed using the 22 min method data . The QuanLynx function in MassLynx v4 . 1 ( Waters Corporation , USA ) was used to integrate extracted ion chromatograms for manually selected acylsugar and internal standard peaks . Variable retention time and chromatogram mass windows were used , depending on the experiment and profile complexity . Peak areas were normalized to the internal standard peak area and expressed as a proportion of mg of dry weight . The concept of Shannon Entropy , originally developed in the field of information theory to quantify the amount of uncertainty or information content of a message ( Shannon , 1948 ) , is used in ecology to quantify species diversity ( Peet , 1974 ) . More recently , this approach was used to quantify transcriptomic and metabolic diversity and specialization ( Martínez and Reyes-Valdés , 2008; Li et al . , 2016 ) . To calculate Shannon Entropy , we explored three different software packages for processing the RAW files from the Waters LCT Premier Mass Spectrometer , namely ( i ) the MarkerLynx function in MassLynx software v4 . 1 ( Waters Corporation , USA ) , ( ii ) Progenesis QI suite ( Nonlinear Dynamics , USA ) and ( iii ) mzMine 2 ( Pluskal et al . , 2010 ) . We found mzMine 2 most appropriate for our use , because it had several options for customization and processing of background data . The batch parameters used for processing 88 RAW files are provided in Figure 1—source data 4 . Two values – peak height and peak areas – were obtained for all peaks with an intensity >500 in each sample . This threshold was set to eliminate most of the background noise , based on empirical observations of raw chromatograms . We further calculated different measures of diversity using the approach highlighted previously ( Martínez and Reyes-Valdés , 2008; Li et al . , 2016 ) . Specifically , using peak intensity as a measure of count , we calculated Shannon Entropy ( H ) as follows:Hj=−∑i=1mPij . log⁡2 ( Pij ) where Pij indicates the relative frequency of the ith m/z peak ( i=1 , 2 , . . . , m ) in the jth sample ( j=1 , 2 , . . . t ) . The average frequency pi of the ith m/z peak among all samples was calculated as:Pi=1t ∑j=1tPij The specificity of the ith m/z peak ( Si ) ws calculated as:Si=1t ∑j=1t ( Pij/Pi ) . log⁡2 ( Pij/Pi ) The specialization index of each sample δj was measured for each jth sample as the average of the peak specificities using the following formula:δj=∑i=1mPijSi For metabolite purification , aerial tissues of 28 Salpiglossis plants ( aged 10 weeks ) were extracted in 1 . 9 L of acetonitrile:isopropanol ( AcN:IPA , v/v ) for ~10 mins , and ~1 L of the extract was concentrated to dryness on a rotary evaporator and redissolved in 5 mL of AcN:IPA . Repeated injections from this extract were made onto a Thermo Scientific Acclaim 120 C18 HPLC column ( 4 . 6 × 150 mm , 5 µm particle size ) with automated fraction collection . HPLC fractions were concentrated to dryness under vacuum centrifugation , reconstituted in AcN:IPA and combined according to metabolite purity as assessed by LC/MS . Samples were dried under N2 gas and reconstituted in 250 or 300 µL of deuterated NMR solvent CDCl3 ( 99 . 8 atom % D ) and transferred to solvent-matched Shigemi tubes for analysis . 1H , 13C , J-resolved 1H , gCOSY , gHSQC , gHMBC and ROESY NMR experiments were performed at the Max T . Rogers NMR Facility at Michigan State University using a Bruker Avance 900 spectrometer equipped with a TCI triple resonance probe . All spectra were referenced to non-deuterated CDCl3 solvent signals ( δH = 7 . 26 and δC = 77 . 20 ppm ) . Total RNA was extracted from 4 to 5 week old ( S . nigrum , S . quitoense ) or 7–8 week old plants ( H . niger , Salpiglossis ) using Qiagen RNEasy kit ( Qiagen , Valencia , California ) with on-column DNA digestion . In addition to trichome RNA , total RNA was extracted from shaved stems of S . nigrum and Salpiglossis , and shaved petioles of S . quitoense and H . niger . The quality of extracted RNA was determined using Qubit ( Thermo Fisher Scientific , USA ) and Bioanalyzer ( Agilent Technologies , Palo Alto , California ) . Total RNA from all 16 samples ( 4 species x 2 tissues x two biological replicates ) was sequenced using Illumina HiSeq 2500 ( Illumina , USA ) in two lanes ( 8X multiplexing per lane ) . Libraries were prepared using the Illumina TruSeq Stranded mRNA Library preparation kit LT , sequencing carried out using Rapid SBS Reagents in a 2 × 100 bp paired end format , base calling done by Illumina Real Time Analysis ( RTA ) v1 . 18 . 61 and the output of RTA was demultiplexed and converted to FastQ format by Illumina Bcl2fastq v1 . 8 . 4 . The mRNA-seq reads were adapter-clipped and trimmed using Trimmomatic v0 . 32 using the parameters ( LEADING:20 TRAILING:20 SLIDINGWINDOW:4:20 MINLEN:50 ) . The quality-trimmed reads from all datasets of a species were assembled de novo into transcripts using Trinity v . 20140413p1 after read normalization ( max_cov = 50 , KMER_SIZE = 25 , max-pct-stdev=200 , SS_lib_type = RF ) ( Grabherr et al . , 2011 ) . We tested three different kmer values ( k = 25 , 27 , 31 ) and selected the best kmer value for each species based on contig N50 values , BLASTX hits to the S . lycopersicum annotated protein sequences and CEGMA results ( Parra et al . , 2007 ) . A minimum kmer coverage of 2 was used to reduce the probability of erroneous or low abundance kmers being assembled into transcripts . After selecting the best assembly for each species , we obtained a list of transcripts differentially expressed between trichomes and stem/petiole for each species using RSEM/EBSeq ( Li and Dewey , 2011; Leng et al . , 2013 ) at an FDR threshold of p<0 . 05 . The differential expression of five transcripts in S . quitoense and four transcripts in Salpiglossis was confirmed using semi-quantitative RT-PCR , along with the PDS as negative control ( Figure 2—figure supplement 3 ) . The protein sequences corresponding to the longest isoform of all expressed transcripts ( read count >10 in at least one dataset in a given species ) were obtained using TransDecoder ( Haas et al . , 2013 ) and GeneWise v2 . 1 . 20c ( Birney et al . , 2004 ) . Only the protein sequences of transcripts with ≥10 reads as defined by RSEM were used for constructing orthologous groups using OrthoMCL v5 ( Li et al . , 2003 ) . We defined orthologous relationships between S . lycopersicum , S . nigrum , S . quitoense , H . niger , N . benthamiana , Salpiglossis and Coffea canephora ( outgroup ) using an inflation index of 1 . 5 . We transferred the tomato gene ontology assignments to the homologs from other species in the same orthologous group . GO enrichment analysis was performed using a custom R script , and enriched categories were obtained using Fisher Exact Test and correction for multiple testing based on Q-value ( Storey , 2002 ) . Primers to specific regions of the targeted transcript were designed with amplicons between 100–200 bp using Primer3 ( Untergasser et al . , 2012 ) . The regions selected for amplification did not overlap with the region targeted in the VIGS analysis . Primer sets for Salpiglossis orthologs of PDS and elongation factor alpha ( EF1a ) were used as controls . We used 1 µg of total RNA from a single VIGS plant with an acylsugar phenotype to generate cDNA using the Thermo Fisher Superscript II RT kit . An initial amplification and visualization on a 1% agarose gel was performed to ensure that the primers yielded an amplicon with the predicted size and did not show visible levels of primer dimers . We first tested multiple primer sets per gene and selected primers within 85–115% efficiency range using a dilution series of cDNA from uninoculated plants . These primers were used for the final qRT-PCR reaction . The Ct values for the transcripts ( on 1x template ) were measured in triplicate , which were averaged for the analysis . Both uninoculated and empty vector controls were measured with all primer sets for ΔΔCt calculations . We confirmed the phylogenetic positions of Salpiglossis and Hyoscyamus niger using the chloroplast ndhF and trnLF marker based phylogenies ( Figure 2—figure supplement 1A , B ) . Specifically , we amplified these regions using locus-specific primers , sequenced the amplicons and assembled the contigs using Muscle ( Edgar , 2004 ) . A neighbor joining tree including ndhF and trnLF sequences from NCBI Genbank was used to confirm the identity of the plant under investigation . Phylogenetic position of S . nigrum was confirmed based on BLASTN vs . all S . nigrum nucleotide sequences from NCBI . Several DNA barcodes ( e . g . trnLF intergenic spacer , atpFH cds ) showed 100% identity to S . nigrum RNA-seq transcripts over 100% of their lengths . Enzyme assays were performed as previously described ( Fan et al . , 2016b ) with the following modifications: Enzyme assays were performed in 30 µL reactions ( 3 µL enzyme +3 µL 1 mM acyl CoA +1 µL 10 mM ( acylated ) sucrose +23 µL 50 mM NH4Ac buffer pH 6 . 0 ) or 20 µL reactions with proportionately scaled components . Reactions that used an NMR-characterized substrate were performed with 0 . 2 µL substrate and 23 . 8 µL buffer . Reaction products were characterized using Waters Xevo G2-XS QToF LC/MS ( Waters Corporation , USA ) using previously described protocols ( Fan et al . , 2016b ) . Identification of SsASAT5 activity required a significant amount of the starting substrate , however , standard sequential reactions used to isolate SsASAT1 , 2 , 3 activities do not produce enough starting substrate for SsASAT5 . Hence , we used S3:15 ( 5 , 5 , 5 ) purified from a back-crossed inbred line ( BIL6180 , [Ofner et al . , 2016] ) whose NMR resolved structure shows R2 , R3 , R4 positions acylated by C5 chains ( unpublished data ) — same as Salpiglossis tri-acylsugars . SsASAT5 could use this substrate to acylate the furanose ring ( Figure 2—figure supplement 8A , B ) , however , the tetra-acylsugar thus produced is not further acetylated to penta-acylsugar , suggesting the C2 acylation by SsASAT5 is not the same as the C2 acylation on tetra-acylsugars in vivo . SsASAT5 also acetylated the acceptors S4:21 ( 5 , 5 , 5 , 6 ) and S4:19 ( 2 , 5 , 6 , 6 ) purified from Salpiglossis trichome extracts on the furanose ring to penta-acylsugars that co-migrated with the most abundant penta-acylsugars purified from the plant ( Figure 2C , D ) . Finally , S5:23 ( 2 , 5 , 5 , 5 , 6 ) and S5:21 ( 2 , 2 , 5 , 6 , 6 ) purified from plant extracts were enzymatically acetylated to hexa-acylsugars , which co-migrated with two low abundance hexa-acylsugars from the plant ( Figure 2—figure supplement 8C , D ) . Positive mode fragmentation patterns suggested that SsASAT5 possessed the capacity to acetylate both pyranose ( weak ) and furanose rings of the penta-acylsugars ( Figure 2—figure supplement 8E , F ) . Primers to amplify fragments of transcripts for VIGS were chosen to minimize probability of cross silencing other Salpiglossis transcripts due to sequence similarity or due to homopolymeric regions ( Supplementary files 2 , 3 ) . This was achieved using a custom Python script ( Moghe , 2017; copy archived at https://github . com/elifesciences-publications/2017_Solanaceae ) , which divided the entire transcript of interest in silico into multiple overlapping 20nt fragments , performed a BLAST against the Salpiglossis transcriptome and tomato genome and flagged fragments with >95% identity and/or >50% homopolymeric stretches . Contiguous transcript regions with >12 unflagged , high-quality fragments were manually inspected and considered for VIGS . 1–2 300 bp regions were cloned into the pTRV2-LIC vector ( Dong et al . , 2007 ) and transformed into Agrobacterium tumefaciens GV3101 . Agro-infiltration of Salpiglossis plants was performed as described previously ( Velásquez et al . , 2009 ) using the prick inoculation method . Salpiglossis phytoene desaturase was used as the positive control for silencing . Empty pTRV2-LIC or pTRV2 vectors were used as negative controls . Each VIGS trial was done slightly differently while modifying the growth , transformation and maintenance conditions of this non-model species . The experimental details , including the optimal growth and VIGS conditions that give the fastest results , are noted in Supplementary file 4 . All steps in the phylogenetic reconstruction were carried out using MEGA6 ( Tamura et al . , 2013 ) . Amino acid sequences were aligned using Muscle with default parameters . Maximum likelihood and/or neighbor joining ( NJ ) were used to generate phylogenetic trees . For maximum likelihood , the best evolutionary model ( JTT[Jones-Taylor-Thornton]+G + I with five rate categories ) was selected based on the Akaike Information Criterion after screening several models available in the MEGA6 software , while for NJ , the default JTT model was used . Support values were obtained using 1000 bootstrap replicates , however trees obtained using 100 bootstrap replicates also showed similar overall topologies . Trees were generated either using ‘complete deletion’ or ‘partial deletion with maximum 30% gaps/missing data’ options for tree reconstruction . Sequences < 350 aa ( eg: Solyc10g079570 ) were excluded from this analysis . Trees were also generated using RAxML v8 . 0 . 6 ( Stamatakis , 2006 ) with 1000 rapid bootstrap replicates , using the best models as specified by the Akaike Information Criterion in MEGA6 . Similarity searches against the 1kp and NCBI nr databases were performed using TBLASTN , using SlASAT1 , SlASAT3 , SsASAT1 and SsASAT2 protein sequences as queries . For the 1kp database , TBLASTN was performed against all Asterid sequences , and the best non-Solanaceae sequences were analyzed using phylogenetic tree reconstruction ( see below; Figure 6—source data 1 ) . The BLAST at NCBI was performed against several specific groups of species , namely ( i ) orders Gentianales + Boraginales + Lamiales + Solanales- ( family Solanaceae ) , ( ii ) Only Convolvulaceae , ( iii ) Solanaceae- ( sub-family Solanoideae ) , ( iv ) Solanoideae- ( genera Solanum + Capsicum ) , ( v ) Solanoideae- ( genus Solanum ) , ( vi ) Solanum- ( section Lycopersicon +species tuberosum ) . The top 10 hits from each search were manually curated and analyzed in a phylogenetic context with SlASATs and SsASATs ( Figure 6—figure supplements 2 and 3 ) . We used a similar approach to search the annotated peptide sequences in C . canephora ( Denoeud et al . , 2014 ) and Ipomoea trifida ( Hirakawa et al . , 2015 ) sequence databases . Sequences that provided additional insights into the evolution of the ASAT clade were integrated into the final gene tree shown in Figure 6A . We used the overall approach employed in Supplementary Note 5 of ( Bombarely et al . , 2016 ) , where the authors obtained dS distributions between S . lycopersicum and Petunia homologs . We identified homologs using an all-vs-all protein BLAST , followed by selection of five top matches . The BLAST results and the genomic locations of the two gene sets were provided as input to MCScanX ( Wang et al . , 2012 ) with default parameters . The collinear blocks identified by MCScanX were used for calculation of dS values using the yn00 utility in PAMLv4 . 4 ( Yang , 2007 ) . Synteny between the Petunia , Capsicum and tomato genomes was determined as previously described using MCScanX ( Moghe et al . , 2014; Wang et al . , 2012 ) . Regions corresponding to PaASAT1 and its best matching homologs were used to make Figure 7B . This and previous studies provide evidence for a number of acyl chains esterified to sucrose , classified into short ( C2 , C3 , C4 , iC5 , aiC5 , iC6 , aiC6 , C8 ) , long ( nC10 , iC10 , nC11 , nC12 ) and non-aliphatic ( malonyl ) . We only focused on aliphatic chains for this estimate . We typically see one long chain and 3–4 short chains on the sugar molecule across different Solanaceae species . In our calculation , we assumed that each acyl chain has the same probability of being incorporated into the acylsugar , with the core being a sucrose core . Under these assumptions , there are eight acyl chains that could be incorporated at three positions on a tetra-acylsugar and 12 acyl chains on the fourth position . This gives a theoretical estimate of 888*12 = 6144 acylsucroses that could be produced with the above acyl chains . All RNA-seq datasets are deposited in the NCBI Short Read Archive under the BioProject PRJNA263038 . Coding sequences of ASATs are provided in Supplementary file 2 and have been deposited in GenBank ( KY978746-KY978750 ) . RNA-seq transcripts and orthologous group membership information has been uploaded to Dryad ( provisional DOI: 10 . 5061/dryad . t7r64 ) .	There are about 300 , 000 species of plant on Earth , which together produce over a million different small molecules called metabolites . Plants use many of these molecules to grow , to communicate with each other or to defend themselves against pests and disease . Humans have co-opted many of the same molecules as well; for example , some are important nutrients while others are active ingredients in medicines . Many plant metabolites are found in almost all plants , but hundreds of thousands of them are more specialized and only found in small groups of related plant species . These specialized metabolites have a wide variety of structures , and are made by different enzymes working together to carry out a series of biochemical reactions . Acylsugars are an example of a group of specialized metabolites with particularly diverse structures . These small molecules are restricted to plants in the Solanaceae family , which includes tomato and tobacco plants . Moghe et al . have now focused on acylsugars to better understand how plants produce the large diversity of chemical structures found in specialized metabolites , and how these processes have evolved over time . An analysis of over 35 plant species from across the Solanaceae family revealed hundreds of acylsugars , with some plants accumulating 300 or more different types of these specialized metabolites . Moghe et al . then looked at the enzymes that make acylsugars from a poorly studied flowering plant called Salpiglossis sinuata , partly because it produces a large diversity of these small molecules and partly because it sits in a unique position in the Solanaceae family tree . The activities of the enzymes were confirmed both in test tubes and in plants . This suggested that many of the enzymes were “promiscuous” , meaning that they could likely use a variety of molecules as starting points for their chemical reactions . This finding could help to explain how this plant species can make such a wide variety of acylsugars . Moghe et al . also discovered that many of the enzymes that make acylsugars are encoded by genes that were originally copies of other genes and that have subsequently evolved new activities . Plant scientists and plant breeders value tomato plants that produce acylsugars because these natural chemicals protect against pests like whiteflies and spider mites . A clearer understanding of the diversity of acylsugars in the Solanaceae family , as well as the enzymes that make these specialized metabolites , could help efforts to breed crops that are more resistant to pests . Some of the enzymes related to those involved in acylsugar production could also help to make chemicals with pharmaceutical value . These new findings might also eventually lead to innovative ways to produce these chemicals on a large scale .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "plant", "biology", "evolutionary", "biology" ]	2017	Evolutionary routes to biochemical innovation revealed by integrative analysis of a plant-defense related specialized metabolic pathway
The hippocampal–entorhinal system encodes a map of space that guides spatial navigation . Goal-directed behaviour outside of spatial navigation similarly requires a representation of abstract forms of relational knowledge . This information relies on the same neural system , but it is not known whether the organisational principles governing continuous maps may extend to the implicit encoding of discrete , non-spatial graphs . Here , we show that the human hippocampal–entorhinal system can represent relationships between objects using a metric that depends on associative strength . We reconstruct a map-like knowledge structure directly from a hippocampal–entorhinal functional magnetic resonance imaging adaptation signal in a situation where relationships are non-spatial rather than spatial , discrete rather than continuous , and unavailable to conscious awareness . Notably , the measure that best predicted a behavioural signature of implicit knowledge and blood oxygen level-dependent adaptation was a weighted sum of future states , akin to the successor representation that has been proposed to account for place and grid-cell firing patterns . Animals efficiently extract abstract relationships between landmarks , events , and other types of conceptual information , often from limited experience . Knowing such regularities can help us act in an environment , because the relationships between items that have never been experienced together can easily be computed and exploited in order to make novel inference . In physical space , spatially tuned cells in the hippocampal–entorhinal system have precise place ( O’Keefe and Dostrovsky , 1971 ) and grid ( Hafting et al . , 2005 ) codes that may form the neural basis of a ‘cognitive map’ ( O’Keefe and Nadel , 1978 ) . It is likely that the particular form of these representations enables rapid computations of spatial relationships such as distances and vector paths ( Bush et al . , 2015; Stemmler et al . , 2015 ) . The potential for such rapid online computations embedded into neuronal representations may explain how animals can find novel paths through space ( McNaughton et al . , 2006; Mittelstaedt and Mittelstaedt , 1980 ) or rapidly reroute when obstacles are introduced ( Alvernhe et al . , 2011 ) or removed ( Alvernhe et al . , 2008 ) . Indeed , in humans , signals that encode distance metrics between landmarks ( Howard et al . , 2014; Morgan et al . , 2011 ) and directions to goals ( Chadwick et al . , 2015 ) can be read out directly from functional magnetic resonance imaging ( fMRI ) data in the entorhinal cortex . The hippocampal formation also encodes non-spatial relationships between objects . When these objects can be laid out in a continuous dimension such as time , hippocampal codes extracted from neuronal ensembles ( Rubin et al . , 2015 ) or fMRI voxel patterns ( Ezzyat and Davachi , 2014 ) reflect proximity along this dimension . fMRI signals also appear to reflect veridical angles when two-dimensional abstract spaces are formed from continuous dimensions ( Constantinescu et al . , 2016; Tavares et al . , 2015 ) . However , many relationships that are encoded by the hippocampal formation reflect associations or relationships between discrete objects ( Horner et al . , 2015; Schapiro et al . , 2013 , 2012; Wimmer and Shohamy , 2012 ) . To be a useful source of knowledge , many associations must be organised within an associative structure , but it is unclear how such structures might be represented in the absence of a continuous organising dimension such as space or time . Highly complex relational structures are often learnt implicitly i . e . unintentionally and without explicit awareness ( Cleeremans et al . , 1998; Reber , 1989; Seger and Augart , 1994 ) . Neurally , implicitly acquired relational knowledge can be reflected as increases in neural similarity for pairwise associations in the temporal cortex ( Schapiro et al . , 2012 ) or for members of a temporal community structure ( Schapiro et al . , 2013 ) . However , it is unclear whether map- or graph-like knowledge structures might be encoded non-consciously , i . e . without subjects being aware of relationships between objects . Here , we explicitly tested this notion using a fMRI adaptation paradigm that allowed us to quantify the relationships between object representations in a neuronal representational space following an implicit learning paradigm . We presented human participants with sequences of objects where stimulus transitions were drawn from random walks along a graph structure . Within the hippocampal-entorhinal system , a map-like organisation of the relationships between object representations could be extracted from fMRI adaptation data acquired on the subsequent day . In this map , associative distance between objects formed a metric that allowed us to extract organising dimensions . This suggests that the brain can automatically organise abstract relational information into map-like structures even if the relationships between objects are non-spatial rather than spatial , discrete rather than continuous , and unavailable to conscious awareness . A signature of map-like encoding was also present behaviourally in a separate group of subjects , demonstrating implicit memory of the structure . We found no evidence for a mapping of discrete relationships into Euclidian space . Instead , the fMRI adaptation pattern as well as behaviour are more consistent with distance measures reflecting the distribution of future states . These principles are consistent with a predictive representation such as the successor representation ( Dayan , 1993 ) . Such a predictive map of state space may facilitate the rapid computation of values in a reinforcement learning world ( Momennejad et al . , 2016; Russek et al . , 2016 ) . It has recently been demonstrated that the successor representation can account for a number of properties of place cell and grid cell activity ( Stachenfeld et al . , 2016 , 2014 ) . We exposed 23 human participants to object sequences whose stimulus transitions , unbeknownst to them , were determined by a random walk in a graph ( Figure 1A ) . Subjects performed a behavioural cover task , in which they learned to associate a random stimulus orientation with a button press . In the task instructions , any reference to a sequence or an underlying structure was avoided . After the fMRI experiment , subjects were debriefed and none reported any explicit knowledge of structure in the task . To test whether this exposure to object sequences induced implicit knowledge about the graph , we scanned the subjects on a subsequent day using fMRI while exposing them to a subset of the same objects presented in a random order ( only a reduced graph was presented to increase statistical power; Figure 1B ) . In 10% of the fMRI trials , subjects performed an unrelated cover task , reporting whether a grey patch had been present on the preceding object . Neither accuracy nor response time in this task depended on the object on screen or the transition structure ( Figure 1—figure supplement 1 ) . 10 . 7554/eLife . 17086 . 003Figure 1 . Experimental design . ( A ) Graph structure used to generate stimulus sequences on day 1 . Trial transitions were drawn from random walks along the graph . ( B ) Objects on reduced graph presented to subjects in the scanner on day 2 . Trial transitions were random . In both sessions , participants performed simple behavioural cover tasks . See Figure 1—figure supplement 1 for behavioural performance during the training and the scan sessions . fMRI: functional magnetic resonance imaging . DOI: http://dx . doi . org/10 . 7554/eLife . 17086 . 00310 . 7554/eLife . 17086 . 004Figure 1—figure supplement 1 . Task performance . ( A ) Response times ( ms ) and ( B ) performance on the orientation judgment cover task performed during training on day 1 for each of the 12 blocks . ( C ) Graph structure indicating the object position . ( D ) Response times ( F6 , 132 = 0 . 68 , p=0 . 67 ) and ( E ) percent correct ( F6 , 132 = 0 . 56 , p=0 . 77 ) on the patch detection cover task performed during the functional magnetic resonance imaging experiment do not differ for the different object locations . ( F ) Response times on the patch detection cover task do not depend on the distance between objects on the graph ( F2 , 44 = 0 . 72 , p=0 . 49 ) . Error bars show mean and standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 17086 . 004 We exploited fMRI adaptation ( Barron et al . , 2016; Grill-Spector et al . , 2006 ) to investigate the representational similarity for different objects on the graph . We reasoned that in regions encoding a map-like representation of the overall task structure , the degree of similarity in neural representation , and therefore the fMRI adaptation , should decrease as a function of distance between items on the graph . Based on this reasoning , we first looked for brain regions whose fMRI responses to each object increased as a linear function of the link distance of the preceding item . This adaptation analysis revealed a cluster bilaterally in the entorhinal cortex ( Figure 2A , family-wise error corrected at peak level within a bilateral entorhinal cortex/subiculum mask , left p=0 . 014 , peak t22 = 4 . 42 , [−18 , –19 , −22] and right p=0 . 006 , peak t22 = 4 . 75 , [24 , −25 , −22] . A right , but not left peak also survived small volume correction ( SVC ) for a larger region of interest [ROI] comprising the hippocampus , parahippocampal cortex , and entorhinal cortex , left p=0 . 058 and right p=0 . 026 , see ROIs in Figure 2—figure supplement 1 ) . This adaptation effect cannot be explained by basic statistics of the object sequence , such as the number of times an object occurred during training , or basic features of the graph structure , such as the number of neighbours an object has on the graph ( Figure 2—figure supplement 3 ) . 10 . 7554/eLife . 17086 . 005Figure 2 . Functional magnetic resonance imaging adaptation in the hippocampal–entorhinal system decreases with distance on the graph . ( A ) Whole-brain analysis showing a decrease in functional magnetic resonance imaging adaptation with link distance in the hippocampal–entorhinal system , thresholded at p<0 . 01 , uncorrected for visualisation . ( B ) Within the hippocampal–entorhinal system , green indicates greater suppression if the preceding stimulus was a neighbour relative to a stimulus two or three links away . Red indicates greater suppression if a preceding stimulus was two links away than three links away . The depicted areas were used as regions of interest for analyses in ( C ) ( green ) and ( D ) ( red ) . ( C ) Parameter estimates for link 2 versus link 3 transitions extracted from the green entorhinal region of interest in Figure 2B ( t22 = 2 . 27 , p=0 . 03 ) . Other brain areas do not show this increase in activity with distance ( Figure 2—figure supplements 2D , E ) . ( D ) Parameter estimates extracted from the red entorhinal region of interest in Figure 2B , sorted according to whether objects were connected on the graph or not ( t22 = 2 . 34 , p=0 . 03 ) . ( E ) Parameter estimates extracted from the peak MNI coordinate reported in Chadwick et al . ( 2015 ) , [−20 , –25 , −24] and sorted according to distance ( F2 , 44 = 10 . 04 , p=0 . 0003 ) . See Figure 2—figure supplement 1 for masks used for small-volume correction in Figure 2A , Figure 2—figure supplement 2 for distance-dependent scaling effects in other brain regions and Figure 2—figure supplement 3 for effects of object familiarity and centrality . Error bars show mean and standard error of the mean . a . u . : arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 17086 . 00510 . 7554/eLife . 17086 . 006Figure 2—figure supplement 1 . Anatomically defined regions of interest used for small-volume correction . ( A ) Mask comprising the bilateral entorhinal cortex and subiculum , received with thanks from Chadwick et al . ( 2015 ) . ( B ) Mask comprising the bilateral entorhinal cortex , hippocampus , and parahippocampal cortex . Regions were defined using the maximum probability tissue labels provided by Neuromorphometrics , Inc . ( http://Neuromorphometrics . com ) . DOI: http://dx . doi . org/10 . 7554/eLife . 17086 . 00610 . 7554/eLife . 17086 . 007Figure 2—figure supplement 2 . Distance-dependent scaling of neural activity is specific to the hippocampal–entorhinal system . ( A ) All areas displaying a decrease in functional magnetic resonance imaging adaptation with graph distance . A cluster in the subgenual cortex did not survive whole-brain correction for multiple comparisons ( punc=0 . 0008 , pFWE=0 . 99 , peak t22 = 3 . 58 [3 , 23 , −10] ) . ( B ) All areas displaying suppression for connected stimuli relative to non-connected stimuli on the graph . In addition to the hippocampal–entorhinal clusters reported in the main text , clusters in the orbitofrontal cortex ( peak t22 = 2 . 93 , [30 , 41 , −10] ) and subgenual cortex ( peak t22 = 3 . 33 , [0 , 23 , −7] ) were used to define regions of interest in order to test distance-dependent scaling , see ( D , E ) . ( C ) All areas displaying greater suppression if a preceding stimulus was two links away rather than three links away . The corollary test for a difference in activity for connected versus non-connected objects was only significant in the hippocampal–entorhinal region of interest , see Figure 2B , D . ( D ) Parameter estimates for link 2 versus link 3 transitions extracted from the orbitofrontal cortex region of interest in ( B ) . The difference is not significant ( t22 = 0 . 85 , p=0 . 41 ) . ( E ) Parameter estimates for link 2 versus link 3 transitions extracted from the subgenual cortex region of interest in ( B ) . The difference is not significant ( t22 = 1 . 29 , p=0 . 21 ) . ( A–C ) are thresholded at p<0 . 01 for visualisation . Error bars show mean and standard error of the mean . a . u . : arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 17086 . 00710 . 7554/eLife . 17086 . 008Figure 2—figure supplement 3 . Effects of object familiarity . ( A ) Number of times each object was presented during training , averaged across participants . Note that this measure is directly related to the number of neighbours an object has on the graph . During scanning/behavioural testing , each object of the reduced graph depicted in dark green was presented equally often ( 180 times ) . Error bars denote standard deviation . ( B ) Decrease of activity with an increase in the number of times an object was presented during training ( general linear model 3 ) , thresholded at p<0 . 01 for visualisation . No cluster survives correction for multiple comparisons . No effect can be detected when testing for an increase of activity with an increase in the number of times an object was presented during training . DOI: http://dx . doi . org/10 . 7554/eLife . 17086 . 008 To confirm the statistical robustness of the effect , and to test whether the effect reflected a gradual increase with distance , we separated the effect into two orthogonal components . These components comprised the difference between connected links ( length 1 ) and all other transitions ( lengths 2 and 3; Figure 2B , green ) , and the difference between transitions of length 2 and those of length 3 ( Figure 2B , red ) . These two independent contrasts were used to define ROIs bilaterally in overlapping regions of the entorhinal cortex ( both thresholded at p<0 . 01 uncorrected; ROI 1: left peak t22 = 3 . 85; [−18 , –19 , −22] and right peak t22 = 3 . 26; [24 , −25 , −22] , ROI 2: left peak t22 = 4 . 55 , [18 , −16 , −25] and right peak t22 = 3 . 38 , [−18 , –25 , −25] ) . Because of their statistical independence , we could use the ROI from one contrast to extract data for the corollary test ( t22 = 2 . 27 , p=0 . 03 for length 2 vs . length 3 in ROI 1 , Figure 2C; and t22 = 2 . 34 , p=0 . 03 for connected vs . all other links in ROI 2 , Figure 2D ) . This pair of tests suggests that the fMRI adaptation faithfully represents the link distance . These tests obviate questions of multiple comparisons , because in each case the data are selected from one contrast , and an orthogonal contrast was used for the test statistic . To further demonstrate this within a single test , we required a coordinate that was independent of all the data . We chose a peak location from an independent study investigating a similar relational measure in spatial maps ( Chadwick et al . , 2015 ) . Extracting data from this coordinate ( ROI 3 ) revealed a linear effect of link distance ( Figure 2E , F2 , 44 = 10 . 04 , p<0 . 001 ) , and correspondingly a significant difference between distances of lengths 1 and 3 ( t22 = 3 . 71 , p=0 . 001 ) and lengths 2 and 3 ( t22 = 3 . 19 , p=0 . 004 ) , but not between distances of lengths 1 and 2 ( t22 = 1 . 67 , p=0 . 11 ) . Although this distance effect is suggestive of a map-like organisation , it might also merely reflect the temporal proximity between two objects during training . When the temporal and distance relationships between pairs of objects were allowed to compete for variance in a multiple linear regression , the number of links ( t22 = 3 . 29 , p=0 . 003 ) , but not time ( t22 = 1 . 27 , p=0 . 22 ) , explained the neural signal extracted from the independently defined ROI 3 ( Figure 3A ) . Furthermore , relationships between items arranged in a map-like structure are non-directional . Our subjects were not constrained to experience each pair of transitions an equal number of times ( Figure 3B ) . Based upon this , we could test whether the fMRI signal was better predicted by the true or symmetrised distance between any two objects . We constructed a measure of the shortest path between each pair of objects according to the actual number of times each transition was experienced by a subject during training ( see 'Materials and methods' section ) . When allowing this measure to compete with its symmetrised , and thereby non-directional , self in a linear model , it was the symmetrised version alone that predicted the fMRI suppression effect ( Figure 3C , t22 = 2 . 78 , p=0 . 01 and t22 = −1 . 64 , p=0 . 11 ) . 10 . 7554/eLife . 17086 . 009Figure 3 . Relational information is organised as a map . ( A ) Linear regression on neural activity with number of links and average time between two objects during training as regressors ( t22 = 3 . 29 , p=0 . 003 and t22 = 1 . 27 , p=0 . 22 ) . ( B ) Absolute difference in the number of times a transition was visited in one versus the other direction ( e . g . 5 preceded by 1 vs . 1 preceded by 5 ) normalised by the total number of visits in either direction for all subjects . ( C ) Multiple linear regression on neural activity with the shortest path between objects and the symmetrised shortest path between objects as regressors ( t22 = −1 . 64 , p=0 . 11 and t22 = 2 . 78 , p=0 . 01 ) . ( D ) 7 × 7 matrix representing the average fMRI signal in response to an object depending on which other object preceded , averaged across subjects and symmetrised . Objects were never repeated during scanning; the diagonal entries are therefore set to 0 . This matrix was used for the multidimensional scaling visualised in ( E ) . ( E ) Visualisation of the localisation of the object representations in a two-dimensional space according to multidimensional scaling . Lines indicate transitions experienced during training . The distances between the resulting locations of nodes in a two-dimensional space are significantly correlated with the link distances of the original graph structure ( r = 0 . 65 , p=0 . 003 , see Figure 3—figure supplement 2 for the null distribution used for the permutation test ) . All analyses were performed on data extracted from the peak MNI coordinate reported in Chadwick et al . [2015] ) , [−20 , –25 , −24] , but also hold in an anatomically defined region of interest including the entorhinal cortex and the subiculum ( Figure 3—figure supplement 3 ) . See Figure 3—figure supplement 1 for results if object-specific activity is removed . Error bars show mean and standard error of the mean . a . u . : arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 17086 . 00910 . 7554/eLife . 17086 . 010Figure 3—figure supplement 1 . The distance-dependent scaling cannot be driven by a main effect of object position . Position-specific activity in the region of interest ( ROI ) defined according to ( A ) a connected < non-connected contrast ( ROI 1 , Figure 2B , green , F6 , 132 = 0 . 88 , p=0 . 5 ) , ( C ) a link 2 < link 3 contrast ( ROI 2 , Figure 2B , red , F6 , 132 = 1 . 96 , p=0 . 08 ) and ( E ) the peak coordinate in Chadwick et al . ( 2015 ) ( ROI 3 , F6 , 132 = 1 . 86 , p=0 . 09 ) . The tests reported in Figure 2C–E are also significant if performed after removing object-specific activity from the neural data . This was achieved by subtracting the mean activity for each object before testing for ( B ) a difference in activity for link 2 versus link 3 transitions in ROI 1 ( t22 = 2 . 10 , p=0 . 048 ) , ( D ) a difference in activity for connected versus non-connected stimuli in ROI 2 ( t22 = 2 . 39 , p=0 . 03 ) , or ( F ) a distance effect in ROI 3 ( F2 , 44 = 6 . 68 , p=0 . 003 ) . Error bars show mean and standard error of the mean . a . u . : arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 17086 . 01010 . 7554/eLife . 17086 . 011Figure 3—figure supplement 2 . Map characteristics in a null distribution generated by permuting the links making up the graph structure . ( A ) Correlation of link distances for graphs making up the null distribution with distances resulting from performing multidimensional scaling on the neural data . The null distribution was constructed by permuting the seven links making up the reduced graph structure to a random set of nodes . The correlation between link distances for the actual graph structure and distances resulting from MDS on the neural data is 0 . 65 . Only 0 . 28% of graphs show a correlation with the mapped data of 0 . 65 or higher . ( B ) Number of line crossings for graphs characterised by nodes in the same location as given by the MDS procedure , but with the seven links randomly distributed between nodes . Only 13 . 17% of all possible graphs generated in this way are characterised by no line crossings; 0 . 25% of all graphs have no line crossing and a correlation of 0 . 65 or higher . The arrows indicate the corresponding measure for the actual data . MDS: multidimensional scaling . DOI: http://dx . doi . org/10 . 7554/eLife . 17086 . 01110 . 7554/eLife . 17086 . 012Figure 3—figure supplement 3 . Distance effects in an anatomically defined region of interest comprising the entorhinal cortex and the subiculum . ( A ) Activity scales linearly with link distance ( F2 , 44 = 8 . 41 , p=0 . 0008 ) . Post-hoc pairwise comparisons revealed a significant difference between distances of lengths 1 and 2 ( t22 = 2 . 36 , p=0 . 03 ) , lengths 1 and 3 ( t22 = 3 . 51 , p=0 . 002 ) , and lengths 2 and 3 ( t22 = 2 . 23 , p=0 . 04 ) . ( B ) In a multiple linear regression , link distance , but not time , predicts the neural activity ( t22 = 2 . 13 , p=0 . 04 and t22 = 1 . 23 , p=0 . 23 ) . ( C ) In a multiple linear regression with symmetrised and non-symmetrised distance measures competing for variance , symmetrised distance rather than non-symmetrised distance explains the neural signal ( directional distance: t22 = −1 . 27 , p=0 . 22 , symmetrised distance: t22 = 2 . 41 , p=0 . 02 ) . ( D ) The graph structure can be recovered by performing multidimensional scaling on the average functional magnetic resonance imaging data across subjects ( r = 0 . 58 , p=0 . 008 , permutation test ) . Data for all analyses are extracted from the anatomically defined regions of interest shown in Figure 2—figure supplement 1A . Error bars show mean and standard error of the mean . a . u . : arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 17086 . 012 In order to test whether these map-like features are a consequence of a map-like organisation , we organised the signal into a 7 × 7 matrix , with each matrix element reflecting the mean fMRI response across subjects to transitions between the corresponding pairs of objects ( Figure 3D ) . For example , element [2 , 7] in this matrix is the response to object 7 when preceded by object 2 on the graph , averaged across all subjects . Because the signal is suppressed for nearby objects , this matrix is analogous to a distance matrix . When we applied multidimensional scaling ( MDS ) in order to visualise the most faithful two-dimensional representation of distances in this matrix , the graph structure of our experimental map was recovered despite the subjects’ professed ignorance of any such organisation ( Figure 3E ) . Permutation tests confirm that the multidimensional scaling-mapped distances are significantly more correlated with link distances of the original graph structure than with link distances of a null distribution consisting of all other complete graphs with seven links ( r = 0 . 65 , p=0 . 003 , Figure 3—figure supplement 2A ) . Furthermore , no links cross in the graph resulting from the MDS mapping . This is only true for 13 . 17% of all possible graphs with nodes in the same location , but seven randomly distributed links ( Figure 3—figure supplement 2B ) . Notably , the data were extracted from an independent ROI taken from an experiment investigating maps in allocentric physical space ( Chadwick et al . , 2015 , ROI 3 ) . Results are comparable if parameter estimates are extracted from an anatomically defined ROI comprising the entorhinal cortex and the subiculum ( Figure 3—figure supplement 3 ) . In the reinforcement learning literature , it has been suggested that a cognitive map of the relationship between states may be most useful if the representation of a state is predictive in nature and reflects the distribution of likely future states . This idea has been formalised as the successor representation ( Dayan , 1993; Momennejad et al . , 2016; Russek et al . , 2016 ) , proposed to be encoded by hippocampal place cells ( Stachenfeld et al . , 2016 , 2014 ) . According to this view , hippocampal place cells do not encode an animal’s current location in space , but instead encode a predictive representation of future locations . The successor representation may facilitate reinforcement learning , because the resulting predictive measure of future states could be flexibly combined with reward representations to enable rapid computation of navigational trajectories ( Baram et al . , 2017; Dayan , 1993; Momennejad et al . , 2016; Russek et al . , 2016 ) . Mathematically , the successor representation can be computed from the adjacency matrix A that defines the relationship between states:∑n=0∞γn An= ( I−γA ) −1 with a discount factor γ<1 . Here , entries aij for each An correspond to the number of possible paths of length n between objects i and j . The successor representation therefore computes the weighted sum of distant future states , with An discounted more heavily for larger n ( i . e . for longer paths between pairs of objects ) . Notably , this same representation is common in graph theory , where the matrix ( I−γA ) −1 is termed the matrix resolvent and is used to measure the proximity or ‘communicability’ between nodes in the graph . Graph theory also proposes a second measure that is closely related , the matrix exponential ( Estrada and Hatano , 2008 , 2010 ) :eA=∑n=0∞Ann ! Both measures compute a weighted sum over future states , which easily generalises from continuous to discrete , and from two-dimensional to high-dimensional spaces . In order to test whether the neural distance effects are consistent with such predictive measures , we tested for areas whose fMRI responses to each object increased as a linear function of communicability , corresponding to the negative of the matrix exponential . Compared to the successor representation , the matrix exponential has the advantage that it does not require the fitting of a free parameter . The matrix exponential is small for nodes that are far away from each other on a graph , such that it scales negatively with distance . Unlike a mapping into Euclidian space , communicability significantly distorts the graph structure by shortening links that form part of many paths around the graph structure and lengthening links that would be less frequently visited by a random navigator ( Figure 4A ) . 10 . 7554/eLife . 17086 . 013Figure 4 . Functional magnetic resonance imaging adaptation in the hippocampal–entorhinal system is consistent with predictive representations of relational knowledge . ( A ) Visualisation of communicability coordinates for the graph structure by performing multidimensional scaling on the communicability matrix . ( B ) Whole-brain regression of communicability onto neural activity . ( C ) Visualisation of the communicability effect . Average parameter estimate for each of the 42 stimulus transitions across subjects extracted from a bilateral region of interest in ( B ) ( thresholded at p<0 . 01 ) . The colours of the dots correspond to link distances . This graph is added for visualisation purposes only as the parameter selection is biased . ( D ) Whole-brain regression of communicability onto neural activity when Euclidian distances are included as an additional regressor in the general linear model . All statistical maps are thresholded at p<0 . 01 for visualisation . a . u . : arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 17086 . 01310 . 7554/eLife . 17086 . 014Figure 4—figure supplement 1 . Activity in the hippocampal–entorhinal system is consistent with the successor representation . ( A ) Visualisation of successor representation coordinates for the graph structure by performing multidimensional scaling on the negative of the successor representation , with the free parameter γ set to 0 . 85λmax ( λmax = largest eigenvalue of the adjacency matrix A ) . ( B ) Whole-brain regression of the negative of the successor representation onto neural activity , with γ=0 . 85λmax , reveals linear scaling with the successor representation in the hippocampal–entorhinal formation ( family-wise error corrected at peak level within a bilateral entorhinal cortex/subiculum mask , left p=0 . 005 , peak t22 = 4 . 80 [−18 , –19 , −25] and right p=0 . 007 , peak t22 = 4 . 66 , [24 , −22 , −25] . Both clusters also survived small volume correction for a larger region of interest comprising the hippocampus , parahippocampal cortex , and entorhinal cortex , left p=0 . 02 and right p=0 . 03 , see regions of interest in Figure 2—figure supplement 1B ) . The statistical maps are thresholded at p<0 . 01 for visualisation . DOI: http://dx . doi . org/10 . 7554/eLife . 17086 . 014 We find that neural activity bilaterally in the hippocampal–entorhinal system scales with communicability ( Figure 4B , C , family-wise error corrected at peak level within a bilateral entorhinal cortex/subiculum mask , left p=0 . 001 , peak t22 = 5 . 47 [−18 , –19 , −25] and right p=0 . 0005 , peak t22 = 5 . 79 , [21 , −19 , −28] ) . Both clusters also survived SVC ( small-volume correction ) for a larger ROI comprising the hippocampus , parahippocampal cortex , and entorhinal cortex ( left p=0 . 004 and right p=0 . 004 , see ROIs in Figure 2—figure supplement 1B ) . Activity in the same areas also scales with the negative of the successor representation if the free parameter γ is set to the commonly used value of γ=0 . 85λmax ( λmax = largest eigenvalue of A in modulus , Aprahamian et al . , 2016; Benzi and Klymko , 2013 , Figure 4—figure supplement 1 ) . In the left hippocampal formation , communicability effects are significant even if Euclidian distances are included as an additional regressor ( Figure 4D , p=0 . 006 , peak t22 = 4 . 72 , [−15 , –13 , −19] , SVC mask 1 and p=0 . 027 , SVC mask 2 ) . This suggests that the hippocampal–entorhinal system does not map the graph structure into a Euclidian space . Instead , these results are consistent with the view that the distance effect we observe in this system may be a consequence of the hippocampal formation encoding a predictive representation of states within a graph structure ( Stachenfeld et al . , 2016 , 2014 ) . A map-like representation in the hippocampal–entorhinal system suggests that subjects acquired implicit knowledge about the graph structure , even in the absence of explicit awareness of any regularities in the object sequence . To reveal such implicit learning behaviourally , we asked an independent group of 26 participants , who were trained in the same way as the scanning cohort on the first graph structure ( Figure 5A ) , to repeat the object orientation cover task on day 2 . As was the case for scanned subjects , object transitions were now random and only objects from a reduced graph were presented ( Figure 5B ) . We hypothesised that implicit knowledge about the graph structure would influence response times , such that subjects would respond faster if a preceding object in the test sequence was closer on the graph structure underlying the train sequence . Indeed , we found that log-transformed response times were longer the further away the preceding object was on the graph ( Figure 5C , D ) . In line with our imaging results , response times did not scale with link or Euclidian distance between objects , but instead with communicability ( communicability: t25 = 2 . 77 , p=0 . 01; link distance: t25 = −0 . 40 , p=0 . 69; Euclidian distance: t25 = −0 . 85 , p=0 . 40 , Figure 5C , D ) . 10 . 7554/eLife . 17086 . 015Figure 5 . Response times reflect graph structure . ( A ) Graph structure used to generate stimulus sequences on day 1 . Trial transitions were drawn from random walks along the graph structure . ( B ) Objects on reduced graph presented to subjects on day 2 . Trial transitions were random . In both sessions , participants performed an object orientation cover task under which response times were measured . ( C ) A regression of communicability , link distance , and Euclidian distance onto log-transformed response times across subjects ( communicability: t25 = 2 . 77 , p=0 . 01; link distance: t25 = −0 . 40 , p=0 . 69; Euclidian distance: t25 = −0 . 85 , p=0 . 40 ) . ( D ) Visualisation of the relationship between communicability and log-transformed response times . Communicability measures were divided into six bins with an equal number of object–object transitions per bin . The y-axis corresponds to the average demeaned log-response time across subjects for each bin . Error bars denote the standard error of the mean . a . u . : arbitrary units . DOI: http://dx . doi . org/10 . 7554/eLife . 17086 . 015 The hippocampal–entorhinal system is engaged when an animal navigates within a physical environment and acquires flexible knowledge about spatial relationships . In mammals , the hippocampal–entorhinal system contributes to spatial navigation by mapping relationships in situations where knowledge is physical , continuous , and consciously available ( Chadwick et al . , 2015; Derdikman and Moser , 2010; Howard et al . , 2014; Spiers and Maguire , 2007 ) . Here , we used a statistical learning paradigm to demonstrate that the hippocampal–entorhinal system also efficiently extracts statistical regularities in a non-spatial task where the relationships between items are discrete , and organises this non-spatial relational knowledge in an abstract relational map , suggesting that the hippocampal–entorhinal system creates metric representations of discrete relationships based on associative strengths ( Eichenbaum and Cohen , 2014 ) . Notably , there were no continuous dimensions in our discrete stimulus sets . The dimensions had to be extracted from the associations . These results add to the notion that the hippocampal formation maps experiences across a wide range of different dimensions , thereby supporting flexible behaviour across many domains of life ( O’Keefe and Nadel , 1978; Tolman , 1948 ) . Recent studies have focused on the human hippocampal formation for storing the relational knowledge that makes up simple world models in reinforcement learning ( Barron et al . , 2013; Boorman et al . , 2016; Bornstein and Daw , 2013; Wimmer and Shohamy , 2012 ) . Our findings extend these simple associative paradigms to more complex associative maps , and demonstrate that these maps may be learnt implicitly . Notably , the resulting map is not Euclidian in nature . Instead , we find that the neural data as well as the behaviour are consistent with a measure corresponding to a weighted sum of future states in the graph structure . This is consistent with the idea that reinforcement learning benefits from knowledge about warped geometries in space , which may be represented as predictive maps , or successor representations , in the hippocampal formation ( Stachenfeld et al . , 2016 , 2014 ) . Such predictive representations of the relationships between discrete objects or states of the world may be combined with reward representations to enable flexible goal-directed behaviour ( Baram et al . , 2017; Dayan , 1993; Momennejad et al . , 2016; Russek et al . , 2016 ) . Our data suggest that the map of the relational structure in discrete abstract space may be encoded by the pattern of firing in the entorhinal cortex , which also encodes maps in continuous physical space . The entorhinal cortex is noted for the presence of grid cells ( Hafting et al . , 2005 ) , which may provide a metric for measuring distances in physical space ( Bush et al . , 2015; Stemmler et al . , 2015 ) , allowing vector navigation . However , recent theoretical treatments suggest that grid-like firing might also be understood as a principal component decomposition of the covariance information between place cells ( Dordek et al . , 2016 ) or the transitions between states ( Stachenfeld et al . , 2016 , 2014 ) . Together with the finding of hippocampal cells in humans that encode individual concepts ( which may be analogous to place cells ) , these theories can explain how grid-like firing patterns could extend to discrete spaces ( such as the one we have used here ) , and also make predictions for how grid-like coding might extend to higher-dimensional spaces ( Baram et al . , 2017 ) . Our behavioural results are consistent with the observation that the human brain can acquire abstract knowledge unconsciously and automatically by extracting statistical regularities from incidental exposure to events generated according to a set of rules ( Berry and Broadbent , 1984; Cleeremans et al . , 1998; Reber , 1989 , 1967; Seger and Augart , 1994 ) . The resulting implicit knowledge can be used to guide behaviour despite an inability to verbalise the underlying regularities . Neurally , implicit learning of sequences involves medial temporal lobe structures , including the hippocampus , subiculum , and entorhinal cortex ( Schendan et al . , 2003 ) . The hippocampal–entorhinal system also responds to the violation of learnt sequence structures ( Kumaran and Maguire , 2006 ) and signals the likelihood of events in learnt sequences ( Strange et al . , 2005 ) . This may be facilitated by an increase in representational similarity as stimuli become embedded into knowledge structures ( Schapiro et al . , 2013 , 2012 ) . Our results add to this literature by proposing a specific way in which implicit knowledge may be organised in maps to facilitate goal-directed behaviour . It is notable that we did not find clear evidence for the map-like structure outside the hippocampal formation , although areas such as the orbitofrontal cortex have been shown to represent cognitive maps of decision spaces ( Schuck et al . , 2016 ) . It is possible that this is because our study relies on implicit learning , and because the subjects do not have to use the associative structure for any task . Indeed , we note that neural signals can be recorded in frontal and parietal cortices , reflecting the ‘state-prediction errors’ that ensue when predicted state relationships are breached during behavioural control ( Gläscher et al . , 2010 ) . Similar prediction errors in the orbitofrontal cortex during active learning predict later changes in hippocampal representations of the stored model ( Boorman et al . , 2016 ) . These and similar ideas have led to a theoretical account of place and grid activity in the hippocampal formation as state representations in reinforcement learning models ( Stachenfeld et al . , 2016 , 2014 ) . It has long been known that the hippocampal formation is important for tasks that rely on associative and relational knowledge . It supports the organisation of stimuli across arbitrary stimulus dimensions such as temporal co-occurrence ( Schapiro et al . , 2013 , 2012 ) or social rank ( Kumaran et al . , 2012 ) , and organises behaviourally relevant stimulus categories in a hierarchy ( McKenzie et al . , 2014 ) . These organisational principles facilitate generalising over individual episodes ( Komorowski et al . , 2013 ) and enable transitive inference by combining newly formed associations between discrete stimuli ( Collin et al . , 2015; Heckers et al . , 2004; Horner et al . , 2015; Preston et al . , 2004; Schlichting et al . , 2015 ) . Value spreading across associated stimulus representations in the hippocampus can then directly influence behaviour in novel decision-making situations ( Wimmer and Shohamy , 2012 ) . We hope that the current findings help to reconcile these results with the spatial functions of the same neural structures . Such an organisation of relational information might be the basis for an animal’s ability to navigate through an abstract concept space and to perform flexible computations without direct experience . Twenty three volunteers ( aged 18–31 years , mean age ± standard deviation 23 . 5 ± 3 . 7 years , 15 males ) with normal or corrected-to-normal vision and no history of neurological or psychiatric disorders participated in the fMRI experiment . All subjects gave written informed consent and the study was approved by the University College London Hospitals Ethics Committee . The study took place at the Wellcome Trust Centre for Neuroimaging . Subjects were naïve to the purpose of the experiment . Thirty one coloured and shaded object images that were similar in terms of their familiarity and complexity were selected from the 'Snodgrass and Vanderwart ‘Like’ Objects’ picture set ( http://wiki . cnbc . cmu . edu/Objects , Rossion and Pourtois , 2004 ) . For each subject , a subset of 12 objects was chosen and randomly assigned to the 12 nodes of the graph shown in Figure 1A . On day 1 , subjects were exposed to object sequences generated from a random walk on the graph , where only objects that were directly connected to another object by a link could follow a presentation of this object . To avoid local repetitions , we constrained sequences such that at least three objects had to occur between any two presentations of the same object . Each object was randomly presented in one of two orientations , which were mirror images of each other . Before the start of the experiment , subjects were shown the entire set of 12 stimuli and instructed to remember which of two buttons to press for a particular object orientation ( normal or mirrored ) . During the actual training , subjects were instructed to press the button associated with the stimulus orientation as quickly and accurately as possible while watching the object sequences . Visual feedback after each button press indicated whether or not a response was correct . Object orientation was randomised across trials and key assignment was counterbalanced across subjects . Subjects learnt to perform the task quickly and accurately ( Figure 1—figure supplement 1 ) . Stimuli were presented for 2 s and each experimental block consisted of 133 object presentations . Subjects performed this experiment for 12 blocks in total . Between blocks ( ca . every 5 min ) , subjects were free to take self-paced breaks . On the next day , subjects were presented with object sequences in the scanner . Only the seven objects corresponding to the locations illustrated in Figure 1B were presented and stimuli were never repeated . This reduced the total number of stimulus–stimulus transitions and thereby increased statistical power for our key question of interest , as this large number of times that each transition was probed provided us with a more accurate estimate of the respective suppression effects . Furthermore , stimulus transitions did not follow the graph structure , but were instead randomised with a constraint that each of the 42 possible object transitions occurred exactly 10 times per block ( objects were never repeated ) . The fMRI experiment consisted of 421 items per run and three experimental runs . Stimuli were presented for 1 s , with a jittered inter-trial interval ( ITI ) generated from a truncated Poisson distribution with a mean of 2 s . While observing the object sequences , subjects performed a cover task of infrequently reporting by button press whether a small grey patch had appeared on a preceding trial . The patch was present on a randomly selected 50% of the objects . Trials on which subjects had to report the existence of a grey patch were signalled by a green cross during the inter-stimulus interval instead of the standard white cross . The cross was green exactly once after each of the 42 possible transitions ( i . e . in 10% of the total number of trials ) . In 50% of those cases , a patch had been present on the preceding trial . Each correct button press was rewarded with £0 . 10 paid out in addition to a £33 show-up fee to ensure that subjects attended to the stimuli . Subjects received brief training on this task before they performed it in the scanner . Key assignment was counterbalanced across subjects . Subjects performed the cover task very well ( correct performance rate across subjects: 94 ± 3% , mean ± standard error of the mean ) , confirming that they paid attention to the presented objects throughout the duration of the scan . After the experiment , subjects were asked whether they noticed any differences between the object sequences presented on day 1 and the object sequences presented in the scanner on day 2 . While subjects realised that some objects were missing on day 2 , none reported any awareness of the fact that the sequence differed in any other way . When asked explicitly , no subject was aware of the fact that the sequences on day 1 were generated according to an underlying structure . Visual stimuli were projected onto a screen via a computer monitor . Subjects indicated their choice using an MRI-compatible button box . MRI data were acquired using a 32-channel head coil on a 3 Tesla Allegra scanner ( Siemens , Erlangen , Germany ) . A T2*-weighted echo-planar sequence was used to collect 43 transverse slices ( ascending order ) of 2-mm thickness with 1-mm gaps and an in-plane resolution of 3 × 3 mm , a repetition time of 3 . 01 s , and an echo time of 70 ms . Slices were tilted by 30° relative to the rostro-caudal axis and a local z-shim with a moment of −0 . 4 mT/m was applied to the orbitofrontal cortex region ( Weiskopf et al . , 2006 ) . The first five volumes of each block were discarded to allow for scanner equilibration . After the experimental sessions , a T1-weighted anatomical scan with 1 × 1 × 1 mm resolution was acquired . In addition , a whole-brain field map with dual echo-time images ( TE1 = 10 ms , TE2 = 14 . 76 ms , resolution 3 × 3 × 3 mm ) was obtained in order to measure and later correct for geometric distortions due to susceptibility-induced field inhomogeneities . We performed slice time correction , corrected for signal bias , and realigned functional scans to the first volume in the sequence using a six-parameter rigid body transformation to correct for motion . Images were then spatially normalised by warping subject-specific images to an MNI ( Montreal Neurological Institute ) reference brain , and smoothed using an 8-mm full-width at half maximum Gaussian kernel . All pre-processing steps were performed with SPM12 ( Wellcome Trust Centre for Neuroimaging , http://www . fil . ion . ucl . ac . uk/spm ) . We implemented three types of event-related general linear models ( GLMs ) in order to analyse the fMRI data . The first set of GLMs contained separate onset regressors for each of the seven objects with a patch and without a patch . Each onset regressor was accompanied by different parametric regressors . These corresponded to the link distance ( defined as the minimum number of links between the pair of items; i . e . distance 1 , 2 , or 3 , Figure 2A and Figure 2—figure supplement 2A ) , the communicability ( see below , Figure 4B ) , and the negative of the successor representation ( Figure 4—figure supplement 1B ) between the object on trial t and the preceding object on trial t – 1 on the graph presented in Figure 1B . For the analysis reported in Figure 4D , both communicability and Euclidian distances were included as parametric regressors . Communicability was computed as the negative of the matrix exponential of the adjacency matrix A , describing the relationship between nodes on the graph:c=−eA=−∑n=0∞Ann ! The successor representation was computed as:∑n=0∞γnAn= ( I−γA ) −1 with γ set to the commonly used value of 0 . 85λmax ( λmax = largest eigenvalue of A in modulus , Aprahamian et al . , 2016; Benzi and Klymko , 2013 ) . Euclidian distances were computed from the graph in Figure 1A , with all distances between objects connected by a link set to 1 . In the second type of GLM ( GLM 2 ) , all 42 possible object transitions ( object 1 preceded by object 2; object 1 preceded by object 3 , … , object 7 preceded by object 6 ) were modelled separately for patch trials and no-patch trials . A third type of GLM contained one onset regressor for all objects with a patch , and a separate onset regressor for objects without a patch . Each onset regressor was accompanied by a parametric regressor indicating the number of times an object was presented during training ( Figure 2—figure supplement 3 ) . All GLMs included a button press regressor as a regressor of no interest . Trials associated with a button press and the two subsequent trials were not included in the main regressors in order to avoid button press-related artefacts . All regressors were convolved with a canonical haemodynamic response function . Because of the sensitivity of the blood oxygen level-dependent signal to motion and physiological noise , all GLMs also included six motion regressors obtained during realignment , as well as 10 regressors for cardiac phase , 6 for respiratory phase and 1 for respiratory volume extracted with an in-house developed Matlab toolbox as nuisance regressors ( Hutton et al . , 2011 ) . Models for the cardiac and respiratory phase and their aliased harmonics were based on RETROICOR ( Glover et al . , 2000 ) . Sessions were modelled separately within the GLMs . The contrast images of all subjects from the first level were analysed as a second-level random effects analysis . We report our results in the hippocampal–entorhinal formation , as this was our a priori ROI , at a cluster-defining statistical threshold of p<0 . 001 uncorrected , combined with SVC for multiple comparisons ( peak-level family-wise error [FWE] corrected at p<0 . 05 ) . For the SVC procedure , we used two different anatomical masks . The first mask consisted of the entorhinal cortex and subiculum alone and was received with thanks from Chadwick et al . ( 2015 ) , ( Figure 2—figure supplement 1A ) . The second mask also contained other medial temporal lobe regions implicated in encoding physical space and comprised the hippocampus , entorhinal cortex , and parahippocampal cortex , as defined according to the maximum probability tissue labels provided by Neuromorphometrics , Inc . ( Figure 2—figure supplement 1B ) . Activations in other brain regions were only considered significant at a level of p<0 . 001 uncorrected if they survived whole-brain FWE correction at the cluster level ( p<0 . 05 ) . While no areas survived this stringent correction for multiple comparisons , other regions are reported in Figure 2—figure supplement 2 at an uncorrected threshold of p<0 . 01 for completeness . While we used masks to correct for multiple comparisons in our ROI , all statistical parametric maps presented in the manuscript are unmasked . To independently test for distance-dependent scaling of activity within the entorhinal cortex , we defined two different ROIs based on two orthogonal contrasts from non-patch trials in GLM 2 . The first ROI was defined on the basis of decreased activity in transitions where the preceding object was directly connected with the current object ( e . g . regressors corresponding to transition 1–2 , 6–4 , or 5–7 , see Figure 1—figure supplement 1C ) relative to all other transitions ( e . g . regressors corresponding to transition 4–2 , 7–4 , or 1–7; i . e . non-connected–connected ) . This contrast revealed that clusters in the bilateral entorhinal cortex show more adaptation if a preceding object is connected with a currently presented object , relative to a situation where the preceding object is 2 or 3 links away ( green in Figure 2B and Figure 2—figure supplement 2B ) . This defined ROI 1 ( thresholded at p<0 . 01 ) , from which we then extracted parameter estimates for each of the 42 no-patch transitions and tested for an orthogonal distance effect , namely whether activity differed for transitions of distance 2 relative to transitions of distance 3 using a two-tailed paired t-test Figure 2C . In a second independent test , we defined a bilateral entorhinal ROI based on the following contrast: [transitions with 3 links between the relevant objects] − [transitions with 2 links between the relevant objects] . This contrast is orthogonal to the first contrast and identified brain regions that responded more strongly on a trial if the preceding object was 3 links rather than 2 links away ( red in Figure 2B and Figure 2—figure supplement 2C ) . Again , we extracted parameter estimates for each of the 42 non-patch onset regressors from ROI 2 and performed an orthogonal test for distant-dependent scaling by investigating whether activity in this region was also significantly different for directly connected versus non-connected objects using a two-tailed paired t-test ( e . g . transition 1–2 , 6–4 , or 5–7 versus transition 4–2 , 7–4 , or 1–7 in Figure 1—figure supplement 1C ) , see Figure 2D . Note that the distance-dependent scaling effects cannot be explained by object-specific differences in activity within these ROIs . While the mean activity for different objects differs slightly , but non-significantly ( Figure 3—figure supplement 1A , C , E ) , removing these main effects by subtracting the mean activity for each object before performing the above-described analyses does not alter the results ( Figure 3—figure supplement 1B , D , F ) . In a further independent test of the distance-dependent scaling of activity in the hippocampal–entorhinal system , we extracted parameter estimates from a ROI defined based on an independent study investigating the representation of a geocentric goal direction in the entorhinal/subicular region ( ROI 3 , Chadwick et al . , 2015 ) . Specifically , we extracted parameter estimates for the 42 non-patch transitions from the peak voxel reported in his study ( MNI coordinates: [−20 , –25 , −24] ) . This definition of a ROI was non-biased and allowed us to test directly for distance-dependent scaling of activity . We first performed a repeated-measures analysis of variance and post-hoc planned two-tailed paired t-tests on the parameter estimates sorted according to distance ( Figure 2E ) . To investigate whether information is organised with respect to the distance relationship or with respect to the average time that passed between the occurrence of two objects during training , we performed a multiple linear regression . In this regression analysis , we included one regressor denoting the distance between object pairs on the graph ( 1 , 2 , and 3 ) and a second regressor accounting for the average number of objects that had occurred between any pair of objects i and j during training . Since the duration of object presentations and the ITI during training were constant , this measure was directly proportional to the time elapsed between the occurrence of the two objects . The dependent variable in the regression analysis was the neural activity for the 42 non-patch transition regressors extracted from this independently defined peak voxel ( ROI 3 , Figure 3A ) . To assess the significance across subjects , we performed two-tailed paired t-tests on the regression coefficients . To test for the directionality of the distance effect in the entorhinal cortex , we exploited the fact that subjects were not exposed to transitions between connected objects in the two directions ( e . g . 5 followed by 3 vs . 3 followed by 5 ) equally often . To assess the variability in the number of times a transition was experienced in one versus the other direction during training , we defined an asymmetry index as:a=\| xy−yx \|xy+yx where xy corresponds to the number of times object y was preceded by object x during training and yx corresponds to the number of times object x was preceded by object y during training . An asymmetry index of 0 corresponds to perfect symmetry ( i . e . the transition was experienced equally often in both directions ) , whereas an asymmetry index of 1 corresponds to maximal asymmetry ( i . e . the transition was only ever experienced in one direction ) . Across subjects and transitions , there was large variability in the asymmetry index ( Figure 3B ) . We could exploit this variability to test for non-directionality in the neural signature , which is a feature of a map-like structure . We converted the number of times each transition was experienced into a distance measure for each individual subject according to the following equation:d=1− c1+cmax Here , d denotes the length of the shortest path between two connected objects . It is computed based on the number of times this particular transition was experienced during training ( c ) relative to the number of times the most visited transition was experienced ( cmax ) . The length of the path between objects that were two or three links away was then computed as the single-source shortest path between these objects ( by adding the path-lengths for connected objects linking these two objects and choosing the shortest one ) . To compute the ‘symmetric shortest path measure’ , the directional path-length measures ( e . g . 5–2 and 2–5 ) were averaged . The directional and the symmetric shortest path measures were used as regressors to predict the neural signal extracted from the peak voxel in ROI 3 ( Figure 3C ) . To assess the significance across subjects , we performed two-tailed paired t-tests on the regression coefficients . To visualise the representation of the graph structure in the entorhinal cortex , we performed MDS on the neural activity extracted from the same peak voxel ( ROI 3 ) . MDS arranges objects spatially such that the distances between them in space correspond to their similarities as defined by the distance matrix as well as possible . Here , we estimated the configuration of objects in two dimensions using the corresponding inbuilt Matlab function 'mdscale' . Specifically , MDS was performed on a matrix denoting the mean neural activity across subjects for each pair of transitions . For example , element 2–5 in the matrix corresponded to the average activity across subjects on trials where object 5 was preceded by object 2 , and element 5–2 corresponded to the average activity across subjects on trials where object 2 was preceded by object 5 . Because neural activity scales with distance , this matrix effectively corresponds to a distance or similarity matrix . Note that MDS can only be performed on symmetric matrices with positive entries . We therefore normalised the matrix by subtracting the minimum value of the matrix and adding 1 , and then symmetrised it by averaging the top and the bottom triangles . We tested the map-like representation for significance by comparing the Euclidian distances resulting from projecting our raw data into a two-dimensional space using MDS to a null distribution of graph structures generated by permuting the links . Specifically , the null distribution was generated by keeping the nodes in the location identified by the MDS , but then permuting the seven links making up the graph structure to random nodes . Only complete graphs were included in the null distribution , that is , graphs where each node was connected to each other node , either directly or indirectly . This results in 68 , 295 unique graphs . We then computed link distances for each graph and correlated the resulting link distance measure with the distances resulting from performing MDS on the average fMRI response . This provided us with a null distribution to which we could compare the correlation between the actual graph’s link distance and the MDS-mapped data Figure 3—figure supplement 1A . As a second test of the mapping , we computed the number of line crossings in this null distribution . A two-dimensional map is characterised by the fact that there are no line crossings between pairs of directly connected nodes . In the null distribution , this is only true for 13 . 17% of all graphs Figure 3—figure supplement 1B ) . We repeated all analyses reported in Figure 3 for parameter estimates extracted from an anatomically defined ROI comprising the entorhinal cortex and subiculum ( Figure 3—figure supplement 3 ) . A separate group of 26 subjects ( aged 19–31 years , mean age ± standard deviation 24 . 9 ± 3 . 7 years , 10 males ) participated in a behavioural version of the experiment . Day 1 of the behavioural experiment was designed to be the same as day 1 of the fMRI experiment , with subjects performing 10 ( n = 14 ) or 12 ( n = 12 ) blocks of the object orientation cover task on object sequences generated according to a random walk along the graph structure . On day 2 , subjects performed the same object orientation cover task on the reduced set of objects presented to subjects participating in the fMRI experiment in the scanner . Trial transitions were pseudo-randomised to ensure that each object was preceded by each other object the same number of times . This enabled us to test for changes in response times with distance between objects on the graph . Subjects performed 10 blocks of test trials with self-paced breaks in between blocks , with 127 objects presented in each block . Thereby , each stimulus–stimulus transition was probed three times per block , or 30 times across the experiment as a whole . All analyses were performed on log-transformed response times in order to normalise response time measures . To account for object-specific effects that are independent from any distance-dependent scaling , we subtracted the mean response time for each object and subsequently computed average demeaned response times for each of the 42 stimulus–stimulus transitions per block ( object 1 preceded by object 2; object 1 preceded by object 3 , … , object 7 preceded by object 6 ) . We averaged these measures across blocks to obtain one representative measure per transition and subject . To test for scaling of response times with distance on the graph , we regressed communicability , link distance , and Euclidian distance for each of the 42 transitions onto subject-specific response time measures . The significance of the regression across subjects was assessed using two-tailed paired t-tests on the resulting regression coefficients ( Figure 5C ) . To visualise the relationship between communicability and response times , we sorted the data according by communicability , created seven bins with an equal number of transitions in each bin , and plotted the mean log response times across subjects for each bin ( Figure 5D ) .	To help us navigate , the brain encodes information about the positions of landmarks in space in a series of maps . These maps are housed by two neighbouring brain regions called the hippocampus and entorhinal cortex . These regions also encode information about non-spatial relationships , for example , between two events that often occur close together in time . However , it was not known whether such non-spatial relationships may also be encoded as a map . To address this question , Garvert et al . showed volunteers a series of objects on a screen . Unbeknown to the volunteers , the order of the objects was not entirely random . Instead , each object could only follow certain others . The objects were thus connected to one another by a network of non-spatial relationships , broadly comparable to the spatial relationships that connect physical locations in the environment . The next day , the volunteers viewed some of the objects again , this time while lying inside a brain scanner . Although the volunteers still believed that the objects had been presented at random , the activity of their hippocampus and entorhinal cortex reflected the non-spatial relationships volunteers had experienced between the objects . The relationships were organised in an abstract map . This suggests that the brain organises knowledge about abstract non-spatial relationships into maps comparable to those used to represent spatial relationships . The brain can use these maps of non-spatial relationships to guide our behaviour , even though we have no conscious awareness of the information they contain . The maps may also enable us to make new inferences , just as we can use our spatial maps to find short cuts or navigate around obstacles . Future studies should investigate the mechanisms underlying our ability to create maps of non-spatial relationships and how we use them to guide decision making .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2017	A map of abstract relational knowledge in the human hippocampal–entorhinal cortex
The visual responses of vertebrates are sensitive to the overall composition of retinal interneurons including amacrine cells , which tune the activity of the retinal circuitry . The expression of Paired-homeobox 6 ( PAX6 ) is regulated by multiple cis-DNA elements including the intronic α-enhancer , which is active in GABAergic amacrine cell subsets . Here , we report that the transforming growth factor ß1-induced transcript 1 protein ( Tgfb1i1 ) interacts with the LIM domain transcription factors Lhx3 and Isl1 to inhibit the α-enhancer in the post-natal mouse retina . Tgfb1i1-/- mice show elevated α-enhancer activity leading to overproduction of Pax6ΔPD isoform that supports the GABAergic amacrine cell fate maintenance . Consequently , the Tgfb1i1-/- mouse retinas show a sustained light response , which becomes more transient in mice with the auto-stimulation-defective Pax6ΔPBS/ΔPBS mutation . Together , we show the antagonistic regulation of the α-enhancer activity by Pax6 and the LIM protein complex is necessary for the establishment of an inner retinal circuitry , which controls visual adaptation . The retina is a primary sensory tissue that receives light stimulus and converts it into electrical signals , which are then sent to the brain for further processing . After light detection by rod and cone photoreceptors , the first step in visual processing occurs in bipolar cells that are either stimulated or inhibited by light-absorbed photoreceptors ( Masland , 2012 ) . The activities of bipolar cells are then tuned by horizontal cells while they receive visual input from the photoreceptors and by amacrine cells while they deliver the signals to retinal ganglion cells ( RGCs ) ( Hoon et al . , 2014; Masland , 2012 ) . The amacrine cells do not simply convey the signals from bipolar cells , but they also invert the signals by releasing inhibitory neurotransmitters such as γ-aminobutyric acid ( GABA ) and glycine . Therefore , even subtle changes in the composition and connectivity of amacrine cell subsets might alter the output of the retina , modifying the visual information sent to the brain . The neurons of the vertebrate retina develop in an ordered fashion from multipotent retinal progenitor cells ( RPCs ) ( Cepko , 2014 ) . A number of transcription factors with precise temporal and spatial expression patterns control the composition of retinal neurons via the hierarchical and reciprocal regulation of other transcription factor expression ( Zagozewski et al . , 2014 ) . Thus , the alterations of transcription factors that specify retinal neuron subtypes should modify visual output of mature retina . Those transcription factors include Pax6 in amacrine cells ( Marquardt et al . , 2001 ) , Vsx2 in bipolar cells ( Liu et al . , 1994 ) , Otx2 in bipolar cells and photoreceptors ( Koike et al . , 2007; Nishida et al . , 2003 ) , and Lhx2 and Sox2 in Müller glia and certain amacrine subtypes ( de Melo et al . , 2012; Gordon et al . , 2013; Lin et al . , 2009 ) . These transcription factors are not only expressed in the earlier optic structures to play critical roles in the eye and brain development ( Danno et al . , 2008; Glaser et al . , 1994; Yun et al . , 2009 ) , but also in the mature retinal neurons to support the survival and functions of the neurons ( de Melo et al . , 2012; Kim et al . , 2015 ) . However , the mechanisms underlying the recurrent expression of transcription factors in the retinal lineage are still largely unknown . Pax6 is one of the earliest transcription factors expressed in the eye field , and as such , it is considered as a master regulator of eye development ( Ashery-Padan and Gruss , 2001; Hanson and Van Heyningen , 1995 ) . Pax6 contains two DNA-binding domains—a paired domain ( PD ) and a homeodomain ( HD ) —linked via a glycine-rich domain , and activates target gene transcription through its C-terminal proline- , serine- , and threonine-rich ( PST ) domain ( Epstein et al . , 1994; Xu et al . , 1999a ) . Multiple cis-regulatory elements govern Pax6 expression in various mouse tissues ( Kammandel et al . , 1999; Xu et al . , 1999b ) . The ‘α-enhancer’ , located within intron 4 of the Pax6 gene , is active in the retina from embryo to adult ( Kammandel et al . , 1999; Marquardt et al . , 2001; Plaza et al . , 1995 ) . This retina-specific enhancer activity sustains in RPCs in the peripheral retina of the embryos and regulates neuronal differentiation in a context-dependent manner ( Marquardt et al . , 2001 ) . In the mature eye , the α-enhancer is active in cells of the ciliary body and amacrine cells of the retina ( Marquardt et al . , 2001 ) . The α-enhancer contains multiple binding sites for transcription factors , including the auto-stimulatory Pax6 ( Kammandel et al . , 1999 ) , the stimulatory Msx1 ( Kammandel et al . , 1999 ) and Pou4f2 ( Plaza et al . , 1999 ) , and the inhibitory Pax2 ( Kammandel et al . , 1999; Schwarz et al . , 2000 ) and Vax1 ( Mui et al . , 2005 ) . Although the inhibition of α-enhancer activity by Vax1 has been shown to be crucial for the development of the retina-optic stalk border ( Mui et al . , 2005 ) , the roles the other transcription factors that bind the α-enhancer in the retina remain unclear . In this study , we show that regulation of Pax6 expression through the α-enhancer fine tunes amacrine cell subtype composition , and consequently , the visual output of the retina . According to DNase footprinting ( DF ) results , the Pax6 α-enhancer contains four retina-specific transcription factor-binding sites called DF1–4 ( Plaza et al . , 1995 ) . It also contains an auto-regulatory Pax6 binding sequence ( PBS; Figure 1A ) . The AT-rich region designated DF4 recruits both positive and negative regulators expressed in the optic vesicle and embryonic retina ( Lakowski et al . , 2007; Mui et al . , 2005; Plaza et al . , 1999; Schwarz et al . , 2000 ) . Still , the transcription factors responsible for regulating α-enhancer activity in the post-natal retina are not yet known . 10 . 7554/eLife . 21303 . 003Figure 1 . Identification of Lhx3 and Tgfb1i1 as Pax6 α-enhancer binding proteins . ( A ) ( Top ) The genomic structure of the mouse Pax6 gene . Exons are shown as boxes , and arrows denote transcription initiation sites . ( Bottom ) The DF3 , PBS , and DF4 sequences in the retina-specific α-enhancer are indicated with their core homeodomain ( HD ) and paired domain ( PD ) binding sites colored red . ( B ) Nuclear extracts from R28 rat retinal precursor cells were incubated with DF4 dsDNA oligomers with single-stranded 5’- ( GT ) 5-3’ overhangs . DF4 oligomer-protein complexes were then added to Sepharose 6B columns conjugated with single-stranded DNA ( ssDNA ) of 5’- ( CA ) 5-3’ , which is complementary to the single-stranded overhang sequence of the oligomer , or 5’- ( TG ) 5-3’ non-specific binding control . Proteins bound to the ssDNA column were eluted for SDS-PAGE and detected by silver staining . Protein bands specifically enriched in the ( CA ) 5 column were then eluted from the gel and digested for mass spectrometric identification . This analysis identified the two bands marked by arrows as Lhx3 and Tgfb1i1 . ( C ) Lhx3 and Tgfb1i1 expression in post-natal day 8 ( P8 ) Pax6 α-enhancer::Cre-IRES-GFP ( P6α-CreiGFP ) mouse retinas stained with rabbit anti-Lhx3 ( top ) and anti-Tgfb1i1 ( bottom ) antibodies ( red ) . These were also co-stained with a chick anti-GFP antibody ( green ) . Scale bars , 100 μm . ( D ) DNA fragments bound to Pax6 , Lhx3 , and Tgfb1i1 in P7 mouse retinas were isolated by chromatin immunoprecipitation ( ChIP ) using rabbit polyclonal antibodies against each protein . The relative enrichment of each protein on the ectoderm enhancer and the α-enhancer of Pax6 gene was determined by PCR amplification of each enhancer sequence from the ChIP DNA fragments . ( E ) qPCR threshold cycle ( Ct ) values for each ChIP sample were compared to those of a protein-A bead only sample to obtain relative expression ( 2-ΔCt ) . The graph shows the ratio of 2-ΔCt values for each sample to those of a pre-immune rabbit IgG ( Rb-IgG ) ChIP sample . Error bars indicate standard deviations ( STD , n = 5 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 00310 . 7554/eLife . 21303 . 004Figure 1—figure supplement 1 . Lhx3 and Tgfb1i1 expression in embryonic and mature mouse retinas . E14 . 5 and P30 P6α-CreiGFP mouse retinas stained with anti-Lhx3 ( A ) and anti-Tgfb1i1 ( B ) antibodies . Lhx3 is absent in E14 . 5 mouse retinas but expressed in bipolar cell subsets in post-natal ( P8 , Figure 1C ) and adult ( P30 ) mouse retinas . Tgfb1i1 is absent in E14 . 5 and P30 mouse retinas , but is expressed in P8 mouse retina ( Figure 1C ) . The specificity of anti-Tgfb1i1 antibody was confirmed by staining P30 Tgfb1i1-ko mouse retinas ( bottom ) . Scale bars , 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 00410 . 7554/eLife . 21303 . 005Figure 1—figure supplement 2 . Binding abilities of Lhx3 and Tgfb1i1 to Pax6 α-enhancer sequence . ( A ) P7 retinal nuclear extracts were incubated with either the wild-type DF4 ( DF4-WT ) dsDNA oligomers used in Figure 1B or mutant DF4 dsDNA oligomers ( DF4-Mut ) in which the homeobox core binding sequence ATTA was replaced with CGGC . Proteins captured by the ( CA ) 5 ssDNA column were eluted for SDS-PAGE and Western blot ( WB ) analyses detecting Lhx3 and Tgfb1i1 . Arrows indicate specific bands and the asterisk marks a non-specific band . ( B ) To evaluate direct binding of Lhx3 and Tgfb1i1 to DF4 sequence in the Pax6 α-enhancer , we performed an EMSA with biotin-labeled DF4 dsDNA oligomers ( Bio-DF4 ) pre-incubated with in vitro translated Lhx3 and Tgfb1i1 . ( C ) Lhx3 binding to the conserved homeodomain binding sequence in DF4 was measured by adding unlabeled competitor DNA ( DF4 ( WT-Comp ) or mutated DF4 ( Mut-Comp , ATTA to CGGC ) ) at 1- , 10- , 100- , and 200-fold the concentration of the Bio-DF4 probe . The asterisk marks non-specific bands . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 00510 . 7554/eLife . 21303 . 006Figure 1—figure supplement 3 . Relationship between LIM domain transcription factor expression and Pax6 α-enhancer activity in mouse retina . ( A ) P14 P6α-CreiGFP mouse retinas stained with rabbit antibodies recognizing LIM domain transcription factors ( LIM-TF ) , Isl1 , Lhx2 , Lhx3 , and Lhx9 , and a mouse antibody recognizing GFP , which represents Pax6 α-enhancer activity . Images in the bottom row are magnified versions of the dotted areas in the top row . Scale bars , 100 μm . ( B ) Population of Pax6 α-GFP-positive cells co-expressing each LIM domain transcription factor in total LIM-TF-expressing cells ( red bars ) or in total GFP-expressing cells ( green bars ) were obtained and shown in a graph . Error bars represent standard deviations ( STD; n = 4 , three litters ) . ( C and D ) EMSA performed with biotin-labeled dsDNA probes for the Pax6 α-enhancer DF4 ( Bio-DF4; C ) or DF3 ( Bio-DF3; D ) sequences . Unbound free DNA probes and LIM domain protein-bound DNA probes are indicated by arrows . An asterisk indicates a non-specific protein-bound probe band . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 006 In a proteomic screen for DF4-binding proteins in R28 rat RPCs , we identified Lhx3 ( LIM domain homeobox 3 ) and Hic-5 ( hydrogen peroxide induced clone 5 ) /Tgfb1i1 ( tumor growth factor-β1 induced transcript one protein ) /Ara55 ( androgen receptor-associated protein 55 ) as potential candidates ( Figure 1B; see Materials and methods for details ) . These proteins share the LIM ( LIN-11 , Isl1 , or MEC-3 ) protein-protein interaction domain ( Karlsson et al . , 1990; Way and Chalfie , 1988 ) . In addition , Lhx3 contains a homeodomain and acts as a transcription factor ( Bridwell et al . , 2001; Roberson et al . , 1994 ) . Tgfb1i1 has four leucine-rich domains ( LDs ) , which mediate interactions with other LD-containing protein , and four LIM domains , which mediate both self-oligomerization and interactions with other LIM domain-containing proteins ( Mori et al . , 2006; Nishiya et al . , 1999 ) . Lhx3 is absent from the embryonic mouse retina , but is expressed in bipolar cells beginning around the first post-natal week ( Figure 1C , top; Figure 1—figure supplement 1A ) ( Balasubramanian et al . , 2014 ) . Tgfb1i1 is expressed in most of post-natal retina , but is absent from the embryonic and adult mouse retinas ( Figure 1C , bottom; Figure 1—figure supplement 1B ) . We also noticed Lhx3- and Tgfb1i1-expressing cells in P8 retinas show no Pax6 α-enhancer activity ( Figure 1C ) , as visualized by an GFP reporter in Pax6 α-enhancer::Cre-IRES-GFP ( P6α-CreiGFP ) mice ( Marquardt et al . , 2001 ) . This suggests a potential negative relationship between these LIM domain proteins and the α-enhancer activity . To validate our screening results , we further examined the binding of those LIM-domain containing proteins in P7 retinal nuclear extracts to DF4 sequence , and found Lhx3 and Tgfb1i1 in these extracts bind wild-type DF4 dsDNA ( DF4-WT ) but not mutant DF4 dsDNA ( DF4-MUT ) in which the 5’-ATTA-3’ homeodomain target sequence is replaced with 5’-CGGC-3’ ( Figure 1—figure supplement 2A ) . Not only the endogenous Lhx3 but also in vitro-translated Lhx3 specifically binds the DF4 oligomer ( Figure 1—figure supplement 2B ) . In vitro-translated Tgfb1i1 , however , lacks a DNA-binding motif , and so does not bind the DF4 oligomer ( Figure 1—figure supplement 2B ) . This suggests Tgfb1i1 binds the α-enhancer indirectly , possibly via an interaction with another DF4-binding protein like Lhx3 . To determine whether Lhx3 and Tgfb1i1 bind the α-enhancer in vivo , we performed a chromatin immunoprecipitation ( ChIP ) analysis using rabbit polyclonal antibodies raised against Lhx3 or Tgfb1i1 . We checked the ChIP DNA fragments isolated from P7 retinas for two mouse Pax6 gene sequences located in the ectodermal enhancer of the 5’-UTR and the α-enhancer of intron 4 using PCR ( Figure 1D ) and quantitative PCR ( qPCR; Figure 1E ) . Since both of these enhancer elements include auto-regulatory Pax6 binding sequences ( Aota et al . , 2003; Kammandel et al . , 1999 ) , we used ChIP DNA fragments obtained with anti-Pax6 rabbit IgG ( α-Pax6 ) as a positive control and those obtained with pre-immune rabbit IgG ( RbIgG ) as a negative control . We found that , in the mouse retina , Lhx3 and Tgfb1i1 interact specifically with the α-enhancer but not the ectodermal enhancer ( Figure 1D , E ) . As other LIM domain-containing transcription factors can target the same DNA sequences as Lhx3 ( Gehring et al . , 1994 ) , we also determined whether other LIM domain transcription factors expressed in the post-natal retina , such as Islet-1 ( Isl1 ) and Lhx2 ( Balasubramanian et al . , 2014 ) ( Figure 1—figure supplement 3A ) , also can bind the Pax6 α-enhancer DF4 sequence . We found Lhx2 , but not Isl1 , shows specific binding to the DF4 sequence ( Figure 1—figure supplement 3C ) . Isl1 instead binds DF3 , which contains the predicted Isl1 binding sequence 5’-CATTAG-3’ ( Lee et al . , 2008; Leonard et al . , 1992 ) ( Figure 1—figure supplement 3D ) . Conversely , the DF4-recognizing LIM transcription factors Lhx2 and Lhx3 do not bind the DF3 sequence ( Figure 1—figure supplement 3D ) . Collectively , these results suggest Tgfb1i1 binds the α-enhancer indirectly , possibly via an interaction with these LIM domain transcription factors . Lhx3 is expressed in cone bipolar cells but not in amacrine cells , including the Pax6 α-GFP-positive subpopulation ( Figure 1C; Figure 1—figure supplement 3A , B ) ( Balasubramanian et al . , 2014 ) . On the contrary , Lhx2 is expressed primarily in Müller glia ( Balasubramanian et al . , 2014 ) but also in amacrine cells , including those with α-enhancer activity ( Figure 1—figure supplement 3A , B ) . Lhx9 is also expressed in amacrine cells ( Balasubramanian et al . , 2014 ) , about 60% of which show Pax6 α-enhancer activity ( Figure 1—figure supplement 3A , B ) . Both Lhx2 and Lhx9 activate the Pax6 α-luciferase reporter in a dose-dependent manner ( Figure 2A , B ) . In contrast , Lhx3 and Lhx4 do not affect α-enhancer activity alone ( Figure 2A ) , but they antagonize Pax6-induced activation of the α-enhancer ( Figure 2B ) . Isl1 is expressed in ON bipolar cells and cholinergic amacrine cells ( Elshatory et al . , 2007; Galli-Resta et al . , 1997; Haverkamp et al . , 2003 ) , but not in Pax6 α-enhancer-active amacrine cells ( Figure 1—figure supplement 3A , B ) . Isl1 does not affect Pax6 α-enhancer activity alone , but it does activate the enhancer in the presence of Pax6 ( Figure 2A , B ) . Together , these results suggest LIM domain transcription factors in the mouse retina can be categorized based on how they affect Pax6 α-enhancer—some are stimulatory ( i . e . , Lhx2 and Lhx9 ) , some are inhibitory ( i . e . , Lhx3 and Lhx4 ) , and some are context-sensitive ( i . e . , Isl1 ) . 10 . 7554/eLife . 21303 . 007Figure 2 . Lhx3 and Isl1 inhibit Pax6 α-enhancer activity in a Tgfb1i1-sensitive manner . ( A ) The effects of LIM domain transcription factors on Pax6 α-enhancer activity were measured with a Pax6 α-enhancer luciferase reporter in HEK293T cells . These cells were co-transfected with DNA constructs encoding cDNAs of the indicated genes as well as the Pax6 α-luciferase reporter ( 0 . 2 μg ) . The triangles denote increasing doses of the indicated constructs ( 0 . 5 μg , 1 μg , and 2 μg ) . The relative luciferase activity of each sample was normalized to co-expressed β-galactosidase activity . ( B ) The effects of LIM domain transcription factors on Pax6-induced activation of the α-enhancer were also examined in the cells transfected with same DNA constructs used in ( A ) plus Pax6 construct ( 0 . 5 μg ) . ( C ) Regulatory effects of Tgfb1i1 and Lmo4 on Pax6 α-enhancer activity were also examined in the transfected cells as described in ( A ) and ( B ) . ( D and E ) Cooperative effects of Isl1 , Lhx3 , and Tgfb1i1 on Pax6 α-enhancer activity were examined with the indicated combinations . ( A – E ) The blue lines indicated relative luciferase activity in samples expressing only the luciferase reporter , while red lines indicate activity of samples expressing the reporter with Pax6 . The values on the Y-axes are averages . Error bars indicate STD ( n > 5 ) ; p<0 . 05 , p<0 . 01 , *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 007 Tgfb1i1 , although it is unable to bind the α-enhancer directly ( Figure 1—figure supplement 2B ) , inhibits Pax6-induced α-enhancer activity upon overexpression ( Figure 2C ) . This inhibition of the α-enhancer is even more significant when Tgfb1i1 is co-expressed with both Lhx3 and Isl1 ( Figure 2D , E ) . We hypothesized that multiple LIM domains of Tgfb1i1 allow it to form a multi-protein complex that blocks α-enhancer-dependent gene expression . To assess this , we co-expressed these LIM domain transcription factors with Lmo4 ( LIM-domain-only 4 ) , which prevents Isl1 and Lhx3 from interacting with one another or with other LIM domain-containing proteins ( Thaler et al . , 2002 ) . Lmo4 alone caused a dose-dependent increase in α-enhancer activity and potentiated Pax6-induced activation of the α-enhancer ( Figure 2C ) . In the presence of Lmo4 , Lhx3 and Isl1 cannot inhibit the α-enhancer ( Figure 2D ) . Thus , Tgfb1i1 and Lmo4 appear to oppositely regulate α-enhancer activity by antagonistically modulating the formation of the LIM domain transcription factor complex . We next used co-immunoprecipitation to determine whether Isl1 , Lhx3 , and Tgfb1i1 form a LIM protein complex in P7 mouse retina . We were able to detect Isl1 in complexes recovered using Lhx3 and Tgfb1i1 , which are also capable of precipitating one another ( Figure 3A ) . This suggests these three proteins may exist as a complex in the retina . To further examine the molecular nature of this LIM protein complex , we used combinatorial transfections of constructs encoding Lhx3 , Isl1 , and Tgfb1i1 into human embryonic kidney 293T ( HEK293T ) cells . The results of these transfections are summarized in Figure 3—source data 1 . 10 . 7554/eLife . 21303 . 008Figure 3 . Pax6 and Tgfb1i1 antagonistically regulate Isl1-Lhx3 complex formation . ( A ) Interactions between endogenous Isl1 , Lhx3 , and Tgfb1i1 in P7 mouse retinas measured by reciprocal co-immunoprecipitation ( co-IP ) and subsequent Western blotting ( WB ) with the indicated antibodies . P7 mouse retinal cell lysates were divided into two input tubes ( 1 and 2 ) in prior to the co-IP with indicated antibodies and subsequent WB detection of co-immunoprecipitated proteins . 5% of input cell lysates were used to compare the relative amount of the proteins in the retinal cell lysates used for co-IP . ( B ) Interactions between epitope-tagged Lhx3 and Isl1 , Lhx3 and Tgfb1i1 , and Isl1 and Tgfb1i1 in HEK293T cells were determined by co-IP and WB . The successful expression of each transfected cDNA was determined by WB for each protein in cell lysates ( 50 μg/lane; 5% of the co-IP input ) with the corresponding epitope-tag antibodies . Arrows indicate specific WB bands , and asterisks indicate non-specific bands . ( C and D ) The effects of Tgfb1i1 and Lmo4 on Isl1-Lhx3 complex formation in HEK293T cells . Triangles denote increasing amounts of each DNA construct ( 1 μg , 2 μg , and 4 μg ) . Interaction between Pax6 and LIM domain proteins ( E ) and effect of Pax6 on LIM domain protein complex formation ( F ) in HEK293T cells were also examined by co-IP and WB analyses . ( G ) Reciprocal effects of LIM domain proteins and Pax6 on the binding to human PAX6 α-enhancer sequence in the transfected HEK293T cells were measured by qPCR amplification of α-enhancer sequences in DNA fragments isolated by ChIP with the indicated epitope tag-specific antibodies . Relative enrichment of each protein on the α-enhancer was determined by comparing the qPCR value of the transfected samples with those produced by antibodies bind non-specifically to the enhancer element in untransfected HEK293T cells . Error bars indicate STD ( n > 5 ) ; p<0 . 05 , p<0 . 01 , p<0 . 001 . ( H ) Schematic model depicting the binding of the Pax6-Isl1 and Isl1-Tgfb1i1-Lhx3 complexes to the Pax6 α-enhancer element . HD , homeodomain; LBD , LIM-binding domain; LD , leucine-rich domain; PD , paired domain . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 00810 . 7554/eLife . 21303 . 009Figure 3—source data 1 . Protein-protein interaction between LIM proteins . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 00910 . 7554/eLife . 21303 . 010Figure 3—figure supplement 1 . Pax6 and Tgfb1i1 antagonistically regulate Isl1-Lhx3 complex formation . ( A ) Schematics for the full-length and deletion mutants of Isl1 , Lhx3 , Pax6 , and Tgfb1i1 used in these experiments . HD , homeodomain; LIM , LIM domain; LBD , LIM binding domain; LD , leucin-rich domain; PD , paired domain; PST , transactivation domain enriched in proline , sereine , and threonine . ( B – I ) 293 T cells ( ~106 ) were transfected with DNA constructs ( 10 μg total ) encoding the indicated protein fragments . Cell lysates collected at 48 hr post-transfection were incubated with antibodies against the epitope tags to immunoprecipitate each protein and its binding partners . Co-immunoprecipitated proteins were then analyzed by SDS-PAGE and subsequent WB with the indicated antibodies . In parallel , the cell lysates ( containing 50 μg protein ) were also analyzed by SDS-PAGE and WB with the indicated antibodies to examine relative levels of the overexpressed proteins in the transfected cells . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 010 In the co-immunoprecipitation experiments , we found that Isl1 binds to Lhx3 with its homeodomain ( HD ) and/or LIM binding domain ( LBD ) , as reported previously ( Thaler et al . , 2002 ) , whereas it interacts with Tgfb1i1 with its LIM domain ( s ) ( Figure 3B [left column]; Figure 3—figure supplement 1B , C ) . Lhx3 binds to Tgfb1i1 and Isl1 via its LIM domain ( s ) ( Figure 3B [center column]; Figure 3—figure supplement 1E , F ) . Tgfb1i1 also uses LIM domain ( s ) to interact with Isl1 and Lhx3 ( Figure 3B [right columns]; Figure 3—figure supplement 1H , I ) . We further tested whether Tgfb1i1 binds Lhx3 and Isl1 separately or whether they form a complex of Lhx3-Tgfb1i1-Isl1 . We found overexpressed Tgfb1i1 further enhanced the association between Lhx3 and Isl1 ( Figure 3C ) . Lmo4 , in contrast , induces a dose-dependent decrease in the association between Isl1 and Lhx3 ( Figure 3D ) . Collectively , these results suggest Tgfb1i1 links Isl1 and Lhx3 to form a hetero-tetrameric ( or larger ) complex while Lmo4 interferes with the complex formation . It is possible Pax6 interacts with the homeodomains of Isl1 and Lhx3 to form a Pax6-LIM protein complex , since Pax6 reportedly interacts with various homeodomain-containing proteins ( Granger et al . , 2006; Mikkola et al . , 2001 ) . We also found Pax6 interacts only with Isl1 , but not Lhx3 , via its paired domain ( PD ) ( Figure 3E; Figure 3—figure supplement 1G ) . Both the HD and LIM domains of Isl1 participated to interact with Pax6 , thus Pax6 might compete with Tgfb1i1 and Lhx3 to bind Isl1 ( Figure 3H , top; Figure 3—figure supplement 1D ) . The DF3 and DF4 , which are separated by an auto-regulatory PBS , are respective targets of Isl1 and Lhx3 ( Figure 1A; Figure 1—figure supplement 3C , D ) . Thus , Pax6 binding to the PBS may hinder the binding of Isl1-Tgfb1i1-Lhx3 complex to the DF3 and DF4 sequences , and vice versa . Using ChIP analyses in the cultured cells , we found Isl1 , Lhx3 , and Tgfb1i1 reduce the binding of Pax6 to the α-enhancer when all three are co-expressed but not when expressed individually ( Figure 3G , three right graphs ) . Conversely , Pax6 expression interferes with the access of Tgfb1i1 to the α-enhancer ( Figure 3G , rightmost graph ) . Pax6 does not affect Lhx3 binding to the α-enhancer , but it promotes Isl1 binding ( Figure 3H , two center graphs ) . Together , these molecular interaction results suggest two different transcription factor complexes occupying the α-enhancer region . The Isl1-Pax6 complex binds to the DF3 and PBS and activates the α-enhancer ( Figure 3H , top ) , whereas the Isl1-Tgfb1i1-Lhx3 complex binds to DF3 and DF4 to cover the area between those two sequences and inhibit the access of Pax6 to DF3 ( Figure 3H , bottom ) . We , next , examined the Pax6 α-enhancer activity by detecting Pax6 α-GFP-positive cells in P14 Tgfb1i1−/− mice in comparison to Tgfb1i1+/+ ( wild-type , WT ) littermates to validate the molecular mechanism proposed by our in vitro data in vivo . In support of the idea that Tgfb1i1 plays an important role in the inhibition of Pax6 α-enhancer , the Tgfb1i1−/− mouse retinas show more cells positive for the Pax6 α-GFP reporter than the retinas of their Tgfb1i1+/+ littermates ( Figure 4A , B ) . We also examined the retinal composition of those littermate mice , and we only observed differences in the amacrine and bipolar cell populations among the major retinal cell types ( Figure 4C–F; other retinal cell types are not shown ) . Tgfb1i1−/− mouse retinas have more Pax6-positive amacrine cells and fewer Vsx2-positive bipolar cells than Tgfb1i1+/+ littermates ( Figure 4C–F ) . 10 . 7554/eLife . 21303 . 011Figure 4 . Elevated GABAergic amacrine cell number in Tgfb1i1−/−mouse retinas . ( A ) Pax6 α-enhancer-active cells in P14 Tgfb1i1+/+ and Tgfb1i1−/−littermate mouse retinas were visualized by immunodetection of GFP expressed from the P6α-CreiGFP transgene . ONL , outer nuclear layer; INL , inner nuclear layer; GCL , ganglion cell layer . ( B ) GFP-positive cell population in 250 μm x 250 μm retinal area . ( C ) P14 Tgfb1i1+/+ and Tgfb1i1−/−littermate mouse retinas stained with antibodies detecting amacrine cell subtype-specific markers . Pax6 , pan-amacrine cells; ChAT , cholinergic amacrine cells; GlyT1 , glycinergic amacrine cells; GABA , Gad67 and Bhlhb5 ( in the bottom half of INL in the images in E ) , GABAergic amacrine cells . ( D ) Fold-changes of amacrine cell numbers in P14 Tgfb1i1−/− retinas compared to Tgfb1i1+/+ littermate retinas . ( E ) P14 Tgfb1i1+/+ and Tgfb1i1−/−mouse retinas stained for bipolar cell-specific markers . Vsx2 , pan-bipolar cell marker; PKCα , rod bipolar cells; G0α , rod and ON-cone bipolar cells; Vsx1 , OFF bipolar cells; Recoverin , photoreceptors ( in the ONL ) and type-2 OFF bipolar cells ( in the INL ) ; Bhlhb5 ( in the top half of INL ) , type-2 OFF bipolar cells . ( F ) Fold-changes in marker-positive cell numbers in Tgfb1i1−/− retinas compared to Tgfb1i1+/+ littermate retinas . Values on the Y-axes of B , D , and F are averages . Error bars indicate STD ( n = 4 , three litters ) ; p<0 . 05; p<0 . 01; p<0 . 001 . Scale bars in the pictures , 100 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 01110 . 7554/eLife . 21303 . 012Figure 4—figure supplement 1 . Elevation of Pax6 α-enhancer-active GABAergic amacrine cells in Tgfb1i1−/−mouse retinas . ( A ) P14 Tgfb1i1+/+;P6α-CreiGFP and Tgfb1i1−/−;P6α-CreiGFP littermate mouse retinas co-stained with antibodies against amacrine cell subtype markers and GFP . Pax6 , pan-amacrine cell marker; ChAT , cholinergic; GlyT1 , glycinergic; Gad67 , GABAergic; GABA , GABAergic subsets; Bhlhb5 , GABAergic subsets ( bottom of the INL ) . Outset images in the bottom row are magnified versions of the dotted box areas in the top row . Scale bar , 100 μm . ONL , outer nuclear layer; INL , inner nuclear layer; GCL , ganglion cell layer . ( B ) Populations of GFP-positive cells co-expressing amacrine cell subset markers are shown in a graph . Values on the Y-axis are averages . Error bars indicate STD ( n = 4 , three litters ) . p<0 . 05; *p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 01210 . 7554/eLife . 21303 . 013Figure 4—figure supplement 2 . Deletion of Lhx3 in the post-natal mouse retina . ( A ) To delete Lhx3 in the post-natal mouse retina , we designed two independent sgRNAs complementary to the sequences near the translation initiation site in the exon2 ( highlighted in red ) , following the suggestion of the CRISPR Design server ( http://crispr . mit . edu ) . The sequences were cloned into pX330 ( pX330-U6-Chimeric_BB-CBh-hSpCas9 ) DNA construct , which express the cloned sgRNA and Cas9 endonuclease . ( B ) P14 mouse retinas , which were electroporated with the indicated pX330 DNA constructs at P0 , were stained for the detection of various amacrine cell markers , including Pax6 ( pan-amacrine ) , Gad67 ( GABAergic ) , GABA ( GABAergic subsets ) , Bhlhb5 ( GABAergic subsets , bottom half of the INL ) , and GlyT1 ( glycinergic ) , and bipolar cell markers , including Vsx2 ( pan-bipolar ) , G0α ( ON bipolar ) , Vsx1 ( OFF bipolar ) , and Bhlhb5 ( type-2 OFF bipolar , top half of the INL ) , as well as for EGFP , which is expressed from co-electroporated pCAGIG DNA construct . Thus , EGFP-positive retinal cells are expected to express sgRNA and Cas9 from the indicated pX330 DNA constructs . Successful loss of Lhx3 in the mouse retinas was examined by immunostaing of Lhx3 . Scale bar , 100 μm . ( C ) Ratio of marker-positive cells to total INL cells of each sample was then compared with that of pX330+pCAGIG ( Mock ) sample . ( D ) The population of EGFP-positive cells co-expressing each amacrine or bipolar cell type-specific marker in total EGFP-positive INL cells were obtained and shown in a graph . Scores on the Y-axis of the graphs in C and D are averages ( n = 6 , two independent batches ) . Error bars indicate STD ( n = 6 , two independent batches ) ; p<0 . 05; p<0 . 01; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 013 We classified amacrine cells positive for Gad67 ( glutamate decarboxylase 67 kDa ) , GABA , and Bhlhb5 ( basic helix-loop-helix domain containing , class B , 5 ) as GABAergic; cells positive for ChAT ( choline acetyl transferase ) as cholinergic; and cells positive for GlyT1 ( glycine transporter 1 ) as glycinergic . Among those amacrine cell subtypes , only GABAergic amacrine cells showed a significant increase , while the numbers of cholinergic and glycinergic amacrine cells remain unchanged ( Figure 4C , D ) . Moreover , the Pax6 α-GFP-positive cells are mainly GABAergic amacrine cells ( Figure 4C , D; Figure 4—figure supplement 1 ) , suggesting a positive relationship between Pax6 α-enhancer activity and GABAergic amacrine cell fate and a negative role of Tgfb1i1 in this . Among bipolar cell subtypes , the Tgfb1i1−/− mouse retinas show fewer Vsx1-positive OFF bipolar cells without significant changes in G0α-positive ON bipolar cells , including PKCα-positive rod bipolar cells ( Figure 4E , F ) . However , the numbers of type-2 OFF bipolar cells , which are positive to Bhlhb5 and Recoverin , are not greatly different between Tgfb1i1+/+ and Tgfb1i1−/− mouse retinas ( Figure 4C–F; Figure 4—figure supplement 1 ) . The results therefore suggest that Tgfb1i1 is necessary for the development of Vsx1-positive OFF bipolar cells , except for type-2 subset , in mouse retina . We also tried to investigate the roles of Lhx3 in the post-natal mouse retina , which cannot develop in Lhx3-deficient mice that die perinatally ( Sheng et al . , 1996 ) . We , thus , electroporated DNA constructs encoding Cas9 endonuclease and single guide RNA ( sgRNA ) targets to mouse Lhx3 sequence , together with the pCAGIG DNA construct expressing EGFP ( enhanced green fluorescent protein ) , into P0 mouse retinas ( Figure 4—figure supplement 2A; see Materials and methods for details ) . We then examined the fates of EGFP-positive cells , which supposedly co-express Cas9 and the Lhx3 sgRNA , in the mouse retinas at P14 . We found GABAergic amacrine cell identities of the retinal cells expressing the constructs were significantly enhanced , whereas OFF bipolar cell identities of the cells were remarkably diminished ( Figure 4—figure supplement 2B–D ) . Collectively , our results suggest that Tgfb1i1 supports the development of OFF bipolar cell subsets , while it antagonizes the development of GABAergic amacrine cell subset , by forming Lhx3-containing protein complex that inhibits Pax6 α-enhancer activity in post-natal mouse retina . In P14 mouse retinas , Lhx3-positive bipolar cells co-expressing Isl1 comprise only 20% of total Lhx3-positive cells ( Figure 5—figure supplement 1B , C ) . In contrast , 82% of Lhx3-positive retinal cells co-express Isl1 at P7 , which is when Tgfb1i1 is expressed in most retinal cell types apart from amacrine cells ( Figure 5—figure supplement 1A , C ) . This suggests the Isl1-Tgfb1i1-Lhx3 complex may form in the retinal cells around the first post-natal week at the peak of bipolar cell development ( Morrow et al . , 2008; Rapaport et al . , 2004 ) . Supporting this , the interaction between Isl1 and Lhx3 is significantly reduced in P7 Tgfb1i1−/− retinas ( Figure 5A , top ) . This might trigger an over-activation of Pax6 transcription , driven by the α-enhancer . 10 . 7554/eLife . 21303 . 014Figure 5 . Pax6 α-enhancer-induced Pax6ΔPD isoform supports GABAergic amacrine cell fate . ( A ) Reciprocal co-IP and WB analyses with the indicated antibodies reveal a reduced interaction between Isl1 and Lhx3 in P7 Tgfb1i1−/−mouse retinas compared with littermate Tgfb1i1+/+retinas ( top two WB images ) . Tgfb1i1−/−retinal lysates show 1 . 6-fold higher Isl1 level than Tgfb1i1+/+retinal lysates and no significant change in the levels of Lhx3 and Actinβ1 ( bottom four WB images ) . ( B ) No significant difference in the assembly of Isl1 and Pax6 was observed in P7 Tgfb1i1+/+and Tgfb1i1−/−littermate mouse retinas ( top two WB images ) . Tgfb1i1−/− retinas show higher expression of the Pax6ΔPD isoform than Tgfb1i1+/+retinas and no change in full-length Pax6 ( bottom two WB images ) . ( C ) Pax6 α-enhancer-active cells were isolated from P14 P6α-CreiGFP retinas by repeated FACS ( see the Materials and Methods ) . Lysates of GFP ( + ) and GFP ( - ) retinal cells were then analyzed by SDS-PAGE and WB with a rabbit anti-Pax6 antibody . Successful purification of the cells was confirmed by WB detection of GFP in each fraction . ( D ) Diagram of pCAGIG DNA constructs encoding V5-tagged Pax6 ( pCAGIG-V5-Pax6 ) and Pax6ΔPD ( pCAGIG-V5-Pax6ΔPD ) . These constructs express EGFP from an IRES linked to the V5-Pax6 or V5-Pax6ΔPD cDNAs . This allowed for the confirmation of successful expression of the cDNAs in in P7 mouse retinas electroporated with the indicated pCAGIG DNA constructs at P0 by WB detection of EGFP and V5 . ( E ) Co-expression of V5-Pax6ΔPD and EGFP in P7 mouse retinas was also determined by immunostaining with mouse anti-V5 ( red ) and chick anti-GFP ( green ) antibodies . ( F ) The identities of EGFP-positive retinal cells co-expressing Pax6 or Pax6ΔPD in P14 mouse retinas were determined by staining with antibodies against various amacrine and bipolar cell-specific proteins . The images are mouse retinal sections stained with anti-GABA ( top ) and anti-Vsx1 ( bottom ) antibodies . Arrowheads indicate cells positive to both of EGFP and the markers . Additional immunostaining results are provided in Figure 5—figure supplement 2 . ( G ) EGFP-positive cells co-expressing each cell type-specific marker are shown as a percentage of total EGFP-positive INL cells . Values on the Y-axis are averages . Error bars indicate STD ( n = 5 ) ; p<0 . 01; *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 01410 . 7554/eLife . 21303 . 015Figure 5—figure supplement 1 . Distribution of Isl1- and Lhx3-expressing cells in Tgfb1i1+/+ and Tgfb1i1−/− mouse retinas . P7 ( A ) and P14 ( B ) Tgfb1i1+/+ and Tgfb1i1−/− littermate mouse retinas stained with a guinea pig anti-Isl1 antibody ( green ) and a rabbit anti-Lhx3 antibody ( red ) . Images in the right columns are magnified versions of the dotted areas in the left columns . Scale bars , 100 μm . Isl1 ( - ) ( red ) and Isl1 ( + ) ( yellow ) cells among Lhx3 ( + ) cells are shown in the graph in ( C ) and populations expressing each marker in total INL cells are shown in the graph in ( D ) . Y-axis values in the graphs are averages and error bars indicate STD ( n = 4 , three independent litters ) . p<0 . 05; *p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 01510 . 7554/eLife . 21303 . 016Figure 5—figure supplement 2 . Ectopic expression of Pax6 isoforms in the post-natal mouse retinas . ( A ) P14 mouse retinas , which had been electroporated with the indicated DNA constructs at P0 , were stained for the detection of various amacrine cell markers , including Syntaxin ( pan-amacrine ) , Gad67 ( GABAergic ) , GABA ( GABAergic subsets; results are in Figure 5G ) , Bhlhb5 ( GABAergic subsets , bottom half of the INL ) , ChAT ( cholinergic ) , and GlyT1 ( glycinergic ) . EGFP cDNA is linked to the Pax6 cDNAs via IRES , thus those two cDNAs are transcribed in a single mRNA . Thus , the cells expressing EGFP together with the amacrine cell markers can be counted to investigate the effects of overexpressed Pax6 isoforms on retinal cell fate determination . Scale bar , 100 μm . ( B ) The retinas were also stained for the detection of bipolar cell markers Vsx2 ( pan-bipolar ) , G0α ( ON bipolar ) , Vsx1 ( OFF bipolar; results are in Figure 5G ) , Recoverin ( type-2 OFF bipolar ) , and Bhlhb5 ( type-2 OFF bipolar , top half of the INL ) . Scale bar , 100 μm . ( C ) Retinal layer distribution of EGFP-positive cells in the indicated electroporated mouse retinas . ( D ) EGFP-positive cells co-expressing each amacrine or bipolar cell type-specific marker are shown as a percentage of total EGFP-positive INL cells . Scores on the Y-axis in the graphs in ( C ) and ( D ) are averages . Error bars indicate STD ( n = 6 , four independent batches ) ; p<0 . 05; p<0 . 01; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 016 However , Pax6 levels did not differ in Tgfb1i1−/− and Tgfb1i1+/+ mouse retinas ( Figure 5B , larger Pax6 bands ) , despite the increase of Pax6-positive cells in Tgfb1i1−/− mouse retina ( Figure 4C , D ) . We did , instead , observe a specific increase in the level of Pax6ΔPD isoform , which is an alternative transcript produced at downstream of the α-enhancer sequence ( Lakowski et al . , 2007; Plaza et al . , 1995 ) , in the Tgfb1i1−/− mouse retina ( Figure 5B , smaller Pax6 bands ) . This Pax6ΔPD isoform is selectively enriched in Pax6 α-GFP-positive cells purified from P7 P6α-CreiGFP retinas by fluorescence activated cell sorting ( FACS ) ( Figure 5C ) . Our results , thus , suggest hyperactivation of the Pax6 α-enhancer in Tgfb1i1−/− retinas triggers ectopic expression of Pax6ΔPD isoform , but not the canonical Pax6 . We , next , investigated the role of α-enhancer-driven Pax6ΔPD expression in retinal cell fate determination by overexpressing Pax6ΔPD , which is connected with EGFP by internal ribosome entry site ( IRES ) , in post-natal mouse retina ( Figure 5D–G; Figure 5—figure supplement 2 ) . About 54% of EGFP-positive INL cells in P14 mouse retinas , which were electroporated with the pCAGIG-Pax6ΔPD DNA at P0 , are identified as Syntaxin-positive amacrine cells , whereas only 26% of EGFP-positive INL cells are amacrine cells in the retinas electroporated with control pCAGIG DNA ( Figure 5—figure supplement 2A [top row] , D ) . This is also contrary to the results of pCAGIG-Pax6-electroporated mouse retinas , in which about 85% of EGFP-positive INL cells are identified as amacrine cells ( Figure 5—figure supplement 2A [top row , middle] , D ) . Moreover , by showing insignificantly different marker positivity with EGFP;Syntaxin double-positive INL cells ( 54% ± 7 . 56% ( Syntaxin ) vs . 46% ± 7 . 33% ( Gad67 ) ) , majority of EGFP-positive amacrine cells in the pCAGIG-Pax6ΔPD-electroporated retinas are predicted as GABAergic amacrine cells , which are approximately half of the EGFP;Syntaxin double-positive amacrine cell population in pCAGIG-Pax6-electroporated mouse retinas ( 85% ± 7 . 2% ( Syntaxin ) vs . 44% ± 9 . 17% ( Gad67 ) ) ( Figure 5F , G; Figure 5—figure supplement 2A [second row] , D ) . The populations of EGFP-positive cholinergic and glycinergic amacrine cells in pCAGIG-Pax6ΔPD-electroporated mouse retinas are not greatly different from those in pCAGIG-electroporated mouse retinas , but are lower than those in pCAGIG-Pax6-electroporated mouse retinas ( Figure 5—figure supplement 2A [bottom two rows] , D ) . Together , these results suggest that Pax6ΔPD preferentially supports GABAergic amacrine cell fate , while full-length Pax6 induces all amacrine cell types in a similar ratio observed in the normal mouse retina ( Voinescu et al . , 2009 ) . Mouse retinas expressing ectopic Pax6ΔPD show almost no EGFP-positive cells co-expressing OFF bipolar cell markers including Vsx1 , Recoverin , and Bhlhb5 ( Figure 5F [bottom row] , G; Figure 5—figure supplement 2B , D ) . On the contrary , significant numbers of EGFP-positive cells co-expressed ON bipolar cell marker G0α in pCAGIG-Pax6ΔPD-electroporated mouse retinas , and the numbers are not significantly different from those in pCAGIG-electroporated samples ( Figure 5—figure supplement 2B , D ) . EGFP-positive cells in pCAGIG-Pax6-electroporated mouse retinas , however , are almost absent of both ON and OFF bipolar cell marker co-expression ( Figure 5—figure supplement 2B , D ) . The results therefore suggest that Pax6ΔPD inhibits only OFF bipolar cell fate , while full-length Pax6 suppress both ON and OFF bipolar cell fates . Next , to inactivate the α-enhancer , we generated Pax6ΔPBS/ΔPBS mice by deleting the auto-stimulatory PBS in the α-enhancer using the CRISPR/Cas9 system ( Figure 6A; see Materials and methods for details ) . Despite the Pax6 α-enhancer being active in the mouse retina from embryo to adult , the gross morphologies of Pax6ΔPBS/ΔPBS mouse eyes are indistinguishable from Pax6+/+ WT eyes ( Figure 6B ) , implicating dispensable roles of Pax6 α-enhancer-induced Pax6ΔPD expression in the eye and retinal development . However , in P14 Pax6ΔPBS/ΔPBS retinas , the α-enhancer-driven GFP and Pax6ΔPD expression are reduced significantly , but not entirely abolished ( Figure 6B–D ) . Since Pax6 does not bind and activate the Pax6ΔPBS α-enhancer ( Figure 6—figure supplement 1 ) , this suggests the presence of positive regulator ( s ) of the α-enhancer in the mouse retina other than Pax6 . 10 . 7554/eLife . 21303 . 017Figure 6 . Pax6-dependent Pax6 α-enhancer activation is positively correlated with GABAergic amacrine cell number . ( A ) Genomic DNA was isolated from the tails of Pax6+/+ ( left ) and Pax6ΔPBS/ΔPBS ( right ) mice for sequencing the Pax6 α-enhancer region . The Pax6 binding sequence ( PBS ) in the α-enhancer is colored red . The Pax6ΔPBS allele is missing six nucleotides ( 5’-TGCATG-3’ ) in the PBS . ( B ) Whole eye images of P30 Pax6+/+;P6α-CreiGFP and Pax6ΔPBS/ΔPBS;P6α-CreiGFP littermate mice ( left ) and the mouse eye sections stained with H&E ( center ) or an anti-GFP antibody ( right ) . Scale bar in the rightmost column is 100 μm . ( C ) Pax6 α-GFP-positive cells in P30 Pax6+/+and Pax6ΔPBS/ΔPBS retinas ( 250 μm x 250 μm ) . Error bars indicate STD ( n = 4 , two independent litters ) . ( D ) Full-length Pax6 and Pax6ΔPD in P14 Pax6+/+ and Pax6ΔPBS/ΔPBS retinal cell lysates were detected by WB with anti-Pax6 antibody and WB band intensities were compared to show the relative values below the WB image . ( E ) Distributions of pan-amacrine cell marker Pax6 , GABAergic amacrine cell subset marker GABA , pan-bipolar cell marker Vsx2 , and OFF bipolar cell marker Vsx1 in P14 Pax6+/+and Pax6ΔPBS/ΔPBS littermate retinas were visualized with immunostaining with antibodies recognizing respective markers . Scale bars , 100 μm . Additional images of amacrine and bipolar cell subtypes are shown in Figure 6—figure supplement 2 . ( F ) Quantification of relative numbers of amacrine and bipolar cell subsets in mouse retinas . Error bars indicate STD ( n = 5 , three independent litters ) . p<0 . 05; p<0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 01710 . 7554/eLife . 21303 . 018Figure 6—figure supplement 1 . Impaired response of the Pax6ΔPBS α-enhancer to Pax6 . ( A ) Luciferase expression at downstream of a Pax6 α-enhancer mutant lacking its PBS ( Pax6-αΔPBS ) was measured by detecting chemiluminescence emitted from the lysates of HEK293T cells combinatorially expressing Pax6 , Lhx3 , Isl1 , and Tgfb1i1 ( n = 4 ) . Bindings of Pax6 , Isl1 , Lhx3 , and Tgfb1i1 to the Pax6 α-enhancer sequence in P30 Pax6+/+ and Pax6ΔPBS/ΔPBS mouse retinas were assessed by qPCR ( B , n = 4 ) and PCR ( C ) amplification of DNA fragments isolated by ChIP with a rabbit IgG recognizing each respective protein . Error bars indicate STD; p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 01810 . 7554/eLife . 21303 . 019Figure 6—figure supplement 2 . Distribution of amacrine and bipolar cell subsets in Pax6+/+ and Pax6ΔPBS/ΔPBS mouse retina . P14 Pax6+/+ and Pax6ΔPBS/ΔPBS littermate retinas co-stained with amacrine cell and bipolar cell subtype marker-specific antibodies . Gad67 , GABAergic amacrine cells; ChAT , cholinergic amacrine cells; GlyT1 , glycinergic amacrine cells; PKCα , rod bipolar cells; Bhlhb5 , OFF bipolar cells and GABAergic amacrine cells . Scale bar , 100 μm . Quantification results are shown in Figure 6F . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 01910 . 7554/eLife . 21303 . 020Figure 6—figure supplement 3 . Fate determination of GABAergic amacrine cells and OFF bipolar cells in the post-natal mouse retinas . ( A ) The effects of deletions of Tgfb1i1 ( Tgfb1i1−/− ) and PBS sequence of Pax6 α-enhancer ( Pax6ΔPBS/ΔPBS ) on GABAergic amacrine cell development were investigated by immunostaining of various GABAergic amacrine cell markers , including Gad67 , GABA , and Bhlhb5 . Distribution of entire amacrine cells was examined by immunostaining of pan-amacrine cell marker Syntaxin . The effects of the gene deletions on Pax6 α-enhancer activity was also determined by detecting cells expressing Pax6 α-GFP . Scale bars , 100 μm ( top ) and 50 μm ( rest ) . ( B ) Relative numbers of marker-positive cells in P4 Tgfb1i1−/−and Pax6ΔPBS/ΔPBS mouse retinas are determined by comparing with those in their WT littermate mice . Error bars denote STD ( n = 4 , two independent litters ) . p<0 . 05 . ( C ) To identify the fate of cells were born in WT , Tgfb1i1−/− , and Pax6ΔPBS/ΔPBS between post-natal day 4 and 7 ( P4 and P7 ) when bipolar cells and Müller glia are predominantly generated , the mice were repeatedly injected with BrdU ( 5 mg/kg ) at P4 , P5 , and P6 . Eye sections of the BrdU-injected mice were obtained at P14 for the immunodetection of Bhlhb5-positive GABAergic amacrine cells and Vsx1-positive OFF bipolar cells , which had exited cell cycle after incorporating BrdU between P4 and P7 . Scale bar , 50 μm . ( D ) To trace the fates of cells produced in the embryonic retina when amacrine cells are generated , pregnant mice were injected with BrdU ( 5 mg/kg ) at 15 dpc ( E15 ) and the identities of cells had exited cell cycle after incorporating BrdU were examined at P7 . Scale bar , 50 μm . ( E and F ) BrdU-labeled cell population in Bhlhb5-positive GABAergic amacrine cells , which locate the bottom half of INL , and that in Vsx1-positive OFF bipolar cell population in P14 mouse retinas as ( C ) and P7 mouse retina as ( D ) are quantified . Values in the Y-axis are average and error bars denote STD ( n = 4 , two independent litters ) . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 020 P14 Pax6ΔPBS/ΔPBS retinas show significantly fewer GABAergic amacrine cells than the retinas of Pax6+/+ WT littermates , despite similar total numbers of Pax6-positive amacrine cells ( Figure 6E [left two columns] , F; Figure 6—figure supplement 2 ) . Conversely , Pax6ΔPBS/ΔPBS retinas show more OFF bipolar cells ( i . e . , Vsx1-positive ) , despite similar total numbers of Vsx2-positive bipolar cells ( Figure 6E [right two columns] , F; Figure 6—figure supplement 2 ) . However , the numbers of GABAergic amacrine cells , which start to develop in the embryonic retina ( Voinescu et al . , 2009 ) , were not significantly different between Pax6+/+ and Pax6ΔPBS/ΔPBS retinas until P4 when the bipolar cells start to develop ( Figure 6—figure supplement 3A [bottom three rows] , B ) . The results therefore suggest that Pax6-dependent activation of Pax6 α-enhancer is not essential for the embryonic development of GABAergic amacrine cells but it might be necessary for the development and/or maintenance of those cells in the post-natal retina . To test a possibility of antagonistic fate determination of newborn retinal neurons between GABAergic amacrine and OFF bipolar cell subsets in the post-natal mouse retinas , we repeatedly injected bromodeoxyuridine ( BrdU ) to WT , Tgfb1i1−/− , and Pax6ΔPBS/ΔPBS mice to label GABAergic amacrine and OFF bipolar cells , which were born between post-natal day 4 and 7 . We failed to find BrdU;Bhlhb5 double-positive GABAergic amacrine cells in P14 WT , Tgfb1i1−/− , and Pax6ΔPBS/ΔPBS mouse retinas , suggesting the lack of newborn GABAergic amacrine cells in the post-natal mouse retinas ( Figure 6—figure supplement 3C , E ) . Furthermore , the number of BrdU;Vsx1 double-positive OFF bipolar cells in those mouse retinas were not significantly different each other ( Figure 6—figure supplement 3C [bottom row] , D ) , despite the remarkable decrease and increase of total Vsx1-positive cell numbers in P14 Tgfb1i1−/− and Pax6ΔPBS/ΔPBS in mouse retinas , respectively ( Figure 4E , F; Figure 6E , F ) . The results therefore suggest that the alteration of OFF bipolar cells in those two mutant mouse retinas was not caused by neurogenic fate changes of newborn retinal cells but may have resulted from fate change of preexisting retinal cells . We , thus , traced the fates of retinal cells born in the embryonic retina by injecting BrdU to pregnant female mice at 15 dpc ( days post coitum ) . The numbers of Bhlhb5;BrdU-labeled GABAergic amacrine cells are significantly decreased in P7 Pax6ΔPBS/ΔPBS mouse retinas in comparison to their WT littermate mouse retinas ( Figure 6—figure supplement 3D [top row] , F ) . Conversely , Vsx1;BrdU-labeled OFF bipolar cell numbers are significantly increased in the Pax6ΔPBS/ΔPBS mouse retinas ( Figure 6—figure supplement 3D [bottom row] , F ) . Taken together , these results suggest Pax6 and the Isl1-Tgfb1i1-Lhx3 complex in the post-natal mouse retina competitively regulate the Pax6 α-enhancer-driven expression of Pax6ΔPD , which maintains GABAergic amacrine cell fate against the transdifferentiation into OFF bipolar cells . We next determined whether these changes in the Pax6 α-enhancer-active ( P6α ) GABAergic amacrine cell and OFF bipolar cell numbers influence visual responses in Tgfb1i1−/− and Pax6ΔPBS/ΔPBS mice . Using the OptoMotry system ( Prusky et al . , 2004 ) , we observed a significant reduction in visual acuity of P60 Tgfb1i1−/− mice compared to age-matched WT and Pax6ΔPBS/ΔPBS mice ( Figure 7A , graph ) . Upon the measurement of light response of a whole retina by electroretinogram ( ERG ) , the amplitudes for the a- and b-waves in dark-adapted ( scotopic ) and light-adapted ( photopic ) ERG responses of P60 Tgfb1i1−/− and Pax6ΔPBS/ΔPBS mouse eyes were , however , unaltered in comparison to those of WT littermate controls ( Figure 7—figure supplement 1 ) . The results suggest that the functions of photoreceptors ( determined by ERG a-waves ) and ON bipolar cells ( determined by ERG b-waves ) are intact in those mutant mice . In support of this , the numbers of photoreceptors and ON bipolar cells in P60 Tgfb1i1−/− and Pax6ΔPBS/ΔPBS mouse retinas were not significantly different from those in their littermate WT mice ( Figure 7—figure supplement 2 ) . Therefore , the reduced visual acuity of Tgfb1i1−/− mice might be caused by either the changes of visual pathway components in the brain or the alterations of amacrine cells and RGCs at downstream of bipolar cells . 10 . 7554/eLife . 21303 . 021Figure 7 . Pax6 α-enhancer-active amacrine cells are important for visual adaptation . ( A ) Visual acuity was measured in P60 mice using the OptoMotry system as previously described ( Prusky et al . , 2004 ) ( for details , see the Materials and Methods ) . Error bars indicate STD ( n = 6 ) . p<0 . 01 . ( B ) Peristimulus time histograms ( PSTHs ) for RGCs in P60 Tgfb1i1+/+ and Tgfb1i1−/− littermate and P60 Pax6+/+ and Pax6ΔPBS/ΔPBS littermate mouse retinas were obtained by multielectrode array ( MEA ) recordings . Maximum and mean numbers of spike were counted from each PSTH . Insets are representative PSTH patterns . Arrowhead indicates the sustained light-ON responses of RGCs . Maximum ( max , C ) and mean ( D ) numbers of spikes were counted from each PSTH . The numbers on the Y-axis are averages ( WT , n = 526 ( in four mice ) ; Tgfb1i1−/− , n = 534 ( in six mice ) ; Pax6+/+ , n = 175; Pax6ΔPBS/ΔPBS , n = 276 ) . Error bars indicate STD . Statistical significance was determined using the D’Agostino and Pearson omnibus normality test followed by one-way ANOVAs and Sidak’s test for multiple comparisons . p<0 . 05; **p<0 . 001 . ( E ) Visual detection in P60 Tgfb1i1−/− , Pax6ΔPBS/ΔPBS , and their WT littermate mice trained to lick water in response to light stimuli . The experimental scheme and task learning curves are provided in Figure 7—figure supplement 3A and B ( for details , see the Materials and Methods ) . ( F ) The mice were also given water in association with a continuous light stimulus ( 2 s ) but not with a continuous light stimulus ( 1 s ) followed by a drifting grate stimulus ( 1 s ) ( see the experimental scheme and task learning curves in Figure 7—figure supplement 3C and D ) . Visual responses were quantified as ratios of hit rates ( HitR , Go ) to false alarm rates ( FAR , Nogo ) . Error bars in ( E ) and ( F ) indicate STD . p<0 . 05; p<0 . 01; p<0 . 001 ( Unpaired t-test ) . ( G ) Diagram depicting the modulation of retinal circuitry important for visual adaptation by Pax6 α-enhancer-active ( P6α ) GABAergic amacrine cells . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 02110 . 7554/eLife . 21303 . 022Figure 7—figure supplement 1 . ERGs of mouse retinas . P60 WT , Tgfb1i1−/− , and Pax6ΔPBS/ΔPBS mice were dark-adapted for 16 hr . Then , their scotopic ERG responses were assessed at a light intensity of 2 . 5 cds ( left ) . Average amplitudes of scotopic ERG a-waves and b-waves measured from WT ( white bars , n = 8 ) , Tgfb1i1−/− ( gray bars , n = 6 ) , and Pax6ΔPBS/ΔPBS ( black bars , n = 4 ) eyes . Photopic ( center ) and flicker ( right ) ERG responses of these mice were also measured after adaptation under room light ( 30 cd/m2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 02210 . 7554/eLife . 21303 . 023Figure 7—figure supplement 2 . Cell composition of P60 WT , Tgfb1i1−/− , and Pax6ΔPBS/ΔPBS mouse retinas . ( A ) Composition of P60 WT , Tgfb1i1−/− , and Pax6ΔPBS/ΔPBS mouse retinas were determined by examining cell type-specific markers . Rhodopsin , rod photoreceptors; M-opsin , M-cone photoreceptors; Calbindin , horizontal cells ( HZ; arrowheads ) ; Vsx2 , bipolar cells ( BP ) ; Sox2 , Müller glia ( MG; arrowheads ) ; Pax6 , amacrine cells ( AC ) ; Brn3b , retinal ganglion cells ( RGCs ) ; glial fibrillary acidic protein ( Gfap ) , astrocytes ( AS ) . Scale bar , 100 μm . ( B ) Relative numbers of marker-positive cells in P60 Tgfb1i1−/−and Pax6ΔPBS/ΔPBS mouse retinas were determined by comparing with those in their WT littermate mice . Error bars denote STD ( n = 4 , two independent litters ) . p<0 . 05; p<0 . 01 . ( C ) Distribution of amacrine cell subtypes in P60 WT , Tgfb1i1−/− , and Pax6ΔPBS/ΔPBS mice were determined by examining cell type-specific markers . ChAT , cholinergic; GlyT1 , glycinergic; Gad67 and Bhlhb5 ( AC in the bottom half of INL ) , GABAergic . Scale bar , 100 μm . ( D ) Relative numbers of marker-positive cells in P60 Tgfb1i1−/−and Pax6ΔPBS/ΔPBS mouse retinas were determined by comparing with those in their WT littermate mice . Error bars denote STD ( n = 4 , two independent litters ) . p<0 . 01; **p<0 . 001 . ( E ) Distribution of bipolar cell subtypes in P60 WT , Tgfb1i1−/− , and Pax6ΔPBS/ΔPBS mice were determined by examining cell type-specific markers . PKCα , rod bipolar cell; Vsx1 and Bhlhb5 ( BP in the top of INL in ( C ) ) , OFF bipolar cells . Scale bar , 100 μm . ( F ) Relative numbers of marker-positive cells in P60 Tgfb1i1−/−and Pax6ΔPBS/ΔPBS mouse retinas were determined by comparing with those in their WT littermate mice . Error bars denote STD ( n = 4 , two independent litters ) . p<0 . 05; p<0 . 01; *p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 02310 . 7554/eLife . 21303 . 024Figure 7—figure supplement 3 . Experimental scheme assessing mouse visual responses . ( A ) P60 mice were trained to associate water rewards with flashing light stimuli . Correct and incorrect lick rates were used to measure visual detection . ( B ) Lick rates during the learning period for mice responding to various intensity of light as shown in Figure 4E . ( C ) P60 mice were trained to associate water rewards only with a continuous ( 2 s ) light stimulus and not a continuous ( 1 s ) light followed by a drifting grate image ( 1 s ) . Correct and incorrect lick rates were used to measure visual discrimination of the drifting grate from various intensities of light stimulus . ( D ) Lick rates during the learning period . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 024 We , therefore , measured the light-evoked activity of individual retinal circuits by performing multi-electrode array ( MEA ) recordings of RGCs , which represent the final circuit component in the retina . We found an increase in the basal firing rate and mean spike number , but no change in maximum spike rate for the light-ON responses of P60 Tgfb1i1−/− retinas when compared to WT littermate controls ( Figure 7B [top] , C ) . Conversely , we observed a reduction in the basal firing rate , maximum spike rate , and mean spike number for the ON responses of P60 Pax6ΔPBS/ΔPBS retinas ( Figure 7B [bottom] , D ) . Interestingly , a significant number of RGCs in P60 Tgfb1i1−/− retinas do not return to the resting state after a transient light response ( Figure 7B , arrowhead ) . Considering the GABAergic identity of P6α amacrine cells and the increase of the cells in the Tgfb1i1−/− retinas ( Figure 4C , D; Figure 4—figure supplement 1 ) , it suggests this specific amacrine cell subset might disinhibit ON response by acting to other inhibitory retinal neurons in light-ON pathway ( Chávez et al . , 2010; Demb and Singer , 2012; Eggers et al . , 2013 ) . The low visual acuity and sustained light response of Tgfb1i1−/− mouse retinas suggest that the hypersensitivity of the mice to light interferes with their detection of dark objects on brighter backgrounds ( Figure 7A , predicted views ) . To test this hypothesis , we presented the mice with two different types of visual stimuli . First , we trained dark-adapted mice to associate a water reward with a flashing light stimulus . Then , we counted correct water-licking events in response to various intensities of flash light ( Figure 7—figure supplement 3A ) . Tgfb1i1−/− mice not only learn this task faster than WT mice ( Figure 7—figure supplement 3B ) , but they also show more sensitive detection of the light stimuli ( Figure 7E ) . However , in a second visual task requiring mice to detect a drifting grate stimulus after a light stimulus , Tgfb1i1−/− mice perform worse than WT mice ( Figure 7F; Figure 7—figure supplement 3C , D ) . This suggests Tgfb1i1−/− retinas are more slowly re-sensitized after light exposure than WT retinas . Conversely , the re-sensitization of Pax6ΔPBS/ΔPBS retinas is significantly faster than WT retinas , despite being less sensitive to light ( Figure 7F , G; Figure 7—figure supplement 3 ) . These results are consistent with our MEA recordings , which showed sustained and transient light responses in Tgfb1i1−/− and Pax6ΔPBS/ΔPBS RGCs , respectively ( Figure 7B–D ) . Collectively , the results suggest that P6α amacrine cells control the tone of light-ON retinal pathway . The overall tone of light-ON pathway was increased in Tgfb1i1−/− mouse retinas , which have extra P6α amacrine cells , whereas it is decreased in Pax6ΔPBS/ΔPBS mouse retinas having reduced P6α amacrine cell number . Consequently , the light-ON pathway is augmented in Tgfb1i1−/− mouse retinas and attenuated in Pax6ΔPBS/ΔPBS mouse retinas to make the retinas hypersensitive and hyposensitive to light stimulus , respectively ( Figure 7B–E ) . On the other hand , the Tgfb1i1−/− mouse retinas cannot be re-sensitized as fast as WT retinas , while the Pax6ΔPBS/ΔPBS mouse retinas can be re-sensitized more quickly than WT retinas after a light stimulus . Given the GABAergic identity of the cells , the P6α amacrine cells may inhibit the activity of post-synaptic partners , which are not identified yet but can be predicted as an inhibitory neuron in light-ON pathway ( Figure 7F ) . Therefore , the results indicate that the proper number of P6α amacrine cells should be present in the retina to respond to light efficiently and adequately . Transcription factors frequently act in combination , allowing relatively few to generate the tremendous cellular diversity of the nervous system ( Jessell , 2000 ) . Especially , the ‘LIM code’ mixes and matches LIM domain-containing transcription factors to direct tissue- and cell-specific gene expression ( Gill , 2003; Shirasaki and Pfaff , 2002 ) . Lhx3 , for example , specifies motor neuron cell fate in the spinal cord by forming a hetero-hexameric complex with Isl1 and nuclear LIM interactor ( NLI ) for the binding to the promoter of the Mnx1/Hb9 gene , whereas it specifies V2 interneuron cell fate by forming a hetero-tetrameric complex with NLI at the promoter of the Vsx2/Chx10 gene ( Thaler et al . , 2002 ) . Given that the various LIM homeodomain transcription factors , including Lhx2 , Lhx3 , Lhx4 , and Lhx9 , share a consensus target sequence ( Gehring et al . , 1994 ) , we speculate Isl1 partners with different LIM homeodomain transcription factors in a cell-context-dependent manner . In contrast to its relationship with Lhx3 , Isl1 cooperates with Lhx2 to activate the α-enhancer in cultured cell lines ( data not shown ) . However , this is unlikely to occur in vivo , because Lhx2 and Isl1 are expressed mutually exclusively in RPCs ( Lhx2 ) and post-mitotic RGCs ( Isl1 ) of the embryonic mouse retina , in GABAergic ( Lhx2 ) and cholinergic amacrine cells ( Isl1 ) in the mature retina , as well as in Müller glia ( Lhx2 ) and ON bipolar cells ( Isl1 ) ( Balasubramanian et al . , 2014; Elshatory et al . , 2007; Gordon et al . , 2013; Pan et al . , 2008 ) . Moreover , Lhx2flox/flox;P6α-Cre retinas , which lack Lhx2 expression in the Cre-active lineages ( Gordon et al . , 2013 ) , show no change in the number of Pax6 α-enhancer-active cells ( data not shown ) . This suggests Lhx2 may be dispensable for the activation of the Pax6 α-enhancer in the mouse retina . We propose that a Tgfb1i1 dimer links Isl1 and Lhx3 to form a hetero-tetrameric complex that represses the Pax6 α-enhancer ( Figures 2 and 3 ) . The effects of Tgfb1i1 on the α-enhancer could be achieved by blocking Pax6’s access the PBS sequence ( Figure 3G , H ) . Alternatively , Tgfb1i1 may also recruit transcriptional co-repressors , such as NCoR ( nuclear receptor co-repressor ) ( Heitzer and DeFranco , 2006 ) , to the Pax6 α-enhancer . These negative effects of Tgfb1i1 on the Pax6 α-enhancer can be antagonized by Lmo4 , which is persistently co-expressed with Pax6 in the retina and interferes with the interactions between Tgfb1i1 and Lhx3 and/or Isl1 ( Duquette et al . , 2010 ) ( Figures 2D and 3D ) . Retinas lacking Lmo4 have fewer GABAergic amacrine cells than controls ( Duquette et al . , 2010 ) , which suggests Lmo4 may positively affect Pax6 α-enhancer-dependent GABAergic amacrine cell fate determination by inhibiting the formation of the LIM complex . However , the antagonistic regulation of the LIM complex by Tgfb1i1 and Lmo4 could not be applied to OFF bipolar cell fate determination , since OFF bipolar cell numbers are decreased commonly in Tgfb1i1−/− and Lmo4-cko mouse retinas . Our results suggest that Tgfb1i1 and Lmo4 might involve in the development of different OFF bipolar cell subsets . The numbers of Bhlhb5-positive OFF bipolar cell subsets were not altered significantly in Tgfb1i1−/− mouse retinas ( Figure 4E , F ) , in contrast to a significant decrease in Lmo4-cko mouse retinas . In addition to its canonical form , two alternative forms of Pax6 , Pax6 ( 5a ) and Pax6ΔPD , are produced by alternative splicing and internal transcription initiation , respectively ( Epstein et al . , 1994; Mishra et al . , 2002 ) . Pax6ΔPD does not affect Pax6 target gene expression via the conserved PBS ( data not shown ) . Instead , as previously reported ( Mikkola et al . , 2001 ) , Pax6ΔPD may potentiate the expression of Pax6 target genes by interacting with full-length Pax6 . This facilitation of Pax6-induced gene transcription by Pax6ΔPD may also occur with the Pax6 α-enhancer , resulting in a feed-forward activation of the α-enhancer . Alternatively , it may bind another promoter element containing the Pax6 homeodomain target DNA sequence ( TAATT ( /C ) NA ( /C ) ATTA ) . Therefore , future studies will be needed to identify the targets of Pax6ΔPD in RPCs and post-mitotic retinal neurons . This will provide a full understanding of the distinctive roles Pax6 and Pax6ΔPD play in the retina . Although the mechanisms of light adaptation and re-sensitization in the photoreceptors are fairly well-understood , how the inner retina contributes to these mechanisms is less clear . Acting downstream of rod bipolar cells that deliver visual signals from rod photoreceptors , A17 GABAergic amacrine cells provide a direct feedback inhibition to the rod bipolar cells ( Chávez et al . , 2010 ) . In parallel , an unidentified subset of GABAergic amacrine cells is also proposed to inhibit rod bipolar cells at the downstream of ON-cone bipolar cells , which can be activated by AII amacrine cells in the rod pathway as well as by daylight ( Demb and Singer , 2012; Eggers et al . , 2013 ) . GABAergic inhibition to the rod bipolar cells could be reduced in Tgfb1i1−/− mouse retinas , leading to sustained ON responses ( Figure 7A ) . Conversely , the ON pathway in Pax6ΔPBS/ΔPBS mouse retinas is activated more transiently and is also more readily re-activated by subsequent visual stimuli ( Figure 7D ) . Therefore , the P6α amacrine cells might attenuate those GABAergic inhibitions to rod bipolar cells and prevent premature inactivation of rod pathway . However , future studies should identify molecular and electrophysiological identities of the P6α amacrine cells and their pre- and post-synaptic partners to fully understand this visual adaptive circuits in the inner retina . cDNA clones were generous gifts from Dr . Motoko Shibanuma ( Hic-5/Tgfb1i1 ) and Dr . Seth Blackshaw ( Lhx2 and Lhx9 ) . The full-length and fragment DNAs used in this study were isolated by PCR amplification from these cDNAs . Tgfb1i1−/− mice were generated as previously described ( Kim-Kaneyama et al . , 2011 ) . Pax6ΔPBS/ΔPBS mice were generated in this study using the CRISPR/Cas9 system ( Cong et al . , 2013 ) . To prepare sgRNA constructs , the pX330 vector was obtained from Addgene and digested with BbsI for insertion of a pair of phosphorylated dsDNA oligos ( 5’-CACCGAAGTCGCTCCGGATCATGCA-3’ , 5-AAACTGCATGATCCGGAGCGACTTC-3’ ) that target the PBS in the Pax6 α-enhancer . A T7 promoter was added to the 5’ end of the sgRNA sequence in the pX330-sgRNA construct . This construct was then used as a template for in vitro transcription using the MEGAshortscript T7 kit ( Life Technologies , CA ) . The in vitro transcribed sgRNAs ( 50 ng/ml ) were injected into C57BL6/J mouse embryos ( 2 cell-stage ) together with Cas9 mRNA ( 100 ng/μl; purchased from Toolgen Inc . , South Korea ) . Then , these embryos were injected into the inner cell mass of ICR embryos . Four resulting F1 chimeric male mice were crossed to C57BL6/J female mice to obtain an F2 generation with the potential to carry deletions in the PBS . Then , tail DNA from each F2 mouse ( n = 51 ) was prepared and used as a template for the PCR-amplification of the Pax6 α-enhancer sequence . Each resulting PCR product was cloned into the pGEM-T vector for sequencing . Four F2 mice carry different heterozygous deletions in the PBS sequence were obtained . Before breeding with littermate Pax6+/ΔPBS female mice , Pax6+/ΔPBS male mice were crossed with C57BL6/J females for more than six generations to dilute any potentially OFF target mutations . All experiments using mice were performed according to the regulations of the KAIST-IACUC ( KA2012-38 ) . HEK293T ( RRID: CVCL_0063 ) and R28 retinal progenitor cells ( RRID: CVCL_5I35 ) were obtained from ATCC and a gift from Dr . Gail Seigel ( University of Rochester School of Medicine and Dentistry ) , respectively . These cell-lines are not in the list of commonly misidentified cell lines ( by the International Cell Line Authentication Committee ) . These cells were regularly checked for mycoplasma contamination . The cells were maintained in DMEM supplemented with 10% fetal bovine serum ( GIBCO , MA ) . Cells were combinatorially transfected with DNA constructs via the PEI ( polyethylenimine ) method ( Polyscience , PA ) . The PCR-amplified mouse Pax6 α-enhancer sequence was fused to the pGL3-luciferase vector ( Promega , WI ) and co-transfected with DNA constructs of interest and pSV-β-gal plasmids . Transfected cells were harvested at 24 hr after transfection , and cell extracts were assessed for luciferase activity followed by normalization using β-galactosidase activity . The ( CA ) 5 or ( TG ) 5 ssDNA oligonucleotides were coupled to CNBr-preactivated Sepharose 4B ( GE Healthcare , IL ) according to the manufacturer’s protocol . R28 cells ( ~108 ) were incubated in a low salt buffer ( 10 mM HEPES , 10 mM KCl , 0 . 1 mM EDTA , 1 mM DTT , and 0 . 5 mM PMSF ) on ice for 10 min to rupture the plasma membranes . Then , the nuclei were collected by centrifugation . After treating the isolated nuclei with 10% ( final v/v ) NP-40 for 20 min , a solution containing 20 mM HEPES , 0 . 4 M NaCl , 1 mM EDTA , 1 mM DTT , and 1 mM PMSF was added to extract nuclear protein complexes . These nuclear extracts were then pre-cleared with a 5% ( final v/v ) slurry of protein A agarose beads ( Invitrogen , CA ) in a binding buffer ( 10 mM Tris-Cl pH 7 . 5 , 0 . 4 M NaCl , 1 mM EDTA , 1 mM DTT , and 5% glycerol ) at 4°C for 30 min . The DF4 dsDNA oligonucleotides with 5’- ( GT ) 5 single-strand overhangs were synthesized with a 5’ terminal amine modification and incubated with the R28 nuclear extracts overnight at 4°C with agitation . The protein-DF4 dsDNA complexes were then incubated with the ssDNA-coupled Sepharose 4B for 6 hr at 4°C , centrifuged , washed twice in binding buffer , washed twice in wash buffer ( 10 mM Tris-Cl pH 7 . 5 , 1 . 2 mM NaCl , 1 mM EDTA , 1 mM DTT , 0 . 1% Nonidet P-40 ) , and washed twice in PBS . Protein-DF4 complexes bound to the column were then eluted in SDS sample buffer for SDS-PAGE and subsequent silver staining . The silver-stained protein bands , which were enriched in the ( CA ) 5-coupled Sepharose 4B relative to the ( TG ) 5-coupled Sepharose 4B , were isolated for trypsin digestion before being subjected to MALDI-TOF MS/MS analysis at the Korean Basic Science Institute ( KBSI ) proteomics core facility . Biotin-labeled and unlabeled dsDNA probes in binding buffer ( 75 mM NaCl , 1 mM EDTA , 1 mM DTT , 10 mM Tris-HCl ( pH 7 . 5 ) , 6% glycerol , 2 mg BSA , and 500 ng poly ( dI-dC ) ) were incubated on ice for 30 min with LIM proteins produced using the TNT Quick Coupled Transcription/Translation kit ( Promega , WI ) . The EMSA was carried out on a 6% polyacrylamide gel in 0 . 5X TBE buffer . The DNA-protein complexes were then transferred to a nylon membrane ( GE Healthcare , IL ) , and the biotin-labeled probes were detected using the Phototop-Star Detection Kit ( New England BioLabs , MA ) according to the manufacturer’s recommendations . P7 mouse retinas and transfected HEK293T cells were lysed in a buffer consisting of 10 mM Tris-HCl ( pH 7 . 4 ) , 200 mM NaCl , 1% Triton X-100 , 1% NP-40 , and a protease inhibitor cocktail ( Invitrogen , CA ) . Cell lysates were centrifuged at 12 , 000 g for 10 min at 4°C . The resulting supernatants were incubated with appropriate antibodies at 4°C for 16 hr , and then pre-washed protein A/G‐sepharose ( GE Healthcare , IL ) was added to the samples . The protein A/G‐sepharose immune complexes were washed five times with cell lysis buffer and subjected to SDS-PAGE and Western blotting ( WB ) for detection of co‐immunoprecipitated proteins . P7 mouse retinas were isolated , chopped , and cross-linked with 1% formaldehyde in PBS for 10 min at room temperature . After a 5 min incubation in 125 mM glycine , the tissues were homogenized and the nuclei were isolated . These nuclei were then subjected to sonication to break their chromatin into ~600 bp fragments in a lysis buffer containing 50 mM Tris-HCl ( pH 7 . 5 ) , 150 mM NaCl , 5 mM EDTA , 0 . 5% NP-40 , 1% Triton X-100 , and a protease inhibitor cocktail ( Invitrogen , CA ) . After pre-clearing with protein A agarose beads for 1 hr , the nuclear extracts were incubated for 16 hr with 1 μg of the appropriate antibody followed by incubation with protein A beads for 45 min at room temperature . The immune complexes were then washed three times with lysis buffer and then three more times with the same wash buffer containing 500 mM LiCl . After adding a Chelex 100 slurry to the washed beads , the DNA fragments were eluted for use as templates for qPCR . We used specific primers to amplify sequences in the ectoderm enhancer ( fp1 , 5’-CTAAAGTAGACACAGCCTT; rp1 , 5’-GGAGACATTAGCTGAATTC ) and the α-enhancer ( fp2 , 5’-GTGACAAGGCTGCCACAAGCGCC , rp2 , 5’- CCGTGTCTAGACAGAAGCCCTCTC ) of the mouse Pax6 gene . qPCR was performed using the iTaq fast SYBR Green Master Mix ( BioRad , CA ) with these same primers and analyzed using the CFX-Manager software ( Bio-Rad , CA ) . Gene expression was normalized to that of a sample containing only protein A beads . Frozen sections ( 12 μm ) of embryonic heads and post-natal mouse eyes were incubated for 1 hr in a blocking solution containing 5% normal donkey serum and 5% normal goat serum in PBS containing 0 . 2% Triton X-100 . The sections were incubated with the antibodies listed in Table 1 for 16 hr at 4°C . Fluorescent images were obtained with a confocal microscope ( Olympus FV100 and Zeiss LSM710 ) after staining with Cy3 , Alexa 647 , and Alexa 488-conjugated secondary antibodies at room temperature for 1 hr . 10 . 7554/eLife . 21303 . 025Table 1 . Antibody used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 21303 . 025Antigen Species Producer Dilution Bhlhb5GoatSanta Cruz1:100Brn3bGoatSanta Cruz1:200CalbindinMouseSigma1:200CalretininMouseMillipore1:1000 ChATGoatMillipore1:200Isl1Rabbitgift from Dr . Mi-Ryoung Song1:500Isl1Guinea Piggift from Dr . Mi-Ryoung Song1:10 , 000 Gad67MouseMillipore1:500GABAGuinea PigMillipore1:300GFAPRabbitAbcam1:500GFPChickAbcam1:200GFPRabbitSanta Cruz1:500GlyT1RabbitAbcam1:200G0αMouseMillipore1:300G/R opsinRabbitMillipore1:200Lhx2GoatSanta Cruz1:200Lhx3RabbitAbcam1:1000 Lhx9RabbitSanta Cruz1:500Pax6RabbitAbcam1:200Pax6RabbitCovance1:300PKCαMouseSigma1:200RecoverinRabbitChemicon1:200RhodopsinMouseMillipore1:500Sox2GoatSanta Cruz1:100Sox9RabbitSanta Cruz1:200Tgfb1i1 ( Hic-5 ) MouseBD1:100Tgfb1i1 ( Hic-5 ) RabbitAbcam1:100Vsx1GoatSanta Cruz1:50Vsx2 ( Chx10 ) MouseSanta Cruz1:200V5MouseGenway Biotech1:1000 P6α-CreiGFP adult mouse eyes were dissected and placed in Hank’s Balanced Salt Solution ( HBSS; Life technologies ) to remove the lens . Retinas were peeled from the eyes and placed in 1 ml HBSS containing activated 10 mg/ml papain ( Sigma-Aldrich ) for 5 min at 37°C . Retinal cells were resuspended in HBSS with 2% FCS followed by centrifugation at 1600 rpm for 2 min . Cell pellets were then gently triturated in HBSS with 2% FCS , filtered through a 70 μm Filcons membrane prior to FACS analysis . GFP-positive retinal cells were then sorted in an Aria Fusion Cell Sorter ( Becton Dickinson ) at 495 nm excitation and 519 nm emission . Following FACS analysis , cells were collected by centrifugation at 1600 rpm for 5 min and the cells were lysed in a buffer containing 10 mM Tris-HCl ( pH 8 . 0 ) , 1 mM EDTA , 1% Triton X-100 , 0 . 1% SDS , and 150 mM NaCl . Electroporation experiments were performed as previously described ( Matsuda and Cepko , 2004 ) . Approximately 0 . 5 μl ( total; 5 μg/μl ) DNA solution mixed with fast green dye was injected into the subretinal space of P0 mouse retinas , and square electric pulses were applied ( 100 V; five 50 ms pulses at 950 ms intervals ) . For CRISPR/Cas9-mediated deletion of Lhx3 gene , dsDNA oligos ( sgRNA-Lhx3-1 , 5'- ( P ) -CACCGGACCCGTCCCGGGAATCCGC-3' and 5'-AAACGCGGATTCCCGGGACGGGTCC-3'; sgRNA-Lhx3-2 , 5'-CACCGTGCTGGCGTTGTTGGCGCGA-3' and 5'-AAACTCGCGCCAACAACGCCAGCAC-3' ) were cloned into the pX330 vector before co-electroporation with the pCAGIG vector ( molar ratio of pX330 constructs to pCAGIG is 1:0 . 5 ) . Mouse retinas were cut into 3 mm x 3 mm patches in artificial cerebrospinal fluid ( ACSF ) solution ( 124 mM NaCl , 10 mM glucose , 1 . 15 mM KH2PO4 , 25 mM NaHCO3 , 1 . 15 mM MgSO4 , 2 . 5 mM CaCl2 , and 5 mM KCl ) bubbled with 95% O2 + 5% CO2 at pH 7 . 3–7 . 4°C and 32°C . Retinal patches were then mounted , ganglion cell layer down , on a planar 8 × 8 MEA , and the light-evoked RGC spikes were recorded using the MEA60 system ( Multi Channel Systems GmbH , Germany ) . White light stimuli were applied with a DLP projector ( Hewlett Packard , ep-7122 ) focused onto the photoreceptor layer of the retina through four convex lenses . Light intensity was 170–200 μW/cm2 ( 116–136 lux ) in 8–10 μW/cm2 ( 5 . 5–6 . 8 lux ) background illumination . Light stimuli were given in 1 s pulses with 6 s inter-pulse intervals to a total of 40 pulses per retina . All experiments were performed after sufficient dark adaptation ( >1 hr ) . Mouse visual acuity was measured with the OptoMotry system ( Cerebral Mechanics Inc . ) as previously described ( Prusky et al . , 2004 ) . Mice , of which genotypes are not determined before the measurement , were adapted to ambient light for 30 min and then placed on the stimulus platform , which is surrounded by four computer monitors displaying grating patterns randomly presented by the OptoMotry software . Mice that stopped moving and began tracking the gratings with reflexive head movements in concert with their rotation were counted as successful visual detection events . The detection thresholds were then obtained from the OptoMotry software .	The retina is a light-sensitive layer of tissue that lines the inside of the eye . This tissue is highly organized and comprises a variety of different nerve cells , including amacrine cells . Together , these cells process incoming light and then trigger electrical signals that travel to the brain , where they are translated into an image . Changes in the nerve cell composition of the retina , or in how the cells connect to each other , can alter the visual information that travels to the brain . The nerve cells of the retina are formed before a young animal opens its eyes for the first time . Proteins called transcription factors – which regulate the expression of genes – tightly control how the retina develops . For example , a transcription factor called Pax6 drives the development of amacrine cells . Several other transcription factors control the production of Pax6 by binding to a section of DNA known as the “α-enhancer” . However , it is not clear how regulating Pax6 production influences the development of specific sets of amacrine cells . Kim et al . reveal that a protein known as Tgfb1i1 interacts with two transcription factors to form a “complex” that binds to the α-enhancer and blocks the production of a particular form of Pax6 . In experiments performed in mice , the loss of Tgfb1i1 led to increased production of this form of Pax6 , which resulted in the retina containing more of a certain type of amacrine cell that produce a molecule called GABA . Mice lacking Tgfb1i1 show a stronger response to light and are therefore comparable to people who are too sensitive to light . On the other hand , mice with a missing a section of the α-enhancer DNA have fewer amacrine cells releasing GABA and become less sensitive to light and are comparable to people who have difficulty detecting weaker light signals . The findings of Kim et al . suggest that an individual’s sensitivity to light is related , at least in part , to the mixture of amacrine cells found in their retina , which is determined by certain transcription factors that target the α-enhancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "neuroscience" ]	2017	The LIM protein complex establishes a retinal circuitry of visual adaptation by regulating Pax6 α-enhancer activity
Experience and training have been shown to facilitate our ability to extract and discriminate meaningful patterns from cluttered environments . Yet , the human brain mechanisms that mediate our ability to learn by suppressing noisy and irrelevant signals remain largely unknown . To test the role of suppression in perceptual learning , we combine fMRI with MR Spectroscopy measurements of GABA , as fMRI alone does not allow us to discern inhibitory vs . excitatory mechanisms . Our results demonstrate that task-dependent GABAergic inhibition relates to functional brain plasticity and behavioral improvement . Specifically , GABAergic inhibition in the occipito-temporal cortex relates to dissociable learning mechanisms: decreased GABA for noise filtering , while increased GABA for feature template retuning . Perturbing cortical excitability during training with tDCs alters performance in a task-specific manner , providing evidence for a direct link between suppression and behavioral improvement . Our findings propose dissociable GABAergic mechanisms that optimize our ability to make perceptual decisions through training . Understanding the structure of the world around us entails extracting and discriminating meaningful patterns from cluttered environments . Effortless as this may seem , it poses for the brain a challenging task that involves suppressing noisy and ambiguous sensory signals . Experience and training have been shown to facilitate perceptual judgments and visual recognition processes ( Fine and Jacobs , 2002; Gilbert et al . , 2001; Goldstone , 1998 ) . For instance , an experienced bird watcher is not only able to break the camouflage and detect a bird in a leafy tree , but also determine whether it is a carrion crow or a hooded crow . Yet , the mechanisms that the human brain employs to suppress task-irrelevant information and optimize perceptual decisions through training remain largely unknown . Theoretical models of perceptual learning ( Dosher et al . , 2013; Dosher and Lu , 1998; Li et al . , 2004 ) posit that experience and training facilitate our ability to a ) detect targets in clutter by filtering external noise , b ) discriminate highly similar objects by suppressing irrelevant features and retuning task-relevant feature templates . Although considerable behavioral evidence supports this framework , its neural implementation remains uncertain . Previous fMRI studies have demonstrated changes in the overall activation of higher visual areas in the occipito-temporal cortex due to training on perceptual decision tasks ( for reviews , see Kourtzi , 2010; Welchman and Kourtzi , 2013 ) . However , fMRI data do not allow us to discern excitatory from suppressive mechanisms of experience-dependent plasticity , as BOLD reflects aggregate activity across large neural populations ( Heeger and Ress , 2002; Logothetis , 2008 ) . Here , we ask whether GABA ( γ-aminobutyric acid ) , the primary inhibitory neurotransmitter in the brain , mediates our ability to improve in making perceptual decisions through training . Previous work in animals has demonstrated that GABAergic inhibition is associated with learning and synaptic plasticity ( Castro-Alamancos et al . , 1995; Trepel and Racine , 2000 ) . Yet , measuring GABA in the human brain has been possible only recently thanks to advances in MR Spectroscopy ( MRS ) . Previous MRS studies have shown that GABA concentrations in the visual cortex relate to homeostatic plasticity ( Lunghi et al . , 2015 ) , while GABA concentrations in the motor cortex relate to individual ability ( Kolasinski et al . , 2017; Stagg et al . , 2011a ) and improved performance ( Blicher et al . , 2015; Floyer-Lea et al . , 2006; O'Shea et al . , 2017; Sampaio-Baptista et al . , 2015 ) in motor learning . To probe the mechanisms that the human brain uses to suppress noisy and irrelevant signals , we employed two tasks that rely differentially on noise filtering vs . template retuning: ( 1 ) a signal-in-noise task that involves extracting a target masked by noise , ( 2 ) a feature-differences task that involves judging fine differences . Interrogating only fMRI signals does not allow us to discern between the brain mechanisms for noise filtering vs . template retuning: our results show similar learning dependent changes in behavioral performance and fMRI activation during training in both learning tasks . However , combining MRS measurements of GABA with fMRI uncovers distinct GABAergic inhibition mechanisms in the posterior occipito-temporal cortex . Learning to detect targets from clutter by noise filtering relates to decreased GABA , while learning to discriminate fine differences by template retuning relates to increased GABA . To move beyond correlative evidence , we then used transcranial direct current stimulation ( tDCs ) in the occipito-temporal cortex to perturb cortical excitability during training . Our findings relating GABAergic inhibition and behavioral improvement lead to opposite predictions for the effect of tDCs on the two learning tasks . In line with these predictions , we find dissociable effects of tDCs stimulation on behavioral improvement between tasks: excitatory anodal stimulation enhances learning to see in clutter , while inhibitory cathodal stimulation enhances learning feature differences . Thus , perturbing visual cortex suppression alters behavioral improvement during training in a task-specific manner , providing evidence for a direct link between suppression and learning . Our findings propose that GABAergic inhibition in the visual cortex underlies dissociable learning mechanisms that optimize our ability to make perceptual decisions . We tested two separate groups of participants on either ( 1 ) a signal-in-noise ( SN ) task that involves extracting shapes ( radial vs . concentric Glass patterns ) from background noise or ( 2 ) a feature-differences ( FD ) task that involves judging fine differences induced by morphing between the two stimulus classes ( Figure 1a ) . For each task , participants improved during a single training session that took place during scanning ( Figure 1b ) , consistent with previous reports showing fast behavioral improvement early in the training ( for a review see Sagi , 2011 ) . A repeated measures ANOVA ( Task ( SN vs . FD ) x Training ( training runs ) showed significantly improved performance –as measured by d’– after training ( main effect of Training: F ( 6 , 192 ) = 3 . 79 , p=0 . 001 ) but no significant effect of Task ( F ( 1 , 32 ) =0 . 01 , p=0 . 91 ) nor Training x Task interaction ( F ( 6 , 192 ) = 0 . 61 , p=0 . 722 ) , suggesting similar improvement in both tasks . Testing participants the following day after training ( transfer test ) showed that performance was significantly different from the first training run for both tasks ( main effect of Session: ( F ( 1 , 32 ) = 10 . 59 , p=0 . 003; Task x Session interaction: ( F ( 1 , 32 ) =0 . 89 , p=0 . 35 ) but not significantly different from the last training run ( main effect of Session: F ( 1 , 32 ) =0 . 95 , p=0 . 34; Task x Session interaction: F ( 1 , 32 ) =0 . 18 , p=0 . 68 ) , suggesting lasting performance improvement due to training . In contrast , no significant changes in performance were observed for a no-training control group who did not receive training in between test sessions ( main effect of Session: F ( 1 , 6 ) = 1 . 13 , p=0 . 33; Task x Session interaction: F ( 1 , 6 ) =0 . 0003 , p=0 . 99 ) ( Figure 1—figure supplement 1 ) . To further quantify behavioral improvement , we computed two complementary measures: a ) delta d prime ( Δd’: last training run minus first training run ) that indicates difference in perceptual sensitivity early vs . late in training , b ) learning rate that indicates the rate with which perceptual sensitivity ( d’ calculated per training run ) changes during training . These measures have been previously used in perceptual learning studies to quantify the effect of training on performance ( Ball and Sekuler , 1987; Chang et al . , 2013; Dosher et al . , 2013 ) . Behavioral improvement was similar between tasks , as indicated by no significant differences between tasks in learning rate ( t ( 34 ) =0 . 03 , p=0 . 974 ) nor Δd’ ( t ( 34 ) =0 . 806 , p=0 . 426 ) . We next tested whether behavioral improvement relates to functional brain changes with learning . First , we tested for learning-dependent changes in functional brain activations during training . GLM analysis of the fMRI data across training runs showed significant changes in occipito-temporal BOLD across tasks ( Figure 1—figure supplement 2 ) , suggesting that BOLD changes at this early stage of learning ( i . e . single training session that resulted in maximum 74% mean performance ) do not differ between tasks ( main effect of Task: F ( 1 , 34 ) = 0 . 20 , p=0 . 66; Task x Block interaction: F ( 1 . 9 , 64 . 6 ) , p=0 . 71 ) . This is consistent with previous fMRI studies showing learning-dependent changes within a single training session ( Mukai et al . , 2007 ) . It is possible that the two tasks may show discriminable BOLD activations after more extensive training resulting in saturated performance , as shown by our previous studies using similar learning paradigms with multiple training sessions ( Kourtzi et al . , 2005; Li et al . , 2012; Mayhew et al . , 2012 ) . Second , we conducted whole-brain voxel-wise covariance analyses using either learning rate or Δd’ as covariates . For these analyses , we pooled the data across tasks , as changes in both behavioral performance and BOLD with training were similar between tasks . Our results showed significant correlations between BOLD change ( late vs . early training runs ) in the posterior occipito-temporal cortex and behavioral improvement ( learning rate , Δd’ ) across tasks ( Figure 1c , Figure 1—figure supplement 3 , Figure 1—source data 1 ) . These results provide evidence for learning-dependent changes in occipito-temporal cortex that relate to behavioral improvement , consistent with our previous studies and the role of this region in visual learning and global shape processing ( Kourtzi et al . , 2005; Kuai et al . , 2013; Zhang et al . , 2010 ) . Therefore , we next focused on the posterior occipito-temporal cortex and tested whether learning-dependent BOLD changes relate to changes in GABA concentration in this region . Previous MRS studies have shown that GABA concentrations in the visual cortex relate to performance in perceptual tasks ( Edden et al . , 2009 ) and homeostatic plasticity ( Lunghi et al . , 2015 ) . Here , we test whether GABAergic inhibition relates to behavioral improvement and learning-dependent functional changes in the visual cortex , by comparing MRS-measurements of GABA in the posterior occipito-temporal cortex before vs . after training . First , we tested whether behavioral improvement –as measured by learning rate and Δd’– relates to changes in visual cortex GABA with training . We recorded GABA concentrations before and after training within a voxel centered on the posterior-occipito-temporal cortex ( Figure 2—figure supplement 1 ) , consistent with the fMRI analysis showing learning-dependent BOLD changes with training in this region . Correlating learning rate and Δd’ with GABA changes showed dissociable effects for the two tasks ( Figure 2; Figure 2—figure supplement 4 ) . In particular , for the Signal-in-noise task we observed a negative correlation of GABA change with learning rate ( r = −0 . 43 , CI=[−0 . 74 , –0 . 07] ) , but no significant correlation with Δd’ ( r = −0 . 14 , CI=[−0 . 49 , 0 . 29] ) . In contrast , for the Feature-differences , task we observed a positive correlation of GABA change with Δd’ ( r = 0 . 54 , CI=[0 . 05 , 0 . 85] ) , but no significant correlation with learning rate ( r = 0 . 13 , CI=[−0 . 38 , 0 . 62] ) . Further , the significant correlations of GABA change with behavioral improvement ( learning rate for SN; Δd’ for FD ) were significantly different between tasks ( Fisher’s z = 2 . 91 , p=0 . 004 ) . These dissociable effects could not be simply explained by differences in overall performance between tasks , as the two tasks resulted in similar behavioral improvement . To ensure that our results were specific to GABA changes in the posterior occipito-temporal cortex due to training , we performed the following controls . First , correlation of GABA change and behavioral improvement remained significant when we corrected for a ) tissue ( grey matter , white matter , cerebrospinal fluid ) composition ( SN , correlation with learning rate: r = −0 . 41 , CI=[−0 . 70 , –0 . 07]; FD , correlation with Δd’: r = 0 . 56 , CI=[0 . 03 , 0 . 83] ) and b ) differences in data quality ( as measured by Cramer-Rao Lower Bounds – see Materials and methods ) between the two GABA measurements ( SN , correlation with learning rate: r = −0 . 44 , CI=[−0 . 71 , –0 . 12]; FD , correlation with Δd’: r = 0 . 46 , CI=[0 . 03 , 0 . 76] ) . Second , correlating percentage GABA change ( GABA change/pre training GABA ) with behavioral improvement to control for pre-training GABA showed significant correlations for both tasks ( SN , correlation with learning rate: r = −0 . 45 , CI=[−0 . 78 , –0 . 002]; FD , correlation with Δd’: r = 0 . 58 , CI=[0 . 18 , 0 . 81] ) . These correlations were significantly different between tasks ( Fisher’s z = 2 . 87 , p=0 . 004 ) and remained so when we referenced GABA to NAA rather than creatine concentration ( Fisher’s z = 2 . 73 , p=0 . 01 ) . Third , changes in Glutamate , the other major cortical neurotransmitter , did not correlate significantly with behavioral improvement ( SN , correlation of Glutamate change with learning rate: r = 0 . 33 , CI=[−0 . 22 , 0 . 67]; FD , correlation of Glutamate change with Δd’: r = −0 . 30 , CI=[−0 . 58 , 0 . 06] ) . These correlations of Glutamate change with measures of behavioral improvement were significantly different from correlations of GABA change with behavioral improvement ( SN , correlations with learning rate: Steiger’s z = 2 . 99 , p=0 . 003; FD , correlations with Δd’: Steiger’s z = 3 . 34 , p=0 . 001 ) . Further , correlations of GABA change and behavioral improvement remained significant after accounting for Glutamate change ( SN , correlation of GABA change with learning rate: r = −0 . 41 , CI=[−0 . 69 , –0 . 08]; FD , correlation of GABA change with Δd’: r = 0 . 54 , CI=[0 . 04 , 0 . 85] ) , suggesting that our results were specific to GABA and do not generalize to glutamate . Finally , to test whether our findings were specific to occipito-temporal cortex GABA , we measured learning related GABA changes in both the posterior occipito-temporal cortex and the posterior parietal cortex ( IPS ) in an independent group of participants ( SN: n = 17; FD: n = 21 ) . We found significant and opposite correlations of occipito-temporal GABA change with behavioral improvement for the two tasks ( SN: r = −0 . 43 , CI=[−0 . 75 , –0 . 02]; FD: r = 0 . 55 , CI=[0 . 10 , 0 . 78] ) . We did not find significant correlations between posterior parietal GABA change and behavioral improvement for either task ( SN: r = −0 . 23 , CI=[−0 . 61 , 0 . 19]; FD: r = 0 . 05 , CI=[−0 . 37 , 0 . 43] ) , suggesting that our findings are specific to local changes in occipito-temporal GABA rather than reflecting changes in global cortical excitability . Our analyses so far showed significant correlations of changes in GABA and behavior due to training . Yet , we did not observe significant differences in mean GABA concentration in occipito-temporal cortex before vs . after training ( main effect of MRS block: F ( 1 , 34 ) = 0 . 06 , p=0 . 81; Task x MRS block interaction: F ( 1 , 34 ) = 0 . 21 , p=0 . 65 ) ( Figure 2—figure supplement 3 ) . Previous studies have reported mean changes in GABA concentration in the motor cortex ( Floyer-Lea et al . , 2006; Sampaio-Baptista et al . , 2015 ) due to training and visual cortex due to changes in homeostatic plasticity ( Lunghi et al . , 2015 ) . The main difference between our study and these previous reports is that participant performance increased but did not saturate during the single training session employed in our study ( i . e . participant reached mean performance 74% ) , in contrast to previous studies that showed saturated performance after training . Thus , it is likely that mean changes in GABA concentration are more pronounced when participant performance has plateaued after training . Further , it is likely that 7T imaging ( rather than 3T imaging used in our study ) affords increased signal-to-noise ratio and time resolution that may benefit measurements of change in GABA concentration ( Barron et al . , 2016; Lunghi et al . , 2015 ) . Next , we tested whether learning-dependent changes in visual GABA ( before vs . after training ) relate to changes in BOLD within the posterior occipito-temporal cortex . We conducted a GLM covariance analysis to test whether BOLD changes ( late vs . early training runs ) relate to GABA changes in this region . This analysis showed opposite correlations between GABA and BOLD change for the two tasks: negative correlation for the Signal-in-Noise , while positive correlation for the Feature-differences task ( Figure 3a ) . We corroborated this result by extracting BOLD signal from the voxel clusters in the posterior occipito-temporal cortex that resulted from the covariance analysis of fMRI with behavioral improvement ( Figure 1c ) . Correlations of change in GABA and BOLD –extracted from this independently defined region of interest ( Figure 1c ) - were opposite and significantly different between the two tasks ( SN: r = −0 . 58 CI=[−0 . 82 , –0 . 22] , FD: r = 0 . 70 CI=[0 . 37 , 0 . 90] , Fisher’s z = 4 . 19 , p<0 . 0001 ) ( Figure 3b ) . Our findings suggest that task-dependent suppression mechanisms relate to functional changes in visual cortex and behavioral improvement . To further test this hypothesis , we performed moderation analyses ( Hayes , 2012 ) ( Figure 3—figure supplement 2 ) that allowed us to test whether the influence that an independent variable ( i . e . BOLD ) has on the outcome ( i . e . behavior ) is moderated by one or more moderator variables ( i . e . GABA , task ) . Our results showed that this model is significant ( F ( 7 , 28 ) =3 . 77 , p=0 . 01 ) and the relationship between BOLD change and behavioral improvement depends multiplicatively on GABA change and task , as indicated by a significant three-way interaction between task , GABA change , and BOLD change ( F ( 1 , 28 ) = 7 . 17 , p=0 . 01; R-square change = 0 . 13 ) . These results suggest that task-dependent GABAergic inhibition moderates the relationship between functional brain plasticity and behavioral improvement in the visual cortex . To ensure that the dissociable correlations we observed between tasks for behavior , GABA and BOLD were not due to differences between the two groups of participants that were each trained on a different task ( SN vs FD group ) , we compared behavioral and imaging data between groups before training . First , our analyses did not show any significant differences in GABA concentration before training ( t ( 34 ) =0 . 11 , p=0 . 91 ) nor in behavioral performance early in training ( i . e . first training run ) ( t ( 34 ) =0 . 23 , p=0 . 82 ) between the two groups . Second , we compared signal-to-noise ratio ( SNR ) between tasks for the first MRS measurement ( i . e . pre-training ) and the first two fMRI runs ( i . e . early in the training , as there were no fMRI measurements before training ) . We did not find any significant differences in MRS SNR ( t ( 34 ) =0 . 77 , p=0 . 45 ) , nor fMRI temporal SNR ( tSNR ) between the two tasks ( t ( 34 ) =0 . 73 , p=0 . 47 ) . These results suggest that the dissociable results we observed between tasks could not be simply due to differences in the two groups . Further , to ensure that the learning-dependent changes we observed were not confounded by changes in the scanner environment during training , we conducted the following control analyses . First , we calculated the variation of the scanner center frequency across training runs for each participant . We found that the mean scanner center frequency variation across participants was very small ( mean and standard deviation across participants: 0 . 0000125 ± 0 . 0000019 MHz ) , and there was no significant interaction between Training ( first two vs . last two fMRI runs ) and Task ( F ( 1 , 34 ) =0 . 68 , p=0 . 42 ) . Second , a similar analysis on tSNR across fMRI runs did not show a significant interaction between Training ( first two vs . last two fMRI runs ) and Task ( F ( 1 , 34 ) = 1 . 62 , p=0 . 21 ) . Further , to control for measurement differences in the MRS before vs . after training we conducted the following analyses . First , to assess measurement quality we calculated spectral SNR for each MRS measurement . This analysis showed no significant interaction between MRS block and task ( F ( 1 , 34 ) = 2 . 37 , p=0 . 13 ) nor a main effect of block ( F ( 1 , 34 ) = 1 . 60 , p=0 . 22 ) . Second , to assess spectral resolution before vs . after training , we calculated peak linewidth for each MRS measurement . This analysis showed no significant interaction between MRS block and task ( F ( 1 , 34 ) =0 . 90 , p=0 . 35 ) nor a significant main effect of MRS block ( F ( 1 , 34 ) = 2 . 97 , p=0 . 09 ) . These results suggest that the MRS data quality was similar before and after training for both tasks . Taken together these analyses suggest that the dissociable correlations between BOLD and GABA we observed between tasks could not be due to differences in the quality of the BOLD or GABA measurements during training . Finally , to ensure that our results were specific to learning-dependent changes , we excluded data from participants who did not show positive improvement during the single training session employed in our study , as indicated by learning rate ( n = 3 ) or Δd’ ( n = 8 ) . Despite the smaller data sample , the following results remained significant: a ) correlations of GABA change with behavioral improvement ( SN: r = −0 . 52 , CI=[−0 . 80 , –0 . 09]; FD: r = 0 . 72 , CI=[0 . 29 , 0 . 94] ) , b ) correlations of BOLD change ( early vs . late training runs ) with behavioral improvement ( learning rate ( r = 0 . 57 , CI=[0 . 24 , 0 . 80] ) and with Δd’ ( r = 0 . 67 , CI=[0 . 33 , 0 . 86] ) . Further , the correlations between GABA change and BOLD change ( extracted from the voxel clusters revealed by the independent covariance analysis with behavioral improvement ) remained significantly different between tasks ( z = 2 . 84 , p=0 . 01 ) . To extend beyond correlative evidence , we sought to perturb cortical excitability using transcranial direct current stimulation ( tDCs ) that has been previously shown to alter overall responsivity of the visual cortex ( i . e . modulate visual evoked potentials ) ( Antal et al . , 2004a ) . Our findings on the relationship of GABA change and behavioral improvement lead to opposite predictions for the effect of tDCs on the two learning tasks . In particular , anodal tDCs is known to be excitatory ( Nitsche and Paulus , 2000 ) and has been shown to result in local GABA reduction in visual ( Barron et al . , 2016 ) and motor cortex ( Stagg et al . , 2009 ) . Further , anodal tDCs has been shown to facilitate learning in motor ( O'Shea et al . , 2017; Stagg et al . , 2011c ) and perceptual tasks ( Fertonani et al . , 2011; Pirulli et al . , 2013; Sczesny-Kaiser et al . , 2016 ) . Our results for the Signal-in-Noise task showed that GABA change correlated negatively with behavioral improvement , suggesting that decreased GABA relates to higher behavioral improvement . Therefore , we hypothesized that excitatory anodal tDCs would enhance performance during training on this task . In contrast , cathodal stimulation is thought to be inhibitory; that is , it has been shown to reduce cortical excitability ( Nitsche and Paulus , 2000 ) by decreasing glutamatergic transmission ( Stagg et al . , 2009 ) . Further , cathodal tDCs on the occipital cortex has been shown to facilitate performance in perceptual judgments by suppressing incorrect sensory input ( Antal et al . , 2004b ) . Our results for the Feature differences task showed that GABA change correlated positively with behavioral improvement , suggesting that increased GABA relates to higher behavioral improvement . Therefore , we hypothesized that the inhibitory cathodal –rather than the excitatory anodal– stimulation would enhance performance during training on this task . To test these hypotheses , participants in different groups received anodal , cathodal or sham stimulation in the posterior occipito-temporal cortex during training on the Signal-in-Noise or the Feature-differences task . Our results ( Figure 4 ) showed significant improvement in behavioral performance , as measured by d’ , for anodal compared to sham stimulation for the Signal-in-Noise ( Training block x Stimulation: F ( 2 , 52 . 5 ) = 3 . 99 , p=0 . 02 ) but not the Feature-differences task ( Training block x Stimulation: F ( 2 . 2 , 57 . 9 ) = 0 . 45 , p=0 . 66 ) . In contrast , we observed improved performance during cathodal compared to sham tDCs for the Feature-differences task ( main effect of Stimulation: F ( 1 , 26 ) = 6 . 13 , p=0 . 02 ) but not the Signal-in-Noise task ( main effect of Stimulation: F ( 1 , 26 ) = 0 . 001 , p=0 . 98 ) . To compare behavioral improvement between tasks , we normalized performance during tDCs ( anodal or cathodal ) to performance during sham stimulation ( Figure 4a ) . A repeated-measures ANOVA showed a significant Task , Stimulation x Training block interaction ( F ( 2 . 5 , 130 ) =3 . 19 , p=0 . 03 ) . We next conducted the following control analyses to ensure that our results relate to learning-dependent changes in behavioral performance rather than differences in task difficulty across stimulation groups . Comparing performance across participants before training ( i . e . pre-training block with no feedback or stimulation ) showed no significant effect of Task ( F ( 1 , 78 ) =0 . 05 , p=0 . 82 ) , nor a significant interaction between Task and Stimulation group ( F ( 2 , 78 ) =0 . 92 , p=0 . 40 ) , suggesting that the tDCs-induced learning effects were not due to differences in difficulty across tasks ( Figure 4 ) . This was further supported by a significant main effect of Session ( F ( 1 , 78 ) =86 . 99 , p<0 . 0001 ) across stimulation groups suggesting that all participants ( including the sham stimulation groups ) were able to learn the task ( Figure 4 ) . Further , the double dissociation we observed between task and stimulation site makes it unlikely that stimulation could produce a non-specific effect on general behavioral performance ( e . g . , through distraction caused by skin irritation ) . In contrast , comparing performance on consecutive days in a no-training control group ( participants were tested twice but without training on the task ) showed no significant main effect of Session ( F ( 1 , 17 ) = 0 . 78 , p=0 . 39 ) nor a significant interaction between Task and Session ( F ( 1 , 17 ) =0 . 30 , p=0 . 59 ) , suggesting that the learning effects we observed were training-specific . Finally , to test whether behavioral improvement was maintained after training , we compared performance in the last training block ( feedback , stimulation ) vs . a post-training test that was conducted on the day following training ( no feedback , no stimulation ) . A repeated-measures ANOVA showed no significant interaction between Session x Stimulation x Task ( F ( 2 , 78 ) =1 . 09 , p=0 . 34 ) , suggesting that improved performance was maintained across all groups when participants were tested without tDCs stimulation . Together , these results demonstrate dissociable effects of tDCs stimulation on behavioral improvement between tasks , suggesting that GABAergic inhibition alters learning and experience-dependent plasticity in the posterior occipito-temporal cortex . In particular , we demonstrate that excitatory stimulation enhances performance during training to detect targets from noise , while inhibitory stimulation enhances fine feature discriminability . These results are consistent with the opposite correlations of change in occipito-temporal GABA and behavioral improvement that we observed across tasks . Taken together , our findings suggest that GABAergic processing in visual cortex optimizes noise filtering for target detection , while retuning of feature templates for fine discrimination . Here , we demonstrate that GABAergic inhibition relates to dissociable learning mechanisms that mediate improved perceptual decisions under uncertainty: when learning to detect targets embedded in clutter or discriminate between highly similar features . Interrogating fMRI signals alone does not allow us to discern between the brain mechanisms that underlie these skills , as BOLD reflects the aggregate activity of excitatory and inhibitory signals at the scale of large neural populations ( Heeger and Ress , 2002; Logothetis , 2008 ) . Our results showed similar learning dependent changes in behavioral performance and BOLD in these tasks at early stages of learning ( i . e . training for a single session ) . This is consistent with previous fMRI studies of perceptual learning that have shown learning-dependent changes in the overall fMRI responses in visual cortex ( e . g . [Kourtzi et al . , 2005; Mukai et al . , 2007; Sigman et al . , 2005] ) , or enhanced discriminability of fMRI patterns with training ( Byers and Serences , 2014; Jehee et al . , 2012; Kuai et al . , 2013; Zhang et al . , 2010 ) . However , combining MRS measurements of GABA with fMRI uncovers distinct suppression mechanisms that moderate the relationship between behavioral improvement and experience-dependent plasticity in visual cortex . Previous studies have investigated the relationship of baseline GABA measurements with performance in the context of visual ( Edden et al . , 2009 ) and sensory-motor tasks ( Heba et al . , 2016; Kolasinski et al . , 2017; Stagg et al . , 2011a ) as well as reward-based learning ( Scholl et al . , 2017 ) . Here , we test whether learning-dependent changes in GABA ( i . e . GABA changes before vs . after training ) relate to changes in performance ( i . e . behavioral improvement ) and functional activation . Previous studies have reported changes in GABA within the range and time scales observed in our study ( 10–15% change observed within 20–30 min ) due to stimulation ( Barron et al . , 2016; O'Shea et al . , 2017; Stagg et al . , 2009 ) or training ( Floyer-Lea et al . , 2006 ) suggesting a role for GABAergic inhibition across stages of learning ( Sampaio-Baptista et al . , 2015; Shibata et al . , 2017 ) . Although , the precise mechanisms that underlie changes in GABA as measured by MRS are still under investigation ( Stagg , 2014 ) , recent animal ( Mason et al . , 2001 ) and human ( Stagg et al . , 2011b ) studies suggest that MRS-measured GABA reflects primarily extra-synaptic GABA concentrations . Our findings provide evidence that changes in GABAergic inhibition in the visual cortex relate to learning-dependent changes in behavior and functional brain plasticity . In particular , for learning to see in clutter , decreased occipito-temporal GABA relates to increased BOLD and improved performance , as indicated by faster learning rate . These findings suggest enhanced noise filtering through gain control that is mediated by learning-dependent changes in visual cortex suppression . This mechanism is consistent with previous animal work linking GABAergic inhibition to neural gain ( Mitchell and Silver , 2003 ) and interventional studies showing that blocking GABAergic inhibition increases neural gain ( Hamann et al . , 2002 ) . It is possible that learning to detect targets in clutter is implemented by decreased local suppression that facilitates recurrent processing for noise filtering and target detection ( Gilbert and Li , 2012; Poort et al . , 2016 ) . In contrast , we demonstrate that for learning to discriminate fine feature differences , increased occipito-temporal GABA relates to increased BOLD and improved performance , as indicated by enhanced sensitivity in visual discrimination after training . This finding suggests that learning to discriminate between highly similar targets by template retuning involves neural tuning to fine feature differences ( Raiguel et al . , 2006; Schoups et al . , 2001; Yang and Maunsell , 2004 ) that is mediated by increased visual cortex suppression . This mechanism is consistent with studies linking GABAergic inhibition to cortical tuning ( Rokem et al . , 2011; Wehr and Zador , 2003 ) and pharmacological interventions showing that GABA agonists enhance orientation selectivity in visual cortex ( Leventhal et al . , 2003; Li et al . , 2008 ) , while blocking GABAergic inhibition results in broader neural tuning ( Leventhal et al . , 2003; Sillito , 1979 ) . Our results showed an intriguing dissociation between tasks; that is , learning rate ( but not Δd’ ) correlated significantly with GABA change for the Signal-in-Noise task , while Δd’ ( but not learning rate ) correlated significantly with GABA change for the Feature differences task . Recent studies characterizing the role of different populations of interneurons in visual learning may shed light into this task-dependent GABAergic plasticity . In particular somatostatin-positive ( SOM ) interneurons have been implicated in spatial summation ( Adesnik et al . , 2012 ) and have been shown to gate plasticity during training by providing contextual information ( van Versendaal and Levelt , 2016 ) . In contrast , parvalbumin-positive ( PV ) interneurons have been implicated in selective inhibition ( Rokem et al . , 2011 ) that sharpens feature representations after training ( Khan et al . , 2018 ) . It is therefore possible , that the dissociable correlations we observed between tasks for GABA change and behavioral improvement may reflect differential involvement of SOM vs . PV interneurons in the two tasks . Specifically , SOM interneurons involved in spatial integration may support learning to detect targets from clutter ( SN task ) through noise filtering . In contrast , PV interneurons involved in selective inhibition may support learning fine differences ( FD task ) through re-tuning of feature templates . Further , SOM vs . PV interneurons are shown to be involved at different stages during the time course of learning . In particular , SOM cells have been shown to gate learning-dependent plasticity during training ( Chen et al . , 2015 ) , while PV cells form stimulus-specific ensembles with pyramidal cells after training on a visual discrimination task ( Khan et al . , 2018 ) . Thus , it is possible that different behavioral measures capture the function of SOM vs . PV interneurons , consistent with the dissociation we observed between tasks for the correlations of GABA change and behavioral improvement . In particular , learning rate ( i . e . the rate with which perceptual sensitivity changes during training ) may capture best the function of SOM interneurons that act during learning to support noise filtering throughout the course of training . In contrast , Δd’ ( i . e . change in perceptual sensitivity after training ) may capture best the function of PV interneurons that are shown to support tuning of stimulus-specific representations after training . Finally , to directly test whether visual cortex suppression mediates behavioral improvement due to training , we employed tDCs to perturb cortical excitability . Our results demonstrate a double dissociation in learning-dependent mechanisms of visual plasticity . Excitatory anodal ( rather than cathodal ) stimulation enhanced learning to detect targets in clutter , consistent with the negative correlation between GABA change and behavior . In contrast , inhibitory cathodal ( rather than anodal ) stimulation enhanced learning to discriminate fine features , consistent with the positive correlation between GABA change and behavior . These findings demonstrate a direct link between GABAergic processing in visual cortex and enhanced visual learning . In sum , our findings provide novel evidence for the role of GABAergic processing in learning , shedding light on the neural implementation of theoretical models of perceptual learning . Future animal studies may probe the micro-circuits that give rise to learning by noise filtering vs . feature template retuning . Recent work has begun to classify cortical interneurons into distinct classes based on morphology , connectivity , and physiology ( Kepecs and Fishell , 2014 ) and link them to distinct cortical computations ( see for example ( El-Boustani and Sur , 2014; Kerlin et al . , 2010; Wilson et al . , 2012 ) . These distinct interneuron types may differentially contribute to learning by noise filtering vs . feature template retuning by changing the gain vs . feature selectivity of pyramidal cells . Thus , our findings propose testable hypotheses linking theoretical models of perceptual learning to the micro-circuits that mediate adaptive behavior and underlie the macroscopic learning-dependent plasticity , as measured by human brain imaging . A hundred and thirty participants took part in this study . Forty six participants ( 21 female; mean age 25 . 04 ± 3 . 69 years ) participated in the brain imaging experiment and eighty four participants ( 45 female; mean age 23 . 8 ± 3 . 41 years ) participated in the brain stimulation experiment . All participants were right-handed , had normal or corrected-to-normal vision and gave written informed consent . The study was approved by the University of Cambridge ethics committee . For the brain imaging experiment , we trained two separate groups of participants . Each group was trained only on one of the two tasks ( SN , FD ) to avoid transfer effects across tasks that have been previously reported when the same individuals were trained sequentially on both tasks ( Chang et al . , 2013; Dosher and Lu , 2007 ) . Data from three participants were excluded from the study due to technical failure during data acquisition resulting in twenty one participants in the Signal-in-Noise experiment and twenty two participants in the Feature-differences experiment . Sample size was determined based on power calculations following previous studies on motor learning showing an effect size of r = 0 . 65 or r = 0 . 60 at 90% power for correlations of GABA with behavior or BOLD , respectively ( Stagg et al . , 2011a ) . For the brain stimulation experiment , we randomly assigned participants into six groups of fourteen participants per group who trained on either the Signal-in-Noise or the Feature-differences task during online anodal , cathodal or sham stimulation . Sample size was determined based on power calculations following previous studies showing a polarity-specific stimulation effect for effect size of Cohen’s d = 1 . 15 at 90% power ( Sczesny-Kaiser et al . , 2016 ) . Previous tDCs studies on visual learning reported significant learning enhancement after stimulation for the same sample size as in our study ( Fertonani et al . , 2011; Pirulli et al . , 2013 ) . We presented participants with Glass patterns ( Glass , 1969 ) generated using previously described methods ( Zhang et al . , 2010 ) . In particular , stimuli were defined by white dot pairs ( dipoles ) displayed within a square aperture on a black background . For the brain imaging experiment stimuli ( size=7 . 9o x 7 . 9o ) were presented in the center of the screen . For the brain stimulation experiment , stimuli ( size=7 . 9o x 7 . 9o ) , were presented at the left hemifield ( 11 . 6 arc min from fixation ) contralateral to the stimulation site to ensure maximal effect of stimulation on stimulus processing . The dot density was 3% , and the Glass shift ( i . e . , the distance between two dots in a dipole ) was 16 . 2 arc min . The size of each dot was 2 . 3 × 2 . 3 arc min2 . For each dot dipole , the spiral angle was defined as the angle between the dot dipole orientation and the radius from the center of the dipole to the center of the stimulus aperture . Each stimulus comprised dot dipoles that were aligned according to the specified spiral angle ( signal dipoles ) for a given stimulus and noise dipoles for which the spiral angle was randomly selected . The proportion of signal dipoles defined the stimulus signal level . We generated radial ( 0o spiral angle ) and concentric ( 90o spiral angle ) Glass patterns by placing dipoles orthogonally ( radial stimuli ) or tangentially ( concentric stimuli ) to the circumference of a circle centered on the fixation dot . Further , we generated intermediate patterns between these two Glass pattern types by parametrically varying the spiral angle of the pattern from 0° ( radial pattern ) to 90° ( concentric pattern ) . We randomized the presentation of clockwise ( 0° to 90° spiral angle ) and counterclockwise patterns ( 0° to −90° spiral angle ) across participants . A new pattern was generated for each stimulus presented in a trial , resulting in stimuli that were locally jittered in their position . For the Signal-in-Noise task , radial and concentric stimuli ( spiral angle: 0° +- 90° ) were presented at 24 ± 1% signal level; that is , 76% of the dipoles were presented at random position and orientation . For the Feature-differences task , stimuli were presented at 100% signal and spiral angle of ±38o ( radial ) or ±52o ( concentric ) ( Figure 1a ) . To control for stimulus-specific training effects , we presented each participant with a newly generated set of stimuli . To control for potential local adaptation due to stimulus repetition , we jittered ( ±1–3° ) the spiral angle across stimuli . These procedures ensured that learning related to global shape rather than local stimulus features .	When searching for a friend in the crowd or telling identical twins apart , your visual system must solve a complex puzzle . It must ignore all irrelevant information ( e . g . , unfamiliar faces in the crowd ) and focus on key features ( e . g . , your friend’s familiar face ) that will allow you to make a decision . We become better at solving complex visual discriminations with practice . But exactly how the brain achieves this improved performance is unclear . To answer this question , Frangou et al . trained healthy volunteers on two such visual tasks . The first ( target detection task ) involved locating a target ( e . g . circular shape made of dots among randomly distributed dots in the background ) , a task similar to identifying a friend in the crowd . The second ( feature discrimination task ) involved assigning highly alike shapes in two different categories , similar to telling apart identical twins . To solve this problem , volunteers had to identify distinct features that allowed them to distinguishthese shapes . During training on this task , they updated and refined the representation of these distinct features in their brain . This enabled them to make finer discriminations and assign each image correctly to one of the two categories . While the volunteers trained on the tasks , Frangou et al . measured levels of a chemical called GABA in brain areas that process visual information . GABA is the brain's main inhibitory molecule and controls the activity of neurons . As the volunteers learned the two tasks , their brains showed opposite changes in GABA levels . In the first , target detection task , individuals did better if their GABA decreased during training . In the second , feature discrimination task , they achieved more if their GABA increased during training . To confirm these findings , Frangou et al . used a second technique to activate or suppress processing in visual areas of the brain . Activating visual areas enhanced performance on the target detection task . Suppressing them enhanced performance on the fine discrimination task . These changes are thus consistent with those seen in GABA levels . As well as revealing how we learn to make decisions based on the information from our eyes , these findings suggest that adjusting brain activity could help patients regain skills lost as a result of eye-related or neurological conditions . Understanding the role of GABA in brain plasticity is also relevant to conditions like autism and psychosis , which have been shown to relate to changes in brain inhibition .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	GABA, not BOLD, reveals dissociable learning-dependent plasticity mechanisms in the human brain
The molecular underpinnings of antibiotic resistance are increasingly understood , but less is known about how these molecular events influence microbial dynamics on the population scale . Here , we show that the dynamics of E . faecalis communities exposed to antibiotics can be surprisingly rich , revealing scenarios where increasing population size or delaying drug exposure can promote population collapse . Specifically , we demonstrate how density-dependent feedback loops couple population growth and antibiotic efficacy when communities include drug-resistant subpopulations , leading to a wide range of behavior , including population survival , collapse , or one of two qualitatively distinct bistable behaviors where survival is favored in either small or large populations . These dynamics reflect competing density-dependent effects of different subpopulations , with growth of drug-sensitive cells increasing but growth of drug-resistant cells decreasing effective drug inhibition . Finally , we demonstrate how populations receiving immediate drug influx may sometimes thrive , while identical populations exposed to delayed drug influx collapse . Antibiotic resistance is a growing public health threat ( Davies and Davies , 2010 ) . Decades of rapid progress fueled by advances in microbiology , genomics , and structural biology have led to a detailed but still growing understanding of the molecular mechanisms underlying resistance ( Blair et al . , 2015 ) . At the same time , recent studies have shown that drug resistance can be a collective phenomenon driven by emergent community-level dynamics ( Vega and Gore , 2014; Meredith et al . , 2015b ) . For example , drug degradation by a sub-population of enzyme-producing cells can lead to cooperative resistance that allows sensitive ( non-producing ) cells to survive at otherwise inhibitory drug concentrations ( Yurtsev et al . , 2013; Sorg et al . , 2016; Yurtsev et al . , 2016 ) . Additional examples of collective resistance include density-dependent drug efficacy ( Brook , 1989; Udekwu et al . , 2009; Tan et al . , 2012; Karslake et al . , 2016 ) , indole-mediated altruism ( Lee et al . , 2010 ) , and increased resistance in dense surface-associated biofilms ( Davies , 2003 ) . The growing evidence for collective resistance underscores the need to understand not just the molecular underpinnings of resistance , but also the ways in which these molecular-level events shape population dynamics at the level of the bacterial community . Indeed , a wave of recent studies are inspiring novel strategies for combating resistance by exploiting different features of the population dynamics , ranging from competition for resources ( Hansen et al . , 2017; Hansen et al . , 2019 ) or synergy with the immune system ( Gjini and Brito , 2016 ) to temporal and spatial features of growth , selection , or the application of drug ( Lipsitch and Levin , 1997; Meredith et al . , 2015a; Fuentes-Hernandez et al . , 2015; Zhang et al . , 2011; Baym et al . , 2016a; Greulich et al . , 2012; Hermsen et al . , 2012; Moreno-Gamez et al . , 2015; De Jong and Wood , 2018; Trindade et al . , 2009; Borrell et al . , 2013; Bonhoeffer et al . , 1997; Bergstrom et al . , 2004; Bonhoeffer et al . , 1997; Yoshida et al . , 2017; Nichol et al . , 2015; Roemhild et al . , 2018; Baym et al . , 2016b; Michel et al . , 2008; Hegreness et al . , 2008; Pena-Miller et al . , 2013; Rodriguez de Evgrafov et al . , 2015; Munck et al . , 2014; Torella et al . , 2010; Imamovic and Sommer , 2013; Kim et al . , 2014; Pál et al . , 2015; Barbosa et al . , 2017; Barbosa et al . , 2018; Nichol et al . , 2019; Maltas and Wood , 2019; Podnecky et al . , 2018; Imamovic et al . , 2018; Dean et al . , 2020 ) . As a whole , these studies demonstrate the important role of community-level dynamics for understanding and predicting how bacteria respond and adapt to antibiotics . Despite the relatively mature understanding of resistance at the molecular level , however , the population dynamics of microbial communities in the presence of antibiotics are often poorly understood . Here we investigate dynamics of E . faecalis populations exposed to ( potentially time-dependent ) influx of ampicillin , a commonly-used β-lactam . E . faecalis is an opportunistic pathogen that contributes to a number of clinical infections , including infective endocarditis , urinary tract infections , and blood stream infections ( Clewell et al . , 2014; Huycke et al . , 1998; Hancock and Gilmore , 2006; Ch'ng et al . , 2019 ) . β-lactams are among the most commonly used antibiotics for treating E . faecalis infections , though resistance is a growing problem ( Miller et al . , 2014 ) . Resistance to ampicillin can arise in multiple ways , including by mutations to the targeted penicillin binding proteins or production of β-lactamase , an enzyme that hydrolyzes the β-lactam ring and renders the drug ineffective . Enzymatic drug degradation is a common mechanism of antibiotic resistance across species and has been recently linked to cooperative resistance in E . coli ( Yurtsev et al . , 2013 ) and S . pneumoniae ( Sorg et al . , 2016 ) . In addition , E . faecalis populations exhibit density-dependent growth when exposed to a wide rang-lactamse of antibiotics ( Karslake et al . , 2016 ) . Increasing population density typically leads to decreased growth inhibition by antibiotics , consistent with the classical inoculum effect ( IE ) ( Brook , 1989 ) . However , β-lactams can also exhibit a surprising ‘reverse’ inoculum effect ( rIE ) characterized by increased growth of the population at lower densities ( Karslake et al . , 2016; Jokipii et al . , 1985 ) . In E . faecalis , the rIE arises from a decrease in local pH at increasing cell densities ( Karslake et al . , 2016 ) , which are associated with increased activity of ampicillin and related drugs ( Yang et al . , 2014 ) . Similar growth-driven changes in pH have been recently shown to modulate intercellular interactions ( Ratzke and Gore , 2018 ) , promote ecological suicide in some species ( Ratzke et al . , 2018 ) , and even to modulate antibiotic tolerance in multispecies communities ( Aranda-Díaz et al . , 2020 ) . In addition to these in vitro studies , recent work shows that E . faecalis infections started from high- and low-dose inocula lead to different levels of immune response and colonization in a mouse model ( Chong et al . , 2017 ) . In this work , we show that density-dependent feedback loops couple population growth and drug efficacy in E . faecalis communities comprised of drug-resistant and drug-sensitive cells exposed to time-dependent concentrations of antibiotic . By combining experiments in computer-controlled bioreactors with simple mathematical models , we demonstrate that coupling between cell density and drug efficacy can lead to rich dynamics , including bistabilities where low-density populations survive while high-density populations collapse . In addition , we experimentally show that there are certain scenarios where populations receiving immediate drug influx may eventually thrive , while identical populations exposed to delayed drug influx–which also experience lower average drug concentrations–are vulnerable to population collapse . These results illustrate that the spread of drug resistant determinants exhibits rich and counterintuitive dynamics , even in a simplified single-species population . To investigate the dynamics of E . faecalis populations exposed to β-lactams , we first engineered drug resistant E . faecalis strains that contain a multicopy plasmid that constitutively expresses β-lactamase ( Materials and methods ) . Sensitive cells harbored a similar plasmid that lacks the β-lactamase insert . To characterize the drug sensitive and drug resistant strains , we estimated the half maximal inhibitory concentration , IC50 , of ampicillin in liquid cultures starting from a range of inoculum densities ( Figure 1A; Materials and methods ) . We found that the IC50 for sensitive strains is relatively insensitive to inoculum density over this range , while β-lactam producing resistant cells exhibit strong inoculum effects ( IE ) and show no inhibition for inoculum densities greater than 10-5 ( OD units ) even at the highest drug concentrations ( 10 µg/mL ) . To directly investigate growth dynamics at larger densities–similar to what can be resolved with standard optical density measurements–we used computer controlled bioreactors to measure per capita growth rates of populations held at constant densities and exposed to a fixed concentration of drug ( as in Karslake et al . , 2016 ) . At these higher densities , we found that resistant strains are insensitive to even very large drug concentrations ( in excess of 103 µg/mL ) . By contrast , sensitive populations are inhibited by concentrations smaller than 1 µg/mL , and the inhibition depends strongly on density , with higher density populations showing significantly decreased growth ( Figure 1B ) –indicative of a reverse inoculum effect ( rIE ) . Taken together , these results illustrate opposing effects of cell density on drug efficacy in sensitive and resistant populations . In what follows , we focus on dynamics in the regime OD > 0 . 05 , where the interplay between these two opposing effects may dictate survival or extinction of resistant populations . Bacteria in natural or clinical environments may often be exposed to drug concentrations that change over time . To introduce non-constant antibiotic concentrations , we grew populations in computer controlled bioreactors capable of precise control of inflow ( e . g . drug and media ) and outflow in each growth chamber ( Figure 1C; see also Toprak et al . , 2012; Toprak et al . , 2013; Karslake et al . , 2016 ) . Cell density is monitored with light scattering ( OD ) , and each chamber received fresh media and drug at a rate µ0 ≈ 0 . 1 hr−1 , which is approximately an order of magnitude slower than the per capita growth rate of sensitive cells in drug-free media . In the absence of drug , cells reach a steady state population size of C⁢ ( 1-μ ) , where C is the carrying capacity ( C≈1 in our experiments ) , μ=μ0/gm⁢a⁢x , and gm⁢a⁢x is the drug-free ( maximum ) per capita growth rate of bacteria . By changing the concentration of drug Dr in the media reservoir , we can expose cells to effective rates of drug influx F=μ0⁢Dr . We first characterized the population dynamics of each cell type ( resistant , sensitive ) alone in response to different influx rates of ampicillin . In each experiment , we started one population at OD = 0 . 6 ( ‘high-density’ ) and one at OD = 0 . 1 ( ‘low density’ ) . Not surprisingly , sensitive only populations exhibit a monotonic decrease in final ( 20 hr ) population size with increasing drug concentration ( Figure 1D , top panel ) , with both high- and low-density populations approaching extinction for F >0 . 1 µg/mL/ per hr . By contrast , high- and low-density populations of resistant cells exhibit divergent behavior , with high-density populations surviving and low-density populations collapsing ( Figure 1D , bottom panel ) . In addition , we note that the resistant strains have dramatically increased minimum inhibitory concentrations ( MIC ) , with high-density populations surviving at Dr=104 µg/mL ( an effective influx of over 1000 µg/mL/hr ) . Indeed , the half-maximal inhibitory concentrations ( IC50 ) for sensitive-only and resistant-only populations differ significantly even at very low densities ( Figure 1A ) , suggesting intrinsic differences in resistance even in the absence of density-dependent coupling . This difference corresponds to a direct benefit provided to the enzyme-producing cells , above and beyond any benefit that derives from drug degradation by neighboring cells . These results , along with those in previous studies ( Karslake et al . , 2016 ) , are consistent with a picture of competing density-dependent feedback loops in populations comprised of both sensitive and resistant sub-populations ( Figure 1E ) . Increasing the total population density potentiates the drug , a consequence of the pH-driven reverse inoculum effect ( rIE ) . On the other hand , increasing the size of only the β-lactamase producing subpopulation is expected to decrease drug efficacy as enzymatic activity decreases the external drug concentration . These opposing effects couple the dynamics of different subpopulations with drug efficacy , which in turn modulates both the size and composition of the community . To investigate the potential impact of these competing density effects on population dynamics , we developed a simple phenomenological mathematical model that ascribes density-dependent drug efficacy to a change in the effective concentration of the antibiotic ( see SI for alternative models ) . Specifically , the dynamics of sensitive and resistant populations are described by ( 1 ) dNsdt=g ( D ) ( 1−Ns+NrC ) Ns−μNs , dNrdt=g ( D′ ) ( 1−Ns+NrC ) Nr−μNrwhere Ns is the density of sensitive cells , Nr the density of resistant cells , C is the carrying capacity ( set to one without loss of generality ) , µ is a rate constant that describes the removal of cells due to ( slow ) renewal of media and addition of drug , D is the effective concentration of drug ( measured in units of MIC of the sensitive cells ) , and D′=D/Kr , where Kr is a factor that describes the increase in drug minimum inhibitory concentration ( MIC ) for the resistant ( enzyme producing ) cells in low-density populations where cooperation is negligible . The function g⁢ ( x ) is a dose response function that describes the per capita growth rate of a population exposed to concentration x of antibiotic and is given by Udekwu et al . ( 2009 ) : ( 2 ) g⁢ ( x ) = ( 1-xh ) ⁢gm⁢a⁢x⁢gm⁢i⁢nxh⁢gm⁢a⁢x+gm⁢i⁢nwhere h is a Hill coefficient that describes the steepness of the dose response function , gm⁢a⁢x is the growth in the absence of drug , and gm⁢i⁢n>0 is the maximum death rate . The function g⁢ ( x ) is a sigmoidal function that equals gm⁢a⁢x at x=0 ( no drug ) , decreases monotonically and crosses the horizontal axis at x=1 , and then approaches the maximum death rate gm⁢i⁢n as x approaches infinity ( g⁢ ( x ) →-gm⁢i⁢n ) . Without loss of generality , we set gm⁢a⁢x=1 , which is equivalent to measuring all rates in time units set by gm⁢a⁢x-1 ( coincidentally , we find that drug-free growth rate under the current experimental conditions is approximately gm⁢a⁢x=1 hr-1 , so measuring rates in units of gm⁢a⁢x-1 is equivalent to measuring time in hours ) . To account for the density dependence of drug efficacy , we model the effective drug concentration as ( 3 ) d⁢Dd⁢t=F+ϵ1⁢ ( Ns+Nr ) ⁢D-ϵ2⁢Nr⁢D-D⁢μwhere ϵ1>0 is an effective rate constant describing the reverse inoculum effect ( proportional to total population size ) , which is modeled as an increase in the effective drug concentration with cell density . We do not mean to imply that the cells physically produce antibiotic; instead , this phenomenological model is intended to capture the increase in drug efficacy due to acidification of the local environment as density increases . Similarly , the parameter ϵ2>0 describes the enzyme-driven ‘normal’ inoculum effect ( proportional to the size of the resistant subpopulation ) , which corresponds mathematically–and in this case , also physically–to a degradation of antibiotic . F=Dr⁢μ is rate of drug influx into the reservoir , which can be adjusted by changing the concentration Dr in the drug reservoir . When ϵ2≤ϵ1–when the per capita effect of the inoculum effect ( IE ) is less than or equal to that of its reverse ( rIE ) counterpart–the ϵ1 term is always larger in magnitude than the ϵ2 term and the net effect of increasing total cell density is to increase effective drug concentration , regardless of population composition . This regime is inconsistent with experiments , where resistant-only populations exhibit a strong IE and sensitive-only populations a rIE ( Figure 1 ) . We therefore focus on the case ϵ2>ϵ1 , where density and composition-dependent trade-offs may lead to counterintuitive behavior . Despite the simplicity of the model , it predicts surprisingly rich dynamics ( Figure 2 ) . At rates of drug influx below a critical threshold ( F<Fc ) , populations reach a stable fixed point at a density approaching C⁢ ( 1-μ ) as influx approaches zero . On the other hand , populations go extinct for large influx rates F≫Fc , regardless of initial density or composition . Between the two regimes lies a region of bistability , where populations are expected to survive or die depending on the initial conditions . To characterize the behavior in this bistable region , we calculated the separatrix–the surface separating regions of phase space leading to survival from those leading to extinction–for different values of the antibiotic influx rate using an iterative bisection algorithm , similar to Cavoretto et al . ( 2017 ) . The analysis reveals that increasing total population size can lead to qualitatively different behavior–survival or extinction–depending on the rate of drug influx . For influx rates at the upper end of the bistable region–and for sufficiently high initial fractions of resistant cells– high-density populations survive while low-density populations go extinct ( Figure 2 , bottom right panel ) . For example , in populations with an initial resistant fraction of 3/4 , small populations approach the extinction fixed point while large populations are expected to survive ( Figure 2 ) . Intuitively , the high-density populations have a sufficiently large number of resistant cells , and therefore produce a sufficient quantity of β-lactamase , that effective drug concentrations reach a steady state value below the MIC of the resistant cells , leading to a density-dependent transition from extinction to survival as the separatrix is crossed ( Figure 2 , bottom right ) . Behavior in the low-influx regime of bistability ( F≈Fc ) is more surprising . In this regime , the model predicts a region of bistability where initially high-density populations go extinct while low-density populations survive ( Figure 2 , bottom left ) . For example , at a resistant fraction of 1/4 , low-density populations will approach the survival fixed point while high-density populations will approach extinction as the separatrix is crossed . These counterintuitive dynamics , which we refer to as ‘inverted bistability’ , are governed in part by the reverse inoculum effect , which leads to a rapid increase in drug efficacy in the high-density populations and a corresponding population collapse . Mathematically , the different behavior corresponds to a translation in the separatrix curve as the influx rate is modulated ( Figure 2; see also Figure 2—figure supplement 1 ) . Interestingly , the stable solutions that correspond to survival are comprised of only resistant cells . Hence , the model is not predicting a stable coexistence of sensitive and resistant strains ( though such coexistence can exist under some conditions; Lenski and Hattingh , 1986 ) ; instead , the initial presence of sensitive cells positions the population within the basin of attraction of states ( like collapse ) that would not be favored in their absence . To further characterize the dynamics of the model , we numerically solved the coupled equations ( Equations 1 , 3 ) for different initial compositions ( resistant cell fraction ) and different drug influx rates . In each case , we considered both high-density ( OD = 0 . 6 ) and low-density ( OD = 0 . 1 ) populations . As suggested by the bifurcation analysis ( Figure 2 ) , the model exhibits bistability over a range of drug influx rates ( Figure 3A ) . The qualitative behavior within this bistable region can vary significantly . For small resistant fractions and low drug influx , bistability favors survival of low-density populations , while large resistant fractions and high drug influx favor survival of high-density populations . The parameter space is divided into four non-overlapping regions , leading to a phase diagram that predicts regions of extinction , survival , and bistabilities . These qualitative features are not unique the specific model described here , but also occur in alternative models that include , for example , more realistic Monod-style growth ( SI; Figure 3—figure supplement 1 through Figure 3—figure supplement 2 ) . It is notable that the dynamics leading to the fixed points can be significantly more complex than simple mononotic increases or decreases in population size ( Figure 3A , top panels ) . To test these predictions experimentally , we first performed a preliminary scan of parameter space in short , 5-hr experiments starting from a wide range of initial population fractions and drug influx rates ( Figure 3—figure supplement 3 ) . Based on these experiments , we then narrowed our focus to a region of ‘high’ influx rate ( F≈600-700 µg/mL/hr ) , where conditions may favor ‘normal’ bistability , and a region of ‘low’ influx rate ( F≈15-20 µg/mL per hour ) , where conditions may favor ‘inverted’ bistability . Then , we performed replicate ( N=3 ) 20 hr experiments starting from a range of population compositions . Note that in the absence of density-mediated changes in drug concentration , these flow rates are expected to produce drug concentrations that increase over time , rapidly eclipsing the low-density limits for IC50’s of both susceptible and resistant cells ( see Figure 1 ) and exponentially approaching steady state values of D=F/μ≈8 . 5⁢F with a time constant of μ-1≈8 . 5 ( and therefore μ0=8 . 5 hr ) . The experiments confirm the existence of both predicted bistable regimes as well as the expected regimes of survival and extinction ( Figure 3B–C ) . At each of the two flow rates ( F1 and F2 ) , we observe a transition from density-independent extinction–where populations starting from both high and low-densities collapse–to density-independent survival–where both populations survive–as the initial resistant population is increased ( Figure 3B–C , left to right ) . However , in both cases , there are intermediate regimes where initial population density determines whether the population will survive or collapse . When drug influx is relatively high ( F2 ) and the population is primarily comprised of resistant cells ( 55 percent ) , initially large populations survive while small populations collapse ( Figure 3B , middle panel ) . On the other hand , when initial populations contain 11–15% resistant cells and drug influx is relatively small ( F1 ) , we observe a clear region of ‘inverted’ bistability ( Figure 3C , middle panel ) . In this regime , high-density populations ( red ) grow initially before undergoing dramatic collapse , while low-density populations ( blue ) initially decay before recovering and eventually plateauing near the carrying capacity . In contrast to predictions of the model , the collapsing populations do not entirely go extinct . We confirmed that these populations do indeed contain living cells , and single colony isolates exhibit dose-response characteristics similar to those of the original sensitive and resistant strains , so there is no evidence that additional resistance has evolved during the experiment ( Figure 3—figure supplement 4 ) . Mathematical models do indicate the existence of long-lived but transient states of non-zero density near the onset of inverted bistability ( Figure 3—figure supplement 4 ) , which may partially explain the lack of complete extinction . However , it is also possible that it reflects features not included in the model . For example , while ampicillin is generally considered to be stable in solution for several days , the degradation rate depends on both temperature and pH ( Hou and Poole , 1969 ) , which could induce new dynamics on timescales of 10 s of hours . Similarly , β-lactamase activity can also vary slightly with pH , adding an additional layer of coupling between the density effects driven by sensitive and resistant cells ( Ohsuka et al . , 1995 ) . The model predicts that the inverted bistability relies on the reverse inoculum effect–specifically , it requires ε1> 0 and is eliminated when ε1 = 0 ( Figure 4 ) . Previous work showed that in this system , the reverse inoculum effect is driven by density-modulated changes in the local pH ( Karslake et al . , 2016 ) . Conveniently , then , it is possible–in principle–to eliminate the effect by strengthening the buffering capacity of the media . To test this prediction , we repeated the experiments in the inverted bistable region in strongly buffered media ( Figure 4 ) . As predicted by the model , we no longer observe collapse of high-density populations , indicating that the region of inverted bistability is now a region of density-independent survival . The competing density-dependent effects on drug efficacy raise the question of whether different time-dependent drug dosing strategies might be favorable for populations with different starting compositions . In particular , we wanted to investigate the effect of delaying the start of antibiotic influx for different population compositions and influx rates . Based on the results of the model , we hypothesized that there would be two possible regimes where delay could dramatically impact survival dynamics: one ( corresponding to ‘normal bistability’ ) where delaying treatment would lead to larger end-point populations , and a second ( corresponding to“inverted bistabillity’ ) where delaying treatment could , counterintuitively , promote population collapse ( see Figure 5—figure supplement 1 ) . To test this hypothesis , we measured the population dynamics in mixed populations starting from an initial OD of 0 . 1 at time zero . We then compared final population size for identical populations experiencing immediate or delayed drug influx , with delay ranging from 0 . 5 to 2 . 5 hr . In experiments with non-zero delays , antibiotic influx was replaced by influx of drug-free media ( at the same flow rate ) during the delay period . In the first case , we chose a relatively small initial resistant fraction ( 0 . 11 ) and a relatively slow drug influx rate ( F=18 µg/mL ) , while in the second case we chose a larger initial resistant fraction ( 0 . 55 ) and a faster drug influx ( F=650 µg/mL ) . Remarkably , we found that delaying treatment can have opposing effects in the two scenarios ( Figure 5 ) . At high drug influx rates and largely resistant populations , immediate treatment leads to smallest final populations ( Figure 5 , right panels ) , consistent with model predictions of bistability . Intuitively , the delay allows the subpopulation of resistant cells to increase in size , eventually surpassing a critical density where the presence of enzyme is sufficient to counter the inhibitory effects of antibiotic . On the other hand , at lower influx rates and lower initial resistant fractions , we find that immediate treatment leads to initial inhibition followed by a phase of rapid growth as the population thrives; by contrast , delays in treatment allow the population to initially grow rapidly before collapsing ( Figure 5 , left panels ) . It is particularly striking that delayed treatments–which also use significantly less total drug–can promote population collapse when immediate treatments appear to fail . Similar to the ‘inverted bistability’ observed earlier , the beneficial effects of delayed treatment can be traced to density-dependent drug efficacy–in words , the delay means the drug is applied when the population is sufficiently large that pH-mediated drug potentiation promotes collapse . We have shown that different types of coupling between cell density and drug efficacy can lead to surprising dynamics in E . faecalis populations exposed to time-dependent ampicillin concentrations . In regimes of relatively fast or slow rates of drug influx , the results are intuitive: populations either survive or collapse , independent of initial population size ( density ) . The intermediate regime , however , is characterized by bistability , meaning that population collapse will depend on initial population size . In regimes where cooperative resistance–in this case , due to enzymatic degradation of drug–dominates , larger populations are favored , similar to results predicted from the classical inoculum effect ( Udekwu et al . , 2009; Karslake et al . , 2016 ) . Under those conditions , it is critical to immediately expose cells to drug influx , as delays lead to increasingly resilient populations . Even more surprisingly , regimes characterized by comparatively smaller resistant populations and slower drug influx can lead to ‘inverted bistability’ where initially small populations thrive while large populations collapse . In this case , delays to drug exposure can paradoxically promote population collapse . It is notable that the mathematical model suggests these results are not simply transient effects but instead reflect asymptotic behavior where the system approaches one of two stable fixed points ( survival or extinction ) with very different biological consequences . Our goal was to understand population dynamics in simple , single-species populations where environmental conditions–including drug influx rate and population composition–can be well controlled . To make sense of experimental results and , more importantly , to generate new testable hypotheses , we developed a minimal mathematical model and analyzed its qualitative behavior using standard tools from dynamical systems and bifurcation theory . We chose to focus on a phenomenological model in an effort to simplify the assumptions and limit the number of unconstrained parameters . However , our model clearly omits a number of potentially relevant biological details . For example , the model neglects evolutionary changes , such as de novo mutations , that would impact behavior on longer time-scales . Similarly , previous work ( Meredith et al . , 2018 ) has shown that lysis of resistant cells can effectively increase the concentration of drug-degrading enzyme . We find that extending our phenomenological model to account for free enzyme leads to qualitatively similar behavior ( see SI ) , but more accurate kinetic models may point to different dynamics in some regimes . Constructing detailed mechanistic models is notoriously difficult , but recent work shows that careful pairing of experiment and theory can be used to systematically overcome many common obstacles ( Hart et al . , 2019 ) . A similar approach could potentially be applied to this system , leading to more accurate quantitative models that account for factors like spontaneous drug degradation ( Hou and Poole , 1969 ) , the pH dependence of β-lactamase activity ( Ohsuka et al . , 1995 ) , and the kinetics of pH-modulated drug activity . It is obvious that the specific in vitro conditions used here fail to capture numerous complexities associated with resistance in clinical settings ( Bonten et al . , 2001 ) , including substantial spatial heterogeneity , potential for biofilm formation , effects of the host immune system , and drug concentrations that differ in both magnitude and time-course from the specific scenarios considered here . In particular , the effects of delayed antibiotic exposure in a clinical setting will depend on many factors not captured here , and there are unquestionably scenarios where such delay could be detrimental to patient well-being . In fact , our results indicate that delaying drug exposure can have differing effects in different parameter regimes , even in laboratory populations . Future work in clinically motivated in vitro systems , such as biofilms , and ultimately in vivo are needed to assess the feasibility of delayed dosing in more realistic scenarios . In addition , we note that β-lactamase producing enterococci are thought to be relatively rare , though they have been associated with multi-drug resistant , high-risk enterococcal infections ( Murray , 1992; Wells et al . , 1992; Arias et al . , 2010 ) and may be more widespread that initially believed because of the difficulty of detection in traditional laboratory tests ( Gagetti et al . , 2019 ) . Finally , our experimental model system is based on plasmid-mediated resistance , and while this fact is not explicitly assumed in any of our mathematical models , horizontal gene transfer may introduce new dynamics ( Lopatkin et al . , 2016; Lopatkin et al . , 2017 ) , particularly in high-density populations where conjugation is frequent . Our results show that the response of microbial populations to antibiotic can be surprisingly complex , suggesting that the spread of resistance alleles may not always follow simple selection dynamics . These findings underscore the need for additional metrics ( similar to the proposed notion of drug resilience; Meredith et al . , 2018 ) that go beyond short-term growth measurements to for population dynamics over multiple timescales . More generally , we hope these results will motivate continued efforts to understand the potentially surprising ways that molecular level resistance events influence dynamics on the scale of microbial populations . Experiments were performed with E . faecalis strain OG1RF , a fully sequenced oral isolate . Ampicillin-resistant strains were engineered by transforming ( Dunny et al . , 1991 ) OG1RF with a modified version of the multicopy plasmid pBSU101 , which was originally developed as a fluorescent reporter for Gram-positive bacteria ( Aymanns et al . , 2011 ) . The plasmid was chosen because it can be conveniently manipulated and propagated in multiple species ( including E . coli ) and contains a fluorescent reporter that provides a redundant control for readily identifying the strains . The modified plasmid , named pBSU101-BFP-BL , expresses BFP ( rather than GFP in the original plasmid ) and also constitutively expresses -lactamase driven by a native promoter isolated from the chromosome of clinical strain CH19 ( Rice et al . , 1991; Rice and Marshall , 1992 ) . The β-lactamase gene and reporter are similar to those found in other isolates of enterococci and streptococci ( Murray and Mederski-Samaroj , 1983; Zscheck and Murray , 1991 ) . Similarly , sensitive strains were transformed with a similar plasmid , pBSU101-DasherGFP , a pBSU101 derivative that lacks the β-lactamase insert and where eGFP is replaced by a brighter synthetic GFP ( Dasher-GFP; ATUM ProteinPaintbox , https://www . atum . bio/ ) . The plasmids also express a spectinomycin resistance gene , and all media was therefore supplemented with spectinomycin . Antibiotics used in this study included Spectinomycin Sulfate ( MP Biomedicals ) and Ampicillin Sodium Salt ( Fisher ) . Experiments to estimate the half-maximal inhibitory concentration ( IC50 ) for each population were performed in 96-well plates using an Enspire Multimodal Plate Reader . Overnight cultures were diluted 102 - 108 fold into individual wells containing fresh BHI and a gradient of 6–14 drug concentrations . After 20 hr of growth , the optical density at 600 nm ( OD ) was measured and used to create a dose response curve , which was fit to a Hill-like function f⁢ ( x ) = ( 1+ ( x/K ) h ) -1 using nonlinear least squares fitting , where K is the half-maximal inhibitory concentration ( IC50 ) and h is a Hill coefficient describing the steepness of the dose-response relationship . Experiments were performed in custom-built , computer-controlled continuous culture devices ( CCD ) as described in Karslake et al . ( 2016 ) . Briefly , bacterial populations are grown in glass vials containing a fixed volume of 17 mL media . Cell density was measured at 1 . 5 s intervals in each vial using emitter/detector pairs of infrared LEDs ( Radioshack ) . Detectors register a voltage output that is then converted to optical density using a calibration curve performed with a table top OD reader . Each vial contains input and output channels connected to silicone tubing and attached to a system of peristaltic pumps ( Boxer 15000 , Clark Solutions ) that add drug and/or media and remove excess liquid on a schedule that can be programmed in advance or determined in real time . The entire system is controlled using a collection of DAQ and instrument control modules ( Measurement Computing ) along with the Matlab ( MathWorks ) Instrument Control Toolbox . In ‘constant flow’ experiments , media ( with drug , when relevant ) is added at a rate of 1 mL/min for a total of 7 . 5 s every 3 . 75 min for an effective flow rate of 2 mL/hr ( corresponding to a rate constant of µ = 10 . 12 hr-1 in 17 mL total volume ) . Media ( plus cells and drug ) is removed at an identical rate to maintain constant volume . While drug influx ( and waste removal ) strictly occurs on discrete on-off intervals , the timescale of those intervals ( 3–4 min ) is an order of magnitude slower than the maximum bacterial growth rate under these conditions , which corresponds to a doubling time of approximately 30–40 min . The influx of drug is therefore approximately continuous on the timescale of bacterial dynamics . We experimentally modulate the influx rate of drug , F , without changing the background refresh rate ( µ ) by changing the drug concentration in the drug reservoir . For experiments involving time-dependent drug influx–for example , those in Figure 5 , the media in the drug reservoirs is exchanged manually at specified times to mimic , for example , delayed treatment start times . All experiments were started from overnight cultures inoculated from single colonies grown on BHI agar plates with streptomycin and incubated in sterile BHI ( Remel ) with streptomycin ( 120 µg/mL ) overnight at 37C . Highly buffered media was prepared by supplementing standard BHI with 50 µM Dibasic Sodium Phosphate ( Fisher ) . Overnight cultures were diluted 100–200 fold with fresh BHI in continuous culture devices and populations were allowed to reach steady state exponential growth at the specified density ( typically OD = 0 . 1 or OD = 0 . 6 ) prior to starting influx and outflow of media and waste . Experiments were typically performed in triplicate .	Antibiotic resistance is a threat to human and animal health worldwide . Although we rely on antibiotics to treat diseases caused by bacteria , such as tuberculosis , some bacteria are already resistant to many of the drugs available . Understanding the basis of resistance is crucial for developing new antibiotics , and for using current drugs more efficiently . One way that bacteria resist antibiotics is by producing enzymes that inactivate specific drugs . If a community of bacteria contains both vulnerable and resistant cells , this can lead to a phenomenon called ‘cooperative resistance’ . When treated with antibiotics , vulnerable cells within the group are shielded by their resistant neighbors , which effectively remove the drugs from the environment . Cooperative resistance can make it difficult for researchers to understand how resistance develops in different bacterial populations . This is because a large group of cells may collectively behave in a different way than individual cells . This means that bacterial populations are a more realistic model for ‘real-world’ infections and disease than studies of single cells . Now , Hallinen , Karslake and Wood show how cooperation between cells affects the way bacterial communities respond to beta-lactams , the most commonly prescribed class of antibiotic drugs . Experiments using cultures of Enterococcus faecalis , a bacterium that often causes hospital infections , revealed that the density of different bacterial populations changes the effectiveness of drugs . Although increased cell density had a protective effect on populations containing only resistant bacteria , it made non-resistant populations even more vulnerable . Mathematical modelling using information from the culture experiments predicted that interactions between vulnerable and resistant bacteria within a mixed community can determine how populations change over time . For example , if the number of antibiotic-sensitive cells is too high , this can cause the entire population to collapse . These predictions contradict the conventional understanding of how antibiotic resistance spreads , where small numbers of resistant cells multiply rapidly at the expense of vulnerable ones . These results shed new light on the complex dynamics of antibiotic resistance within bacterial populations as a whole . In the future , they may inspire new ecology-based strategies for slowing the spread of resistance , ultimately helping reduce the burden of disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems", "microbiology", "and", "infectious", "disease" ]	2020	Delayed antibiotic exposure induces population collapse in enterococcal communities with drug-resistant subpopulations
Information processing in the sensory periphery is shaped by natural stimulus statistics . In the periphery , a transmission bottleneck constrains performance; thus efficient coding implies that natural signal components with a predictably wider range should be compressed . In a different regime—when sampling limitations constrain performance—efficient coding implies that more resources should be allocated to informative features that are more variable . We propose that this regime is relevant for sensory cortex when it extracts complex features from limited numbers of sensory samples . To test this prediction , we use central visual processing as a model: we show that visual sensitivity for local multi-point spatial correlations , described by dozens of independently-measured parameters , can be quantitatively predicted from the structure of natural images . This suggests that efficient coding applies centrally , where it extends to higher-order sensory features and operates in a regime in which sensitivity increases with feature variability . Sensory receptor neurons encode signals from the environment , which are then transformed by successive neural layers to support diverse and computationally complex cognitive tasks . A normative understanding of these computations begins in the periphery , where the efficient coding principle—the notion that a sensory system is tuned to the statistics of its natural inputs—has been shown to be a powerful organizing framework ( Barlow , 2001; Simoncelli , 2002 ) . Perhaps the best-known example is that of redundancy removal via predictive coding and spatiotemporal decorrelation . In insects , this is carried out by neural processing ( Laughlin , 1981; van Hateren , 1992b ) ; in vertebrates , fixational eye movements—which precede the first step of neural processing ( Srinivasan et al . , 1982; Atick and Redlich , 1990; Atick et al . , 1992 ) —play a major role ( Kuang et al . , 2012 ) . This approach was later extended to describe population coding , retinal mosaic structure ( Barlow , 2001; Karklin and Simoncelli , 2001; Borghuis et al . , 2008; Balasubramanian and Sterling , 2009; Liu et al . , 2009; Garrigan et al . , 2010; Ratliff et al . , 2010; Kuang et al . , 2012 ) , adaptation of neural responses ( Brenner et al . , 2000; Fairhall et al . , 2001; Schwartz and Simoncelli , 2001 ) , and early auditory processing ( Smith and Lewicki , 2006 ) . Taken together , normative theories based on efficient coding have been successful in explaining aspects of processing in the sensory periphery that are tuned to simple statistical features of the natural world . Can we extend such theories beyond the sensory periphery to describe cortical sensitivity to complex sensory features ? Normative theories have been successful in predicting the response properties of single cells , including receptive fields in V1 ( Olshausen and Field , 1996; Bell and Sejnowski , 1997; van Hateren and Ruderman , 1998; van Hateren and van der Schaaf , 1998; Hyvarinen and Hoyer , 2000; Vinje and Gallant , 2000; Karklin and Lewicki , 2009 ) and spectro-temporal receptive fields in primary auditory cortex ( Carlson and DeWeese , 2002 , 2012 ) , as well as distributions of tuning curves across individual cells in a population ( Lewicki , 2002; Ganguli and Simoncelli , 2011 ) . Some complex features , however , might not be represented by the tuning properties of individual cells in any direct way , but rather emerge from the collective behavior of many cells . Instead of trying to predict individual cell properties , we therefore focus on the sensitivity of the complete neural population . Is there an organizing principle that determines how resources within the population are allocated to representing such complex features ? When the presence of complex features is predictable ( i . e . , can be accurately guessed from simpler features along with priors about the environment ) , mechanisms are best devoted elsewhere ( See Discussion , van Hateren , 1992a ) . In contrast , sensory features that are highly variable and not predictable from simpler ones can serve to determine their causes ( e . g . , to distinguish among materials or objects ) , a first step in guiding decisions . We will show that these ideas predict a specific organizing principle for aggregate sensitivities arising in cortex: the perceptual salience of complex sensory signals increases with the variability , or unpredictability , of the corresponding signals over the ensemble of natural stimuli . To test this hypothesis , we focus on early stages of central visual processing . Here , early visual cortex ( V1 and V2 ) is charged with extracting edges , shapes , and other complex correlations of light between multiple points in space ( Morrone and Burr , 1988; Oppenheim and Lim , 1981; von der Heydt et al . , 1984 ) . We compare the spatial variation of local patterns of light across natural images with human sensitivity to manipulations of the same patterns in synthetic images . This allows us to determine how sensitivity is distributed across many different features , rather than simply determining the most salient ones . ( We will say that a feature is more salient if it is more easily discriminated from white noise . ) To this end , we parametrize the space of local multi-point correlations in images in terms of a complete set of coordinates , and we measure the probability distribution of coordinate values sampled over a large ensemble of natural scenes . We then use a psychophysical discrimination task to measure human sensitivity to the same correlations in synthetic images , where the correlations can be isolated and manipulated in a mathematically rigorous fashion by varying the corresponding coordinates ( Chubb et al . , 2004; Victor et al . , 2005; Victor and Conte , 2012; Victor et al . , 2013 ) . Comparing the measurements , we show that human sensitivity to these multi-point elements of visual form is tuned to their variation in the natural world . Our result supports a broad hypothesis: cortex invests preferentially in mechanisms that encode unpredictable sensory features that are more variable , and thus more informative about the world . Namely , variance is salience . The analysis of natural scenes is schematized in Figure 1 . We collect an ensemble of image patches from the calibrated Penn natural image database ( PIDB ) ( Tkačik et al . , 2011 ) . We preprocess the image patches as shown in Figure 1A . This involves first averaging pixel luminances over a square region of N × N pixels , which converts an image of size L1 × L2 pixels into an image of reduced size L1/N × L2/N pixels . Images are then divided into R × R square patches of these downsampled pixels and whitened ( see ‘Materials and methods’ , Image preprocessing , for further details ) . Since the preprocessing depends on a choice of two parameters , the block-average factor N and patch size R , we report results for multiple image analyses performed using the identical preprocessing pipeline but for various choices of N and R . After preprocessing , we binarize each patch to have equal numbers of black and white pixels ( black = −1 , white = +1 ) . We characterize each patch by the histogram of 16 binary colorings ( 22×2 ) seen through a square 2 × 2 pixel glider ( Figure 1B ) . Translation invariance imposes constraints on this histogram , reducing the number of degrees of freedom to 10 ( Victor and Conte , 2012 ) . These degrees of freedom can be mapped to a set of image statistic coordinates that separates correlations based on their order: ( i ) one first-order coordinate , γ , describes overall luminance , ( ii ) four second-order coordinates , {β\| , β− , β/ , β\} , describe two-point correlations between pixels arranged vertically , horizontally , or diagonally , ( iii ) four third-order coordinates , {θ⌞ , θ⌜ , θ⌝ , θ⌟} , describe three-point correlations between pixels arranged into ⌞-shapes of different orientations , and ( iv ) one fourth-order coordinate , α , describes the single four-point correlation between all four pixels in the glider ( Figure 1C ) . The binarization step of the preprocessing pipeline forces γ to zero , leaving nine coordinates . Each image patch is thus characterized by a vector of coordinate values {β\| , β− , β/ , β\ , θ⌜ , θ⌝ , θ⌟ , θ⌞ , α} , that is , a point within the multidimensional space of image statistics . Accumulating these points across patches yields a multidimensional probability distribution that characterizes the local correlations in natural scenes ( schematized in Figure 1D ) . A total of 724 images ( up to 249780 patches , depending on the choice of N and R ) , was used to construct this distribution . 10 . 7554/eLife . 03722 . 003Figure 1 . Extracting image statistics from natural scenes . ( A ) We first block-average each image over N × N pixel squares , then divide it into patches of size R × R pixels , then whiten the ensemble of patches by removing the average pairwise structure , and finally binarize each patch about its median intensity value ( see ‘Materials and methods’ , Image preprocessing ) . ( B ) From each binary patch , we measure the occurrence probability of the 16 possible colorings as seen through a two-by-two pixel glider ( red ) . Translation invariance imposes constraints between the probabilities that reduce the number of degrees of freedom to 10 . ( C ) A convenient coordinate basis for these 10° of freedom can be described in terms of correlations between pixels as seen through the glider . These consist of one first-order coordinate ( γ ) , four second-order coordinates ( β\| , β− , β/ , β\ ) , four third-order coordinates ( θ⌞ , θ⌜ , θ⌝ , θ⌟ ) , and one fourth-order coordinate ( α ) . Since the images are binary , with black = −1 and white = +1 , these correlations are sums and differences of the 16 probabilities that form the histogram in panel B ( Victor and Conte , 2012 ) . ( D ) Each patch is assigned a vector of coordinate values that describes the histogram shown in ( B ) . This coordinate vector defines a specific location in the multidimensional space of image statistics . The ensemble of patches is then described by the probability distribution of coordinate values . We compute the degree of variation ( standard deviation ) along different directions within this distribution ( inset ) . ( E ) Along single coordinate axes , we find that the degree of variation is rank-ordered as {β\| , β−}>{β/ , β\}>α>{θ⌞ , θ⌜ , θ⌝ , θ⌟} , shown separately for different choices of the block-average factor N and patch size R used during image preprocessing . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 00310 . 7554/eLife . 03722 . 004Figure 1—figure supplement 1 . Two-component decomposition of natural image distribution . ( A ) The 9-dimensional distribution of natural image statistics is shown projected onto the α−β− plane , where each point represents a single image patch . Note that it is not possible to see all points in the distribution due to their overlap . ( B ) This distribution is well described by a mixture of two components in which each image patch is assigned to one of the two components . Inspection of the image patches assigned to each component reveals that one component ( light gray ) contains in-focus patches , while the other component ( black ) contains blurred patches . Note that the two components are separated in the full 9-dimensional space but appear overlapping when projected onto a single coordinate plane . Insets show semi-transparent versions of the out-of-focus B-1 and in-focus ( B-2 components . We highlight the coordinate values of specific images that are C fully in focus , ( D ) blurred due to variations in field of depth , and ( E ) blurred due to camera motion . Spatial distributions of patch assignments ( left ) and original image patches ( right ) are shown below each distribution . ( C ) A sharp image is composed of patches that are uniformly assigned to the ‘in-focus’ component . ( D ) An image that is partially out of focus due to variations in field of depth has patches that are assigned to each of the two components . ( E ) An image that is blurred due to camera motion is composed of patches that are uniformly assigned to the “blurry” component . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 00410 . 7554/eLife . 03722 . 005Figure 1—figure supplement 2 . Filtering via defocus or motion blur reassigns sharp image patches to the ‘blurry’ component . ( A ) Images can be blurred due to variations in field of depth ( upper row ) or camera motion ( lower row ) . A mixture of components ( MOC ) method separates blurry ( black ) from in-focus ( gray ) image patches . Patches assigned to the ‘blurry’ component have larger positive coordinate values ( red ) , showing saturated values of second- and fourth-order coordinates . Blurring due to variations in field of depth tends to uniformly increase all second- and fourth-order statistics . In comparison , motion blurring tends to more strongly increase both the fourth-order statistic and the second-order statistic aligned with the direction of motion ( here , β\| ) . ( B ) The application of a Gaussian blur filter ( middle row ) or a motion filter ( bottom row ) to an in-focus image ( top row ) produces similar effects; with a sufficiently strong filter ( Gaussian blur of σ=2 pixels or motion of Δh=6 pixels ) , all patches in the original ‘in-focus’ image are reassigned to the ‘blurry’ component . Furthermore , both the Gaussian blur and motion filters alter the distribution of image statistics in a consistent manner . Gaussian blur filters increase the values of all second- and fourth-order coordinates , while motion filters more strongly increase the values of the fourth-order coordinate and the second-order coordinate aligned with the direction of motion . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 00510 . 7554/eLife . 03722 . 006Figure 1—figure supplement 3 . Image statistics along single coordinate axes for white-noise patches . The robustly observed statistical structure of natural scenes ( open circles ) is completely absent from the same analysis performed on samples of white noise ( shaded circles ) . The inset shows that this holds across analysis parameters . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 006 To summarize this distribution , we compute the degree of variation ( standard deviation ) along each coordinate axis ( Figure 1E ) . As is shown , the degree of variation along different coordinate axes exhibits a characteristic rank-ordering , given by {β\| , β−}>{β/ , β\}>α>{θ⌞ , θ⌜ , θ⌝ , θ⌟}; that is , the most variable correlations are pairwise correlations in the cardinal directions , followed by pairwise correlations in the oblique directions , followed by fourth-order correlations . Interestingly , third-order correlations are the least variable across image patches . An analogous analysis performed on white noise yields a flat distribution with considerably smaller standard deviation values ( See ‘Materials and methods’ , Analysis variants for Penn Natural Image Database , and Figure 1—figure supplement 3 for comparison ) , and performing the analysis on a colored Gaussian noise ( e . g . 1/fk spectrum ) would also yield a flat distribution because of the whitening stage in the image preprocessing pipeline . These ( and subsequent ) findings are preserved across different choices of image analysis parameters ( shown in Figure 1E for block-average factors N = 2 , 4 and patch sizes R = 32 , 48 , 64; see ‘Materials and methods’ , Analysis variants for Penn Natural Image Database , and Figure 3—figure supplement 5A for a larger set of parameters ) and also across other collections of natural images ( see ‘Materials and methods’ , Comparison with van Hateren Database , and Figure 3—figure supplement 5B for a parallel analysis of the van Hateren image dataset ( van Hateren and van der Schaaf , 1998 ) , which gives similar results ) . To characterize perceptual sensitivity to different statistics , we isolated them in synthetic visual images and used a figure/ground segmentation task ( Figure 2B ) . We used a four-alternative forced-choice task in which stimuli consisted of a textured target and a binary noise background ( or vice-versa ) . Each stimulus was presented for 120ms and was followed by a noise mask . Subjects were then asked to identify the spatial location ( top , bottom , left , or right ) of the target . Experiments were carried out for synthetic stimuli in which the target or background was defined by first varying image statistic coordinates independently ( Figure 2A shows examples of gamuts from which stimuli are built ) . Along each coordinate axis , threshold ( 1/sensitivity ) was defined as the coordinate value required to support a criterion level of performance ( Figure 2C , inset ) . We then performed further experiments in which the target or background was defined by simultaneously varying pairs of coordinates . For measurements involving each coordinate pair ( to which we will refer as a ‘coordinate plane’ ) , we traced out an isodiscrimination contour ( Figure 2C ) that describes the threshold values not only along the cardinal directions , but also along oblique directions . Measurements were collected for four individual subjects in each of 11 distinct coordinate planes ( representing all distinct coordinate pairs up to 4-fold rotational symmetry; see ‘Materials and methods’ , Psychophysical methods , for further details ) . Each subject performed 4320 judgements per plane , for a total of 47 , 520 trials per subject . 10 . 7554/eLife . 03722 . 007Figure 2 . Measuring human sensitivity to image statistics . ( A ) Synthetic binary images can be created that contain specified values of individual image statistic coordinates ( as shown here ) or specified values of pairs of coordinates ( Victor and Conte , 2012 ) . ( B ) To measure human sensitivity to image statistics , we generate synthetic textures with prescribed coordinate values but no additional statistical structure , and we use these synthetic textures in a figure/ground segmentation task ( See Victor and Conte , 2012 and ‘Materials and methods’ , Psychophysical methods ) . ( C ) For measurements along coordinate axes , test stimuli are built out of homogeneous samples drawn from the gamuts shown in A ( e . g . the target shown in B was generated from the portion of the gamut indicated by the red arrow in A; See ‘Materials and methods’ , Psychophysical methods , and Victor et al . , 2005; Victor and Conte , 2012; Victor et al . , 2013 ) . We assess the discriminability of these stimuli from white noise by measuring the threshold value of a coordinate required to achieve performance halfway between chance and perfect ( inset ) . A similar approach is used to measure sensitivity in oblique directions; here , two coordinate values are specified to create the test stimuli . The threshold values along the axes and in oblique directions define an isodiscrimination contour ( red dashed ellipse , main panel ) in pairwise coordinate planes . ( D ) Along individual coordinate axes , we find that sensitivities ( 1/thresholds ) are rank-ordered as {β\| , β−}>{β/ , β\}>α>{θ⌜ , θ⌝ , θ⌟ , θ⌞} , shown separately for four individual subjects . A single set of perceptual sensitivities is shown for ( β\| , β− ) , ( β/ , β\ ) , and ( θ⌞ , θ⌜ , θ⌝ , θ⌟ ) , since human subjects are equally sensitive to rotationally-equivalent pairs of second-order coordinates and to all third-order coordinates ( Victor et al . , 2013 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 007 Figure 2D shows perceptual sensitivities measured along each coordinate axis . For each of four subjects , a similar pattern emerges for sensitivities as was observed for variation in natural image statistics: sensitivities are rank-ordered as {β\| , β−}>{β/ , β\}>α>{θ⌜ , θ⌝ , θ⌟ , θ⌞} . Note that the difference between the sensitivities in the horizontal and vertical directions ( β− and β\| ) vs the diagonal directions ( β\ and β/ ) is not simply an ‘oblique effect’ , that is , a greater sensitivity to cardinally- vs obliquely-oriented contours ( Campbell et al . , 1966 ) . Horizontal and vertical pairwise correlations differ from the diagonal pairwise correlations in more than just orientation: pixels involved in horizontal and vertical pairwise correlations share an edge , while pixels involved in diagonal pairwise correlations only share a corner . Correspondingly , the difference in sensitivities for horizontal and vertical correlations vs diagonal correlations is approximately 50% , which is much larger than the size of the classical oblique effect ( 10–20% ) ( Campbell et al . , 1966 ) . Figures 1E and 2D show a rank-order correspondence between natural image statistics and perceptual sensitivities . This qualitative comparison can be converted to a quantitative one ( Figure 3A ) , as a single scaling parameter aligns the standard deviation of natural image statistics with the corresponding perceptual sensitivities . In this procedure , each of the six image analyses is scaled by a single multiplicative factor that minimizes the squared error between the set of standard deviations and the set of subject-averaged sensitivities ( see ‘Materials and methods’ , Image preprocessing , and Figure 3—figure supplement 1 for additional details regarding scaling ) . The agreement is very good , with the mismatch between image analyses and human psychophysics comparable to the variability from one image analysis to another , or from one human subject to another . 10 . 7554/eLife . 03722 . 008Figure 3 . Variation in natural images predicts human perceptual sensitivity . ( A ) Scaled degree of variation ( standard deviation ) in natural image statistics along second- ( β ) , third- ( θ ) , and fourth-order ( α ) coordinate axes ( blue circular markers ) are shown in comparison to human perceptual sensitivities measured along the same coordinate axes ( red square markers ) . Degree of variation in natural image statistics is separately shown for different choices of the block-average factor ( N ) and patch size ( R ) used during image preprocessing . Perceptual sensitivities are separately shown for four individual subjects . As in Figure 2C , A single set of perceptual sensitivities is shown for {β\| , β−} , {β/ , β\} , and {θ⌞ , θ⌜ , θ⌝ , θ⌟} . ( B ) For each pair of coordinates , we compare the precision matrix ( blue ellipses ) extracted from natural scenes ( using N = 2 , R = 32 ) to human perceptual isodiscrimination contours ( red ellipses ) . Coordinate planes are organized into a grid . The set of ellipses in each pairwise plane is scaled to maximally fill each portion of the grid; agreement between the variation along single coordinate axes and the corresponding human sensitivities ( shown in A ) guarantees that no information is lost by scaling . Across all 36 coordinate planes , there is a correspondence in the shape , size , and orientation of precision matrix contours and perceptual isodiscrimination contours . ( C ) Quantitative comparison of a single image analysis ( N = 2 , R = 32 ) with the subject-averaged psychophysical data . For single coordinates depicted in A , we report the standard deviation in natural image statistics ( upper row ) and perceptual sensitivities ( middle row ) . For sets of coordinate planes depicted in ( B ) , we report the ( average eccentricity , angular tilt ) of precision matrix contours from natural scenes ( upper row ) and isodiscrimination contours from psychophysical measurements ( middle row ) . The degree of correspondence between predictions derived from natural image data and the psychophysical measurements can be conveniently summarized as a scalar product ( see text ) , where 1 indicates a perfect match . In all cases , the correspondence is very high ( 0 . 938–0 . 999 ) and is highly statistically significant ( p ≤ 0 . 0003 for both single coordinates and pairwise coordinate planes; see ‘Materials and methods’ , Permutation tests , for details ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 00810 . 7554/eLife . 03722 . 009Figure 3—figure supplement 1 . Scaling of natural image analyses . We scale each image analysis by a single scale factor that minimizes the squared error between the set of nine standard deviations and the set of nine psychophysical sensitivities . The scale factors are shown here as a function of block-average factor N for different choices of the patch size R . We find that the variance of image statistics decreases with increasing values of N , and thus larger values of N require a larger scale factor . Similarly , for a given value of N , the variance of image statistics increases with increasing R , and thus larger values of R require a larger scale factor . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 00910 . 7554/eLife . 03722 . 010Figure 3—figure supplement 2 . Covariation in natural image statistics predicts human isodiscrimination contours . ( A ) For each pair of coordinates , we compare the precision matrix ( blue ellipses ) extracted from natural scenes ( using N =2 , R = 32 ) to human perceptual isodiscrimination contours ( red ellipses ) . Coordinate planes are organized into a grid , with subject-averaged and subject-specific isodiscrimination contours shown respectively above and below the diagonal of the grid . Across all 36 coordinate planes , there is a correspondence in the shape , size , and orientation of precision matrix contours and perceptual isodiscrimination contours . The quality of the match is quantified by computing the angular tilt ( B ) and eccentricity ( C ) of image-statistic contours ( blue circular markers; shown for variations in the block-average factor ( N ) and patch size ( R ) used during image preprocessing ) and of perceptual isodiscrimination contours ( red square markers; shown for individual subjects ) . Since contours are highly similar within subsets of coordinate planes ( denoted by blocks in A; e . g . the set of θα planes ) , contour properties have been averaged within such subsets . Angular tilt and eccentricity are highly consistent between precision matrix contours and perceptual isodiscrimination contours ( except for near-circular contours , for which tilt is poorly-defined , as in the case denoted by an arrow ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 01010 . 7554/eLife . 03722 . 011Figure 3—figure supplement 3 . Principal axes of variation in natural images predict principal axes of perceptual sensitivity . Principal axes {ξ→NI} of variation in the distribution of natural image statistics are shown in comparison to the principal axes {ξ→PP} of human sensitivity . Each of the nine principal axes is represented by a vertical gray/white column . Markers ( circular = variation in natural image coordinates; square = human perceptual sensitivity ) represent the fractional power of the contributions of ( A ) second-order cardinal ( β\| , β− ) , ( B ) second-order oblique ( β/ , β\ ) , ( C ) third-order ( θ ) , and ( D ) fourth-order ( α ) coordinates to each principal axis; all contributions within each column sum to 1 . Principal axes components , and the range of variability observed across image analysis variants or across subjects ( see legend ) , are shown in blue for natural scene statistics and in red for perceptual sensitivities . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 01110 . 7554/eLife . 03722 . 012Figure 3—figure supplement 4 . Mapping ellipse shapes to the quarter unit sphere . We describe an ellipse by the unit vector ω→=sin α cos δ x^+sin α sin δ y^+cos α z^ , where ϵ=sin α is the eccentricity and δ is the angular tilt . In spherical coordinates , the tilt δ is the polar angle defined in the x−y plane , and the angle α=sin−1 ( ϵ ) is the azimuthal angle measured from the z-direction . In this representation , the unit vector z^ corresponds to a circle , and the unit vectors x^ and y^ correspond , respectively , to the ellipses that have been maximally elongated ( i . e . , into lines ) in the x^ and y^ directions . Points between the equator ( in the x−y plane ) and the pole correspond to ellipses of intermediate eccentricities . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 01210 . 7554/eLife . 03722 . 013Figure 3—figure supplement 5 . Single coordinate axes: variation in natural images predicts human perceptual sensitivities . Scaled variation in natural image statistics measured along second- ( β ) , third- ( θ ) , and fourth-order ( α ) coordinate axes ( blue circular markers ) are shown in comparison to human perceptual sensitivities measured along the same coordinates ( red square markers ) . Natural image statistics are extracted from the Penn natural image database ( A ) and the van Hateren image database ( B ) . Ranges of variation and human sensitivities are robustly rank-ordered as β\|−>β\/>α>θ . When each image analysis is scaled by a single factor , ranges match sensitivities . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 01310 . 7554/eLife . 03722 . 014Figure 3—figure supplement 6 . Pairwise coordinate planes in Penn Natural Image Database: covariation in natural images predicts human isodiscrimination contours . ( A ) For each pair of coordinates , we compare the precision matrix ( blue ellipses ) extracted from natural scenes ( using N = 2 , R = 32 ) to human perceptual isodiscrimination contours ( red ellipses ) . A precision matrix is represented by the contour lines of its inverse ( the covariance matrix M ) ; these are the points ( x , y ) at which Mxxx2+2Mxyxy+Myyy2= constant . A short distance of the blue contour from the origin thus indicates a large value of M and a small value of the precision matrix . This in turn denotes a direction in which prior knowledge of the image statistic is imprecise . Our prediction is that psychophysical thresholds ( red ellipses ) should match these contours . Coordinate planes are organized into a grid , with subject-averaged and subject-specific isodiscrimination contours shown respectively above and below the diagonal of the grid . Across all 36 pairwise coordinate planes , there is a correspondence in the shape , size , and orientation of precision matrix contours and perceptual isodiscrimination contours . The quality of the match is quantified by computing the ( B ) angular tilt and ( C ) eccentricity of image-statistic contours ( circular markers ) and of perceptual isodiscrimination contours ( square markers ) . Since contours are highly similar within subsets of pairwise planes ( denoted by blocks in A; e . g . the set of θα planes ) , contour properties have been averaged within such subsets . Angular tilt and eccentricity are highly consistent between precision matrix contours and perceptual isodiscrimination contours . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 01410 . 7554/eLife . 03722 . 015Figure 3—figure supplement 7 . Pairwise coordinate planes in van Hateren Image Database: covariation in natural images predicts human isodiscrimination contours . ( A ) For each pair of coordinates , we compare the precision matrix ( blue ellipses ) extracted from natural scenes ( using N = 2 , R = 32 ) to human perceptual isodiscrimination contours ( red ellipses ) . A precision matrix is represented by the contour lines of its inverse ( the covariance matrix M ) ; these are the points ( x , y ) at which Mxxx2+2Mxyxy+Myyy2= constant . A short distance of the blue contour from the origin thus indicates a large value of M and a small value of the precision matrix . This in turn denotes a direction in which prior knowledge of the image statistic is imprecise . Our prediction is that psychophysical thresholds ( red ellipses ) should match these contours . Coordinate planes are organized into a grid , with subject-averaged and subject-specific isodiscrimination contours shown respectively above and below the diagonal of the grid . Across all 36 pairwise coordinate planes , there is a correspondence in the shape , size , and orientation of precision matrix contours and perceptual isodiscrimination contours . The quality of the match is quantified by computing the ( B ) angular tilt and ( C ) eccentricity of image-statistic contours ( circular markers ) and of perceptual isodiscrimination contours ( square markers ) . Since contours are highly similar within subsets of pairwise planes ( denoted by blocks in A; e . g . the set of θα planes ) , contour properties have been averaged within such subsets . Angular tilt and eccentricity are highly consistent between precision matrix contours and perceptual isodiscrimination contours . Coordinates extracted from the van Hateren database show larger variability in the β\|β− and β\β/ planes than those extracted from the Penn Natural Image Database ( Figure 3—figure supplement 6 ) , exhibiting a larger number of low-eccentricity contours for which tilt is poorly defined . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 01510 . 7554/eLife . 03722 . 016Figure 3—figure supplement 8 . Principal axes of variation across natural images predict principal axes of human perceptual sensitivity in the full coordinate space . Principal axes {ξ→NI} of variation in the distribution of natural image statistics are shown in comparison to the principal axes {ξ→PP} of human sensitivity . Each of the nine principal axes is represented by a vertical gray/white column . Markers ( circular = variation in natural image coordinates; square = human perceptual sensitivity ) represent the fractional power of the contributions of ( A , E ) second-order cardinal ( β\|− ) , ( B , F ) second-order oblique ( β/\ ) , C , G third-order ( θ ) , and ( D , H ) fourth-order ( α ) coordinates to each principal axis; all contributions within each column sum to 1 . Principal axes components , and the range of variability observed across image analysis variants or across subjects ( see legend ) , are shown in blue for natural scene statistics and in red for perceptual sensitivities . There is an excellent match between the blue and red components for both the Penn and van Hateren image databases . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 01610 . 7554/eLife . 03722 . 017Figure 3—figure supplement 9 . Asymmetries in natural image statistics . ( A ) Probability distributions of natural image statistics . Projections of the distribution along second- and fourth-order coordinate axes are asymmetric about the origin , being shifted toward positive values . ( B ) We compute the ratio of standard deviations measured along positive vs negative coordinate axes ( circular markers ) to the ratio of human sensitivities measured along positive vs negative coordinate axes ( square markers ) . Natural images show larger asymmetries in second- and fourth-order coordinate values than is observed in human sensitivities . This is particularly notable for the α coordinate , which shows a 2–6 fold asymmetry in natural images variation but at most a 1 . 2-fold asymmetry in human sensitivity . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 017 We quantify the correspondence between image analyses and psychophysical analyses by computing the scalar product between the normalized vector of standard deviations ( extracted separately from each image analysis ) and the normalized vector of subject-averaged sensitivities ( extracted from the set of psychophysical analyses ) . A value of 1 indicates perfect correspondence , and 0 indicates no correspondence . This value ranges from 0 . 987 to 0 . 999 across image analyses and is consistently larger than the value measured under the null hypothesis that the apparent correspondence between statistics and sensitivities is chance ( p ≤ . 0003 for each image analysis; see Tables 1–2 and ‘Materials and methods’ , Permutation tests , for details regarding statistical tests ) . These findings support our hypothesis that human perceptual sensitivity measured along single coordinate axes ( assessed using synthetic binary textures ) is predicted by the degree of variation along the same coordinate axes in natural scenes . The correspondence shown in Figure 3A considers each image statistic coordinate in isolation . However , it is known that image statistics covary substantially in natural images ( as diagrammed in Figure 1D ) and also that they interact perceptually ( as diagrammed in Figure 2C ) . When pairs of natural image statistics covary , thus sampling oblique directions not aligned with the coordinate axes in the space of image statistics , our hypothesis predicts that human perceptual sensitivity is matched to both the degree and the direction of that covariation ( we are referring here to the orientation of a distribution in the coordinate plane of a pair of image statistics , and not to an orientation in physical space ) . To test this idea , we proceeded as follows . First , we fit the distribution of image statistics with a multidimensional Gaussian . When projected into pairwise coordinate planes , the isoprobability contours of this Gaussian capture the in-plane shape and orientation of the covariation of the distribution . Along single coordinate axes , the variation in natural image statistics predicts human perceptual sensitivities , as we have shown ( Figure 3A ) . More generally , we would predict that sensitivity should be be high along directions in which the distribution of natural image statistics has high standard deviation , because in those directions , the position of a sample cannot be guessed . Within coordinate planes , the quantitative statement of this idea is that the inverse covariance matrix , or precision matrix , predicts perceptual isodiscrimination contours . Sensitivity is expected to be low ( and therefore threshold high ) along directions in which the precision matrix has a high value and the position of a sample can be guessed a priori . Results in each coordinate plane are shown in Figure 3B . Across all subjects and all coordinate planes , we find that the shape and orientation of perceptual isodiscrimination contours ( red ellipses ) are predicted by the distribution of image statistics extracted from natural scenes ( blue ellipses ) . As in Figure 3A , the correspondence is very good , with mismatch that is comparable to the variability observed across image analyses and across subjects . To quantify the correspondence between natural image and psychophysical analyses , we describe each ellipse by a single vector ω→ that combines information about shape ( eccentricity ) and orientation ( angular tilt ) , and we compute the scalar product between the image analysis vector ω→NI and the subject-averaged psychophysical vector ω→PP . This value , averaged across coordinate planes , ranges from 0 . 953 to 0 . 977 across image analyses . We compared this correspondence to that obtained under the null hypotheses that ( i ) the apparent correspondence between image statistic covariances and isodiscrimination contours is chance , or ( ii ) the apparent covariances in image statistics are due to chance . The observed correspondence is much greater than the value measured under either null hypothesis ( p ≤ . 0003 for each image analysis under both hypotheses; see ‘Materials and methods’ , Analysis of image statistics in pairwise coordinate planes , and Figure 3—figure supplement 2 for comparisons of eccentricity and tilt , and Tables 1–3 and ‘Materials and methods’ , Permutation tests , for statistical tests ) . These findings confirm that the shape and orientation of human isodiscrimination contours , measured across all pairwise combinations of coordinates , can be quantitatively predicted from the covariation of image statistics extracted from natural scenes . The observed correspondence is maintained within the full 9-dimensional coordinate space ( see ‘Materials and methods’ , Analysis of the full 9-dimensional distribution of image statistics , and Figure 3—figure supplement 3 for principal component analyses , and Tables 1–3 and ‘Materials and methods‘ , Permutation tests , for statistical tests ) , confirming that our hypothesis describes human sensitivity in the full 9-dimensional space of local image statistics extracted from natural scenes . Although we did not record cortical responses directly , several lines of evidence indicate that that the perceptual thresholds we measured are determined by cortical processes . First , the stimuli had high contrast ( 100% ) and consisted of pixels that were readily visible ( 14 arcmin ) , so retinal limitations of contrast sensitivity and resolution were eliminated . Second , the task requires pooling of information over wide areas ( 100–200 pixels , that is , a region whose diameter is 10–15 times the width of an image element; see Figure 7 in Victor and Conte , 2005 ) . Retinal receptive fields are unlikely to do this , as the ratio of their spatial extent ( surround size ) to their resolution ( center size ) is typically no more than 4:1 ( Croner and Kaplan , 1995; Kremers et al . , 1995 ) . Third , to account for the specificity of sensitivity to three- and four-point correlations , a cascade of two linear-nonlinear stages is required ( Victor and Conte , 1991 ) ; retinal responses are quite well-captured by a single nonlinear stage ( Nirenberg and Pandarinath , 2012 ) , and cat retinal populations show no sensitivity to the four-point correlations studied used here ( Victor , 1986 ) while simultaneous cortical field potentials do . Conversely , macaque visual cortical neurons ( Purpura et al . , 1994 ) , especially those in V2 , manifest responses to three- and four-point correlations ( Yu et al . , 2013 ) . Successive stages of sensory processing share the same broad goals: invest resources in encoding stimulus features that are sufficiently informative , and suppress less-informative ones . In the periphery , this is exemplified by the well-known suppression of very low spatial frequencies; in cortex , this is exemplified by insensitivity to high-order correlations that are predictable from lower-order ones . Previous work has shown that such higher-order correlations can be separated into two groups—informative and uninformative—and only the informative ones are encoded ( Tkačik et al . , 2010 ) . We used this finding to select an informative subspace for the present study , and we asked how resources should be efficiently allocated amongst features within this informative subspace . A simple model of efficient coding by neural populations is shown in Figure 4A ( details in ‘Materials ans methods’ , Two regimes of efficient coding ) . Here , to enable analytical calculations , we used linear filters of variable gain and subject to Gaussian noise to model a population of neural channels encoding different features . The optimal allocation of resources to maximize information transmitted by the population depends on the amount of input noise , the amount of output noise , the input signal variability , and the total resources available to the system , here quantified as a constraint on the total output power ( i . e . , sum of response variances ) in the neural population . The constrained output power and the output noise together determine the ‘bandwidth’ of the system—that is , the expressive capacity of its outputs . Consider a neural population with input noise , output noise , and a fixed amount of output power . We find that when input signal variability is sufficiently large compared to the input noise , the gain of neurons should decrease with the variance of the input ( regions to the right of the peaks in the right-hand panel of Figure 4A ) . This is a regime where the output bandwidth is low compared to the input range , and efficient coding predicts that signals should be ‘whitened’ by equalizing the variance in different channels . Conversely , consider input signals with a smaller range , which are thus more disrupted by input noise . In this case , the gain of neurons should increase with the variance of the input ( regions to the left of the peaks in the right-hand panel of Figure 4A ) . This is a regime where the input noise dominates , and efficient coding predicts that the system should invest more resources in more variable , and hence more easily detectable , input signals . The relative sizes of input and output noise ( controlled by Λ in Figure 4A ) determines the input ranges over which the two qualitatively different regimes of efficient coding apply . 10 . 7554/eLife . 03722 . 018Figure 4 . Regimes of efficient coding . ( A ) To analyze different regimes of efficient coding , we consider a set of channels , where the kth channel carries an input signal with variability sk . Gaussian noise is added to the input . The result is passed through a linear filter with gain \|Lk\| , and then Gaussian noise is added to the filter output . We impose a constraint on the total power output of all channels , that is , a constraint on its total resources . With these assumptions , the set of gains that maximizes the transmitted information can be determined ( see ‘Materials and methods’ , Two regimes of efficient coding , and ( van Hateren , 1992a; Doi and Lewicki , 2011; Doi and Lewicki , 2014 ) ) . This set of gains depends on the relative strengths of input and output noise and on the severity of the power constraint , quantified here by the dimensionless parameter Λ ( right-hand panel ) . As Λ decreases from 1 to 0 , the system moves from a regime in which output noise is limiting to one in which input noise is limiting . ( B ) The efficient coding model applied to the sensory periphery . Raw luminances from natural images are corrupted with noise ( e . g . shot noise resulting from photon incidence ) and passed through a linear filter . The resulting signal is carried by the optic nerve , which imposes a strong constraint on output capacity . In the bandwidth limited case where output noise dominates over input noise ( e . g . , under high light conditions when photon noise is not limiting ) , the optimal gain decreases as signal variability increases . Since channel input and channel gain vary reciprocally , channel outputs are approximately equalized , resulting in a ‘whitening’ , or decorrelation . ( C ) The efficient coding model applied to cortical processing . Informative image features resulting from early cortical processing , caricatured by our preprocessing pipeline as applied to the retinal output , are sampled from a spatial region of the image . This sampling acts as a kind of input noise , because it only provides limited count-based estimates for the true statistical properties of the image source . When this input noise is limiting , the optimal gain increases as signal variability increases . Rather than whiten , the output signals preserve the correlational structure of the input . Note that in both regimes ( B ) and ( C ) , there is a range of signals that are not encoded at all . These are the signals that are not sufficiently informative to warrant an allocation of resources . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 01810 . 7554/eLife . 03722 . 019Figure 4—figure supplement 1 . Schematic representation of channel optimization problem . We consider a set of channels , each of which is dedicated to processing an independent signal sk . Sampling noise ( taken here to be unity ) is added to the signal sk , which is then passed through a linear filter Lk with gain \|Lk\| . Channel noise ( taken here to be unity ) is added to the output of Lk . The total dynamic range of all channels is constrained . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 01910 . 7554/eLife . 03722 . 020Figure 4—figure supplement 2 . Optimal coding regimes . Optimal gain \|Lk\| is shown as a function of signal strength sk for different choices of the output constraint Λ . For signals below a critical strength Λ/ ( 1−Λ ) , the optimal gain is zero , and signals are not encoded . The limit Λ→1 from below defines the transmission-limited regime , while the limit Λ→0 from above defines the sampling-limited regime . ( A ) Transmission-limited regime . For signal strengths much larger than the critical value , the main constraint is output power , and the optimal gain is inversely proportional to the signal strength ( as indicated by the dotted line with negative slope ) . As Λ→1 , there is an increasingly sharp transition between signals that are not encoded , and signals that are encoded in inverse proportion to their size ( ‘whitened’ ) . ( B ) Sampling-limited regime . As Λ→0 , there is a broadening of the transition between signals that are not encoded , and signals that are whitened . This broadening results in a regime in which sampling-noise is the dominant constraint , and the optimal gain increases with signal strength ( as indicated by the dotted line with positive slope ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 02010 . 7554/eLife . 03722 . 021Figure 4—figure supplement 3 . Noise-dependent transition between efficient coding regimes . Total noise is the sum of sampling noise ( x-axis ) and channel noise ( y-axis ) . In the case considered here ( d = 2 ) , channel noise cannot exceed 0 . 5 , but sampling noise can . For total noise below 0 . 5 , the optimal filter L is antialigned with the signal , and the optimal strategy is decorrelation via whitening ( white region , transmission-limited regime ) . For total noise above 0 . 5 , the optimal filter is aligned with the signal ( black region , sampling-limited regime ) , consistent with our findings that perceptual sensitivity is tuned to the direction and degree of variation in natural image statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 02110 . 7554/eLife . 03722 . 022Figure 4—figure supplement 4 . Optimal filter shape and orientation . Tilt and eccentricity of the optimal linear filter L∗ for random choices of the input signal s . As the magnitudes of sampling and channel noises vary , there emerge two regimes of efficient coding: a transmission-limited regime ( A–B ) and a sampling-limited regime ( C–D ) . In the transmission-limited regime , the maximum filter eigendirection is aligned with the minimum signal eigendirection ( and hence there is a difference in tilt of π/2 ) . In contrast , in the sampling-limited regime , the maximum filter eigendirection is aligned with the maximum signal eigendirection . Note that a direct comparison of eccentricities between these two regimes can be misleading , due to a reversal of the maximal eigendirections . ( A ) Sampling noise Ξ=0 , channel noise Σ=0 . 2 . The optimal strategy is decorrelation via whitening using a filter aligned perpendicularly to the input signal ( right panel ) with an eccentricity that matches that of the input signal ( dashed line , left panel ) . ( B ) Sampling noise Ξ=0 . 1 , channel noise Σ=0 . At very low total noise , even with zero channel noise , the optimal strategy is still decorrelation ( right panel ) using a filter whose eccentricity is less than the eccentricity of the input signal . ( C ) Sampling noise Ξ=4 , channel noise Σ=0 ( the low input SNR regime identified in ( van Hateren , 1992a ) ) . The tilt of the optimal filter is aligned to the tilt of the signal ( right panel ) , and the filter eccentricity is approaching the prediction of the square-root gain relation ( curved dotted line , left panel ) with decreasing SNR . ( D ) Sampling noise Ξ=0 . 4 , channel noise Σ=0 . 35 ( dominating sampling noise ) . For increasing sampling noise strength , the filter eccentricities match the signal eccentricities ( dashed line , left panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 022 To make these abstract considerations concrete , we first considered coding in the sensory periphery . A common strategy employed in the periphery is ‘whitening’ , where relatively fewer resources are devoted ( yielding lower gain ) to features with more variation ( Olshausen and Field , 1996 ) . As an example , within the spatial frequency range that the retina captures well , sensitivity is greater for high spatial frequencies than for low ones , that is , sensitivity is inversely related to the degree of variation in natural scenes ( the well-known 1/f2 power spectrum [Olshausen and Field , 1996] ) . Figure 4B illustrates how this strategy can emerge from the simple efficient coding scheme discussed above as applied to peripheral sensory processing . Spatiotemporal correlations of light undergo filtering before passing through the optic nerve bottleneck ( a constraint on bandwidth ) . Such a constraint on bandwidth is equivalently understood as a regime where output noise is relatively large compared to input noise . In this limit , where output noise dominates over input noise , the optimal strategy is whitening ( See Srinivasan et al . , 1982 and Figure 4A ) . Of course , real neural systems contend with both input and output noise; indeed recent work has shown that simply whitening to deal with output noise underestimates the optimal performance that the sensory periphery can achieve ( Doi and Lewicki , 2014 ) . An alternative regime arises when input noise limits performance . In this regime , relatively more resources are devoted to features with more variation . This regime was discussed in early work of van Hateren , ( 1992a ) and was also recognized in ( Doi and Lewicki , 2011 , 2014 ) , although it has received much less attention than the ‘whitening’ regime . Our results suggest that this is the regime is relevant to cortex , where it predicts the relative allocation of resources to higher-order image statistics . Figure 4C illustrates the simple efficient coding scheme in this context . We use our image preprocessing pipeline to mimic early visual processing , and we consider the downstream coding of higher-order image features . Because these features must be sampled from a finite patch of an image , they are subject to input noise arising from fluctuations in statistical estimation . When such input noise is limiting , the ability to detect a signal from noise increases with the variability of that signal . In this limit , efficient coding predicts that resources should be allocated in proportion to feature variability ( Figure 4C ) . This captures the intuition that when signal reliability is in question , more reliable signals warrant more resources . Furthermore , if two or more channels have covarying signals , resources should be devoted in relation to the direction and degree of maximum covariance ( see ‘Materials and methods’ , Two regimes of efficient coding , Figure 4—figure supplement 3 , and Figure 4—figure supplement 4 ) . The difference between these two efficient coding regimes is a consequence of the form of noise—output vs input noise—that is limiting . Our finding that cortex operates in a different regime than the well-known peripheral whitening reflects the fact that different stages and kinds of processing can face different constraints . While information transmission by the visual periphery is limited by a bottleneck in the optic nerve , cortex faces no such transmission constraint . Furthermore , while faithful encoding may be an immediate goal of early visual processing , cortical circuits have to interpret image features from a complex and crowded visual scene and perform statistical inference . For example , to discriminate between various textures , the cortex cannot perform pixel-by-pixel comparisons , but must rely on the estimation of local correlations ( image statistics ) instead . Because these correlations must be sampled from a finite patch of the visual scene , any estimate will be limited by sampling fluctuations . Sampling fluctuations constitute a source of input noise , the magnitude of which depends on the size of the sampled region . For natural images , this gives rise to a tradeoff: small regions lead to large fluctuations in the estimated statistics , while large regions blur over local details . This blurring may obscure the boundaries between objects with different surface properties . While the brain must implement such sampling , the size , scale , and potentially dynamic nature of the sampling region is not known . Interestingly , our predictions of human sensitivities do not change substantially over a wide range of spatial scales and image patch sizes , perhaps reflecting a scaling property of natural images ( Stephens et al . , 2013 ) . An avenue for future research is to determine whether there is an optimal region size , and if so , whether it could be estimated from images themselves . Sampling limitations alone do not suffice to account for the observed differential sensitivity of the brain to local image statistics . Were sampling limitations the only consideration , perceptual sensitivity would be the same along each coordinate axis , and perceptual isodiscrimination contours would be circular in each coordinate plane . This follows from an ideal observer calculation ( See Appendix B of Victor and Conte , 2012 ) . In contrast , we find that human observers have a severalfold variability in sensitivity along different coordinate axes ( Figure 3A ) and have isodiscrimination contours that are elongated in oblique directions ( Figure 3B ) . The efficient coding principle can account for these findings by taking into consideration the fact that a real observer has finite processing resources . In this context ( finite resources and substantial input noise ) , the efficient coding principle predicts that resources are invested in relation to the range of signal values that are typically present ( van Hateren , 1992a ) , as we find . Interestingly , resource limitations seem to play an important role in the cortex despite the vast expansion in the number of neurons compared to the optic nerve . Presumably , this reflects the large number of complex features that could be computed and the corresponding need for a large overrepresentation of the stimulus space ( Olshausen and Field , 1997 ) . While we find a close match between the variation in natural image statistics and human psychophysical performance , some aspects of the distribution of natural image statistics do not match psychophysical data . These differences are not readily apparent when we examine the variances and covariances ( Figure 3 ) of the distribution of natural image statistics but emerge only when one considers its detailed shape ( see ‘Materials and methods’ , Asymmetries in distributions of natural image statistics ) . For example , the distribution of α-coordinate values has a longer tail in the positive vs negative direction ( see Figure 3—figure supplement 9 and ( Tkačik et al . , 2010 ) ) . In contrast , human perceptual sensitivity is symmetric , or very nearly so ( within ∼20% ) , for positive vs negative values of α ( Victor et al . , 2005; Victor and Conte , 2012; Victor et al . , 2013 ) . This suggests that limitations imposed by ‘neural hardware’ force the system to use heuristics instead of matching the natural image distribution exactly . For example , an opponent mechanism responsible for detecting variations along , example , the α coordinate , might be a useful and easy ( although imperfect ) way to process the asymmetric distribution of four-point correlations found in natural scenes . Such a mechanism could be matched to the variance of the natural image distribution along the α coordinate , but not to its skew or other odd moments . An opponent mechanism would necessarily give rise to equal sensitivities to positive vs negative values of α , as observed in psychophysical results . Further study of deviations from a perfect match to the distribution of natural image statistics might provide additional insight into these or other possible neural mechanisms , and into the goals of the computations . Independently , our results also raise an interesting theoretical question about the optimal representation of non-gaussian , multidimensional signals under resource-limited conditions . Looking forward , we hypothesize that the principle of efficient coding might apply to cortical processing at higher levels . For example , more complex image features , such as shapes , are represented as conjunctions of contour fragments ( Brincat and Connor , 2004 ) , where each contour fragment is a local image object defined by particular multi-point correlations . We might speculate that the joint statistics of contour fragments in natural scenes can predict , through appropriate formulation of the same efficient coding principle used here , the properties of neurons in area IT ( Hung et al . , 2012; Yau et al . , 2012 ) or the associated perceptual sensitivities of human observers . Finally , although we have focused on perception of image statistics , we do this with the premise that this process is in the service of inferring the materials and objects that created an image and ultimately , guiding action . Thus , it is notable that we found a tight correspondence between visual perception and natural scene statistics without considering a specific task or behavioral set . Indeed , the emergence of higher-order percepts without explicit task specification was the original hope of the efficient coding framework as first put forward by Barlow and Attneave ( Attneave , 1954; Barlow , 1959 , 1961 ) . Doubtless , these ‘top-down’ factors also influence the visual computations that underlie perception , and the nature and site of this influence are an important focus of future research . We determined perceptual sensitivity to local image statistics via a texture segmentation paradigm adapted from ( Chubb et al . , 2004 ) , and in standard use in our lab ( Victor et al . , 2005; Victor and Conte , 2012; Victor et al . , 2013 ) ; we describe it briefly here . These measurements were carried out in parallel with the natural scene analysis described above . Some of the psychophysical results have been previously reported ( Victor and Conte , 2012; Victor et al . , 2013 ) ; see ‘Subjects’ below . In pairwise coordinate planes , our hypothesis predicts that the inverse covariance matrix , or precision matrix , matches human isodiscrimination contours . A precision matrix is represented by the contour lines of its inverse ( the covariance matrix M ) ; these are the points ( x , y ) at which Mxxx2+2Mxyxy+Myyy2= constant . A short distance of this contour from the origin thus indicates a large value of M and a small value of the precision matrix . This in turn denotes a direction in which prior knowledge of the image statistic is imprecise . Figure 3B shows a correspondence between contours of the precision matrix ( extracted from natural images ) and human isodiscrimination contours . This is shown again here in Figure 3—figure supplement 2A for subject-specific ( lower half grid ) and subject-averaged ( upper half grid ) isodiscrimination contours . This correspondence can be made quantitative by computing the angular tilt ( Figure 3—figure supplement 2B ) and eccentricity ( Figure 3—figure supplement 2C ) of each ellipse . Across all 36 pairwise coordinate planes , we find a detailed quantitative match between the shape and orientation of precision matrix contours and human isodiscrimination contours . Our results , shown in Figure 3 for single coordinates and pairwise coordinate planes , and extended to the full 9-dimensional distribution in Figure 3—figure supplement 3 , show a consistent match between the variation in natural image statistics and psychophysical sensitivities . We quantify this match by first assigning vectors to the quantities shown in Figure 3 and Figure 3—figure supplement 3 , and then computing the overlap between natural image vectors and the corresponding psychophysical vectors . We consider the following vector quantities:Single coordinates: We describe the range of variation in natural image statistics by the normalized 9-component vector of standard deviations σ→NI/\|\|σ→NI\|\| , where \|\|v→\|\| denotes the L2 norm 1N∑i=1Nvi2 of a vector v→ . Similarly , we describe the set of perceptual sensitivities by the normalized vector s→PP/\|\|s→PP\|\| . In both cases , the vector components are measured with respect to the coordinates {β\| , β− , β\ , β/ , θ⌜ , θ⌝ , θ⌟ , θ⌞ , α} . Pairwise coordinate planes: We describe each ellipse by the unit vector ω→ that is a combined measure of eccentricity ( ∈ ) and tilt ( δ ) . We define ω→ on one quarter of the unit sphere: ω→=sin α cos δ x^+sin α sin δ y^+cos α z^ , where ϵ=sin α and cos δ are defined on the interval [0 , 1] ( the second follows from the 180° rotational symmetry of ellipses ) . Note that this definition of ω→ captures the ellipse property that when ϵ=sin α=0 ( circular ellipses ) , δ is not defined . See Figure 3—figure supplement 4 for a schematic of this representation . Principal components: We consider two related measures for describing principal components . As shown in Figure 3—figure supplement 3 , we describe each principal component {ξ→ ( i ) } by the normalized vector f→ ( i ) /\|\|f→ ( i ) \|\| , which measures the fractional contribution of sets of statistics to the principal components ξ→ ( i ) . For a more detailed comparison , we can similarly describe each principal component by the normalized vector F→ ( i ) /\|\|F→ ( i ) \|\| , where F→ ( i ) =[fβ\| ( i ) , fβ− ( i ) , fβ\ ( i ) , fβ/ ( i ) , fθ⌜ ( i ) , fθ⌝ ( i ) , fθ⌟ ( i ) , fθ⌞ ( i ) , fα ( i ) ] . This measures the fractional contribution of individual statistics ( rather than sets of statistics ) to the principal components ξ→ ( i ) . For each vector quantity ( σ→ , ω→ , f→ , and F→ ) , we compute the scalar product between a given image analysis vector and the subject-averaged psychophysical vector . We then report the overlap values ( scalar products ) measured for the six image analyses considered Figures 1 and 3 ( N = 2 , 4 and R = 32 , 48 , 64 ) . In computing the scalar product between ω→NI and ω→PP , we report the overlap averaged over all 36 pairwise coordinate planes . Similarly , in computing the overlap between f→NI and f→PP and between F→NI and F→PP , we report the overlap averaged over all 9 principal components . Note that , for each vector σ→ , ω→ , f→ , and F→ , the maximum overlap is 1 . We find that natural image analyses show consistently high overlap with the set of psychophysical results ( see Tables 1–3 ) . The overlap , as measured across image analyses , ranges from 0 . 988 to 0 . 999 for single coordinates ( σ→ ) , from 0 . 953 to 0 . 977 for pairwise coordinate planes ( ω→ ) , from 0 . 987 to 0 . 993 for fractional principal axes ( f→ ) , and from 0 . 829 to 0 . 917 for the full principal axes ( F→ ) . We test the significance of this overlap by comparing our results to the following two null models:1A . Shuffled coordinate labels: sets of coordinates . This model ( and model 1b ) tests the null hypothesis that the apparent correspondence between image statistic covariances and isodiscrimination contours is chance . We examine the 23 permutations of the sets of coordinates {β\|− , β\/ , θ , α} . We apply these permutations to the psychophysical data , as human subjects are equally sensitive to coordinates within each set ( {β\| , β−} , {β\ , β/} and all θ's ) . This shuffling creates a new set of subjects whose second-order cardinal , second-order oblique , third-order , and fourth-order coordinate values are randomly permuted ( transforming the original vector [β\|− , β\/ , θ , α] into , example , the shuffled vector [β\/ , θ , β\|− , α] ) . If the correspondence between quantities derived from image analysis and psychophysics is statistically significant , we expect that the shuffled vectors σ→ , ω→ , f→ , and F→ will show less overlap with the image analysis vectors than do the original psychophysical vectors ( note that the limited number of permutations restricts the minimum p-value to be 0 . 04 ) . 1B . Shuffled coordinate labels: individual coordinates . Here , we expand the test described in 1a to randomly shuffle the full set of coordinate labels {β\| , β− , β\ , β/ , θ⌜ , θ⌝ , θ⌟ , θ⌞ , α} . In an analogous manner to that described in 2A , we expect that the shuffled vectors σ→ , ω→ , f→ , and F→ will show less overlap with the image analysis vectors than do the original psychophysical vectors if the correspondence between quantities derived from image analysis and psychophysics is statistically significant . 2 . Shuffled patch labels . This model tests the null hypothesis that the apparent covariances in image statistics are due to chance . For each coordinate , we randomly shuffle image patch labels . This shuffling creates a new set of null patches whose second- , third- , and fourth-order coordinate values are randomly drawn from a subset of the original image patches ( e . g . a given null patch can be described by a β/-value measured from patch m but an α value measured from patch n ) . This shuffling destroys correlations between coordinate values measured within individual patches . Note that this shuffling does not alter the range of variation measured along single coordinate axes and will therefore not alter the values of precision matrix ellipses measured along coordinate axes . As a result , this test is not applicable to σ→ , which measures natural image variation and human sensitivities along individual coordinate axes . However , shuffling will destroy correlations along oblique directions in coordinate planes , thereby aligning each ellipse along a single coordinate axis . Note that the eccentricity of each ellipse ( in , e . g . , the A-B plane ) is then trivially related to the ratio of variances σ2 measured along the corresponding coordinate axes: ϵ=1−σA2/σB2 . We therefore expect that this shuffling will most strongly affect the tilt and eccentricity in pairwise planes in which ellipses are oriented along oblique directions ( β\|β− , β\β/ , and θθ planes ) . Finally , in destroying correlations between pairs of coordinates , this shuffling creates a diagonal covariance matrix , such that principal components are aligned with single coordinate axes . If the correspondence between quantities derived from image analysis and psychophysics is statistically significant , we expect that the shuffled vectors ω→ , f→ , and F→ will show less overlap with the psychophysical vectors than do the original image analysis vectors . Each null model is constructed by randomly selecting permuted indices that independently shuffle coordinate labels for subject-averaged psychophysical data ( Null Model 1 ) and independently shuffle image patch labels for a given statistic ( Null Model 2 ) . For null model 1a , we perform the full set of 23 non-identity permutations . For models 1B and 2 , we perform 10 , 000 permutations . For each permutation , we compute a set of shuffled vectors {σ→ , ω→ , f→ , F→} , and we measure the overlap ( defined as the scalar product ( * ) →NI⋅ ( * ) →PP ) between each shuffled vector and the corresponding subject-averaged psychophysical vector . Note that , when assigning shuffled principal components to symmetry classes , no hand-tuning was performed . However , as described previously , such hand-tuning was only applied to a very small fraction of components for select image analyses . When repeated for many permutations , this procedure yields a distribution of shuffled overlap values against which we measure the significance of the true ( observed ) overlap . Significance values ( p-values ) are estimated by computing the fraction of permutations for which the shuffled overlap exceeds the true overlap . We find that the original image analyses show significantly higher overlap with psychophysical data than do the analyses produced by either of the null models . Results are significant for each measure of overlap and for each of the six analyses presented in Figures 1 and 3 ( p <0 . 0005 , or as small as possible given the number of possible permutations , in all cases ) ; see Tables 1–3 for full results . 10 . 7554/eLife . 03722 . 023Table 1 . Permutation tests for null model 1a: shuffled coordinate labelsDOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 023Measures of overlapImage analysisObserved overlapShuffled overlap ValuesSignificanceMeanstdminmaxRange/Sensitivity σ→NI⋅s→PPN = 2R = 320 . 9990 . 8590 . 9 × 10−10 . 7040 . 983<0 . 04R = 480 . 9930 . 8321 . 1 × 10−10 . 6510 . 978<0 . 04R = 640 . 9870 . 8091 . 1 × 10−10 . 6140 . 974<0 . 04N = 4R = 320 . 9980 . 8251 . 1 × 10−10 . 6380 . 969<0 . 04R = 480 . 9940 . 8121 . 1 × 10−10 . 6460 . 990<0 . 04R = 640 . 9910 . 7941 . 1 × 10−10 . 6170 . 985<0 . 04Inverse Range/Threshold 〈ω→NI⋅ω→PP〉N = 2R = 320 . 9710 . 7091 . 5 × 10−10 . 5080 . 924<0 . 04R = 480 . 9690 . 6921 . 6 × 10−10 . 4690 . 924<0 . 04R = 640 . 9530 . 6851 . 7 × 10−10 . 4500 . 913<0 . 04N = 4R = 320 . 9670 . 6791 . 7 × 10−10 . 4470 . 908<0 . 04R = 480 . 9750 . 6321 . 5 × 10−10 . 4000 . 880<0 . 04R = 640 . 9770 . 6481 . 6 × 10−10 . 4110 . 894<0 . 04Fractional Principal Components f→NI⋅f→PPN = 2R = 320 . 9940 . 3821 . 5 × 10−10 . 1600 . 657<0 . 04R = 480 . 9950 . 4851 . 2 × 10−10 . 2870 . 727<0 . 04R = 640 . 9910 . 4870 . 7 × 10−10 . 3720 . 632<0 . 04N = 4R = 320 . 9950 . 4591 . 4 × 10−10 . 2380 . 732<0 . 04R = 480 . 9960 . 4441 . 0 × 10−10 . 2770 . 601<0 . 04R = 640 . 9960 . 4501 . 1 × 10−10 . 2790 . 614<0 . 04Full Principal Components 〈F→NI⋅F→PP〉N = 2R = 320 . 9170 . 3161 . 3 × 10−10 . 1230 . 578<0 . 04R = 480 . 8280 . 4011 . 0 × 10−10 . 2280 . 611<0 . 04R = 640 . 9110 . 3630 . 7 × 10−10 . 2820 . 532<0 . 04N = 4R = 320 . 8820 . 3761 . 2 × 10−10 . 1800 . 618<0 . 04R = 480 . 9170 . 3621 . 0 × 10−10 . 2010 . 520<0 . 04R = 640 . 9190 . 3571 . 0 × 10−10 . 1960 . 522<0 . 04We separately permute the sets of coordinate labels {β\|− , β\/ , θ , α} . We apply these permutations to the psychophysical data , therein examining all 23 non-identity permutations of the four labels . This shuffling significantly decreases the overlap between image analyses and psychophysical data . Results are significant across all six analyses considered in Figures 1 and 3 ( N = 2 , 4 and R = 32 , 48 , 64 ) . p-values , estimated as the fraction of permutations for which the shuffled overlap exceeds the true overlap , are less than 0 . 04 ( the minimum value given 23 permutations ) for each image analysis . 10 . 7554/eLife . 03722 . 024Table 2 . Permutation Tests for null model 1b: shuffled coordinate labelsDOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 024Measures of overlapImage analysisObserved overlapShuffled overlap ValuesSignificanceMeanstdminmaxRange/Sensitivity σ→NI⋅s→PPN = 2R = 320 . 9990 . 8066 . 8 × 10−20 . 6590 . 9990 . 0003R = 480 . 9930 . 7757 . 7 × 10−20 . 6100 . 993<0 . 0001R = 640 . 9870 . 7628 . 0 × 10−20 . 5790 . 987<0 . 0001N = 4R = 320 . 9980 . 8286 . 0 × 10−20 . 7070 . 998<0 . 0001R = 480 . 9940 . 7987 . 1 × 10−20 . 6600 . 9940 . 0002R = 640 . 9910 . 7807 . 6 × 10−20 . 6300 . 991<0 . 0001Inverse Range/Threshold 〈ω→NI⋅ω→PP〉N = 2R = 320 . 9710 . 6938 . 1 × 10−20 . 4990 . 9720 . 0002R = 480 . 9690 . 6828 . 4 × 10−20 . 4760 . 9690 . 0003R = 640 . 9530 . 6718 . 5 × 10−20 . 4460 . 9540 . 0002N = 4R = 320 . 9670 . 6967 . 6 × 10−20 . 5210 . 964<0 . 0001R = 480 . 9750 . 6928 . 0 × 10−20 . 5090 . 9760 . 0002R = 640 . 9770 . 6898 . 2 × 10−20 . 4930 . 9780 . 0003Fractional Principal Components 〈f→NI⋅f→PP〉N = 2R = 320 . 9940 . 5921 . 2 × 10−10 . 2710 . 9950 . 0003R = 480 . 9950 . 6041 . 3 × 10−10 . 2810 . 9950 . 0004R = 640 . 9910 . 5911 . 2 × 10−10 . 2780 . 9910 . 0003N = 4R = 320 . 9950 . 5901 . 2 × 10−10 . 2180 . 9950 . 0001R = 480 . 9960 . 5771 . 2 × 10−10 . 2510 . 9960 . 0002R = 640 . 9960 . 5811 . 2 × 10−10 . 2660 . 9960 . 0004Full Principal Components 〈F→NI⋅F→PP〉N = 2R = 320 . 9170 . 3911 . 2 × 10−10 . 1000 . 9270 . 0002R = 480 . 8280 . 3911 . 2 × 10−10 . 0860 . 8560 . 0008R = 640 . 9110 . 3961 . 2 × 10−10 . 1200 . 9530 . 0003N = 4R = 320 . 8820 . 3811 . 2 × 10−10 . 0660 . 9890 . 0003R = 480 . 9170 . 3801 . 2 × 10−10 . 0900 . 902<0 . 0001R = 640 . 9190 . 3871 . 2 × 10−10 . 0950 . 9370 . 0004We separately permute all nine coordinate labels {β\| , β− , β\ , β/ , θ⌜ , θ⌝ , θ⌟ , θ⌞ , α} . This shuffling , applied to the psychophysical data , significantly decreases the overlap between image analyses and psychophysical data . Results are significant across all six analyses considered in Figures 1 and 3 ( N = 2 , 4 and R = 32 , 48 , 64 ) . p-values , estimated as the fraction of permutations for which the shuffled overlap exceeds the true overlap , are less than 0 . 0005 for all image analyses . 10 . 7554/eLife . 03722 . 025Table 3 . Permutation tests for null model 2: shuffled patch labelsDOI: http://dx . doi . org/10 . 7554/eLife . 03722 . 025ComparisonsImage analysisObserved overlapShuffled overlap ValuesSignificanceMeanstdminmaxInverse Range/Threshold 〈ω→NI⋅ω→PP〉N = 2R = 320 . 9710 . 9240 . 70 × 10−30 . 9210 . 926<0 . 0001R = 480 . 9690 . 9211 . 1 × 10−30 . 9170 . 925<0 . 0001R = 640 . 9530 . 9121 . 3 × 10−30 . 9080 . 917<0 . 0001N = 4R = 320 . 9670 . 9191 . 7 × 10−30 . 9140 . 926<0 . 0001R = 480 . 9750 . 9221 . 9 × 10−30 . 9160 . 930<0 . 0001R = 640 . 9770 . 9242 . 8 × 10−30 . 9160 . 935<0 . 0001Fractional Principal Components 〈f→NI⋅f→PP〉N = 2R = 320 . 9940 . 8069 . 1 × 10−60 . 8060 . 806<0 . 0001R = 480 . 9950 . 8068 . 3 × 10−60 . 8060 . 806<0 . 0001R = 640 . 9910 . 8063 . 7 × 10−60 . 8060 . 806<0 . 0001N = 4R = 320 . 9950 . 8072 . 5 × 10−40 . 8060 . 809<0 . 0001R = 480 . 9960 . 8074 . 1 × 10−40 . 8060 . 810<0 . 0001R = 640 . 9960 . 8073 . 5 × 10−40 . 8060 . 810<0 . 0001Full Principal Components 〈F→NI⋅F→PP〉N = 2R = 320 . 9170 . 4485 . 8 × 10−20 . 4060 . 596<0 . 0001R = 480 . 8280 . 5025 . 9 × 10−20 . 4080 . 675<0 . 0001R = 640 . 9110 . 4584 . 8 × 10−20 . 4070 . 591<0 . 0001N = 4R = 320 . 8810 . 4894 . 9 × 10−20 . 4090 . 638<0 . 0001R = 480 . 9170 . 4543 . 0 × 10−20 . 4080 . 637<0 . 0001R = 640 . 9190 . 4924 . 2 × 10−20 . 4110 . 648<0 . 0001Within each image analyses , we separately permute image patch labels along individual coordinate axes . This shuffling does not alter the range of variation observed along individual coordinates; as a result , this test only applies to ω→ , f→ and F→ . We find that this shuffling significantly decreases the overlap between image analyses and psychophysical data . Results are significant across all six analyses considered in Figures 1 and 3 ( N = 2 , 4 and R = 32 , 48 , 64 ) . p-values , estimated as the fraction of permutations for which the shuffled overlap exceeds the true overlap , are less than 0 . 0001 for each image analysis . In Figures 1 and 3 , we reported results using image analyses with varying values of the block-average factor N ( N = 2 , 4 ) and patch size R ( R = 32 , 48 , 64 ) . In Figure 1—figure supplement 3 , we show that the relative variation in different image statistics ( first shown in Figure 1E ) is not an artifact of our image analysis pipeline , as the pattern of variation is destroyed if white-noise image patches are instead used . In Figures 3–figure supplement 5-3–figure supplement 8 , we show that the comparison between natural image and psychophysical analyses is consistent across a wider range of image preprocessing parameters: N = 2 , 4 , 8 , 12 , 16 , 20 and R = 32 , 48 , 64 , 80 , 128 . Note that sampling limitations restrict some combinations of N and R ( e . g . for sufficiently large N , we must choose sufficiently small R to have a statistically significant number of image patches ) . All analyses reported in Results and shown in Figures 1 and 3 were performed on a set of images from the UPenn Natural Image Database ( Tkačik et al . , 2011 ) . Here , we extend our analyses to a set of 2300 images from the van Hateren image database ( van Hateren and van der Schaaf , 1998 ) , using the same set of parameters used to analyze images from the UPenn database , with block-average factors N = 2 , 4 , 8 , 12 , 16 , 20 and patch sizes R = 32 , 48 , 64 , 80 , 128 . Note that we are able to perform a larger number of analyses ( specific combinations of N and R ) than was performed using the Penn database , as we have a larger selection of images and therefore do not face the same sampling limitations . Figures 3—figure supplement 5-3–figure supplement 8 confirm that our results are consistent across image databases . We find systematic asymmetries in the distributions of natural image statistics when examined beyond their second moments . Figure 3—figure supplement 9 shows the distributions of single coordinates for the image analysis N = 2 , R = 32 . All distributions are shifted toward positive coordinate values , and there is larger variation in positive vs negative coordinate values . We assess this asymmetry in natural image analyses by computing the ratio of the standard deviations measured along positive vs negative coordinate axes . We similarly assess asymmetry in psychophysical analyses by computing the ratio of human sensitivities to positive vs negative deviations of coordinate values . This comparison is shown in Figure 3—figure supplement 9 . The mismatch provides potential clues for the neural mechanisms responsible for processing local image statistics ( See Discussion ) . In this section , we illustrate how two contrasting regimes emerge from the efficient coding principle: ( i ) the well-known transmission-limited regime , in which ‘whitening’ is optimal , and ( ii ) the sampling-limited regime , which is the focus of this paper . To enable exact calculations of optimal behavior , we consider a simplified scenario , in which all signals and noises are Gaussian , and all filters are linear . We consider a set of channels dedicated to processing independent signals of varying sizes . The channels , which are indexed by k , are abstract and general . For example , each k can represent a different spatial or temporal frequency in the input , as in the traditional analysis of visual coding in the periphery . Here , we take the signal on each channel k to represent a complex image feature , that is the result of a specific local nonlinear transformation applied to the input image . Figure 4—figure supplement 1 shows the setup of a single channel dedicated to processing the signal sk . Sampling noise , which is assumed to be identical for each channel , is added to this signal; without loss of generality , we can take its value to be unity . Note that for the parametrization of local image statistics used here , sampling noise is in fact identical for each parameter at the origin of the parameter space ( see Equations B19-B20 in Victor and Conte , 2012 ) . The result is passed through a linear filter Lk , characterized by a gain \|Lk\| . The output of Lk then has intrinsic channel noise added , and the total dynamic range of all channels is constrained . All channels are assumed to have the same intrinsic noise . Again , without loss of generality , we take this value to be unity ( as any scale associated with this noise can be absorbed into an overall multiplier for the filters Lk and the constraint on total dynamic range of the channels ) . We seek to find the optimal set of gains {\|Lk\|} that maximize the mutual information ∑kHk between the signals {sk} and the channel input , subject to a constraint Q on total output power . Using a Lagrange multiplier Λ for the constraint , the problems translates into extremizing P=∑kHk+ΛQ by setting ∂P/∂Lk=0 . The solution can be found in Equation 8 of van Hateren , ( 1992a ) , noting the following correspondences between the setup of Figure 4—figure supplement 1 and the scenario considered in that paper . Referring to the notation in ( van Hateren , 1992a ) , the input and channel noises , Np and Nc , respectively correspond here to the sampling and channel noises ( both taken to be unity ) . The prefiltered stimulus power Sp corresponds here to signal variance sk2 . The power transfer function pn of the neural filter corresponds here to the filter power \|Lk\|2 . Finally , the negative Lagrange multiplier −λ corresponds here to the positive Lagrange multiplier +Λ . With these correspondences , the optimal filter for channel k has a gain \|Lk\| given by: ( 0 . 7 ) \|Lk\|2=− ( 2+sk2 ) +sk4+4sk2/Λ2 ( 1+sk2 ) provided that the above quantity is non-negative , and has a gain of zero otherwise . The range of values of sk for which the above quantity is ≤0 corresponds to signals that are not worthwhile to code , because the signal-to-noise is too small given the constraint on the channel dynamic range . More specifically , the above quantity is positive ( and hence \|Lk\| is nonzero ) provided that sk>Λ/ ( 1−Λ ) . Note that this critical value becomes infinite as Λ approaches one from below , indicating that Λ near one is the transmission-limited regime . Conversely , the critical value of sk approaches zero as Λ approaches zero from above , indicating that this is the sampling-limited regime . We further discuss these regimes below . We interpret the gain \|Lk\| as representing the amount of resources devoted to a given signal sk . Since it is a direct measure of signal-to-noise for a unit-size input , it therefore corresponds to perceptual sensitivity . In the psychophysical experiments here , we measure sensitivity for each of the image statistic coordinates {β\| , β− , β/ , β\ , θ⌜ , θ⌝ , θ⌟ , θ⌞ , α} , using a highly artificial set of stimuli . As predicted from the sampling-limited regime , we find that gains \|Lk\| are larger for the channels in which the natural environment provides larger values of the signal sk . While this analysis provides a rigorous identification of a regime in which gain increases with signal strength , we caution that it is an asymptotic analysis of a simplified model of feature coding . It therefore stops short of making the quantitative prediction that gain ( sensitivity ) is proportional to the square root of the signal strength of each image statistic . On the other hand , the analysis does translate into a quantitative prediction about perceptual axes ( i . e . , about the orientations of the isodiscrimination contours ) . As shown in Figure 3 ( blue contours ) , the image statistic coordinates {β\| , β− , β/ , β\ , θ⌜ , θ⌝ , θ⌟ , θ⌞ , α} have substantial covariances . A rotation of the coordinates will thus yield a new set of coordinates with zero covariance and independent sampling errors . If these new coordinates are independently coded , then the perceptual axes will share the same axes as the image statistics which is what we find ( Figure 3B ) . 0≤Ξ≪Σ ( dominating channel noise ) . The optimal strategy is decorrelation by whitening ( Figure 4—figure supplement 4A ) ; the tilt of the filter relative to the signal is π/2 , and the eccentricities are equal ( i . e . , the small eigenvalue of L∗L∗T is proportional to the inverse of the large eigenvalue of S and vice versa , indicating that the gain scales as the inverse of the input power ) . 0<Ξ≪1 , Σ=0 ( zero channel noise , small sampling noise ) . The optimal strategy is still decorrelation ( Figure 4—figure supplement 4B ) with signal components of higher power being suppressed by the gain , but the suppression does not follow the inverse law as above . Ξ≥1 , Σ=0 ( zero channel noise , large sampling noise ) . The tilt of the filter matches the tilt of the signal , and the gain scales with input power . For high sampling noise and zero channel noise , the gain scales as the square-root of the input power ( Figure 4—figure supplement 4C ) . Ξ>Σ>0 ( dominating sampling noise ) . In a broad regime of noise strengths where sampling noise dominates over non-zero channel noise , the tilt of the gain matches the tilt of the signal , and the gain roughly scales with the input power ( Figure 4—figure supplement 4D ) . This regime is consistent with the correspondence that we observe between the natural scenes statistics and the psychophysical measurements .	Our senses are constantly bombarded by sights and sounds , but the capacity of the brain to process all these inputs is finite . The stimuli that contain the most useful information must therefore be prioritized for processing by the brain to ensure that we build up as complete a picture as possible of the world around us . However , the strategy that the brain uses to select certain stimuli—or certain features of stimuli—for processing at the expense of others is unclear . Hermundstad et al . have now provided new insights into this process by analyzing how humans respond to artificial stimuli that contain controllable mixtures of features that found in natural stimuli . To do this , Hermundstad et al . selected photographs of the natural world , and measured the brightness of individual pixels . After adjusting images in a way that mimics the human retina , the brightest 50% of the pixels in each photograph were colored white and the remaining 50% were colored black . Hermundstad et al . then used statistical techniques to calculate the degree to which the color of pixels could be used to predict the color of their neighbors . In this way , it was possible to calculate the amount of variation throughout the images , and then make computer-generated images in which pixel colorings were more or less predictable than in the natural images . Volunteers then performed a task in which they had to locate a computer-generated pattern against a background of random noise . The volunteers were able to locate this target most easily when it contained the same kinds of patterns and features that were meaningful about natural images . While this shows that the brain is adapted to prioritize features that are more informative about the natural world , understanding exactly how the brain implements this strategy remains a challenge .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Variance predicts salience in central sensory processing
Genetic redundancy and pleiotropism have limited the discovery of functions associated with miRNAs and other regulatory mechanisms . To overcome this , we performed an enhancer screen for developmental defects caused by compromising both global miRISC function and individual genes in Caenorhabditis elegans . Among 126 interactors with miRNAs , we surprisingly found the CED-3 caspase that has only been well studied for its role in promoting apoptosis , mostly through protein activation . We provide evidence for a non-apoptotic function of CED-3 caspase that regulates multiple developmental events through proteolytic inactivation . Specifically , LIN-14 , LIN-28 , and DISL-2 proteins are known miRNA targets , key regulators of developmental timing , and/or stem cell pluripotency factors involved in miRNA processing . We show CED-3 cleaves these proteins in vitro . We also show CED-3 down-regulates LIN-28 in vivo , possibly rendering it more susceptible to proteasomal degradation . This mechanism may critically contribute to the robustness of gene expression dynamics governing proper developmental control . The robustness of animal development is ensured by multiple regulatory mechanisms with overlapping roles acting on specific cellular processes , often manifested as genetic redundancy ( Fay et al . , 2002; Kitano , 2004; Felix and Wagner , 2008; Hammell et al . , 2009 ) . miRNAs mostly exert repression of gene expression by blocking target mRNA translation and/or through mRNA decay as part of the miRNA-induced-silencing complex ( miRISC ) , which includes GW182 and argonaute proteins ( Ding and Han , 2007; Fabian and Sonenberg , 2012 ) . miRNA-mediated gene silencing is a critical regulatory mechanism that ensures dynamic changes in gene expression during animal development or other physiological processes ( Ambros , 2004; Bartel and Chen , 2004 ) . However , specific physiological roles of individual miRNAs are often executed through the combinatory effects of multi-miRNA , multi-target mRNA networks ( Brenner et al . , 2010; Karp et al . , 2011; Kudlow et al . , 2012; Miska et al . , 2007; Parry et al . , 2007; Than et al . , 2013; Alvarez-Saavedra and Horvitz , 2010 ) . Moreover , these miRNA–mRNA interaction networks may act in concert , and often semi-redundantly , with other regulatory mechanisms to limit the expression of many genes involved in animal development or other physiological functions ( Figure 1A ) . Therefore , tackling genetic redundancy would be critical to uncover many specific functions associated with miRNAs and other gene expression regulatory mechanisms . 10 . 7554/eLife . 04265 . 003Figure 1 . Genome-wide RNAi screen for genes that cooperate with miRISCs to regulate development . ( A ) Rationale of enhancer screen strategy ( detailed in Figure 1—figure supplement 1A ) . ( B ) The number of genes identified as interactors of either/both ain-1 ( lf ) and/or ain-2 ( lf ) . ( C ) Distribution of the 126 interactors into functional categories ( interactors listed in Supplementary file 2 ) . ( D ) The proportion of genes exhibiting singular vs pleiotropic RNAi phenotypes with ain-1 ( lf ) or ain-2 ( lf ) ( detailed phenotypic frequencies shown in Figure 1—figure supplement 1B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 00310 . 7554/eLife . 04265 . 004Figure 1—figure supplement 1 . RNAi screen strategy and frequencies of phenotypes . ( A ) Cartoon diagram illustrating steps of the screen performed for the entire ORF RNAi library using liquid culture 96-well format . We included the RNAi-sensitizing mutation , rrf-3 ( lf ) , with ain-1 ( lf ) and ain-2 ( lf ) to increase screen sensitivity and therefore used rrf-3 ( lf ) alone for the control . In the double blind , we identified genes where the RNAi effect for the control was mostly normal but where ain-1 ( lf ) or ain-2 ( lf ) showed an obvious enhancer phenotype ( as defined in Supplementary file 1 ) . Example phenotypes body morphology defect ( Bmd ) , embryonic lethality ( Emb ) , and reduced brood size ( Red ) are depicted for illustration purposes only . Confirmations were performed in quadruplicate in the double blind . These interactors were all then revealed and sequence-verified . ( B ) Frequency of phenotypes observed in the RNAi screen . The three letter phenotypes indicated here are all defined in Supplementary file 1 and are depicted here as the frequency of occurrence amongst interactors for either ain-1 or ain-2 . Due to pleiotropism , the sum of the phenotypes will exceed 100% . All genes identified in the screen with individual phenotypes are listed in Supplementary file 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 004 We have carried out a genome-wide enhancer screen for genes that when knocked down would generate a strong developmental defect when general miRISC function is compromised . Among a large number of interactors identified from the screen is the ced-3 gene that encodes a caspase , well-characterized as a key component of the apoptotic pathway ( Conradt and Xue , 2005 ) . While ced-3 is absolutely required for the apoptotic process , null mutations of the gene are not associated with obvious developmental defects ( Hengartner , 1997 ) . However , two recent studies have reported different non-apoptotic roles of the ced-3 pathway , namely in stress-related neuronal function ( Pinan-Lucarre et al . , 2012 ) and aging ( Yee et al . , 2014 ) . Because no specific downstream targets of CED-3 were found in these studies , the mechanistic detail of such non-apoptotic functions of CED-3 remains unclear . Moreover , whether the CED-3 system is widely utilized to regulate animal development and other functions is a question of high significance . To uncover specific physiological functions of miRNAs and other regulatory mechanisms acting with miRNAs during development , we performed a genetic enhancer screen for developmental defects that manifested only when miRISC function and another regulatory mechanism were both compromised ( Figure 1A ) . We chose to use loss-of-function ( lf ) mutations of the ain-1 and ain-2 genes ( GW182 orthologs ) that each alone significantly compromises but does not eliminate global miRISC function ( Ding et al . , 2005; Zhang et al . , 2007 , 2009 ) . While the ain-1 ( lf ) mutant has a mild heterochronic phenotype and the ain-2 ( lf ) mutant is superficially wild-type , loss of both genes results in severe pleiotropic defects including alteration in temporal cell fate patterning . Therefore , an enhancer screen using the ain-1 ( lf ) or ain-2 ( lf ) mutant can potentially detect functions associated with most miRNAs . Using the entire Caenorhabditis elegans ORFeome RNAi feeding library ( Rual et al . , 2004 ) , we performed a double-blind screen that identified 126 genetic interactors ( Figure 1A–D , Figure 1—figure supplement 1 and Supplementary files 1 , 2 ) , of which only eight have been reported to interact with miRNA regulatory pathways ( Parry et al . , 2007 ) . Many interactions were confirmed by testing candidate mutants for phenotypes when treated with ain-1 and ain-2 RNAi ( Supplementary file 3 ) . Nearly two-thirds of the 126 genetic interactors were found to interact with both ain-1 and ain-2 genes ( Figure 1B ) . Gene ontology analysis revealed that these genes belong to a broad range of functional groups ( Figure 1C ) . Over-representation of genes associated with protein stability is consistent with the hypothesis that miRNAs act in concert with other repressive mechanisms to limit gene expression ( Figure 1A , C ) . We found that ain-1 ( lf ) displayed more pronounced pleiotropism with its interactors than ain-2 ( lf ) ( Figure 1D ) and that the two GW182 homologs have distinct frequencies of phenotypes with their interactors ( Figure 1—figure supplement 1B ) , arguing against general sickness being the cause for the enhancement ( further elaborated in Figure 2—figure supplement 1 ) . The pleiotropic nature of ain-1 interactions is consistent with the diverse physiological functions associated with AIN-1 or possibly its expression patterns or levels . We were most surprised to identify the C . elegans cell-killing caspase , ced-3 , as an interactor of the miRISC GW182 homolog , ain-1 . Using multiple alleles of each gene , we found that ced-3 ( lf ) ;ain-1 ( lf ) double mutants have pleiotropic developmental phenotypes including delays in larval growth rate , smaller brood size , abnormal adult body morphology , egg-laying defect ( accumulation of eggs inside the animal ) , sluggish movement , embryonic lethality , and laid oocytes ( failure to fertilize ) ( Figure 2A–D and Figure 2—figure supplement 2A , B ) . The penetrance of abnormal phenotypes increased as the adults continued to age ( Figure 2—figure supplement 2C ) and was therefore best quantified in a synchronized population . Combining mutations of miRISC components such as ain-1 ( GW182 ) ( lf ) or alg-1 ( argonaute ) ( lf ) with the cell death pathway factors ced-3 ( caspase ) ( lf ) or its upstream activator , ced-4 ( apaf-like ) ( lf ) , results in abnormal adults ( Figure 2E ) but ced-3 ( lf ) ;ain-2 ( lf ) animals did not show a significant defect ( Figure 2—figure supplement 2D ) . To test the involvement of other core cell death pathway factors , we also examined the interaction of ain-1 with egl-1 that has been shown to act upstream of the CED-3 caspase to promote apoptosis ( Figure 2—figure supplement 3A ) and egl-1 ( lf ) is known to cause a strong cell death defect ( Conradt and Xue , 2005 ) . We found that , like ced-3 ( lf ) and ced-4 ( lf ) , egl-1 ( RNAi ) also significantly enhanced the developmental defects of ain-1 ( lf ) ( Figure 2—figure supplement 3B ) . 10 . 7554/eLife . 04265 . 005Figure 2 . C . elegans strains compromised in both miRISC and ced-3 functions have significant pleiotropic developmental phenotypes . ( A and B ) Microscopic images showing the pleiotropic phenotypes of the ced-3 ( lf ) ;ain-1 ( lf ) double mutant , including egg-laying defect ( Egl ) , sluggish movement ( Slu ) , body morphology defects ( Bmd ) , larval arrest ( Lva ) , and embryonic lethality ( Emb ) . Asterisk in ( A ) indicates an Egl animal that was devoured by internally hatched progeny , and the arrow indicates an adult animal with multiple defects ( Egl , Slu and Bmd ) . Figure 2—figure supplement 1 shows the phenotype of another interactor , ceh-18 , which is very different from ced-3 , supporting distinct physiological relevance of the identified interactors . ( C ) ced-3 ( RNAi ) significantly enhanced the frequency of ain-1 ( lf ) phenotypes . Mean values ± SD for percent normal ( p < 0 . 001 , compared to wt with mock RNAi , compared to all others , Chi-square test comparing the distributions of phenotypes ) . Number of worms tested indicated above each bar ( same for all figures ) . ( D ) Mean values ± SD of embryonic lethality ( p < 0 . 05 compared to all , Mann–Whitney test ) . ( E ) Enhancement of miRISC phenotypes by ced-3 ( lf ) and ced-4 ( lf ) . Mean values ± SD for percent normal ( p < 0 . 0001 , compared to each of the relevant single mutants , Chi-square test comparing the distributions of phenotypes ) . Other ain-1 and ced-3 alleles ( Figure 2—figure supplement 2 ) and the ain-1 interaction with egl-1 ( Figure 2—figure supplement 3 ) were also tested . ( F ) Rescue effects of expressing ain-1 or ain-2 in specific tissues ( driven by tissue-specific promoters for the four principal tissues of C . elegans including the hypodermis , gut , muscle , and nerve; see ‘Materials and methods’ ) in the ced-3 ( lf ) ;ain-1 ( lf ) double mutants . ‘All tissues’ indicates a genomic ain-1 transgene . Mean values ± SD for percent normal [p < 0 . 0001 , Fisher's Exact test comparing the distribution of normal and abnormal animals for each rescue to ced-3 ( lf ) ;ain-1 ( lf ) without rescue ( see ‘Materials and methods’ for statistical rationale ) ] . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 00510 . 7554/eLife . 04265 . 006Figure 2—source data 1 . Source data quantifying genetic interactions between the miRISC and cell death pathways . ( A ) Source data for Figure 2C , ( B ) Source data for Figure 2D , ( C ) Source data for Figure 2E , ( D ) Source data for Figure 2F , ( E ) Source data for Figure 2—figure supplement 1B , ( F ) Source data for Figure 2—figure supplement 2A , ( G ) Source data for Figure 2—figure supplement 2B , ( H ) Source data for Figure 2—figure supplement 2C , ( I ) Source data for Figure 2—figure supplement 2D , ( J ) Source data for Figure 2—figure supplement 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 00610 . 7554/eLife . 04265 . 007Figure 2—figure supplement 1 . ain-1 ( lf ) ;ceh-18 ( lf ) double mutants have reduced oocytes . DIC images of gonads from single and double mutant animals . ( A ) Arrows indicate the farthest point for the gonad turn . Asterisks are located near the first identifiable oocyte in the given gonad arm . The expanded segment is a digital zoom-in to show the morphological detail of the double mutant gonad for the indicated segment . ( B ) Dot plot of the oocyte counts per gonad arm ( n = 40 for each strain ) . Each dot represents the number of oocytes in one gonad arm and the median values are given by black bars for each strain ( p < 0 . 001 , compared to ain-1 ( lf ) and compared to both ain-1 ( lf ) and ceh-18 ( lf ) alone , Mann–Whitney test ) . The distinct phenotypes observed amongst some of the genetic interactors , such as ced-3 ( shown in Figure 2 ) vs ceh-18 suggest distinct physiological functions and argue against general sickness of the single mutants . This conclusion is also supported more broadly by the frequency of phenotypes observed ( Figure 1—figure supplement 1B ) such that equal frequency is not observed across the various phenotypes . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 00710 . 7554/eLife . 04265 . 008Figure 2—figure supplement 2 . Additional phenotypes of ced-3 ( lf ) ;ain-1 ( lf ) and test of other alleles . ( A ) ain-1 ( lf ) animals treated with ced-3 ( RNAi ) have a significant increase in oocytes laid ( p < 0 . 0001 , compared to wt mock , compared to all others , Mann–Whitney test ) . ( B ) Developmental defects associated with ced-3 ( lf ) ;ain-1 ( lf ) double mutants were observed for other ain-1 ( lf ) and ced-3 ( lf ) alleles . Bar graph showing the synergistic effect between ain-1 ( tm3681 ) and two ced-3 ( lf ) alleles on the egg-laying defective ( Egl ) phenotype . Animals were scored 5 days after eggs were placed on plates . The mean values are shown ( p < 0 . 001 compared to wt and each of the relevant single mutants , Fisher's exact test ) . ( C ) Phenotypes of adults scored at two time points after synchronized first stage larvae were placed on OP50 food . In panel ( C ) , and elsewhere in the study , unless noted , the ain-1 ( ku322 ) and ced-3 ( n1286 ) alleles were used ( p < 0 . 001 , relative to wt and single mutants , Fisher's exact test comparing the distribution of normal and abnormal animals ) . ( D ) The ced-3 ( lf ) ;ain-2 ( lf ) double mutant adults are phenotypically comparable to the ced-3 ( lf ) single mutant adults in ( C ) at 96 hr on OP50 food following synchronization ( p < 0 . 001 , Chi-square analysis ) . For all panels: Slu , sluggish or immobile; Egl , egg-laying defective; Rup , ruptured through vulva . Number of worms tested is indicated above each bar . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 00810 . 7554/eLife . 04265 . 009Figure 2—figure supplement 3 . The core apoptotic regulatory pathway acts in parallel to miRISC for normal development . ( A ) Core apoptotic regulatory pathway with C . elegans gene names indicated ( mammalian counterparts in parentheses ) . ( B ) The phenotypes observed for ced-3 ( lf ) ;ain-1 ( lf ) were observed when combining ain-1 ( lf ) with RNAi of upstream components of the ced-3 pathway . The ain-1 ( lf ) single mutant shows enhanced defects when treated with ced-3 , ced-4 , or egl-1 RNAi for two RNAi generations . Significance of phenotypes when wt and ain-1 ( lf ) animals were fed the indicated RNAi was determined [p < 0 . 001 , relative to both ain-1 ( lf ) fed mock RNAi and to wt fed the given RNAi , Fisher's exact test comparing the distributions of normal and abnormal animals ( see ‘Materials and methods’ for statistical rationale ) ] . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 009 To better characterize these defects , we tested the interaction in specific tissues . Expressing either ain-1 or ain-2 in the intestine or hypodermis alone partially rescued the defects of the ced-3 ( lf ) ;ain-1 ( lf ) double mutant ( Figure 2F ) . These findings suggest that these two tissues are the major sites for miRNA functions in this interaction and likely also CED-3 function given that ced-3 acts cell autonomously ( Yuan and Horvitz , 1990 ) . Expressing ced-3 with strong tissue-specific promoters has been shown to kill those tissues , even in cells that do not normally die , due to the resulting high level of CED-3 accumulation ( Shaham and Horvitz , 1996; Hengartner , 1997 ) thus preventing the reciprocal rescue experiments . The ced-3 caspase has been well-characterized for its role in apoptosis but not demonstrated to have a broad , non-apoptotic function in development ( Yuan et al . , 1993; Xue et al . , 1996; Conradt and Xue , 2005; Peden et al . , 2008 ) . The fact that strong ced-3 ( lf ) alleles cause robust defects in programmed cell death but not the developmental defects described above suggests that the functions of ced-3 with miRISCs uncovered in our screen are non-apoptotic . To further address this question , we first used an assay previously shown to effectively identify apoptotic functions of genes , such as mcd-1 encoding a zinc-finger containing protein , for which mutations caused subtle apoptotic defects alone , but significantly enhanced the cell death defect of a ced-3 reduction-of-function allele ( ced-3 ( rf ) ) ( Reddien et al . , 2007 ) ( Figure 3A ) . We found that , in contrast to the positive control , mcd-1 ( lf ) , the ain-1 ( lf ) mutation did not enhance the apoptotic defect of ced-3 ( rf ) animals as assayed by observing the perdurance of lin-11::GFP positive undead P9-11 . aap cells ( Figure 3A–B ) . Because nuc-1 encodes an effector nuclease important for the proper execution of apoptosis ( Wu et al . , 2000 ) , we then tested if the ain-1 ( lf ) mutation was able to enhance any subtle nuc-1 ( lf ) phenotype and found no significant defect beyond the phenotypes of the single mutants ( Figure 3C ) . Finally , ain-1 ( RNAi ) did not affect the number of apoptotic cell corpses accumulating in the heads of ced-1 ( lf ) first stage larvae ( Figure 3D ) , which are defective in cell corpse engulfment allowing for visualization of dead cell corpses . Therefore , the ain-1 and ced-3 interaction described above is non-apoptotic . 10 . 7554/eLife . 04265 . 010Figure 3 . ain-1 ( lf ) does not alter cell-death phenotypes . ( A ) Cartoon illustrating a previously established enhancer assay using a reduction-of-function ( rf ) ced-3 allele ( Reddien et al . , 2007 ) . ( B ) ain-1 ( lf ) does not enhance the cell death defect of a ced-3 ( rf ) mutation ( p < 0 . 0001 , compared to ced-3 ( rf ) , Mann–Whitney test ) . ( C ) No enhanced interaction between ain-1 ( lf ) and nuc-1 ( lf ) . Mean values ± SD ( no significant difference , Fisher's Exact test comparing the distributions of normal and abnormal animals of the ain-1 ( lf ) ;nuc-1 ( lf ) double mutant to the single mutants ) . ( D ) ain-1 ( RNAi ) does not alter apoptotic events as indicated by L1 head corpses that fail to occur in ced-3 ( lf ) mutants . The ced-1 ( lf ) mutation was used to enhance visualization of head corpses ( Ledwich et al . , 2000 ) . Mean values ± SD ( no significant difference , Mann–Whitney test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 01010 . 7554/eLife . 04265 . 011Figure 3—source data 1 . Source data quantifying apoptotic assays . ( A ) Source data for Figure 3B , ( B ) Source data for Figure 3C , ( C ) Source data for Figure 3D . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 011 Further analysis indicated that the ced-3 ( lf ) and ced-4 ( lf ) single mutants have mild reduction in their rates of post-embryonic growth similar to the ain-1 ( lf ) and alg-1 ( lf ) mutants ( Figure 4A–C and also Figure 4—figure supplement 1 for more ced-3 ( lf ) data ) . Additionally , the ced-3 ( lf ) ;ain-1 ( lf ) and ced-3 ( lf ) ;alg-1 ( lf ) double mutants , but not ced-3 ( lf ) ;ain-2 ( lf ) , have significantly slower growth rates beyond either single mutant ( Figure 4A–C and Figure 4—figure supplement 1 ) , suggesting cooperativity in regulating the related developmental programs . 10 . 7554/eLife . 04265 . 012Figure 4 . Loss of ced-3 function slows the rate of post-embryonic development . ( A ) Percent of animals reaching adulthood at 96 hr after hatching is shown . Mean ± SD ( p < 0 . 0001 , compared to wt , *compared to the relevant single mutants , Fisher's Exact test comparing the distributions of adult to larval-stage animals at this time ) . ( B and C ) Distribution of stages at 48 hr and 72 hr with food ( p < 0 . 0001 , compared to wt , *compared to the relevant single mutants , Chi-square test comparing the distributions of all stages ) . Also see Figure 4—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 01210 . 7554/eLife . 04265 . 013Figure 4—source data 1 . Source data quantifying post-embryonic growth rates . ( A ) Source data for Figure 4A , ( B ) Source data for Figure 4B , ( C ) Source data for Figure 4C , ( D ) Source data for Figure 4—figure supplement 1A , ( E ) Source data for Figure 4—figure supplement 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 01310 . 7554/eLife . 04265 . 014Figure 4—figure supplement 1 . ced-3 ( lf ) mutants displayed a mild but significant reduction in the rate of post-embryonic development . ( A and B ) Animals were synchronized 36 hr in M9 buffer at 20°C then placed on standard bacteria food ( OP50 ) and staged every 24 hr thereafter . Data for 24 and 96 hr are shown here and data for 48 and 72 hr are shown in Figure 4B–C . The distribution of animals from first larval stage ( L1 ) through young adult/adult ( indicated as YA+ ) is shown ( p < 0 . 0001 , compared to wt , Chi-square analysis comparing the distributions of stages ) . Number of worms indicated above each set . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 014 To interrogate the genetic interaction further , we screened all of the available C . elegans miRNA deletion strains in the blind ( Figure 5A , strains listed in Supplementary file 4 ) for synthetic interactions with ced-3 by depleting ced-3 in each miRNA mutant background by RNA interference . After finding pronounced RNAi effects associated with several miRNA deletions , we then generated double or triple mutants containing ced-3 ( lf ) and the miRNA mutations , and observed phenotypes similar to those seen in ced-3 ( lf ) ;ain-1 ( lf ) ( Figure 5B , and refer to Figure 2A–E ) Specifically , mutations in the let-7-family members , mir-48 and mir-84 , had the strongest effect with a fully penetrant egg-laying defect observed in the ced-3 ( lf ) ;mir-48 ( lf ) ;mir-84 ( lf ) triple mutant ( Figure 5B ) . Interestingly , the ced-3 ( lf ) ;mir-1 ( lf ) ;mir-84 ( lf ) triple mutant displayed some developmental defects not seen in the mir-1 ( lf ) ;mir-84 ( lf ) , ced-3 ( lf ) ;mir-1 ( lf ) , or the ced-3 ( lf ) ;mir-84 ( lf ) double mutants ( Figure 5B ) . Since ced-3 ( lf ) had the strongest developmental defects with the let-7-family members , and since both lin-14 and lin-28 mRNAs are well-known targets of the let-7-family of miRNAs , we thus tested the possibility that ced-3 ( lf ) may enhance specific temporal cell fate patterning defects of these miRNA mutants by examining their adult alae . Normal adult-specific alae are generated by seam cells and defects in adult alae formation are commonly used as a sensitive assay for defects in temporal cell fate patterning ( Ambros and Horvitz , 1984 ) . We found that ced-3 ( lf ) significantly enhanced adult alae defects ( Figure 5C , D ) . This effect was observed for both the miR-48 ( lf ) , miR-84 ( lf ) ;ced-3 ( lf ) triple mutant and the ced-3 ( lf ) ;ain-1 ( lf ) double mutant , but not the ced-3 ( lf ) ;mir-1 ( lf ) ;mir-84 ( lf ) triple mutant ( Figure 5D ) . These findings suggested the hypothesis that the expression of some developmental timing regulators is co-regulated by miRISCs and ced-3 . 10 . 7554/eLife . 04265 . 015Figure 5 . Identification of specific miRNAs that cooperate with ced-3 caspase to regulate development . ( A ) Diagram for screening miRNA deletion mutants ( listed in Supplementary file 4 ) when fed mock or ced-3 RNAi to identify overt developmental phenotypes when ced-3 was depleted . let-7 ( lf ) and lin-4 ( lf ) mutants were excluded due to significant defects alone . ( B ) miRNA deletion ( s ) [indicated by the miR number ( s ) ] identified in ( A ) were combined with ced-3 ( lf ) . ‘+’ and ‘−’ indicate wild-type and ced-3 ( null ) , respectively . Phenotypes including egg-laying defect ( Egl ) , ruptured vulva ( Rup ) , and sluggish movement ( Slu ) were quantified . Mean values ± SD for percent normal ( p < 0 . 05 , when compared to ced-3 ( lf ) and the relevant miRNA deletion ( s ) alone , Fisher's Exact test comparing the distributions of normal and abnormal animals ) . ( C and D ) ced-3 ( lf ) enhances adult-specific alae defects including low quality ( thin and rough ) and gapped alae [bracket in ( C ) near the mid-body shows a gap] . Percent of adults with alae defects ( p < 0 . 001 , compared to the relevant single or double mutants , Chi-square test comparing the distributions of adult alae phenotypes ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 01510 . 7554/eLife . 04265 . 016Figure 5—source data 1 . Source data quantifying genetic interactions between miRNA mutants and ced-3 . ( A ) Source data for Figure 5B , ( B ) Source data for Figure 5D . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 016 To better analyze the mechanism underlying this non-apoptotic temporal cell fate patterning function of ced-3 , we tested its effect on seam cell development . The division and differentiation pattern of the stem cell-like seam cells are regulated by a well-described genetic pathway that includes several miRNAs and the LIN-28 pluripotency factor that blocks the maturation of pre-let-7 miRNA ( Viswanathan and Daley , 2010 ) . During each larval stage , lateral seam cells ( V1–V4 and V6 ) divide in an asymmetric , stem-cell like manner with additional stems cells only produced in the L2 stage by an additional symmetric division pattern that duplicates V1–V4 and V6 seam cell numbers ( Sulston and Horvitz , 1977; Ambros and Horvitz , 1984 ) . Wild-type animals consistently have 16 seam cells on both the left and right sides by adulthood ( Joshi et al . , 2010 ) . The dynamic changes in the expression levels of several conserved pluripotency factors are critical for proper temporal cell fate patterning . LIN-14 is highly expressed during L1 to promote L1-specific developmental programs , whereas LIN-28 is highly expressed from late embryonic to L2 stages and acts to promote the L2-specific programs including the only normal symmetric division of V1–V4 and V6 seam cells ( Ambros and Horvitz , 1984; Ambros , 1989; Ruvkun and Giusto , 1989; Moss et al . , 1997; Rougvie and Moss , 2013 ) ( Diagrammed in Figure 6—figure supplement 2A ) . Expression of LIN-14 and LIN-28 rapidly diminishes after L1 and L2 , respectively , which is necessary for animals to progress to the next stage ( Figure 6—figure supplement 2A ) . Loss-of-function ( lf ) mutations in lin-14 and lin-28 result in animals skipping the L1- and L2-specific programs , respectively ( precocious phenotype ) ( Figure 6—figure supplement 2A ) . In contrast , hyperactive ( gain-of-function , gf ) mutations leading to prolonged expression of each gene cause the animals to reiterate the corresponding stage ( retarded phenotype ) ( Figure 6—figure supplement 2A ) . Because of the additional symmetric cell division of V1–V4 and V6 seam cells in L2 , skipping or reiterating the L2 stage in lin-28 ( lf ) or lin-28 ( gf ) mutations lead to a decrease or increase of total seam cell number , respectively ( Ambros and Horvitz , 1984; Moss et al . , 1997 ) and diagrammed in Figure 6—figure supplement 2A . Mammalian DIS3L2 was recently annotated as the ribonuclease that degrades the uridylated pre-let-7 miRNA following binding by LIN-28 and 3′-oligo-uridylation by a polyU polymerase ( Chang et al . , 2013 ) . We identified the likely C . elegans ortholog of Dis3l2 and named it disl-2 ( Figure 6—figure supplement 1 ) . The effects for disl-2 on seam cell development have not been determined . As previously published ( Ding et al . , 2005; Zhang et al . , 2007 ) , we also found that the ain-1 ( lf ) mutant alone has a mild increase in the number of seam cells by late larval development ( Figure 6A , B and Figure 6—figure supplement 2 ) consistent with the well-established role of miRNAs in regulation of temporal cell fate patterning; whereas the ced-3 ( lf ) mutant alone rarely shows altered seam cell numbers ( Figure 6A , B and Figure 6—figure supplement 2 ) . Strikingly , the ced-3 ( lf ) ;ain-1 ( lf ) double mutants have both a markedly increased number of seam cells and an increased range of seam cell number by late larval development ( Figure 6A , B ) with a mean value ( ±SD ) of 25 . 9 ( ±5 . 5 ) per side . Notably , the ced-3 ( lf ) ;ain-1 ( lf ) double mutants hatch with the correct number of seam cells but they continue to increase inappropriately throughout later larval development ( Figure 6—figure supplement 2A , B ) . The production of supernumerary seam cells indicates a previously unknown role for ced-3 in cooperating with miRISC-regulated seam cell differentiation and temporal cell fate patterning ( Figure 6A , B and Figure 6—figure supplement 2C ) . 10 . 7554/eLife . 04265 . 017Figure 6 . ced-3 may act upstream of multiple conserved pluripotent factors to affect differentiation of stem cell-like seam cells . ( A and B ) Pseudocolored GFP from DIC images of a seam cell reporter and dot plot quantitation . The tick line depicts 16 seam cells that are normally found in wild-type animals . Black bars indicate the median values for each strain ( p < 0 . 0001 , compared to wt , compared to single mutants , Mann–Whitney test ) . ( C ) Effect of RNAi treatment beginning at L2 on the seam-cell-number phenotype of the ced-3 ( lf ) ;ain-1 ( lf ) double mutant ( p < 0 . 0001 , compared to mock RNAi , Mann–Whitney test ) . C . elegans disl-2 is homologous to mammalian Dis3l2 ( Figure 6—figure supplement 1 ) . ( D ) Effect of the same RNAi on the ced-3 ( lf ) ;ain-1 ( lf ) double mutant defects . Mean values ± SD for percent normal [p < 0 . 0001 , compared to mock RNAi , Fisher's Exact test comparing the distributions of normal and abnormal animals ( see ‘Materials and methods’ for statistical rationale ) ] . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 01710 . 7554/eLife . 04265 . 018Figure 6—source data 1 . Source data quantifying temporal cell fate patterning and other phenotypes . ( A ) Source data for Figure 6B , ( B ) Source data for Figure 6C , ( C ) Source data for Figure 6D , ( D ) Source data for Figure 6—figure supplement 2B , ( E ) Source data for Figure 6—figure supplement 2C , ( F ) Source data for Figure 6—figure supplement 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 01810 . 7554/eLife . 04265 . 019Figure 6—figure supplement 1 . Protein sequence alignment of human DIS3L2 and C . elegans DISL-2 . Domain prediction and sequence alignment of the mammalian DIS3L2 protein with the C . elegans DISL-2 protein . Domain prediction was done by Interpro ( Hunter et al . , 2012 ) and Pfam ( Punta et al . , 2012 ) , and the alignment was done using Clustal W ( Larkin et al . , 2007 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 01910 . 7554/eLife . 04265 . 020Figure 6—figure supplement 2 . Additional analyses of seam cells for the ced-3 ( lf ) ;ain-1 ( lf ) double mutant . ( A ) Diagram depicting the symmetric and asymmetric cell divisions of V1–V4 and V6 lineages ( based on content detailed in a recent review [Rougvie and Moss , 2013] ) . One cell is shown beginning at L1 ( red ) . In wt , the L2 stage has a symmetric division followed by asymmetric divisions thereafter . The lin-28 ( lf ) mutation generates a precocious phenotype without the L2-specific symmetric division; whereas , the lin-28 ( gf ) mutation generates a retarded phenotype with further L2-like reiterations . Thus , a normal animal hatches with 10 seam cells that result in 16 terminally differentiated seam cells by adulthood . ( B ) The ced-3 ( lf ) ;ain-1 ( lf ) double mutant animals hatch with the correct number of seam cells which continue to develop during late larval stages . 10 seam cells are observed for all strains with no significant differences observed . Black bars indicate the median values for each strain ( p > 0 . 05 , Mann–Whitney test , all single and double mutants compared to wild-type ) . ( C ) Quantitation of third and fourth larval stages of the ced-3 ( lf ) ;ain-1 ( lf ) double mutant animals suggests supernumerary seam cells continue to arise during late larval stages . Black bars indicate the median values for each strain ( p = 0 . 008 , compared to L3 stage ced-3 ( lf ) ;ain-1 ( lf ) , Mann–Whitney test ) . The data shown for the L4 animals in panel ( B ) here are unique from the main Figure 6B and are not repeated . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 02010 . 7554/eLife . 04265 . 021Figure 6—figure supplement 3 . ced-3 ( lf ) mutants enhance lin-66 ( RNAi ) ruptured vulva phenotype . lin-66 was previously shown to limit the expression of LIN-28 by an incompletely understood mechanism ( Morita and Han , 2006 ) . We therefore tested if loss of ced-3 could enhance the vulva defect in lin-66 ( RNAi ) . ( A ) Images and ( B ) bar graph showing that lin-66 ( RNAi ) increases the frequency of ruptured vulva of both ced-3 ( lf ) and ain-1 ( lf ) mutants . L2 stage animals were fed either mock or lin-66 ( RNAi ) and the subsequent generation was scored for ruptured vulva ( Rup ) at adulthood ( multiple ruptured animals are indicated by arrows ) . RNAi was used due to the fourth larval stage lethality of the lin-66 ( lf ) alleles . ain-1 in miRISC is already known to negatively regulate LIN-28 expression and its enhancer phenotype here with lin-66 ( RNAi ) serves as a positive control . The percent mean values are shown ( number of worms indicated above each bar , *p < 0 . 001 , Chi-square analysis ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 021 We found that the increased number of seam cells in the ced-3 ( lf ) ;ain-1 ( lf ) double mutants was partially suppressed by down-regulating lin-14 , lin-28 , or disl-2 ( Dis3l2 ) through RNAi treatment beginning at L2 ( Figure 6C ) , suggesting that an abnormally high level of any of the three proteins could be a significant contributor to the phenotype . A lin-14 ( lf ) or lin-28 ( lf ) mutation would not be effective for such a suppression test because of the strong defects associated with them at the early larval stage ( Rougvie and Moss , 2013 ) . LIN-66 was previously shown to act in parallel to miRNAs to repress LIN-28 expression ( Morita and Han , 2006 ) . Consistent with a ced-3 function in lin-28-mediated temporal cell fate patterning regulation , we also observed that ced-3 ( lf ) enhanced the heterochronic defect of lin-66 reduction ( Figure 6—figure supplement 3 ) . We further found that down-regulation of lin-14 , lin-28 , or disl-2 ( Dis3l2 ) by RNAi beginning at L2 could significantly suppress the defects in the ced-3 ( lf ) ;ain-1 ( lf ) double mutants ( Figure 6D ) . These findings suggest that ced-3 cooperates with miRNAs to regulate the lin-14-lin-28-disl-2 ( Dis3l2 ) axis during development . The above genetic data suggest that ced-3 normally represses lin-28 , disl-2 , and/or lin-14 in development . As a caspase , we thought that CED-3 may directly repress the expression of these genes through proteolytic cleavage , which is consistent with our observation that LIN-14 , LIN-28 , and DISL-2 contain multiple consensus CED-3 cleavage sites that consist of a tetra-peptide sequence usually ending in an aspartic acid residue ( Xue et al . , 1996 ) . To test this hypothesis , we performed an in vitro CED-3 cleavage assay as previously described ( Xue et al . , 1996 ) . We found that the DIS3L2 ribonuclease homolog , DISL-2 , was robustly cleaved by the CED-3 caspase while LIN-14 and LIN-28 were partially cleaved ( Figure 7A ) . The multiple cleavage products generated by CED-3 cleavage of DISL-2 ( Figure 7A , B ) suggest a clear role for CED-3-mediated inactivation of this target protein . We further tested the specificity of the partial LIN-28 cleavage by CED-3 and found that it was completely blocked by addition of the caspase-specific-inhibitor zDEVD-fmk ( Figure 7C ) . We then determined the proteolytic cleavage site for LIN-28 by mutagenesis and identified the CED-3-specific recognition sequence ( Figure 7D and Figure 7—figure supplement 1 ) . Numerous possible cleavage sites were found for LIN-14 and DISL-2 but were not pursued further ( Figure 7—figure supplement 2 ) . The identified sequence DVVD fits the canonical CED-3 recognition motif ( DxxD ) ( Xue et al . , 1996 ) and mutating the second aspartic acid residue to an alanine ( D31A in Figure 7D ) entirely eliminated CED-3 cleavage . CED-3 proteolysis of LIN-28A generates an N-terminal asparagine in the remaining protein ( Figure 7E ) . Asparagine is known to function generally as a destabilizing residue at the N-terminus of eukaryotic proteins resulting in proteasomal degradation in a phenomenon termed the N-end rule ( Sriram et al . , 2011 ) . 10 . 7554/eLife . 04265 . 022Figure 7 . CED-3 cleavage of LIN-14 , LIN-28 , and DISL-2 ( DIS3L2 ) in vitro . ( A ) Established in vitro CED-3 cleavage assay ( Xue et al . , 1996 ) of 35S-labeled proteins . CED-9 served as a positive control throughout . Red asterisks indicate cleavage products ( same in B–D ) . ( B ) Result from a longer-run gel showing near quantitative cleavage of full-length DISL-2 ( arrow indicates the full-length protein ) . ( C ) In vitro cleavage assay with the zDEVD-fmk caspase-specific irreversible inhibitor ( Rickers et al . , 1998 ) . The arrow and arrowhead ( and red asterisks ) indicate the full-length protein and a predominant CED-3 cleavage product , respectively . ( D ) Effect of the D31A mutation on CED-3 cleavage ( for other mutants see Figure 7—figure supplement 1 ) . ( E ) Diagram showing the position and consequence of LIN-28A cleavage by CED-3 in vitro ( 22 kDa with an N-terminal asparagine ) . Each panel was performed as an independent experiment . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 02210 . 7554/eLife . 04265 . 023Figure 7—figure supplement 1 . Mutagenesis of LIN-28A to identify the CED-3 cleavage site . Several possible cleavage sites exist in the C . elegans LIN-28A protein corresponding to the tetra-peptide D ( E , N , G ) xxD ( E ) cleavage sequence with the alternate residues indicated in parentheses and the proximal residue for cleavage underlined ( aspartic acid in this position strongly favored ) ( Xue et al . , 1996 ) . ( A–B ) Mutants were made for the second acidic residue in the tetra-peptide sequence for all such sites in the LIN-28A N-terminal region but only the DVVD to DVVA mutation ( noted as D31A in panel B and Figure 7D ) abolished CED-3 cleavage . Cleavage products are indicated by red asterisk . These experiments were run independently of those shown in Figure 7 and thus constitute replicates for the mutant D31A cleavage assay . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 02310 . 7554/eLife . 04265 . 024Figure 7—figure supplement 2 . Identification of possible CED-3 cleavage sites in LIN-14 and DISL-2 . Numerous possible CED-3 cleavage sites are found in LIN-14B and DISL-2 based on the consensus motif detailed in Figure 7—figure supplement 1 ( Xue et al . , 1996 ) . ( A–B ) Potential CED-3 substrate tetrapeptides are shown in red font , some of which overlap . Closed triangles and open triangles indicate potential cleavage sites of high and moderate probability , respectively . The numbers in parentheses above the sequences indicate the proximal acidic residue . The LIN-14B isoform is shown in ( A ) since this is the isoform we cloned . However , all of the predicted cleavage sites shown here are also present in the LIN-14A isoform , which differs in the amino terminus prior to the first predicted site . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 024 To examine CED-3-mediated turnover of the LIN-28 protein in vivo , we generated a polyclonal antibody against a C-terminal peptide in LIN-28 that recognizes both LIN-28 isoforms reported previously ( Seggerson et al . , 2002 ) ( Figure 8—figure supplement 1A , B ) . We found that the dynamic decrease in LIN-28 abundance during L2–L4 stages was similarly delayed by two different ced-3 ( lf ) mutations ( Figure 8A and quantitation shown in Figure 8—figure supplement 1C ) . At late L4 ( 48 hr in Figure 8A ) , LIN-28 was almost completely absent in both wild type and ced-3 ( lf ) mutants , indicating the role of general , non-CED-3-mediated , proteolysis during late larval stages . Interestingly , the 22-kDa cleavage product observed in the in vitro assay ( Figure 7D , E ) was not observable in vivo ( Figure 8A ) , consistent with the idea that the cleavage product with an asparagine at its N-terminus was possibly degraded by an additional proteolytic process . It is possible that the delayed down-regulation of LIN-28 seen in Figure 8A is the consequence of the slower post-embryonic growth rate observed for ced-3 ( lf ) mutants ( Figure 4 ) . To address this question , we first used a LIN-28::GFP transgenic strain previously shown to have functional LIN-28 activity ( Moss et al . , 1997 ) to monitor stage-matched L3 larvae with or without a ced-3 ( lf ) mutation by DIC microscopy . We observed that the ced-3 ( lf ) mutation delayed the proper down-regulation of the LIN-28::GFP reporter at L3 in the hypodermis ( Figure 8B–D and Figure 8—figure supplement 2A–C ) . We also found that down-regulation of LIN-28::GFP expression was delayed in neuronal cells in the head ( Figure 8—figure supplement 2D , E ) . These findings support the hypothesis for the delayed down-regulation of LIN-28 by ced-3 ( lf ) . The difference in magnitude between the Western blot results and the number of fluorescent cells seen by DIC microscopy may suggest that the observed fluorescence levels do not linearly reflect the protein levels and that the two methods may have different dynamic ranges . 10 . 7554/eLife . 04265 . 025Figure 8 . In vivo confirmation that CED-3 caspase negatively regulates LIN-28 expression in late larval stages . ( A ) Western blot with the LIN-28 antibody we developed ( validation shown in Figure 8—figure supplement 1 ) to see the effects of ced-3 ( lf ) mutations on LIN-28 protein expression during developmental transitions . Notice that the cleavage product of the larger isoform of LIN-28 observed in the in vitro assay ( Figure 7 ) is not detectable in the in vivo analysis , suggesting that the cleavage product , which has an Asn instead of Met as the N-terminal end residue , may potentially be sensitive to the N-end rule proteasomal degradation pathway . The pattern and timing of LIN-28 expression and downregulation we show here for wt are similar to previous findings ( Seggerson et al . , 2002; Morita and Han , 2006 ) and two independent ced-3 ( lf ) mutant strains are shown here ( quantitation is shown in Figure 8—figure supplement 1C ) . ( B–C ) Pseudocolored GFP from DIC images of L3 larvae near the mid-body . Also see DIC images of these same animals without GFP illumination ( similar length gonads shown in Figure 8—figure supplement 2A , B ) and test of similar staging ( Figure 8—figure supplement 2C ) . Size bars are indicated . ‘wt’ indicates the lin-28 ( + ) ::gfp integrated transgene alone previously shown to be functional ( Moss et al . , 1997 ) and ‘ced-3 ( lf ) ’ indicates this same transgene combined with a ced-3 ( lf ) mutation . ( D ) Quantitation of the LIN-28::GFP expression between the strains within the L3 stage . ( p = 0 . 0038 , significant compared to wt , Mann–Whitney test comparing the integrated intensity of LIN-28::GFP hypodermal expression in L3 larvae ) . Persistent expression of LIN-28::GFP in head cells ( Figure 8—figure supplement 2D , E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 02510 . 7554/eLife . 04265 . 026Figure 8—source data 1 . Source data quantifying effects of ced-3 ( lf ) on LIN-28::GFP expression . ( A ) Source data for Figure 8D , ( B ) Source data for Figure 8—figure supplement 2C , ( C ) Source data for Figure 8—figure supplement 2E . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 02610 . 7554/eLife . 04265 . 027Figure 8—figure supplement 1 . Validation of our newly generated LIN-28 antibody and quantitation of Western blot data . ( A ) Western blot to demonstrate the specificity of our peptide-purified rabbit-anti-C . elegans LIN-28 antibody . The peptide we chose to immunize the rabbits with is found near the C-terminus ( reported in ‘Materials and methods’ ) and therefore should recognize both the a and b isoforms equally well . Equivalent amounts of extracts from mixed staged wt , lin-28 ( n719 , lf ) , and a strain with an integrated transgene of lin-28::GFP were resolved by SDS-PAGE then probed with our purified LIN-28 antibody . The Is[lin-28::gfp] strain shows both the endogenous forms of LIN-28 and the shifted fusion protein . ( B ) The large isoform of LIN-28 ( corresponding to the a isoform ) was synthesized in vitro then added at increasing amounts into the lin-28 ( n719 , lf ) strain to simulate a complex mixture for all other background proteins . Three exposures of the same gel are shown . Two predominant background bands from the in vitro lysate are indicated by dashed blue lines . LIN-28-specific bands ( identified in panel A ) are indicated . The lin-28 ( n719 ) + mock lane has as much in vitro lysate as the most +LIN-28a lane but with a mock vector to show lysate background . This background is not present in the worm extracts ( compare the lin-28 ( n719 ) lane vs lin-28 ( n719 ) + mock ) . ( C ) Quantitation of Western blot data from Figure 8A . Arbitrarily , 100% was defined as the intensity of total LIN-28 at 0 hr normalized to actin for the wt strain . Both ced-3 ( lf ) strains were set relative to the wt 0 hr . The subsequent time points for all strains were compared to this 0 hr value . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 02710 . 7554/eLife . 04265 . 028Figure 8—figure supplement 2 . Larval staging and persistent LIN-28::GFP expression . ( A–C ) Analysis of gonad length in LIN-28 ( + ) transgenic lines with ced-3 ( wt ) or ced-3 ( lf ) . We first picked L3 stage animals in the blind on a non-fluorescent microscope . Prior to GFP illumination , we assessed the sub-stages of animals by measuring the gonad length of the animals . This was done since the animals had roller phenotype making P cell division difficult to observe and strong defects in LIN-28 , directly , are known to have only minor effects on gonad development ( Ambros , 1997 ) . Thus , these strains should not be expected to have significant defects in gonad extension and gonad length should serve as a quantifiable metric of larval stage similar to previous reports ( Ambros and Horvitz , 1984; Moss et al . , 1997; Abbott et al . , 2005 ) . The images shown here ( A–B ) are from the same animals shown in the main Figure 8B–C . The white bars indicate 20 μM . The white brackets indicate the gonads of equal length . ( C ) No significant difference in gonad length for the L3 samples was observed between the two strains suggesting similar sub-stage . Mean ± SD is shown ( p > 0 . 05 , Mann–Whitney test ) . ( D–E ) Pseudocolored GFP from DIC head images of L3 larvae and quantitation of the number of cells expressing LIN-28::GFP ( p < 0 . 0001 , compared to wt; ( + ) , Chi-square test comparing the distributions of L3 larvae based on the number of LIN-28::GFP positive cells ) . ‘+’ indicates the lin-28 ( + ) ::gfp integrated transgene previously shown to be functional ( Moss et al . , 1997 ) . This set of experiments was done completely independently of the ones shown in main Figure 8B–D . A similar method for selection of L3 animals in the blind on a non-fluorescent microscope was used . All animals were visualized for GFP positive cells through multiple focal planes of the head . All images were taken with identical exposure times , and all cells were counted by an individual who did not take the images . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 028 We then further addressed the question by testing the physiological impact of the lin-28 ( D31A ) mutation . Specifically , we made the point mutation in the previously published lin-28 ( + ) ::gfp fusion protein ( Moss et al . , 1997 ) . To ensure that the LIN-28 ( D31A ) mutation did not disrupt the global function of the protein , we tested its ability to overcome the highly penetrant protruding vulva ( Pvl ) phenotype in lin-28 ( n719 , lf ) animals and found that it was able to rescue the Pvl phenotype ( Figure 9—figure supplement 1 ) . Following integration and outcrossing , we found the copy number of the lin-28 ( D31A ) ::gfp transgene to be slightly lower than that of the non-mutated lin-28 ( + ) ::gfp transgene ( Figure 9—figure supplement 2 ) . We then examined the developmental profile and found that the lin-28 ( D31A ) ::gfp transgene alone caused a delay in larval development similar to that caused by the combination of the lin-28 ( + ) ::gfp transgene with ced-3 ( lf ) ( Figure 9A ) . Western blot analysis showed that the lin-28 ( D31A ) ::gfp integration had less basal expression than the non-mutated lin-28 ( + ) ::gfp integration , consistent with the lower copy number estimate . We observed a quantifiable difference in the down-regulation of the lin-28 ( D31A ) ::gfp transgene compared to the lin-28 ( + ) ::gfp transgene ( Figure 9B , C ) . This finding provides evidence that a failure in CED-3 cleavage of LIN-28 leads to slower degradation of LIN-28 and is one of the causes of slower development , since the D31A point mutation alone resulted in both a slower growth rate ( Figure 9A ) and delayed LIN-28 down-regulation ( Figure 9B , C ) . Additionally , in this Western blot ( Figure 9B–C ) , down-regulation of the wild-type LIN-28 transgene in ced-3 ( lf ) worms seems to be delayed more than LIN-28 ( D31A ) in wild-type worms . Such a difference could be due to roles of CED-3 on other targets such as LIN-14 and DISL-2 , which is also expected to contribute to the larval developmental defect in ced-3 ( lf ) ( Figure 9A ) . 10 . 7554/eLife . 04265 . 029Figure 9 . CED-3 caspase represses LIN-28 in vivo to ensure proper temporal cell fate patterning regulation . ( A ) Effects of disrupting CED-3 activity on LIN-28 in vivo on the rate of post-embryonic growth . Percent of animals reaching adulthood at 96 hr after hatching is shown . ‘+’ indicates the lin-28 ( + ) ::gfp integrated transgene described in Figure 8B–C . D31A indicates a transgene integration with the CED-3-cleavage-resistant D31A point mutation in the first exon of LIN-28 but is otherwise identical to the original ( + ) transgene . Test of lin-28 ( lf ) rescue ( Figure 9—figure supplement 1 ) and copy number of the transgenes ( Figure 9—figure supplement 2 ) . Mean values ± SD ( p < 0 . 0001 , compared to wt; ( + ) , Fisher's Exact test comparing the distributions of adult to larval-stage animals at this time ) . ( B–C ) Western blot for the LIN-28::GFP transgenes described in ( A ) and quantitation from three independent Western blot experiments of the LIN-28::GFP transgenes [one Western blot shown in ( B ) ] . Here , 1 . 0 was defined as the intensity of total LIN-28 ( D31A ) ::GFP at 0 hr normalized to actin . Both the lin-28 ( + ) and the ced-3 ( lf ) ;lin-28 ( + ) strains and the 30 hr time point for all strains were compared to this value . Mean ± SEM for the two time points ( dashed lines are used only to indicate the net change in relative expression for the three strains ) . ( D ) Disrupting CED-3 activity on LIN-28 enhances adult alae defects of the strains described in ( A ) ( p < 0 . 01 , compared to wt; ( + ) , Chi-square test comparing the distributions of adult alae phenotypes ) . Figure 9—figure supplement 3 shows examples of the adult alae phenotypes for these three strains . Data for increased expression of LIN-14 in ced-3 ( lf ) mutants at the first larval stage is shown in Figure 9—figure supplement 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 02910 . 7554/eLife . 04265 . 030Figure 9—source data 1 . Source data quantifying effects of ced-3 ( lf ) and LIN-28 ( D31A ) mutation on protein levels and developmental phenotypes . ( A ) Source data for Figure 9A , ( B ) Source data for Figure 9C , ( C ) Source data for Figure 9D , ( D ) Source data for Figure 9—figure supplement 1B . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 03010 . 7554/eLife . 04265 . 031Figure 9—figure supplement 1 . Test for lin-28 ( D31A ) function in overcoming the lin-28 ( n719 , lf ) protruding vulva defect . The full name of the extrachromosomal array that we tested: Ex[lin-28 ( D31A ) ::gfp::lin-28 3′UTR; myo-3::mCherry; pBSIIKS ( − ) ] where myo-3::mCherry served as the transgenic marker . ( A ) Myo-3::mCherry positive animals are rescued from the Pvl phenotype similar to the original lin-28 ( + ) ::gfp described previously ( Moss et al . , 1997 ) . The top panel shows two adults not carrying the array which displayed the protruding vulva ( Pvl ) and ruptured vulva ( Rup ) ( arrows ) phenotype , respectively . Three adults carrying the array are indicated in this same panel by an asterisk near their well-developed vulva ( also myo-3::mCherry positive ) . The bottom two panels show magnified images of two different animals , one without array , and therefore Pvl , and one with the array , and therefore not Pvl . ( B ) The array-positive animals have a significant rescue compared to animals without the array ( 89% rescued on average , p < 0 . 0001 , when compared to animals without the array , Fisher's Exact test ) showing that the D31A mutation does not alter the gross function of the protein . Rather , we conclude that the D31A mutation only abolishes the direct cleavage of LIN-28 as supported by our in vitro experiments ( Figure 7 and Figure 7—figure supplement 1 ) . The offspring from parents bearing the rescue array may have some maternal effect with ∼10% of adults not carrying the array without Pvl but with severe egg-laying defect ( also suggests a non-functional vulva ) and the remaining 5% with apparently normal vulva . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 03110 . 7554/eLife . 04265 . 032Figure 9—figure supplement 2 . Transgene copy number determination . + indicates the [lin-28 ( + ) ::gfp::lin-28 3′UTR; rol-6] integrated transgene which was previously published as functional ( Moss et al . , 1997 ) . D31A indicates the transgene integration we generated which contains the D31A point mutation in the first exon of LIN-28 . Following integration , total DNA was used as input for qPCR to quantify the copy number . Mean ± SD is from two sets of lin-28 primers normalized to two sets of unrelated endogenous genes ( ain-1 and hrp-1 ) . These findings suggest low copy number integration for the two transgenes . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 03210 . 7554/eLife . 04265 . 033Figure 9—figure supplement 3 . Loss of ced-3 function or mutating the CED-3 cleavage site of LIN-28 enhances adult alae defects by a multi-copy lin-28 transgene . ( A–D ) Adult alae were scored using DIC optics ( quantitation of findings is shown in Figure 9D ) . These images serve only as visual reference for the phenotypes scored . The strains are described in Figure 9—figure supplements 1–2 . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 03310 . 7554/eLife . 04265 . 034Figure 9—figure supplement 4 . LIN-14 protein levels are increased in ced-3 ( lf ) mutants at the first larval stage . An integrated transgenic strain Is[lin-14::gfp] previously published ( Olsson-Carter and Slack , 2010 ) was crossed with a ced-3 ( lf ) mutation . Sibling offspring were isolated to obtain the Is[lin-14::gfp] ( with wild-type ced-3 gene ) and the ced-3 ( lf ) ;Is[lin-14::gfp] . Results from three independent Western blots of three independent synchronous first stage larvae are shown here . Samples were synchronized , collected , processed , probed by Western blot with an anti-GFP antibody ( Clontech , Antibody JL8 ) independently at different times . Though the difference between wild-type and ced-3 ( lf ) is subtle , it is repeatable . The subtle effects seen for LIN-14 are also of note since it is , in part , an upstream positive regulator of lin-28 . Thus , subtle effects may be physiologically significant . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 034 Examination of adult-specific alae is a sensitive physiological readout that should overcome any limitations of monitoring delays in the down-regulation of LIN-28 expression levels since scoring adult alae ensures stage-matching and accounts for any perdurance . To further test the functional outcome of both the LIN-28 ( D31A ) transgene and the LIN-28 ( + ) transgene combined with ced-3 ( lf ) , we examined the adult-specific alae and found significant defects including low quality and gapped alae ( Figure 9D and Figure 9—figure supplement 3 ) . This is consistent with the data described above that ced-3 ( lf ) enhances adult-specific alae defect of let-7-family miRNA mutants and ain-1 ( lf ) ( Figure 5C–D ) . We should note that the original report of the LIN-28 ( + ) transgene indicated that some of the adults were observed to have gapped alae ( Moss et al . , 1997 ) . Though we did observe rough and very thin sections of alae for this strain ( scored as low quality alae ) , we did not observe any gapped adult alae . This subtle difference is likely explained by a different threshold since we scored alae using a sensitive camera ( See ‘Materials and methods’ ) . Nonetheless , the relative enhancement of ced-3 ( lf ) with this transgene is quite obvious and similar to that of the caspase-cleavage resistant LIN-28 ( D31A ) point mutant transgene ( Figure 9D ) . Altogether , our data support a causal role for CED-3 cleavage of LIN-28 in the regulation of temporal cell fate patterning . CED-3 appears to facilitate the stereotypical transition of LIN-28 to enhance the robustness of the L2 to L3 developmental transition . Consistent with LIN-14 being modestly cleaved by CED-3 in vitro ( Figure 7A ) , we found that the LIN-14::GFP level was modestly increased in ced-3 ( lf ) mutants in vivo at the L1 stage ( Figure 9—figure supplement 4 ) . This result may not be explained by slower growth rate since these animals were obtained as synchronous L1s without food . Our attempts to monitor DISL-2 protein levels including developing an antibody to endogenous DISL-2 were impeded by technical difficulties . Moreover , N- and C-terminal GFP fusions to DISL-2 had exceedingly low levels of expression beyond detection by common methods suggesting that DISL-2 protein levels are kept exquisitely low for physiological significance . Therefore , our in vitro and in vivo data show that developmental timing regulators are proteolytic targets of the CED-3 caspase , likely resulting in their inactivation . This role of CED-3 cleavage is in contrast to known apoptotic functions of CED-3 caspase activity in two major aspects: CED-3 inactivates its targets rather than activates them as in its apoptotic function ( Conradt and Xue , 2005 ) ; and it acts with other regulatory systems , including miRNAs and possibly the N-end rule proteasomal system , to maintain robust developmental functions . We report the discovery of a new gene expression regulatory mechanism whereby a non-apoptotic activity of the CED-3 caspase functions to inactivate and repress the expression of key developmental regulators , significantly contributing to the robustness of gene expression dynamics and animal development ( Figure 10A ) . Consistent with this , a previous report showed that CED-3 is capable of cleaving more than 22 C . elegans proteins in an in vitro proteomics survey ( Taylor et al . , 2007 ) and two recent genetics-based findings showed that ced-3 may play important roles in neural regeneration ( Pinan-Lucarre et al . , 2012 ) and aging ( Yee et al . , 2014 ) . Second , the described CED-3 function in repressing gene expression is likely in contrast to the role of CED-3 in promoting apoptosis through activation of protein targets by cleavage at specific sites ( Nakagawa et al . , 2010; Chen et al . , 2013 ) . Here , the CED-3 cleavage alone may already destroy the target protein activity . Additionally , the cleavage products may be further degraded by other degradation systems notably via N-terminal destabilizing residues which may make the target more susceptible to additional degradation mechanisms , such as proteasomal degradation ( Sriram et al . , 2011 ) ( Figure 10B ) . We hypothesize that this function operates continually during development to facilitate rapid turnover of these regulatory proteins at the post-translational level and in cooperation with other regulatory mechanisms ( Figure 10—figure supplement 1 ) . We should note that it is curious that only the LIN-28A isoform was found to be cleaved by CED-3 in vitro yet expression of both LIN-28A and LIN-28B isoforms was altered by ced-3 ( lf ) in vivo . This may imply that ced-3 has potential indirect effects on other factors within the heterochronic pathway that could alter LIN-28 isoform expression but further experiments are required to satisfactorily explain this . 10 . 7554/eLife . 04265 . 035Figure 10 . Model for CED-3 function in temporal cell fate patterning regulation . ( A ) Model of CED-3 collaborating with miRNAs to repress the expression of LIN-14 , LIN-28 , and DISL-2 . Red blocks indicate the new findings . For simplicity , many other factors in the pathway were not included here including additional regulators associated with this pathway that were also identified in our genomic enhancer screen ( see Figure 10—figure supplement 1 ) . ( B ) Hypothetical model for the biochemical role of CED-3 cleavage in protein turnover during development whereby a new N-terminus is generated which could potentially destabilize the protein according to the N-end rule ( see text ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 03510 . 7554/eLife . 04265 . 036Figure 10—figure supplement 1 . A more detailed genetic model for roles of CED-3 caspase in regulating the heterochronic pathway and for showing other genes from our genomic screen in this pathway . Modified from the discussion provided by a recent review ( Rougvie and Moss , 2013 ) , this model is still simplified given the complex and dynamic processes ( Ambros , 1989; Slack and Ruvkun , 1997; Ambros , 2000; Pasquinelli and Ruvkun , 2002; Resnick et al . , 2010; Rougvie and Moss , 2013 ) . Black arrows and blocks indicate aspects of the pathway that were previously known . Red blocks and font indicate the major findings of this study . Blue font indicates factors known to function in this pathway that were also identified in our enhancer screen ( Supplementary file 2 ) as genetic interactors of GW182 ( ain-1 and/or ain-2 ) . The blue asterisk associated with let-7 is to indicate that we also identified the let-7 family miRNAs in our miRNA-ced-3 enhancer screen ( Figure 5A , B ) and that ced-3 ( lf ) was able to enhance their specific temporal cell fate patterning phenotypes as seen with adult-specific alae defects ( Figure 5D ) . The blue asterisks and dashed red lines for miR-1 and miR-246 indicate additional miRNAs we identified in our miRNA-ced-3 enhancer screen ( Figure 5A , B ) . Both miR-1 and miR-246 are predicted to target lin-14 mRNA ( See ‘Materials and methods’ ) but have not been experimentally demonstrated to do so here . They failed to show enhanced specific developmental timing defects and were not pursued further . Dashed arrows and blocks are intended to indicate a combination of complex direct and indirect effects ultimately affecting developmental fate and timing . * Previous works have indicated DAF-12 functions on several components of the pathway ( Rougvie and Moss , 2013; Hochbaum et al . , 2011 ) , providing the basis for us to have observed the enhanced developmental defects associated with ain-1 ( lf ) and daf-12 ( RNAi ) ( Supplementary file 2 ) . The exact mechanism of LIN-66 function is still unclear ( dashed block ) though it is able to negatively regulate LIN-28 function ( Morita and Han , 2006 ) . MSI-1 ( Musashi-1 ) was identified as an RNA-binding protein that may be important for LIN-28-mediated regulation of pre-miRNA processing as well as translational regulation ( Sakakibara et al . , 1996; Kawahara et al . , 2011 ) . Interestingly , we also identified the C . elegans musashi-1 ortholog , msi-1 , as an enhancer in our screen ( Supplementary file 2 ) . Following binding by LIN-28 , pre-let-7 is poly-uridylated by a TUTase ( poly-U-polymerase ) to mark it for destruction . PUP-2 is indicated here as a C . elegans ortholog for TUTase and it was also identified in our screen ( Supplementary file 2 ) . Recently , the RNAse , Dis3l2 ( Dis-3-like RNAse gene 2 ) , was identified as the 3′-5′ exonuclease that degrades poly-uridylated-pre-let-7 miRNA and is the RNAse mutated in Perlman syndrome ( Kawahara et al . , 2011; Chang et al . , 2013; Ustianenko et al . , 2013 ) . We suggest that the C . elegans gene F48E8 . 6 is the C . elegans ortholog of Dis3l2 and have therefore named it DISL-2 ( Dis-3-like RNAse gene 2 ) ( See Figure 6—figure supplement 1 ) . The intricate network of regulatory factors shown here emphasizes the importance of this pathway and how CED-3 caspase cooperates with the miRISC and numerous other factors to control temporal cell fate patterning . DOI: http://dx . doi . org/10 . 7554/eLife . 04265 . 036 We find that the altered LIN-28 expression levels in a ced-3 ( lf ) background or with the caspase-cleavage resistant mutant [LIN-28 ( D31A ) ] in a ced-3 ( wt ) background , are subtle compared to previous findings regarding a lin-28 ( gf ) transgene with deleted lin-4 and let-7 miRNA-binding sites in the 3′ UTR ( Moss et al . , 1997 ) . Consistent with this subtlety , ced-3 ( lf ) alone displays essentially no defect in seam cell numbers ( Figure 6 ) . The physiological effect of this subtle regulation is clearly seen in seam cell temporal patterning when miRNA function is compromised in the ain-1 ( lf ) mutant background . This prominent enhancement indicates that ced-3 has an important role in supporting the robustness of the larval transitions . Based on the pleiotropic phenotypes associated with ced-3 ( lf ) ;ain-1 ( lf ) , such roles may potentially extend to a broad range of cellular processes . Previous studies using model organisms , including our own , have indicated that genetic redundancy by structurally unrelated genes is commonly associated with genes with regulatory functions ( Ferguson et al . , 1987; Fay et al . , 2002; Suzuki and Han , 2006; Costanzo et al . , 2011 ) . Asking the same question for the global miRISC function , our screen , by identifying 118 previously unknown miRISC interactors , thus identified new roles for miRISC in normal developmental processes that are otherwise masked by redundancy and/or pleiotropism , as well as identifying other regulatory mechanisms that collaborate with miRNAs . Examples we found for the latter in this study include genes encoding the POU-homeodomain protein ( ceh-18 , Figure 2—figure supplement 1 ) , the histone acetyltransferase ( pcaf-1 ) , the ras-related GTPase homolog ( ral-1 ) , the homeodomain transcription factor ( unc-39 ) , and the cell-killing ced-3 caspase ( the majority of this study ) ( and others listed in Supplementary file 2 ) . However , the interactions identified in this study most likely reflect only a small portion of miRNA functions because screening for obvious developmental defects under well-fed conditions only permitted us to identify limited physiological functions . Applying various assays , including behavioural assays or animals under various growth or stress conditions , is expected to identify many more miRNA functions . Furthermore , although feeding RNAi has important advantages for such a screen , it is not effective for many genes especially for genes functioning in certain tissues such as neurons . Therefore , genetic screens or analyses under sensitized backgrounds will continue to play a major role in identifying miRNA functions . See Supplementary file 4 for the list of all strains used in this study . In this screen , we wanted to identify genetic pathways that may redundantly cooperate with the miRISC in development . Since the loss of most miRISC function resulted in highly pleiotropic phenotypes ( Zhang et al . , 2009 ) , we chose to score multiple obvious phenotypes ( defined in Supplementary file 1 ) . The ORFeome RNAi feeding library ( Rual et al . , 2004 ) was screened using a 96-well liquid culture format in the double blind . Here , double blind means that no identities for interactors were revealed to anyone setting up the plates , anyone phenotyping the plates , nor anyone processing the scored data until all candidate interactors were confirmed in a secondary screen performed in quadruplicate ( see below ) . Similar to a previously reported method ( Lehner et al . , 2006 ) , a 2 day set up for each screening session was employed ( Figure 1—figure supplement 1A ) . For each scoring session , rrf-3 ( pk1426 , lf ) , ain-1 ( ku322 , lf ) ;rrf-3 ( pk1426 , lf ) , and ain-2 ( tm2432 , lf ) ;rrf-3 ( pk1426 , lf ) were each fed with mock , ain-1 , and ain-2 RNAi cultures in parallel which served as the experimental controls . These controls were set up in 4 sets of triplicate ( n = 12 total for each ) . We identified potential interactors whenever ain-1 ( ku322 ) ;rrf-3 ( pk1426 ) or ain-2 ( tm2432 ) ;rrf-3 ( pk1426 ) showed a significant defect ( Figure 1—figure supplement 1A ) . All candidates were then retested in quadruplicate liquid format . Any gene showing effect in three or more replicates was considered a bona fide interactor by RNAi and their identities were then revealed and confirmed by sequence analysis . Multiple interactors were confirmed by testing the corresponding mutant strains when treated with ain-1 or ain-2 RNAi ( Supplementary file 3 ) . Before any statistical analyses were made , all relevant data sets were first tested for normality using the D'Agostino-Pearson omnibus test . This test also informed us for sufficient sample sizes . We analyzed our results in the following ways: ( 1 ) the Mann–Whitney test was used for pair-wise comparisons , ( 2 ) Chi-square analysis was used to compare distributions of categorical data , and ( 3 ) Fisher's Exact test was used to analyze cases where two categories were most important between two strains ( e . g . , the frequency of normal animals to the pooled frequency of all abnormal animals in each of the tissue-specific rescues or in the RNAi suppression test ) . Use of Fisher's Exact test in such cases prevented outcomes where Chi-square analysis of the same data may identify a rescue as significant only because the abnormal phenotypic categories had changed in distribution relative to the unrescued mutant , but where the fraction of normal animals was not improved . p values and statistical tests were reported throughout the study . Statistics source data have been provided . Similar to the liquid culture format , positive and negative controls were run in parallel to ensure effectiveness of the culturing conditions . RNAi cultures and plates were prepared as previously described ( Fraser et al . , 2000; Timmons et al . , 2001 ) with 100 μg/ml ampicillin . Depending on the experiment , strains were added to RNAi plates in one of the following ways: ( 1 ) bleached strains were synchronised in M9 for 36 hr at 20°C , counted , then added to plates or ( 2 ) either eggs , L2 stage animals , or L4 stage animals were carefully added to lawns . Gravid young adult ain-1 ( lf ) hermaphrodites fed either mock or ced-3 RNAi since hatching were transferred to a new RNAi plate and allowed to lay eggs for 4–5 hr . The young adults were removed and the eggs laid were counted . After 40 hr at 20°C , unhatched eggs and larvae were scored . 64 hr after removing young adults , very few additional larvae were observed for ain-1 ( lf ) animals treated with ced-3 RNAi . Data are from eight independent trials . Synchronous L1 stage animals were added to normal food ( OP50 bacteria ) ( 150–200 worms per trial ) and incubated at 20°C . Animals were scored for developmental stages every 24 hr thereafter . Data are from three to five independent trials . The ain-1 ( lf ) defects were rescued using a fragment of genomic DNA containing ain-1 sequence and native promoter ( Ding et al . , 2005 ) , an ain-1 sequence driven by a dpy-5 ( hypodermal-specific [Thacker et al . , 2006] ) or a ges-1 ( gut-specific [Egan et al . , 1995] ) promoter , and genome-integrated constructs with tissue-specific promoters driving ain-2 expression which has previously been shown by our lab to rescue ain-1 ( lf ) in those tissues ( Zhang et al . , 2011; Kudlow et al . , 2012; Than et al . , 2013 ) . Data are from four to ten independent trials . In the blind , all available miRNA deletion mutants were tested for enhancer phenotypes with ced-3 using the RNAi feeding method on solid agar . The let-7 ( lf ) and lin-4 ( lf ) mutants were excluded since they are very sick . One person picked 10 eggs or 2 L4 stage animals onto mock or ced-3 RNAi in replicates according to a key that was kept confidential . 4 to 5 days later , another person then examined the plates for phenotypes ( defined in Supplementary file 1 ) . All mutants showing an RNAi phenotype were revealed for identity and then crossed with ced-3 ( lf ) mutants in single or in combination and tested for enhancer phenotypes . We employed a published assay to identify subtle apoptotic enhancers using a reporter line: ced-3 ( n2427 , reduction of function ) ;nls106 [lin-11::GFP + lin-15 ( + ) X] ( Reddien et al . , 2007 ) . The ced-3 ( n717 ) ;nIs106 strain served as the positive control for complete loss of ced-3 function , and the mcd-1 ( n3376 ) ;ced-3 ( n2427 , rf ) ;nIs106 strain was the positive control for enhanced ablation of programmed cell death comparable to the strong ced-3 ( n717 ) loss of function allele for accumulation of P9-11 . aap cells , consistent with the previous findings ( Reddien et al . , 2007 ) . Young adults of all strains were scored in the blind for the number of GFP-positive undead P9-11 . aap ventral cord cells . Three independent lines of ced-3 ( n2427 , rf ) ;ain-1 ( lf ) ;nIs106 were scored in the blind ( data for these three lines were combined in Figure 3B ) . This standard method was done as previously described ( Ledwich et al . , 2000 ) . The ced-1 ( e1735 ) mutation was used to enhance visualization of corpses . DIC optics were used to count the head corpses . These predictions were all made by TargetScan 6 . 2 release June 2012 ( Jan et al . , 2011; Lewis et al . , 2005 ) . Adult alae were scored using DIC optics ( Zeiss Axioplan 2 , Thornwood , NY ) at 630× magnification ( Ding et al . , 2005; Zhang et al . , 2007 , 2009 ) . One side of each non-roller adult was scored ( the side facing up ) . All roller phenotype animals ( the three LIN-28::GFP transgenic lines ) were scored in the same way such that all alae that could be viewed were assessed for gaps and quality . Each animal was scored as either normal , low quality alae ( very thin and rough sections ) or gapped alae ( discontinuous alae ) . Animals with both low quality and gapped alae were counted as only gapped alae so that each animal was represented only once . Any thin region of alae that appeared as a gap through the oculars was imaged by the camera ( Zeiss Axiocam MRm ) and evaluated on a large screen . Only alae observed as truly discontinuous by aid of the camera were scored as gapped . This method was applied equally to all strains throughout the study . All seam cell lines were counted on a fluorescent microscope with DIC optics ( Zeiss Axioplan 2 ) at 110× and 630× magnification ( Zhang et al . , 2009 ) at the L1 , L3 , or L4 stage . To prevent over-representation of our sample size , we reported only one side of each animal . We randomly chose to report the top or the left side of the animal , depending on the orientation in the microscopy field . We followed this convention for the single mutants as well . Therefore , one dot corresponds to one side of one animal and each animal is plotted only once ( Figure 6A–C and Figure 6—figure supplement 2 ) . Data are from five independent trials . We hypothesized that loss of both ain-1 and ced-3 resulted in the upregulation of LIN-14 , LIN-28 , and DISL-2 . These factors are normally expressed at high levels beginning in late embryonic development and down-regulated toward the end of the second larval stage . We therefore decided to begin RNAi treatment of ced-3 ( lf ) ;ain-1 ( lf ) animals at the second larval stage and score for phenotypes 48–54 hr later . Animals were considered normal if they were only mildly-to-moderately egg-laying defective and capable of normal motility . Data are from three to six independent trials . The LIN-14 , LIN-28 , and DISL-2 coding sequence templates for in vitro synthesis were each generated first by reverse transcription from mixed stage WT ( N2 ) C . elegans total RNA and then PCR amplified before subcloning into pTNT vector ( Promega , Madison WI ) . The primer sequences are as follows ( Restrictions sites indicated in bold-type , start codons underlined in FWD primers ) : lin-14 FWD , attacgcgtACCATGGCTATGGATCTGCCTGGAACGTCTTCGAAC; REV , attggtaccCTATTGTGGACCTTGAAGAGGAGGAG; lin-28 FWD , attacgcgtACCATGGCTATGTCGACGGTAGTATCGGAGGGA; REV , attggtaccCTCAGTGTCTAGATGATTCTATTCATC; disl-2 FWD , attacgcgtACCATGGCTATGTCAGCAGTTGAAAGTCCCGTT; REV , attggtaccCTACTGAAGAATTGTTGAGCCCGTTTC . Point mutations were generated using Quick Change II kit ( Agilent Technologies , Santa Clara , CA ) . All constructs were sequence-verified . As previously published ( Xue et al . , 1996 ) , cleavage substrates were freshly synthesised with L-35S-Methionine in vitro and used immediately . For caspase inhibitor reactions , zDEVD-fmk caspase-specific inhibitor ( ApexBio , Houston , TX ) or DMSO was added . All cleavage reactions were incubated at 30°C in a thermocycler with heated lid for up to 6 hr . Each panel shown in Figure 7 was performed independently with freshly synthesized L-35S-labeled substrates and independent cleavage reactions for each experiment . Antibody against a LIN-28 C-terminal peptide ( RKHRPEQVAAEEAEA ) was produced by Spring Valley Laboratories ( Sykesville , MD ) using rabbit as the host and purified using a peptide column . Validation of the specificity of the antibody is shown in Figure 8—figure supplement 1A , B . Synchronous L1 stage animals were added to normal food ( OP50 bacteria ) and incubated at 20°C then collected at the indicated hours with food . For each time-point , equivalent protein input from wt , ced-3 ( n717 ) , and ced-3 ( n1286 ) staged animal lysates were resolved by SDS-PAGE and then detected by Western blot using the anti-LIN-28 antibody . Actin was used as loading control ( Anti-Actin antibody , A2066 , Sigma–Aldrich , St . Louis , MO ) . Similarly sized L3 stage animals were picked on a non-fluorescent dissecting scope to blind the selection of animals . Prior to fluorescent illumination , gonad length was observed and measured to ensure animals were of comparable developmental stage ( Ambros and Horvitz , 1984; Moss et al . , 1997; Abbott et al . , 2005 ) . This method should provide a similar distribution of developmental sub-stages for both backgrounds within the L3 stage . No significant difference in gonad extension was found ( Figure 8—figure supplement 2A–C ) . Gonad length was measured and recorded prior to GFP illumination to ensure no bias . All animals were illuminated for 5 s for each picture by DIC optics . Multiple planes through the animal were examined by one person to ensure all GFP positive cells were identified . Another person , who did not take the images , then used ImageJ to obtain integrated GFP intensity values which were reported relative to the gonad length to account for stage ( Figure 8B–D ) or counted the number of GFP positive head cells ( Figure 8—figure supplement 2D , E ) . Data for all animals viewed by DIC were kept and reported . Data for the hypodermal and head cell expression assays come from two and three independent experiments , respectively .	For an organism to develop from a single cell into a collection of many different , specialized cells , different genes must be switched on or off at particular times . However , some of these genes involved in development are ‘redundant’ and carry out the same or similar tasks . This acts like a backup system , so if one of the genes is unable to complete a task , the others can compensate and the organism will still develop correctly . To produce a protein from a gene , the DNA sequence that makes up the gene is used as a template to create another molecule called messenger RNA . Genes can also be ‘silenced’—prevented from making proteins—by small molecules called microRNAs , which bind to messenger RNA molecules and mark them for destruction . MicroRNA molecules therefore play an important role in controlling development . However , as many microRNA molecules often work together , and as many genes are redundant , it can be difficult to discover the effects of specific microRNAs . It is also difficult to discover whether any other mechanisms work alongside the microRNAs to control development . Weaver , Zabinsky et al . used mutant forms of the nematode worm Caenorhabditis elegans , in which microRNA gene regulation did not work correctly , to investigate the mechanisms that work alongside microRNAs to control development . Genes in these worms were silenced; those silenced genes that caused additional developmental defects were considered likely to work ‘redundantly’ in the same role as a microRNA molecule . This revealed over one hundred genes that were previously unknown to work with microRNA molecules . Weaver , Zabinsky et al . focused on one of these genes , called ced-3 . The CED-3 protein produced from this gene is known to execute programmed cell death , a carefully controlled process also known as apoptosis , but was not known to have other developmental functions . However , the worms with mutant forms of the ced-3 gene already have problems performing apoptosis but are otherwise relatively normal , so Weaver , Zabinsky et al . reasoned that the CED-3 protein must also have another role in development . Further investigation revealed that ced-3 mutations most severely disrupt development when they are combined with mutations in one particular family of microRNAs . These microRNAs are particularly important for controlling both when cells specialize into a particular type of cell , and the timing of when certain stages of development happen . Experiments using purified proteins showed that CED-3 breaks down three proteins that are produced from genes controlled by this family of microRNA molecules , and one of these proteins was also broken down by CED-3 in experiments with mutant worms . Weaver , Zabinsky et al . therefore propose that CED-3 is part of a semi-redundant system that ensures the proteins are produced at the right level and at the right time even if the microRNAs insufficiently regulate them . This finding demonstrated both a specific role and specific targets for the CED-3 protein during development , entirely distinct from its role in apoptosis . Although Weaver , Zabinsky et al . have identified a large number of genes that work alongside microRNAs to control development , these are only the genes that cause obvious developmental defects in healthy worms . Further experiments using similar techniques performed on worms under stress may reveal yet more such genes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology" ]	2014	CED-3 caspase acts with miRNAs to regulate non-apoptotic gene expression dynamics for robust development in C. elegans
Efficient regulation of internal homeostasis and defending it against perturbations requires adaptive behavioral strategies . However , the computational principles mediating the interaction between homeostatic and associative learning processes remain undefined . Here we use a definition of primary rewards , as outcomes fulfilling physiological needs , to build a normative theory showing how learning motivated behaviors may be modulated by internal states . Within this framework , we mathematically prove that seeking rewards is equivalent to the fundamental objective of physiological stability , defining the notion of physiological rationality of behavior . We further suggest a formal basis for temporal discounting of rewards by showing that discounting motivates animals to follow the shortest path in the space of physiological variables toward the desired setpoint . We also explain how animals learn to act predictively to preclude prospective homeostatic challenges , and several other behavioral patterns . Finally , we suggest a computational role for interaction between hypothalamus and the brain reward system . Survival requires living organisms to maintain their physiological integrity within the environment . In other words , they must preserve homeostasis ( e . g . body temperature , glucose level , etc . ) . Yet , how might an animal learn to structure its behavioral strategies to obtain the outcomes necessary to fulfill and even preclude homeostatic challenges ? Such , efficient behavioral decisions surely should depend on two brain circuits working in concert: the hypothalamic homeostatic regulation ( HR ) system , and the cortico-basal ganglia reinforcement learning ( RL ) mechanism . However , the computational mechanisms underlying this obvious coupling remain poorly understood . The previously developed classical negative feedback models of HR have tried to explain the hypothalamic function in behavioral sensitivity to the ‘internal’ state , by axiomatizing that animals minimize the deviation of some key physiological variables from their hypothetical setpoints ( Marieb & Hoehn , 2012 ) . To this end , a direct corrective response is triggered when a deviation from setpoint is sensed or anticipated ( Sibly & McFarland , 1974; Sterling , 2012 ) . A key lacuna in these models is how a simple corrective action ( e . g . ‘go eat’ ) in response to a homeostatic deficit might be translated into a complex behavioral strategy for interacting with the dynamic and uncertain external world . On the other hand , the computational theory of RL has proposed a viable computational account for the role of the cortico-basal ganglia system in behavioral adaptation to the ‘external’ environment , by exploiting experienced environmental contingencies and reward history ( Sutton & Barto , 1998; Rangel et al . , 2008 ) . Critically , this theory is built upon one major axiom , namely , that the objective of behavior is to maximize reward acquisition . Yet , this suite of theoretical models does not resolve how the brain constructs the reward itself , and how the variability of the internal state impacts overt behavior . Accumulating neurobiological evidence indicates intricate intercommunication between the hypothalamus and the reward-learning circuitry ( Palmiter , 2007; Yeo & Heisler , 2012; Rangel , 2013 ) . The integration of the two systems is also behaviorally manifest in the classical behavioral pattern of anticipatory responding in which , animals learn to act predictively to preclude prospective homeostatic challenges . Moreover , the ‘good regulator’ theoretical principle implies that ‘every good regulator of a system must be a model of that system’ ( Conant & Ashby , 1970 ) , accentuating the necessity of learning a model ( either explicit or implicit ) of the environment in order to regulate internal variables , and thus , the necessity of associative learning processes being involved in homeostatic regulation . Given the apparent coupling of homeostatic and learning processes , here , we propose a formal hypothesis for the computations , at an algorithmic level , that may be performed in this biological integration of the two systems . More precisely , inspired by previous descriptive hypotheses on the interaction between motivation and learning ( Hull , 1943; Spence , 1956; Mowrer , 1960 ) , we suggest a principled model for how the rewarding value of outcomes is computed as a function of the animal's internal state , and of the approximated need-reduction ability of the outcome . The computed reward is then made available to RL systems that learn over a state-space including both internal and external states , resulting in approximate reinforcement of instrumental associations that reduce or prevent homeostatic imbalance . The paper is structured as follows: After giving a heuristic sketch of the theory , we show several analytical , behavioral , and neurobiological results . On the basis of the proposed computational integration of the two systems , we prove analytically that reward-seeking and physiological stability are two sides of the same coin , and also provide a normative explanation for temporal discounting of reward . Behaviorally , the theory gives a plausible unified account for anticipatory responding and the rise-fall pattern of the response rate . We show that the interaction between the two systems is critical in these behavioral phenomena and thus , neither classical RL nor classical HR theories can account for them . Neurobiologically , we show that our model can shed light on recent findings on the interaction between the hypothalamus and the reward-learning circuitry , namely , the modulation of dopaminergic activity by hypothalamic signals . Furthermore , we show how orosensory information can be integrated with internal signals in a principled way , resulting in accounting for experimental results on consummatory behaviors , as well as the pathological condition of over-eating induced by hyperpalatability . Finally , we discuss limitations of the theory , compare it with other theoretical accounts of motivation and internal state regulation , and outline testable predictions and future directions . A self-organizing system ( i . e . an organism ) can be defined as a system that opposes the second law of thermodynamics ( Friston , 2010 ) . In other words , biological systems actively resist the natural tendency to disorder by regulating their physiological state to fall within narrow bounds . This general process , known as homeostasis ( Cannon , 1929; Bernard , 1957 ) , includes adaptive behavioral strategies for counteracting and preventing self-entropy in the face of constantly changing environments . In this sense , one would expect organisms to reinforce responses that mitigate deviation of the internal state from desired ‘setpoints’ . This is reminiscent of the drive-reduction theory ( Hull , 1943; Spence , 1956; Mowrer , 1960 ) according to which , one of the major mechanisms underlying reward is the usefulness of the corresponding outcome in fulfilling the homeostatic needs of the organism ( Cabanac , 1971 ) . Inspired by these considerations ( i . e . preservation of self-order and reduction of deviations ) , we propose a formal definition of primary reward ( equivalently: reinforcer , economic utility ) as the approximated ability of an outcome to restore the internal equilibrium of the physiological state . We then demonstrate that our formal homeostatic reinforcement learning framework accounts for some phenomena that classical drive-reduction was unable to explain . We first define ‘homeostatic space’ as a multidimensional metric space in which each dimension represents one physiologically-regulated variable ( the horizontal plane in Figure 1 ) . The physiological state of the animal at each time t can be represented as a point in this space , denoted by Ht= ( h1 , t , h2 , t , . . , hN , t ) , where hi , t indicates the state of the i-th physiological variable . For example , hi , t can refer to the animal's glucose level , body temperature , plasma osmolality , etc . The homeostatic setpoint , as the ideal internal state , can be denoted by H= ( h1 , h2* , . . , hN* ) . As a mapping from the physiological to the motivational state , we define the ‘drive’ as the distance of the internal state from the setpoint ( the three-dimensional surface in Figure 1 ) : ( 1 ) D ( Ht ) =∑i=1N\|hi−hi , t\|nmm and n are free parameters that induce important nonlinear effects on the mapping between homeostatic deviations and their motivational consequences . Note that for the simple case of m = n = 1 , the drive function reduces to Euclidian distance . We will later consider more general nonlinear mappings in terms of classical utility theory . We will also discuss that the drive function can be viewed as equivalent to the information-theoretic notion of surprise , defined as the negative log-probability of finding an organism in a certain state ( D ( Ht ) =−ln p ( Ht ) ) . 10 . 7554/eLife . 04811 . 003Figure 1 . Schematics of the model in an exemplary two-dimensional homeostatic space . Upon performing an action , the animal receives an outcome Kt from the environment . The rewarding value of this outcome depends on its ability to make the internal state , Ht , closer to the homeostatic setpoint , H , and thus reduce the drive level ( the vertical axis ) . This experienced reward , denoted by r ( Ht , Kt ) , is then learned by an RL algorithm . Here a model-free RL algorithm is shown in which a reward prediction error signal is computed by comparing the realized reward and the expected rewarding value of the performed response . This signal is then used to update the subjective value attributed to the corresponding response . Subjective values of alternative choices bias the action selection process . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 003 Having defined drive , we can now provide a formal definition for primary reward . Let's assume that as the result of an action , the animal receives an outcome ot at time t . The impact of this outcome on different dimensions of the animal's internal state can be denoted by Kt= ( k1 , t , k2 , t , . . , kN , t ) . For example , ki , t can be the quantity of glucose received as a result of outcome ot . Hence , the outcome results in a transition of the physiological state from Ht to Ht+1=Ht+Kt ( See Figure 1 ) and thus , a transition of the drive state from D ( Ht ) to D ( Ht+1 ) =D ( Ht+Kt ) . Accordingly , the rewarding value of this outcome can be defined as the consequent reduction of drive: ( 2 ) r ( Ht , Kt ) =D ( Ht ) −D ( Ht+1 ) =D ( Ht ) −D ( Ht+Kt ) Intuitively , the rewarding value of an outcome depends on the ability of its constituting elements to reduce the homeostatic distance from the setpoint or equivalently , to counteract self-entropy . As discussed later , the additive effect ( Kt ) of these constituting elements on the internal state can be approximated by the orosensory properties of outcomes . We will also discuss how erroneous estimation of drive reduction can potentially be a cause for maladaptive consumptive behaviors . We hypothesize in this paper that the primary reward constructed as proposed in Equation 2 is used by the brain's reward learning machinery to structure behavior . Incorporating this physiological reward definition in a normative RL theory allows us to derive one major result of our theory , which is that the rationality of behavioral patterns is geared toward maintaining physiological stability . Here we show that our definition of reward reconciles the RL and HR theories in terms of their normative assumptions: reward acquisition and physiological stability are mathematically equivalent behavioral objectives . More precisely , given the proposed definition of reward and given that animals discount future rewards ( Chung & Herrnstein , 1967 ) , any behavioral policy , π , that maximizes the sum of discounted rewards ( SDR ) also minimizes the sum of discounted deviations from the setpoint , and vice versa . In fact , starting from an initial internal state H0 , the sum of discounted deviations ( SDD ) for a certain behavioral policy π that causes the internal state to move in the homeostatic space along the trajectory p ( π ) , can be defined as: ( 3 ) SDDπ ( H0 ) =∫p ( π ) γt . D ( Ht ) . dt Similarly , the sum of discounted rewards ( SDR ) for a policy π can be defined as: ( 4 ) SDRπ ( H0 ) =∫p ( π ) γt . rt . dt=∫p ( π ) γt . ( D ( Ht ) −D ( Ht+dt ) ) . dt It is then rather straightforward to show that for any initial state H0 , we will have ( See ‘Materials and methods’ for the proof ) : ( 5 ) if γ<1 : argminπSDDπ ( H0 ) =argmaxπSDRπ ( H0 ) where γ is the discount factor . In other words , the same behavioral policy satisfies optimal reward-seeking as well as optimal homeostatic maintenance . In this respect , reward acquisition sought by the RL system is an efficient means to guide an animal's behavior toward fulfilling the basic objective of defending homeostasis . Thus , our theory suggests a physiological basis for the rationality of reward seeking . In the domain of animal behavior , one fundamental question is why animals should discount rewards the further they are in the future . Our theory indicates that reward seeking without discounting ( i . e . , if γ = 1 ) would not lead , and may even be detrimental , to physiological stability ( See ‘Materials and methods’ ) . Intuitively , this is because a future-discounting agent would always tend to expedite bigger rewards and postpone punishments . Such an agent , therefore , tries to reduce homeostatic deviations ( which is rewarding ) as soon as possible , and thus , tries to find the shortest path toward the setpoint . A non-discounting agent , in contrast , can always compensate for a deviation-induced punishment by reducing that deviation any time in the future . While the formal proof of the necessity of discounting is given in the ‘Materials and methods’ , let us give an intuitive explanation . Imagine you had to plan a 1-hr hill walk from a drop-point toward a pickup point , during which you wanted to minimize the height ( equivalent to drive ) summed over the path you take . In this summation , if you give higher weights to your height in the near future as compared to later times , the optimum path would be to descend the hill and spend as long as possible at the bottom ( i . e . homeostatic setpoint ) before returning to the pickup point . Equation 5 shows that this optimization is equivalent to optimizing the total discounted rewards along the path , given that descending and ascending steps are defined as being rewarding and punishing , respectively ( Equation 2 ) . In contrast , if at all points in time you give equal weights to your height , then the summed height over path only depends on the drop and pickup points , since every ascend can be compensated with a descend at any time . In other words , in the absence of discounting , the rewarding value of a behavioral policy that changes the internal state only depends on the initial and final internal states , regardless of its trajectory in the homeostatic space . Thus , when γ = 1 , the values of any two behavioral policies with equal net shifts of the internal state are equal , even if one policy moves the internal state along the shortest path , whereas the other policy results in large deviations of the internal state from the setpoint and threatens survival . These results hold for any form of temporal discounting ( e . g . , exponential , hyperbolic ) . In this respect , our theory provides a normative explanation for the necessity of temporal discounting of reward: to maintain internal stability , it is necessary to discount future rewards . A paradigmatic example of behaviors governed by the internal state is the anticipatory responses geared to preclude perturbations in regulated variables even before any physiological depletion ( negative feedback ) is detectable . Anticipatory eating and drinking that occur before any discernible homeostatic deviation ( Woods & Seeley , 2002 ) , anticipatory shivering in response to a cue that predicts the cold ( Mansfield et al . , 1983; Hjeresen et al . , 1986 ) , and insulin secretion prior to meal initiation ( Woods , 1991 ) , are only a few examples of anticipatory responding . One clear example of a conditioned homeostatic response is animals' progressive tolerance to ethanol-induced hypothermia . Experiments show that when ethanol injections are preceded ( i . e . , are predictable ) by a distinctive cue , the ethanol-induced drop of the body core temperature of animals diminishes along the trials ( Mansfield & Cunningham , 1980 ) . Figure 2 shows that when the temperature was measured 30 , 60 , 90 , and 120 min after daily injections , the drop of temperature below the baseline was significant on the first day , but gradually disappeared over 8 days . Interestingly , in the first extinction trial on the ninth day where the ethanol was omitted , the animal's temperature exhibited a significant increase above normal after cue presentation . This indicates that the enhanced tolerance response to ethanol is triggered by the cue , and results in an increase of temperature in order to compensate for the forthcoming ethanol-induced hypothermia . Thus , this tolerance response is mediated by associative learning processes , and is aimed at regulating temperature . Here we demonstrate that the integration of HR and RL processes accounts for this phenomenon . 10 . 7554/eLife . 04811 . 004Figure 2 . Experimental results ( adapted from Mansfield & Cunningham , 1980 ) on the acquisition and extinction of conditioned tolerance response to ethanol . ( A ) In each block ( day ) of the experiment , the animal received ethanol injection after the presentation of the stimulus . ( B ) The change in the body temperature was measured 30 , 60 , 90 , and 120 min after ethanol administration . Initially , the hypothermic effect of ethanol decreased the body temperature of animals . After several training days , however , animals learned to activate a tolerance response upon observing the stimulus , resulting in smaller deviations from the temperature setpoint . If the stimulus was not followed by ethanol injection , as in the first day of extinction ( E1 ) , the activation of the conditioned tolerance response resulted in an increase in body temperature . The tolerance response was weakened after several ( four ) extinction sessions , resulting in increased deviation from the setpoint in the first day of re-acquisition ( R1 ) , where presentation of the cue was again followed by ethanol injection . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 004 We simulate the model in an artificial environment where on every trial , the agent can choose between initiating a tolerance response and doing nothing , upon observing a cue ( Figure 3A ) . The cue is then followed by a forced drop of temperature , simulating the effect of ethanol ( Figure 3B ) . We also assume that in the absence of injection , the temperature does not change . However , if the agent chooses to initiate the tolerance response in this condition , the temperature increases gradually ( Figure 3D ) . Thus , if ethanol injection is preceded by cue-triggered tolerance response , the combined effect ( Figure 3F , as superposition of Figure 3B , D ) will have less deviation from the setpoint as compared to when no response is taken ( Figure 3B ) . As punishment ( as the opposite of reward ) in our model is defined by the extent to which the deviation from the setpoint increases , the ‘null’ response will have a bigger punishing value than the ‘tolerance’ response and thus , the agent gradually reinforces the ‘tolerance’ action ( Figure 3C ) ( More precisely , the rewarding value of each action is defined by the sum of discounted drive-reductions during the 24 hr upon taking that action ) . This results in gradual fade of the ethanol-induced deviation of temperature from setpoint ( Figure 3E; See Figure 3—source data 1 for simulation details ) . 10 . 7554/eLife . 04811 . 005Figure 3 . Simulation result on anticipatory responding . ( A ) In every trial , the simulated agent can choose between initiating a tolerance response and doing nothing , upon observing the stimulus . Regardless of the agent's choice , ethanol is administered after 1 hr , followed by four temperature measurements every 30 min . ( B ) Dynamics of temperature upon ethanol injection . ( C ) Learning curve for choosing the ‘tolerance’ response . ( D ) Dynamics of temperature upon initiating the tolerance response . ( E ) Temperature profile during several simulated trails . ( F ) Dynamics of temperature upon initiating the tolerance response , followed by ethanol administration . Plots c and e are averaged over 500 simulated agents . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 00510 . 7554/eLife . 04811 . 006Figure 3—source data 1 . Free parameters for the anticipatory responding simulation . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 006 Clearly , if after this learning process cue-presentation is no longer followed by ethanol injection ( as in the first extinction trial , E1 ) , the cue-triggered tolerance response increases the temperate beyond the setpoint ( Figure 3E ) . In general , these results show that the tolerance response caused by predicted hypothermia is an optimal behavior in terms of minimizing homeostatic deviation and thus , maximizing reward . Thus , this optimal homeostatic maintenance policy is acquired by associative learning mechanisms . Our theory implies that animals are capable of learning not only Pavlovian ( e . g . shivering , or tolerance to ethanol ) , but also instrumental anticipatory responding ( e . g . , pressing a lever to receive warmth , in response to a cold-predicting cue ) . This prediction is in contrast to the theory of predictive homeostasis ( also known as allostasis ) where anticipatory behaviors are only reflexive responses to the predicted homeostatic deprivation upon observing cues ( Woods & Ramsay , 2007; Sterling , 2012 ) . The definition of the drive function ( Equation 1 ) in our model has two degrees of freedom: m and n are free parameters whose values determine the properties of the homeostatic space metric . Appropriate choice of m and n ( n > m > 1 ) permits our theory to account for the following four key behavioral phenomena in a unified framework . First , it accounts for the fact that the reinforcing value of an appetitive outcome increases as a function of its dose ( Kt ) ( Figure 4A ) : ( 6 ) ∂r ( Ht , Kt ) ∂kj , t>0 : for Kt= ( 0 , 0 , … , kj , t , … , 0 ) and kj , t>010 . 7554/eLife . 04811 . 007Figure 4 . Schematic illustration of the behavioral properties of the drive function . ( A ) excitatory effect of the dose of outcome on its rewarding value . ( B , C ) excitatory effect of deprivation level on the rewarding value of outcomes: Increased deprivation increases the rewarding value of reducing drive ( B ) , and increases the punishing value of increasing drive ( C ) . ( D ) inhibitory effect of irrelevant drives on the rewarding value of outcomes . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 007 This is supported by the fact that in progressive ratio schedules of reinforcement rats maintain higher breakpoints when reinforced with bigger appetitive outcomes , reflecting higher motivation toward them ( Hodos , 1961; Skjoldager et al . , 1993 ) . Secondly , the model accounts for the potentiating effect of the deprivation level on the reinforcing value ( i . e . , food will be more rewarding when the animal is hungrier ) ( Figure 4B , C ) : ( 7 ) ∂r ( Ht , Kt ) ∂\|hj−hj , t\|>0 : for Kt= ( 0 , 0 , … , kj , t , … , 0 ) and kj , t>0 This is consistent with experimental evidence showing that the level of food deprivation in rats increases the breakpoint in a progressive ratio schedule ( Hodos , 1961 ) . Note that this point effectively establishes a formal extension for the ‘incentive’ concept as defined by incentive salience theory ( Berridge , 2012 ) ( Discussed later ) . Thirdly , the theory accounts for the inhibitory effect of irrelevant drives , which is consistent with a large body of behavioral experiments showing competition between different motivational systems ( See Dickinson & Balleine , 2002 for a review ) . In other words , as the deprivation level for one need increases , it inhibits the rewarding value of other outcomes that satisfy irrelevant motivational systems ( Figure 4D ) : ( 8 ) ∂r ( Ht , Kt ) ∂\|hi−hi , t\|>0 : for all i≠j , where Kt= ( 0 , 0 , … , kj , t , … , 0 ) and kj , t>0 Intuitively , one does not play chess , or even search for sex , on an empty stomach . As some examples , calcium deprivation reduces the appetite for phosphorus , and hunger inhibits sexual behavior ( Dickinson & Balleine , 2002 ) . Finally , the theory naturally captures the risk-aversive nature of behavior . The rewarding value in our model is a concave function of the corresponding outcome magnitude: ( 9 ) ∂2r ( Ht , Kt ) ∂kj , t2<0 : for Kt= ( 0 , 0 , … , kj , t , … , 0 ) and kj , t>0 It is well known that the concavity of the economic utility function is equivalent to risk aversion ( Mas-Colell et al . , 1995 ) . Indeed , simulating the model shows that when faced with two options with equal expected payoffs , the model learns to choose the more certain option as opposed to the risky one ( Figure 5; See Figure 5—source data 1 for simulation details ) . This is because frequent small deviations from the setpoint are preferable to rare drastic deviations . In fact , our theory suggests the intuition that when the expected physiological instability caused by two behavioral options are equal , organisms do not choose the risky option , because the severe , though unlikely , physiological instabilities that it can cause might be life-threatening . 10 . 7554/eLife . 04811 . 008Figure 5 . Risk aversion simulation . In a conditioned place preference paradigm , the agent's presence in the left and the right compartments has equal expected payoffs , but different levels of risk ( A ) Panel ( B ) shows the Markov decision process of the same task . In fact , in every trial , the agent chooses whether to stay it the current compartment , or transit to the other one . The average input of energy per trial , regardless of the animal's choice , is set such that it is equal to the animal's normal energy expenditure . Thus , the internal state stays close to its initial level , which is equal to the setpoint here ( D ) . The model learns to prefer the non-risky over the risky compartments ( C ) in order to avoid severe deviations from the setpoint . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 00810 . 7554/eLife . 04811 . 009Figure 5—source data 1 . Free parameters for the risk-aversion simulations . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 009 Our unified explanation for the above four behavioral patterns suggests that they may all arise from the functional form of the mapping from the physiological to the motivational state . In this sense , we propose that these behavioral phenomena are signatures of the coupling between the homeostatic and the associative learning systems . We will discuss later that m , n , and H* can be regarded as free parameters of an evolutionary process , which eventually determine the equilibrium density of the species . Note that the equations in this section hold only when the internal state remains below the setpoint . However , the drive function is symmetric with respect to the setpoint and thus , analogous conclusions can be derived for other three quarters of the homeostatic space . Since learning requires experience , learning whether an action in a certain internal state decreases or increases the drive ( i . e . is rewarding or punishing , respectively ) would require our model to have experienced that internal state . Living organisms , however , cannot just experience internal states with extreme and life threatening homeostatic deviations in order to learn that the actions that cause them are bad . For example , once the body temperature goes beyond 45°C , the organism can never return . We now show how our model manages this problem; that is , it avoids voluntarily experiencing extreme homeostatic deviations and hence ensures that the animal does not voluntarily endanger its physiological integrity ( simulations in Figure 6 ) . In the simplest case , let us assume that the model is tabula rasa: it starts from absolute ignorance about the value of state–action pairs , and can freely change its internal state in the homeostatic space . In a one-dimensional space , it means that the agent can freely increase or decrease the internal state ( Figure 6—figure supplement 1 ) . As the value of ‘increase’ and ‘decrease’ actions at all internal states are initialized to zero , the agent starts by performing a random walk in the homeostatic space . However , the probability of choosing the same action for z times in a row decreases exponentially as z increases ( p ( z ) =2−z ) : for example , the probability of choosing ‘increase’ is 2−1 = 0 . 5 , the probability of choosing two successive ‘increases’ is 2−1 = 0 . 25 , the probability of choosing three successive ‘increases’ is 2−3 = 0 . 125 , and so on . Thus , it is highly likely for the agent to return at least one step back , before getting too far from its starting point . When the agent returns to a state it had previously experienced , going in the same deviation-increasing direction will be less likely than the first time ( i . e . , than 50–50 ) , since the agent has already experienced the punishment caused by that state–action pair once . Repetition of this process results in the agent gradually getting more and more attracted to the setpoint , without ever having experienced internal states that are beyond a certain limit ( i . e . the brink of death ) . 10 . 7554/eLife . 04811 . 010Figure 6 . Simulations showing that the model avoids extreme deviations . Starting from 30 , the agent can either decrease or increase its internal state by one unit in each trial . ( A ) The number of visits at each internal state after 106 trials . ( B ) The drive function in the one-dimensional homeostatic space . ( setpoint = 0 ) . The mean ( C ) and standard deviation ( D ) of the internal state of 105 agents , along 1500 trials . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 01010 . 7554/eLife . 04811 . 011Figure 6—source data 1 . Free parameters for the simulations showing that the model avoids extreme homeostatic deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 01110 . 7554/eLife . 04811 . 012Figure 6—figure supplement 1 . The Markov Decision Process used for simulation results presented in Figure 6 and Figure 6—figure supplements 2–7 . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 01210 . 7554/eLife . 04811 . 013Figure 6—figure supplement 2 . Value function ( A ) and choice preferences ( B ) for state–action pairs after simulating one agent for 106 trials ( As in Figure 6 ) . The parameters of the model where as follows: α = 0 . 4 , β = 0 . 05 , γ = 0 . 9 , n = 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 01310 . 7554/eLife . 04811 . 014Figure 6—figure supplement 3 . Simulation results replicating Figure 6 , with the difference that the initial internal state was zero . ( A ) The number of visits at each internal state after 106 trials . ( B ) The drive function in the one-dimensional homeostatic space ( setpoint=0 ) . Value function ( C ) and choice preferences ( D ) for state-action pairs after simulating one agent for 106 trials . The mean ( E ) and standard deviation ( F ) of the internal state of 105 agents , along 1500 trials . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 01410 . 7554/eLife . 04811 . 015Figure 6—figure supplement 4 . Simulation results replicating Figure 6 , with the difference that the initial internal state was zero , and the rate of exploration , β , was 0 . 03 . ( A ) The number of visits at each internal state after 106 trials . ( B ) The drive function in the one-dimensional homeostatic space ( setpoint=0 ) . Value function ( C ) and choice preferences ( D ) for state-action pairs after simulating one agent for 106 trials . The mean ( E ) and standard deviation ( F ) of the internal state of 105 agents , along 1500 trials . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 01510 . 7554/eLife . 04811 . 016Figure 6—figure supplement 5 . Simulation results replicating Figure 6 , with the difference that the initial internal state was zero , and also m = n = 1 . ( A ) The number of visits at each internal state after 106 trials . ( B ) The drive function in the one-dimensional homeostatic space ( setpoint=0 ) . Value function ( C ) and choice preferences ( D ) for state-action pairs after simulating one agent for 106 trials . The mean ( E ) and standard deviation ( F ) of the internal state of 105 agents , along 1500 trials . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 01610 . 7554/eLife . 04811 . 017Figure 6—figure supplement 6 . Simulation results replicating Figure 6 , with the difference that the initial internal state was zero , and the discount factor , γ , was zero . ( A ) The number of visits at each internal state after 106 trials . ( B ) The drive function in the one-dimensional homeostatic space ( setpoint=0 ) . Value function ( C ) and choice preferences ( D ) for state-action pairs after simulating one agent for 106 trials . The mean ( E ) and standard deviation ( F ) of the internal state of 105 agents , along 1500 trials . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 01710 . 7554/eLife . 04811 . 018Figure 6—figure supplement 7 . Simulation results replicating Figure 6 , with the difference that the initial internal state was zero , and the discount factor , γ , was one ( no discounting ) . ( A ) The number of visits at each internal state after 106 trials . ( B ) The drive function in the one-dimensional homeostatic space ( setpoint=0 ) . Value function ( C ) and choice preferences ( D ) for state-action pairs after simulating one agent for 106 trials . The mean ( E ) and standard deviation ( F ) of the internal state of 105 agents , along 1500 trials . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 018 Simulating the model in a one-dimensional space shows that even after starting from a rather deviated internal state ( initial state = 30 , setpoint = 0 ) , the agent never visits states with a deviation of more than 40 units after 106 trials ( every action is assumed to change the state by one unit ) ( Figure 6A; See Figure 6—figure supplements 1 , 2 , and Figure 6—source data 1 for simulation details ) . Also , simulating 105 agents over 1500 trials ( starting from state 30 ) shows that the mean value of the internal state across all agents converges to the setpoint ( Figure 5C ) , and its variance converges to a steady-state level ( Figure 5D ) . This shows that all agents stay within certain bounds around the setpoint ( The maximum deviation from the setpoint among all the 105 agents over the 1500 trials was 61 ) . Also , this property of the model is shown to be insensitive to the parameters of the model , like the initial internal state ( Figure 6—figure supplement 3 ) , the rate of exploration ( Figure 6—figure supplement 4 ) , m and n ( Figure 6—figure supplement 5 ) , or the discount factor ( Figure 6—figure supplements 6 , 7 ) . These parameters only affect the rate of convergence or the distribution over visited states , but not the general property of never-visiting-drastic-deviations ( existence of a boundary ) . Moreover , this property can be generalized to multi-dimensional homeostatic spaces . Therefore , our theory suggests a potential normative explanation for how animals ( who might be a priori naïve about potential dangers of certain internal states ) would learn to avoid extreme physiological instability , without ever exploring how good or bad they are . As mentioned , we hypothesize that orosensory properties of food and water provide the animal with an estimate , K^t , of their true post-ingestive effect , Kt , on the internal state . Such association between sensory and post-ingestive properties could have been developed through prior learning ( Swithers et al . , 2009; Swithers et al . , 2010; Beeler et al . , 2012 ) or evolutionary mechanisms ( Breslin , 2013 ) . Based on this sensory approximation , the only information required to compute the reward ( and thus the reward prediction error ) is the current physiological state ( Ht ) and the sensory-based approximation of the nutritional content of the outcome ( K^t ) : ( 10 ) r ( Ht , K^t ) =D ( Ht ) −D ( Ht+K^t ) Clearly , the evolution of the internal state itself depends only on the actual ( Kt ) post-ingestive effects of the outcome . That is Ht+1=Ht+Kt . According to Equation 10 , the reinforcing value of food and water outcomes can be approximated as soon as they are sensed/consumed , without having to wait for the outcome to be digested and the drive to be reduced . This proposition is compatible with the fact that dopamine neurons exhibit instantaneous , rather than delayed , burst activity in response to unexpected food reward ( Schneider , 1989; Schultz et al . , 1997 ) . Moreover , it might provide a formal explanations for the experimental fact that intravenous injection ( and even intragastric intubation , in some cases ) of food is not rewarding even though its drive reduction effect is equal to when it is ingested orally ( Miller & Kessen , 1952 ) ( See also Ren et al . , 2010 ) . In fact , if the post-ingestive effect of food is estimated by its sensory properties , the reinforcing value of intravenously injected food that lacks sensory aspects will be effectively zero . In the same line of reasoning , the theory suggests that animals' motivation toward palatable foods , such as saccharine , that have no caloric content ( and thus no need-reduction effect ) is due to erroneous over-estimation of their drive-reduction capacity , misguided by their taste or smell . Note that the rationality of our theory , as shown in Equation 5 , holds only as long as K^t is an unbiased estimation of Kt . Otherwise , pathological conditions could emerge . Last but not least , the orosensory-based approximation provides a computational hypothesis for the separation of reinforcement and satiation effects . A seminal series of experiments ( McFarland , 1969 ) demonstrated that the reinforcing and satiating ( i . e . , need reduction ) effects of drinking behavior , dissociable from one another , are governed by the orosensory and alimentary components of the water , respectively . Two groups of water-deprived animals learned to press a green key to self-administer water orally . After this pre-training session , pressing the green key had no consequence anymore , whereas pressing a novel yellow key resulted in the oral delivery of water in one group , and intragastric ( through a fistula ) delivery of water in the second group . Results showed that the green key gradually extinguished in both groups ( Figure 7A , B ) . During this time , responding on the yellow key in the oral group initially increased but then gradually extinguished ( rise-fall pattern; Figure 7A ) . The second group , however , showed no motivation for the yellow key ( Figure 7B ) . This shows that only oral , but not intragastric , self-administration of water is reinforcing for thirsty animals . Our model accounts for these behavioral dynamics . 10 . 7554/eLife . 04811 . 019Figure 7 . Experimental results ( adapted from McFarland , 1969 ) on learning the reinforcing effect of oral vs . intragastric delivery of water . Thirsty animals were initially trained to peck at a green key to receive water orally . In the next phase , pecking at the green key had no consequence , while pecking at a novel yellow key resulted in oral delivery of water in one group ( A ) , and intragastric injection of the same amount of water through a fistula in a second group ( B ) . In the first group , responding was rapidly transferred from the green to the yellow key , and then suppressed . In the fistula group , the yellow key was not reinforced . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 019 Simulating the model shows that the agent's subjective probability of receiving water upon pressing the green key gradually decreases to zero in both groups ( Figure 8C , D ) . As this predicted outcome ( alimentary content ) decreases , its approximated thirst-reduction effect ( equal to reward in our framework ) decreases as well , resulting in the extinction of pressing the green key ( Figure 8A , B ) . As for the yellow key , the oral agent initially increases the rate of responding ( Figure 8A ) as the subjective probability of receiving water upon pressing the yellow key increases ( Figure 8C ) . Gradually , however , the internal state of the animal reaches the homeostatic setpoint ( Figure 8E ) , resulting in diminishing motivation ( thirst-reduction effect ) of seeking water ( Figure 8A ) . Thus , our model shows that whereas the ascending limb of the response curve represents a learning effect , the descending limb is due to mitigated homeostatic imbalance ( i . e . , unlearning vs . satiation ) . Notably , classical RL models only explain the ascending , and classical HR models only explain the descending pattern . 10 . 7554/eLife . 04811 . 020Figure 8 . Simulation results replicating the data from McFarland ( 1969 ) on learning the reinforcing effect of oral vs . intragastric delivery of water . As in the experiment , two groups of simulated agents were pre-trained to respond on the green key to receive oral delivery of water . During the test phase , the green key had no consequence , whereas a novel yellow key resulted in oral delivery in one group ( A ) and intragastric injection in the second group ( B ) . All agents started this phase in a thirsty state ( initial internal state = 0; setpoint = 0 ) . In the oral group , responding transferred rapidly from the green to the yellow key and was then suppressed ( A ) as the internal state approached the setpoint ( E ) . This transfer is due to gradually updating the subjective probability of receiving water outcome upon responding on either key ( C ) . In the fistula group , as the water was not sensed , the outcome expectation converged to zero for both keys ( D ) and thus , responding was extinguished ( B ) . As a result , the internal state changed only slightly ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 02010 . 7554/eLife . 04811 . 021Figure 8—source data 1 . Free parameters for the reinforcing vs . satiation simulations . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 02110 . 7554/eLife . 04811 . 022Figure 8—figure supplement 1 . A model-based homeostatic RL system . Upon performing an action in a certain state , the agent receives an outcome , Kt , which results in the internal state to shift from Ht to Ht+Kt . At the same time , sensory properties of the outcome are sensed by the agent . Based on this information , the agent updates the state-action-outcome associations . In fact , the agent learns to predict the sensory properties , K^t , of the outcome that is expected to be received upon performing a certain action . Having learned these associations , the agent can estimate the rewarding value of different options . That is , when the agent is in a certain state , it predicts the outcome K^t , expected to result from each behavioral policy . Based on K^t and the internal state Ht , the agent can approximate the drive-reduction reward . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 02210 . 7554/eLife . 04811 . 023Figure 8—figure supplement 2 . The Markov Decision Process used for simulating the reinforcing vs . satiation effects of water . At each time point , the agent can choose between doing nothing ( nul ) or pecking at either the green or the yellow key . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 023 In contrast to the oral agent , the fistula agent never learns to press the yellow key ( Figure 8B ) . This is because the approximated alimentary content attributed to this response remains zero ( Figure 8D ) and so does its drive-reduction effect . Note that as above , the sensory-based approximation ( K^t ) of the alimentary effect of water in the oral and fistula cases is assumed to be equal to its actual effect ( Kt ) and zero , respectively ( See Figure 8—figure supplements 1 , 2 , and Figure 8—source data 1 for simulation details ) . Our theory also suggests that in contrast to reinforcement ( above ) , satiation is independent of the sensory aspects of water and only depends on its post-ingestive effects . In fact , experiments show that when different proportions of water were delivered via the two routes in different groups , satiation ( i . e . , suppression of responding ) only depended on the total amount of water ingested , regardless of the delivery route ( McFarland , 1969 ) . Our model accounts for these data ( Figure 9 ) , since the evolution of the internal state only depends on the actual water ingested . For example , whether water is administered completely orally ( Figure 9 , left column ) or half-orally-half-intragastrically ( Figure 9 , right column ) , the agent stops seeking water when the setpoint is reached . As only oral delivery is sensed , the subjective outcome magnitude converges to 1 ( Figure 9C ) and 0 . 5 ( Figure 9D ) units for the two cases , respectively . When the setpoint is reached , consuming more water results in overshooting the setpoint ( increasing homeostatic deviation ) and thus , is punishing . Therefore , both agents self-administer the same total amount of water , equal to what is required for reaching the setpoint . 10 . 7554/eLife . 04811 . 024Figure 9 . Simulation results of the satiation test . Left column shows results for the case where water was received only orally . Rate of responding drops rapidly ( A ) as the internal state approaches the setpoint ( E ) . Also , the agent learns rapidly that upon every key pecking , it receives 1 . 0 unit of water ( C ) . On the right column , upon every key-peck , 0 . 5 unit of water is received orally , and 0 . 5 unit is received via the fistula . As only oral delivery is sensed by the agent , the subjective outcome-magnitude converges to 0 . 5 ( D ) . As a result , the reinforcing value of key-pecking is less than that of the oral case and thus , the rate of responding is lower ( B ) . This in turn results in slower convergence of the internal state to the setpoint ( F ) . The MDP and the free parameters used for simulation are the same as in Figure 8 . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 024 However , as the sensed amount of water is bigger in the completely-oral case , water-seeking behavior is approximated to have a higher thirst-reduction effect . As a result , the reinforcing value of water-seeking is higher in the oral case ( as compared to the half-oral-half- intragastric case ) and thus , the rate of responding is higher . This , in turn , results in faster convergence of the internal state to the setpoint ( compare Figure 9E , F ) . In this respect , we predict that the oral/fistula proportion affects the speed of satiation: the higher the proportion is , the faster the satiety state is reached and thus , the faster the descending limb of responding emerges . Homeostatic regulation critically depends on sensing the internal state . In the case of energy regulation , for example , the arcuate nucleus of the hypothalamus integrates peripheral hormones including leptin , insulin , and ghrelin , whose circulating levels reflect the internal abundance of fat , abundance of carbohydrate , and hunger , respectively ( Williams & Elmquist , 2012 ) . In our model , the deprivation level has an excitatory effect on the rewarding value of outcomes ( Equation 7 ) and thus on the reward prediction error ( RPE ) . Consistently , recent evidence indicates neuronal pathways through which energy state-monitoring peptides modulate the activity of midbrain dopamine neurons , which supposedly carry the RPE signal ( Palmiter , 2007 ) . Namely , orexin neurons , which project from the lateral hypothalamus area to several brain regions including the ventral tegmental area ( VTA ) ( Sakurai et al . , 1998 ) , have been shown to have an excitatory effect on dopaminergic activity ( Korotkova et al . , 2003; Narita et al . , 2006 ) , as well as feeding behavior ( Rodgers et al . , 2001 ) . Orexin neurons are responsive to peripheral metabolic signals as well as to the animal's deprivation level ( Burdakov et al . , 2005 ) , as they are innervated by orexigenic and anorexigenic neural populations in the arcuate nucleus where circulating peptides are sensed . Accordingly , orexin neurons are suggested to act as an interface between internal states and the reward learning circuit ( Palmiter , 2007 ) . In parallel with the orexinergic pathway , ghrelin , leptin and insulin receptors are also expressed on the VTA dopamine neurons , providing a further direct interface between the HR and RL systems . Consistently , whereas leptin and insulin inhibit dopamine activity and feeding behavior , ghrelin has an excitatory effect on them ( See Palmiter , 2007 for a review ) . The reinforcing value of food outcome ( and thus RPE signal ) in our theory is not only modulated by the internal state , but also by the orosensory information that approximates the need-reduction effects . In this respect , endogenous opioids and μ-opioid receptors have long been implicated in the hedonic aspects of food , signaled by its orosensory properties . Systemic administration of opioid antagonists decreases subjective pleasantness rating and affective responses for palatable foods in humans ( Yeomans & Wright , 1991 ) and rats ( Doyle et al . , 1993 ) , respectively . Supposedly through modulating palatability , opioids also control food intake ( Sanger & McCarthy , 1980 ) as well as instrumental food-seeking behavior ( Cleary et al . , 1996 ) . For example , opioid antagonists decrease the breakpoint in progressive ratio schedules of reinforcement with food ( Barbano et al . , 2009 ) , whereas opioid agonists produce the opposite effect ( Solinas & Goldberg , 2005 ) . This reflects the influence of orosensory information on the reinforcing effect of food . Consistent with our model , these influences have mainly been attributed to the effect of opiates on increasing extracellular dopamine levels in the Nucleus Accumbens ( NAc ) ( Devine et al . , 1993 ) through its action on μ-opioid receptors in the VTA and NAc ( Noel & Wise , 1993; Zhang & Kelley , 1997 ) . Such orosensory-based approximation of nutritional content , as discussed before , could have been obtained through evolutionary processes ( Breslin , 2013 ) , as well as through prior learning ( Beeler et al . , 2012; Swithers et al . , 2009 , 2010 ) . In the latter case , approximations based on orosensory or contextual cues can be updated so as to match the true nutritional value , resulting in a rational neural/behavioral response to food stimuli ( de Araujo et al . , 2008 ) . Above , we developed a normative theory for reward-seeking behaviors that lead to homeostatic stability . However , animals do not always follow rational behavioral patterns , notably as exemplified in eating disorders , drug addiction , and many other psychiatric diseases . Here we discuss one prominent example of such irrational behavior within the context of our theory . Binge eating is a disorder characterized by compulsive eating even when the person is not hungry . Among the many risk factors of developing binge eating , a prominent one is having easy access to hyperpalatable foods , commonly defined as those loaded with fat , sugar , or salt ( Rolls , 2007 ) . As an attempt to explain this risk factor , we discuss one of the points of vulnerability of our theory that can induce irrational choices and thus , pathological conditions . Over-seeking of hyperpalatable foods is suggested to be caused by motivational systems escaping homeostatic constraints , supposedly as a result of the inability of internal satiety signals in blocking the opioid-based stimulation of DA neurons ( M . Zhang & Kelley , 2000 ) . Stimulation of μ-opioid receptors in the NAc , for example , is demonstrated to preferentially increase the intake of high-fat food ( Glass et al . , 1996; Zhang & Kelley , 2000 ) , and hyperpalatable foods are shown to trigger potent release of DA into the NAc ( Nestler , 2001 ) . Moreover , stimulation of the brain reward circuitry ( Will et al . , 2006 ) , as well as DA receptor agonists ( Cornelius et al . , 2010 ) are shown to induce hedonic overeating long after energy requirements are met , suggesting the hyper-palatability factor to be drive-independent . Motivated by these neurobiological findings , one way to formulate the overriding of the homeostatic satiety signals by hyperpalatable foods is to assume that the drive-reduction reward for these outcomes is augmented by a drive-independent term , T ( T > 0 for palatable foods , and T = 0 for ‘normal’ foods ) : ( 11 ) r ( Ht , Kt ) =D ( Ht ) −D ( Ht+Kt ) +T In other words , even when the setpoint is reached and thus , the drive-reduction effect of food is zero or even negative , the term T overrides this signal and results in further motivation for eating ( See ‘Materials and methods’ for alternative formulations of Equation 11 ) . Simulating this hypothesis shows that when a deprived agent ( initial internal state = −50 ) is given access to normal food , the internal state converges to the setpoint ( Figure 10C ) . When hyperpalatable food with equal caloric content ( K is the same for both types of food ) is made available instead , the steady level of the internal state goes beyond the setpoint ( Figure 10C ) . Moreover , the total consumption of food is higher in the latter case ( Figure 8D ) , reflecting overeating . In fact , the inflated hedonic aspect of the hyperpalatable food causes it to be sought and consumed to a certain extent , even after metabolic demands are fulfilled . One might speculate that such persistent overshoot would result in excess energy storage , potentially leading to obesity . 10 . 7554/eLife . 04811 . 025Figure 10 . Simulating over-eating of hyperpalatable vs . normal food . ( A ) The simulated agent can consume normal ( T = 0 ) or hyperpalatable ( T > 0 ) food . The nutritional content , K , of both foods are equal . In the single-option task ( C , D ) , one group of animals can only choose between normal food and nothing ( nul ) , whereas the other group can choose between hyperpalatable food and nothing . Starting the task in a deprived state ( initial internal state=−50 ) , the internal state of the second , but not the first , group converges to a level above the setpoint ( C ) and the total consumption of food is higher in this group ( D ) . In the multiple-choice task , the agents can choose between normal food , hyperpalatable food , and nothing ( B ) . Results show that the hyperpalatable food is preferred over the normal food ( E ) and the internal state is defended at a level beyond the setpoint ( F ) . See Figure 10—source data 1 for simulation details . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 02510 . 7554/eLife . 04811 . 026Figure 10—source data 1 . Free parameters for the over-eating simulations . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 026 Simulating the model in another condition where the agent has ‘concurrent’ access to both types of foods shows significant preference of the hyperpalatable food over the normal food ( Figure 10E ) , and the internal state again converges to a higher-than-setpoint level ( Figure 10F ) . This is in agreement with the evidence showing that animals strongly prefer highly palatable to less palatable foods ( McCrory et al . , 2002 ) . ( See Figure 10—source data 1 for simulation details ) Our model is inspired by the drive reduction theory of motivation , initially proposed by Clark Hull ( Hull , 1943 ) , which became the dominant theory of motivation in psychology during the 1940s and 1950s . However , major criticisms have been leveled against this theory over the years ( McFarland , 1969; Savage , 2000; Berridge , 2004; Speakman et al . , 2011 ) . Here we propose that our formal theory alleviates some of major faults of the classical drive-reduction . Firstly , the classical drive-reduction does not explain anticipatory responding in which animals paradoxically voluntarily increase ( rather than decrease ) their drive deviation , even in the absence of any physiological deficit . As we demonstrated , such apparently maladaptive responses are optimal in terms of both reward-seeking and ensuring physiological stability , and are thus acquired by animals . Secondly , the drive reduction could not explain how secondary reinforcers ( e . g . , money , or a light that predicts food ) gain motivational value , since they do not reduce the drive per se . Because our framework integrates an RL module with the HR reward computation , the drive reduction-induced reward of primary reinforcers can be readily transferred through the learning process to secondary reinforcers that predict them ( i . e . , Pavlovian conditioning ) as well as to behavioral policies that lead to them ( i . e . , instrumental conditioning ) . Finally , the original Hull's theory is in contradiction with the fact that intravenous injection of food is not rewarding , despite its drive-reduction effect . As we showed , this could be due to the orosensory-based approximation mechanism required for computing the reward . Despite its limitations ( discussed later ) , we would suggest that our modern re-formulation of the drive-reduction theory subject to specific assumptions ( i . e . , orosensory approximation , connection to RL , drive form ) can serve as a framework to understand the interaction between internal states and motivated behaviors . Several previous RL-based models have also tried to incorporate the internal state into the computation of reward by proposing that reward increases as a linear function of deprivation level . That is , r=wr¯ , where r¯ is a constant and w is proportional to the deprivation level . Interestingly , a linear approximation of our proposed drive-reduction reward is equivalent to assuming that the rewarding value of outcomes is equal to the multiplication of the deprivation level and the magnitude of the outcome . In fact , by rewriting Equation 2 for the continuous case we will have: ( 12 ) r ( Ht , Kt ) ≡dD ( Ht+Kt ) dKt Using Taylor expansion , this reward can be approximated by: ( 13 ) r ( Ht , Kt ) ≅−Kt . ∇DH ( Ht ) +O ( ∇2DH ( Ht ) ) Where ∇ is the gradient operator , and ∇2 is the Laplace operator . Thus , a linear approximation of our proposed drive-reduction reward is equivalent to assuming that the rewarding value of outcomes is linearly proportional to their need-reduction capacity ( Kt ) , as well as a function ( the gradient of drive ) of the deprivation level . In this respect , our framework generalizes and provides a normative basis to multiplicative forms of deprivation-modulated reward ( e . g . , decision field theory ( Busemeyer et al . , 2002 ) , intrinsically motivated RL theory ( Singh et al . , 2010 ) , and MOTIVATOR theory ( Dranias et al . , 2008 ) ) , where reward increases as a linear function of deprivation level . Moreover , those previous models cannot account for the non-linearities arising from our model; that is the inhibitory effect of irrelevant drives and risk aversion . Whether the brain implements a nonlinear drive-reduction reward ( as in Equation 2 ) or a linear approximation of it ( as in Equation 13 ) can be examined experimentally . Assuming that an animal is in a slightly deprived state ( Figure 11A ) , a linear model predicts that as the magnitude of the outcome increases , its rewarding value will increase linearly ( Figure 11B ) . A non-linear reward , however , predicts an inverted U-shaped economic utility function ( Figure 11B ) . That is , the rewarding value of a large outcome can be negative , if it results in overshooting the setpoint . 10 . 7554/eLife . 04811 . 027Figure 11 . Behavioral predictions of the model . ( A ) Differential predictions of the multiplicative ( linear ) and drive-reduction ( non-linear ) forms of reward . In our model , assuming that the internal state is at ht ( A ) , outcomes larger than h−ht result in overshooting the setpoint and thus a declining trend of the rewarding value ( B ) . Previous models , however , predict the rewarding value to increase linearly as the outcome increases in magnitude . ( C ) Our model predicts that when given a choice between two options with equal net effects on the internal state , animals choose the option that first results in reducing the homeostatic deviation and then is followed by an increase in deviation ( green ) , as compared to a reversed-order option ( red ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 027 A more recent framework that also uses a multiplicative form of deprivation-modulated reward is the incentive salience theory ( Berridge , 2012; Zhang et al . , 2009 ) . However , in contrast to the previous models and our framework , this model assumes that the rewarding value of outcomes and conditioned stimuli is learned as if the animal is in a reference internal state ( ψ=1 ) . Let's denote this reward by r ( s , ψ=1 ) for state s . At the time of encountering state s in the future , the animal uses a factor , ψt , related to its current internal state , to modulate the real-time motivation of the animal: r ( s , ψt ) =ψt . r ( s , ψ=1 ) . In the case of conditioned tolerance to hypothermic agents , however , heat-producing response is motivated at the time of cue presentation , when the hypothermic agent is not administered yet . At this time , the animal's internal state is not deviated and thus , the motivational element , ψt , in the incentive salience theory does not provoke the tolerance response . Therefore , in our reading and unlike our framework , the incentive salience theory cannot give a computational account of anticipatory responding . Another approach to integrate responsiveness to both internal and external states appeals to approximate inference techniques from statistical physics . The free energy theory of brain ( Friston , 2010 ) proposes that organisms optimize their actions in order to minimize ‘surprise’ . Surprise is an information-theoretic notion measuring how inconceivable it is to the organism to find itself in a certain state . Assume that evolutionary pressure has compelled a species to occupy a restricted set of internal states , and p ( Ht ) indicates the probability of occupying state Ht , after the evolution of admissible states has converged to an equilibrium density . Surprise is defined as the negative log-probability of Ht occurring; −ln p ( Ht ) . We propose that our notion of drive is equivalent to surprise as utilized in the free energy ( Friston , 2010 ) and interoceptive inference ( Seth , 2013 ) frameworks . In fact , we propose that an organism has an equilibrium density , p ( . ) , with the following functional form: ( 14 ) p ( Ht ) ∝e−D ( Ht ) =e−∑i=1N\|hi−hi , t\|nm In order to stay faithful to this probability density ( and ensure the survival of genes by remaining within physiological bounds ) , the organism minimizes surprise , which is equal to −ln p ( Ht ) =∑i=1N\|hi−hi , t\|nm . This specific form of surprise is equivalent to our definition of drive ( Equation 1 ) . The equivalency of reward maximization and physiological stability objectives in our model ( Equation 5 ) shows that optimizing either homeostasis or sum of discounted rewards corresponds to prescribing a principle of least action applied to the surprise function . Although our homeostatic RL and the free-energy theory are similar in spirit , several major differences can be mentioned . Most importantly , the two frameworks should be understood at different levels of analysis ( Marr , 1982 ) : the free-energy theory is a computational framework , whereas our theory fits in the algorithmic/representational level . In the same line , the two theories use different mathematical tools as their optimization techniques . The free energy approach uses variational Bayes inference . Thus , rationality in that model is bounded by the simplifying assumptions for doing ‘approximate’ inference ( namely , factorization of the variational distribution over some partition of the latent variables , Laplace approximation , etc . ) . Our approach , however , depends on tools from optimal control theory and thus , rationality is constrained by the capabilities and weaknesses of the variants of the RL algorithm being used ( e . g . model-based vs . model-free RL ) . In this sense , while the notion of reward is redundant in the free energy formulation , and physiological stability is achieved through gradient descent function , homeostasis in our model can only be achieved through computing reward . In fact , the associative learning component in our model critically depends on receiving the approximated reward from the upstream regulatory component . As a result , our model remains faithful to and exploits the well-developed conditioning literature in behavioral psychology , with its strengths and weaknesses . A further approach toward adaptive homeostatic regulation is the predictive homeostasis ( otherwise known as allostasis ) model ( Sterling , 2012 ) where the classical negative-feedback homeostatic model is coupled with an inference system capable of anticipating forthcoming demands . In this framework , anticipated demands increase current homeostatic deviation ( by adjusting the setpoint level ) and thus , prepare the organism to meet the predicted need . Again , the concept of reward is redundant in this model and motivated behaviors are directly controlled by homeostatic deviation , rather than by a priori computed and reinforced rewarding values . As alternative to the homeostatic regulation theories phrased around maintenance of setpoints , another theoretical approach toward modeling regulatory systems is the ‘settling point’ theory ( Wirtshafter & Davis , 1977; Berridge , 2004; Müller et al . , 2010; Speakman et al . , 2011 ) . According to this theory , by viewing organisms as dynamical systems , what looks like a homeostatic setpoint is just the stable state of the system caused by a balance of different opposing effectors on the internal variables . However , one should notice that mathematically , such dynamical systems can be re-formulated as a homeostatically regulated system , by writing down a potential functional for the system ( or an energy function ) . Such an energy function is equivalent to our drive function whose setpoint corresponds to the settling point of the dynamical system formulation . Thus , there is equivalence between the two methods , and the setpoint approach summarizes the outcome of the underlying dynamical system on the regulated variables . Note that nothing precludes our framework to treat the setpoint conceptually as maintained internally by an underlying system of effectors and regulators . However , the setpoint/drive-function formulation conveniently allows us to derive our normative theory . Here we list the testable predictions of our theory , some of which put our model to test against alternative proposals . Firstly , as mentioned before ( Figure 9 ) , our theory predicts that the oral vs . fistula proportion in the water self-administration task ( McFarland , 1969 ) affects the speed of satiation: the higher the oral portion is , the faster the setpoint will be reached . Secondly , as discussed before , our model predicts an inverted U-shaped utility function ( Figure 11A , B ) . This is in contrast to the multiplicative formulations of deprivation-modulated reward . Thirdly , our model predicts that if animals are offered with two outcomes where one outcome reduces the homeostatic deviation and the other increases the deviation , the animal chooses to first take the deviation-reducing and then the deviation-increasing outcome ( Figure 11C , green sequence ) , but not the other way around ( Figure 11C , red sequence ) . This is due to the fact that future deviations ( and rewards ) are discounted . Thus , the animal tries to postpone further deviations and expedite drive-reducing outcomes . Fourthly , as explained earlier , we predict that animals are capable of learning not only Pavlovian , but also instrumental anticipatory responding . This is in contrast to the prediction of the predictive homeostasis theory ( Woods & Ramsay , 2007; Sterling , 2012; Stephen C ) . Finally , our theory predicts that upon reducing the magnitude of the outcome , a transitory burst of responding should be observed . We simulate both our model ( Figure 12 , left ) and classical homeostatic regulation models ( Figure 12 , right ) in an artificial environment where pressing a lever results in the agent receiving a big outcome ( 1 g ) during the first hour , and a significantly smaller outcome ( 0 . 125 g ) during the second hour of the experiment . According to the classical models , the corrective response ( lever-press ) is performed when the internal state drops below the setpoint . Thus , during the first hour , the agent responds with a stable rate ( Figure 12E , F ) in order maintain the internal state above the setpoint ( Figure 12D ) . Upon decreasing the dose , the agent waits until the internal state again drops below the setpoint . Thereafter , the agent presses the lever with a new rate , corresponding to the new dose . Therefore , according to this class of models , response rate switches from a stable low level to a stable high level , with no burst phase in between ( Figure 12F ) . 10 . 7554/eLife . 04811 . 028Figure 12 . Simulation results , predicting a transitory burst of responding upon reducing the dose of outcome . Our model ( left column ) and negative-feedback models ( right column ) are simulated is a process where responding yields big and small outcomes , during the first and second hours of the experiment , respectively . In our model , the objective is stay as close as possible to the setpoint ( A ) , whereas in previous homeostatic regulation models , the objective is to stay above the setpoint ( D ) . Thus , our model predicts a short-term burst of responding after the dose reduction , followed by regular and escalated response rate ( B , C ) . Classical HR models , however , predict an immediate transition from a steady low to a steady high response rate ( E , F ) . See Figure 12—figure supplements 1 and Figure 12—source data 1 for simulation details . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 02810 . 7554/eLife . 04811 . 029Figure 12—source data 1 . Free parameters for the within-session dose-change simulation . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 02910 . 7554/eLife . 04811 . 030Figure 12—figure supplement 1 . The Markov Decision Process used for the within-session dose-change simulation . DOI: http://dx . doi . org/10 . 7554/eLife . 04811 . 030 According to our model , however , when the unit dose decreases from 1 g to 0 . 125 g , the agent requires at least some new experiences with the outcome in order to realize that this change has happened ( i . e . , in order to update the expected outcome associated with every action ) . Thus , right after the dose is decreased , the agent still expects to receive a big outcome upon pressing the lever . Therefore , as the objective is to minimize deviation from the setpoint ( rather that staying above the setpoint ) , the agent waits for a period equal to the normal inter-infusion interval of the 1 g unit-dose . During this period , the internal state reaches the same lower bound as in previous trials ( Figure 12A ) . Afterward , when the agent presses the lever for the first time , it receives an unexpectedly small outcome , which is not sufficient for reaching the setpoint . Thus , several further responses will be needed to reach the setpoint , resulting in a burst of responding after decreasing the unit dose ( Figure 12B , C ) . After the setpoint is achieved , the agent presses the lever with a lower ( -than-burst ) rate , in order to keep the internal state close to the setpoint . In sum , in contrast to the classical HR models , our theory predicts a temporary burst of self-administration after dose reduction ( See Figure 12—source data 1 for simulation details ) . From an evolutionary perspective , physiological stability and thus survival may themselves be seen as means of guaranteeing reproduction . These intermediate objectives can be even violated in specific conditions and be replaced with parental sacrifice . Still , we believe that homeostatic maintenance can explain a significant proportion of motivated behaviors in animals . It is also noteworthy that our theory only applies to rewards that have a corresponding regulatory system . How to extend our theory to rewards without a corresponding homeostatic regulation system ( e . g . , social rewards , novelty-induced reward , etc . ) remains a key challenge for the future . In order to put forth our formal theory we had to put forward several key constraints and assumptions . As further future directions , one could relax several constraining assumptions of our formal setup of the theory . For example , redesigning the model in a partially observable condition ( as opposed to the fully-observable setup we used ) where the internal state observation is susceptible to noise could have important implications for understanding some psychiatric diseases and self-perception distortion disorders , such as anorexia nervosa . Also , relaxing the assumption that the setpoint is fixed and making it adaptive to the animal's experiences could explain tolerance ( as elevated perception of desired setpoint ) and thus , drug addiction and obesity . Furthermore , relaxing the restrictive functional form of the drive function and introducing more general forms could explain behavioral patterns that our model does not yet account for , like asymmetric risk-aversion toward gains vs . losses ( Kahneman & Tversky , 1979 ) . In a nutshell , our theory incorporates a formal physiological definition of primary rewards into a novel homeostatically regulated reinforcement learning theory , allowing us to prove that economically rational behaviors ensure physiological integrity . Being inspired by the classic drive-reduction theory of motivation , our mathematical treatment allows for quantitative results to be obtained , predictions that make the theory testable , and logical coherence . The theory , with its set of formal assumptions and proofs , does not purport to explain the full gamut of animal behavior , yet we believe it to be a credible step toward developing a coherent mathematical framework to understand behaviors that depend on motivations stemming from internal states and needs of the individual . Furthermore , this work puts forth a meta-hypothesis that a number of apparently irrational behaviors regain their rationality if the internal state of the individual is taken into account . Among others , the relationship between our learning-based theory and evolutionary processes that shape animal a priori preferences and influence behavioral patterns remains a key challenge . Here we show analytically that maximizing rewards and minimizing deviations from the setpoint are equivalent objective functions . For the especial case that m/n = 1 , Equation 11 can be rewritten as follows: ( S6 ) r ( Ht , Kt ) =D ( Ht ) −D ( Ht+Kt ) +T= ( Ht−H ) 2− ( Ht+Kt−H* ) 2+T= ( Ht− ( H+T2Kt ) ) 2− ( Ht+Kt− ( H+T2Kt ) ) 2 This means that the effect of T is equivalent to having a simple HRL system ( without term T ) whose drive function is shifted such that the new setpoint is equal to H+T2Kt , where H is the setpoint of the original system . This predicts that the bigger the hyper-palatability factor T is , the higher the new steady state is , and the higher the real nutritional content Kt of the food outcome is , the less divergence of the new setpoint from the original setpoint is . Equation 5 can also be re-written as: ( S7 ) r ( Ht , Kt ) =D ( Ht ) −D ( Ht+Kt ) +T= ( Ht−H* ) 2− ( Ht+Kt−H* ) 2+T= ( ( Ht−T2Kt ) −H* ) 2− ( ( Ht−T2Kt+Kt ) −H* ) 2 This can be interpreted as the effect of T being equivalent to a simple HRL system ( without term T ) whose internal state Ht is underestimated by T2Kt units . That is , hyper-palatability makes the behavior look like as if the subject is hungrier than what they really are .	Our survival depends on our ability to maintain internal states , such as body temperature and blood sugar levels , within narrowly defined ranges , despite being subject to constantly changing external forces . This process , which is known as homeostasis , requires humans and other animals to carry out specific behaviors—such as seeking out warmth or food—to compensate for changes in their environment . Animals must also learn to prevent the potential impact of changes that can be anticipated . A network that includes different regions of the brain allows animals to perform the behaviors that are needed to maintain homeostasis . However , this network is distinct from the network that supports the learning of new behaviors in general . These two systems must , therefore , interact so that animals can learn novel strategies to support their physiological stability , but it is not clear how animals do this . Keramati and Gutkin have now devised a mathematical model that explains the nature of this interaction , and that can account for many behaviors seen among animals , even those that might otherwise appear irrational . There are two assumptions at the heart of the model . First , it is assumed that animals are capable of guessing the impact of the outcome of their behaviors on their internal state . Second , it is assumed that animals find a behavior rewarding if they believe that the predicted impact of its outcome will reduce the difference between a particular internal state and its ideal value . For example , a form of behavior for a human might be going to the kitchen , and an outcome might be eating chocolate . Based on these two assumptions , the model shows that animals stabilize their internal state around its ideal value by simply learning to perform behaviors that lead to rewarding outcomes ( such as going into the kitchen and eating chocolate ) . Their theory also explains the physiological importance of a type of behavior known as ‘delay discounting’ . Animals displaying this form of behavior regard a positive outcome as less rewarding the longer they have to wait for it . The model proves mathematically that delay discounting is a logical way to optimize homeostasis . In addition to making a number of predictions that could be tested in experiments , Keramati and Gutkin argue that their model can account for the failure of homeostasis to limit food consumption whenever foods loaded with salt , sugar or fat are freely available .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Homeostatic reinforcement learning for integrating reward collection and physiological stability
Cytoplasmic dynein is a molecular motor responsible for minus-end-directed cargo transport along microtubules ( MTs ) . Dynein motility has previously been studied on surface-immobilized MTs in vitro , which constrains the motors to move in two dimensions . In this study , we explored dynein motility in three dimensions using an MT bridge assay . We found that dynein moves in a helical trajectory around the MT , demonstrating that it generates torque during cargo transport . Unlike other cytoskeletal motors that produce torque in a specific direction , dynein generates torque in either direction , resulting in bidirectional helical motility . Dynein has a net preference to move along a right-handed helical path , suggesting that the heads tend to bind to the closest tubulin binding site in the forward direction when taking sideways steps . This bidirectional helical motility may allow dynein to avoid roadblocks in dense cytoplasmic environments during cargo transport . Cytoskeletal motors transport a wide variety of intracellular cargos by processively moving along linear tracks . These motors do not always follow a linear trajectory . Rather , they generate torque perpendicular to their direction of motion , resulting in a helical movement relative to the filament . Such helical movement was first observed in a filament gliding assay in which surface-immobilized Tetrahymena axonemal dynein motors rotated MTs about their principal axes while translocating them ( Vale and Toyoshima , 1988 ) . Helical movement of cargoes has subsequently been demonstrated for several members of the myosin and kinesin superfamilies ( Nishizaka et al . , 1993; Nitzsche et al . , 2008; Yajima et al . , 2008; Bormuth et al . , 2012; Brunnbauer et al . , 2012 ) . Cytoplasmic dynein is the primary MT minus-end directed motor responsible for diverse cellular processes in cargo transport , nuclear positioning , and cell division ( Roberts et al . , 2013 ) . Despite its importance , key aspects of dynein's mechanism remain unclear , including whether or not the motor can produce torque . In contrast to kinesin-1 , which follows a single protofilament track ( Ray et al . , 1993; Yajima and Cross , 2005 ) , dynein takes frequent sideways steps ( Reck-Peterson et al . , 2006 ) . However , the full trajectory of dynein motors in three dimensions ( 3D ) remains unknown , because the motors are sterically prevented from moving around the circumference of surface-immobilized MTs . We next investigated the helical motility of cytoplasmic dynein . We used a tail-truncated yeast dynein artificially dimerized with glutathione S-transferase ( GST-Dyn331kD ) , which has similar motile properties to full-length dynein ( Reck-Peterson et al . , 2006 ) . The motors were fused to GFP at the N-terminus and attached to cargo beads coated with anti-GFP antibodies ( see ‘Materials and methods’ ) . At 1 mM ATP , dynein-coated beads moved in helical trajectories with a pitch of 591 ± 32 nm ( mean ± SEM , 67 rotations , 15 beads , Figure 2A–C ) . This rotation corresponds to a sideways movement to a neighboring protofilament ( 6 nm ) for every six tubulin dimers ( 48 nm ) in the forward direction . Unlike axonemal dynein and kinesin motors , which primarily rotate along their tracks in only one direction ( Vale and Toyoshima , 1988; Brunnbauer et al . , 2012 ) , beads coated with cytoplasmic dynein exhibited both left- and right-handed helical movement on MTs ( N = 67 rotations , Figure 2A–B , Figure 2—figure supplements 1 and 2 , Videos 2 , 3 ) . The speeds of left- and right-handed movements along the helical path ( 79 ± 4 . 7 nm/s and 89 ± 17 nm/s , mean ± SEM , respectively ) were statistically indistinguishable ( t-test , p = 0 . 28 ) . Bidirectional helical movement was also evident from traces of single beads , which occasionally switch direction during a run ( 4 out of 67 rotations , Figure 2D ) . In contrast , reversal of bead motility along the MT axis was never observed . 10 . 7554/eLife . 03205 . 006Figure 2 . Dynein moves in both left- and right-handed helical paths along MT bridges . ( A and B ) ( top ) Representative three-dimensional trace of a cargo bead-driven by GST-Dyn331kD motors shows left- ( A ) and right-handed ( B ) helical motion . ( bottom ) Two-dimensional projections of the traces shown at top . ( C ) Histogram of observed pitches per complete rotation . The average pitch is 591 ± 32 nm ( mean ± SEM ) . The average pitch of the left-handed movement ( 546 ± 42 nm , SEM , N = 32 ) was shorter ( t-test , p = 0 . 01 ) than that of the right-handed movement ( 749 ± 81 nm , SEM , N = 10 ) . ( D ) Change in handedness of rotation during the transport of a cargo bead . An example trace shows that a cargo bead initially moves along GMP-CPP MTs with a right-handed helical motion . At around t = 10 s , the bead reverses its helical motion for half of the period . At t = 20 s , the bead switches back to right-handed rotation and takes another half turn around the MT . Finally , at t = 25 s , the bead resumes left-handed helical motion until it disassociates from the MT . Arrows show the transitions from one type of helical motion to the other . DOI: http://dx . doi . org/10 . 7554/eLife . 03205 . 00610 . 7554/eLife . 03205 . 007Figure 2—figure supplement 1 . Movement of a GST-Dyn331kD coated bead along an MT bridge . ( A ) First 20 s of the 3D trace shown in Figure 2A is plotted in x , y , and z directions as a function of time . ( B ) The brightfield image of the bead shows the changes in the position and intensity of the bead as a function of time . DOI: http://dx . doi . org/10 . 7554/eLife . 03205 . 00710 . 7554/eLife . 03205 . 008Figure 2—figure supplement 2 . Additional example of the right-handed helical movement of a GST-Dyn331kD-coated bead along an MT bridge . ( top ) Representative three-dimensional trace of a cargo bead-driven by GST-Dyn331kD motors shows right-handed helical motion . ( bottom ) Two-dimensional projection of the trace shown at top . DOI: http://dx . doi . org/10 . 7554/eLife . 03205 . 00810 . 7554/eLife . 03205 . 009Figure 2—figure supplement 3 . Bidirectional helical motility of cargo beads driven by full-length dynein along MT bridges . ( A and B ) ( top ) Representative three-dimensional trace of a cargo bead driven by full-length dynein motors shows left- ( A ) and right-handed ( B ) helical motion . ( bottom ) Two-dimensional projections of the traces shown at top . The average pitch of the left-handed movement ( 576 ± 66 nm , SEM , N = 14 ) was longer ( t-test , p <0 . 01 ) than that of the right-handed movement ( 399 ± 94 nm , SEM , N = 19 ) . ( C ) Histogram of observed pitches per complete rotation . The average pitch is 500 ± 36 nm ( mean ± SEM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03205 . 00910 . 7554/eLife . 03205 . 010Figure 2—figure supplement 4 . Change in handedness of rotation during the transport of a cargo bead driven by full-length dynein motors . An example trace shows that a cargo bead initially moves along a GMP-CPP MT bridge with a right-handed helical motion . At around t = 5 s , the bead reverses its helical directionality ( arrow ) and moves in a left-handed helical path . DOI: http://dx . doi . org/10 . 7554/eLife . 03205 . 01010 . 7554/eLife . 03205 . 011Figure 2—figure supplement 5 . GST-Dyn331kD on taxol stabilized MTs . ( A ) Representative two-dimensional trace of cargo beads driven by GST-Dyn331kD on taxol stabilized MTs which contains the mixture of 12 , 13 , 14 protofilaments with different rotational pitches . ( B ) Histogram of observed pitches per complete rotation . The results are similar to the cargo beads driven on GMP-CPP MTs which have 14 protofilaments with ∼6400 nm rotational pitch . The average pitch is 607 ± 50 nm ( mean ± S . E . M . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03205 . 01110 . 7554/eLife . 03205 . 012Video 2 . Example recording of left-handed rotations of GST-Dyn331kD-coated bead at 100 ms time resolution via bright-field microscopy . DOI: http://dx . doi . org/10 . 7554/eLife . 03205 . 01210 . 7554/eLife . 03205 . 013Video 3 . Example recording of right-handed rotations of GST-Dyn331kD-coated bead at 100 ms time resolution via bright-field microscopy . DOI: http://dx . doi . org/10 . 7554/eLife . 03205 . 013 We next investigated the helical motility of a full-length yeast cytoplasmic dynein ( Reck-Peterson et al . , 2006 ) . The motors were fused to GFP at the N-terminus and attached to cargo beads coated with anti-GFP antibodies . Dynein-coated beads moved in helical trajectories with a pitch of 500 ± 36 nm ( mean ± SEM , Figure 2—figure supplement 3 , Video 4 ) and exhibited both left- and right-handed helical movement . The velocities of left- and right-handed movement along the helical path ( 42 ± 11 nm/s and 43 ± 10 nm/s , mean ± SEM , respectively ) were statistically indistinguishable ( t-test , p = 0 . 21 ) . Similar to GST-Dyn331kD , beads driven by full-length dynein occasionally switch direction during a run ( 8 out of 33 rotations , Figure 2—figure supplement 4 ) . Unlike GST-Dyn331kD , which has a net preference for left-handed helical movement ( 75% ) , full-length dynein has a net preference for right-handed helical movement ( 58% , t-test , p = 10−5 ) . This difference may be related to GST dimerization , in which the heads may be oriented differently relative to the MT surface . 10 . 7554/eLife . 03205 . 014Video 4 . Example recording of right- and left-handed rotations of full-length dynein-coated bead at 100 ms time resolution via bright-field microscopy . DOI: http://dx . doi . org/10 . 7554/eLife . 03205 . 014 The pitch of dynein-driven rotation is much shorter than the supertwist of the GMP-CPP MTs ( ∼6400 nm ) , suggesting that helical motility of dynein is independent from the helicity of the MT track . To verify this , we repeated the assay with GST-Dyn331kD on taxol-stabilized MTs , which contain a mixture of 12 ( 77% ) and 13 ( 11% ) and 14 ( 2% ) protofilaments ( Ray et al . , 1993 ) . Out of 34 rotations , 68% were left-handed and 32% were right-handed . On average , the pitch of helical movement was 607 ± 50 nm ( mean ±SEM ) ( Figure 2—figure supplement 5 ) , similar to that of GMP-CPP MTs ( t-test , p=0 . 77 ) . The results demonstrate that our findings are not an artifact of the MT polymerization method . To test the possibility whether bidirectional helical motility is driven by a rotational tug-of-war between dynein motors ( i . e . , some motors strictly rotate in a right-handed helix and the others rotate in the opposite direction ) , we tracked individual GST-Dyn331kD motors labeled with a quantum-dot on MT bridges ( Figure 3A ) . The average run length of single motors on MT bridges was 1 . 5 ± 0 . 2 µm ( mean ± SEM , N = 10 ) , which is similar to that measured on surface-immobilized MTs ( Reck-Peterson et al . , 2006 ) . The velocity of single dimers was 64 ± 8 nm/s ( mean ± SEM , N = 10 ) , similar ( t-test , p = 0 . 12 ) to that of cargo beads carried by multiple dimers ( 80 ± 10 nm/s , N = 15 ) . The traces of single motors show high variability along the perpendicular axis of MTs and switch directions in their sideways movement more frequently than the beads carried by multiple motors ( Figure 3B ) . These results exclude rotational tug-of-war . 10 . 7554/eLife . 03205 . 015Figure 3 . Single dynein motors frequently switch the direction of their sideways movement . ( A ) Schematic representation of quantum-dot labeled single dynein motors on the MT bridges ( not to scale ) . Expected amplitude of rotations is ∼50 nm . ( B ) Two example traces show 2D projection of dynein motors along the MT , using fluorescent tracking . MT filaments remain nearly straight between the bridges ( persistence length is 5 . 2 mm ) and oscillate due to the thermal fluctuation . The red trace represents the fluctuation of the MT bridge in the perpendicular axis , determined by the position of a quantum dot stably bound to a MT . The red trace was subtracted from the traces of quantum dots attached to single dynein motors ( blue trace ) . Single motors do not show signs of regular helical movement . DOI: http://dx . doi . org/10 . 7554/eLife . 03205 . 015 We next tested the possibility that the relative orientation of the two motor domains in a dimer can influence the handedness of helical motility . To eliminate the contribution of interhead orientation , we used a cargo bead coated with monomeric Dyn331kD , which is not processive on its own ( Reck-Peterson et al . , 2006 ) . We observed that multiple monomers were able to drive processive motility of cargo beads at 61 ± 6 nm/s ( mean ± SEM ) , consistent with the ability of multiple non-processive kinesins or myosins to drive processive motility ( Kamei et al . , 2005; Furuta et al . , 2013 ) . The trajectories of beads driven by dynein monomers ( Figure 4 , Video 5 ) also displayed a helical component . The average pitch ( 579 nm ± 38 nm , 57 rotations , 11 beads ) was similar to that of GST-Dyn331kD ( t-test , p = 0 . 78 ) and higher than full-length dynein ( t-test , p = 0 . 16 ) . The majority ( 59% ) of the rotations was right-handed , similar to full-length dimers ( N-1 two proportion test , p = 0 . 47 ) . The speeds of left- and right-handed movements along the helical path ( 66 ± 10 nm/s and 57 ± 6 nm/s , respectively ) were statistically indistinguishable ( t-test , p = 0 . 18 ) . The results indicate that right-handed preference of full-length dynein for helical movement is not due to the head–head orientation of the dimer . 10 . 7554/eLife . 03205 . 016Figure 4 . Dynein monomers prefer to move in a right-handed helix . ( A ) Representative three-dimensional trace of a cargo bead driven by monomers shows right-handed helical motion . ( B ) ( top ) Representative two-dimensional trace for monomeric Dyn331kD . ( bottom ) Histogram of the periods of rotations shows that the average pitch is 579 ± 38 nm ( mean ±SEM ) . The average pitch of the left-handed movement ( 658 ± 92 nm , SEM , N = 17 ) was longer ( t-test , p=0 . 05 ) than that of the right-handed movement ( 490 ± 40 nm , SEM , N = 24 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03205 . 01610 . 7554/eLife . 03205 . 017Video 5 . Example recording of right-handed rotations of monomeric Dyn331kD-coated bead . Video recorded via bright-field microscopy with time resolution of 100 ms . DOI: http://dx . doi . org/10 . 7554/eLife . 03205 . 017 In this study , we showed that cytoplasmic dynein moves along MTs in a helical trajectory with a pitch of ∼500 nm . Single dynein dimers take frequent sideways steps ( Reck-Peterson et al . , 2006 ) , whereas multiple dynein dimers persistently move in a helical path with a net preference for a right-handed rotation . What is the molecular basis of this preference ? Kinesin-1 monomers are believed to move to the closest possible site on the microtubule which stays to the left of the previous binding site ( Yajima and Cross , 2005 ) . However , kinesin-1 dimers only walk along a single protofilament because their short neck-linker prevents off-axis stepping ( Brunnbauer et al . , 2012 ) . In dynein , monomers may prefer to attach to the nearest tubulin binding site towards the MT minus end , favoring a rightward step ( Figure 5A ) . In the case of a dimer , the heads are attached to neighboring protofilaments and the leading head prefers to be on the right side at 30° relative to the trailing head ( DeWitt et al . , 2012; Qiu et al . , 2012 ) . This orientation would make a rightward step more favorable for the trailing head , resulting in a net rightward bias ( Figure 5B ) . 10 . 7554/eLife . 03205 . 018Figure 5 . A model for the helical movement of cytoplasmic dynein . ( A ) Top view of a monomeric dynein ( red oval ) stepping toward the MT-minus end ( arrows ) . The yellow circles represent the putative binding sites for the highlighted dyneins . The closest available binding sites are numbered from 1 to 3 . The nearest ( 8 nm ) binding site is along the same protofilament ( 1 ) . The binding site on the right ( 2 ) has a shorter distance ( 9 . 3 nm ) than the one in left ( 3 , 10 . 8 nm ) , resulting in a net preference to step rightward . ( B ) A dynein dimer prefers to orient on an MT with the leading head positioned on the right of the trailing head . When the trailing head ( bright red oval ) moves forward , it prefers to step rightwards to be positioned on the right hand side of its partner . ( C ) When multiple dimers carry a cargo bead , helical directionality may be affected by the number and orientation of the motors associated with an MT track . In this orientation , tubulin binding sites to the right for the motor in the middle ( bright red ovals ) may be obstructed for the motor in the lead . This results in a tendency to move in a left-handed helical pattern . ( D ) Due to the finite run length of dynein motors , MT-associated motors dissociate and new ones attach to the track . Changes in the orientation of MT-bound motors , switch the directionality of helical movement . DOI: http://dx . doi . org/10 . 7554/eLife . 03205 . 018 In contrast to other motors studied to date , dynein moves along both left- and right-handed helical paths . The molecular basis of the switches in helical directionality remains unclear . The irregular stepping pattern of individual motors rules out the possibility of a rotational tug-of-war . Instead , we propose that the helical pattern of bead movement may be determined by the conformations of dynein motors associated with the MT track . It is likely that dyneins that are bound in sufficiently close proximity to other dyneins experience steric exclusion effects . If a bead is carried by motors which are oriented such that one is bound immediately forward and to the right of the other , the tubulin binding sites ahead and to the right of the trailing head will be obstructed . Therefore , these motors may prefer to step to the left , and the entire cargo will eventually trace out a left-handed spiral ( Figure 5C ) . The number and orientation of the motors that are simultaneously in contact with the MT may change over the course of a recording . Individual dynein motors have 1400 nm run length ( Reck-Peterson et al . , 2006 ) , indicating that motors on the bead detach and reattach during a processive run of a cargo bead . These may alter the orientation of the MT bound motors , resulting in the reversal of helical directionality during a processive run ( Figure 5D ) , as occasionally observed . Inside cells , the MT surface is crowded with associated proteins , which act as roadblocks during the transport of intracellular cargos ( Stamer et al . , 2002 ) . Furthermore , the intracellular space is crowded with large structures , such as vesicles and organelles . The ability of molecular motors to produce torque and axial force may allow the motors to switch protofilaments and avoid these obstacles . In cells , the same MT track is used for both plus- and minus-end-directed transport . Sideways movement may prevent traffic jams on MTs . In vitro assays have shown that dynein can bypass roadblocks whereas kinesin-1 stalls when it encounters an obstacle and eventually releases ( Dixit et al . , 2008 ) . Bidirectional helical movement may provide additional flexibility to dynein to transport cargos in dense cellular environments . Saccharomyces cerevisiae strains expressing mutant forms of cytoplasmic dynein ( Dyn1 ) gene were generated by homologous recombination . Proteins were expressed and purified as described ( Reck-Peterson et al . , 2006 ) . MTs polymerized under 10 mM taxol contain a mixture of 12 , 13 , or 14 protofilaments . In MTs with 13 protofilaments , the protofilament long axes align with the MT long axis , whereas MTs with 12 and 14 protofilaments have a right-handed supertwist with 4000 nm pitch and left-handed supertwist with 6400 nm pitch , respectively ( Hyman et al . , 1995 ) . GMP-CPP MTs were grown for 3 hr at 37°C from a 50 µl BRB80 buffer ( 80 mM PIPES pH 6 . 8 with KOH , 2 mM MgCl2 , 1 mM EDTA ) solution supplemented with 5 µM tubulin ( 80% unlabeled porcine tubulin , 20% HiLyte 647 labeled porcine tubulin ) , 1 mM GMP-CPP and 2 mM MgCl2 . Assembled MTs were pelleted at 40 , 000×g with Beckman Ti 102 . 1 rotor and resuspended in 60 µl BRB80 buffer . The average length of HiLyte 647 labeled MTs was 15 µm . Carboxylated polystyrene beads were coated with anti-rabbit polyclonal GFP antibodies ( Covance , Emeryville CA ) . The beads were initially pelleted and resuspended in the activation buffer ( 100 mM MES , 100 mM NaCl , pH 6 . 0 ) . Carboxyl groups on the surface of the bead were functionalized with amine reactive groups via EDC and sulfo-NHS crosslinking for 30 min at room temperature . The beads were then washed with phosphate buffer saline ( PBS ) at pH 7 . 4 , and anti-GFP antibodies were added to the beads and reacted for 3 hr in room temperature . Excess antibodies were removed by centrifugation . The beads were resuspended in PBS along with 0 . 1% azide for storage purposes . eGFP was fused to the N-terminus of the SRS—dynein MTBD chimeric construct ( GFP-SRS85:82 ) . GFP was used for attachment to an anti-GFP antibody-coated bead , and the MTBD stably binds to a MT ( Gibbons et al . , 2005; Carter et al . , 2008 ) . GFP-SRS85:82 does not generate motility on its own ( Gibbons et al . , 2005; Carter et al . , 2008 ) . Saturating amount of GFP-SRS85:82 was incubated with anti-GFP-coated carboxyl beads ( 2 µm diameter ) on ice for 10 min in dynein assay buffer ( DLB; 80 mM HEPES pH 7 . 4 , 1 mM EGTA , 2 mM MgCl2 , and 10% glycerol ) containing 1 mg/ml casein . Excess SRS protein was removed by centrifugation at 15 , 000×g , and the beads were resuspended in DLB . SRS85:82-coated beads were nonspecifically adsorbed to the coverslip . After 10 min of incubation , unbound beads were removed by washing the chamber twice with 30 µl DLB + 1 mg/ml casein . Casein was used to pre-block nonspecific surface attachment of MTs and motor proteins . Next , fluorescently labeled MTs were flowed and incubated for 10 min . Free MTs were washed with 30 µl DLB + 1 mg/ml casein buffer . We observed less than 20 bridges in 1000 µm × 1000 µm area . The bead density was kept high ( four beads on average in 20 µm × 20 µm area ) , and MT concentration was kept low to ensure that each bridge was formed by a single MT . Motor-coated beads always moved unidirectional without changing the direction of motion along the MT long axis during processive motility , excluding the possibility of bridges containing multiple MTs pointing in the opposite directions . Finally , a solution containing dynein-coated 0 . 5-µm diameter beads in DLB buffer supplemented with 2 . 5 mM PCA ( protocatechuic acid ) and 50 nM PCD ( protocatechuate-3 , 4-dioxygenase ) oxygen scavenging system ( Aitken et al . , 2008 ) , 1 mg/ml casein and 1 mM ATP were flowed to the chamber and sides of the chamber were sealed with nail polish to prevent evaporation of the assay solution . The assays were performed with a custom-built optical trapping microscope equipped with Nikon TiE microscope body , Nikon 100× 1 . 49 NA plan apochromat objective and Ixon+ electron multiplied charge coupled device ( EM-CCD ) camera ( Andor , United Kingdom ) . HiLyte-labeled MTs were excited with 632 nm laser beam in epifluorescence mode , and the fluorescent signal was detected by the EM-CCD camera with an effective pixel size of 160 nm . The videos were recorded at 100 ms frame rate . The surface of the flow chamber was scanned to find MT bridges between the two 2-µm-diameter beads . We performed our bead tracking assays on the bridges , in which MTs are 10–15 µm long between the beads , appear steady by the resolution of a fluorescence microscope ( 250 nm ) at 10 Hz frame rate , and the entire MT fluorescence appears in focus . A 0 . 5-µm cargo bead freely diffusing in solution was trapped by a focused 1064 nm laser beam . The trap was steered with a pair of acousto-optical deflectors ( AA Opto-Electronic , France ) , and bead position was detected by a position sensitive detector using back–focal plane interferometry . Leakage of the intense trapping beam to fluorescence detection was blocked by a 708/75 nm bandpass filter . A single MT bridge was used over the course of one experiment . The dynein-coated beads were captured with an optical trap and brought to the proximity of the MT . When the motors on the cargo bead attached to the MT and started to move , the optical trap was turned off . The movement of beads was recorded using bright-field microscopy . MT polarity was first tested by placing the bead on a MT at the center of the bridge . Once the directionality is determined , the bead was moved away from the MT and placed at the plus-end tip of the bridge to explore the motility throughout the entire length of the bridge . All of the beads moved towards the same direction on a single bridge , without any reversals in axial direction . The movement of single quantum dots on MT bridges was determined by labeling GST-Dyn331kD motors at the C-terminus with a quantum dot 655 using HaloTag attachment ( DeWitt et al . , 2012 ) . Because the amplitude of side to side movements was smaller in the case of Q-dots ( ∼50 nm ) compared to beads ( ∼250 nm ) , the MT fluctuation was subtracted from Quantum-dot labelled dynein data . To measure the oscillation of the MT bridges , the MTs were sparsely labeled with a quantum dot 585 . The standard deviation of the position of 585 quantum dots was 35 nm in perpendicular direction and 17 nm in parallel direction to the long axis of the MTs . The position of 585 quantum dots was subtracted from the quantum dot 655 labeled dyneins to correct for MT oscillations . Cargo beads were tracked by using custom-written software in Matlab ( The Mathworks , Natick MA ) , which utilizes Gaussian fitting to determine the xy position of the bead . The Matlab code used in data analysis is available at http://research . physics . berkeley . edu/yildiz/SubPages/code_repository . html and in Supplementary file 1 . The precision of bead tracking was 3 nm in x and 5 nm in y directions . The z position of the bead was determined by the intensity of the bead center . To calibrate the bead intensity as a function of z position , the surface of a sample chamber was decorated with 0 . 5-µm diameter beads . The microscope objective was moved ±250 nm in the z direction with 25 nm increments using PIFOC objective scanner ( Physik Instrumente , Germany ) . Corresponding intensities of peaks at each frame were plotted with the z position relative to the microscope objective . This routine was repeated for 20 times to obtain a calibration curve shown in Figure 1D . The bead intensity profile was fitted with a third order polynomial function and the z position of the bead was calculated from the calibration curve . The bead image could not be well-fit by a Gaussian when the z position was between −25 nm and +25 nm . To avoid sample-to-sample variability in calibration procedures , the background was subtracted and the intensities were normalized in the bright-field images of the bead . Traces of bead motility were smoothed by the moving average filter of window of five data points . The pitches of the helical motion of the beads are corrected for the 6400 nm left-handed supertwist of GMP-CPP MTs . The corrected pitch is calculated by 1pitchmeasured=±1pitchMT+1pitchcorrected . The corrected pitch is longer than the measured pitch for left-handed rotations and shorter for right-handed rotations .	Cells rely on ‘molecular motors’ travelling along tracks called microtubules to move proteins and other cargoes between different parts of a cell . Dynein is a molecular motor that moves along the microtubules by taking “steps” towards the slowly growing end of these tracks . The trajectories of dynein motors have been studied extensively using techniques that can follow their movements in two dimensions . However , some molecular motors can also rotate as they travel , creating a twisting force called a torque that causes the motor to spiral around the microtubule in a helix . To assess the torque that dynein can generate and to better understand its movements in three dimensions , Can et al . used a length of microtubule to build a ‘bridge’ between two polystyrene beads . The dynein motors were made to carry a smaller polystyrene bead as cargo , and the movement of this smaller bead was tracked using a computer algorithm to interpret the motion recorded by a microscope . Can et al . found that dynein moves in a helical trajectory around the microtubule , rather than travelling along it in a straight line . As it travels it can twist in one direction or the other , generating torque in either direction . This is unlike other types of molecular motor , which produce torque in just one direction . However , dynein prefers to rotate to the right , suggesting that with every step along a microtubule , it binds to the closest available binding site in the forward direction . Why might it be useful for molecular motors to behave in this way ? Can et al . propose that the ability to rotate in both directions may allow dynein to avoid roadblocks or other obstacles in the dense and busy cellular environment in which it has to operate .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2014	Bidirectional helical motility of cytoplasmic dynein around microtubules
The West Nile Virus ( WNV ) envelope protein , E , promotes membrane fusion during viral cell entry by undergoing a low-pH triggered conformational reorganization . We have examined the mechanism of WNV fusion and sought evidence for potential intermediates during the conformational transition by following hemifusion of WNV virus-like particles ( VLPs ) in a single particle format . We have introduced specific mutations into E , to relate their influence on fusion kinetics to structural features of the protein . At the level of individual E subunits , trimer formation and membrane engagement of the threefold clustered fusion loops are rate-limiting . Hemifusion requires at least two adjacent trimers . Simulation of the kinetics indicates that availability of competent monomers within the contact zone between virus and target membrane makes trimerization a bottleneck in hemifusion . We discuss the implications of the model we have derived for mechanisms of membrane fusion in other contexts . Cell entry of enveloped viruses requires membrane fusion , catalyzed by a viral surface protein . The fusion protein of flaviviruses—the group that includes yellow fever ( YFV ) , West Nile ( WNV ) , dengue ( DV ) , and tick-borne encephalitis ( TBEV ) viruses—is the envelope protein ( E ) , which becomes fusogenic when exposed to reduced pH in an endosome ( Allison et al . , 1995; van der Schaar et al . , 2007 ) . This step merges viral and endosomal membranes and releases the viral genome into the cytosol . Although thermodynamically favorable , lipid-bilayer fusion has a very high kinetic barrier ( ∼50 kcal/mol ) ( Rand and Parsegian , 1984 ) . Modeling and experiment suggest that hydration-force repulsion occurs when two bilayers approach more closely than 15–20 Å . Fusion proteins such as E reduce this barrier by inducing the distortion needed to form a hemifusion stalk—the short-lived intermediate that resolves to form a pore . They do so through a series of membrane-coupled conformational rearrangements ( Harrison , 2008 ) . Mature flavivirus particles have an ordered icosahedral lattice of E dimers on the surface of a ∼500 Å diameter virion ( Figure 1A ) ( Zhang et al . , 2013a ) . Cleavage by furin of prM , the E-protein ‘chaperone’ , during transit through the trans-Golgi network converts an immature , non-infectious , and fusion incompetent particle , with trimer-clustered prM-E spikes , into an infectious and fusion competent particle , with dimer-clustered E ( Rey et al . , 1995; Stadler et al . , 1997; Li et al . , 2008 ) . This rearrangement primes the E protein to undergo a sequence of fusion-inducing conformational changes when exposed to acidic pH . 10 . 7554/eLife . 04389 . 003Figure 1 . Molecular structure and conformational rearrangements in flavivirus membrane fusion . ( A ) Left: domains of flavivirus E protein in the dimer on the surface of a mature virion or virus-like particle ( VLP ) , monomers shown in solid and outline forms . right: arrangement of dimers on a small ( 60-subunit ) VLP . ( B ) Schematic diagram of the low-pH induced , fusogenic transition in E , with postulated intermediates enclosed in brackets . The steps illustrated are: dissociation of the dimer and outward projection of each monomer , membrane engagement of the fusion loops ( asterisks ) , trimerization of membrane-interacting monomers , collapse of trimer into the postfusion conformation . ( C ) Schematic diagram of the single-particle fusion assay used in these studies . Left: DiD-labeled VLPs are captured on the supported bilayer at neutral pH through a surrogate receptor ( Fab ) anchored to Ni2+/NTA-headgroup lipids; center: low-pH induced fusion and DiD dequenching; right: idealized plots of DiD fluorescence within the diffraction limited zone of the captured VLP , in which lateral diffusion and loss of intensity follows an initial burst when dequenching reports merger of proximal leaflets of the two membranes; a pH-dependent fluorophore in the membrane reports the time of pH drop . DOI: http://dx . doi . org/10 . 7554/eLife . 04389 . 003 Flavivirus E proteins have a conserved , three-domain architecture , with a C-terminal connector ( ‘stem’ ) linking the third domain to the transmembrane anchor ( Rey et al . , 1995 ) . Domain I is a central β-barrel that positions the other two domains; domain II , two long , clustered extensions from domain I , bears the fusion loop at its distal tip; domain III is an immunoglobulin-like domain that may have a receptor-binding surface . In the prefusion , neutral-pH conformation , a cavity at the junction between domains I and III sequesters the fusion loop of the E dimer partner ( Figure 1B ) . Protonation of key histidine residues induces dimer dissociation , exposing the fusion loop and allowing the ectodomain to hinge outwards from the virion surface ( Fritz et al . , 2008 ) . Engagement of the fusion loop with the target membrane facilitates trimerization , accompanied by repositioning of domain III relative to domain I ( Allison et al . , 1995; Rey et al . , 1995; Modis et al . , 2004 ) . Rearrangement of the membrane-proximal stem , from a conformation buried in the outer leaflet of the viral membrane to one that stabilizes the inter-protomer trimer interface , can then ‘zip’ the trimer together and induce collapse of the extended intermediate , causing the transmembrane anchors to approach the trimer-clustered fusion loops ( Pangerl et al . , 2011; Klein et al . , 2013 ) . This last step favors formation of a hemifusion stalk between the apposed leaflets of the two membranes . Disruption of domain III repositioning by exogenous free domain III or inhibition of stem zippering by stem-derived peptides inhibits infection ( Liao and Kielian , 2005; Schmidt et al . , 2010 ) . Resolution of the hemifused state into an open pore may require further rearrangement of the fusion loops and the C-terminal transmembrane anchors . We have studied the kinetics of WNV membrane fusion , in a single-particle format developed previously to analyze influenza virus fusion ( Floyd et al . , 2008 ) ( Figure 1C ) . Kinetic data can define rate-limiting steps preceding an observed process , and coupled with site-directed mutation they can identify key intermediates ( Floyd et al . , 2010; Ivanovic et al . , 2013 ) . We have used this approach to probe the conformational transition of WNV E on the virion surface . Our data support a mechanism in which a pH-dependent dimer-monomer equilibrium creates , at the contact between virus and target membrane , a pool of monomers competent to cluster into target-membrane engaged trimers . Formation of at least two adjacent trimers allows progression to hemifusion . The corresponding two-stage structural picture is ( i ) that stochastic activation of E monomers ( by dimer dissociation and monomer outward extension ) leads to trimerization and essentially irreversible target-membrane engagement , whenever three neighboring monomers are active , and ( ii ) that two adjacent trimers can together exert enough force on the membranes they bridge ( target membrane and viral membrane ) to produce a hemifusion stalk . Formation of adjacent trimers is limited by the availability of competent monomers . This mechanism—like the one shown previously to fit influenza-virus fusion kinetics—does not require defined trimer–trimer interactions , because resistance of the two membranes to deformation toward a hemifusion stalk couples conformational changes in the trimers that bridge them ( Ivanovic et al . , 2013 ) . Fusion proceeds rapidly whenever a sufficient number of them can overcome the deformation barrier . We suggest that this description may apply more generally to fusion of intracellular vesicles and to fusion of two cells . Recombinant expression of flavivirus proteins prM and E yields mature , non-infectious , empty virus-like particles ( VLPs ) . The properties of these particles are essentially the same as those of virions ( Figure 2A–C ) ( Schalich et al . , 1996 ) . WNV VLPs were prepared as described in Methods . Electron microscopy shows particles in two principal size classes , corresponding to the 60- and 180-E subunit shells described for TBE VLPs ( Figure 2C ) ( Allison et al . , 2003 ) . We measured a bulk liposome hemifusion pH threshold of ∼6 . 4 for WNV VLPs , as previously reported ( Moesker et al . , 2010 ) ( Figure 2A ) . 10 . 7554/eLife . 04389 . 004Figure 2 . Characterization of VLPs and single-particle fusion . ( A ) Bulk liposome fusion activity of purified WNV VLPs as function of pH , measured by DiD dequenching upon fusion with liposomes . ( B ) Discontinuous Optiprep gradient of WNV VLPs . Immunoblot at right ( left lane , 25/30% fraction; right lane , 30/35% fraction ) , showing >95% cleavage of prM to M . ( C ) Negative-stain ( left ) and cryo- ( right ) electron microscopy of WNV VLPs , with histogram of size distribution of cryo-electron micrograph particles . The smaller size corresponds to 60-subunit VLPs; the larger , to 180-subunit , virion-size VLPs . ( D ) Fluorescence recovery after photobleaching for a region of supported bilayer , showing recovery with a diffusion constant D = 0 . 59 μm2/s , consistent with a fluid bilayer , calculated from D = 0 . 22 R2/t ( Takamori et al . , 2006 ) , where R is radius of photobleached area . ( E ) Representative single-particle traces for fluorescein bleaching upon pH decrease ( green ) and DiD dequenching and diffusion ( red ) . The bar shows the lag time ( or ‘dwell time’ , td ) between pH drop and time of half-maximal dequenching . ( F ) The peak intensity of the hemifused particles does not correlate with the hemifusion lag time . DOI: http://dx . doi . org/10 . 7554/eLife . 04389 . 004 We collected single-particle data for WNV VLP hemifusion by total internal reflection fluorescence ( TIRF ) microscopy to obtain single-particle histograms of hemifusion delay times ( the interval between fluorescein signal decrease and DiD dequenching ) for VLPs ( Video 1 ) . The VLP or virus preparation was introduced at neutral pH , and the pH then lowered to induce fusion . To uncouple attachment from fusion-loop exposure , we incorporated either the E16 Fab or the lectin domain of DC-SIGN-R as a receptor ( Davis et al . , 2006 ) , with a C-terminal His6 tag that bound a Ni-NTA headgroup lipid in the fluid ( Figure 2D ) , glass-supported , lipid bilayer ( Figure 1C ) . In these experiments , loss of fluorescence from a fluorescein-conjugated lipid in the bilayer reported the pH drop , and dequenching of a hydrophobic fluorophore ( DiD ) introduced into the particle membrane reported the time of membrane merger and dye diffusion into the bilayer ( Figure 2E ) . 10 . 7554/eLife . 04389 . 005Video 1 . Video of WNV VLP hemifusion events at pH 6 . 25 , recorded at 640 nm channel sped up 20 times actual speed . DOI: http://dx . doi . org/10 . 7554/eLife . 04389 . 005 The extent of labeling did not affect the course of VLP hemifusion , as shown by the lack of correlation between peak intensity and hemifusion time ( Figure 2F ) . We collected WNV VLP single particle data over a pH range from 5 . 0 to 6 . 25 ( Figure 3A ) . We also collected data for live-virus Kunjin , a variant of West Nile , and found similar kinetics over the pH range from 5 . 0 to 6 . 0 ( Figure 3B ) . Labeling of the Kunjin virus with DiD did not influence infectivity ( Figure 3C ) . The absence of correlation between intensity of DiD fluorescence and VLP hemifusion times and the agreement between data for WNV VLPs and intact Kunjin virus show that hemifusion kinetics do not depend on particle size ( i . e . , on whether they are 60- or 180-subunit particles ) ( Figures 2F and 3B ) . 10 . 7554/eLife . 04389 . 006Figure 3 . Single-particle WNV VLP and Kunjin virus hemifusion . ( A ) Histograms of single-particle dwell times for WNV VLPs , pH 5 . 0 to 6 . 25 , with fitted curves calculated for a process with a single exponential decay ( pH 5 . 0–5 . 5 ) or a gamma distribution with two sequential or parallel steps ( N = 2 , pH 5 . 75–6 . 25 ) ( fits calculated over the first 150 s of the data ) . The number of events detected and the mean dwell time ( tmean ) are noted ( mean dwell time determined from all data ) . ( B ) Histograms of single-particle dwell times of Kunjin virus particles , with fits to a single exponential decay ( pH 5 . 0–5 . 5 ) or a gamma distribution with two sequential or parallel steps ( N = 2 , pH 5 . 75–6 . 25 ) . ( C ) Comparison of infectivity of carrier ( DMSO ) and DiD-treated Kunjin virus . DOI: http://dx . doi . org/10 . 7554/eLife . 04389 . 006 We compared bulk hemifusion rates over a range of temperatures , 20–36°C , at pH 6 . 25 , just below the pH threshold , and found an Arrhenius dependence for the time to hemifusion ( Figure 4A ) . The absence of abrupt changes in rate allows us to conclude that the protein conformational changes relevant for inducing hemifusion are largely the same across this temperature interval and that there are no detectable effects of lipid phase transitions . We therefore chose to work at 20°C for single-particle data collection , to slow the reaction and thereby facilitate detection of transient intermediates . 10 . 7554/eLife . 04389 . 007Figure 4 . Data for temperature dependence , receptor dependence , and sensitivity to fusion-loop mutation for fusion of WNV VLPs . ( A ) Kinetics of bulk hemifusion for WNV VLPs . Left: time to hemifusion as a function of temperature between 20°C and 36°C; right: Arrhenius plot and calculated activation energy . ( B ) Overall bulk fusion activity for WNV and Dengue serotype 4 ( DV4 ) VLPs produced at 28°C and 37°C . ( C ) Single-particle kinetics of WNV VLP fusion with DCSIGN-R as receptor at pH 5 . 5; compare with Figure 3A , upper right-hand panel . ( D ) Single-particle fusion kinetics ( left ) and fusion yield ( right ) for WNV VLPs containing a 1:2 mixture of mutant ( W101A ) and wild-type E; compare with Figure 7D . DOI: http://dx . doi . org/10 . 7554/eLife . 04389 . 007 Production temperature can affect the stability of Dengue E packing on the virion surface and alter epitope accessibility of E in WNV ( Zhang et al . , 2013b ) . We tested WNV VLPs produced at 37°C and 28°C and found no difference in the extent of prM processing or in the bulk liposome fusion activity . Dengue virus serotype 4 VLPs were significantly less active when produced at 37°C ( Figure 4B ) , consistent with previous observations ( Zhang et al . , 2013b ) . The mean delay time in the WNV VLP single particle distributions decreased with increasing proton concentration ( Figure 3A ) . We could fit the data at lower pH ( pH 5 . 0–5 . 5 ) with a probability density function describing a single exponential decay and the data at higher pH ( pH 5 . 75–6 . 25 ) with a function describing a process with two steps of equal rate , either parallel or sequential ( see ‘Materials and methods’ ) . These qualitative conclusions were independent of the surrogate receptor , as data collected with DC-SIGN-R had only slightly shorter delay times and slightly higher hemifusion rates ( Figure 4C ) . We observed from single-particle data for Kunjin-virus fusion between pH 5 . 0 and 6 . 0 mean delay times similar to those we measured for WNV VLPs , with exponential fits to a formal rate constant for the lower pH range and two-step fits at pH 5 . 75 and above ( Figure 3B ) . WNV VLPs that incorporated a fusion-loop mutant ( W101A ) did not dequench upon pH drop and instead released and washed away . This result is consistent with a requirement for fusion-loop insertion into the supported bilayer , as assumed in models for the reaction . It also shows that attachment and fusion-loop interaction are distinct steps in these experiments . In bulk fusion studies , in which acidification induces fusion with liposomes that do not bear a receptor or surrogate , exposure of the fusion loops may be necessary even for initial interaction of the virus or VLPs with the target bilayer . Mixed VLPs , with a 50:50 mixture of wild-type:W101A E protein , had a mean dwell time ( 52–53 s at pH 5 . 5 ) and an overall yield of fusion events ( ∼30% ) similar to those of fully wild-type particles ( Figure 4D ) . The first step in assembly of trimeric fusion complexes is release of E protein ectodomains from lateral contacts , especially those between dimers , in the surface of a mature virion . Some epitopes inaccessible in the pre-fusion surface lattice as visualized by cryoEM ( Mukhopadhyay et al . , 2003 ) bind antibodies even at neutral pH , showing that E subunits transiently expose buried surfaces ( Dowd et al . , 2011 ) . The temperature dependence for expansion of dengue virions suggests that thermal fluctuations can potentiate access to such cryptic epitopes ( Zhang et al . , 2013b; Zhang et al . , 2014 ) . The antibody-captured and temperature-induced states are probably on-pathway intermediates to subsequent hemifusion and pore formation . We measured by dynamic light scattering the hydrodynamic radius , R , of Kunjin virus as a function of pH , to detect initial low pH-induced rearrangement of E proteins on the virion surface ( Figure 5A ) . The uniform size of the virus particles ( as opposed to VLPs ) allowed us to interpret an increased R as projection of monomeric E subunits away from the particle surface , exposing the fusion loops . R increased reversibly as we lowered the pH , with a transition midpoint at pH 6 . 8 ( Figure 5B ) The value of R at high pH is larger than the outer radius of the fully ordered , compact cryoEM structure , suggesting that at 20°C and pH 8 there is enough outward ‘breathing’ to influence hydrodynamic drag , but some of the effect could be due to residual polydispersity . The Hill coefficient of the transition is ∼3 . 0 , indicating local cooperativity at the level of an E dimer and probably its immediate neighbors , but not an all-or-none transition over the entire particle surface . 10 . 7554/eLife . 04389 . 008Figure 5 . An initial reversible pH-dependent transition . ( A ) Reversible increase in hydrodynamic radius of Kunjin virus upon treatment with a low pH buffer ( pH 5 . 25 ) . The back-neutralized ( to pH 7 . 8 ) sample shows the same initial radius . ( B ) Hydrodynamic radius measurements of Kunjin virus over a pH range 5 . 25–7 . 8 . Half maximal extension observed at pH 6 . 8 . The transition has a Hill coefficient of ∼3 . 0 . ( C ) pH half-maximal point of soluble , West Nile virus , E-protein trimerization as measured by liposome co-floatation . DOI: http://dx . doi . org/10 . 7554/eLife . 04389 . 008 Trimerization of soluble flavivirus E ectodomains ( sE ) at low pH generally requires the presence of liposomes , probably to accelerate subunit association when the fusion loops of monomeric sE insert into the lipid bilayer . Trimer formation under these conditions is irreversible . Trimerization of TBEV E on the virion surface is likewise irreversible , with an overall pH threshold of 6 . 5 ( Allison et al . , 1995 ) . We measured irreversible trimerization of WNV sE ( which is monomeric even at neutral pH ) , by following co-floatation with liposomes , and found a threshold pH of 6 . 1 ( Figure 5C ) . Formal kinetic fits do not contain information about the underlying molecular mechanisms . We sought to relate the steps mediating hemifusion to molecular interactions between E-protein subunits on the VLP by carrying out a series of stochastic simulations for the activity of an array of E-proteins in contact with the supported membrane ( Figure 6A ) . The design of the simulation came from our understanding of structural features of flavivirus virions , physical properties of the E protein , and direct measurements for different pH dependent transitions . 10 . 7554/eLife . 04389 . 009Figure 6 . Simulation of time course to hemifusion . ( A ) Schematic for simulation of a flavivirus contact patch . E monomers are arranged in a hexagonal lattice ( an idealization of the more complex arrangement on the virion surface ) . Each monomer can be in one of four different states: inactive monomer , active monomer , trimer member ( with the condition that two adjacent monomers are active; if the condition holds , all three become trimer members ) , hemifusion mediator ( with the conditions that it completes a trimer adjacent to another trimer ) . Dimers are defined explicitly as adjacent monomers in the lattice . The dimer-monomer transition is reversible; the trimerization and hemifusion steps are irreversible . The forward rate constant , kact , reflects the dimer to monomer transition , while the reverse rate constant , kret , reflects the monomer to dimer transition . The pH-dependent equilibrium constant is Kdm = ( kact/kret ) [H+] . The assumption of reversibility in the dimer-monomer transition is based on dynamic light scattering observations ( Figure 5A ) . Cooperativity of dimer activation is described by a factor , Pdim , by which the probability for dimer-to-monomer transition is multiplied . The rate ktri is the rate constant for completing one trimer . The rate constant kcomp reflects the transition to hemifusion upon completion of the defined number of adjacent trimers . ( B ) Flow diagram of simulation , with final values optimized to the mean dwell time and fit over the experimental range of pH indicated in the figure . Free parameters shown in bold and red; fixed parameters , in gray . ( C ) Histograms for simulations of virions with kact , ktri and kcomp set to values shown in panel B , over the pH range 5 . 0–6 . 5 . The curves for each histogram are a single exponential for pH 5 . 0–5 . 5 and a gamma distribution with N = 2 for pH 5 . 75–6 . 25 . DOI: http://dx . doi . org/10 . 7554/eLife . 04389 . 009 The contact patch in these simulations is a hexagonal array of 30 E monomers within a circular zone on the VLP ( or virion ) surface . The geometry is an idealization of the actual surface lattice , with explicitly defined dimers that reorganize during the dimer-to-trimer transition . We chose 30 monomers for the contact patch , because the small VLPs have about 60 subunits ( depending on how perfect their surface lattice ) , and 30 subunits , a hemisphere , is the largest possible contact without major distortion of the target bilayer . The contact is unlikely to be much smaller , because particles with half of their E proteins containing ‘weakened’ fusion loops ( the W101A mutation ) had fusion properties similar to those of wild-type particles . A single number for the size of the contact patch is a reasonable approximation for the somewhat heterogeneous population of VLPs , because observed hemifusion kinetics did not depend on the extent of labeling with DiD and hence did not depend on the area of particle membrane or the diameter of the particle ( Figure 2F ) . We initialize the simulation by distributing the monomers between a ‘prefusion’ state and an ‘activated’ state . The former we take to be dimer-like; the latter we take to represent a conformation in which the monomer extends outward so that its fusion loops can contact the target membrane . The reversible transition between these states corresponds to the dimer-monomer equilibrium of soluble flavivirus E ectodomain ( sE ) and to the reversible E extension detected in our dynamic light scattering observations . It has a pH-dependent equilibrium constant , Kdm , and a forward rate constant , kact ( and hence a probability Pact = [1−exp ( −kactΔt ) ] for each time step , Δt , in the simulation ) . Thus , Kdm = ( kact/kret ) [H+] , where kret reflects the reverse rate constant ( for the monomer to dimeric transition ) . We set the pK for Kdm to agree with our measurements of the pH at half-maximal increase in the hydrodynamic radius ( pH 6 . 8 for the ∼30 Å expansion of Kunjin virus ) . We describe cooperativity in the activation transition with a factor ( Pdim ) , by which the activation of one protomer increases the activation rate for its neighboring dimer partner . That is , for a given neighbor , its probability for activation , Pact ( neighbor ) = Pdim[1−exp ( −kactΔt ) ] , if ( and only if ) the partner is already activated ( Figure 6A ) . The extended monomers that bridge to the target membrane can then cluster as trimers . We define explicitly all possible trimer combinations for the 30 protomer hexagonal lattice . We assume that in this second transition , a trimer forms with probability Ptri = [1−exp ( −ktriΔt ) ] , whenever three adjacent monomers ( a triangle of positions in the idealized lattice of the simulations—Figure 6A ) become activated; this step includes essentially irreversible capture of the fusion loops by the supported bilayer , as observed experimentally ( Stiasny and Heinz , 2004 ) . It corresponds to the observed irreversible trimerization and membrane capture of E ectodomains when the protein is exposed to reduced pH in the presence of liposomes . We used the measured pH for half-maximal trimerization ( 6 . 1 ) , as determined by liposome co-floatation of recombinant soluble WNV E ( Figure 5C ) , to set the pH dependence for the trimerization step . Finally , when T adjacent trimers have formed , the simulation allows hemifusion to proceed to completion with a probability Pcomp = [1−exp ( −kcompΔt ) ] . A simulation for a given virion exits on execution of this step , after recording the total number of steps ( i . e . , the time to hemifusion ) ( Figure 6B ) . We ran hemifusion simulations for 500 particles over a range of pH values . We found a unique set of parameters ( Figure 6B ) that generated histograms very similar in shape and in pH-dependence of dwell time to our experimental data from WNV VLPs ( Figure 5C ) . The three values we varied were kact , ktri and kcomp . We converged on these parameters following sequential rounds of optimization , testing fits for conditions in which one , two or three trimers defined hemifusion . We also tested a range of values for Pdim from 2–50 , thus altering the cooperativity of subunit activation , and values for T from 1 to 3 . We found that the data fit best to models in which two trimers mediate hemifusion and that this number was insensitive to the assumed pKa for monomer activation and to the assumed threshold pH for trimer formation . We investigated if accounting for cooperativity between a subunit and its neighbors in the surface lattice other than its dimer partner would alter the outcome of our simulations . Incorporation of an additional factor did not make any changes in the properties of the simulation histograms and their fits . For all subsequent simulations , we fixed the parameter Pdim = 4 , well within the tested range . With these parameters , the rate-limiting step is trimer formation . We noticed that as we increased the pH in these simulations , we lowered the yield of total particles reaching hemifusion during the fixed time frame of the simulation . We found the same trend in the experimental data ( Figure 7B ) . In the simulations , the lower yield came from a reduced pool of activated monomer available to form trimers and a resulting geometric penalty for trimer possibilities . Relaxing this constraint , by increasing the target patch size to 37 monomers , increased overall event yield in the simulation; decreasing the target patch size to 23 decreased the yield ( Figure 7C ) . 10 . 7554/eLife . 04389 . 010Figure 7 . Simulations varying rate constants and contact-patch size and including mixture of wild-type and mutant E . ( A ) Overall hemifusion rate at pH 5 . 5 from simulations with all parameters fixed at values shown in Figure 6B except for ktri or kcomp ( left and right panels , respectively ) . ( B ) Yield of single particle hemifusion events ( total number of fusion events/total number of identified particles in the field ) as measured at different pH ( left ) and simulated with a 31-monomer contact patch ( right ) , with Kdm set at 6 . 8 and ktri set with a pH 6 . 1 half-maximal transition point . ( C ) Yield in simulations with 23-monomer and 37-monomer contact patches . ( D ) Simulation of time course and yield for mixed particles in which one-third of the E monomers could not stably engage the membrane but could be part of a trimer . One third was chosen for the proportion of dead subunits in simulation , because it is expected that the proportion of the mixed particles in the experiment with predominantly W101A subunits in their contact area would flow away upon pH drop and not be recorded . We assumed that trimers with one inactive monomer could participate in induction of a hemifusion stalk but that trimers with two or three inactive monomers could not . Compare with Figure 4D . DOI: http://dx . doi . org/10 . 7554/eLife . 04389 . 010 We generated a series of mutant WNV VLPs , with mutations chosen to target different conformational states , in order to relate the sequential conformational transitions in E to changes in rate constants derived from simulation ( Table 1 ) . Mutation of key histidine residues identified to affect fusion eliminated all activity , while many mutations targeting the dimeric prefusion state had little effect on hemifusion kinetics or yield , consistent with previous observations ( Fritz et al . , 2008 ) . 10 . 7554/eLife . 04389 . 014Table 1 . Summary of mutant kinetic fits and ratesDOI: http://dx . doi . org/10 . 7554/eLife . 04389 . 014tmean ( s ) fitk ( s ) WT46 . 1single exponential2 . 9 ± 0 . 4 × 10−21:1 WT W101A Mix53 . 4single exponential2 . 2 ± 0 . 4 × 10−2F408V95 . 8gamma N = 22 . 0 ± 0 . 2 × 10−2N193A85 . 3single exponential1 . 2 ± 0 . 2 × 10−2V434W65 . 5gamma N = 23 . 9 ± 0 . 3 × 10−2F450W54 . 2single exponential1 . 2 ± 0 . 2 × 10−1 Residue F408 is part of a conserved contact that bridges the N-proximal segment of the ‘zipped’ stem of one subunit with domain II of the adjacent subunit in the postfusion trimer ( Klein et al . , 2013 ) . The interaction contributes to robust trimer formation in vitro of secreted , soluble TBEV E ectodomain ( Allison et al . , 1999 ) . We measured the bulk fusion activity for F408V and found no variation in pH threshold , ruling out any global effect of the mutations on proton binding and restricting potential variation in the simulation parameters corresponding to ktri and kcomp ( Figure 8A ) . An F408V mutation in WNV VLPs slowed hemifusion at pH 5 . 5 , broadening and extending the distribution of hemifusion delay times from a roughly single-exponential fall-off for wild-type at that pH ( Figure 3 , top right panel ) to a rise-and-decay approximated reasonably by a two-step gamma distribution ( Figure 8A ) . Under conditions expected to speed up the overall reaction ( pH 5 . 0 , 22°C ) , the F408V mutation reduced the single exponential rate for the distribution of dwell times from that of WT ( Figure 8B ) . Simulations in which the trimerization rate constant ( ktri ) for the F408V mutant was set to one-tenth its best value for the wild-type simulation generated a good fit to the observed hemifusion times , when all other parameters were held constant ( Figure 8C ) . Simulations altering the kcomp value generated poorer fits . The affected step in the simulation ( trimerization ) is consistent with the known contacts of F408 in the trimeric intermediate and with the known trimerization properties of E when those interactions are absent ( Pangerl et al . , 2011; Klein et al . , 2013 ) . 10 . 7554/eLife . 04389 . 011Figure 8 . Effects of mutations in E on VLP hemifusion . ( A ) Schematic of conformational state targeted by F408V mutation ( residue mutated indicated by star ) , bulk hemifusion titration and the single particle data histogram; compare with Figure 3A , upper right-hand panel . ( B ) Single particle data for wild type and F408V WNV VLPs under conditions to accelerate the hemifusion reaction ( pH 5 . 0 , 22°C ) ( C ) Simulation of WNV fusion with ktri parameter reduced to 0 . 1 s−1 . ( D ) Schematic of N193A conformational rearrangement ( residue mutated indicated by star ) , bulk hemifusion data and single particle data; compare with Figure 3A , upper right-hand panel . ( E ) Single particle data for wild type and N193A under conditions to further slow the reaction ( pH 6 . 25 , 18°C ) . ( F ) Simulation of WNV fusion with ktri parameter reduced to 0 . 25 s−1 . DOI: http://dx . doi . org/10 . 7554/eLife . 04389 . 011 Residue N193 , in the hinge region between domains I and II , contributes in the prefusion conformation to the ‘bottom’ of a cavity occupied by n-octyl-β-D-glucopyranoside ( octyl glucoside ) in a crystal structure of E from dengue virus serotype 2 . In the post-fusion conformation , N193 makes a ring of hydrogen bonds in the center of the domain II trimer contact , where it probably contributes to clustering of the fusion loops . WNV N193A VLPs showed no change in bulk fusion threshold . Yet , single particle data of the WNV N193A mutant broadened the distribution of hemifusion delay times for the VLPs at pH 5 . 5 , although somewhat less markedly than F408V ( Figure 8D ) . Under conditions in which the reaction slows ( pH 6 . 25 , 18°C ) , the N193A data further broaden to a distribution of hemifusion times that fits a process with two steps of equal rate ( Figure 8E ) . Reduction of ktri , from its best fit to the wild-type distribution , produced only an approximate fit to this distribution ( Figure 8F ) . Thus , the properties of the mutant are consistent with lower stability of the trimer relative to monomer and hence with an effect on late-stage trimerization of E , but it might also affect any other stage during which domain II rotation occurs . Residues V434 and F450 , both in the ‘stem’ , are near the N-terminus of a membrane-embedded helix ( ‘helix 2’ in older terminology; ‘α3’ in the more recent assignment from fitting a high-resolution cryoEM map ) and at the junction between stem and transmembrane anchor , respectively ( Figure 9A ) . We collected single particle data for the V434W and F450W mutants , both of which showed no change in their bulk fusion threshold ( Figure 9B–C ) . A single exponential best fit data from the F450W mutant , with a rate constant , k = 1 . 2 ± 0 . 2 × 10−1 . The F450W single particle data also had an extended tail in the dwell time distribution . The model from cryoEM reconstruction shows that F450 packs tightly into a hydrophobic cluster that includes a set of contacts with the N-proximal region of the M transmembrane anchor , close to the dimer twofold axis ( Zhang et al . , 2013a ) . Trimerization of E requires disruption of the packing around F450 , to pull it away from M; subsequent zippering of the stem requires some perturbation of the way α3 interacts with lipid . We expect substitution of a bulkier residue for F450 to destabilize the local packing with M and therefore to lower the barrier to trimer formation; indeed , the observed rate constant for the mutant was substantially higher than the rate constant for wild-type particles ( Figure 3A , top right-hand panel ) . The V434W mutant was most closely fit with a gamma distribution , N = 2 , and a rate constant k = 3 . 9 ± 0 . 3 × 10−2 , faster than wild-type , but without further data we cannot assign structural correlates for the two steps . 10 . 7554/eLife . 04389 . 012Figure 9 . Effects of mutations in membrane-proximal segment of E and temperature on VLP hemifusion . ( A ) Schematic of stem region in pre-fusion state with mutations indicated by stars . ( B ) Schematic , bulk fusion data , and single particle data for V434W mutant at pH 5 . 5 , 20°C; compare with Figure 3A , upper right-hand panel . ( C ) Schematic , bulk fusion data , and single particle data for F450W mutant; compare with Figure 3A , upper right-hand panel . ( D ) Single particle WNV VLP data at 22 and 24°C fit to a single exponential . ( E ) Yield of single particle events from 16 to 24°C . DOI: http://dx . doi . org/10 . 7554/eLife . 04389 . 012 Structures of viral fusion proteins , from x-ray crystallography and ( more recently ) from cryoEM , correspond either to an immature conformation , a mature , primed conformation , or a rearranged , postfusion conformation . The single-particle fusion experiments described here probe the transient , intervening states , inaccessible to direct structural analysis . By correlating fusion kinetics with specific , site-directed alterations in E , we have sought to determine the rate-limiting molecular events , the number of WNV E trimers needed to fuse , and the mechanism for coordinating conformational change among the several trimers that generate a single fusion pore . Previous work on influenza virus , taking a similar approach , has produced the following description of hemagglutinin ( HA ) catalyzed fusion ( Floyd et al . , 2008; Ivanovic et al . , 2013 ) ( Figure 9A ) . The rate-limiting step in the hemagglutinin ( HA ) conformational change is exposure of the fusion peptide and its engagement with the target bilayer . The contact zone between a typical influenza virus particle and the membrane with which it will fuse includes about 100 closely packed HA trimers . When the pH drops , stochastic rearrangement of trimers within that contact zone creates a random pattern of extended HA bridges between the two membranes . The extended intermediates have a detectable lifetime only because hydration force and resistance to membrane deformation prevent collapse of any single trimer to a postfusion conformation . The free energy recovered from collapse of three neighboring trimers is enough to overcome this barrier , and rapid collapse and hemifusion ensue as soon as three adjacent trimers in the contact zone have created intermembrane bridges . The sequence of molecular events in flavivirus fusion is somewhat more elaborate , because of the change in oligomeric association of E . E dimers , which initially form a regular icosahedral array , dissociate , exposing the fusion loops at the tip of domain II and allowing the subunits to project outwards from the virion surface . These events are fast , and under many conditions they may be in rapid equilibrium . Our results , both from experiment and simulation , suggest that the rate-limiting molecular step is formation of extended trimers . This conclusion is consistent with observations on low-pH induced trimer formation from soluble E ectodomain dimers: liposomes greatly enhance the yield of trimers , probably by transiently immobilizing the monomers in the right orientation . The rate of trimerization depends in turn on the effective surface concentration of activated monomers in the contact zone between virus particle and target membrane , and the rate of hemifusion , on the occurrence of a suitable number ( at least two ) of adjacent trimers within that zone . Since activation of individual monomers is stochastic , a critical constraint comes from the geometric requirements for availability of activated E to trimerize . Previous , bulk-phase studies of flavivirus and alphavirus fusion with liposomes using pyrene-modified lipids incorporated into the virion followed the decrease in pyrene excimer fluorescence that accompanies dilution into the target liposome ( Chatterjee et al . , 2000; Waarts et al . , 2002; Fritz et al . , 2011 ) . Our single particle analysis has extracted mechanistic detail from mutants that have no detectible bulk phenotype . The observed increase in single particle hemifusion yield at higher temperatures ( ∼40% at 24°C ) suggests that local rearrangements in the particle surface can help satisfy the geometric requirements for trimerization , increasing overall yield with little change in overall reaction rate ( Figure 9D–E ) . Experimental data for the 1:1 WT:W101A mixed particle showed no decrease in hemifusion yield ( Figure 4D ) . A likely explanation is that mixed trimers , with one or two Trp-containing fusion loops , can participate in facilitating hemifusion , perhaps in some cases with participation of a third trimer . A simulation assuming that any trimer with one or more mutated E monomers would be inactive greatly decreases yield , while relaxing this constraint recovers wild-type levels , as observed experimentally ( Figure 7D ) . The mechanism just outlined has some formal similarities to the one described for SNARE-catalyzed synaptic vesicle fusion , but the latter has an additional regulatory feature leading to a pause immediately before initial membrane mixing . Fast ( ms ) synaptic vesicular membrane fusion proceeds by rearrangement of multiple copies of the trans-membrane SNAP/SNARE complexes ( Takamori et al . , 2006; Hernandez et al . , 2014 ) , docked in a poised state analogous to the fusion-loop inserted , extended E-protein intermediate ( Figure 10B ) . The synaptic fusion machinery may also act in a stochastic fashion , in which a minimum number of complexes are sufficient to catalyze fusion . Although subsequent events are very fast , the poised state precedes any membrane merger: the two bilayers remain distinct until triggering , and activation by calcium-dependent co-factors allows coordinated , rapid progression from the extended state to full pore formation , probably through a transient hemifusion stalk ( Diao et al . , 2012; Lai et al . , 2014 ) . The pre-assembly step in synaptic vesicle fusion thus avoids the penalty imposed on viral fusion by the requirement that a critical number of neighboring bridges accumulate after the triggering event ( increased proton concentration ) before they can zipper and collapse . 10 . 7554/eLife . 04389 . 013Figure 10 . Kinetic mechanisms of membrane fusion . Darker arrows indicate faster steps . ( A ) Model of the sequence of events in influenza hemagglutin-mediated fusion . A representative schematic cross-section is shown ( an actual contact zone will include 100 or more trimers ) . Stochastic release of a sufficient minimum number of fusion peptides is rate limiting . ( B ) Schematic of calcium-triggered SNAP/SNARE-mediated fusion . Rapid triggering and activation of a minimum number of assembled complexes by calcium . ( C ) Schematic of flavivirus fusion: monomer activation is fast ( defined by pH ) , trimerization is rate limiting , and final collapse step is fast . DOI: http://dx . doi . org/10 . 7554/eLife . 04389 . 013 Synaptic vesicles and virus particles both have rather sharply bent lipid bilayers , with curvature imposed by the proteins that drive their budding from a cellular membrane . Moreover , both have high local protein density within the bilayer—at least 15% for flaviviruses and nearly 25% for synpatic vesicles ( Takamori et al . , 2006 ) . We find that WNV VLP fusion kinetics are independent of particle size and hence of curvature within the relevant range , which spans a substantial difference in outer:inner area ratios . The area of the outer polar-group layer in small VLPs is nearly three times that of the inner polar-group layer; the ratio is just over 1 . 5 for the virion-sized particles . The wedge-shaped transmembrane hairpins of M and E , neither of which extends fully into the lumen of the particle , may compensate in part for curvature and help drive budding into the ER; insertion of amphipathic α3 of the stem into the outer leaflet of the bilayer , occupying about 10% of the total outer-leaflet surface area , probably stabilizes curvature as well . With these compensations , even the rather sharp curvature of the small VLPs probably contributes very little to overcoming the barrier to fusion . A potential contribution to catalyzing membrane merger is the perturbation in the target membrane that comes from introducing fusion loops or fusion peptides into its lipid bilayer . The heterogeneity of fusion-loop and fusion-peptide structures and sequences among the various kinds of enveloped viruses appears , however , to rule out a common mechanism for such a disturbance . Our analysis therefore concentrates on the distortions imposed by linking large-scale conformational changes in the fusion protein with comparably strong deformations of the lipid bilayers . In the absence of regulation at a primed hemifusion step , like that regulated by synaptotagmin or complexin for SNAREs , the transition to fusion is very fast , once the necessary number of trimers has formed ( Figure 10B ) . The requirement that several neighboring monomers undergo domain rearrangements to assemble into trimers implies that the extended conformation will have a finite lifetime and that blocking trimerization will markedly reduce the likelihood of hemifusion . The potency of exogenous stem-derived peptide inhibitors of dengue virus ( Schmidt et al . , 2010 ) suggests that these properties provide an inhibitory strategy yet to be fully exploited by conformation-specific targeted antibodies or small-molecule inhibitors . West Nile Virus virus-like particles ( VLPs ) were produced from a stable 293T cell line transfected with the pVRC8400 expression vector with a structural cassette containing prM-E sequence from the genotype NY99 sequence ( produced by Angelica Medina–Selby , Doris Coit and Colin McCoin ( Lanciotti , 1999 ) ) preceded by the tissue plasminogen activator signal sequence . WNV VLPs were harvested at 37°C from Gibco FreeStyle 293 medium ( Life Technologies , Grand Island , NY ) , clarified from debris by low-speed centrifugation and precipitated with Polyethylene glycol 8000 . Following resuspension in buffer containing 20 mM Tricine ( N- ( 2-Hydroxy-1 , 1-bis ( hydroxymethyl ) ethyl ) glycine ) pH 7 . 8 , 140 mM NaCl and 0 . 005% Pluronic F-127 , VLPs were purified over a Optiprep density gradient ( SW41 rotor , 34 , 000 rpm , 4°C , 2 hr . 20 min . ) with 55%-45%-35%-30%-25%-20–10% steps . We collected the band between the 35% and 30% densities and found this material to contain >95% fully processed M and to consist of particles 35 and 50 nm in diameter as assessed by cryo- and negative-stain electron microscopy ( Allison et al . , 2003 ) . Particles were labeled with DiD ( 1 , 1′-Dioctadecyl-3 , 3 , 3′ , 3′-Tetramethylindodicarbocyanine Perchlorate ) at ∼20 μM or 20-fold the protein concentration . Excess dye was removed using NAP-10 desalting column ( GE Healthcare , United Kingdom ) . C6/36 cells were maintained in L-15 medium ( Mediatech , Manassas , VA ) supplemented with 10% fetal bovine serum . For plaque-forming assays ( PFAs ) , BHK-21 cells were in minimum essential medium ( α-MEM ) supplemented with penicillin , streptomycin , and 5% fetal bovine serum ( FBS ) . Aliquots of Kunjin virus were purified by PEG precipitation and an Optiprep gradient and labeled at ∼20 μM DiD ( or with equivalent concentration DMSO ) . Ten-fold dilutions in EBSS were prepared in 48-well plates , and 100 μl of each dilution added to the cells . The plates were incubated for 1 hr at 37°C , unabsorbed virus removed by two washes with PBS , and 1 ml of α-MEM , supplemented with carboxymethyl cellulose ( CMC ) , penicillin , streptomycin , and 2% FBS , was added to each well . After a 4-day incubation , the CMC overlay was removed , and the cells were washed with PBS and stained with crystal violet . The plates were washed with water to remove excess crystal violet and dried overnight . Glass coverslips were cleaned by sonication in ‘7X’ detergent , 1M potassium hydroxide , acetone and ethanol , and dried for 1 hr at 100°C . Polydimethylsiloxane ( PDMS ) flow cells with 0 . 5 mm wide and 70 μm high channels ( 5 per cell ) were prepared as described previously ( Ivanovic et al . , 2012 ) and bonded to plasma-treated glass . Teflon FEP tubing ( 0 . 2 mm , Upchurch Scientific ) connected an eppendorf tube with solution to the channel , and Intramedic polyethylene tubing ( 0 . 76 mm ) connected the channel to a syringe pump ( Harvard Pump 11; Harvard Apparatus , Holliston , MA ) . Liposomes for preparing planar bilayers contained 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine ( POPE ) , 1-oleoyl-2-palmitoyl-sn-glycero-3-phosphocholine ( POPC ) , cholesterol , and 1 , 2 , dioleoyl-sn-glycero-3-phosphocholine ( DOPC ) , 1 , 2-dioleoyl-sn-glycero-3-phosphoethanolamine-N- ( carboxyfluorescein ) ( FL-PE ) and 1 , 2-dioleoyl-sn-glycero-3-[ ( N- ( 5-amino-1-carboxypentyl ) iminodiacetic acid ) succinyl] ( Ni-NTA DOGS ) ( Avanti Polar Lipids , Alabaster , AL ) in a ratio of 4:2:2:2:0 . 02:1% . Liposomes at 10 mg/ml were extruded through a 200 nm pore-size polycarbonate membrane filter . Liposomes were loaded into the flow cell and the flow then stopped to allow bilayers to form . We preformed fluorescence recovery after photobleaching experiments to confirmed the fluidity of the bilayer ( Figure 2D ) . Unattached liposomes were washed away , and E16 Fab ( or DCSIGN-R ) with a C-terminal His6 tag was introduced at 50 nM for 2 min . His-tagged E16 was produced from a stable 293T line expressing both heavy and light chains from the pVRC8400 vector , purified by Ni-affinity chromatography and S200 size-exclusion chromatography . DCSIGN-R and soluble WNV constructs was expressed from Hi-5 cells infected with recombinant baculovirus . Labeled virus particles were loaded onto pseudo-receptor decorated bilayer . To initiate fusion , we introduced acetate buffer ( 100 mM sodium acetate , pH 5 . 0–5 . 5 ) or MES ( 100 mM , pH 5 . 75–6 . 25 ) , with 140 mM sodium chloride and 0 . 005% Pluronic F-127 . End-point bulk fusion data were collected using a GE Amersham Typhoon plate reader at 633 nm and 670 nm excitation and emission wavelengths respectively in 96-well clear-bottom plates with 2 mg/ml final lipid concentration ( 200 nm liposomes prepared as described above ) . VLPs were prepared and labeled with DiD as previously described . Kinetic Bulk liposome fusion data were collected on a PTI ( Photon technology International , Edison , NJ ) 814 Fluorimeter at 648 nm and 669 nm excitation and emission wavelengths respectively . Data were collected with Cole–Parmer digital polyStat temperature controlled thermo-jacket at 2 Hz over 10 min and at 0 . 2 mM final lipid concentration ( 200 nm liposomes prepared as described above ) . Single-particle fusion data were collected on an inverted Olympus IX71 fluorescence microscopy with a high numerical aperture objective ( 60× , N . A . = 1 . 3 ) . VLPs were illuminated with 488 and 640-nm Coherent ( Wilsonville , OR ) lasers . A custom-fabricated water-chilled temperature collar ( Bioptecs , Butler , PA ) was fitted on the objective turret . Each time-lapsed fluorescence Video was recorded at 1 Hz for 300 s using 3i Slidebook software . The position of each particle was determined by particle tracking analysis . Fluorescence trajectories were calculated by integrating the intensities from a 4 × 4 pixel region around each particle . Data were analyzed with software written in MatLab ( The Mathworks , Natick , MA ) , and Igor Pro ( WaveMetrics , Lake Oswego , OR ) . The square-root of the number of observations was chosen for the number of bins when generating histograms . As described in Floyd et al . ( Floyd et al . , 2010 ) , a waiting time distribution contains information regarding the mechanism of the process in its shape . If a process proceeds through an intermediate , the waiting time distribution is the joint probability density , or the convolution of each individual process . If each transition is a single-exponential decay:k1e[−k1τ]then the Gamma distribution is the convolution of N exponential decays:p ( τ ) =kNτN−1Γ ( N ) e−ktwhere Γ ( N ) = ( N−1 ) ! for integral N . For a process with two steps , the fit can be described ( Floyd et al . , 2010 ) by the equation:p ( τ ) ={k1k2k1−k2 ( e−k2τ−e−k1τ ) , k1≠k2k2τe−kτ , k1=k2For three steps , with three different rates:p ( τ ) =1 ( k2−k1 ) ( k1−k3 ) ( k2−k3 ) k1k2k3e−τ ( k1+k2+k3 ) ( ( k2−k1 ) eτ ( k1+k2 ) + ( k1−k3 ) eτ ( k1+k3 ) + ( k3−k2 ) eτ ( k2+k3 ) ) When two of the three rates are very similar , the distribution becomes:p ( τ ) =a2be−bτ−e−aτ+ ( a−b ) τe−aτ ( a−b ) 2where k1=k2=a , and k3=b . We compared fits for models for a single exponential , a process with two rates ( equal or not equal to one another ) , three rates ( each independent ) , and three rates ( with two set equal to one another ) . In all data presented in this study , we used a simple Akaike information criterion test ( AIC = χ2 + 2k , where k = the number of parameters; it measures the relative quality of a statistical model by comparing the trade-off between goodness of fit and the number of parameters ) . Decisions on the model used for fitting were made on a case-by-case basis . Dynamic light scattering measurements were made at dilute concentrations ( <0 . 5 mg/ml ) to avoid aggregation and at 20°C using a DynaPro Protein Solutions instrument ( Wyatt Technology , Santa Barbara , CA ) . Solvent refractive index and viscosity parameters were defined using the instrument PBS standard values . We calibrated the instrument using polystyrene beads ( 5 nm radius ) , tomato busy stunt virus in compact and expanded states ( R = 170 Å and 190 Å , respectively ) , and rotavirus double-layered particle as standards ( 350 Å radius ) . Each data point is the average of three measurements , each of which was determined by twenty consecutive 10 s acquisitions Data were filtered based on stability of light intensity and sample polydispersity criteria ( <20% ) with the program DYNAMICS v6 for data analysis . Simulations were written in MatLab ( Mathworks ) . Code found in Source code 1 .	Flaviviruses are a group of viruses that cause serious diseases in humans , including yellow fever , West Nile fever and dengue fever . Like all viruses , flaviviruses protect their genetic material with a protein shell and , like many other viruses , that shell also has a lipid membrane . Flaviruses use one of their surface membrane proteins , known as ‘envelope protein’ or simply ‘E’ , to bind to the surface of host cells . Once the virus has attached to the host cell membrane , it becomes engulfed within a bubble-like structure called an endosome , which also has a surrounding membrane . The interior of an endosome is acidic . Under these conditions the E protein undergoes a series of changes that bring the two membranes into close contact , so that the membrane of the virus can fuse with the membrane of the endosome . This membrane fusion allows the genome of the virus to escape the endosome and hijack the cell to make new copies of the virus . The E proteins on a mature flavivirus particle are found in pairs , but previous work showed that these proteins must work together in groups of three ( called ‘trimers’ ) for the viral and endosomal membranes to fuse . Chao et al . have now asked: what are the rate-limiting steps that lead to the formation of trimers ? And how many trimers are necessary to cause the membranes to fuse ? Chao et al . have investigated these questions using virus-like particles containing the E protein of West Nile Virus . They used techniques that can track individual particles , which their laboratory had previously used to investigate the influenza virus , to model changes in the E protein before , during and after membrane fusion . Chao et al . then made mutant versions of the envelope protein and used virus-like particles containing them to test the model . The data that Chao et al . obtained and computer simulations they carried out suggest that exposure to acidic conditions encourages the pairs of E proteins to separate and extend towards the endosome membrane . Individual E proteins then group together into trimers , and at least two trimers are needed to exert enough force to allow the membranes to fuse . The experimental design used by Chao et al . will now allow them to study the action of molecules that inhibit membrane fusion by West Nile Virus and other viruses .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2014	Sequential conformational rearrangements in flavivirus membrane fusion
We have developed a generally adaptable , novel high-throughput Viral Chromosome Conformation Capture assay ( V3C-seq ) for use in trans that allows genome-wide identification of the direct interactions of a lytic virus genome with distinct regions of the cellular chromosome . Upon infection , we found that the parvovirus Minute Virus of Mice ( MVM ) genome initially associated with sites of cellular DNA damage that in mock-infected cells also exhibited DNA damage as cells progressed through S-phase . As infection proceeded , new DNA damage sites were induced , and virus subsequently also associated with these . Sites of association identified biochemically were confirmed microscopically and MVM could be targeted specifically to artificially induced sites of DNA damage . Thus , MVM established replication at cellular DNA damage sites , which provide replication and expression machinery , and as cellular DNA damage accrued , virus spread additionally to newly damaged sites to amplify infection . MVM-associated sites overlap significantly with previously identified topologically-associated domains ( TADs ) . DNA viruses that replicate in the nucleus depend on host cellular functions for transcriptional and replication machinery to express and amplify their genomes . Accessing these functions is critical to productive infection , yet successful establishment of replication must also overcome cellular antiviral activity , which for larger DNA viruses includes innate immune responses , epigenetic silencing , the cellular DNA-damage response ( DDR ) , and antiviral activity found associated with PML bodies ( Weitzman et al . , 2010 ) . Replication of many DNA viruses takes place in distinct micro-nuclear compartments termed replication centers that are rich in factors viruses must interact with - either positively or negatively - to productively replicate ( Schmid et al . , 2014 ) . It is not fully clear , however , how nuclear-replicating viruses initiate replication centers in order to optimize access to factors and functions they need to either utilize or inactivate . Parvoviruses are small non-enveloped icosahedral viruses that are important pathogens in many animal species including humans . Minute Virus of Mice ( MVM ) is an autonomously replicating parvovirus that is lytic in murine cells and transformed human cells ( Cotmore and Tattersall , 2014 ) . The viral genome is approximately 5 kb and possesses inverted terminal repeats at each end that serve as origins of replication ( Cotmore and Tattersall , 2014 ) . MVM encodes two non-structural proteins: the larger non-structural phosphoprotein NS1 performs a number of functions required for viral replication , while NS2 plays important , currently undefined , roles during infection of the normal murine host ( Cotmore and Tattersall , 2014 ) . Parvoviruses are the only known viruses of vertebrates that contain single-stranded linear DNA genomes , and thus , they present novel replicative DNA structures to cells during infection . They depend heavily on cellular functions for replication , and unlike the DNA tumor viruses , do not drive quiescent cells into S-phase . However , following S-phase entry , cellular DNA polymerase δ converts the single stranded viral DNA genome into a double stranded molecule that serves as a template for transcription of the viral genes ( Cotmore and Tattersall , 2013 ) . As MVM infection progresses through S-phase , it induces substantial cellular DNA damage and evokes a robust , ATM-dependent DNA damage response ( DDR , [Adeyemi et al . , 2010] ) . Infection is characterized by a pre-mitotic cell cycle arrest that is both p21 and CHK1 independent ( Adeyemi and Pintel , 2012; 2014 ) . During this block virus replication proceeds for many hours , and ATM inhibitors reduce ongoing viral replication ( Adeyemi et al . , 2010 ) . Parvoviruses establish replication factories in the nucleus ( termed Autonomous Parvovirus-Associated Replication , or APAR , bodies ) where active transcription of viral genes and viral replication takes place ( Bashir et al . , 2001 ) . Low resolution confocal microscopy originally showed that DDR sensor and response proteins , cell cycle regulators , DNA polymerases and RNA polymerase II accumulate in MVM APAR bodies where they co-localize with replicating viral DNA and NS1 ( Adeyemi et al . , 2010; Bashir et al . , 2001; Ruiz et al . , 2011 ) . More recently , higher magnification microscopic studies revealed that phosphorylated histone H2A variant γ-H2AX , as well as other DDR proteins ( [Ruiz et al . , 2011] see Figure 1 , below ) , seemed to reside adjacent to , rather than within , early APAR bodies . This gave rise to the suggestion that these DDR factors may reside on cellular DNA in the vicinity of viral replication centers ( Ruiz et al . , 2011 ) . The induction of cellular DNA damage leads to almost instantaneous recruitment of DDR factors to the break site which coordinate complex signaling cascades that recruit DNA damage sensors , repair mediators and effectors ( Hashiguchi et al . , 2007; Polo and Jackson , 2011 ) . Endogenous sites of damage are also often associated with interference between replication and transcription polymerases ( Durkin and Glover , 2007 ) . DNA breaks , therefore , serve as cellular depots of DDR proteins , and factors involved in DNA replication and expression ( Hashiguchi et al . , 2007; Polo and Jackson , 2011 ) . While it is plausible that during infection damaged cellular DNA is relocated to sites of MVM replication , it seemed possible , alternatively , that the virus initially established its replication centers at damaged cellular sites where these DDR , replication , and expression factors were already present . It has been suggested that a number of viruses , including hepatitis B virus ( HBV ) and human papillomavirus ( HPV ) , associate with sites of DNA damage , including early-replicating or common , fragile sites ( ERFs , and CFSs , respectively ) at early times during their infections ( Durkin and Glover , 2007; Jang et al . , 2014; Tubbs and Nussenzweig , 2017 ) . The HPV genome was shown to be tethered to CFSs by the chromatin modifier BRD4 , which facilitates subsequent integration into the host genome utilizing the cellular DDR machinery ( Feitelson and Lee , 2007; Jang et al . , 2014 ) . These studies utilized chromatin immunoprecipitation ( ChIP ) assays of the HPV E2 protein to identify sites of HPV localization to cellular CFSs , which were then validated by 3D-FISH assays of the viral and cellular DNA ( Jang et al . , 2014 ) . Consistent with these findings , crosslinked-ChIP assays have recently demonstrated that FANCD2 , required for maintaining fragile site stability and coordinating their replication , also associates with HPV genomes at replication centers ( Madireddy et al . , 2016; Spriggs and Laimins , 2017 ) . However , an unbiased way to map the interaction between the viral and cellular genomes directly has been largely unavailable until recently . The development of Chromosome Conformation Capture ( 3C ) technologies has enabled detailed analysis of cis-interactions between separated regions of the genome , as well as aspects of chromatin packaging ( Dixon et al . , 2016 ) . When combined with high-throughput sequencing , these techniques ( termed 4C , 5C , and Hi-C assays ) have become valuable tools for studying the details of nuclear configuration ( Denker and de Laat , 2016 ) . Specifically , 4C and Hi-C assays provide genome-wide interaction data through unbiased deep sequencing . Hi-C assays are designed to provide information on all nuclear interactions whereas 4C assays enable higher resolving power and a deeper interrogation of genomic associations by utilizing inverse PCRs to amplify only the ‘bait’ ( Lajoie et al . , 2015 ) . In this study , we report the novel adaptation of high-throughput circular chromosome conformation capture assay ( 4C ) for use in trans which we term V3C ( Viral Chromosome Conformation Capture ) . This assay , which should be generally adaptable , has allowed us to characterize , on a genome-wide scale , the direct association of the linear MVM genome with discreet regions of the cellular genome . These sites , termed Virus Association Domains ( VADs ) , correlated initially with sites of cellular DNA damage that in mock-infected cells also exhibited damage as cells progressed through S-phase . As infection progressed these sites expanded , and additional sites of DNA damage were induced . MVM subsequently associated with the newly induced sites as infection amplified . Sites of association identified biochemically were confirmed microscopically , and in addition , MVM could be targeted specifically to sites of DNA damage artificially engineered into the cellular chromosome . Development of the V3C assay has allowed us to suggest the following model . Soon after nuclear entry MVM homes first to sites of pre-existing endogenous DNA damage to initiate infection at sites that provide cellular factors necessary for its replication . Subsequently , as cellular DNA damage accrues , virus spreads additionally to these sites of damage to amplify infection . The V3C-seq assay should be useful for characterizing the interaction of many DNA viruses that associate with the cellular genome , and provide a useful tool to characterize the molecular events leading to the initiation of infection . As MVM infection progresses through S-phase , it induces cellular DNA damage and evokes a robust , ATM-dependent DNA damage response ( DDR ) characterized by a pre-mitotic cell cycle arrest that is both p21 and CHK1 independent ( Adeyemi and Pintel , 2012 , 2014 ) . During this block , virus replication proceeds for many hours , and ATM inhibitors reduce ongoing viral replication ( Adeyemi et al . , 2010 ) . Standard confocal microscopy demonstrated previously that numerous cell cycle and DDR effector proteins , RNA polymerase II , as well as DNA polymerase-α and δ are found associated with MVM replication centers called APAR bodies ( Adeyemi et al . , 2010; Bashir et al . , 2000; Kollek et al . , 1982; Ruiz et al . , 2011 ) . Similar examples of such confocal images of DDR proteins associated with APAR bodies , but not the irrelevant transcription factor NR5A2 ( Duggavathi et al . , 2008 ) , processed with deconvolution , can be seen in Figure 1A . Comparison of the median three-dimensional distance between FANCD2 and γ-H2AX - which localize to stalled replication forks where they facilitate DNA repair ( Kim et al . , 2018; Lossaint et al . , 2013; Madireddy et al . , 2016 ) , and the center of APAR bodies - identified by MVM NS1 staining , indicated that at 16 hr post infection ( hpi ) and release ( representing approximately 8–10 hr into S-phase in our para-synchronization protocol ) , they localized closely with MVM replication centers ( Figure 1B ) . This was in contrast to the irrelevant transcription factor NR5A2 ( Duggavathi et al . , 2008 ) which exhibited a diffuse localization ( Figure 1B ) . However , confocal super resolution imaging ( Airyscan ) demonstrated that γ-H2AX seemingly localized to the periphery of APAR bodies ( Figure 1C ) . Super-resolution imaging using the GSD-STORM platform also demonstrated that γ-H2AX , and FANCD2 ( Figure 1D ) , localized to the periphery of APAR bodies ( Figure 1D ) . γ-H2AX characteristically amplifies on damaged DNA as megabase ( Mb ) -sized platforms , making the possibility of marking the 5 kb MVM genome with phosphorylated histone H2AX less likely ( Rogakou et al . , 1999 ) . These results suggested that , MVM replication centers may localize to and expand adjacent to sites of cellular chromatin undergoing DNA damage , where replication , expression and DDR factors reside . To characterize the association of the MVM genome with the cellular genome during lytic infection more directly we have developed a high-throughput chromosome conformation capture ( 3C ) assay for use in trans . 3C assays have been typically used to identify long-range interactions between regions of a single chromosome ( Dekker et al . , 2013 ) . Our analysis , which we term V3C-seq ( Viral Chromosome Conformation Capture Sequencing ) allowed us to identify direct interactions between the linear MVMp genome and the cellular chromosome in an infected cell population on a genome-wide scale in an unbiased manner . V3C-seq assays utilize formaldehyde-mediated crosslinking to first ‘freeze’ the locations of the viral and cellular genomes at various points during infection . Samples are then digested and ligated under conditions that favor intramolecular interaction , and the resultant novel virus-cell DNA fragments are subjected to high-throughput sequencing ( Figure 2A ) . The assay provides a precise genomic map of the sites with which viral DNA interacts , and the frequency with which unique individual linked fragments are generated provides quantification of these interactions . V3C-seq was performed in parasynchronized mouse A9 fibroblasts , the traditional host for MVM , at various times post-infection . A typical time-course of MVM infection is shown in Figure 2—figure supplement 1A . Assays utilized a viral viewpoint at a HindIII site at nucleotide 2651 and NlaIII site at 1899 in the MVMp genome , thereby capturing the interaction of the MVMp fragment containing both the viral P4 and P38 promoters upstream of the HindIII site . Clustering algorithms and visualization of interaction sites on the UCSC Genome Browser ( Kent et al . , 2002; Ramírez et al . , 2016 ) revealed that by 12 hr post-infection and release ( representing approximately 4–6 hr into S-phase ) MVM genomes associated with discrete regions on most cellular chromosomes [Figure 2—figure supplements 1B and 2 , ( Kent et al . , 2002 ) ] . These cellular sites served as initial amplification points for MVM , and upon progression to 16 hpi , the virus both expanded at these regions and associated with new sites ( Figure 2B , Figure 2—figure supplement 2 ) . We term these sites of association Virus Associated Domains , or VADs . The clusters of interacting sites on each chromosome ranged in density and in size ( Figure 2—figure supplement 2 ) , from approximately 1–2 Mb to larger 5–15 Mb-size domains ( chromosomes 17 and 19 are shown in Figure 2B ) . While most chromosomes contained multiple small VADS , many contained 1–3 larger VADs of 5–15 Mb size . Chromosomes 1 , 13 , 18 , X and Y had fewer discernible interaction sites . The larger VADs in Figure 2B are boxed for comparison purposes but is not meant to restrict the designation of VADs to a particular size . Comparison of MVM interaction sites across the entire mouse genome showed that approximately 51% of VADs identified at 12 hpi were retained at 16 hpi , indicating that approximately 84% of VADs identified at 16 hpi were newly generated ( Figure 2C ) . Approximately 39% of MVM interaction sites detected at 16 hpi were retained at 20 hpi , while only approximately 18% of the interaction sites identified at 20 hpi were newly generated ( Figure 2C ) . Clustering analysis of MVM interaction sites in multiple replicates over the time-course of infection showed that they were reproducible across replicates ( Figure 2—figure supplement 1B ) . Importantly , MVM interaction sites clustered together at 16 and 20 hpi in a characteristic manner that was distinct from early ( 12 hpi ) , and late ( 24 hpi ) infection . These results suggested that interactions of MVM with the cellular chromosome increased as infection progressed . By 24 hpi , MVM interaction with the host chromosome was extensive ( Figure 2—figure supplement 2 ) . This latter observation indicated that MVM interaction with the cellular genome at VADs during early stages of infection was not an artifact of preferential sequencing at the VAD sites , and is consistent with previously published profiling of APAR bodies by microscopy ( Ruiz et al . , 2011 ) . Additionally , at these late times , in the presence of saturating amount of MVM DNA , we observed interactions of the viral genome with sites throughout the mouse genome . This further suggested that VADs identified earlier during infection featured properties that enabled viral recruitment and replication . It is noteworthy , however , that the cellular genome undergoes substantial DNA damage by late stages of infection ( Figure 3—figure supplement 2A and as described below ) , likely precluding detection of some interactions effectively by V3C-seq . In addition , viral packaging would be expected to reduce available viral genomes by approximately 20 hpi ( Cotmore and Tattersall , 2014 ) , which would also be predicted to contribute to the decreased interaction seen at 20 hpi in Figure 2B . We chose to validate the V3C assay by confirming the association of MVM with one of these VADs , murine 19qA , using a focused Taqman-based assay , in which the frequency of novel ligation junctions were determined by quantitative PCR ( qPCR ) . MVM association with the VAD at 19qA was readily detectable in our standard protocol using Taqman probes complementary to the MVM genome ( forward ) and the cellular genomic site ( reverse ) ( Figure 2D ) . This association was substantially diminished , either in the absence of intramolecular ligation , or when cross-links were reversed prior to intramolecular ligation ( Figure 2D ) . These experiments defined the lower limit of background levels generated by our V3C assay , and suggested that MVM associations with cellular VADs were specific , and mediated by DNA-DNA and/or DNA-protein intramolecular crosslinks . Focused 3C-qPCR in parasynchronized NIH-3T3 fibroblasts infected with MVMp ( Figure 2E ) , and EL4 lymphocyte cells infected with the lymphotrophic variant MVMi ( Figure 2F ) both showed association with a subset of the VADs identified in A9 cells ( Figure 2B , Figure 2—figure supplement 2 ) , which suggested a common mechanism may exist for the establishment of MVM replication sites in these mouse cell lines . Differential rates of MVM replication and cell cycle kinetics in A9 compared to NIH-3T3 and EL4 cells precluded performance of 3C-qPCR assays simultaneously in these systems . Together , we interpret our results as establishing that V3C-seq is a valid means to map the interaction of a lytic linear DNA virus with specific sites on the host cell genome . As described above , super-resolution microscopy suggested that MVM replication centers seemingly associated adjacent to genomic sites containing factors involved in replication , expression , and the DDR . ERFs have such characteristics in uninfected cells , and as mentioned , MVM continues to induce DNA damage as infection proceeds ( Adeyemi et al . , 2010; Barlow et al . , 2013 ) . Therefore , as we identified sites of viral interaction with the cellular genome , we looked for association with sites of cellular DNA damage . To identify sites of cellular DNA damage , we initially performed chromatin immunoprecipitation coupled with high-throughput sequencing ( ChIP-seq , [Landt et al . , 2012] ) for γ-H2AX in parasynchronized A9 cells , either mock infected , infected with MVM , or mock infected and treated with hydroxyurea ( HU ) . Results for algorithm-called peaks for chromosomes 17 and 19 are shown in Figure 3A , and for the complete murine genome is shown in Figure 3—figure supplement 1 . Mock infected A9 cells ( taken 12 hr post release , hpr ) showed a significant number of sites of damage , as identified by γ-H2AX , as they passed into S-phase . These were likely ERFs , which accrue damage during replication ( Figure 3A ) . Whole-genome peak analysis of all γ-H2AX bound regions revealed that approximately 55% of the sites identified in mock infected cells overlapped with those identified at 16 hr post-infection ( hpi ) ; approximately 55% of sites identified at 16 hpi were newly generated ( Figure 3B , peak calling using EPIC and intersection using BEDtools , ( Quinlan and Hall , 2010 ) , as described in Materials and methods ) . By 16 hpi , sites of damage concentrated in distribution , and expanded in number ( Figure 3A ) . At this point in infection MVM had begun to induce additional sites of DNA damage , as evident both by ChIP-seq and increased tail moments in Comet assays ( Figure 3—figure supplement 2A ) . The majority of γ-H2AX-containing sites at 20 hpi were newly generated , coinciding with only 10% of the sites identified at 16 hpi , indicating the widespread induction of DNA damage by this point of infection . This can be seen more clearly in the magnified view of chromosomes 17 and 19 shown in Figure 3—figure supplement 2C . Interestingly , the γ-H2AX ChIP-seq regions identified at 16 hpi correlated well with γ-H2AX ChIP-seq performed following 12 hr treatment with HU ( Figures 3A and 16 hpi vs HU ) . Approximately 51% of γ-H2AX peaks detected 16 hpi were shared with those induced after 12 hr treatment with HU , and conversely , approximately 26% of the peaks identified following treatment with HU were shared at 16 hpi . In order to confirm the statistical significance of the intersection analyses , the γ-H2AX peaks at indicated time points were intersected with randomly permuted peaks across the mouse genome and visualized as a Jaccard Plot ( Figure 3B , far right ) . The MVM genome initiated infection at sites of cellular DNA damage that in mock infected cells also exhibited DNA damage as the cells cycled through S-phase , and as infection progressed , localized to additional sites of induced damage . Comparisons of the ChIP-seq results with V3C-seq assays showed that MVM associated directly with sites of cellular DNA damage , as identified by the presence of γ-H2AX at the same region , in a manner that increased as infection progressed . Figure 3A compares MVM VADs at 16 hpi , to sites of DNA damage ( as determined by γ-H2AX ChIP-seq ) for chromosomes 17 and 19 as infection progressed . Large VAD regions in Figure 3A are boxed for comparison purposes , but are not meant to restrict overlap only to VADs of that size . Comparisons for the full mouse genome are shown in Figure 3—figure supplement 1 and while there is significant variation , the overlap between VADs and sites positive for γ-H2AX ChIP-seq was strikingly consistent . Figure 3C summarizes the genome-wide correlation at the nucleotide level of VADs and γ-H2AX ChIP-seq data presented in Figure 3—figure supplement 1 . For the composite comparisons at 12 , 16 and 20 hr post-infection , data was taken from the same experiment ( comparing VADs at various time points as shown in Figure 2B to ChIP-seq sites at those times as shown in Figure 3A ) . At 12 hpi , MVM associated with approximately 55% of sites that in mock infected cells exhibited DNA damage upon progression into S-phase . By 16 hpi , this association rose to close to 80% . By 16 hpi , close to 90% of γ-H2AX occupied sites overlapped with VADs ( Figure 3C ) , which included γ-H2AX sites present in uninfected cells ( Figure 3A ) . By the late time point of 20 hpi , approximately 70% of the γ-H2AX sites co-localized with VADs . Visualization of MVM association in the vicinity of γ-H2AX-positive sites using hierarchical clustering further revealed that MVM association with damaged sites increased from 16 hpi to 20 hpi; however , increased incidence of cellular DNA breaks after 20 hpi led to a decrease in the proportion of MVM-associated damaged sites ( Figure 3D ) . Notably , approximately 25% of the VADs identified at 12 hpi , and approximately 95% of VADs identified at 16 hpi , associated with the γ-H2AX sites identified following 12 hr of treatment with HU . Randomly generated peaks showed less than 1% overlap with γ-H2AX peaks identified 16 hpi ( Figure 3C ) . A magnified view of the large VADs at 19qA and 17qA/B outlined in Figure 3A is also provided in Figure 3—figure supplement 2C ( left ) , while further magnifications of VAD regions demarcated by red rectangles in Figure 3—figure supplement 2C ( left panel ) at Narfl , Vwa7 , Ehd1 and Slc29a2 genes are shown on the right panel . The strong correlation of MVM interaction sites with sites that in uninfected cells exhibit DNA damage upon replication is consistent with the notion that MVM may have initially established replication at cellular fragile sites that are susceptible to DNA damage as cells cycle through S-phase , although it cannot be formally ruled out that these sites were virally-induced at the earliest times in infection . It is also important to note that VADs also correlated strongly with γ-H2AX ChIP-seq sites identified on cellular chromosomes of non-infected , HU-treated A9 cells ( Figure 3A , red rectangles ) . This strongly implied , although does not prove , that the γ-H2AX identified associated with MVM during infection resides on cellular DNA . It is striking that sites of damage in uninfected cells , sites of damage induced by virus , and sites of damage induced by treatment of the DNA-damaging agent HU overlap so significantly . Cellular sites of DNA damage also often contain BRCA1 , which binds DNA and can co-localize with γ-H2AX in DNA double-strand break repair foci , although typically in a more narrow pattern ( Barlow et al . , 2013 ) . As expected , VADs also strongly associated with sites identified by BRCA1 ChIP-seq at 16 hr post-infection ( Figure 4A , row 2 ) . Furthermore , VADs also overlapped with BRCA1 and γ-H2AX sites in primary mouse cells induced with replication stress agents ( Figure 4—figure supplement 1A and B ) , and characterized as ERFs , in previously published studies ( Barlow et al . , 2013 ) . As MVM can infect transformed human cells , we also performed focused V3C-qPCR at 16 hpi in parasynchronized SV40-transformed human NB324K cells . As shown in Figure 4—figure supplement 1C , MVM localized to the previously characterized human fragile site FRA5H , but not FRA11F . MVM also associated weakly at this time point with the prototypical human fragile site , FRA3B . As an indirect confirmation of MVM recruitment to the cellular genome , we performed ChIP-seq for the MVM-NS1 protein , which binds covalently to the viral genome , and non-covalently to additional ACCAACCA consensus sequences throughout the MVM genome ( Christensen et al . , 1995 ) . We reasoned that ChIP-seq assays for NS1 would confirm cellular sites associated with the viral DNA by secondary crosslinking of NS1-bound MVM DNA to cellular DNA . Figure 4A shows NS1 binding profiles to cellular chromosome 17 and 19 that are concordant with the VADs at 16 hpi , further validating our findings from V3C-seq assays . A genome-wide analysis of called peaks indicated that approximately 90% of the peaks identified by NS1- , BRCA1 - , and γ-H2AX ChIP-seq overlapped with VADs identified by V3C ( Figure 4B and C ) , while overlap was undetectable when intersected with a randomly generated library of ChIP-seq peaks of equivalent size ( Figure 4B ) . In concordance with these findings , the binding profile of NS1 , BRCA1 and γ-H2AX around a VAD site at 16 hpi centered within 1 Mb of the MVM associated cellular site ( Figure 4B ) . Taken together , our V3C-seq and ChIP-seq experiments are consistent with a model that upon infection , MVM first localized to cellular sites susceptible to DNA damage as cells progressed into S-phase , and as infection progressed , localized to additional sites of damage that were virally induced , to amplify its replication . We next sought to confirm the association of MVM replication with sites of cellular DNA damage using super-resolution ( STORM ) microscopy . For these assays , we designed PCR-based FISH probes complementary to the MVM genome , to a VAD regions at 19qA , and to a control , VAD-negative , site at 6 pA ( Figure 5A ) . 3D-FISH combined with confocal imaging of multiple nuclei was performed at 16 hpi . Representative examples are shown in Figure 5B and D , which demonstrated close localization between the MVM genome and 19qA-VAD probes ( represented by red and green probes respectively ) , in contrast to the lack of direct localization of MVM with the control probe at chromosome 6 pA ( Figure 5C and D; represented by red and cyan foci respectively ) . The 3D distances between the VAD probe and MVM genome in multiple nuclei were calculated using confocal imaging . As shown in Figure 5E , the median distance between the MVM genome and the 19qA-VAD and 15qE-VAD probes were approximately 0 . 7 , and 0 . 6 µm , respectively . This is similar to the median radius of Type II APAR bodies , suggesting that on average VAD sites coincide with APAR bodies . In contrast , non-VAD control sites on chromosome 12 and 17 ( 12qA3 and 17qA2 ) exhibited a much greater range of co-location and were separated from its nearest MVM genome by median distances of 1 . 1 and 1 . 1 µm , respectively . Taken together , our representative super resolution imaging and quantitative 3D-FISH analyses support V3C results demonstrating that MVM localized with VADs . If MVM preferentially associates with cellular sites of DNA damage , one might expect that MVM could be targeted to artificially–engineered sites of cellular DNA damage . We tested this in two ways . First , we used laser micro-irradiation of MVM-infected A9 cells at 18 hpi to induce focused cellular DNA damage , which is evident as a γ-H2AX ‘stripe’ in the nucleus ( Figure 6A ) . Anti-NS1 staining of these cells suggested that MVM distinctly co-localized with irradiation-induced damaged cellular DNA , and in doing so , viral replication centers adapted to the shape of the damaged DNA stripe ( Figure 6A , top two panels ) , rather than the distinct foci characteristic of APAR bodies ( representative example shown in Figure 6A , bottom panel ) . Localization of the cellular transcription factor NR5A2 , which is not found in MVM replication centers ( Figure 1B ) , was not affected by micro-irradiation ( Figure 6A , third panel ) . These experiments demonstrated that at least NS1 localized to induced sites of DNA damage . The V3C-seq/NS1 ChIP-seq experiments described above suggested that NS1 was a useful surrogate for virus replication in our assays; however , we chose to further assess the direct interaction of the MVM genome with sites of artificially induced cellular DNA damage using directed cleavage of the genome by CRISPR/Cas9 ( Sanjana et al . , 2014 ) . Guide RNAs were designed that targeted a gene desert in chromosome 9 ( Figure 6B , cytogenetic location at 9qE1 ) . These guides were transfected into A9 fibroblasts stably expressing CRISPR/Cas9 . When these cells were then infected with MVM and assayed by focused 3C-qPCR , we detected substantially more amplicons between MVM and the DNA break site in cells transfected with 9qE1 desert-specific guide RNAs , compared to scrambled control guides ( Figure 6C ) . Interaction was further confirmed using complementary TaqMan probes that recognize the chromosome 9 CRISPR cleavage site ( Figure 6D ) . As expected , MVM interaction with a previously identified VAD on Chromosome 19 ( 19qA ) was not significantly affected in this assay ( Figure 6E ) . As an independent verification of the localization of MVM to this site , we also detected NS1 binding to the induced damage site using ChIP , suggesting that NS1 bound to the MVM genome at the break site is secondarily crosslinked and detected by ChIP-qPCR ( Figure 6F ) . NS1 binding to the chromosome 19 VAD at 19qA was unaffected in these experiments ( Figure 6G ) , consistent with our V3C findings in Figure 6C and D . When DNA viruses enter the nucleus they must locate to sites suitable to sustain replication . Small DNA viruses , such as parvoviruses , require multiple cellular factors for the expression and replication of their genomes . It may be that DNA viruses set up replication centers essentially randomly , and factors necessary for replication are recruited to these sites . An alternative model , suggested by the present work , is that incoming DNA viruses can initially locate to cellular sites that maintain factors necessary for virus replication . Sites of cellular DNA damage present such an opportunity ( Hashiguchi et al . , 2007; Polo and Jackson , 2011 ) . Using a high-throughput conformational capture assay developed here for use in trans , we show that at early times post infection MVM interacted directly with sites of cellular DNA damage that in mock infected cells also exhibited DNA damage upon entry into S-phase . As infection progressed , interaction of the MVM genome , as well as the presence of the viral replication protein NS1 , increased at these sites , suggesting they were sites of ongoing viral replication . ChIP-seq analysis for γ-H2AX and BRCA1 demonstrated that DNA damage increased during infection , and MVM subsequently associated also with newly induced sites . This model is supported by our findings that MVM NS1 and the MVM genome could be re-localized to artificially induced sites of cellular DNA damage . These observations support a model consistent with the notion that MVM initially establishes replication at cellular DNA damage sites that provide replication and expression machinery , as well as other DDR factors , and as infection progresses induces additional sites of cellular DNA damage , using these to amplify infection . It should be noted , however , that even at the earliest time points examined , multiple virus genomes have accumulated at sites of replication . Thus , specific localization of input genomes remains circumstantial . Consistent with our overall model , we have previously demonstrated that an ATM inhibitor applied during infection specifically reduced virus replication ( Adeyemi et al . , 2010 ) . Also , as might be expected , we find that HU pretreatment of permissive rat F111 cells , which have lower levels of endogenous DNA damage as indicated by lower levels of γ-H2AX , resulted in increased MVM replication by 20 hpi ( Figure 3—figure supplement 2B ) . Because VADs correlated strongly with γ-H2AX ChIP-seq sites identified on mock-infected cells as they progressed through S-phase , and on non-infected , HU-treated A9 cells , the predominantly identified γ-H2AX signal associated with MVM during infection very likely resides on cellular DNA . However it remains possible that γ-H2AX and/or other DDR signaling proteins are directly associated with the viral genome during infection . In this regard , while there is a clear potential role in MVM replication for DNA polymerase-δ and gene expression factors potentially present at DNA damage sites , the possible roles of other DDR proteins in parvovirus replication warrants additional study . Recently Shah and O’Shea have elegantly demonstrated a bipartite cellular DDR to adenovirus ( Ad ) infection that initially targets the virus while sparing the cell . The Mre11-Rad50-Nbs1 ( MRN ) complex first inhibits adenovirus replication without inducing a global response that could interfere with cellular proliferation or viability . As Ad overcomes this block , utilizing the Ad E4Orf6/E1B 55 kDa complex , and replicates to high levels , a global DDR is subsequently induced , and cellular DNA breaks were found to sequester DDR proteins from adenovirus preventing its replication ( Shah and O'Shea , 2015 ) . While MVM localizes to sites of cellular DNA damage as replication ensues , the model that we propose for parvovirus infection is different . It is consistent with the parvovirus life cycle , which depends upon the induction of a cell cycle arrest , is susceptible to ATM inhibitors , and which continually induces cellular DNA damage during its infection . Additionally , in contrast to adenovirus , MVM is much more dependent on host cell factors for its replication and expression ( Cotmore and Tattersall , 2014 ) , and it does not have an extensive genetic capacity to encode functions that inactivate the cellular DDR as does adenovirus ( Ou et al . , 2012; Querido et al . , 2001; Stracker et al . , 2002 ) . Chromosome conformation assays have been traditionally used to analyze the 3D folding principles of the cellular genome , including the regulation of promoter-enhancer loops and structural folding loops ( Dixon et al . , 2016 ) . However , these assays can also serve as valuable tools to study the interaction between the host genome and invading virus . In the context of virus-infected cells , Chromatin Interaction Analysis by Paired-End Tag Sequencing ( ChIA-PET ) assays have enabled the comprehensive mapping of EBV interaction with the cellular genome ( Jiang et al . , 2017 ) . These studies , which combine ChIP-seq with chromosome conformation capture techniques , utilize the proximal interaction of distally located DNA regions bound by shared protein elements . They have demonstrated that Epstein Barr Virus enhancers can regulate the expression of the cellular Myc oncogene in lymphoblastoid cells via long-range promoter-enhancer looping , thereby contributing the EBV-mediated cellular transformation ( Jiang et al . , 2017 ) . However , mapping of the virus-host interactome by ChIA-PET experiments can be limited by the necessity of having a priori knowledge of the proteins mediating this interaction . The EBV interactome has also been characterized using in-situ Hi-C assays , which generate chromosome conformation maps that provide a snapshot of how every restriction enzyme site associates with every other site throughout the genome . These studies showed that the latent EBV episome associates with gene poor regions , but relocalizes to gene-rich regions of the genome upon reactivation ( Moquin et al . , 2017 ) . While highly informative , these studies did not have the resolving power of the assays described here for MVM . For MVM , a single inverse-PCR viewpoint was sufficient to map its interactome . Moreover , using inverse PCR using primers complementary to the viral genome to generate the sequencing library ensured that the V3C-seq assay detects only MVM-host hybrid junctions . This enabled higher resolving power and a deeper interrogation of genomic associations , allowing the detection of both frequent and infrequent MVM interaction sites . The drawback of V3C-seq , however , is that it does not assay changes in nuclear architecture in response to viral infection , which invariably is altered during parvovirus replication , particularly in late stages . Looping interactions have previously been observed at distinct loci in genomic regions proximal to some integrated viral genomes . For example , the integrated Murine Leukemia Virus ( MuLV ) genome has beRen , BRen , Ben shown to influence the folding properties of its proximally located Myc promoter , which in turn contributes to cellular oncogenesis ( Zhang et al . , 2012 ) . In other studies , it has been shown in cell line models that the integrated HIV genome associates distally with an uncharacterized chromosomal region to promote reactivation of latent HIV ( Dieudonné et al . , 2009 ) . Chromosome conformation capture analyses focused on elucidating the topological conformation of the viral genome have also revealed the conformational structure adopted by the gammaherpesvirus Kaposi’s Sarcoma Associated Herpesvirus ( KSHV ) , which forms distinct promoter-enhancer loops mediated by the cellular architectural proteins CTCF and Cohesin ( Kang et al . , 2011 ) . In this study we have utilized chromosome conformation capture technology for the first time to map the trans interaction of a lytic virus with the cellular genome in a non-biased way . However , in addition to assaying the inter-chromosomal interaction cellular sites , long-range interactions of integrated viruses such as MuLV and HIV with distal elements , or TADs , may also be assayed using V3C-seq . Inspection of VADs show that regions of MVM interaction vary greatly in size . The size of the peaks correlates with the number of interactions , and so indicate that some regions are more densely populated by virus than others . As can be seen in the whole-genome analysis , there is quite a variation in VADs across different chromosomes ( Figure 2—figure supplement 2 ) . While most chromosomes have multiple VADS , most have only a few larger VADs of 5–15 Mb . Whether the size of the VADs correlates to the success of infection at that site is not known . There are a number of chromosomes [chromosomes 1 , 13 , 14 , 18 , X and Y ( Figure 2—figure supplement 2 and Figure 3—figure supplement 1 ) ] that have few sites of damage , as assessed by γ-H2AX ChIP-seq , as well as few VADs , reinforcing their correlation . VADs identified by V3C-seq showed striking overlap with sites of damage in infected cells , at sites of damage incurred in uninfected cells as they progressed into S-phase , but surprisingly , also at sites in uninfected cells treated with HU . This overlap suggested that there may be a predilection in the cellular chromosome for sites sensitive to the induction of damage for which MVM has an affinity . In this vein , we find that most VADs ( and by implication sites of endogenous and infection-induced damage and damage induced by HU ) also overlap strikingly with Topologically Associated Domains ( TADs ) , determined for murine CH12 B-cells , by Hi-C assays ( Figure 4—figure supplement 2A–D , [Rao et al . , 2014] ) . Such studies , in combination with high-resolution imaging , have shown that the mammalian genome is folded into megabase-sized cis-interacting regions ( TADs ) . TADs have been divided into two main compartments: A ( primarily euchromatin-like ) , and B ( primarily heterochromatin-like ) ( Rao et al . , 2014 ) . The A compartment is characterized by being gene dense , containing highly expressed genes , and active chromatin marks . The A1 subcompartment was seen to finish replicating at the beginning of S-phase , while the A2 subcompartment was described as continuing to replicate into the middle of S-phase . Overlaying published chromatin marks from murine 3T3 cells induced with the DNA damaging agent aphidicolin ( Kraushaar et al . , 2013 ) suggests that MVM associations may be with the A compartment ( Figure 4—figure supplement 2E ) . This compartment of cellular DNA would provide the necessary machinery for the expression and replication of MVM within S-phase , and would be consistent with MVM initially associating with sites that emerge as ERFs during cellular replication , which would be predicted to replicate in compartment A . The VADs identified in our study occupy multiple adjacent TADs . This suggests that several adjacent TADs may contain the necessary environment to support MVM expression and replication . The mechanisms that maintain TAD borders are elusive , but structural proteins such as CTCF and cohesin have been implicated in these processes ( Dixon et al . , 2016; Rao et al . , 2014 , 2017 ) . Indeed , Canela et al . , have shown that CTCF and RAD21 form the anchors to cis-interacting cellular loops that are susceptible to topoisomerase TOP2B mediated DNA breaks ( Canela et al . , 2017 ) . The MVM NS1 protein , which also interacts at TADs , is a known DNA binding protein with double-stranded nickase activity; however , the mechanisms by which the viral genome associates with the cellular sites remains to be determined . Our results suggest that host chromatin states may play a significant role in permissiveness to DNA damage , and thereby influence MVM localization for replication . Consistent with this model , the HPV E2 protein has been found associated with actively transcribing genes and active chromatin , presumably to facilitate expression of HPV through hijacking the host transcriptional machinery at these sites ( Jang et al . , 2009 ) . Indeed , profiling of sites on the cellular genomic that undergo replication stress have shown that these regions are encased in a protective chromatin environment to facilitate efficient repair ( Kim et al . , 2018 ) . Such regions , either pre-existing prior to infection , or induced by virus , may provide the supportive environment necessary for successful infection . The generally adaptable , high-throughput V3C-seq assay allows genome-wide identification of the direct associations of viral genomes with distinct regions of the cellular chromosome . It should be useful for characterizing the interaction of many DNA viruses that associate with the cellular genome , and thus provide a useful tool to characterize the molecular events leading to the initiation of their infections . Further information and requests for resources and reagents should be directed to and will be fulfilled by David Pintel ( pinteld@missouri . edu ) . Cell lines were cultured in 5 percent FBS-containing DMEM media ( 5 percent CO2 and 37 degrees Celsius ) . Murine EL4 cells were cultured in RPMI media with 5 percent FBS . Cell lines are routinely authenticated for mycoplasma contamination , and background levels of DNA damage detected by γ-H2AX . Further information on cell line authentication and parvovirus replication are available at ATCC and published studies ( Tattersall and Bratton , 1983 ) respectively . Male Murine A9 , NIH-3T3 , EL4 and human NB324K cells were propagated and wild-type MVMp and MVMi were produced as previously described ( Adeyemi et al . , 2010 ) . Infection was carried out at a Multiplicity Of Infection ( MOI ) of 5 unless otherwise stated , leading to infection rates of 70–80% as detected by NS1 staining . LentiCRISPRv2 plasmid was obtained from Addgene ( plasmid# 52961 , [Sanjana et al . , 2014] ) and pseudotyped viruses were generated in 1 × 106 293 T cells transfected with 1 µg of LentiCRISPRv2 , 1 µg of HIV Gag/Pol and 1 µg VSV-G proteins using Lipo293D ( SignaGen ) . Supernatant containing the lentivirus was collected at 48 hr post-transfection . Independent preparations of LentiCRISPRv2 lentivirus were used to transduce A9 cells with 750 µl of lentiviral supernatant for 48 hr before selecting cells in 1 µg/ml of puromycin for 10 days . The resulting polyclonal puromycin-selected LentiCRISPRv2 A9 cells were validated for Cas9 expression by western blot , and were utilized for induced DNA break assays ( described below ) . Chromosome Conformation Capture assays were performed using 107 cultured A9 , EL4 , NIH-3T3 and NB-324K cells . Briefly , samples were cross-linked in 2 percent formaldehyde for 10 min , before quenching them in 0 . 125 M glycine . Cells were lysed in NP40 lysis buffer ( 0 . 1% NP40 , NaCl , Tris-HCl ) and the resulting nuclei were resuspended in restriction enzyme buffer ( NEB Buffer 2 . 1 ) . The nuclei were permeabilized in 0 . 3% SDS for an hour , followed by sequestration of SDS in 2% Triton X-100 . The samples were digested in 400U of Hind III restriction enzyme overnight . Digestion was continued with a further 300U of Hind III on the next day , before inactivating the enzyme with 1% SDS at 65°C . SDS was sequestered with 1% Triton X-100 , and 3C chromatin was resuspended in 1 . 15X T4 DNA Ligase Reaction Buffer . 50U of T4 DNA Ligase was added to the samples . Intramolecular ligation was carried out at room temperature for 4 hr , before reversing the crosslinks and digesting protein at 65 degrees C overnight with Proteinase K . 3C DNA was purified by phenol:chloroform:isoamyl alcohol extraction , isopropanol precipitation and finally using a PCR purification kit . The 3C-DNA was eluted in 200 microliters of Buffer EB ( Qiagen ) . Cross-linking efficiencies were measured using Taqman-qPCR assays with primers and probes shown in Supplementary file 1 . Relative crosslinking between two distally located HindIII fragments was determined by the ratio of the novel ligation junction to that of nearest neighbor interaction on the Ercc3 locus , as described previously ( Hagège et al . , 2007 ) . V3C-seq assays were performed with Hind III as the primary restriction enzyme to digest cross-linked MVM infected A9 fibroblast chromatin . The Hind III-digested DNA was intramolecular-ligated using the 3C procedure , before resuspending in Buffer EB ( 100 µl , Qiagen ) . 3C-DNA was secondary-digested with Nla III ( 100U , overnight at 37°C ) , before being heat inactivated and circularized with 100U of T4 DNA Ligase at room temperature overnight in 6 ml of ligation reaction . The V3C samples were precipitated by phenol:chloroform extraction , precipitated in isopropanol , resuspended in Qiagen Buffer EB ( 100 µl ) , and . Inverse PCR was performed on the circularized DNA using primers within the Hind III - Nla III fragments on the MVM genome using inverse PCR primers described in Supplementary file 1 . Inverse PCR products were diluted 1:100 in TE buffer and used as templates for nested inverse PCRs ( described in Supplementary file 1 ) , yielding V3C-seq DNA libraries . Sequencing libraries were prepared using the NEB Ultra Kit , and twelve samples were pooled per run for 75 base-pair single end sequencing using an Illumina Next Seq 500 sequencer . V3C-seq samples were trimmed and aligned to the mouse reference genome ( mm10 build ) using Bowtie2 ( Langmead and Salzberg , 2012 ) . The Biostrings package in RStudio was used to generate a genome-wide map of HindIII restriction fragments for the assignment of reads ( Pagès et al . , 2017 ) . To compare between different timepoints , reads for each fragment were averaged and quantile normalized using preprocessCore package on RStudio ( Bolstad , 2013 ) . For visualization of the V3C-seq data , a running mean was calculated using a window size of five contiguous HindIII fragments ( Medvedovic et al . , 2013 ) . Bioinformatic codes provided in Table 1 . The indicated cells were cross-linked with 1% formaldehyde for 10 mins at room temperature and then quenched with 0 . 125 M glycine . The cells were collected and lysed using a ChIP lysis buffer ( 1% SDS , 10 mM EDTA , 50 mM Tris-HCl , pH 8 , protease inhibitors ) for 20 min on ice . The lysates were sonicated using a Diagenode Bioruptor for 75 cycles ( 30 s on and 30 s off per cycle ) , before being incubated overnight at 4°C with the indicated antibodies bound to Protein A Dynabeads ( Invitrogen ) , in ChIP dilution buffer ( 0 . 01% SDS , 1 . 1% Triton X-100 , 1 . 2 mM EDTA , 16 . 7 mM Tris-HCl pH8 , 167 mM NaCl ) . Samples were washed for 3 min each at 4 degrees Celsius with low salt wash ( 0 . 01% SDS , 1% Triton X-100 , 2 mM EDTA , 20 mM Tris-HCl pH8 , 150 mM NaCl ) , high salt wash ( 0 . 01% SDS , 1% Triton X-100 , 2 mM EDTA , 20 mM Tris-HCl pH8 , 500 mM NaCl ) , lithium chloride wash ( 0 . 25M LiCl , 1% NP40 , 1% DOC , 1 mM EDTA , 10 mM Tris-HCl pH8 ) and twice with TE buffer before being eluted with SDS- elution buffer ( 1% SDS , 0 . 1M Sodium bicarbonate ) . Following elution , the chromatin-antibody-DNA complexes , and the input chromatin were subjected to proteinase K treatment at 65°C overnight . The ChIP DNA was purified using a PCR purification kit ( Qiagen ) , and eluted in 100 ul of Buffer EB ( Qiagen ) . ChIP assays were analyzed by quantitative PCR ( qPCR ) with iTaq universal SYBR green mastermix ( Bio-Rad ) , using primer sets described in Supplementary file 1 , or sequenced as described below . Percent input was calculated as described previously ( Fuller et al . , 2017 ) . Sequencing libraries were generated from ChIP DNA using the NEBNext Ultra II Library Prep Kit for Illumina , and the sonication quality was determined using Agilent Bioanalyser . For ChIP-seq , twelve samples were pooled and sequenced on an Illumina Next Seq 500 using 75 base-pair Single End sequencing . ChIP-seq samples were aligned to the mouse genome ( build mm10 ) using Bowtie2 ( Langmead and Salzberg , 2012 ) . Peaks were called with EPIC analysis software ( using the SICER algorithm ( Zang et al . , 2009 ) according to default parameters . Called-peaks that were shared between replicates were identified using BEDtools software ( Quinlan and Hall , 2010 ) . Comparison between ChIP-seq and V3C-seq peaks were performed using Deeptools package ( Ramírez et al . , 2016 ) . In order to compare the magnitudes of ChIP-seq peaks between different timepoints of MVM infection and mock versus Hydroxyurea treatment , rpm values were calculated ( using Galaxy , [Afgan et al . , 2016] ) on the bedgraph files generated from EPIC , and were quantile normalized using preprocessCore package on RStudio ( Bolstad , 2013 ) . Bioinformatic codes provided in Table 1 . Laser micro-irradiation was performed on 1 million A9 cells cultured on glass bottom dishes ( MatTek Corp . ) infected with MVMp at an MOI of 10 for 18 hr . Cells were sensitized with 2 microliters of Hoechst dye ( ThermoFisher Scientific ) 5 min prior to irradiation . Samples were irradiated using a Leica TCP SP8 confocal microscope with a 405 nm laser using 25% power at 40 Hz frequency for 2 consecutive frames per field-of-view . Regions of interest ( ROIs ) were selected within the nucleus without traversing the nuclear membrane . Samples were processed for immunofluorescense imaging without CSK pre-extraction immediately after micro-irradiation . Stable A9 cells expressing LentiCRISPRv2 were co-transfected with guide RNAs targeting chromosome 9 at 9qE1 ( labelled as TGT ) , or scrambled control guide RNAs ( labelled as CTRL ) and human CD4 expressing vector during parasynchronization . CD4-positive cells were purified using an EasySep CD4+ T Cell Enrichment Kit ( StemCell ) prior to release into complete DMEM media and MVM infection . Infected cells were harvested and processed for ChIP and 3C assays at the indicated timepoints . The MVMp genome and indicated cellular regions were labelled with the DNA FISH-Tag Multicolor Kit ( ThermoFisher ) . Briefly , 1 µg of DNA was labelled with aminoallyl-modified dNTP by nick-translation using the manufacturer’s instructions before being labelled with amine-modified Alexa-Fluor dyes ( AlexaFluor 488 and AlexaFluor 555 ) . The dye combinations were resuspended at equimolar amounts in hybridization buffer ( 50% formamide , 2X SSC , 40% dextran sulfate , 10% Denhardt’s solution ) prior to hybridizing to the sample . Parasynchronized MVMp-infected A9 cells were harvested at the indicated timepoints by pre-extracting with CSK Buffer ( 10 mM PIPES pH 6 . 8 , 100 mM Sodium Chloride , 300 mM Sucrose , 1 mM EGTA , 1 mM Magnesium Chloride ) for 3 min followed by CSK Buffer with 0 . 5% Triton for 3 min . Cells were crosslinked with 4% paraformaldehyde for 10 min at room temperature , before being washed with PBS . The nuclei were dehydrated by sequential treatments with 50% ethanol , 70% ethanol and 100% ethanol for 3 min each . Nuclei were subsequently rehydrated with sequential treatment with 70% ethanol , 50% ethanol and PBS for 3 min each . Cells and nuclei were permeabilized with 0 . 5% Triton X-100 in PBS for 15 min , before being washed with PBS . The samples were treated with 2 μg of RNAse A ( Roche ) in PBS for 1 hr at 37°C . Samples were denatured in 50% formamide ( Ambion ) before being hybridized to the suspended fluorescently labelled probes overnight at 37°C . Samples were washed in 2X SSC with 0 . 1% Tritox X-100 at 37°C followed by three times for 5 min each , followed by 2 washes in 2X SSC at 37°C for 5 min each . Samples were then mounted on slides and imaged on the indicated microscope . For GSD/dSTORM super-resolution imaging , coverslips with adherent immunostained cells were mounted on cavity microscope slides with PBS ( 0 . 01 M , pH 7 . 4 ) imaging buffer containing 100 mM beta-mercaptoethylamine , 10% w/v glucose , 0 . 5 mg/ml glucose oxidase and 40 µg/ml catalase . GSD super-resolution imaging was performed on a Leica SR GSD 3D microscope ( Leica Microsystems , Inc . ) using a 560 nm ( AlexaFluor 555 ) or a 632 nm ( AlexaFluor 647 ) excitation lasers and a 405 nm back-pumping ( activation ) laser . A 160 × 1 . 43 NA oil-immersion objective lens was used for imaging . Two-color GSD images were acquired sequentially with an Andor iXon Ultra 897 EMCCD camera at exposure times 7–8 ms using a QGSD 561 quad filter cube and emission bandpasses 605/45 nm ( AlexaFluor 555 ) and 695/85 nm ( AlexaFluor 647 ) . Approximately 8000 images per channel of a 18 × 18 µm field-of-view were acquired . The coordinates of single molecules were localized in all recorded raw images and high-resolution GSD images were constructed using Leica LAS X software ( version 1 . 9 ) . For confocal imaging , samples were mounted on slides using Pro-Long Diamond anti-fade media with DAPI ( Invitrogen ) . Confocal z-stacks were acquired using a Leica TCP SP8 confocal microscope with 488 nm ( AlexaFluor 488 ) and 552 nm ( AlexaFluor 555 ) excitation lasers and a 100 × 1 . 4 NA objective lens . 3D-FISH images were analyzed using ImageJ . Background noise was filtered out using the Kalman Stack Filter plugin to determine the coordinates ( x , y , z ) of the centers of the foci . The coordinates of the viral and cellular foci were measured using Sync Measure 3D . The 3D-distance was calculated by computing the displacement vector between the two locations as described previously ( Shih and Krangel , 2010 ) . Parasynchronized MVMp-infected A9 cells were harvested at the indicated time points processed as described above till permeabilization with 0 . 5% Triton X-100 in PBS . Samples were blocked with 3% BSA in PBS for 1 hr , incubated with the indicated antibodies for 1 hr , and incubated with the indicated secondary antibodies ( tagged with Alexa Fluor fluorophores ) for 1 hr . Samples were washed and mounted on slides with ProLong Diamond Antifade Mountant with DAPI ( Invitrogen ) . Alkaline Comet Assays were performed using Trevigen Comet Assay kits . Murine A9 fibroblasts were grown on 10 centimeter dishes and mock infected , induced with Doxorubicin ( 200 nM ) for 9 hr , or infected with MVMp at an MOI of 10 for 20 hr , before detaching them from the flask by scraping . Cells were washed with ice-cold PBS and resuspended at a density of 105 cells/ml in ice cold PBS . Cells were combined with molten LM Agarose at 37°C at a ratio of 1:10 and pipetted onto Comet Slides . Slides were placed at 4°C in the dark for 10 mins . Slides were immersed in 4°C Lysis solution for 30–60 min , before placing in Alkaline Unwinding Solution for 1 hr at 4°C in the dark . 850 ml Alkaline Electrophoresis Solution was added to the slide tray , and 21 Volts were applied for 30 min . Slides were immersed in water twice for 5 min each , followed by immersion in 70% ethanol for 5 min . Samples were dried at 37 degrees Celsius for 15 min , and subsequently stained with 100 ul of SYBR Gold for 30 min in the dark . Slides were briefly rinsed in water and completely dried at 37°C . Slides were imaged on a Leica widefield microscope . Cells grown and infected in 60 mm dishes were harvested at the indicated timepoints , followed by lysis in modified RIPA buffer ( 20 mM Tris-HCl pH 7 . 5 , 150 mM NaCl , 10% glycerol , 1% NP-40 , 1% Sodium Deoxycholate , 0 . 1% SDS , 1 mM EDTA , 10 mM trisodium pyrophosphate , 20 mM Sodium Fluoride , 2 mM Sodium Orthovanadate and 1X Protease Inhibitor cocktail ( Sigma ) . Protein concentrations were quantified using Bradford assay and equal amounts of lysates were loaded per well for Western blot analysis . Cells were grown on 25 mm plates and infected at an MOI of 5 . Cells were harvested at the indicated timepoints , pelleted and resuspended in Southern Lysis Buffer . Cells were proteinase K treated overnight at 37°C , and sheared using 25 G X 5/8 inch 1 mL needle-syringe ( BD Biosciences ) . Total DNA content was quantified using Nanodrop , equal amount of DNA loaded per well and electrophoresed on a 1 percent agarose gel . Samples were transferred to a nitrocellulose membrane and hybridized with completely homologous genomic clones . Commercially available antibodies were used for ChIP assays and Immunofluorescence , and are described in the Antibody Table ( Table 2 ) and Key Resources Table . Lenti-CRISPRv2 plasmid was produced by Feng Zhang ( Addgene plasmid 52961 , [Sanjana et al . , 2014] ) . pgRNA-humanized plasmid was produced by Stanley Qi ( Addgene plasmid 44248 , [Qi et al . , 2013] ) . pCMV-CD4 was a gift from Dr . Marc Johnson ( University of Missouri ) . Plasmids and reagents are available upon request . Imaging studies ( 3D-FISH and Immunofluorescense ) were quantified using ImageJ . Background noise was filtered out using the Kalman Stack Filter plugin , and the 3D distance between viral and cellular genome probes were calculated using Sync Measure 3D plugin to calculate the location of the center of mass between the imaged foci . The 3D-distance was calculated by computing the displacement vector between the two locations . The distances between foci were measured for multiple cells in preparations of viral infections , and were statistically analyzed using GraphPad Prism software . Statistical tests were performed using GraphPad Prism for imaging studies and chromosome conformation capture assays . The relevant statistical tests have been indicated in the respective figure legends . The code for bioinformatics analyses used to process V3C-seq and ChIP-seq data have been tabulated below: The V3C-seq and ChIP-seq data generated have been deposited in the Gene Expression Omnibus ( GEO ) under the accession codes GSE112957 .	Viruses are small infectious particles that can only reproduce with the help of a host . Once they are inside their victim , they hijack the cells’ genetic material and reprogram it to become a virus factory that produces more virus particles . Parvoviruses , for example , are among the simplest of viruses and need all resources a cell has to offer to successfully replicate . This process often takes place at so-called replication centers that contain these necessary factors . It was previously thought that parvoviruses set up such centers randomly , and gather the required molecules such as proteins to these sites . However , it was not well understood how they do this . Now , Majumder et al . have developed a new method that enabled them to study in detail how parvoviruses gain access to the resources of the cell they need to initiate and amplify replication . The results show that parvoviruses set up their replication centers at sites on the host DNA that are already rich in proteins needed to repair and then replicate damaged DNA . Some of these sites already exist in the cell’s genetic material as a consequence of naturally occurring processes , but others are created during infection by the virus . These findings may have important implications for how other viruses may establish their replication . Viruses , including parvoviruses , are important pathogens . Like many microbes , viruses can be beneficial for our health and environment . Others , however , can be harmful . A clearer understanding of how viruses establish and amplify an infection may provide new treatment opportunities .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "microbiology", "and", "infectious", "disease" ]	2018	Parvovirus minute virus of mice interacts with sites of cellular DNA damage to establish and amplify its lytic infection
Lrp4 , the muscle receptor for neuronal Agrin , is expressed in the hippocampus and areas involved in cognition . The function of Lrp4 in the brain , however , is unknown , as Lrp4−/− mice fail to form neuromuscular synapses and die at birth . Lrp4−/− mice , rescued for Lrp4 expression selectively in muscle , survive into adulthood and showed profound deficits in cognitive tasks that assess learning and memory . To learn whether synapses form and function aberrantly , we used electrophysiological and anatomical methods to study hippocampal CA3–CA1 synapses . In the absence of Lrp4 , the organization of the hippocampus appeared normal , but the frequency of spontaneous release events and spine density on primary apical dendrites were reduced . CA3 input was unable to adequately depolarize CA1 neurons to induce long-term potentiation . Our studies demonstrate a role for Lrp4 in hippocampal function and suggest that patients with mutations in Lrp4 or auto-antibodies to Lrp4 should be evaluated for neurological deficits . In humans , mutations in genes encoding synaptic organizing complexes have been implicated in numerous and diverse neurological diseases , ranging from congenital myasthenia to autism spectrum disorders ( Sudhof , 2008; Burden et al . , 2013 ) . Lrp4 plays a key role in the formation and maintenance of neuromuscular synapses , as a loss of Lrp4 leads to a failure to form neuromuscular synapses , and mutations in Lrp4 or auto-antibodies to Lrp4 cause congenital myasthenia and myasthenia gravis , respectively ( Shen et al . , 2013; Ohkawara et al . , 2014; Tsivgoulis et al . , 2014 ) . Lrp4 functions bidirectionally at neuromuscular synapses , where it responds to neuronal Agrin , stimulating MuSK , a receptor tyrosine kinase that functions as a master regulator of synapse formation , and functions in a retrograde manner to stimulate differentiation of motor nerve terminals ( Yumoto et al . , 2012 ) . Lrp4 belongs to the low-density lipoprotein receptor ( LDLR ) family , an ancient group of endocytic type 1 , single-pass transmembrane proteins . Although LDLR family members were initially studied for their roles in receptor-mediated endocytosis , multiple other physiological roles have been described . Lrp4 has multifunctional roles in tissues other than the nervous system , including bone homeostasis , limb patterning , kidney formation , and placode development ( Johnson et al . , 2005; Weatherbee et al . , 2006; Ohazama et al . , 2008; Li et al . , 2010; Ahn et al . , 2013 ) . Lrp4 is expressed in the central nervous system ( CNS ) as well as in the peripheral nervous system ( Visel et al . , 2004; Tian et al . , 2006; Weatherbee et al . , 2006; Lein et al . , 2007 ) . Within the CNS , Lrp4 is expressed prominently in the hippocampus , olfactory bulb , cerebellum , and neocortex and present in postsynaptic membranes ( Tian et al . , 2006 ) . The role of Lrp4 in the CNS is not understood , as Lrp4 mutant mice die at birth from neuromuscular and respiratory failure , before synapse formation in the CNS ensues ( De Felipe et al . , 1997; Tian et al . , 2006; Weatherbee et al . , 2006; Kim et al . , 2008; Yumoto et al . , 2012 ) . Previously , we generated mice that lack Lrp4 in all tissues except skeletal muscle and found that muscle-selective expression of Lrp4 ( Lrp4m ) rescued the neuromuscular deficits of Lrp4 mutant mice , allowing the mice to survive as adults ( Gomez and Burden , 2011 ) . To learn whether Lrp4 plays a role in the CNS , we used multiple behavioral paradigms to study the behavior of these muscle-rescued mice . Next , we examined the synaptic transmission and the anatomical organization of inputs onto CA1 hippocampal pyramidal neurons . Our data show that the rescued mice perform poorly in several learning and memory paradigms , demonstrating that Lrp4 has a critical role in the CNS . Moreover , we show that Lrp4 is enriched in postsynaptic membranes from the hippocampus , and our electrophysiological studies demonstrate a dramatic loss in long-term potentiation ( LTP ) , accompanied by a reduction in synapses on apical dendrites of CA1 neurons . Newborn mice , which lack Lrp4 in all tissues except skeletal muscle ( Lrp4−/−; Lrp4m ) , retained the fused digit and appendage defects found in Lrp4 mutant mice . In other respects , the rescued mice appeared indistinguishable from their wild-type littermates ( Figure 1A , inset ) . By three weeks after birth the growth rate of Lrp4−/−; Lrp4m mice began to slow and by 6 weeks the mice were modestly runted ( Figure 1B ) . Nonetheless , Lrp4−/−; Lrp4m mice were fertile and lived a normal lifespan , indicating that Lrp4 is not required in tissues other than muscle for postnatal survival . The macroscopic morphology of the brain from adult Lrp4−/−; Lrp4m mice appeared normal , although brain size , like body mass , was modestly reduced ( Figure 1C , D ) . 10 . 7554/eLife . 04287 . 003Figure 1 . Restoring Lrp4 expression selectively in muscle of Lrp4 mutant mice rescues neonatal lethality . ( A ) Lrp4−/−; Lrp4m mice are fertile and live a normal lifespan . ( B ) The body mass of six-week old Lrp4−/−; Lrp4m mice is reduced by 16% ( wild-type , 21 . 4 ± 0 . 8 g , n = 10; Lrp4−/−; Lrp4m , 18 . 0 ± 0 . 5 g , n = 5 ) . ( C ) The gross morphology of the adult brain is similar in wild-type and Lrp4−/−; Lrp4m mice . ( D ) The size of the adult brain is reduced by 11% in Lrp4−/−; Lrp4m mice ( wild-type , 0 . 45 ± 0 . 005 g , n = 17; Lrp4−/−; Lrp4m , 0 . 4 ± 0 . 008 g , n = 15 ) . ( E , F , G ) The locomotor activity of Lrp4−/−; Lrp4m mice in an open field test was normal as measured by distance traveled ( E ) , mean velocity ( F ) , and maximum velocity ( G ) ( wild-type , n = 17; Lrp4−/−; Lrp4m , n = 15 ) . ( H ) Lrp4−/−; Lrp4m mice showed reduced exploratory behavior ( wild-type , 22 . 3 ± 2 . 2% , n = 29; Lrp4−/−; Lrp4m , 14 . 9 ± 1 . 2% , n = 25 ) . ( I ) Representative heat maps of wild-type and Lrp4−/−; Lrp4m mice during a 30 min open field test . DOI: http://dx . doi . org/10 . 7554/eLife . 04287 . 003 Motor function is required to execute behavioral paradigms , thus we first asked if locomotion was normal in Lrp4−/−; Lrp4m mice using an open-field test . When placed in the open-field arena , Lrp4−/−; Lrp4m mice traveled as far and fast as control animals ( Figure 1E–G ) , indicating that Lrp4−/−; Lrp4m mice do not have obvious motor or skeletal defects . However , Lrp4−/−; Lrp4m mice exhibited different open-field behavior compared to control animals ( Figure 1H , I ) . The heat map in Figure 1I shows that control mice roamed throughout the open-field , whereas Lrp4−/−; Lrp4m mice avoided the center of the arena ( Figure 1H , I ) . Additionally , when suspended by the tail , Lrp4−/−; Lrp4m mice display a stereotyped limb clasping behavior , similar to other mouse models of neurological disorders , including Rett Syndrome ( Figure 2 ) ( Guy et al . , 2001 ) . 10 . 7554/eLife . 04287 . 004Figure 2 . Forelimb and hindlimb clasping in Lrp4−/−; Lrp4m mice . Wild-type mice splay their limbs when suspended by their tail , whereas Lrp4−/−; Lrp4m mice clasp their limbs . DOI: http://dx . doi . org/10 . 7554/eLife . 04287 . 004 To examine cognitive function and associative learning in these animals , we next assessed their behavior with classical fear-conditioning and passive avoidance paradigms ( LeDoux , 2003; Lai and Ip , 2013 ) . In the fear-conditioning assay , mice were trained to associate an auditory cue with a foot shock , which elicited a freezing response . During training , both control and Lrp4−/−; Lrp4m mice responded similarly to the tone and foot shock ( Figure 3A ) , demonstrating that these sensory systems remained intact in the rescued mice . The next day , mice were exposed to the same tone in the absence of the foot shock , and the time spent freezing was measured . As expected , wild-type mice froze in response to the tone alone . In contrast , Lrp4−/−; Lrp4m mice spent much less time freezing , suggesting impaired learning or memory for the tone–shock pairing ( Figure 3A ) . 10 . 7554/eLife . 04287 . 005Figure 3 . Mice lacking Lrp4 in the CNS display defects in learning and memory . ( A ) A schematic representation of the fear-conditioning paradigm . Lrp4−/−; Lrp4m mice exhibit a decrease in freezing behavior , compared to littermate control mice , when presented with an aversive conditioned stimulus ( Lrp4−/−; Lrp4m , n = 12; littermate controls , n = 14 ) . ( B ) A schematic representation of the passive avoidance paradigm . Lrp4−/−; Lrp4m mice were less hesitant to enter a dark chamber associated with an aversive stimulus ( Lrp4−/−; Lrp4m , n = 13; littermate controls , n = 17 ) . ( C ) Lrp4−/−; Lrp4m mice showed spatial learning deficits and reduced cognitive flexibility in the Morris water maze . Both control and Lrp4−/−; Lrp4m mice were able to locate the escape platform during the visible version of the water maze ( Lrp4−/−; Lrp4m , n = 8; littermate controls , n = 9 ) . ( D ) Lrp4−/−; Lrp4m and control mice displayed comparable swimming velocity during the Morris water maze . ( E ) Lrp4−/−; Lrp4m and control mice spent more time searching in the target quadrant region than other quadrants during the probe trial . ( F ) Lrp4−/−; Lrp4m and control mice crossed the platform site with similar frequency during the probe trial . ( G ) Lrp4−/−; Lrp4m mice spent less time in the new target quadrant during reversal training . See also Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 04287 . 005 We next assessed the rescued mice using a passive avoidance paradigm , which exploits an innate preference of mice to avoid a well-lit environment . Mice were placed in a well-lit chamber and freely allowed to enter a dark chamber , where they received a foot shock ( Figure 3B ) . During training , control and Lrp4−/−; Lrp4m mice showed a similar latency to enter the dark chamber . 2 days later , mice were once again placed in a well-lit chamber , and the time to enter the dark chamber was recorded . Control mice were slow to enter the dark chamber after training , whereas Lrp4−/−; Lrp4m mice entered the dark chamber with a shorter latency than control mice , indicating an impaired association of the dark chamber with the foot shock ( Figure 3B ) . Fear conditioning and passive avoidance paradigms involve activity in the hippocampus among other brain areas ( LeDoux , 2003 ) . Because the hippocampus plays a major role in spatial learning , we used the Morris water maze to more specifically test whether Lrp4−/−; Lrp4m mice have spatial learning and memory deficits . Lrp4−/−; Lrp4m mice were trained for several days to associate visible spatial cues with the position of a hidden platform in an opaque pool ( Figure 3C and Figure 4A ) . During this acquisition phase , the latency of control animals to locate the hidden platform decreased ( Figures 3C and 4B ) . Although the latency for Lrp4−/−; Lrp4m mice also decreased during this training period , they consistently took longer to find the platform than control mice ( Figures 3C and 4B ) . After training , the hidden platform was removed and a trial was conducted to assess recall for the position of the hidden platform . Recall was quantitated by measuring the time spent in the target quadrant that previously contained the hidden platform . Lrp4−/−; Lrp4m mice displayed a preference for the correct target quadrant and crossed the former platform location with similar frequency to control mice ( Figures 3D and 4C ) . These findings indicate that while the rescued mice had difficulty learning the position of the hidden platform; however , with repeated training , they were able to learn the position and eventually behave in a manner similar to control mice . 10 . 7554/eLife . 04287 . 006Figure 4 . Mice lacking Lrp4 in the CNS display defects in learning and memory in the Morris water maze . ( A ) A schematic representation of the Morris water maze training protocol . Mice were trained for 5 days to locate a hidden platform . A probe trial was performed on the fifth day , when the platform was removed . The hidden platform was moved to the opposite quadrant during reversal training . A flag was placed on the hidden platform during the visible training phase . Representative trajectories of control and Lrp4−/−; Lrp4m mice during the acquisition ( B ) , probe ( C ) , reversal ( D ) , and visible ( E ) tests of the Morris water maze trial . DOI: http://dx . doi . org/10 . 7554/eLife . 04287 . 006 In order to assess cognitive flexibility of Lrp4−/−; Lrp4m mice , we placed a hidden platform in the quadrant opposite to the original target quadrant and tested the mice . Control mice rapidly learned the new location of the platform , whereas Lrp4−/−; Lrp4m mice showed a significant impairment ( Figures 3C and 4D ) and spent less time than control mice in the new target region ( Figure 3G ) . These data indicate that the rescued mice were slower to extinguish their memory of the original target , learn the position of the new target , or both . Control and Lrp4−/−; Lrp4m mice were similarly able to locate a visible platform and had comparable swimming velocities , indicating that the impaired performance of Lrp4−/−; Lrp4m mice was not due to compromised visual acuity , swimming ability , or motivation to escape the water ( Figure 3C , D , 4E ) . To determine whether the cognitive deficits observed in Lrp4−/−; Lrp4m mice are associated with synaptic dysfunction , we examined synaptic transmission using whole-cell recordings in acute hippocampal slices . We focused on examining excitatory postsynaptic currents ( EPSCs ) in CA1 pyramidal cells evoked by stimulation of CA3 Schaffer collaterals ( SC; Figure 5A ) because of the extensive characterization of CA3–CA1 synapses in many mouse models ( Malenka and Bear , 2004 ) . We first determined whether Lrp4 is critical for synaptic transmission . SCs were briefly stimulated twice with varied inter-stimulus intervals , and EPSCs were recorded . At wild-type synapses , the second stimulus , delivered 50 ms later , elicited a greater postsynaptic response ( Figure 5B , inset ) , possibly due to higher basal levels of residual calcium in nerve terminals following the first stimulus ( Katz and Miledi , 1968; Zucker and Regehr , 2002 ) . Paired-pulse facilitation was similar in control and Lrp4−/−; Lrp4m mice , indicating that Lrp4 is not essential for this form of evoked transmitter release and short-term plasticity ( Figure 5B ) . 10 . 7554/eLife . 04287 . 007Figure 5 . Lrp4 is required for normal synaptic transmission . ( A ) The panel shows the configuration of whole-cell voltage-clamp recordings made from acute hippocampal slices of young , adult mice . Postsynaptic responses in CA1 pyramidal neurons were measured following stimulation of Schaffer collaterals ( SC ) . ( B ) Representative traces from CA1 neurons to paired stimuli show that paired-pulsed facilitation is normal in Lrp4−/−; Lrp4m mice ( wild-type , n = 9; Lrp4−/−; Lrp4m , n = 6 ) . ( C ) Representative traces of spontaneous miniature excitatory postsynaptic currents ( mEPSC ) of wild-type or Lrp4 mutant CA1 neurons . ( D ) mEPSC frequency is reduced in Lrp4 mutant CA1 neurons ( wild-type , n = 12; Lrp4−/−; Lrp4m , n = 12 . ( E , F ) mEPSC ampitudes of Lrp4 mutant CA1 neurons are comparable to wild-type ( wild-type , n = 12; Lrp4−/−; Lrp4m , n = 12 ) . ( G , H ) Representative traces from individual CA1 neurons following a TBS delivered to neurons , which were voltage-clamped at −70 mV , show LTP from wild-type but not Lrp4 mutant neurons ( upper panel ) . Excitation , measured as the amplitude of an EPSC ( pA ) , is shown below , and input resistance ( Ri ) is shown in the bottom panel . There is considerable variability in the baseline EPSC amplitudes in slices from wild-type and Lrp4−/−; Lrp4m mice . However , there was no significant difference in the baseline EPSC amplitudes between wild-type and mutant mice ( wild-type , 52 . 6 ± 9 . 7 pA , n = 21; Lrp4−/−; Lrp4m , 44 . 9 ± 4 . 9 pA , n = 27 , p = 0 . 44 ) . Further , there was not a significant correlation between the magnitude of LTP and initial synaptic strength ( wild-type , R2 = 0 . 05 , n = 21 , p = 0 . 35; Lrp4−/−; Lrp4m , R2 = 0 . 09 , n = 21 , p = 0 . 13 ) . ( I ) Representative traces from individual CA1 neurons during TBS show a reduction in the integral of the summed EPCSs recorded during a single TBS in Lrp4-deficient neurons , top panel . Induction volume , quantified as the integral of the postsynaptic response , is shown in the bottom panel ( wild-type , n = 21; Lrp4−/−; Lrp4m; n = 32 ) ( J , K ) Representative responses from individual CA1 neurons following a TBS delivered to neurons , which were voltage-clamped at 0 mV , show a restoration of LTP in Lrp4 mutant neurons ( upper panel ) . Excitation , measured as the amplitude of an EPSC ( pA ) , is shown in the middle panel , and input resistance ( Ri ) is shown in the bottom panel . ( L ) Averaged data from control cells ( n = 21 ) and Lrp4 mutant ( n = 32 ) neurons shows that the response from wild-type neurons is potentiated 1 . 5-fold , whereas Lrp4 mutant CA1 neurons potentiate little , if at all 30 min following a TBS delivered at Vc = −70 mV . ( M ) Depolarization of Lrp4 mutant CA1 neurons during TBS restores LTP ( wild-type , n = 11; Lrp4−/−; Lrp4m , n = 9 ) . ( N ) Quantitation of potentiation at 10-20 min after TBS demonstrates a lack of LTP in Lrp4 mutant CA1 neurons , which is restored upon depolarization ( Vc = −70 mV: wild-type , n = 21; Lrp4−/−; Lrp4m , n = 32; Vc = 0 mV: wild-type , n = 11; Lrp4−/−; Lrp4m , n = 9 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04287 . 007 Next , we examined basal activity by measuring spontaneous miniature excitatory postsynaptic currents ( mEPSC ) in CA1 neurons . mEPSC frequency was reduced twofold in Lrp4 mutant neurons ( Figure 5C , D ) , although the mEPSC amplitude was normal ( Figure 5C , E , F ) . The lower mEPSC frequency could be due to a reduced probability of vesicle fusion or a reduction in the number of synapses on CA1 neurons . To determine whether Lrp4 is essential for induction or expression of LTP at CA3–CA1 synapses , where long-term synaptic plasticity is known to require postsynaptic depolarization and activation of postsynaptic NMDA receptors ( Malenka and Nicoll , 1999 ) , theta burst stimulation ( TBS ) was used to strongly activate SCs . EPSCs were recorded before and after repetitive TBS . TBS induced robust LTP from wild-type CA1 neurons ( Figure 5G , L , N ) , whereas TBS failed to induce LTP in Lrp4−/− CA1 neurons ( Figure 5H , M , N ) . During the induction procedure the amplitudes of TBS-evoked EPSCs were reduced in Lrp4−/− CA1 neurons compared to wild-type ( Figure 5I ) . This dramatic loss of LTP demonstrates that Lrp4 is critical for a form of synaptic plasticity that has been linked to learning and memory . We hypothesized that the smaller EPSCs during TBS in Lrp4−/− CA1 neurons prevented LTP induction in these cells . Induction of LTP in CA1 neurons requires that SC-released glutamate binds to AMPA receptors to depolarize postsynaptic neurons to a level sufficient to release the Mg2+-block of NMDA receptors and drive Ca2+ influx into postsynaptic compartments ( Nowak et al . , 1984; Collingridge et al . , 1988 ) . To determine whether CA3 Schaeffer collaterals are unable to adequately depolarize CA1 neurons during TBS , we paired repetitive TBS with direct postsynaptic depolarization to 0 mV during LTP induction . We found that this brief period of depolarization of CA1 neurons was sufficient to restore normal TBS-induced LTP ( Figure 5J , K , M , N ) . Importantly , these data show that expression of LTP in Lrp4−/− CA1 neurons is intact as long as CA1 neurons are adequately depolarized during TBS . Thus , CA3 SCs appear unable to depolarize CA1 neurons to a level required to recruit NMDA receptors . Consistent with the idea that AMPA and NMDA receptors are available to participate in LTP , AMPA , and NMDA receptors are expressed at normal levels in synaptosomes isolated from Lrp4−/−; Lrp4m hippocampus ( Figure 6C , D ) . 10 . 7554/eLife . 04287 . 008Figure 6 . Lrp4 is enriched in synaptic membranes . ( A ) Quantitative analysis of proteins in hippocampal lysates from wild-type and Lrp4−/−; Lrp4m mice . The expression level of each protein , normalized to actin , was determined and assigned a value of 1 . 0 in wild-type mice . The graph shows the ratio of values in Lrp4−/−; Lrp4m mice compared to wild-type mice . Expression of most proteins is not dependent upon Lrp4 , but expression of the synaptic vesicle-associated protein , Synaptophysin , was modestly elevated ( 30% ) and expression of the GABAARγ2 subunit was modestly decreased ( 20% ) in Lrp4 mutant hippocampi . ( B ) Lrp4 expression is detected in cultured hippocampal neurons , grown in cell culture for 21 days , and in hippocampal tissue from wild-type but not from Lrp4−/−; Lrp4m mice . ( C ) Lrp4 co-isolates with synaptosomes , the pH 6-solubilzed fraction , which is enriched for Synaptophysin , a presynaptic marker , and the postsynaptic density ( PSD ) fraction , which is highly enriched for NR1 . ( D ) The expression levels of presynaptic and postsynaptic proteins present in the synaptosomal fraction ( C ) are not altered in Lrp4−/−; Lrp4m mutant hippocampi . DOI: http://dx . doi . org/10 . 7554/eLife . 04287 . 008 However , we also considered the possibility that in the absence of Lrp4 , inhibition on CA1 neurons may have been strengthened , and depolarization-induced suppression of inhibition ( DSI ) may have removed the enhanced inhibition , unmasking LTP ( Pitler and Alger , 1994; Wilson et al . , 2001 ) . We therefore directly measured inhibition to determine whether it was strengthened in Lrp4−/−; Lrp4m mice ( Figure 7 ) . Contrary to this notion , we found that inhibition was modestly reduced at low stimulation intensities and unchanged at higher stimulation intensities in Lrp4−/−; Lrp4m mice ( Figure 7G ) . Because excitation was reduced while inhibition remained largely unaffected ( Figures 5 , 7 ) , the E/I ratio at CA1 synapses was diminished ( Figure 7H ) . Together , these data are inconsistent with the idea that enhanced inhibition is responsible for a failure to elicit LTP and instead favor the idea that a failure of presynaptic input to adequately depolarize CA1 neurons underlies the LTP deficit . 10 . 7554/eLife . 04287 . 009Figure 7 . The strength of inhibition is unchanged in CA1 neurons from Lrp4−/−; Lrp4m mice . ( A ) Representative traces of spontaneous miniature inhibitory postsynaptic currents ( mIPSC ) from CA1 neurons from wild-type or Lrp4−/−; Lrp4m mice , which were voltage-clamped at 0 mV . ( B ) The mIPSC frequency is similar in Lrp4 mutant and wild-type CA1 neurons ( wild-type , 0 . 017 ± 0 . 005 /sec , n = 7; Lrp4−/−; Lrp4m , 0 . 026 ± 0 . 009 /sec , n = 6 ) . ( C , D ) mIPSC amplitudes are increased in CA1 neurons from Lrp4−/−; Lrp4m mice ( wild-type , 25 . 4 ± 1 . 6 pA , n = 7; Lrp4−/−; Lrp4m , 34 . 7 ± 3 . 7 pA , n = 6 ) . ( E ) Representative traces of excitation ( black ) and inhibition ( blue ) at varied stimulus intensities . ( F ) Excitation is decreased in CA1 neurons from Lrp4−/−; Lrp4m mice . ( G ) Inhibition is similar in CA1 neurons from wild-type and Lrp4−/−; Lrp4m mice . ( H ) The E/I ratio is decreased in CA1 neurons from Lrp4−/−; Lrp4m mice . DOI: http://dx . doi . org/10 . 7554/eLife . 04287 . 009 To assess whether the defects in synaptic transmission and failure of LTP induction were due to disorganization of hippocampal synaptic circuitry , we stained hippocampal slices from adult Lrp4−/−; Lrp4m hippocampus with probes that label nuclei , nerve endings , and excitatory postsynaptic membranes . The distribution of DAPI-stained nuclei and NeuN , a neuronal transcription factor , were comparable in slices from wild-type and Lrp4−/−; Lrp4m mice ( Figure 8A ) . Additionally , the distributions of the presynaptic marker Synapsin and the excitatory postsynaptic marker PSD95 were similar in sections from wild-type and rescued mice ( Figure 8A ) . Together , these data indicate that the defects in synaptic transmission and plasticity in Lrp4−/−; Lrp4m mice were not accompanied by a gross morphological disorganization of the hippocampus . 10 . 7554/eLife . 04287 . 010Figure 8 . A loss of Lrp4 decreases spine density in primary apical dendrites of CA1 neurons . ( A ) The organization of the hippocampus appears normal as assessed by staining sections of the adult hippocampus for DNA ( DAPI ) , neuronal nuclei ( NeuN ) , presynaptic terminals ( synapsin ) or excitatory postsynaptic membranes ( PSD95 ) . ( B ) Representative images of dendrites of Thy1-YFP labeled CA1 Lrp4 mutant pyramidal neurons near ( basal and apical ) and far ( oblique apical and tufts ) from the cell body . Scale bar: 5 μm . ( C ) The spine density is reduced selectively at primary apical dendrites ( basal: wild-type , n = 45; Lrp4−/−; Lrp4m , n = 45; primary apical: wild-type , n = 43; Lrp4−/−; Lrp4m , n = 46; oblique apical: wild-type , n = 42; Lrp4−/−; Lrp4m , n = 46; tufts: wild-type , n = 42; Lrp4−/−; Lrp4m , n = 45 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 04287 . 010 Dendritic spines contain the majority of excitatory synapses on CA1 pyramidal neurons , and changes in spine density lead to alterations in mEPSC frequency and are associated with aberrations in synaptic plasticity ( Sala et al . , 2001; Tada and Sheng , 2006 ) . The inputs to CA1 pyramidal neurons are topographically organized at stereotyped positions along their apical–basal arborization ( Spruston , 2008 ) . Thus , we generated mice that are mutant for Lrp4 , carry the Lrp4m transgene that restores Lrp4 expression in muscle , and also carry the Thy1::YFP-H transgene , which sparsely labels neurons , to examine spine density along the apical–basal axis ( Feng et al . , 2000 ) . The spine density at several locations ( basal , oblique apical , and tufts ) was similar in Lrp4−/−; Lrp4m and wild-type mice ( Figure 8B , C ) . However , the density of spines on primary apical dendrites on CA1 neurons was reduced ( ∼20% ) in Lrp4−/−; Lrp4m mice ( Figure 8B , C ) . This decrease in spine density in CA1 neurons of Lrp4−/−; Lrp4m mice is consistent with the reduction in mEPSC frequency and impaired synaptic plasticity . During neuromuscular synapse formation , Lrp4 organizes synaptic differentiation by bidirectional signaling from the postsynaptic membrane ( Gomez and Burden , 2011 ) . To determine whether Lrp4 is enriched at synapses in the hippocampus we fractionated membranes from wild-type and Lrp4 mutant hippocampus . We found that Lrp4 co-isolates with synaptosomes and is enriched in postsynaptic membranes , containing NMDA receptors , as well as presynaptic membranes , containing Synaptophysin ( Figure 6 ) . Defects in LTP and changes in spine density are often associated with changes in the activity of NMDA receptors and/or AMPA receptors , two components critical for synaptic plasticity ( Zoghbi and Bear , 2012; Ebert and Greenberg , 2013 ) . To determine whether a loss of Lrp4 alters the expression of synaptic components , we measured the expression of presynaptic and postsynaptic proteins from adult hippocampal lysates ( Figure 6 ) . Wild-type and Lrp4−/− hippocampi expressed similar levels of NMDAR subunits , AMPAR subunits , synaptic scaffolding proteins , and cell-adhesion organizing molecules ( Figure 6 ) . These data indicate that the defects in synaptic transmission and cognition are not caused by a change in expression of key synaptic proteins . Our study reveals that Lrp4 has a critical role in hippocampal function . Specifically , our results indicate that defects in synaptic transmission and postsynaptic integration may contribute to deficits in long-term plasticity , learning , and memory . Stimulation of CA3 inputs fails to induce LTP in Lrp4−/− CA1 neurons . Importantly , direct depolarization of Lrp4−/− CA1 neurons during TBS can rescue LTP . Thus , CA3–CA1 synapses in Lrp4 mutant neurons have the capacity to express LTP . Because direct stimulation by-passes the normal synaptic mechanisms for depolarizing CA1 neurons , the absence of LTP in Lrp4−/− CA1 neurons may be due to a failure of SC inputs to adequately depolarize CA1 neurons and remove the Mg++-dependent block of NMDA receptors . The reduction in the number of CA3–CA1 synapses on apical dendrites is consistent with this view . We do not yet know when and where Lrp4 is required for normal CNS function . Further studies will be required to learn whether Lrp4 is required for neurogenesis , early steps in synapse formation , later stages in synaptic differentiation , and the construction and function of Lrp4-dependent circuits that are important for cognition . It seems likely that the loss of neuronal Lrp4 is responsible for the changes in cognition and synaptic function described here . Consistent with this idea , Lrp4 is strongly expressed in the dentate gyrus and CA fields of the hippocampus , as well as in the olfactory bulb , and cerebral cortex ( Tian et al . , 2006; Lein et al . , 2007 ) . Moreover , Lrp4 co-fractionates with the postsynaptic density ( Tian et al . , 2006 ) ( Figure 6C ) . Further , Lrp4 is expressed on the cell surface of cultured cortical neurons ( Tian et al . , 2006 ) . Nonetheless , it may also be the case that other cell types , including glia , may express Lrp4 and contribute to the cognitive deficits found in muscle-rescued Lrp4 mutant mice . Because glia are known to regulate the efficiency of synaptic transmission ( Eroglu and Barres , 2010 ) , a potential defect in glia–neuron signaling could contribute to a failure of synaptic transmission , postsynaptic integration , and/or synapse formation on CA1 apical dendrites . Lrp4-deficient mice rescued for neuromuscular synapses appeared to have a range of cognitive defects . Mice displayed anxiety-like behavior in the open-field test and perseverative behavior in the Morris water maze . Stereotypic and restricted repetitive behaviors are symptoms often observed in patients with autism ( Lord et al . , 2000 ) and in mouse models of autism spectrum disorders ( Banerjee et al . , 2014 ) . Accordingly , it will be interesting to explore whether Lrp4-deficient mice are predisposed to additional autism-like phenotypes such as altered social behavior , hyperactivity , and epilepsy . Mutations in Lrp4 are responsible for Cenani–Lenz syndrome , characterized by bone abnormalities and fusions in hand , limb , and other bones ( Karner et al . , 2010; Li et al . , 2010; Kariminejad et al . , 2013; Khan et al . , 2013 ) . Lrp4 binds , sequesters , and presents negative regulators of Wnt- and BMP-signaling , such as Dickhopf , Sclerostin , and Wise ( Ohazama et al . , 2008; Choi et al . , 2009; Ahn et al . , 2013 ) . In certain instances , Cenani–Lenz syndrome is caused by mutations that prevent Lrp4 from interacting with these negative regulators , leading to excessive Wnt and/or BMP signaling ( Leupin et al . , 2011 ) . Mutations in Lrp4 that reduce Agrin–Lrp4–MuSK signaling , without perturbing Wnt signaling , cause a neuromuscular disease , termed congenital myasthenia ( Ohkawara et al . , 2014 ) . Moreover , auto-antibodies to Lrp4 are responsible for one form of myasthenia gravis ( Higuchi et al . , 2011; Zisimopoulou et al . , 2014 ) . It is currently unclear whether patients with Cenani–Lenz syndrome , Lrp4 congenital myasthenia , or auto-immune Lrp4 myasthenia gravis have cognitive deficits or neurological complications . Given our data showing that Lrp4 has an important role in cognition in mice , it will be interesting and important to evaluate the neurological status of these patients . Mice that are null for Lrp4 , or carry muscle-specific::Lrp4 ( Lrp4m ) or Thy1::YFP-H transgenes , have been described previously ( Feng et al . , 2000; Weatherbee et al . , 2006; Gomez and Burden , 2011 ) . Procedures were approved by the New York University School of Medicine Institutional Animal Care and Use Committee ( Protocol 140406-01 ) . Mice were placed in an open field box ( 40 × 40 × 30 cm ) for 30 min , and movement was recorded and analyzed with ANY-maze video tracking software ( Stoelting , Wood Dale , IL ) . Open fields were thoroughly washed with water and ethanol between sessions . Mice were trained and tested in a sound-attenuated cage using the FreezeFrame system ( Coulbourn Instruments , Whitehall , PA ) , and behavior was recorded using low-light video cameras . Stimulus presentation was automated using Actimetrics FreezeFrame software ( version 2 . 2; Coulbourn Instruments ) . Test cages were equipped with stainless-steel shocking grids , which were connected to a precision feedback current-regulated shocker ( Coulbourn Instruments ) . Mice were allowed to roam for 2 min , without a shock , on a stainless-steel shocking grid connected to a precision feedback current-regulated shocker in ethanol-scented cages ( Coulbourn Instruments ) . Fear conditioning was established by three auditory tone ( 30 s , 4000 Hz , 80 dB ) /foot shock ( 2 s , 0 . 5 mA ) pairings , separated by 15 s without stimuli . 2 min after conditioning , mice were returned to their home cages . 24 hr after training , the mice were placed in a 1% Pine-Sol-scented cage on non-shocking grids with a different texture than the stainless-steel shocking grids used during training . Following 2 min without an auditory stimulus , mice were presented with the training tone ( 4000 Hz , 80 dB ) for 2 min . The percent of time spent in a frozen stature , before and after training , was measured . All equipments were thoroughly cleaned with detergent and water between sessions . Mice were trained and tested in a two-chamber passive avoidance cage , in which one chamber was permanently darkened and separated from the lit chamber by a sliding door . The darkened chamber was equipped with a stainless-steel shocking grid connected to a precision feedback current-regulated shocker ( Coulbourn Instruments ) . Equipment control was automated with Graphic State ( Coulbourn Instruments ) . During training , mice were placed in the brightly-illuminated chamber and allowed to move freely for 1 min . Following this period , the door to the dark chamber was opened , and mice were allowed to move into the dark chamber . Latency to enter the dark chamber was recorded . Once the mice had completely entered the dark chamber , the door shut and a 2 s , 0 . 5 mA shock was delivered . After 10 s in the dark chamber , mice were returned to their home cage . During the test period , 48 hr after training , the training protocol was repeated , but the mice did not receive a foot shock upon entry into the dark chamber . Latency to enter the dark chamber was recorded . All equipment was thoroughly cleaned with detergent and water between sessions . Mice were given four trials ( 60 s/trial; swim-start position randomized ) each day to find a hidden platform in a circular pool of water , rendered opaque with white tempera paint , using visual cues placed outside the pool . The trajectories of mice were recorded and analyzed with Ethovision XT software ( Noldus , Attleboro , MA ) . The time required to find the escape platform ( escape latency ) was measured . At the end of the fifth day , the platform was removed and a 1 min probe trial was conducted to measure the percent of time spent in each quadrant as well as the frequency which mice crossed the former location of the platform . For reversal training , the hidden platform was moved to the opposite quadrant , and mice were given four trials ( 60 s/trial; swim-start position randomized ) for 3 days to locate the new platform position . Escape latencies from four trials each day were recorded . To control for motivation to escape the water , visual acuity , and swimming ability mice were trained on a visible platform for 2 days ( 60 s/trial; swim-start position randomized ) . Coronal hippocampal slices ( 350 μm ) were prepared from one hemisphere of age-matched mice ( P21–P34 ) anesthetized with intraperitoneal injection of ketamine/xylazine ( ketamine , 100 mg/kg; xylazine , 10 mg/kg ) . Slices were cut with a vibratome ( VT1200S , Leica , Buffalo Grove , IL ) and placed in ice-cold oxygenated ( 95% O2/5% CO2 ) dissection buffer , which was ( in mM ) : 75 sucrose , 87 NaCl , 2 . 5 KCl , 1 . 25 NaH2PO4 , 0 . 5 CaCl2 , 7 MgCl2 , 25 NaHCO3 , 1 . 2 ascorbic acid , and 10 dextrose , pH 7 . 4 . After approximately 30 min , the solution was gradually warmed to room temperature . Slices were then transferred to artificial cerebrospinal fluid ( ACSF ) , which was ( in mM ) : 124 NaCl , 2 . 5 KCl , 1 . 25 NaH2PO4 , 2 . 5 CaCl2 , 2 MgSO4 , 26 NaHCO3 , 10 dextrose , and 4 sucrose , and incubated at room temperature for at least 30 min to allow for recovery . Slices were then transferred to the recording chamber and perfused ( 2 . 0–2 . 5 ml/min ) with oxygenated ACSF at 32°C with a TC-344B in-line solution heater and controller ( Warner Instruments , Hamden , CT ) . Somatic whole-cell recordings were made from CA1 pyramidal hippocampal neurons , which were voltage clamped with an Axoclamp 2B amplifier ( Molecular Devices , Sunnyvale , CA ) and imaged using infrared-differential interference contrast video microscopy , digitized by Digidata 1440a ( Molecular Devices ) . Patch pipettes ( 4–8 MΩ ) were filled with ( in mM ) : 125 Cs-gluconate , 2 CsCl , 5 TEA , 4 ATP , 0 . 3 GTP , 10 phosphocreatine , 10 HEPES , 0 . 5 EGTA , and 3 . 5 QX-314 . Data were filtered at 2 kHz , digitized at 10 kHz , and analyzed with Clampfit 10 ( Molecular Devices ) . SC afferents were stimulated with a small glass bipolar electrode ( S88 Stimulator , Grass Instruments , Warwick , RI ) . Paired-pulse facilitation was induced with two stimuli of equal intensity presented at variable interstimulus intervals , ranging from 10 ms to 1 s , and quantified as the ratio of second to first EPSC . Once a baseline for synaptic transmission was stable for 10 min , LTP was induced with a TBS . TBS consisted of four trains , separated by 20 s intervals . Each train was comprised of ten bursts at 5 Hz , and each burst included 4 stimuli at 100 Hz . During TBS , CA1 neurons were held either at the resting potential ( −70 mV ) or depolarized to 0 mV . We computed the magnitude of LTP as the average synaptic strength 10–20 min after pairing , normalized by the average synaptic strength before pairing ( Feldman , 2000; Froemke et al . , 2005; Gambino et al . , 2014 ) . During TBS , we integrated four consecutive EPSCs and termed this value the TBS-evoked EPSC; for each cell , all TBS-evoked EPSCs were averaged , and the mean value was determined ( Figure 3I ) . We measured inhibitory postsynaptic currents by voltage-clamping CA1 neurons to 0 mV , near the reversal potential for excitation . Data are presented as mean ± SEM . p values are derived from unpaired , two-tailed Student's t-tests or two-way analysis of variance ( ns = not significant , p < 0 . 05 , p < 0 . 01 , p < 0 . 001 , **p < 0 . 0001 ) . Hippocampi were isolated in cold PBS , flash frozen in liquid nitrogen , and stored at −80°C . Brain lysates , synaptosomes , and PSD fractions were prepared as described previously ( Phillips et al . , 2001; Jordan et al . , 2004 ) . An equal amount of protein ( 10 μg ) from each fraction was separated by SDS-PAGE . The following antibodies were diluted in TBST ( 0 . 2% Tween-20 ) , with 2% BSA: Lrp4 ( 1:2500; Neuromab , N207/27 , Davis , CA ) , Synaptophysin ( 1:20 , 000; Life Technologies , Grand Island , NY ) , PSD95 ( 1:1000; Neuromab , K28/43 ) , GluA1 ( 1:1000; Millipore , Billerica , MA ) , GluA2/3 ( 1:500; Millipore ) , pan-Shank ( 1:1000; Santa Cruz Biotechnology , C-20 , Dallas , TX ) , NR1 ( 1:1000; Neuromab , N308/48 ) , GABAAR γ2 ( 1:1000; PhosphoSolutions , Aurora , CO ) , NeuN ( 1:1000; Millipore ) , Gephryin ( 1:1000; SYnaptic Systems , Goettingen , Germany ) , actin ( 1:2000; Sigma , AC74 , St . Louis , MO ) , N-cadherin ( 1:1000; BD Transduction Laboratories , 610920 , San Jose , CA ) , PTPR σ ( 1:500; Protein Tech Group , Chicago , IL ) , GFAP ( 1:1000; Sigma ) , β-tubulin III ( 1:6000; SYnaptic SYstems ) . The following antibodies were gifts from P . Scheiffele: Neuroligin 1 ( 1:2000 ) , Neuroligin 2 ( 1:2000 ) , Neuroligin 3A ( 1:2000 ) , and pan-Neuroligin ( 1:3000 ) . Deeply anesthetized ( ketamine , 100 mg/kg; xylazine , 10 mg/kg , i . p . ) mice were transcardially perfused briefly with phosphate buffer saline ( PBS ) followed by 4% paraformaldehyde in PBS . Coronal slices ( 40 μm ) were blocked and permeabilized in PBS with 2% normal goat serum and 0 . 2% Triton X-100 , followed by overnight incubation at 4°C with primary antibodies . The sections were subsequently washed in PBS and incubated with secondary antibody . Dissociated primary hippocampal neuron cultures were prepared from embryonic day 18 mouse embryos , as described previously ( Osten et al . , 1998 ) . Neurons were plated at a density of 1 × 106 cells in poly-L-lysine coated 60-mm tissue culture dishes and grown in Neurobasal Medium , supplemented with B-27 ( Life Technologies ) . 3 days after plating cells , 2 μM Ara-C was added to the medium to minimize growth of dividing cells . The medium , including Ara-C , was replaced once per week .	LRP4 is a muscle protein that is found in the hippocampus , a region of the brain that controls cognitive processes such as learning and memory . However , we know very little about what exactly LRP4 does in the hippocampus , and how it affects learning and memory . A standard way to figure out what a protein does is to study mice that have been genetically modified so that they cannot produce that protein . However , deleting the gene for LRP4 leads to muscle problems that kill these mutant mice at birth . To get around this problem , Gomez et al . have developed a method to restore the production of LRP4 in the muscles of mutant mice but not in their brains . These mutant mice were then subjected to a battery of tests to measure their ability to learn and recall new memories . These tests showed that LRP4 must be present in the brain , otherwise learning and memory are impaired . Gomez et al . also explored a process known as long-term potentiation . This process , which involves strengthening the functional connections between neurons , is believed to be essential for learning and other cognitive process . Gomez et al . demonstrated that long-term potentiation was disrupted by the lack of LRP4 . Further experiments are needed to work out how LRP4 controls the learning process in the hippocampus and to explore the connection between LRP4 and various neuromuscular and neurological diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	Synaptic plasticity and cognitive function are disrupted in the absence of Lrp4
Extensive evidence suggests that people use base rate information inconsistently in decision making . A classic example is the inverse base rate effect ( IBRE ) , whereby participants classify ambiguous stimuli sharing features of both common and rare categories as members of the rare category . Computational models of the IBRE have posited that it arises either from associative similarity-based mechanisms or from dissimilarity-based processes that may depend on higher-level inference . Here we develop a hybrid model , which posits that similarity- and dissimilarity-based evidence both contribute to the IBRE , and test it using functional magnetic resonance imaging data collected from human subjects completing an IBRE task . Consistent with our model , multivoxel pattern analysis reveals that activation patterns on ambiguous test trials contain information consistent with dissimilarity-based processing . Further , trial-by-trial activation in left rostrolateral prefrontal cortex tracks model-based predictions for dissimilarity-based processing , consistent with theories positing a role for higher-level symbolic processing in the IBRE . Does this patient have influenza or Ebola virus ? Categorization is a fundamental process that underlies many important decisions . Categories , such as viruses , often have different relative frequencies or base rates . Influenza , for example , is very common and infects millions of people worldwide each year , whereas Ebola virus tends to have infection rates that are orders of magnitude lower . One critical question is how people use such base rate information when making categorization decisions . Research so far has suggested that people tend to be , at best , inconsistent in their use of base rate information . In both realistic studies with medical professionals and artificial categorization tasks in the lab , when confronted with examples that share characteristics with both rare and common categories , people show a tendency to predict the rare category much more often than the base rates would suggest ( Tversky and Kahneman , 1974; Casscells et al . , 1978; Bravata , 2000 ) . In an extreme case , known as the inverse base rate effect ( IBRE ) , people may even predict rare categories as more likely than common ones ( Medin and Edelson , 1988 ) . For example , in an IBRE context , a patient presenting with cough ( a characteristic feature of influenza ) and unexplained bleeding ( a characteristic feature of Ebola ) , may be more likely to be diagnosed with Ebola than influenza . The mechanisms that lead to base rate neglect are currently undetermined at both the cognitive and neural levels . Computationally , according to influential work with similarity-based categorization models ( Medin and Edelson , 1988; Kruschke , 1996 , Kruschke , 2001 ) , the IBRE arises from differential selective attention to features for common and rare categories . Specifically , participants learn to attend more strongly to features of rare categories , making ambiguous cases seem more similar to rare categories and thus more likely to be rare category members . In terms of the flu example , participants may attend more to the unexplained bleeding feature of the rarer Ebola virus category , and thus predict Ebola when confronted with a patient with both features . Similarity-based category learning models have strong support in the neurobiological category learning literature . Model-based predictions for how similar items are to stored category representations have been shown to correlate with activation in the medial temporal lobes ( MTL; Davis et al . , 2012a , Davis et al . , 2012b ) . Moreover , at a finer-grained level , multivoxel activation patterns in the MTL have been shown to contain information associated with higher-order similarity relationships between category members anticipated by similarity-based models ( Davis and Poldrack , 2014 ) , including those predicted by differences in selective attention ( Mack et al . , 2016 ) . The dorsolateral prefrontal cortex ( dlPFC ) tends to track predictions of choice uncertainty from similarity-based models , whereas ventromedial PFC ( vmPFC ) tends to track estimates of high choice accuracy or model-confidence ( Davis et al . , 2017 ) . Despite the strong cognitive and neural evidence for similarity-based models , it remains an open question whether they provide a complete account of IBRE-like phenomena . One alternative proposition is that people’s choice of rare categories when confronted with conflicting information may stem from reliance on dissimilarity processes , either solely , or in addition to similarity-based processes . According to theories that focus on dissimilarity-based processes , people build strong expectations of the common category; thus they view items containing features inconsistent with these expectations as more likely to be members of the rare category ( Juslin et al . , 2001; Winman et al . , 2005 ) . For example , a doctor may have seen thousands of cases of flu , none with unexplained bleeding , and thus rule out influenza and choose Ebola virus based on these expectations . In these cases , it is dissimilarity to members of the common ( unchosen ) category that drives choice , rather than the similarity to rare ( chosen ) category members per se . Formal models positing dissimilarity processes have so far been explicitly dual-process oriented . For example , ELMO , a computational model that incorporates a choice elimination decision based on dissimilarity , argues that such elimination depends on explicit reasoning processes that are separate from similarity-based processes that arise in other trials ( Juslin et al . , 2001 ) . In the present study , we propose a new account based on a recently proposed dissimilarity-based extension of the generalized context model , the dissGCM ( Stewart and Morin , 2007 ) . This account uses the same basic similarity computations as standard similarity-based models ( e . g . Nosofsky , 1986 ) , but allows similarities and dissimilarities to stored exemplars to be used as evidence for a category . In terms of the above example , dissimilarity to influenza can be used as evidence for Ebola ( and vice versa ) . As specified computationally , the dissGCM is agnostic about whether using dissimilarity-based evidence constitutes a different cognitive or neurobiological mechanism from using similarity-based evidence . On one hand , the dissGCM has no fundamentally different computations from a basic similarity process; as detailed below , dissimilarity is a simple transformation of similarity . On the other hand , it is possible that dissimilarity processes require manipulation of similarity relationships between category representations in a more symbolic or abstract manner , as anticipated by previous dissimilarity theories . Given that the dissGCM makes separable estimates for the relative contributions of similarity- and dissimilarity-based evidence to choice in a given trial , these model predictions can be used to explicitly test whether employing dissimilarity-based evidence against unchosen categories engages brain regions beyond those associated with the use of similarity-based evidence , consistent with dual-process accounts of IBRE . Specifically , we can test whether regions that are known to be critical for using higher-level abstract rules track dissGCM’s predicted trial-by-trial used of dissimilarity-based evidence , and whether these regions diverge from those typically found to track estimates of similarity-based evidence . Higher-level cognitive control mechanisms are thought to depend on a hierarchy of abstraction in the lateral PFC along the rostral-caudal axis ( Badre and D'Esposito , 2007; Badre and D'Esposito , 2009 ) . At the apex of this hierarchy is the rostrolateral PFC ( rlPFC ) , a region often implicated in tasks that require people to generalize across abstract , symbolic representations . For example , relational reasoning tasks such as Raven’s progressive matrices and rule-based tasks involving abstract relations are thought to depend on left rlPFC ( Christoff et al . , 2001; Bunge et al . , 2005 , Bunge et al . , 2009; Davis et al . , 2017 ) . In addition to its role in generalizing abstract , relational rules , we have recently found left rlPFC to be involved in rule evaluation and novel generalization processes for simpler feature-based rules in categorization tasks ( Paniukov and Davis , 2018 ) . In the present study , dissimilarity-based generalization to novel feature pairings may depend on rule evaluation processes in the rlPFC more so than simple similarity-based processing , if studies anticipating that dissimilarity-based processes depend more upon higher-level symbolic rules are correct ( Juslin et al . , 2001; Winman et al . , 2005 ) . Alternatively , pure similarity-based accounts suggest that generalization patterns in an IBRE task do not depend on the existence of a separate , higher-level mechanism ( Medin and Edelson , 1988; Kruschke , 1996 , Kruschke , 2001 ) , and would thus expect a single neurobiological network associated with similarity-based processing to be engaged for choice across trials . Here we test the dissGCM by incorporating its predictions into an analysis of fMRI data collected from participants completing a standard IBRE task ( Medin and Edelson , 1988; Kruschke , 1996 ) . We first examine whether multivoxel activation patterns elicited during conflicting trials in the IBRE task are consistent with participants activating information associated with the rare category , as predicted by pure similarity-based accounts , or activating information associated with ( dissimilarity to ) the common category , as predicted by the dissGCM . To this end , we use representational similarity analysis ( RSA; Kriegeskorte et al . , 2008 ) to decode which features of the stimuli are most strongly activated while participants are categorizing the conflicting items . This analysis is based on recent work in the broader memory literature establishing that it is possible to decode whether participants are retrieving particular object categories from memory based on their activation patterns in ventral temporal cortex ( Rissman and Wagner , 2012; Haxby et al . , 2014 ) . To facilitate the multivoxel analysis , here we use the real world visual categories faces , scenes , and objects as stimulus features . These visual categories have a well-defined representational topography across the cortex ( Haxby et al . , 2001; Grill-Spector and Weiner , 2014 ) , allowing us to predict whether participants are differentially activating particular stimulus features ( faces , scenes , or objects ) by computing similarities between activation patterns elicited for the key IBRE trials and feature-specific patterns from an independent localizer scan . By crossing the visual stimulus features with our category structure ( Figure 1 ) , we create situations where a rare category is associated with one feature type ( e . g . a scene ) and a common category is associated with another feature type ( e . g . an object ) . The extent to which each type of information is active can then be compared to determine whether participants are representing stimulus features associated with the common or rare category on a trial , and thus answer whether their BOLD activation patterns are more consistent with pure similarity or dissGCM’s combined dissimilarity and similarity processes . In this context , we anticipate that the multivoxel pattern analysis will index an interactive process between feature-based attention and memory retrieval: the dissGCM , pure similarity-based GCM , and previous dissimilarity-based inference models all predict that categorization decisions are driven by an attention-weighted memory process whereby a stimulus is compared with the contents of memory ( Nosofsky , 1986; Juslin et al . , 2001; Stewart and Morin , 2007 ) . This prediction suggests that during categorization , the multivoxel patterns activated for a particular stimulus will reflect both direct perceptual processing and retrieval of information from memory . Because the dissGCM predicts greater contributions from the common , unchosen category during this retrieval process , we expect multivoxel patterns during ambiguous trials to reveal greater activation of information associated with the common , unchosen category . In addition to our multivoxel analysis , we also test whether using dissimilarity-based evidence against unchosen categories may tap distinct brain regions , such as the rlPFC , beyond those involved with similarity-based computations . To this end , we take trial-by-trial predictions for how much similarity- and dissimilarity-based evidence contribute to the winning category and use these predictions as regressors in fMRI analysis . We anticipated that the MTL and vmPFC would be positively associated with similarity-based evidence , whereas dlPFC would be negatively associated with similarity-based evidence for the winning category . Contrastingly , we expected rlPFC to track estimates of dissimilarity-based evidence against alternative options . Learning curves over the 12 learning blocks for common and rare disease item pairs are shown in Figure 2 . All subjects reached greater than 90% accuracy over the last four blocks ( M = 98 . 1% , SD = 2 . 4% , range = 93 . 5–100% ) . Mean choice performance in the first block was above chance ( 25% ) for both common ( M = 63 . 6% ) and rare ( M = 43 . 2% ) feature pairs . Consistent with previous IBRE studies , a linear mixed effects model revealed a significant block by trial type interaction ( F ( 1 , 262 ) =20 . 7 , p<0 . 001 ) , suggesting that the common diseases were learned more quickly than the rare diseases . Paired t-tests revealed that participants were significantly more accurate on common compared with rare disease trials in the first ( t ( 21 ) =2 . 26 , p=0 . 034 ) , second ( t ( 21 ) =2 . 85 , p=0 . 010 ) , third ( t ( 21 ) =2 . 72 , p=0 . 013 ) , fourth ( t ( 21 ) =2 . 46 , p=0 . 023 ) , and 12th blocks ( t ( 21 ) =2 . 23 , p=0 . 037 ) . For the test phase , participants were asked to categorize the original category exemplars in addition to a number of other novel feature combinations in the absence of feedback . We then fit the dissGCM to the group choice probabilities for each test item . The dissGCM is based on the original generalized context model ( Nosofsky , 1986 ) , but allows for dissimilarity to be used as evidence for a decision ( Stewart and Morin , 2007 ) . The model posits that people represent stimuli as points in a multidimensional feature space , and that categorization judgments are based on distances between probe stimuli and stored exemplars . As for the standard GCM , similarities to all exemplars of each category are summed into evidence for each category . However , in the dissGCM , evidence that an item is dissimilar to other categories is also used as evidence for a category . For example , evidence for Disease 1 includes not only an item’s similarity to members of Disease 1 , but also its dissimilarity to other diseases . Choice probabilities and dissGCM-derived predictions for each of the test items are summarized in Table 1 . Consistent with an inverse base rate effect , participants were numerically more likely to classify ambiguous test stimuli ( combinations of rare and common features ) as members of the relevant rare category ( M = 49 . 8% ) than the relevant common category ( M = 43 . 5% ) combined across object-scene , scene-scene , and object-object pairs . A one-sample t-test revealed that the percentage of rare responding on ambiguous trials was significantly higher than the 1/4 base rate for the rare category ( t ( 21 ) =8 . 11 , p<0 . 001 ) . Likewise , participants chose the rare category for ambiguous pairs significantly more often than for the imperfect predictors ( faces: M = 30 . 0% ) , ( t ( 21 ) =3 . 85 , p<0 . 001 ) . In addition to response probabilities , we tested whether reaction times differed on the ambiguous test trials depending on whether a rare or common response was made . On these trials of interest , a linear mixed effects model revealed that RTs were slower when participants made rare responses ( M = 1 . 47 s ) than common responses ( M = 1 . 27 s ) , ( t ( 21 ) =10 . 48 , p<0 . 001 ) . The observation of slowed RTs on ambiguous trials receiving rare responses suggests that rare selections may be more cognitively demanding relative to common selections , consistent with previous dissimilarity-based theories of IBRE that posit a role of higher-level , inferential reasoning in base rate neglect . By revealing a link between activation of common feature patterns and the IBRE , our multivoxel results suggest that dissimilarity-based evidence against unchosen categories contributes to choice behavior in the present task . However , it remains an open question whether such dissimilarity processes involve distinct neural or cognitive mechanisms beyond those thought to underlie basic similarity processes . Importantly , similarity-based theories propose that a single , non-inferential cognitive process is responsible for generalization patterns across trials in the IBRE task ( Medin and Edelson , 1988; Kruschke , 1996 , Kruschke , 2001 ) , and thus it is anticipated that a network of brain regions associated with similarity-based generalization underlies choice across task contexts in the present study . Although the dissGCM is agnostic as to whether using dissimilarity as evidence is more cognitively demanding than relying on similarity alone , previous theories of IBRE positing dissimilarity processes propose that the use of contrastive evidence is inherently inferential ( Juslin et al . , 2001; Winman et al . , 2005 ) . Accordingly , the latter account would predict a unique neural topography associated with dissimilarity-based evidence , including regions known to be involved in higher-level , symbolic reasoning . To test whether similarity- and dissimilarity-based evidence rely on different brain regions , we modeled univariate voxel-wise activation using trial-by-trial estimates of similarity- and dissimilarity-based evidence derived from the dissGCM . Specifically , the overall evidence v for the winning category on each test trial was decomposed into two separate regressors: one for summed similarity to the winning category , and the other for summed dissimilarity to the non-winning categories . The regions associated with dissimilarity-based evidence in this analysis are thus distinct from those negatively associated with similarity-based evidence because they are derived from evidence against the alternative , non-winning category . Our analysis showed that greater similarity-based contributions to the winning category were associated with activation in the MTL ( left hippocampus ) vmPFC , and primary motor cortex ( Figure 6A , depicted in red; Table 2 ) . These results are consistent with findings from other model-based fMRI studies suggesting that the MTL is involved in similarity-based retrieval ( Davis et al . , 2012a , Davis et al . , 2012b ) . Likewise , the engagement of vmPFC corroborates recent studies suggesting that this region tracks higher relative evidence for categorization decisions ( Davis et al . , 2017; O'Bryan et al . , 2018 ) . The positive relationship between vmPFC and similarity processes may also be reflective of attention to strong predictors ( Sharpe and Killcross , 2015; Nasser et al . , 2017 ) or the application of familiar category rules ( Boettiger and D'Esposito , 2005; Liu et al . , 2015 ) , both of which are consistent with similarity-based accounts of IBRE that attribute choice to a well-established association between perfect predictors and their outcomes that is driven by attention ( e . g . Kruschke , 1996 ) . We also found that greater contributions of similarity-based evidence were positively associated with activation in the primary motor cortex . While more anterior motor-planning regions such as pre-SMA and SMA tend to be associated with rule acquisition processes ( e . g . Boettiger and D'Esposito , 2005 ) , primary motor cortex has been found to track increasing levels of response automaticity in categorization tasks ( Waldschmidt and Ashby , 2011 ) . The dlPFC , dorsomedial PFC , and posterior parietal cortex were found to be negatively correlated with similarity-based evidence for the chosen category ( Figure 6A , depicted in blue; Table 2 ) . This fronto-parietal network is generally associated with rule-based category learning ( Filoteo et al . , 2005; Seger and Cincotta , 2006; Soto et al . , 2013 ) , and is thought to play a critical role in representing the uncertainty associated with categorization decisions ( DeGutis and D'Esposito , 2007; Seger et al . , 2015; Davis et al . , 2017 ) . Thus , our results are consistent with these findings , and moreover , suggest that dlPFC and functionally related fronto-parietal regions may be engaged in cases where probes fail to elicit a strong similarity-based match with stored category exemplars . Contrastingly , dissimilarity-based evidence was positively correlated with activation in the left rlPFC ( Figure 6B; Table 2 ) , consistent with our hypothesis that this type of evidence might encourage more symbolic processes believed to underlie the rlPFC’s contribution to category learning ( Davis et al . , 2017; Paniukov and Davis , 2018 ) . No clusters were significantly negatively associated with dissimilarity-based evidence . Comparing the statistical maps in Figure 6A and B , it is apparent that greater relative contributions of dissimilarity-based evidence do not necessitate smaller contributions of similarity-based evidence in the dissGCM: if these regressors were anticorrelated , one would expect the regions associated with dissimilarity processes to resemble the fronto-parietal network we found to be negatively associated with similarity . Instead , our results show that using contrastive evidence uniquely engages the rlPFC , in line with dual-process theories that suggest dissimilarity-based processing may distinctly depend on higher-level , abstract reasoning . Despite obvious differences between the activation patterns elicited for each contrast , it is notable that all three maps ( positive similarity , negative similarity , and positive dissimilarity ) revealed significant activation in portions of ventral occipitotemporal cortex . These regions ( lateral occipital cortex , inferior temporal gyrus , and fusiform gyrus ) have well-established roles in representing visual object categories , including those used as stimulus dimensions in the present study ( Grill-Spector and Weiner , 2014 ) . Accordingly , it is possible that the engagement of these regions in our study reflects feature-based attention , exemplar retrieval , or a combination of both processes that occurs regardless of the respective contributions that similarity- and dissimilarity-based evidence make to a decision . The present study employed model-based fMRI to test how similarity and dissimilarity contribute to the inverse base rate effect ( IBRE ) and how these types of evidence relate to neural mechanisms that support category learning . The dominant theory behind the IBRE suggests that it arises from attentional processes that make ambiguous items containing features of rare and common categories seem more similar to members of the rare category . Here we find support for the hypothesis that dissimilarity-based evidence also contributes to the IBRE: people may categorize the ambiguous stimuli as members of the rare category not only because of their similarity to the rare category , but also because of their dissimilarity to members of the common category . The dissGCM , an extension of the GCM that allows for the use of dissimilarity-based evidence in categorization behavior , predicted two novel observations in the neuroimaging data . First , as predicted by the dissGCM’s relative contribution of similarity- and dissimilarity-based evidence during the ambiguous trials , multivoxel analysis suggested stronger activation of patterns associated with features of the common category when participants classified ambiguous stimuli as rare . Second , model-based univariate analysis revealed that measures of similarity- and dissimilarity-based evidence had unique neural topographies . Similarity-based evidence for the winning category was positively correlated with regions of the hippocampus , vmPFC , and primary motor cortex . In contrast , dlPFC , dorsomedial PFC , and posterior parietal cortex were negatively correlated with similarity-based contributions . Dissimilarity-based evidence against non-winning categories was positively correlated with the left rlPFC . The present results raise several important questions about the cognitive and neural mechanisms underlying people’s use of base rate information . Previous theories arguing for dissimilarity-like processes as explanations of IBRE have argued that they arise from mechanisms rooted in higher-level propositional logic that fundamentally differ from the similarity-based mechanisms posited by dominant theories ( Juslin et al . , 2001 ) . As illustrated by the dissGCM , such dissimilarity-based processes can be viewed as simple extensions of similarity-based processing and need not depend on the existence of a functionally separate categorization system . At the same time , our neuroimaging results suggest that dissimilarity , but not similarity-based evidence may arise from processing in rlPFC regions that are known to be involved with higher-level reasoning and problem solving ( Christoff et al . , 2001; Bunge et al . , 2005 , Bunge et al . , 2009 ) . One possibility for reconciling these theories is that the dissimilarity-based evidence involves more abstract or symbolic feature processing than pure similarity processes , and this additional processing taps rlPFC regions . This is consistent with our recent model-based fMRI results , which demonstrate that rlPFC tracks measures of relational encoding in category learning , but otherwise this type of category learning may rely on the same basic similarity-based mechanisms as simpler feature-based learning ( Davis et al . , 2017 ) . By establishing that the rlPFC is engaged when participants incorporate dissimilarity-based evidence into categorization decisions , our research adds to a growing literature aiming to pinpoint a domain-general computational role for this region . A common thread among tasks shown to engage the rlPFC is that they tend to involve combining across disparate representations to form the basis for a decision – whether those representations comprise confidence estimates and subjective value ( De Martino et al . , 2013 ) , visual features and their relations ( Bunge , 2004; Bunge et al . , 2009; Davis et al . , 2017 ) , or expected rewards and their relative uncertainties ( Boorman et al . , 2009; Badre et al . , 2012 ) . Likewise , in the case of the current study , the evidence that an ambiguous stimulus is similar to a given category must be combined with the evidence that the stimulus is dissimilar to the other possible categories . Although the dissGCM instantiates dissimilarity as a simple transformation of similarity , the involvement of rlPFC when participants place more reliance on dissimilarity-based evidence may be attributable to increasing demands for integrating evidence across several abstract representations . A decision made on pure similarity-based evidence would require no such integration . This hypothesis accords with recent findings implicating the rlPFC in evaluative processes for categorization tasks that require candidate rules to be weighed over the course of several trials , relative to matching tasks where a rule can be known with certainty following a single correct trial ( Paniukov and Davis , 2018 ) . One question that has arisen repeatedly in the literature on the IBRE is whether it reflects an inherent irrationality in decision making . When viewed through the lens of basic similarity-based attentional processes ( e . g . Medin and Edelson , 1988; Kruschke , 1996 , Kruschke , 2001 ) , the IBRE appears to arise from very simple learning mechanisms that are not particularly tied to higher-level rationality , and rare choices seem to indicate a lack of knowledge of the base rates . Indeed , in a separate model fit , we attempted to fit the standard similarity-based GCM to the key pattern on the ambiguous trials . However , the standard GCM was only able to predict a greater proportion of rare choices if accurate knowledge of the exemplar base rates was eliminated ( all values of tj = 1 or fit as free parameters ) . In contrast , accurate knowledge of the category base rates directly contributes to the greater dissimilarity-based evidence against the common category . Thus from the dissGCM perspective , participants are perfectly knowledgeable about the base rates in the present task , but they use this knowledge in a way not anticipated by pure similarity-based models . However , whether or not this use of dissimilarity-based evidence constitutes irrationality is a deeper question that cannot be answered based purely on the present results . How or whether the use of dissimilarity is encouraged by the standard IBRE design , compared with other types of categorization problems , is an open question . The dissGCM was originally developed to explain sequential effects in categorization ( Stewart and Morin , 2007 ) , and its success in this domain suggests that dissimilarity processes , such as those revealed here , may be present in many categorization tasks that are more familiar in the neuroimaging literature . However , how much a task encourages dissimilarity-based processing may vary considerably and depend on a number of factors . For example , purely attentional accounts posit that strong initial learning of the common category leads people to learn the rare category by the features that distinguish it from the common ( Markman , 1989; Kruschke , 1996 ) . Learning-order effects do appear to play a role in the IBRE: previous studies have shown that using blocked or unequally distributed category frequencies during training leads participants to favor later-learned categories on ambiguous test probes , even when overall base rates are held constant ( Medin and Bettger , 1991; Kruschke , 1996 ) . Early studies on the role of order and blocking on IBRE are similar to current work on blocked versus interleaved learning in the broader categorization literature ( Birnbaum et al . , 2013; Carvalho and Goldstone , 2014 , Carvalho and Goldstone , 2015; Goldwater et al . , 2018 ) . In blocked learning , categories are learned by viewing a number of items from the same category before switching to other categories . For example , in the present case , if we had used blocked learning , participants may see an entire block of Disease 1 examples , and then blocks of Disease 2 , 3 , and 4 , but the examples of the Diseases would not be intermixed . In interleaved learning , the standard for category learning , items from all categories are presented in a random order such that the examples of the different categories are intermixed . Blocked learning tends to lead participants to focus more on stimulus features that are shared with members of the same category , whereas interleaved learning tends to lead participants to focus more on features that differentiate categories . Interestingly , while not an interleaved versus blocked manipulation per se , frequency manipulations such as those used in the present study have an effect of creating more blocking within common categories – common categories are more likely to follow examples of the same category , and interleaving within rare categories . Although discovered long after the initial IBRE studies , blocked versus interleaved learning theories may offer a concurring explanation of the IBRE that does not depend on differences in the rate at which common and rare categories are learned . However , formal computational models of blocked versus interleaved learning have thus far focused on how these scenarios produce differences in selective attention to stimulus features that are characteristic ( blocked ) or diagnostic ( interleaved ) of a category , and are pure similarity-based models such as the original GCM ( Carvalho and Goldstone , 2017 ) . Contrastingly , our MVPA and univariate fMRI results show that pure similarity-based processing cannot fully explain the IBRE , and thus strongly suggest that dissimilarity processes contribute to the IBRE . To investigate how learning manipulations , such as blocked versus interleaved , or individual differences in learning influence the mechanisms we propose here , it will be critical for future research to build full learning models of the dissGCM . The dissGCM , like the standard GCM , is a model of asymptotic categorization performance and generalization , and thus is not well-equipped to account for learning dynamics or individual differences . For these reasons , our individual difference analysis focused on using MVPA estimates from individual stimuli rather than formal model-based analysis . Nonetheless , these analyses reveal important individual differences that are consistent with dissimilarity-based theories more broadly . Dissimilarity-based theories posit that one of the reasons IBRE arises is because the common category becomes more thoroughly established in memory during learning , which leads participants to retrieve this information more readily at test . From this perspective , participants who learn the common category more strongly should consequently exhibit base rate neglect more frequently . Consistent with these predictions of dissimilarity-based theories , pattern similarity analyses revealed that participants who more strongly activated information associated with common categories during learning engaged in base rate neglect more often at test . While these results suggest that individual differences in learning contribute to IBRE , they nonetheless point to a critical need to develop a full learning-based version of the dissGCM that can be applied at the individual level to capture these differences . For example , one possibility is that participants’ weighting of similarity and dissimilarity ( the s parameter ) changes over learning based on the participants’ learning rates and factors related to blocking versus interleaved presentation ( Carvalho and Goldstone , 2017 ) . However , such a model would require extensive additional data to validate , and thus is beyond the scope of the present study . The IBRE exemplifies a case in cognitive neuroscience where independent models that predict essentially the same behavioral patterns make very different assumptions about the cognitive processes , and accordingly , brain states , involved in producing the behavior . Our findings from the test phase represent a critical step forward in an emerging area of research using multivariate fMRI to reveal that qualitatively distinct brain states may reflect the use of multiple response strategies in the face of identical stimuli ( e . g . Mack et al . , 2013 ) . Consistent with past research using MVPA to decode learned selective attention ( Mack et al . , 2013 , Mack et al . , 2016; Leong et al . , 2017; O'Bryan et al . , 2018 ) , multivoxel patterns associated with predictive features were more strongly activated than imperfectly predictive features during the learning phase . Using the same approach to decode which information participants were focusing on during ambiguous test trials , we found stronger activation of patterns associated with common compared with rare stimulus features , but importantly , this pattern only emerged in cases where participants chose the rare category . Moreover , rare category selections were accompanied by slower RTs relative to common selections . These results are consistent with a higher-level , dissimilarity-based process where activating information associated with common exemplars provides contrastive evidence against the well-established common category . Alternatively , it is possible that participants are more likely to respond according to the base rates when the ambiguous stimuli elicit a strong similarity-based match: given our RT results along with the correlation between similarity-based evidence and motor cortex engagement , in these cases subjects may revert to habitual response patterns from the learning phase and simply choose the more well-established ( common ) category . However , understanding the precise cognitive mechanisms that contribute to these response-dependent activation patterns remains a direction for future research . Interestingly , while our findings argue against the prediction from similarity-based models that the IBRE arises because rare features become more similar to their associated category , the observed attention weight parameters wk from the model fits are consistent with a key part of similarity theory – that there is greater selective attention allocated to the rare feature dimension . Indeed , the rare feature dimensions outweighed the common features for both sets of categories in our data . However , these larger attention weights did not seem to drive greater pattern similarity to the rare feature dimension in our multivoxel results . We predict that our multivoxel results are not driven directly by simple feature-based attention , but instead indicate some combination of attention and memory-based retrieval of the category exemplars . Pattern similarity measures in ventral temporal cortex have been shown to effectively index both dimensional selective attention ( Leong et al . , 2017; O'Bryan et al . , 2018 ) and the retrieval of non-present , associated stimuli ( e . g . Zeithamova et al . , 2012; for review , see Rissman and Wagner , 2012 ) . Rather than adjudicating between whether the multivoxel patterns in the current study are more likely to indicate attention or memory , a possibility that accords with both potential explanations is that these pattern similarity indices reflect information that is actively represented in working memory , either by way of visual cueing or reinstated long-term memories ( Lewis-Peacock and Postle , 2008 ) . In cases in which multiple or competing stimulus representations are present in WM , as may be expected for the ambiguous IBRE trials , multivoxel patterns should be most similar to whichever representation is consciously attended ( Lewis-Peacock et al . , 2012 ) . However , given the design of the current study we are unable to rule out the possibility that implicitly activated or post-decisional feature representations contribute to our pattern similarity results . Future studies may wish to combine multivoxel pattern analysis with eye-tracking ( e . g . Leong et al . , 2017 ) to better understand the unique contributions that attention and memory make to the present results . In conclusion , using model-based fMRI analysis , we found evidence that extreme cases of base rate neglect such as the IBRE may arise from a combination of similarity- and dissimilarity-based processes . Accordingly , measures of neural activation suggest that people may be more strongly relying on evidence about how dissimilar an item is to common categories when faced with ambiguous stimuli . Furthermore , dissimilarity processes have a unique cortical topography that includes the rostrolateral PFC , a region believed to be involved with more symbolic feature processing . The study consisted of three phases: localizer , learning , and test . The localizer phase consisted of two scanning runs ( run length = 5 min 10 s ) in which participants classified images based on whether they contained a face , an object , or a scene . Each image was presented for 2 . 5 s during which participants were asked to respond ‘Scene ( 1 ) , Face ( 2 ) , or Object ( 3 ) ? ' Each trial was separated with random fixation drawn from a truncated exponential distribution with mean = 3 s . Over the duration of the localizer phase , subjects categorized 38 examples of each stimulus type . The face , object , and scene images used were black-and-white squares presented on a white background with black text . The stimuli used during the localizer runs were presented in a random order , and did not include any of the images used for the experimental task . In the learning phase , participants learned a classic IBRE category structure ( Medin and Edelson , 1988; see Figure 1 ) . The features used for the stimuli included examples of faces , objects , and scenes not shown in the localizer phase . Participants were given an epidemiological cover story asking them to predict whether hypothetical patients would contract a disease based on the people they have been in contact with ( faces ) , the objects they have used ( objects ) , and the places they have been ( scenes ) . On each trial of the learning phase , participants would see a stimulus for 3 s and were asked to answer ‘Disease 1 , 2 , 3 , or 4 ? ’ This was followed by random fixation , feedback ( 1 . 75 s ) in which they were told whether they were right or wrong and the correct answer , and additional fixation . The same distribution was used to generate fixations as in the localizer phase . Faces were always assigned to the imperfectly predictive feature dimensions , whereas objects and scenes were perfectly predictive and associated with only one disease ( Figure 1 ) . To ensure that no visual stimulus category differed in overall frequency , one common disease was always associated with objects and the other scenes , and likewise for rare diseases . Participants were randomly assigned to one of two conditions to balance which images were presented together during learning and test , and disease labels were randomized across participants . Within-pair stimulus position ( left or right ) was randomized on each trial , and the presentation order of feature pairs was randomized within each block for every participant . The learning phase was spread over three scanning runs ( run length = 5 min 10 s ) , and four blocks of the stimulus set were presented per run , resulting in a total of 12 blocks and 96 trials for the learning phase . The progression of a learning trial is depicted in the bottom panel of Figure 1 . During the test phase , participants completed trials with both new and old exemplars and classified them as ‘Disease 1 , 2 , 3 , or 4 ? ' , but no longer received feedback . New items included all possible single and two-feature combinations of the perfectly predictive features ( see Table 1 , Results ) . Trials were 3 s and separated by random fixation as described above . Like the learning phase , the test phase occurred over three consecutive scanning runs ( run length = 5 min 10 s ) . Each item in the stimulus set was encountered twice per run , with the exception of the ambiguous perfect predictor pairs which were repeated four times per run . This resulted in 24 instances of ambiguous scene-object pairs , and 48 instances of the ambiguous trials overall for each participant . Presentation order of the test items was randomized for each of the three runs , with participants rating two test sets per run , resulting in a total of 156 test trials . The dissimilarity generalized context model ( dissGCM; Stewart and Morin , 2007 ) is an extension of the generalized context model ( Nosofsky , 1986 ) that accounts for choice using a combination of similarity- and dissimilarity-based evidence . Like the original GCM , stimuli are represented as points in a multidimensional feature space . The model computes distances in this space between probe stimuli Si and stored exemplars Sj along each dimension k: ( 1 ) dij= ( ∑k=1Kwk\|Sik−Sjk\|r ) 1/r , where r defines the metric of the space , here assumed to be one ( city-block ) . The wk indicates dimensional attention weights , which have the function of stretching the distance along strongly attended dimensions , and are constrained to sum to one . Distances are converted to similarities via an exponential transform: ( 2 ) simij=e−cdij , where c is a specificity parameter that controls the rate at which similarity decays as a function of distance . The first contribution of evidence for a given category comes from the summed similarity between a probe and all stored exemplars for that category , consistent with the original GCM . DissGCM then combines this similarity-based contribution with the summed dissimilarity between a probe and the exemplars from all other categories . The overall evidence , v , for a category CA , given stimulus Si is: ( 3 ) viA=s∑sj∈CAtjsimij+ ( 1−s ) ∑sj∈¬CAtj ( 1−simij ) , where s is a free parameter that determines how much the model weights similarity versus dissimilarity . The parameter tj reflects exemplar-specific memory strength , which we fix at each exemplar’s true base rate during learning ( 1 for rare category exemplars , 3 for common category exemplars ) . Here , we also make the assumption that exemplars only contribute evidence ( similarity or dissimilarity ) if they have at least one positive feature match with a probe stimulus . The model makes a prediction for how likely an item is to be classified as a member of a given category CA by: ( 4 ) pr ( resp=CA\|Si ) =viA+b∑viC+4b , where b is a free parameter that reflects the baseline level of similarity for a category that has 0 positive feature matches . More generally , this parameter ensures that no predicted probabilities are 0 or 1 , which interferes with the maximum likelihood-based model fits . The model was fit to the group response frequencies for each option by minimizing the −2 * Log Likelihood using a differential evolution function optimizer . The overall fit was 4 , 314 . 588 . The best fitting parameters for each of the dimension weights were w1 ( face 1 ) =0 . 277 , w2 ( common scene ) =0 . 665 , w3 ( rare object ) =0 . 887 , w4 ( face 2 ) =0 . 170 , w5 ( common object ) =0 . 712 , and w6 ( rare scene ) =0 . 879 ) ; c = 9 . 05; s = 0 . 946; b = 0 . 023 . Imaging data were acquired on a 3 . 0 T Siemens Skyra MRI scanner at the Texas Tech Neuroimaging Institute . Structural images were acquired in the sagittal plane using MPRAGE whole-brain anatomical scans ( TR = 1 . 9 s; TE = 2 . 44 ms; θ = 9°; FOV = 250 × 250 mm; matrix = 256 × 256 mm; slice thickness = 1 . 0 mm , slices = 192 ) . Functional images were acquired using a single-shot T2*-weighted gradient echo EPI sequence ( TR = 2 . 5 s; TE = 25 ms; θ = 75°; FOV = 192 × 192 mm; matrix = 64 × 64; slice thickness = 3 mm ) . Functional data were preprocessed and analyzed using FSL ( www . fmrib . ox . ac . uk/fsl ) . Anatomical images were preprocessed using Freesurfer ( autorecon1 ) . Functional images were skull stripped , motion corrected , prewhitened , and high-pass filtered ( cutoff: 60 s ) . For the model-based univariate analysis , functional images were spatially smoothed using a 6 mm FWHM Gaussian kernel . No smoothing was performed on functional data used for the multivoxel analysis . First-level statistical maps were registered to the Montreal Neurological Institute ( MNI ) −152 template using 6-DOF boundary-based registration to align the functional image to the Freesurfer-processed high-resolution anatomical image , and 12-DOF affine registration to the MNI-152 brain . The model-based univariate analysis employed a standard three-level mixed effects model carried out in FSL’s FEAT program . The first-level model included an EV for stimulus presentation and two model-based parametric modulators: similarity- and dissimilarity-based evidence , computed from the dissGCM . Specifically , these regressors were obtained on a trial-by trial basis using equation 3 ( see Model section ) , where the evidence contribution of summed similarity to the winning category ( CA; most probable category according to the model ) is calculated as: ( 5 ) s∑Sj∈CAtjsimij , and the evidence contribution of summed dissimilarity to non-winning categories with a positive feature match is calculated as: ( 6 ) ( 1−s ) ∑Sj∈¬CAtj ( 1−simij ) , Both parametric modulators were centered and scaled ( z-scored ) within run . Additional explanatory variables ( EVs ) of no interest included motion parameters , their temporal derivatives , EVs to censor volumes exceeding a framewise displacement of 0 . 9 mm ( Siegel et al . , 2014 ) , and an EV to account for trials in which participants failed to make a behavioral response . Final statistical maps were corrected for multiple comparisons using a non-parametric cluster-mass-based correction with a cluster-forming threshold of t ( 21 ) =3 . 52 ( p<0 . 001 , one-tailed ) . RSA was conducted using the PyMVPA toolbox ( Hanke et al . , 2009 ) and custom Python routines . To obtain trial-by-trial estimates of the hemodynamic response , we computed a β-map ( Rissman et al . , 2004 ) for each stimulus onset using an LS-A procedure ( Mumford et al . , 2012 ) , simultaneously modeling the trials of interest as separate regressors in a GLM . These estimates were anatomically restricted to three ventral temporal ROIs that were maximally responsive to scene , object , and face information in the localizer data . Specifically , pattern estimates were spatially localized in visual stimulus category-specific ROIs by creating 6 mm spheres around subjects’ peak activation within anatomically defined regions in the Harvard-Oxford Atlas associated with category selectivity ( objects: left inferior posterior temporal gyrus; scenes: bilateral parahippocampal gyrus; faces: right temporal occipital fusiform gyrus; Ishai et al . , 1999; Lewis-Peacock and Postle , 2008; Lewis-Peacock et al . , 2012; Grill-Spector and Weiner , 2014 ) . The last trial of each run was automatically discarded from the multivoxel analysis to ensure stable estimation of the activation patterns for all trials . Additional explanatory variables ( EVs ) of no interest included motion parameters , their temporal derivatives , and EVs to censor volumes exceeding a framewise displacement of 0 . 9 mm . For the primary pattern similarity analyses , we measured how much participants were activating scene , object , and face information on individual test phase trials by calculating mean correlation distance ( 1 – Pearson’s r ) between activation patterns on each test trial and those elicited for each visual category during the localizer phase . For interpretative ease , the distances were converted to similarities using exp ( - distance ) , and then standardized ( z-scored ) within participants . Source data and scripts used to create all figures and tables ( e . g . R code , PyMVPA scripts , statistical maps for the model-based fMRI analysis ) are freely available online at https://osf . io/atbz7/ .	Is a patient with muscle aches , headache and fever more likely to have influenza or Ebola ? Most people correctly choose ‘influenza’ because it is the more common of the two diseases . But what about someone with a cough and unexplained bleeding ? Coughing is a symptom of influenza but not of Ebola . Unexplained bleeding is a symptom of Ebola but not of influenza . Faced with ambiguous symptoms such as these , many people would diagnose ‘Ebola’ despite knowing that influenza is more common . Indeed , when a situation shares characteristics with both a common and a rare category , we often tend to predict that it belongs to the rare group . This phenomenon is known as base rate neglect , but why does it occur ? One theory is that we pay more attention to features that belong to rare categories ( such as unexplained bleeding ) because they are distinctive and unusual . But another possibility is that we use our knowledge of the common category to rule out examples that do not conform to our expectations . Of the many cases of influenza that you have heard about or experienced , probably none of them featured unexplained bleeding . To distinguish between these possibilities , O’Bryan et al . trained healthy volunteers on a categorization task that included ambiguous stimuli . The participants performed the task inside a brain scanner . O’Bryan et al . then programmed a computer to solve the same problems . The simulation either used similarity-based judgments ( how similar is this to the rare category ? ) , dissimilarity-based judgments ( how dissimilar is this from the common category ? ) , or both . The results suggested that when people show base rate neglect , they rely more on dissimilarity-based evidence than on similarity-based evidence . In other words , they focus more on how a test item differs from the common category . Consistent with this , whenever the volunteers chose the rare category , their brains were processing information about the common category . The imaging results also revealed that when the volunteers used dissimilarity-based evidence , they activated a brain region involved in abstract thinking and reasoning . How people use information about likelihoods is relevant to all aspects of decision-making . Beyond helping us to understand how we assign items to categories , the work by O’Bryan et al . could also inform future research in areas such as learning and memory .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2018	Model-based fMRI reveals dissimilarity processes underlying base rate neglect
The G9a/GLP complex mediates mono- and dimethylation of Lys9 of histone H3 at specific gene loci , which is associated with transcriptional repression . However , the molecular mechanism by which the G9a/GLP complex is targeted to the specific gene loci for H3K9 methylation is unclear . In this study , with unbiased protein affinity purification , we found ZNF644 and WIZ as two core subunits in the G9a/GLP complex . ZNF644 and WIZ interact with the transcription activation domain of G9a and GLP , respectively . Moreover , both ZNF644 and WIZ contain multiple zinc finger motifs that recognize consensus DNA sequences . ZNF644 and WIZ target G9a and GLP to the chromatin and mediate the G9a/GLP complex-dependent H3K9 methylation as well as gene repression . Thus , our studies reveal two key subunits in the G9a/GLP complex that regulate the function of this histone methyltransferase complex . Post-translational modifications of histones , especially lysine methylation , play important roles in diverse biological processes , such as gene transcription , chromatin packaging , and cellular differentiation . Different lysine residues of histones are methylated by different histone lysine-specific methyltransferases ( HKMTs ) . Among HKMTs , G9 and GLP specifically catalyze H3K9 mono- ( H3K9me1 ) and dimethylation ( H3K9me2 ) ( Tachibana et al . , 2001 , 2005; Kubicek et al . , 2007 ) . G9a and GLP are paralogs with similar domain architecture . Both contain a transcription activation domain ( TAD ) , a glutamate-rich domain , a cysteine-rich domain , 7 tandem ankyrin repeats ( ANKs ) , and a methyltransferase domain ( Roopra et al . , 2004; Dillon et al . , 2005; Lee et al . , 2006; Purcell et al . , 2011; Shinkai and Tachibana , 2011; Bittencourt et al . , 2012 ) . Interestingly , G9a and GLP form a heterodimer via the interaction between the C-terminal catalytic domains ( Tachibana et al . , 2005 ) . The major function of the G9a/GLP complex is to catalyze H3K9me1 and H3K9me2 in euchromatin , which is associated with transcriptional repression ( Tachibana et al . , 2005 , 2008 ) . Accumulated evidence has shown that this methyltransferase complex regulates multiple biological processes , such as meiosis , embryonic development , immune response , and tumorigenesis ( Tachibana et al . , 2002 , 2007; Schaefer et al . , 2009; Chen et al . , 2010; Huang et al . , 2010; Shinkai and Tachibana , 2011 ) . Interestingly , if one of these two methyltransferases is deleted , the other one alone has little enzymatic activity in vivo , suggesting that the heterodimer formation is important for the function of this enzyme complex ( Tachibana et al . , 2005 , 2008 ) . Downstream functional partners of the G9a/GLP complex have been examined . Since HP1 recognizes H3K9me2 ( Bannister et al . , 2001; Lachner et al . , 2001; Nielsen et al . , 2001 ) , it is likely that the G9a/GLP complex mediates the recruitment of HP1 to specific gene loci for transcriptional repression ( Ogawa et al . , 2002; Nishio and Walsh , 2004; Shinkai and Tachibana , 2011 ) . Moreover , the ANKs of G9a and GLP also recognize H3K9me1 and H3K9me2 ( Collins et al . , 2008 ) , which may facilitate the chromatin spreading of the G9a/GLP complex . Recent study suggests that the G9a/GLP complex is associated with Polycomb Repressive Complex 2 ( PRC2 ) ( Mozzetta et al . , 2014 ) . EZH2 , the catalytic subunit in the PRC2 , regulates histone H3K27 methylation . Thus , it is likely that these methyltransferases function together to modulate histone codes during transcription . In addition to histone methylation , the G9a/GLP complex also regulates DNA methylation during early embryogenesis , which is independent of their methyltransferase activity . It has been reported that the G9a/GLP complex associates with DNMT1 through PCNA ( Esteve et al . , 2006 ) . Moreover , the ANKs of G9a interact with DNMT3A and 3B for the de novo DNA methylation ( Epsztejn-Litman et al . , 2008; Chang et al . , 2011 ) . Although the G9a/GLP complex plays an important role in epigenetic modification and gene transcription , the molecular mechanism by which the G9a/GLP complex is regulated in vivo remains elusive . Although several partners of G9a have been identified , it is unclear whether these partners form a stable complex with G9a and GLP , and directly control the G9a/GLP-dependent H3K9me1 and H3K9me2 . In this study , we searched other possible subunit ( s ) in the G9a/GLP complex . With unbiased protein affinity purification , we found that ZNF644 and WIZ , two zinc finger proteins , interact with G9a and GLP , respectively . These two zinc finger proteins target G9a and GLP to genomic loci for the regulation of gene transcription . To explore the regulation mechanism of the G9a/GLP complex , we have searched the functional partner ( s ) of G9a using tandem protein affinity purification . Cell lysates of 293T cells stably expressing SFB-tagged G9a were subjected to two rounds of affinity purification . Since G9a and GLP form a heterodimer in vivo ( Tachibana et al . , 2005 ) , we could easily detect GLP as a partner of G9a in this purification , which was served as a positive control . Interestingly , besides GLP , G9a also interacted with two other proteins . Mass spectrometry analysis revealed that these two proteins were ZNF644 and WIZ , two zinc finger proteins . Between these two proteins , WIZ has been known to regulate the stability of the G9a/GLP heterodimer ( Ueda et al . , 2006 ) , while the function of ZNF644 has not been characterized yet ( Figure 1A ) . To validate our initial purification results , we performed reciprocal affinity purification using ZNF644 as the bait . Again , we identified G9a , GLP , and WIZ as binding partners of ZNF644 , suggesting that ZNF644 and WIZ co-exist in the same complex with G9a and GLP ( Figure 1B ) . 10 . 7554/eLife . 05606 . 003Figure 1 . ZNF644 and WIZ associate with G9a . ( A ) Silver staining of affinity-purified G9a complex . Cell lysates of 293T cells stably expressing SFB-G9a were subjected to affinity purification . Eluted proteins were visualized by silver staining . Arrows indicate proteins corresponding to G9a , GLP , ZNF644 and WIZ . Peptide coverage is shown in the table . ( B ) Silver staining of affinity-purified ZNF644 partners . ( C ) ZNF644 and WIZ co-exist in the same complex with G9a and GLP . U2OS cell lysates were analyzed by co-immunoprecipitation ( co-IP ) and Western blotting with the antibodies indicated . The whole cell lysates of U2OS was used as the input . An irrelevant IgG was used as the IP control . ( D ) Down-regulation of G9a or GLP impairs the interaction between WIZ and ZNF644 . G9a and GLP were down-regulated by siRNAs in U2OS cells . The cell lysates were analyzed by IP and Western blotting with the antibodies indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 00310 . 7554/eLife . 05606 . 004Figure 1—figure supplement 1 . ZNF644 and WIZ antibodies have been generated and specifically recognize the endogenous ZNF644 and WIZ respectively . ( A ) Anti-ZNF644 antibody recognizes endogenous ZNF644 from U2OS cell lysates . In cell lysates , the antibody specifically recognized a band around 150 kD . When cells were treated with siZNF644 , this band was disappeared . Anti-actin was used as protein loading control . ( B ) Anti-WIZ antibody recognizes endogenous WIZ from U2OS cell lysates . In cell lysates , the antibody specifically recognized a band around 175 kD . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 004 To further confirm the interaction between these proteins in vivo , we first raised antibodies against endogenous ZNF644 or WIZ . Anti-ZNF644 and WIZ antibodies specifically recognized bands around 150 kDa and 175 kDa , respectively . Moreover , siZNF644 and siWIZ treatment diminished the expression of these two proteins , indicating that both antibodies specifically recognize the endogenous proteins ( Figure 1—figure supplement 1 ) . We next performed co-immunoprecipitation ( co-IP ) assays using U2OS cell lysates and found that one protein associated with the other proteins in this complex ( Figure 1C ) , suggesting that these proteins are core subunits in the G9a/GLP complex . To further characterize the interactions between these subunits , we knocked down G9a or GLP by siRNAs . Interestingly , down-regulation of G9a or GLP impaired the interaction between WIZ and ZNF644 ( Figure 1D ) , suggesting that the association between WIZ and ZNF644 is mediated by G9a and GLP . Collectively , by unbiased protein affinity purification and co-IP assays , we found that ZNF644 and WIZ are two important subunits in the G9a/GLP complex . Next , we examined the interaction domain in each subunit in this complex . Based on the domain architecture ( Lee et al . , 2006; Shinkai and Tachibana , 2011 ) , we generated four internal deletion mutants of G9a to delete the TAD , the Glu-rich and Cys-rich domains , the ANKs and the catalytic domain , respectively ( Figure 2A ) . Interestingly , the D1 mutant of G9a abolished the interaction with ZNF644 ( Figure 2B ) . Since the D1 mutant of G9a lacks the TAD , it suggests that ZNF644 interacts with the TAD of G9a . Moreover , lacking the catalytic domain of G9a disrupted the interaction with WIZ ( Figure 2C ) . However , the catalytic domain of G9a also interacts with the catalytic domain of GLP for a heterodimer . Thus , it is possible that the association between G9a and WIZ is mediated by GLP . To confirm this hypothesis , we knocked down G9a by siRNA ( Figure 2—figure supplement 1A ) . Lacking G9a , WIZ still interacted with GLP ( Figure 2—figure supplement 1B ) , suggesting that WIZ is likely to directly interact with GLP . To further elucidate the interactions between these subunits , we generated two internal deletion mutants of GLP to delete either the TAD or the catalytic domain ( Figure 2D ) . Only the TAD of GLP , but not the catalytic domain of GLP , is required for the interaction with WIZ ( Figure 2E ) , suggesting that WIZ recognizes the TAD of GLP . In contrast , the catalytic domain of GLP is required for the interaction with ZNF644 ( Figure 2F ) . Since the catalytic domains of G9a and GLP form a heterodimer , it is likely that ZNF644 directly recognizes the TAD of G9a and associates with GLP via the interactions between G9a and GLP . Taken together , ZNF644 and WIZ interact with the TADs of G9a and GLP , respectively . 10 . 7554/eLife . 05606 . 005Figure 2 . Mapping the interaction regions of ZNF644 , WIZ , G9a and GLP . ( A ) A series of deletion mutants of SFB-tagged G9a were generated to map the interaction region of G9a . CD: catalytic domain . ( B ) The D1 mutant of G9a abolishes the interaction with ZNF644 . SFB-tagged wild-type G9a and deletion mutants were expressed in 293T cells together with Myc-ZNF644 . The cell lysates were subjected to streptavidin beads pull-down and Western blotting with the indicated antibodies . The whole cell lysates were used as the input . Cells only expressing Myc-ZNF644 were used for pull-down control . ( C ) Lacking the catalytic domain of G9a ( CD ) disrupts the interaction with WIZ . ( D ) The TAD and catalytic domain deletion mutants of GLP are generated . ( E ) The TAD domain of GLP is important for the interaction with WIZ . ( F ) Lacking the catalytic domain of GLP abolishes the interaction with ZNF644 . ( G ) The N-terminus deletion mutant of ZNF644 abolishes the interaction with G9a . Myc-tagged ZNF644 and deletion mutants were co-expressed together with SFB-G9a in 293T cells . IP and Western blotting were performed with indicated antibodies . ( H ) The C-terminus deletion mutant of WIZ abolishes the interaction of WIZ and GLP . SFB-tagged WIZ and deletion mutants were co-expressed with Myc-GLP in 293T cells . ( I ) A model shows that the N-terminus of ZNF644 interacts with the TAD of G9a , while the C-terminus of WIZ interacts with the TAD of GLP . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 00510 . 7554/eLife . 05606 . 006Figure 2—figure supplement 1 . Down-regulation of G9a doesn’t affect the interaction between WIZ and GLP . ( A ) The siRNA targeting G9a is used to down-regulate G9a in U2OS cells . ( B ) The cell lysates were analyzed by co-IP and Western blotting with indicated antibodies . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 006 We also mapped the interaction regions on WIZ and ZNF644 by generating series of internal deletion mutants of ZNF644 and WIZ , respectively ( Figure 2G , H ) . The D1 mutant of ZNF644 abolished the interaction with G9a ( Figure 2G ) , suggesting that the N-terminus of ZNF644 interacts with the TAD of G9a . In contrast , the C-terminus of WIZ is required for the interaction with GLP ( Figure 2H ) . Taken together , with the analyses on the internal deletion mutants , we found that the N-terminus of ZNF644 interacts with the TAD of G9a , while the C-terminus of WIZ interacts with the TAD of GLP ( Figure 2I ) . Since both ZNF644 and WIZ have multi zinc-finger motifs , it is likely that both ZNF644 and WIZ regulate the function of G9 and GLP . Since zinc finger motif is a DNA-binding module ( Klug and Rhodes , 1987 ) , we ask if ZNF644 and WIZ associate with chromatin . We lysed cells with NETN100 solution ( 0 . 5% NP-40 , 2 mM EDTA , 10 mM Tris–HCl pH 8 . 0 , and 100 mM NaCl ) . However , both ZNF644 and WIZ could not be eluted into soluble fraction under low salt conditions ( Figure 3A ) . Interestingly , after treating the insoluble pellets with Benzonase to digest the genomic DNA , both ZNF644 and WIZ were eluted into the soluble fraction , suggesting that ZNF644 and WIZ are chromatin-bound proteins . Usually , chromatin-bound proteins could be eluted from genomic DNA by 300 mM NaCl treatment ( Zhang et al . , 2009; Chen et al . , 2013 ) . With increased sodium concentration in the lysis buffer , ∼ 50% of ZNF644 was eluted out . However , only a small fraction of WIZ could be eluted from the chromatin , and the remaining WIZ was still tightly associated with genomic DNA ( Figure 3A ) . Thus , these results indicate that both ZNF644 and WIZ tightly bind to chromatin . It is also consistent with our affinity purification results that WIZ is slightly difficult to be identified in the mass spectrometry analyses . 10 . 7554/eLife . 05606 . 007Figure 3 . ZNF644 and WIZ are important for the chromatin localization of G9a . ( A ) Both ZNF644 and WIZ tightly bind to chromatin . U2OS cells were lysed by NETN100 ( lysis buffer with 100 mM NaCl ) and NETN300 ( lysis buffer with 300 mM NaCl ) respectively . After harvesting the soluble fractions , the pellets were digested by Benzonase to extract the chromatin fraction . Each fraction was examined by Western blotting . Tubulin and histone H3 were used as loading control for the soluble fraction and chromatin fraction respectively . ( B ) Knockdown of ZNF644 or/WIZ impairs the chromatin association of G9a and GLP . U2OS cells were lysed with NETN100 buffer . The soluble fraction and chromatin fraction were separated and each fraction was examined with Western blotting . Tubulin and histone H3 were used as loading control in the soluble fraction and chromatin fraction respectively . ( C ) In the cells with siRNA-resistant ZNF644 or WIZ , G9a is retained in the chromatin fraction . But the D1 mutant of ZNF644 or the D8 mutant of WIZ is unable to target G9a to chromatin . ( D ) A model shows that ZNF644 and WIZ facilitate the chromatin localization of the G9a/GLP complex . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 007 Since ZNF644 and WIZ are subunits in the G9a/GLP complex , we ask if ZNF644 and WIZ target the G9a/GLP complex to chromatin . We used siRNAs to knock down ZNF644 and/or WIZ . Lacking either ZNF644 or WIZ impaired the chromatin association of G9a and GLP ( Figure 3B ) . Moreover , when we knocked down ZNF644 and WIZ simultaneously , chromatin-bound G9a and GLP were remarkably reduced , but the levels of soluble G9a and GLP were not affected ( Figure 3B ) . Reconstituted cells with siRNA-resistant ZNF644 or WIZ retained G9a in the chromatin fraction . However , either the D1 mutant of ZNF644 or the D8 mutant of WIZ was able to target G9a to chromatin ( Figure 3C ) . Collectively , these results suggest that ZNF644 and WIZ facilitate the chromatin localization of the G9a/GLP complex ( Figure 3D ) . To determine if ZNF644 and WIZ target G9a to specific genomic loci , we performed high-throughput ChIP sequencing ( ChIP-seq ) to examine the genome-wide localization of G9a , ZNF644 , and WIZ in 293T cells . We identified 14 , 153 G9a enriched regions , 12 , 777 ZNF644 enriched regions , and 11 , 853 WIZ enriched regions , respectively . The ChIP-seq results were validated using ChIP-qPCR to examine 30 randomly picked loci that represent a broad range of ChIP-seq fragment counts ( Figure 4—figure supplement 1 ) . To analyze genome-wide distribution of those enriched regions , the whole genome was partitioned into three regions: intragenic region , promoter region ( 5 kb upstream or downstream of the TSS ) , and distal intergenic region not encoding any genes ( Figure 4—figure supplement 2 ) . Approximately , 40% of G9a peaks , 45% of ZNF644 peaks , and 43% of WIZ peaks were distributed in gene promoter region ( Figure 4A ) . We found around 54% of WIZ-enriched regions were bound by G9a , and around 58% of ZNF644 enriched regions were bound by G9a , while around 63% of G9a enriched regions were bound by ZNF644 and/or WIZ ( Figure 4B ) . These results indicate that most G9a-enriched regions are associated with ZNF644 and/or WIZ , which is in agreement with our results that the chromatin loading of G9a is dependent on the ZNF644 and/or WIZ . It has been shown that G9a regulates gene transcription via catalyzing H3K9me2 at promoter regions ( Su et al . , 2004; Barski et al . , 2007; Kubicek et al . , 2007; Chen et al . , 2012; Fang et al . , 2012 ) . Thus , we analyzed G9a peaks in promoter regions and found that around 82% of G9a-enriched peaks in promoter region were bound by ZNF644 and/or WIZ ( Figure 4C ) . Further analyses across G9a peaks in promoter regions show that ZNF644 and WIZ profiles are also associated with the G9a profiles in promoter region ( Figure 4D ) . Thus , accumulated evidence suggests that G9a is clearly associated with ZNF644 and WIZ , especially in promoter regions . To further analyze the co-localization of ZNF644 and WIZ with G9a at specific gene loci , we studied several genes with promoter enrichment of G9a . At CWH43 , DIP2C and ROCK1 loci , G9a , ZNF644 and WIZ co-localized together at the promoter regions ( Figure 4E ) . At CACNA2D1 , ANKRD26P1 , USP14 , and HCN1 loci , only ZNF644 , but little WIZ , significantly co-localized with G9a in the promoter regions ( Figure 4F , Figure 4—figure supplement 3A ) . In contrast , at PARD3 , ABCA13 , SENP5 , and NRXN3 loci , WIZ , but little ZNF644 , co-localized with G9a ( Figure 4G , Figure 4—figure supplement 3B ) . Taken together , ZNF644 and/or WIZ associate with G9 at the promoter regions of specific loci . 10 . 7554/eLife . 05606 . 008Figure 4 . WIZ and ZNF644 associate with G9a at specific genomic loci . ( A ) Summary of genome-wide distribution of G9a , ZNF644 and WIZ in different regions . Y-axes: percentage of each region in the genome . ( B ) Venn diagram shows a significant overlap between G9a , ZNF644 and WIZ enriched peaks . ( C ) The G9a-enriched peaks were bound with ZNF644 and/or WIZ , especially in promoter region . ( D ) G9a , ZNF644 and WIZ ChIP-seq read counts in 100-bp window were plotted against the distance ( −2 kb , +2 kb ) from the center of G9a enriched regions in promoter region . Y-axes: mean tag density . ( E ) ChIP-seq results show the co-occupancy of ZNF644 , WIZ and G9a at CWH43 , DIP2C and ROCK1 loci . ( F ) ZNF644 and G9a are co-localized at the promoter regions of CACNA2D1 , ANKRD26P1 , USP14 and HCN1 . ( G ) WIZ and G9a are co-localized at the promoter regions of PARD3 , ABCA13 , SENP5 and NRXN3 . ( H ) The consensus DNA-binding motif of ZNF644 is analyzed according to ChIP-seq result . The binding sequences in CACNA2D1 , ANKRD26P1 , USP14 and HCN1 loci are shown in red . ( I ) The specific DNA binding sequence of WIZ is obtained according to the ChIP-seq results , and is confirmed at PARD3 , ABCA13 , SENP5 and NRXN3 loci . ( J ) Both ZNF644 and WIZ-binding sequences are identified at CWH43 , DIP2C and ROCK1 loci , which are co-occupied by ZNF644 and WIZ . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 00810 . 7554/eLife . 05606 . 009Figure 4—figure supplement 1 . Validation of ChIP-seq results by qPCR . ChIP-seq fragment densities of G9a ( x-axis ) are plotted against ChIP-qPCR fold-enrichment of G9a ( percentage of input ) ( y-axis ) at 30 selected loci in 293T cells that represent a broad range of ChIP-seq fragment counts . The 30 selected loci contain 10 loci that are G9a positive ( ChIP-seq signal >5 ) , 10 loci that are G9a , ZNF644 and WIZ positive , and the other 10 loci are G9a negative . The same methods were used to analyze the ChIP-seq result of ZNF644 and WIZ . 20 loci identified as significantly enriched by ChIP-seq were clearly different from 10 unenriched loci in the plots . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 00910 . 7554/eLife . 05606 . 010Figure 4—figure supplement 2 . Genome-wide analysis of ChIP-seq peaks . Average genome-wide occupancies of G9a , ZNF644 and WIZ along the transcription unit . TSS and TES , the transcription start and end sites , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 01010 . 7554/eLife . 05606 . 011Figure 4—figure supplement 3 . The gene loci occupied by ZNF644 or WIZ are confirmed by ChIP-qPCR . ( A ) ChIP-qPCR confirms the occupancy of ZNF644 , but not WIZ , at CACNA2D1 , ANKRD26P1 , USP14 and HCN1loci . p < 0 . 05; n . s . , not significant . ( B ) The occupancy of WIZ , but not ZNF644 , is shown at PARD3 , ABCA13 , SENP5 and NRXN3 loci . p < 0 . 05; n . s . , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 01110 . 7554/eLife . 05606 . 012Figure 4—figure supplement 4 . DNA binding motifs of ZNF644 or WIZ concluded from ChIP-seq results were validated by Electrophoretic Mobility Shift Assay ( EMSA ) . ( A ) GST-tagged full length ZNF644 ( GST-ZNF644 ) , N-terminus of ZNF644 ( a . a . 1-300 ) ( ZNF644N300 ) , full length human WIZ ( GST-WIZ ) or N-terminus of WIZ ( a . a . 1-200 ) ( WIZN200 ) were purified from Sf9 insect cells and used for EMSA . The proteins were purified by GST beads and examined by coomassie blue staining . ( B ) Recombinant GST-ZNF644 or N terminus of ZNF644 without Zinc finger motif ( GST-ZNFN300 ) was incubated with 32P-labeled 48-mer sequence motif-contained DNA oligonucleotides . Only GST-ZNF644 , but not ZNF644N300 , could bind to the DNA containing sequence motif . The 32P-labeled 48-mer DNA oligonucleotides containing “mutant” DNA target was used as the negative control . ( C ) Recombinant GST-WIZ or N terminus of WIZ without Zinc finger motif ( GST-WIZN200 ) was incubated with 32P-labelled 48-mer sequence motif-contained DNA oligonucleotides . Only GST-WIZ , but not GST-WIZN200 , could bind to the DNA containing consensus motif . The 32P-labeled 48-mer DNA oligonucleotides containing “mutant” DNA target was used as the negative control . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 012 Since ZNF644 and WIZ have 8 and 12 zinc finger motifs , respectively , we further analyzed the DNA-binding sequences of ZNF644 and WIZ based on the information from ChIP-seq . Using Peak-motifs software ( http://floresta . eead . csic . es/rsat/peak-motifs_form . cgi ) , we examined only ZNF644-enriched regions ( lacking WIZ ) , by which we excluded the possible loading of ZNF644 onto chromatin via WIZ . A specific DNA-binding sequence was concluded by the software ( Figure 4H ) , and this sequence was confirmed at CACNA2D1 , ANKRD26P1 , USP14 , and HCN1 loci only occupied by ZNF644 . Similarly , a specific DNA binding sequence of WIZ was also obtained from software analyses ( Figure 4I ) and was confirmed at PARD3 , ABCA13 , SENP5 , and NRXN3 loci . Moreover , both ZNF644 and WIZ-binding sequences were identified at the loci co-occupied by ZNF644 and WIZ , such as CWH43 , DIP2C , and ROCK1 loci ( Figure 4J ) . We performed electrophoretic mobility shift assays ( EMSA ) and found that the consensus DNA-binding motifs of ZNF644 and WIZ showed strong binding with full-length recombinant proteins ( Figure 4—figure supplement 4 ) . Since the G9a/GLP complex catalyzes methylation of H3K9 in euchromatin and represses gene transcription ( Tachibana et al . , 2005 , 2008 ) , we next explored the function of ZNF644 and WIZ in G9a-dependent gene transcriptional repression with ChIP assays . In agreement with the ChIP-seq results , ZNF644 , WIZ , and G9a localized at CWH43 , DIP2C , and ROCK1 loci ( Figure 5—figure supplement 1 ) . Moreover , down-regulation of ZNF644 and WIZ impaired the localization of G9a at these loci ( Figure 5A ) , suggesting that ZNF644 and WIZ are important for targeting G9a to this specific loci . However , down-regulation of G9a did not affect the chromatin localization of ZNF644 and WIZ at these loci ( Figure 5—figure supplement 2 ) . Since G9a catalyzes H3K9me2 that is recognized by HP1α , we found that both H3K9me2 and HP1α were enriched at these loci . Lacking G9a abolished the enrichment of H3K9me2 and HP1α ( Figure 5B ) . Similarly , loss of ZNF644 and WIZ also impaired the enrichment of H3K9me2 and HP1α ( Figure 5B ) , suggesting that ZNF644 and WIZ are important for the G9a-dependent H3K9 methylation at G9a targeting genes . Since G9a mainly occupied the promoter regions at these gene loci , G9a-dependent H3K9 methylation is likely to repress gene transcription . Down-regulation of G9a by siRNA indeed increased gene transcription at these loci ( Figure 5C ) . Again , loss of ZNF644 and WIZ also facilitated gene transcription ( Figure 5C ) . Taken together , these results suggest that ZNF644 and WIZ regulate the function of G9a during transcription . Moreover , we examined only ZNF644 or WIZ occupied gene loci , and similar results were obtained ( Figure 5—figure supplement 3 ) . 10 . 7554/eLife . 05606 . 013Figure 5 . ZNF644 and WIZ target G9a for gene repression . ( A ) Down-regulation of ZNF644 and WIZ by siRNAs ( siZ + W ) impairs the localization of G9a at CWH43 , DIP2C and ROCK1 loci . p < 0 . 05 compared to IgG . ( B ) Knockdown G9a abolishes the enrichment of H3K9me2 and HP1α at CWH43 , DIP2C and ROCK1 loci . Loss of ZNF644 and WIZ also impairs the enrichment of H3K9me2 and HP1α at these loci . p < 0 . 05 compared to IgG . ( C ) Down-regulation of G9a by siRNA increases gene transcription at CWH43 , DIP2C and ROCK1 loci , and loss of ZNF644 and WIZ also facilitates gene transcription at these loci . p < 0 . 05 compared to Mock . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 01310 . 7554/eLife . 05606 . 014Figure 5—figure supplement 1 . Co-occupancy of ZNF644 , WIZ and G9a is shown at CWH43 , DIP2C and ROCK1 loci . ChIP-qPCR was performed to confirm the ChIP-seq results . p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 01410 . 7554/eLife . 05606 . 015Figure 5—figure supplement 2 . Down-regulation of G9a does not affect the chromatin localization of ZNF644 ( A ) and WIZ ( B ) at CWH43 , DIP2C and ROCK1 loci . ChIP-qPCR was performed in the siG9a-treated U2OS cells . p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 01510 . 7554/eLife . 05606 . 016Figure 5—figure supplement 3 . ZNF644 and WIZ target G9a for gene repression at only ZNF644 or WIZ occupied gene loci . ( A ) Knockdown ZNF644 by siZNF644 impairs the recruitment of G9a at CACNA2D1 , ANKRD26P1 , USP14 and HCN1 loci . Knockdown WIZ abolishes the localization of G9a at PARD3 , ABCA13 , SENP5 and NRXN3 loci . p < 0 . 05 compared to IgG . ( B ) Knockdown ZNF644 facilitates the gene transcription at CACNA2D1 , ANKRD26P1 , USP14 and HCN1 loci , while knockdown WIZ induces the gene transcription at PARD3 , ABCA13 , SENP5 and NRXN3 loci . p < 0 . 05 compared to Mock . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 016 To study if the interaction domains in ZNF644 and WIZ are important for the G9a-dependent function , we reconstituted siRNA-treated cells with siRNA-resistant ZNF644 and WIZ . ZNF644 and WIZ rescued the recruitment of G9a to the gene loci , facilitated the enrichment of H3K9me2 and HP1α , and repressed gene transcription ( Figure 6 ) . However , expression of the D1 mutant of ZNF644 and the D8 mutant of WIZ that abolish the interactions with G9 and GLP , failed to restore the enrichment of H3K9me2 and HP1α as well as transcription repression ( Figure 6 ) . Thus , our results demonstrate that ZNF644 and WIZ are two key subunits in the G9/GLP complex to target G9a and GLP to genomic loci for transcriptional repression . 10 . 7554/eLife . 05606 . 017Figure 6 . The interaction domains in ZNF644 and WIZ are important for the G9a-dependent function . ( A ) Wild-type ZNF644 and WIZ , but not the D1 mutant of ZNF644 and the D8 mutant of WIZ , rescue the recruitment of G9a to CWH43 , DIP2C and ROCK1 loci . p < 0 . 05 , *p < 0 . 01 compared to IgG . ( B-D ) Wild-type ZNF644 and WIZ , but not the D1 mutant of ZNF644 and the D8 mutant of WIZ , restore the enrichment of H3K9me2 and HP1α as well as gene transcription at CWH43 , DIP2C and ROCK1 loci . p < 0 . 05 , **p < 0 . 01 compared to IgG ( B , C ) or the control U2OS cells without siRNAs treatment ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05606 . 017 In this study , with unbiased protein affinity purification , we identified ZNF644 and WIZ as two core subunits in the G9a/GLP complex . With the analyses on the internal deletion mutants , we found that the N-terminus of ZNF644 interacts with the TAD of G9a , while the C-terminus of WIZ interacts with the TAD of GLP . In the previous study ( Ueda et al . , 2006 ) , WIZ was found to interact with the catalytic domain of G9a . We obtained the similar result ( Figure 2C ) . However , the catalytic domain of G9a also interacts with the catalytic domain of GLP to form a heterodimer . Lacking G9a did not impair the interaction between WIZ and GLP ( Figure 2—figure supplement 1B ) , suggesting that the interaction between WIZ and G9a might be indirect and mediated by the catalytic domain heterodimer of G9a and GLP . The G9a/GLP complex is known to associate with specific gene promoter and catalyze local H3K9 methylation for transcription repression ( Tachibana et al . , 2005 , 2008 ) . However , both G9a and GLP lack the DNA recognition domains for targeting specific genes . Here , we show the evidence that ZNF644 and WIZ , two multi zinc finger motif-containing proteins , mediate the recruitment of G9a and GLP to the specific gene loci . ZNF644 has 8 zinc finger motifs , while WIZ contains 12 zinc finger motifs . It is likely that these zinc finger motifs function together to recognize specific DNA sequences . Based on our ChIP-seq analyses , we identified the DNA sequences recognized by ZNF644 and WIZ . Thus , ZNF644 and WIZ act as two hands to grab the genomic DNA and target enzymatic subunits G9 and GLP for H3K9 methylation . It is also interesting to notice that ZNF644 and WIZ bind the TADs of G9a and GLP , respectively . Thus , the complex may form a symmetric structure with ZNF644 and G9a on one side , and WIZ and GLP on the other side . These two parts are linked by the interaction between the catalytic domains of G9a and GLP in the middle ( Figures 2I and 3D ) . The double DNA recognition may reduce the flexibility of the complex on the chromatin and allow the G9a and GLP to precisely catalyze histone methylation at gene loci . In the ChIP-seq analysis , we also notice that a fraction of G9a and GLP only associate with either ZNF644 or WIZ ( Figure 4A , B ) . Thus , it is possible that ZNF644 or WIZ alone is sufficient for targeting G9a and GLP to certain loci for H3K9 methylation and transcription repression . Since ZNF644 and WIZ bind different DNA sequences , in these cases , different DNA-binding subunits target G9a and GLP to different loci . With only one DNA-binding arm , the G9a/GLP complex may have more flexibility to methylate targets . Alternatively , we cannot rule out the possibility that a small amount of G9a or GLP only form homodimers . Since ZNF644 only recognizes the TAD of G9a , and WIZ interacts with the TAD of GLP , only ZNF644 or WIZ is sufficient to target the homodimer to the chromatin . Moreover , a small set of G9a associates with neither ZNF644 nor WIZ . Thus , it is possible that a small set of G9a may interact with other regulators . It has been shown that G9 associates with other zinc finger proteins , such as Blimp-1 , which may also target G9a to the substrates ( Gyory et al . , 2004 ) . However , Blimp-1 is mainly expressed in plasma cells as the major function of Blimp-1 is to regulate plasma cell differentiation ( Shaffer et al . , 2002 ) . Thus , Blimp-1 mainly regulates G9a's activity in plasma cells . It has been shown that G9a regulates gene transcription via catalyzing H3K9me2 at promoter regions ( Su et al . , 2004; Barski et al . , 2007; Kubicek et al . , 2007; Chen et al . , 2012; Fang et al . , 2012 ) . Consistently , we found that G9a associated with ZNF644 and WIZ , especially in the promoter regions , to regulate transcription . Interestingly , Wen et al . 2009 examined H3K9me2-enriched loci in the differentiated tissues and found that large chromatin regions associate with H3K9me2 . These regions were named as large organized chromatin K9 modifications ( LOCKs ) . However , the function of LOCKs remains unclear . Interestingly , LOCKs are dynamically regulated during development , and the size of LOCKs varies in different types of cells during differentiation , suggesting that LOCKs might be regulated by not only histone methyltransferases but also demethylases . It is possible that the G9a complex plays a key role for LOCKs formation . However , LOCKs do not exist in cancer cells ( Wen et al . , 2009 ) . Future analysis of the G9a complex during tissue development and differentiation may reveal the mechanism and function of LOCKs . It is possible that , besides ZNF644 and WIZ , other functional partners of G9a regulate LOCKs . Nevertheless , in this study , we have demonstrated that ZNF644 and WIZ are two major functional partners of G9a and GLP . ZNF644 and WIZ target the G9a/GLP complex to genomic loci for H3K9 methylation and transcription repression . Full-length cDNA of G9a , GLP , and WIZ was cloned into pS-FLAG-SBP ( SFB ) vector , respectively , the full-length cDNA of ZNF644 , G9a , and GLP was cloned into pCMV-Myc vector , and the full-length cDNA of WIZ was also cloned into pCMV-HA vector . For protein co-immunoprecipitation experiments , G9a deletion mutants , WIZ deletion mutants , and GLP deletion mutants were cloned into SFB vector , respectively . ZNF644 deletion mutants were cloned into the pCMV-Myc vector . Primary antibodies used in this study include: mouse anti-G9a monoclonal antibody ( Abcam , Cambridge , UK ) , mouse anti-HA and anti-Myc monoclonal antibodies ( Covance , Princeton , NJ ) , rabbit anti-H3K9me2 polyclonal antibody ( Upstate , Billerica , MA ) , rabbit anti-human ZNF644 antibody ( raised against N-terminus a . a . 50–602 ) , rabbit anti-human WIZ antibody ( raised against N-terminus a . a . 220–750 ) . The siRNAs targeting G9a , GLP , ZNF644 , and WIZ were ordered from Dharmacon ( Lafayette , CO ) . Purification of SFB triple-tagged protein was described previously ( Zhang et al . , 2009 ) . To search for binding partners of G9a or ZNF644 , we harvested 50 10 cm2 plates of 293T cells stably expressing SFB-G9a or ZNF644 and washed cells with PBS . Cells were lysed with 30 ml ice-cold NETN300 buffer ( 0 . 5% NP-40 , 50 mM Tris–HCl pH 8 . 0 , 2 mM EDTA , and 300 mM NaCl ) . The soluble fraction was incubated with 0 . 5 ml streptavidin-conjugated agarose beads . The beads were washed with NETN buffer three times . Associated proteins were eluted with 2 mM biotin in PBS and further incubated with 50-ml S beads ( Novagen , Billerica , MA ) . The bound proteins were eluted with SDS sample and analyzed with 10% SDS-PAGE and mass spectrometry . For immunoprecipitation assays , 293T cells or U2OS cells were lysed with ice-cold NETN400 buffer ( 0 . 5% NP-40 , 50 mM Tris–HCl pH 8 . 0 , 2 mM EDTA , and 400 mM NaCl ) containing 10 mM NaF and 50 mM β-glycerophosphate . The soluble fractions were collected and diluted to 100 mM NaCl , then directly subjected to electrophoresis or immunoprecipitation with indicated antibodies followed by Western blotting analysis with indicated antibodies . For the SFB-tagged protein , streptavidin beads were used to perform the pull-down assay followed by Western blotting analysis . Chromatin immunoprecipitation assays ( ChIP ) were performed according to the protocol described by Upstate ( Billerica , MA ) . The genomic DNA isolated from 293T cells was sonicated to an average size between 300 and 600 bp . Solubilized chromatin was immunoprecipitated with the antibody against WIZ , G9a , or ZNF644 . Antibody–chromatin complexes were pulled-down using protein A-sepharose , washed , and then eluted . After cross-link reversal and proteinase K treatment , immunoprecipitated DNA was extracted with phenol-chloroform , ethanol precipitated , treated with RNase , and dissolved with TE buffer . ChIP DNA was qualified using PicoGreen . DNA fragments isolated from ChIP were repaired to blunt ends by T4 DNA polymerase and phosphorylated with T4 polynucleotide kinase using the END-IT kit ( Epicentre , Madison , WI ) . A single ‘A’ base was added to 3′ end with Klenow . Double-stranded adaptors ( 75 bp with a ‘T’ overhang ) were ligated to the fragments with DNA ligase . Ligation products between 200 and 600 bp were gel purified to remove unligated adaptors and subjected to 20 PCR cycles . Completed libraries were quantified with PicoGreen . The DNA libraries were analyzed by Solexa/Illumina high-throughput sequencing . The read quality of each sample was determined by FastQC software . After prefiltering the raw data by removing sequence adaptors and low quality reads , the tags were mapped to the human genome ( hg19 ) by Bowtie software . Parameters settings were listed as follows: -v , 3 ( reported alignments with at most 3 mismatches ) , -5 , 3 and -3 , 7 ( trim 3 bases from 5′ end and 7 from 3′ end to remove low-quality bases ) . Peak detection was performed using MACS software from Galaxy browser ( http://galaxyproject . org/ ) . Parameters settings were as follows: IgG ChIP-seq aligned reads were used as control file , tag size with 25 bp , band width with 300 bp . When comparing peaks from different samples , peaks were considered to be overlapping if they were within 2 kb of each other . The peaks obtained from ChIP-seq were matched to the annotated reference genome ( human hg19 ) using Cisgenome 2 . 0 . To view the peak density and position , Cisgenome 2 . 0 was used . To obtain the binding motif of ZNF644 and WIZ , the online software Peak-motifs http://floresta . eead . csic . es/rsat/peak-motifs_form . cgi ) was used . A set of 30 PCR primer pairs ( Supplementary file 1 ) were designed to amplify ∼200 bp fragments from genomic regions showing a wide range of signals for G9a , ZNF644 , and WIZ . ChIP-qPCR values reflect two independent ChIP assays , and each was evaluated in duplicate by qPCR . To examine the genome distribution of G9a , ZNF644 , and WIZ , the whole genome was partitioned into three regions: intragenic region , promoter region ( 5 kb upstream or downstream of the TSS ) , and distal intergenic region that does not encode any genes . Genes not uniquely mapped to the genome were excluded . To avoid redundancy , only the longest transcript variant of each gene was used to define chromosomal locations of the intragenic region , promoter region , and intergenic region . The read counts around the center of G9a-enriched peaks in promoter region were analyzed by SEQMINER software ( Ye et al . , 2011 ) . The center of G9a-enriched peaks in promoter region was used as the reference . Tag densities from each ChIP-seq were collected within a window of 4 kb around reference coordinates . The tag density of each ChIP-seq in a 200 bp window was calculated and plotted against distance from the center . For Figure 4—figure supplement 2 , genes were profiled 5 kb upstream of the transcriptional start site ( TSS ) , through the gene body and 5 kb downstream of the transcriptional end site ( TES ) . 5 kb upstream of the TSS and 5 kb downstream from the TES were divided into windows of 200 bp , and read counts were calculated in each window . For gene body plots , each gene was segmented into 300 non overlapping windows . Plots were made using a 1 kb moving average . Values are tag-normalized and reflect the number of tags observed in each window . ChIP-seq data have been deposited in the Gene Expression Omnibus under accession number GSE62616 . Recombinant proteins were purified from Sf9 insect cells . For generating baculovirus , DNA fragments containing full-length human ZNF644 , N-terminus of ZNF644 ( a . a . 1–300 ) ( ZNF644N300 ) , full-length human WIZ , and N-terminus of WIZ ( a . a . 1–200 ) ( WIZN200 ) were subcloned into pFastBac Vector with a GST tag . Baculoviruses were generated in accordance with the manufacturer's instructions ( Invitrogen , Carlsbad , CA ) . After Sf9 cells were infected with baculoviruses for 48 hr , the cells were harvested , washed with cold PBS three times and lysed with ice-cold NETN100 buffer ( 20 mM Tris–HCl pH 8 . 0 , 100 mM NaCl , 1 mM EDTA , 0 . 5% Nonidet P-40 ) . The soluble fraction was incubated with Glutathione–Sepharose beads and eluted with Glutathione . Oligonucleotide substrates were obtained from IDT ( IDT , Coralville , IA ) and were purified by polyacrylamide gel electrophoresis ( PAGE ) . The following oligonucleotides containing ZNF644 binding motif and WIZ binding motif were used , 5′- GAGTAAGATCATGCCACTGGGAATCATCGAACACAGAGTGAGGCTGGG -3′ ( ZNF644 ‘WT’ DNA target ) ; 5′- GAGTCTCACTCACGCGCCATTCCATTCCATTCAGATACTAGTACGGTCAG -3′ ( WIZ ‘WT DNA target’ ) . The oligonucleotides containing ZNF644 binding motif mutation and WIZ binding motif mutation were used as control , 5′- GAGTAAGATCATGCCACTGGCATTGTTGACTCACAGAGTGAGGCTGGG -3′ ( ZNF644 ‘mutant’ DNA target ) ; 5′- GAGTCTCACTCACGCGCTGCAATCAGGAACAGATACTAGTACGGTCAG -3′ ( WIZ ‘mutant’ DNA target ) . 48-mer oligonucleotides were annealed at 1:1 molar ratio to its complementary oligonucleotides to generate the dsDNA and then radio-labeled with 32P at the 5′-end . GST-ZNF644 , GST-ZNF644N300 , GST-WIZ , or GST-WIZN200 was incubated with 0 . 2 nM ( molecules ) radio-labeled DNA substrates for 2 hr at 4°C in buffer D ( 20 mM HEPES-KOH ( pH 7 . 9 ) , 20% glycerol ( vol/vol ) , 0 . 2 mM EDTA , 0 . 1 M KCl , 0 . 5 mM PSMF , 1 mM DTT ) with 1 . 25 μg/μl Bovine serum albumin , 1 mM DTT , 5 mM MgCl2 . The samples were resolved by electrophoresis on a 7 . 5% polyacrylamide gel in TBE buffer for 70 min at 60 V . The gel was then dried and exposed to autoradiography film overnight . In all cases , multiple independent experiments were performed on different days to verify the reproducibility of experimental findings . Two-way comparison was performed using the t-test , and ANOVA was used for more than two groups . For all analyses , a p value of less than 0 . 05 was considered significant . Results are given as means ± s . d .	Genes encode instructions for processes within cells , but only a small subset of the genes within a cell will be switched on ( or expressed ) at any given time . The other genes are kept switched off until their instructions are needed . For example , some genes are switched on when it is time for a cell to divide or in response to changes in the environment . In humans and other eukaryotes , DNA is packaged within cells in proteins called histones . The level of gene expression can be altered by how tightly the DNA is packaged; if the DNA is more tightly packed around the histones , the gene will be expressed at lower levels than if the DNA is only loosely packed . A group of proteins called the G9a/GLP complex can alter histones to reduce the expression of some genes during embryo development , immune responses , and the formation of tumors . The complex works by attaching ‘methyl’ tags to the histones associated with particular genes , but it is not clear how it is able to specifically target these histones . Bian , Chen , and Yu used a technique called unbiased protein affinity purification to search for other proteins that can bind to the G9a/GLP complex . The experiments found two proteins called ZNF644 and WIZ , both of which are required for the G9a/GLP complex to be able to add methyl tags to histones . Further experiments revealed that ZNF644 and WIZ both contain regions called zinc finger motifs that enable them to identify and bind to specific sequences of DNA . Therefore , these proteins can guide the G9a/GLP complex to specific sites in the genome to switch off the expression of particular genes . A future challenge will be to try to modify these zinc finger motifs and guide the G9a/GLP complex to switch off other genes . This may allow us to develop therapies that could alter the expression of genes involved in cancer and other diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology" ]	2015	The zinc finger proteins ZNF644 and WIZ regulate the G9a/GLP complex for gene repression
Survival of Trypanosoma brucei depends upon switches in its protective Variant Surface Glycoprotein ( VSG ) coat by antigenic variation . VSG switching occurs by frequent homologous recombination , which is thought to require locus-specific initiation . Here , we show that a RecQ helicase , RECQ2 , acts to repair DNA breaks , including in the telomeric site of VSG expression . Despite this , RECQ2 loss does not impair antigenic variation , but causes increased VSG switching by recombination , arguing against models for VSG switch initiation through direct generation of a DNA double strand break ( DSB ) . Indeed , we show DSBs inefficiently direct recombination in the VSG expression site . By mapping genome replication dynamics , we reveal that the transcribed VSG expression site is the only telomeric site that is early replicating – a differential timing only seen in mammal-infective parasites . Specific association between VSG transcription and replication timing reveals a model for antigenic variation based on replication-derived DNA fragility . The growth and propagation of pathogens in vertebrates requires strategies to survive the host immune responses , in particular adaptive immunity . One such survival strategy , found widely in biology , is antigenic variation , which involves periodic switches in exposed pathogen antigens , thereby allowing a fraction of the infecting population to escape immune clearance . A number of strategies for antigenic variation have been described , though normally only one is employed in any given pathogen . In this regard , antigenic variation in the African trypanosome , Trypanosoma brucei , is unusual , since here two apparently distinct approaches are adopted: recombination and transcription . Antigenic variation in T . brucei involves switches in the identity of the Variant Surface Glycoprotein ( VSG ) expressed on the cell surface , where the protein forms a dense ‘coat’ that is believed to shield invariant antigens from immune recognition ( Higgins et al . , 2013 ) . At any given time an individual T . brucei cell in the mammal expresses only one VSG gene , due to transcriptional control mechanisms that ensure only one of ~15 VSG transcription sites , termed bloodstream expression sites ( BES ) , is active . Such monoallelic expression is found in other antigenic variation systems , such as that involving the ~60 var genes in Plasmodium falciparum ( Guizetti and Scherf , 2013 ) , as is the ability to switch the gene that is actively transcribed , eliciting antigenic variation . The nature of the monoallelic control and transcriptional switch mechanisms in T . brucei , and whether they share features with other pathogens , are still being unraveled ( Horn , 2014 ) . One complexity in understanding VSG transcriptional switching is the elaborate structure of the BES ( Hertz-Fowler et al . , 2008 ) , where the VSG is co-transcribed with many other genes , termed expression site-associated genes ( ESAGs ) , from an RNA Polymerase I promoter . Despite some variation in ESAG composition between BES , two features appear invariant in all these sites: the VSG is always proximal to the telomere and is separated from the upstream ESAGs by an array of 70 bp repeats ( which appear to be only found adjacent to VSGs in the T . brucei genome ) ( Marcello and Barry , 2007 ) . Transcriptional switching occurs between the VSGs that occupy the BES , and is therefore limited by BES number . However , a second route of VSG switching relies upon recombination and can over-write the BES-resident VSGs , generating new VSG coats from a genomic archive of ~2000 VSGs ( Cross et al . , 2014; Marcello and Barry , 2007 ) . In numerical terms , therefore , recombination is the major route for VSG switching . Indeed , recombination is a very widespread strategy for antigenic variation in eukaryotic and bacterial pathogens ( Palmer and Brayton , 2007 ) , most likely because it drives antigen diversity , which prolongs infection and facilitates transmission ( Hall et al . , 2013; Mugnier et al . , 2015 ) . The VSG archive is distributed across the three chromosome classes that comprise the T . brucei nuclear genome . A small part of the archive is the BES ( Hertz-Fowler et al . , 2008 ) , which are found in the 11 diploid megabase chromosomes as well as in the ~5 aneuploid intermediate chromosomes . A larger part of the archive is found at the telomeres of ~100 minichromosomes ( Wickstead et al . , 2004 ) , where ESAGs and BES promoters have not been found , suggesting this part of the archive is simply a store of silent VSGs for recombination . The largest silent store is composed of arrays of VSGs in the subtelomeres of the megabase chromosomes , where the majority of the VSGs are pseudogenes or partial genes ( Berriman et al . , 2005 ) . The strategies for VSG recombination in antigenic variation reflect the archive location and gene composition ( McCulloch et al . , 2015 ) . A minor route for switching is termed reciprocal VSG recombination , where telomeres are exchanged between two chromosomes , moving the VSG out of the active BES and moving a previously silent VSG into the active BES ( Rudenko et al . , 1996 ) . More common is VSG gene conversion , which can involve both intact and impaired VSGs , and involves deletion of the VSG in the BES and replacement by VSG sequence copied from the silent archive . Early in infections gene conversion of intact VSGs predominates ( Marcello and Barry , 2007; Morrison et al . , 2005 ) and , since the VSGs share little sequence homology , the reaction relies on flanking homology . ~90% of VSGs are flanked by 70 bp repeats ( Marcello and Barry , 2007 ) , which provide upstream homology to guide recombination of virtually all genes in the archive . In addition , gene conversion of VSGs between BES can use extensive upstream homology: gene conversion can extend to downstream homology within and around the VSG open reading frame ( ORF ) or , if the silent VSG is telomeric , to the chromosome end . Impaired VSGs are seen as recombination substrates later in infections and here gene conversion differs from intact VSGs , since the reaction involves the production of a functional gene using homology within the ORF; indeed , multiple VSG donors are frequently recombined to generate novel ‘mosaic’ VSGs in a reaction termed segmental gene conversion ( Hall et al . , 2013; Mugnier et al . , 2015 ) . All available evidence suggests switching of intact VSGs by recombination is catalyzed by homologous recombination ( HR ) , a universally conserved reaction that directs repair of DNA damage and maintains replication fork progression genome-wide . Mutation of the central catalytic enzyme of HR , RAD51 , impairs ( but does not abolish ) VSG recombination , including gene conversion ( McCulloch and Barry , 1999 ) . Consistent with that phenotype , mutation of T . brucei BRCA2 ( Hartley and McCulloch , 2008 ) and at least one of four RAD51 paralogues ( Dobson et al . , 2011 ) - factors that aid RAD51 function - has the same outcome . More recently , mutation of TOP3α or RMI1 , which interact and may be components of the T . brucei RTR ( RecQ/Sgs1-Top3/TOPO3α- Rmi1/BLAP75/18 ) complex ( Mankouri and Hickson , 2007 ) , was shown to result in increased VSG switching , an effect that is RAD51-dependent ( Kim and Cross , 2010; 2011 ) . The conclusion that antigenic variation can be executed by a non-specific , general repair pathway is not limited to T . brucei , as similar gene knockout studies in Neisseria gonorrhoeae implicate HR in pilin antigenic variation ( Cahoon and Seifert , 2011 ) . However , VSG switching can occur at rates substantially higher than might be predicted for background mutation ( Turner , 1997 ) and may be focused to target the active BES , features that may suggest some mechanistic specialization or locus-specificity . As a result , recent work has explored how VSG switching might be initiated in T . brucei , leading to an association between elevated rates of switching and DNA double strand breaks ( DSBs ) . The evidence for this association is two-fold . First , controlled induction of the endonuclease I-SceI to specifically generate a DSB adjacent to the 70 bp in the active BES leads to a ~250 fold increase in VSG switching by recombination ( Boothroyd et al . , 2009 ) , an effect not seen when a DSB is induced in other locations in the active BES ( Boothroyd et al . , 2009; Glover et al . , 2013 ) or when the 70 bp repeats have been deleted from the active BES ( Boothroyd et al . , 2009 ) . Second , ligation-mediated PCR is able detect DNA breaks in the BES , with the lesions initially reported to be limited to the vicinity of the 70 bp repeats in the active BES ( Boothroyd et al . , 2009 ) , though later also reported in the silent BES ( Glover et al . , 2013; Jehi et al . , 2014 ) and found to be more widely distributed in the transcription units ( Glover et al . , 2013 ) . Despite the emerging association between DNA DSBs and VSG switching , questions remain about the detailed mechanism ( s ) of VSG switch initiation . For instance , are DSBs generated directly in the active BES , such as through the action of an endonuclease , as occurs during Saccharomyces cerevisiae mating type switching ( Lee and Haber , 2015 ) ? Alternatively , might other processes lead more indirectly to break formation and elicit switching , such as the transcription , replication and DNA nicking events that initiate locus-directed recombination reactions during , respectively , immunoglobulin gene switching in mammals ( Roth , 2014 ) , mating type switching in Schizosaccharomyces pombe ( Klar et al . , 2014 ) and pilin antigenic variation in N . gonorrhoeae ( Obergfell and Seifert , 2015 ) ? In this study , we have examined VSG switch initiation in two ways . First , we describe the impact on DNA repair and VSG switching caused by mutation of one of two T . brucei RecQ-like helicases , which we have named TbRECQ2 . We show that loss of TbRECQ2 impairs DSB repair , consistent with the observation that the protein localizes to such lesions . Conversely , TbRECQ2 mutants display elevated rates of VSG switching , indicating it is unlikely that the direct formation of DSBs is the initiating event in VSG switching . Second , we provide evidence for strong association between replication timing and BES transcription , indicating that VSG switch initiation may be mechanistically linked to DNA replication . Helicases are molecular motors that use energy released from nucleoside triphosphate hydrolysis to unwind RNA , DNA or RNA:DNA hybrids . RecQ-like helicases are widespread Superfamily 2 DNA helicases ( Fairman-Williams et al . , 2010 ) , identifiable by homology with the first RecQ helicase described in Escherichia coli ( Umezu et al . , 1990 ) . S . pombe and S . cerevisiae , single-celled yeast , encode a single RecQ helicase: Rqh1 and Sgs1 , respectively . In contrast , multicellular eukaryotes such as humans and Drosophila menaogaster possess multiple RecQ helicases , with five discernible in mammals ( Bernstein et al . , 2010; Hickson , 2003 ) . RecQ helicases function in diverse aspects of genome maintenance . For example , human RECQ4 interacts with the replication factors MCM10 and MCM2-7 , while RECQL4 interacts with CDC45 and GINS ( Im et al . , 2009; Xu et al . , 2009 ) . RecQ helicases also play roles in non-homologous end-joining , since human WRN interacts with XRCC4-ligase IV ( Kusumoto et al . , 2008 ) , while human RECQ1 binds Ku70/80 and its depletion leads to reduced repair ( Parvathaneni et al . , 2013 ) . Finally , RecQ helicases act in both the initiation and execution of HR ( Haber , 2015 ) , since the RTR complex promotes DNA DSB resection ( Mimitou and Symington , 2008; Zhu et al . , 2008 ) , controls DNA annealing during strand invasion ( Fasching et al . , 2015; Spell and Jinks-Robertson , 2004 ) , and ‘dissolves’ Holliday junction intermediates to limit crossover genetic exchange between chromosomes ( Cejka et al . , 2010; Hickson and Mankouri , 2011 ) . BLAST searches with multiple eukaryotic RecQ helicase sequences consistently revealed two well-aligned proteins encoded in the T . brucei genome , which we arbitrarily named TbRECQ1 ( TriTrypDB gene accession number Tb427 . 06 . 3580 ) and TbRECQ2 ( Tb427 . 08 . 6690 ) . RNAi analysis suggests that TbRECQ1 , which appears more distantly related to eukaryotic RecQ helicases than TbRECQ2 , is essential ( Devlin et al , unpublished ) . Protein domain predictions ( Figure 1A ) suggest that TbRECQ2 contains a conserved DEAD/DEAH box helicase domain , indicating potential ATP and nucleic acid binding activity , and a more C-terminal helicase domain that is found in helicases of multiple families ( Linder , 2006 ) . In addition , an HRDC ( helicase and RNaseD C-terminal ) domain is predicted close to the C-terminus . In contrast to the two other domains , the HRDC domain appears limited to some RecQ helicases and RNase D homologues ( Morozov et al . , 1997 ) , where it is probably involved in DNA binding ( Bachrati and Hickson , 2003 ) . However , HRDC domains are not found in all RecQ helicases; for example , three human RecQ helicases , WRN , BLM and RECQ1 , each contain an HRDC , but it is absent in human RECQ4 and RECQ5 ( Bernstein et al . , 2010 ) . Thus , the prediction that TbRECQ2 contains an HRDC might suggest a function closer to the former human RecQ helicases . One domain that is limited to RecQ helicases is termed RQC ( RecQ C-terminal ) , which may be involved in protein-protein interactions ( Bernstein et al . , 2010 ) , as well as in binding and unwinding dsDNA at branch points ( Kitano et al . , 2010 ) . Though this domain could not be predicted in TbRECQ2 , it is also absent from or highly diverged in some other validated RecQ proteins ( Bachrati and Hickson , 2003 ) and the evolutionary distance between T . brucei and the most characterised model eukaryotes may confound identification . 10 . 7554/eLife . 12765 . 003Figure 1 . T . brucei RECQ2 is non-essential and acts in genome repair . ( A ) Representation of predicted protein domains in TbRECQ2 . Approximate position ( in amino acids , aa , from the N-terminus , N ) of predicted functional domains ( boxed ) is shown underneath the diagram ( not to scale ) . ( B ) The cell density of wild type ( WT ) cells and recq2 +/- and -/- mutants cultures was counted every 24 hr up to a maximum of 96 hr , starting from a cell density of 1 x 104 cells . mL1 . The mean cell density from three independent experiments is shown on a Log10 Y-axis graph; error bars depict standard error of the mean . ( C ) Cell cycle analysis of WT cells and recq2 +/- and -/- mutants . DNA content was evaluated after DAPI staining of fixed cells and the number of cells with one nucleus and one kinetoplast ( 1N1K , white box ) , one nucleus and two kinetoplasts ( 1N2K , hatched box ) , two nuclei and two kinetoplasts ( 2N2K , black box ) and cells that did not fit into any of these categories ( other , grey box ) were counted . The proportion of each cell type is represented as a percentage of the total cells counted ( N ) . ( D ) Western blotting of whole cell extracts from WT and recq2-/- mutants grown in the absence ( - ) of methyl methane sulphonate ( MMS ) , or for 18 hr in media containing 0 . 0003% MMS ( + ) . Blots were probed with peptide antiserum recognizing Thr130 phosphorylated T . brucei histone H2A ( γ-H2A ) and , as loading control , polyclonal antiserum recognizing T . brucei EF1α . ( E ) Clonal survival ofwild type ( wt ) cells and recq2 heterozygous ( +/- ) or homozygous ( recq2-/- ) mutants is shown in the presence of varying concentrations of MMS , phleomycin or hydroxyurea ( HU ) . Mean survival ( % ) is plotted of the treated cells relative to untreated from three independent experiments , with vertical lines representing standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 00310 . 7554/eLife . 12765 . 004Figure 1—figure supplement 1 . Generation of recq2 null mutants in bloodstream form T . brucei . ( A ) PCR was used to amplify a 5’ and a 3’ region ( blue arrows ) of the RECQ2 ORF ( orange arrow ) . These PCR products were the ‘targeting regions’ cloned into the constructs shown below , which weremade to generate heterozygous ( +/- ) and knockout ( -/- ) mutants by replacing the ORF with resistance cassettes . Generalised representation of the linearised blasticidin and neomycin ( G418 ) resistance constructs relative to the RECQ2 ORF ( orange ) and the targeting flanking regions ( UTR; grey ) , which allow HR-mediated exchange ( crosses ) . β/α tub , β/ α tubulin intergenic region; actin IR , actin intergenic region; BSD , blasticidin resistance; NEO , neomycin ( G418 ) resistance . Not to scale . ( B ) The upper diagram shows diagnostic PCRs to confirm replacement of RECQ2 with the knockout constructs ( positions of primers are indicted by arrows ) : a region of the ORF was amplified ( “ORF PCR” ) and , in addition , a region was amplified using a forward primer lying upstream of the 5’ UTR region in the knockout construct and a reverse primer specific to the BSD or NEO genes ( “BSD PCR” and “NEO PCR” , respectively ) . The lower diagram shows agarose gels of PCR products generated from genomic DNA from wild type ( WT ) cells and heterozygote ( +/- ) and knockout ( -/- ) clones using primers described above . Distilled water was used as a negative control ( - ) . Broken lines indicate different images aligned in this figure; size markers are shown ( bp ) . ( C ) . RT-PCR confirmation of recq2 mutants . A region of the PIF6 ORF and a region of the RECQ2 ORF was PCR-amplified using cDNA ( +RT ) synthesised from RNA extracted from wild type ( wt ) cells and from recq2+/- and recq2-/- mutants; control samples in which no reverse transcriptase had been added ( -RT ) are shown , as are controls using distilled water rather than substrate ( - ) ; size markers on the agarose gels are shown ( bp ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 004 To evaluate the role of TbRECQ2 in T . brucei bloodstream form ( BSF ) cells , heterozygous ( +/- ) and homozygous ( -/- ) knockout mutants were generated by sequential transformation with constructs that replace nearly all of the TbRECQ2 ORF with cassettes that express resistance to blasticidin or G418 ( Figure 1—figure supplement 1A ) . Integration of the constructs and deletion of both wild type ( WT ) TbRECQ2 alleles in the -/- mutants was confirmed by PCR ( Figure 1—figure supplement 1B ) , while reverse transcription PCR ( RT-PCR ) ( Figure 1—figure supplement 1C ) showed no intact TbRECQ2 mRNA could be detected in the -/- mutants . The successful generation of null mutants shows TbRECQ2 is not essential in BSF T . brucei , though growth analysis revealed that the recq2-/- mutants had an increased population doubling time compared with WT cells ( Figure 1B ) . Indeed , slowed growth was also apparent in the req2+/- mutants , suggesting a growth impediment after loss of one allele that becomes more severe in the null mutant . To ask if the growth change results from an impediment in completing a cell cycle stage or traversing between sequential stages , cells were stained with DAPI to visualise nuclear ( N ) and kinetoplast ( K ) DNA . Counting the relative numbers of the two T . brucei genomes allows a cytological assessment of the cell cycle stage of individual cells in the population ( McKean , 2003 ) , and we found no change in the proportion of 1N1K ( G1-S phase ) , 1N2K ( G2-M phase ) or 2N2K ( post M phase ) cells in mutants relative to WT ( Figure 1C ) . Thus , the growth impairment of the RECQ2 mutants is not due to detectable stalling at a discernible cell cycle stage or transition . Western blotting to detect the levels of Thr130 phosphorylated histone H2A ( γ-H2A ) revealed an increased signal in the recq2-/- mutants relative to WT ( Figure 1D ) , indicating this modification accumulates in the absence of the helicase to an extent comparable to that seen in WT or -/- mutants cells grown for 18 hr in 0 . 0003% MMS ( see below ) . This form of phosphorylation of histone H2A is a modification seen after various genotoxic treatments in T . brucei ( Glover and Horn , 2012 ) , suggesting it is the kinetoplastid variant of a conserved eukaryotic chromatin alteration that acts a prelude to repair . Thus , accumulation of the histone variant in the recq2-/- mutants indicates an increased level of nuclear DNA damage , which appears not to impair cell cycle progression but may impede cell growth or survival . To ask if TbRECQ2 contributes to genome repair in T . brucei survival of the mutants was compared with WT cells following exposure to three DNA damaging compounds: hydroxyurea ( HU ) , methyl methanesulfonate ( MMS ) and phleomycin . HU depletes the cellular dNTP pool ( Bianchi et al . , 1986 ) , resulting in stalled replication forks that can subsequently collapse and generate DNA breaks . MMS methylates purines ( Brookes and Lawley , 1961 ) , which causes DNA breaks , at least in part through DNA repair activities targeting the alkylation ( Lundin et al . , 2005; Wyatt and Pittman , 2006 ) . MMS damage also perturbs replication , due to alkylated nucleotides blocking replication fork progression ( Groth et al . , 2010 ) . Phleomycin blocks the activity of DNA polymerase , inhibiting DNA synthesis and resulting in the formation of , primarily , DNA DSBs ( Falaschi and Kornberg , 1964; Reiter et al . , 1972 ) . Figure 1E shows clonal survival assays , which revealed that the recq2-/- cells displayed a much increased sensitivity to MMS compared with WT . Indeed , at MMS concentrations at and above 0 . 0002% , recq2+/- survival was lower than WT . These data are consistent with the increased MMSsensitivity previously reported in S . cerevisiae SGS1 mutants ( Mullen and Brill , 2000 ) , as well as human and chicken DT40 blm-/- mutants ( Imamura et al . , 2001 ) , and indicate TbRECQ2 is involved in the T . brucei response to MMS-induced damage . Clonal survival showed that TbRECQ2 also contributes to the T . brucei response to phleomycin and HU damage , since in both cases the recq2-/- mutants were more sensitive than WT cells ( Figure 1E ) . However , in neither case was there clear evidence that the recq2+/- cells were more sensitive than WT , perhaps indicating a more pronounced role for the putative helicase in tackling MMS damage . To examine the subcellular localisation of TbRECQ2 , the protein was N-terminally tagged with 12 copies of the myc epitope ( 12myc ) using a modified version ( gift , A . Trenaman ) of the pEnT6B construct ( Kelly et al . , 2007 ) , which allowed the variant protein to be expressed from the endogenous TbRECQ2 locus . A western blot of a cell lysate from a transformant clone showed expression of a myc-tagged protein of the expected size ( 182 kDa; Figure 2A ) . To test the functionality of the 12myc-TbRECQ2 variant , the untagged TbRECQ2 allele was deleted by replacement with a G418-resistance cassette ( Figure 1—figure supplement 1A ) . MMS sensitivity of the resulting recq2 12myc/- cells was then assessed by clonal survival ( Figure 2B ) . As survival of the cells expressing only the 12myc tagged variant of TbRECQ2 was comparable with WT cells and recq2+/- mutants in the presence of MMS , and notably better than recq2-/- mutants , addition of the epitope does not impair TbRECQ2 function in repair . 10 . 7554/eLife . 12765 . 005Figure 2 . TbRECQ2 is a nuclear factor that relocalises to foci in the presence of DNA damage , when it colocalises with RAD51 . ( A ) Western blot analysis of WT cells relative to cells expressing 12myc-tagged RECQ2 from the endogenous locus; size markers are shown ( kDa ) . ( B ) Clonal survival of cells expressing myc-tagged RECQ2 ( 12myc and 12myc/- ) is shown relative to WT , recq2+/- and recq2-/- cells in the presence of varying concentrations of MMS . Mean survival ( % ) is plotted of the treated cells relative to untreated from three independent experiments , with vertical lines representing standard error of the mean . ( C ) Representative examples of 12myc-RECQ2 and RAD51 cellular localisation in fixed cells , including after 18 hr growth in the presence or absence of phleomycin ( 1 μg . mL-1 ) ; bar: 13 μm . The tagged protein was detected by direct immunofluorescence using an anti-myc antiserum coupled with the Alexa Fluor 488 flurophore ( myc , green ) , while RAD51 was localised by indirect immunofluorescence using a rabbit anti-RAD51 antisera and an Alexa Fluor 594 goat anti-rabbit IgG antiserum; DNA was visualized with DAPI , and differential interference contrast ( DIC ) was used to visualise whole cells . ( D ) Percentage of cells containing 12myc-TbRECQ2 and RAD51 foci , as well as the number of detectable foci either in the absence ( - phleomycin ) or presence ( + phleomycin ) of phleomycin is shown . ( E ) Cells containing 12myc-RECQ2 and RAD51 foci following phleomycin treatment were categorised according to the degree of foci co-localisation , represented as percentage of cells that contained foci . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 005 Localisation of 12myc-TbRECQ2 was examined by direct immunofluorescence with anti-myc antiserum conjugated with Alexa-Fluor 488 ( Figure 2C ) . In most cells no signal could be detected , though in the very small proportion ( 0 . 2% ) that did show a signal ( Figure 2D ) , this was seen as a discrete puncta in the nucleus ( data not shown ) . Such a pattern is reminiscent of localisation described for T . brucei RAD51 ( Dobson et al . , 2011; Glover et al . , 2008; Hartley and McCulloch , 2008; Proudfoot and McCulloch , 2005; Trenaman et al . , 2013 ) , which is normally not seen in the cell , but localises in what have been described as foci in a small number of cells in the absence of induced damage . RAD51 foci are thought to be repair-related structures , as their numbers increase after damage , both in T . brucei and in many other cells ( Bergink et al . , 2013; Haaf et al . , 1995; Tarsounas et al . , 2004 ) . As a result , we examined co-localisation of 12myc-TbRECQ2 and TbRAD51 , detecting the latter by indirect immunofluorescence with polyclonal anti-RAD51 antiserum . To ask if the 12myc-TbRECQ2 signals might represent repair–related foci and , indeed , might be structurally associated with TbRAD51 foci , the cells were treated for 18 hr with phleomycin at 1 μg . mL-1 , which has been shown to generate a majority of cells with TbRAD51 foci ( Dobson et al . , 2011; Hartley and McCulloch , 2008; Trenaman et al . , 2013 ) . TbRAD51 foci were observed in ~2% of untreated cells ( Figure 2D ) , which is similar to previous studies and comparable with 12myc-TbRECQ2 . After growth in phleomycin , 47% of cells contained one or more 12myc-TbRECQ2 foci and 44% of cells contained one or more TbRAD51 foci ( Figure 2D ) . Indeed , not only was the relative proportion of cells with 12myc-TbRECQ2 and TbRAD51 foci comparable in these conditions , but the pattern of foci accumulation was highly related: most cells contained a single focus of either protein , though some contained 2 or 3 discrete foci , while others had larger numbers ( difficult to count accurately ) . Moreover , there was substantial overlap in the two signals , examples of which are shown in Figure 2C . In the majority of cells in which 12myc-TbRECQ2 and TbRAD51 foci were seen ( irrespective of the number of foci ) , the signals co-localised fully ( ~60% of cells with both foci , Figure 2E ) . In ~20% of cells with foci , the signal overlap was partial because the numbers of 12myc-TbRECQ2 and TbRAD51 foci were not equivalent in a single cell . Finally , in ~20% of cells there was no overlap , because 12myc-TbRECQ2 displayed foci but TbRAD51 did not , or vice versa . In summary , there appears to be pronounced similarity in behaviour and overlap in signal between 12myc-TbRECQ2 and TbRAD51 before and after phleomycin- induced damage . Whether the non-overlapping signals merely reflect incomplete resolution of one or other signal , or tell us that the proteins can act in subtly different manners ( perhaps temporally or spatially ) , is unclear . In order to test directly if TbRECQ2 acts in DNA DSB repair , we utilised two cell lines ( Figure 3A ) in which a single DNA DSB can be controllably induced in the genome ( gift , David Horn ) . Both the HR1 ( or INT ) ( Glover and Horn , 2014; Glover et al . , 2008 ) and HRES ( or TEL , VSGup ) ( Glover et al . , 2013; Glover and Horn , 2014 ) cells have been modified such that expression of the I-SceI meganuclease is dependent upon addition of tetracycline ( Tet ) to alleviate transcriptional repression by the Tet repressor . HR1 and HRES cells differ in the location of the I-SceI recognition site ( Figure 3A ) . In HR1 , the I-SceI site is located on chromosome 11 , between genes Tb927 . 11 . 4530 and Tb927 . 11 . 4540 ( tritrydb . org ) , and >1 Mbp from the nearest telomere . Here , the I-SceI site is embedded within an RFP ( red fluorescent protein ) :PUR ( puromycin N-acetyl transferase ) fusion gene . The RFP:PUR gene is flanked by tubulin sequences for mRNA transplicing and polyadenylation . Thus , HR-directed repair after I-SceI-induced DSB formation could occur by recombination between chromosome 11a ( containing the I-SceI site ) and its homologue ( 11b ) , but could also occur ectopically with chromosome 1 ( where the tubulin locus is found ) using the short tubulin sequences on the RFP:PUR cassette ( Glover et al . , 2008 ) . In HRES , the I-SceI recognition site is located upstream of VSG221 and downstream of the 70 bp repeats in the active BES ( BES1 ) , fused to a PUR gene ( Glover et al . , 2013 ) , an organisation similar to that described by Boothroyd et al . ( 2009 ) . Here , DSB induction has been proposed to mimic VSG switching , by initiating HR through available homology ( e . g . other VSGs , 70 bp repeats , telomere repeats , ESAGs ) ( Boothroyd et al . , 2009; Glover et al . , 2013 ) . In both HR1 and HRES , the presence or absence of the PUR gene at the I-SceI recognition site provides a means to assay for repair after Tet induction: due to the proximity of the PUR gene to the I-SceI target , the PUR sequence must be degraded after I-SceI cutting by DSB end resection to access the flanking homology that drives HR-directed DSB repair DNA , resulting in puromycin sensitivity . 10 . 7554/eLife . 12765 . 006Figure 3 . Mutation of TbRECQ2 impairs survival of T . brucei after induction of a DNA double strand break , either in the active telomeric VSG expression site or in the core of a chromosome . ( A ) I-SceI target sequences in HR1 and HRES cells . HR1 cells contain an I-SceI recognition site embedded within an RFP:PUR fusion gene ( black ) , flanked by tubulin sequences ( white ) , located between genes Tb . 11 . 02 . 2110 and Tb . 11 . 02 . 2020 on one copy of chromosome 11; HRES cells contain an I-SceI recognition site upstream of a PUR gene , flanked by tubulin sequences , located downstream of the 70 bp repeats of the active VSG221 expression site on chromosome 6 . B and C show clonal survival following I-SceI induction in HR1 and HRES cells , respectively . In both cases , wild type and two recq2-/- clones were distributed in three 96 well plates at a concentration of 0 . 26 cells per well either in the absence ( I-SceI uninduced ) or presence ( I-SceI induced ) of 2 μg . mL-1 tetracycline . The number of wells with surviving cells after 7–10 days growth is depicted as percentage of survivors following I-SceI induction relative to survivors without I-SceI induction; error bars represent standard error of the mean between three experimental repeats . Puromycin sensitivity of surviving I-SceI induced and uninduced clones was then tested , and is represented as the percentage of tested clones that grew in the presence ( + ) or absence ( - ) of 1 μg . mL-1 puromycin ( N: number of clones analysed ) . ( D ) Clones from ( C ) , excluding those that were puromycin resistant , were assayed for ESAG1 and VSG221 presence by PCR; data are shown as the percentage that were PCR positive ( N: number of clones analysed ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 00610 . 7554/eLife . 12765 . 007Figure 3—figure supplement 1 . Generation of TbRECQ2 mutants in T . brucei HR1 and HRES cells . ( A ) Agarose gels of PCR products generated from genomic DNA from HRES or HR1 wild type ( WT ) cells and RECQ2 heterozygote ( +/- ) and knockout ( -/- ) clones using primers described in the Figure S1B . Gaps indicate that lanes have been aligned in this figure after excision from multiple gels or from disparate parts of the same gel; size markers are shown ( bp ) . ( B ) Confirmation by PCR of RECQ2 knockout in HR1 and HRES cell lines . Agarose gels showing products generated when PCR was performed on a region of the PIF6 ORF or the RECQ2 ORF using cDNA ( +RT ) synthesised from RNA extracted from recq2-/- mutants; control reactions in which no substrate was added ( - ) , or where cDNA from HR1 WT cells was used , are shown . Gaps indicate that lanes have been aligned in this figure after excision from multiple gels or from disparate parts of the same gel; size markers are shown ( bp ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 007 TbRECQ2 mutants were generated in HR1 and HRES BSF cells as before , with integration of the constructs and loss of intact RECQ2 in two -/- mutant clones confirmed by PCR ( Figure 3—figure supplement 1A ) and RT-PCR ( Figure 3—figure supplement 1B ) . In order to understand if loss of TbRECQ2 affected DSB repair , cell survival following Tet induction was assayed by determining clonal survival efficiency relative to uninduced cells . The survival rate of HR1 WT cells following I-SceI induction was ~60% ( Figure 3B ) , equivalent to that reported previously ( Glover and Horn , 2014; Glover et al . , 2008 ) . In the absence of TbRECQ2 , the survival rate decreased 2-fold , with only ~30% of wells displaying growth in both of the two HR1 recq2-/- clones examined . These data indicate that recq2-/-mutants are less able to survive a chromosome-internal DNA DSB than WT T . brucei cells . Analysis of the puromycin sensitivity of surviving clones showed , for both WT and recq2-/- HR1 cells , that all Tet-induced clones ( n = 25 ) were puromycin sensitive and all uninduced clones ( n = 6 ) were puromycin resistant ( Figure 3B ) . Thus , a functional PUR gene was lost in all cases after I-SceI expression was induced , showing that DSB formation was successful and indicating repair is possible , though less efficient , in the absence of TbRECQ2 . Broadly the same outcome was seen for the HRES cells ( Figure 3C ) . Consistent with previous observations ( Glover et al . , 2013 ) , the survival rate of HRES WT cells after I-SceI induction ( ~24% ) was >2-fold lower than HR1 cells ( Figure 3C ) , indicating greater lethality when a DSB is made in the active BES . Nonetheless , the survival rate of the two HRES recq2-/- clones ( 13% and 14 . 8% ) was again ~50% of HRES WT cells , suggesting loss of TbRECQ2 impairs survival in both loci . In fact , the level of impairment after loss of TbRECQ2 may be greater than the clonal survival assay predicts since , unlike in HR1 , evaluating the puromycin sensitivity of recovered clones showed that 14% of survivors ( n = 7 ) in one HRES recq2-/- clone and 70% ( n = 10 ) in the other were puromycin resistant ( Figure 3C ) . These data are most simply explained by a greater number of HRES recq2-/- mutants being recovered ( relative to HR1 mutants ) in which a DSB has not been induced , reflecting the very limited survival capacity of recq2-/- mutants after a DSB is made in the active BES . Despite this , PCR of the puromycin sensitive recq2-/- survivor clones showed that all had lost the VSG221 gene ( also called VSG 427–2 ) ( Hertz-Fowler et al . , 2008 ) and most had retained the ESAG1 gene variant specific to the targeted BES ( Figure 3D ) , a pattern consistent with previous analysis in WT HRES ( Glover et al . , 2013 ) cells and relatives ( Boothroyd et al . , 2009 ) . Thus , loss of TbRECQ2 does not result in a major shift in repair pathway after induction of a DNA DSB in the BES , meaning reduced survival in the mutants is best explained by less efficient execution of a predominant repair reaction . In order to analyse the effect of TbRECQ2 loss onVSG switching we adapted an in vitro strategy of Povelones et al ( Povelones et al . , 2012 ) , which is conceptually related to an assay established by Kim and Cross ( Kim and Cross , 2010; 2011 ) . In this strategy , a herpes simplex virus thymidine kinase ( TK ) gene fused to a hygromycin resistance gene ( HYG-TK ) was inserted between the 70 bp repeats and VSG221 in the active BES ( Figure 4A ) . Additionally , enhanced GFP and PUR genes were integrated downstream of the active BES promoter . Integration of the marker genes and the expected expression of GFP and VSG221 protein was confirmed by PCR and western blotting ( Figure 4—figure supplement 1A ) . The parental cell line generated by these manipulations , GFP221hygTK , allows the nucleotide analogue ganciclovir ( GCV ) to be used to eliminate cells from the population that have not inactivated TK , which can occur through VSG switching events that can be distinguished by the presence and expression of the VSG221 and GFP genes in the BES ( Figure 4A ) . In a transcriptional ( in situ ) switch VSG221 and GFP proteins are no longer expressed from the BES but both genes are retained . In contrast , cells that have switched by a gene conversion downstream of PUR-GFP ( here termed VSG GC ) retain GFP expression from the BES but have deleted TK and VSG221 from the transcription unit . Longer range gene conversions are also possible that encompass the whole BES and lead to removal of both GFP and VSG221 , though in the approach used here this reaction cannot be distinguished from events in which the BES is deleted without gene conversion and cells survive through a transcriptional switch ( ES GC or in situ+ES del , respectively ) ( Cross et al . , 1998; Rudenko et al . , 1998 ) . Finally , in VSG switching by telomere exchange ( telomere XO ) GFP continues to be expressed from the BES , whereas VSG221 protein expression is silenced by moving the gene to another telomere . Cells that have inactivated TK through mutation , rather than VSG switching , can also be selected for in this assay ( Povelones et al . , 2012 ) . However , such cells , which can be identified by continued expression of VSG221 , were rare in this study ( Figure 4C ) . 10 . 7554/eLife . 12765 . 008Figure 4 . Mutation of TbRECQ2 leads to elevated VSG switching and increased recombination . ( A ) Strategy for determining VSG switching mechanisms; adapted from Povelones et al . ( 2012 ) . The active VSG BES of GFP221hygTK cells is shown , within which the PUR , eGFP , HYG-TK and VSG221 genes are represented as coloured boxes . In addition , one of the ~14 silent BES containing a distinct VSG ( turquoise box ) is shown , as are multiple silent VSGs elsewhere in the genome ( various colours; for convenience these are shown as a single array , but could also be at the telomere of silent mini-chromosomes ) . 70 bp repeats upstream of the VSGs are denoted by hatched boxes . Different switching strategies allow survival after ganciclovir treatment and can be distinguished by analysis of VSG221 and GFP presence by PCR , and expression of the proteins by western blot ( profiles detailed under each mechanism ) . Switchers that arise by in situ switching , telomere crossover ( XO ) or VSG gene conversion ( VSG GC ) can be detected unambiguously , while events that occur by BES gene conversion or in situ switching coupled with BES deletion ( ES GC/ in situ+ES del ) are indistinguishable . Note , only in situ+ES del reaction is shown , and not ES GC ( where all sequence of a silent BES is duplicated and replaces the VSG221 BES ) ; in addition , for VSG GC the silent grey array donor VSG gene is shown as being copied , but the reaction could also use a BES VSG gene . Non-switcher TK mutants can also allow ganciclovir survival . ( B ) The mean switching rate of GFP221hygTK WT and recq2 mutants ( +/- and -/- ) was inferred from the mean number of survivors from two experiments , each with three replicates , following treatment with ganciclovir and after culture with ( + ) or without ( - ) puromycin; error bars represent standard error of the mean . ( C ) Profiles of WT and recq2 mutants ( +/- and -/- ) survivors in the non-puromycin experiments , represented as a percentage of total surviving clones analysed from the two datasets; number of clones ( N ) analysed is indicated . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 00810 . 7554/eLife . 12765 . 009Figure 4—figure supplement 1 . Generation of TbRECQ2 mutants in T . brucei GFP221hygTK cells . ( A ) Confirmation of the generation of GFP221hygTK cells by PCR and western blot . The upper diagram shows PCR performed on genomic DNA extracted from GFP221hygTK transformants or wild type ( WT ) cells , testing for integration of the 221GP1 and HYG-TK constructs . Size markers ( bp ) are shown . The lower diagram shows westerns blots of total protein extract of GFP221hygTK cells probed with rabbit anti-VSG221 antisera or with rabbit anti-GFP antiserum; size markers ( kDa ) are shown . ( B ) PCR and western blot analysis of GFP221hygTK recq2 mutants . The upper diagrams show PCR performed on genomic DNA extracted from GFP221hygTK recq2+/- and GFP221hygTK recq2-/- cells , testing for the RECQ2 ORF , integration of the BSD or NEO resistance cassettes , and retention of the 221GP1 and HYG-TK constructs; size markers are shown ( bp ) . The lower diagram shows westerns blots of total protein extract of GFP221hygTK recq2+/- and GFP221hygTK recq2-/- cells cells probed with rabbit anti-VSG221 antisera or with rabbit anti-GFP antiserum; size markers ( kDa ) are shown . In all gels , gaps indicate that lanes have been aligned in this figure after excision from multiple gels/membranes , or from disparate parts of the same gel/membrane . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 00910 . 7554/eLife . 12765 . 010Figure 4—figure supplement 2 . Summary table of ganciclovir survival mechanisms . Summary of PCR and western analysis of clones from the first ( left ) and second ( right ) ganciclovir survival experiments . GFP221hygTK wildtype ( WT ) , GFP221hygTK recq2+/- and GFP221hygTK recq2-/- cells are shown , as are whether or not clonal survival in ganciclovir was conducted in the presence ( + ) or absence ( - ) of puromycin . ‘Survival in’ denotes whether or not the clones that were recovered could grow in the presence of puromycin ( PUR; 1 μg . mL-1 ) or hygromycin ( HYG; 10 μg . mL-1 ) . The following columns summarise the results of PCR to assay for the presence or absence of RNA POLI , GFP and VSG221 genes , and westerns to detect expression of GFP or VSG221; blue indicates a negative result and yellow indicates a positive result , while completely white rows indicate clones that could not be analysed , either due to absence of detectable genomic DNA or of protein in the cell lysate . The final column summarises the inferred strategy for survival , as detailed in Figure 5 . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 010 To assess the contribution of TbRECQ2 to VSG switching , req2+/- and req2-/- mutants were generated in the GFP221hygTK cells , with integration of the constructs and loss of intact TbRECQ2 in the -/- mutants confirmed by PCR , and continued expression of GFP and VSG221 shown by western blotting ( Figure 4—figure supplement 1B ) . The switching rate of the WT cells was then compared with the TbRECQ2 mutants: cultures were grown for 48 hr in media lacking hygromycin or puromycin ( allowing switch variants to arise ) and then cloned by limiting dilution in antibiotic-free media containing GCV , allowing measurement of the number of cells in the diluted population that had inactivated TK . In the WT cells GCV resistant clones arose at a rate of ~1 . 5 x 10–5cells/generation ( Figure 4B ) , consistent with rates determined in comparable studies Kim and Cross , 2010a , 2011a; Povelones et al . , 2012b ) . No change in rate was seen in the recq2+/- mutants ( Figure 4B ) , but GCV resistant cells arose around 2-fold more frequently in the recq2-/- mutants ( ~3 . 2 x 10–5 resistant cells/generation ) . To ask if this change could be explained by an alteration in VSG switching strategy , we used PCR and western blotting to determine the profile of GFP and VSG gene presence and protein expression in a selection of GCV resistant clones for each cell line ( Figure 4C; Figure 4—figure supplement 2 ) . In broad agreement with observations made by Povelones et al . ( 2012 ) , in WT cells most GCV resistant clones had arisen either by in ES GC or in situ+ES del ( ~60% ) , with in situ switching the next most common process ( ~25% ) ; VSG GC was rare ( ~10% ) , and we found no examples of telomere XO events . A virtually identical pattern of events was seen in the recq2+/- cells , consistent with the unaltered rate at which GCV resistant cells arose . In contrast , ~90% of GCV resistant clones in the recq2-/- mutants arose either by VSG GC ( ~50% of total ) or telomere XO ( ~45% ) , indicating that the elevated rate in the null mutants is due to increased use of these recombination strategies . To test this interpretation further , the switching experiment was conducted in the presence of puromycin , which should prevent any events that inactivate expression of PUR ( Figure 4B ) . In these conditions , GCV resistant cells arose ~2–3 fold less frequently in the WT and recq2+/- cells ( Figure 4C ) , consistent with their predominant use of in situ and ES GC or in situ+ES del events , whilst there was less impact on the recq2-/- cells , where VSG switching is largely downstream of PUR . Taken together , these data indicate that loss of TbRECQ2 results in increased VSG switching by a change in repair strategy . Given the dichotomy between the effects of TbRECQ2 loss on DNA DSB repair and VSG switching , we next characterised in more detail the response of T . brucei BSF cells to induction of an I-SceI-mediated DSB . We first compared the cell cycle response to DSB induction in HR1 and HRES cells ( Figure 5A , B ) . Expression of I-SceI was induced with Tet and , 12 and 24 hrlater , cells were stained with DAPI to visualize N and K DNA . Virtually all cells were categorised as either 1N1K , 1N2K or 2N2K , irrespective of whether I-SceI expression was induced or not . However , consistent with previous reports ( Glover et al . , 2013; Glover and Horn , 2014; Glover et al . , 2008 ) , addition of Tet resulted in increased numbers of 1N2K cells , indicating impaired G2-M cell cycle progression ( Figure 5A , B ) . In HR1 , 1N2K cell numbers increased ( from ~10% of the uninduced population ) to ~20% 12 hr after Tet induction and then returned to ~10% after 24 hr ( Figure 5A ) , indicating the cell cycle impairment was transient . In contrast , increased 1N2K cell numbers ( ~20% of the population ) persisted until 24 hr after Tet addition in HRES ( Figure 5B ) , suggesting the response to a DSB differs if the lesion is in the active BES or a chromosome-internal site ( Glover and Horn , 2014 ) . To examine this further , we used quantitative real-time PCR ( qPCR ) to assess the dynamics of I-SceI site cleavage . 10 . 7554/eLife . 12765 . 011Figure 5 . Analysis of cell cycle progression , DNA repair kinetics and VSG expression after I-SceI-mediated DNA double strand break formation . Cell cycle analysis of HRES ( A ) and HR1 ( B ) cells following I-SceI induction . DNA content is shown 12 and 24 hr post I-SceI induction ( +T ) after visualisation by DAPI staining of fixed cells; uninduced cells ( -T ) were analysed as a control . The number of cells with one nucleus and one kinetoplast ( 1N1K ) , one nucleus and two kinetoplasts ( 1N2K ) , two nuclei and two kinetoplasts ( 2N2K ) and cells that did not fit into any of these categories ( other ) were counted . The proportion of each cell type is represented as a percentage of the total cells counted ( N ) . ( C ) Relative efficiency of PCR amplification of the I-SceI target sequence is shown at various time points after induction of I-SceI in HR1 and HRES cells; values are shown at each post-induction time point as a percentage of the amount of PCR product generated at 0 hr; values are the mean of three experimental repeats and vertical lines denote standard deviation . ( D ) Relative PCR amplification of the VSG221 gene downstream of the I-SceI target is shown after I-SceI induction in HRES cells; values were determined and are represented as in C . ( E ) VSG221 expression in HRES cells 24 hr post I-SceI induction was visualised by indirect immunofluorescence of fixed cells with anti-VSG221 antiserum ( + ) , and are compared with control cells in which only secondary antiserum was used ( - ) . The graphs below show the proportion of cells expressing VSG221 on their surface 24 hr post I-SceI induction ( +T ) , or without I-SceI induction ( -T ) ; data are represented as the percentage of total cells counted ( N ) . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 01110 . 7554/eLife . 12765 . 012Figure 5—figure supplement 1 . Break formation over 72 hr after ISceI induction in T . brucei HR1 cells . Relative efficiency of PCR amplification of the ISceI target sequence is shown at various time points up to 72 hr after induction of ISceI in HR1 cells: values are shown at each post-induction time point as a percentage of the amount of PCR product generated at 0 hr; values are the mean of three experimental repeats and vertical lines denote standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 012 Genomic DNA was prepared at multiple timepoints after Tet induction in HR1 or HRES cells and qPCR was performed with primers spanning the I-SceI target sequence , determining the amount of PCR product after Tet induction relative to uninduced cells and normalized by a control locus ( ΔΔCt method ) ( Livak and Schmittgen , 2001 ) . In HR1 ( Figure 5C ) , a reduction in product was seen from 2 hr after Tet addition ( 20% ) and this effect increased until around 8 hr ( 80% ) , with little change in PCR efficiency thereafter ( 8–72 hr; Figure 5C and Figure 5—figure supplement 1 ) . These data suggest that I-SceI cleavage in HR1 cells is rapid and reaches a maximum within a cell cycle ( ~8 hr ) , an interpretation consistent with the timing of single-stranded DNA formation at I-SceI breaks detected by Southern blotting ( Glover and Horn , 2014; Glover et al . , 2008 ) . Southern blotting suggests that allelic repair products in HR1 accumulate slowly over ~16–72 hr ( Glover et al . , 2008 ) , consistent with qPCR to detect duplication of the intact allele ( data not shown ) . I-SceI target qPCR revealed a different response to I-SceI induction in HRES cells ( Figure 5C ) . Early in the time course , loss of PCR product was more rapid and more complete , with <10% of uninduced product by 2–3 hr after Tet addition , suggesting I-SceI cleavage was at least as efficient in HRES cells as in HR1 . However , in contrast with HR1 , the levels of PCR product later in the reaction increased , in two distinct peaks ( at ~20 and 30 hr post induction ) , to levels approaching that of the uninduced cells . Given that I-SceI cleavage is rapid , these data either suggest repair occurs efficiently in a manner that regenerates the I-SceI target , or repair is inefficient and cells that are subjected to I-SceI cleavage are killed , allowing those that have not suffered a DNA DSB to outgrow . Unlike for HR1 , where most repair is mediated by allelic HR , I-SceI cleavage in HRES ( or related cells ) predominantly results in cells expressing a new VSG , whose identity cannot be easily predicted ( Figure 3 ) ( Boothroyd et al . , 2009; Glover et al . , 2013 ) . Thus , we used qPCR to ask about the fate of the BES VSG ( VSG221 ) after I-SceI cleavage ( Figure 5D ) . The abundance of the VSG221 PCR product very closely matched that of the I-SceI target over the timecourse , and the recurrence of product indicates the initial , rapid loss of VSG221 is not due to replacement by another VSG . To test this further , we performed immunofluorescence with anti-VSG221 antiserum , revealing that virtually all cells retained VSG221 on their surface 24 hr after Tet induction ( Figure 5E ) . Taken together , these data indicate that induction of a DSB by I-SceI in the active BES does not elicit rapid repair ( within 32 hr ) that removes the downstream VSG . The analyses above question the association between an induced DNA DSB and VSG switch initiation . In order to ask if any other feature of genome maintenance might correlate with antigenic variation , we examined the dynamics of T . brucei nuclear DNA replication using marker frequency analysis coupled with next generation sequencing ( MFAseq ) ( Tiengwe et al . , 2012 ) . MFAseq compares the relative depth of sequence read mapping in replicating ( S phase ) and non-replicating ( here , G2 ) cells , allowing the sites and relative efficiencies of origins of replication to be determined , as well as inference on the timing and direction of replication genome-wide . In T . brucei , MFAseq has so far only been performed in procyclic cell forms ( PCF ) , the insect stage of the parasite , and in the strain TREU927 ( Tiengwe et al . , 2012 ) , in which the repertoire of telomeric BES has not been characterized . Here , we performed MFAseq in PCF and BSF cells of T . brucei strain Lister 427 , where all BES have been sequenced ( Hertz-Fowler et al . , 2008 ) , allowing us to ask if differences between the life cycle stages , including gene expression changes , result in alterations in replication dynamics . Figure 6 shows MFAseq mapping for the eleven megabase chromosomes of T . brucei , excluding the BES , and comparing the patterns seen when early S or late S cells are compared with G2 phase cells . The MFAseq pattern of peak location was invariant when comparing early S BSF and PCF cells , and when comparing late S phase BSF and PCF cells . Even more strikingly , the relative heights of the MFAseq peaks in each chromosome were invariant between the two life cycle stages ( Figure 6 ) , as well as being invariant between T . brucei strains Lister427 and TREU927 ( data not shown ) . Thus , neither differentiation between life cycle stages in T . brucei , nor extended growth of different T . brucei strains , leads to changes in the genomic sites used as origins , or changes in the timing programme of origin activation in the chromosome cores . Late S MFAseq has not previously been reported for T . brucei , though it was inferred that the number of origins ( 42 ) mapped using early-mid S phase cells might be an underestimate of ~2-fold , due to replication initiation at late acting origins ( Tiengwe et al . , 2012 ) . This prediction appears to be inaccurate , as most of the peaks detected in the late S samples were merely wider than the early S peaks; indeed , in several locations early S peaks had merged as replication forks converged ( Figure 6 ) . Only five origins ( dashed lines , Figure 6 ) were observed in the present data , both in the BSF and PCF cells , which were not predicted previously ( Tiengwe et al . , 2012 ) . All these were ‘weak’ origins , with low MFAseq peak heights , and it is possible that they were observed here due to the more compressed graphical representation , rather than being origins that are active in Lister 427 cells and not in TREU927 . Using the localisation of TbORC1/CDC6 ( Tiengwe et al . , 2012 ) and histone H4K10Ac ( Siegel et al . , 2009 ) binding sites in the TREU927 genome as a guide ( Figure 6 ) , it is clear these five origins localise to the boundaries of the polycistronic transcription units , as expected . Taken as a whole , the above MFAseq analysis suggests pronounced rigidity in the coordination of nuclear DNA replication in T . brucei . 10 . 7554/eLife . 12765 . 013Figure 6 . Replication timing throughout the core genome is stringently conserved between BSF and PCF T . brucei cells . Each set of four graphs shows the distribution of replication origins in the 11 megabase chromosomes ( depicted as Chr1 to Chr11 ) , assessed by MFAseq ( Tiengwe et al . , 2012 ) . At the top of each set of graphs is a track representing the genes in the chromosome: in blue the open reading frames ( ORFs ) are transcribed from the left to the right , and in red they are transcribed from right to left . Below each set of graphs is a track depicting histone H4K10ac-enriched sites ( Siegel et al . , 2009 ) . The four graphs in each case show the ratio between the coverage ( read-depth ) of DNA derived from Illumina sequencing of early S phase and G2 phase cells , or late S phase and G2 phase cells , where each point represents the median S/G2 ratio ( y-axis ) in 2 . 5 Kbp bins across the chromosome ( x-axis; bars indicate 500 Kbp intervals ) . All graphs are scaled according to chromosome size . The light red graph shows MFAseq for BSF early S cells , while dark red represents the data from late S phase . PCF MFAseq data is shown for early S cells in light green , and in dark green for late S . Vertical , solid grey lines represent the origins identified previously ( Tiengwe et al . , 2012 ) , while dashed lines highlight replication origins only observed in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 013 The BES repertoire of T . brucei Lister 427 is composed of 14 distinct BES ( Hertz-Fowler et al . , 2008 ) , most of which have not been allocated to specific megabase or intermediate chromosomes . The PCF and BSF S and G2 sequence reads were used to perform MFAseq mapping to the 16 available contigs representing the 14 different BES , as shown in Figure 7 . In these data , peaks cannot be discerned , as the sizes of the BESs are smaller than the distance covered by the replication forks at most origins ( Figure 6 ) . Thus , at the higher resolution used here , the MFAseq mapping is seen as multiple discrete points , corresponding to the median S/G2 read depth ratio in each of the 2 . 5 kbp ‘bins’ that span the BES ( Figure 7B ) . For all but one of the BES , there was no clear difference between the MFAseq mapping in the BSF and PCF cells , either for early or late S phase . Moreover , there was no evidence that 13 of the 14 BES had been replicated , even in the BSF and PCF late S samples , since there was no consistent increase in S phase reads relative to G2: comparing the S/G2 ratios for each bin in each BES showed that the overall median S/G2 ration for most BES was ~1 . 0 ( Figure 7C ) . These data suggest that 13 of 14 BES are replicated very late in S phase , similar to telomeres in other eukaryotes ( Rhind and Gilbert , 2013 ) . BES1 was the single exception to the above trend . In this contig , BSF S phase reads ( both early and late ) across the BES were markedly elevated relative to G2 ( Figure 7B ) , with overall median S/G2 ratios of 1 . 25 and 1 . 3 in early and late S , respectively ( Figure 7C ) . This effect was limited to the BSF cells , however , since the PCF MFAseq data for BES1 ( again in either early or late S ) was comparable with all other BES , with S/G2 ratios ~1 . 0 ( significantly different from BSF cells: p-value < 0 . 0001 ) . These data suggest that BES1 , alone amongst the 14 BES , is replicated early , and this deviation from the late replication of other BES is limited to mammal-infective cells . BES1 differs from the other BES in being the actively transcribed site , encoding VSG221 in the BSF cells used in this study , as shown by indirect immunofluorescence ( Figure 7A ) . In PCF cells full transcription of all BES , including BES1 , is silenced ( Rudenko et al . , 1994 ) and the VSG coat is replaced with procyclin ( Figure 7A ) . Thus , the unique early replication of BES1 only in BSF cells suggests that replication timing of T . brucei telomeres displays a precise association with transcription . 10 . 7554/eLife . 12765 . 014Figure 7 . The active VSG expression site in bloodstream form T . brucei cells is the only telomeric site that is early replicating . ( A ) Immunofluorescence of PCF or BSF Lister 427 cells and BSF strain Lister 427 with anti-VSG 221 antiserum or with anti-EP procyclin antiserum; top panels show the cells stained with DAPI , while the bottom panel shows the cells’ outline by DIC . Images were acquired with the Axioskop 2 imaging system and the scale bar represents 5 μm . ( B ) The Lister 427 bloodstream VSG expression site ( BES ) TAR clones sequenced by ( Hertz-Fowler et al . , 2008 ) were used to map the MFAseq data from BSF and PCF cells; note that two BES are represented by duplicate TAR clones: BES 7 ( ϕ – TAR 65; ϕϕ – TAR 153 ) , and BES 17 ( ϕ – TAR 51; ϕϕ – TAR 59 ) . The ratio between sequence coverage ( read-depth ) in early S phase and G2 phase cells , or late S phase and G2 phase samples , is plotted , where each point represents the median S/G2 ratio ( y-axis ) per 2 . 5 Kbp bin across the BES ( x-axis ) . The size of each BES is shown on each x-axis in 10 Kbp intervals , and all graphs are scaled according to BES size . The y-axis scale is the same for all graphs , but the legend is only shown on the ones at the far left . BSF early S data is represented as light red , BSF late S as dark red , PCF early S as light green , and PCF late S as dark green . The red dashed box highlights BES 1 . ( C ) The S/G2 values used to generate the graphs in ( B ) are shown plotted per sample ( BSF early S – light red , BSF late S – dark red , PCF early S – light green , and PCF late S – dark green ) , rather than by genomic location , for each BES ( numbered as before ) . Horizontal bars ( black ) represent the median of the S/G2 values , and error bars the interquartile range . In order to infer statistical significance , the values were analysed with the non-parametric , unmatched , Kruskal-Wallis test; statistical significance is only shown for differences between the BSF and PCF samples: ( ) p-value <0 . 01; ( * ) p-value <0 . 001; ( **** ) p-value <0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 014 To test the MFAseq mapping in the BES we used qPCR to examine the predicted early replication of the actively transcribed VSG gene . First , S/G2 ratios were derived using qPCR on DNA from cells expressing BES1 ( VSG221 ) , which had been used in the MFAseq ( Figure 6 and 7 ) . Using an origin-associated locus and a non-origin locus from chromosome 5 ( Figure 8A ) as controls , we observed early and late S/G2 qPCR ratios consistent with the former locus being early replicating and the latter late replicating ( Figure 8B ) . Furthermore , qPCR of VSG221 in the same cells revealed early and late S/G2 ratios higher than the chromosome 5 origin locus , consistent with the VSG replicating very early ( Figure 8B ) . To ask if the early replication of VSG221 is determined by transcription of BES1 , we next examined cells generated by Glover et al ( Glover et al . , 2007 ) in which a Tet operator is inserted downstream of the BES1 promoter ( Figure 8B ) . In the absence of Tet , binding of the Tet repressor ( TetR ) blocks transcription elongation in BES1 and cells are selected that have switched to transcribing another VSG BES ( Aresta-Branco et al . , 2016; Glover et al . , 2007 ) . A recently derived clone of the BES1 TetR blockade cells was sorted into early S , late S and G2 populations , and qPCR performed on recovered DNA . Early and late S/G2 qPCR ratios for the chromosome 5 controls were lower in these experiments compared with the qPCR from the BES1 ( VSG221 ) expressers , probably as a result of sorting differences in the selection of cells within the S phases or within G2 . Nonetheless , the S/G2 ratio of the origin locus increased from early to late S , indicating replication progression . For VSG221 , the early and late S/G2 ratios in the TetR blockade cells were much lower than in the BES1 ( VSG221 ) expressing cells , being indistinguishable from the origin locus , indicating the pronounced early replication of this VSG is not seen when it is no longer transcribed . Immunofluorescence indicated that VSG121 , whose gene is present in BES3 in this T . brucei strain ( Hertz-Fowler et al . , 2008 ) , could be detected on the cell surface of most of the BES1 TetR blockade cells at the outset of the experiment ( Figure 8B ) and so we used qPCR to test the replication timing of this VSG in both cell types . qPCR of VSG121 is complicated relative to VSG221 because the gene is not only located in a BES: at least four VSG121 copies are found within the subtelomeric VSG arrays ( Trenaman et al . , 2013 ) , whose replication timing is unclear ( Tiengwe et al . , 2012 ) . Despite this , early and late S/G2 qPCR ratios for VSG121 in BES1 ( VSG221 ) expressing cells were lower than both the VSG221 and chromosome 5 origin control locus values and were more comparable with the non-origin control ( Figure 8B ) , consistent with late replication . In contrast , in the BES1 TetR blockade cells the VSG121 S/G2 ratios were higher than both VSG221 and the non-origin locus and , instead , were comparable with the origin control ( Figure 8B ) . Thus , despite the potentially confounding effect of VSG121 array copies that may be late replicating , as well as uncertainty about the transcriptional status of all the VSG BES in these cells , these data suggest earlier replication of VSG121 when the VSG221-containing BES1 is silenced and BES3 is at least one of the BES expressed in the T . brucei population . Taken together , these VSG-focused qPCR experiments validate the MFAseq association between replication timing and transcription status of the telomeric BES . 10 . 7554/eLife . 12765 . 015Figure 8 . Determination of telomere replication timing in T . brucei cells expressing distinct bloodstream VSG expression sites . ( A ) MFAseq of chromosome 5 , as shown in Figure 6 , comparing S/G2 read depth ratios in 2 . 5 kbp bins in bloodstream form ( early S – light red; late S – dark red ) and procyclic form ( early S – light green; late S – dark green ) cells . Arrows highlight the locations of an early replicating ( origin ) locus and a late replicating ( non-origin ) locus , which were used in real-time quantitative ( q ) PCR validation . ( B ) qPCR to determine replication timing of VSG221 , VSG121 and chromosome 5 origin and non-origin loci in cells in which BES1 ( containing VSG221 , red box ) is actively transcribed ( left ) , or in which ( left ) elongation of BES1 transcription in blocked by Tet repressor ( TetR , black circle ) binding to a Tet operator ( black box ) adjacent to the BES promoter ( arrow ) , leading to transcription ( dotted arrow ) of BES3 ( containing VSG121 , green box ) . In each representation of the BES only the VSG genes are shown and black arrows denote the approximate location of primers used in qPCR; below each diagram immunoflouresence microscopy with anti-VSG221 ( left , red ) and anti-VSG121 antiserum ( right , green ) is shown ( cells are shown by differential interference contrast , DIC ) . Graphs depict the relative abundance of PCR product from VSG221 , VSG121 , origin and non-origin loci in the two cell types shown above; in each case qPCR was used to determine the amount of the PCR products in DNA from early S phase cells relative to G2 ( upper graph ) , or in late S phase cells relative to G2 ( lower graph ) . S/G2 ratios are the mean of three qPCR repeats . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 015 Understanding the initiation event ( s ) of VSG switching by recombination is important , since this element of the reaction may be lineage-specific , and might explain both the elevated rate of the reaction relative to general HR ( which catalyses the change in VSG ) and the potential focus on the active BES . VSG switch initiation has been modelled by the direct generation of a DSB in the active BES through the controlled expression and targeting of I-SceI ( Boothroyd et al . , 2009; Glover et al . , 2013; Glover and Horn , 2014 ) , which results in VSG switching . However , to date there is little evidence for the direct formation of a DSB being the initiating strategy in VSG switching , despite data showing that breaks can be detected around the telomeric VSG ( though with uncertainty as to whether these are limited to the active BES or to the 70 bp repeats ) ( Boothroyd et al . , 2009; Glover et al . , 2013; Jehi et al . , 2014 ) . As a result , the nature of the lesions that first form in the BES to drive VSG switching , and the route by which the lesions are generated , is unclear . Indeed , several initiation models having been proposed ( Barry and McCulloch , 2009 ) : an uncharacterised endonuclease ( Barry , 1997 ) , telomere instability ( Dreesen et al . , 2007; Hovel-Miner et al . , 2012; Jehi et al . , 2014 ) and transcription-derived instability ( Kim and Cross , 2010; 2011 ) . Here , we have tested the association between I-SceI-mediated DSB formation and VSG switching , revealing that endonuclease mediated DSB formation is unlikely to be the route for initiation of T . brucei antigenic variation . Instead , we reveal a remarkably precise association between DNA replication timing and transcription of the single active VSG BES in BSF cells , suggesting that replication may drive VSG switching . To test the role of endonuclease-generated , BES-focused DSBs in the initiation of VSG switching , we first analysed the genome repair functions of one of two T . brucei RecQ-like proteins , TbRECQ2 . Several lines of evidence indicate that TbRECQ2 localises to and repairs DNA breaks , including DSBs . First , T . brucei recq2 null ( -/- ) mutants display impaired survival in the presence of three compounds that can cause DNA damage , including phleomycin , which is known to generate DSBs ( Falaschi and Kornberg , 1964; Reiter et al . , 1972 ) . Second , TbRECQ2 protein relocalises to subnuclear foci after exposure to phleomycin and , critically , in these conditions the putative helicase displays pronounced colocalisation with TbRAD51 , an enzyme that binds single-stranded DNA formed at DSBs to catalyse HR repair . Finally , T . brucei recq2-/- mutants display reduced survival relative to WT cells after I-SceI induction of DSBs , both in the active BES and elsewhere in the genome , arguing for impaired repair of this lesion . All the above findings are consistent with the activities of related RecQ helicases in other cells , where the enzymes are known to contribute widely to HR . In this regard , the repair and VSG switching ( see below ) phenotypes observed for TbRECQ2 are highly reminiscent of those described for T . brucei Topo3α ( Kim and Cross , 2010 ) and RMI1 ( Kim and Cross , 2011 ) . Thus , it seems plausible that TbRECQ2 , and not the other putative T . brucei RecQ-like helicase ( TbRECQ1 ) , interacts with these factors to form the T . brucei homologue of the RTR complex ( Mankouri and Hickson , 2007 ) . Given the above evidence for TbRECQ2 repair functions , the enzyme provided a means to test the role of endonuclease-generated DSBs in the initiation of VSG switching . Loss of TbRECQ2 results in an increased rate of VSG switching , as a consequence of altered pathways of recombination: increased levels of telomere exchange and VSG-proximal gene conversion . Both these phenotypic outcomes are incompatible with the effects of TbRECQ2 loss on repair of I-SceI-mediated DSBs in the active BES , where the rate of repair is reduced and there is no change in repair pathway . The effects of TbRECQ2 loss on VSG switching are reminiscent of the phenotypes seen after mutation of Topo3α or RMI1 ( Kim and Cross , 2010; , 2011 ) , providing more evidence that these factors act together in T . brucei . Though VSG switching rates increase in each of the null mutants , the extent of this change is somewhat variable: a 2–3 fold increase in recq2-/- mutants , compared with 4-fold and 10–40 fold increases in rmi1 and topo3α null mutants , respectively ( Kim and Cross , 2010; 2011 ) . The differing extent of VSG switching increase may be consistent with findings that S . cerevisiae Sgs1 and Top3-Rmi1 can act independently on strand exchange intermediates ( Fasching et al . , 2015 ) , but may equally be explicable by subtle differences in the strains used or the growth conditions . In this regard , although the switching rate of the WT cells used by Kim and Cross is broadly similar to the data presented here , the switching profile is somewhat different . For example , in situ switchers were almost entirely absent ( <2% of total switchers ) in the Kim and Cross studies ( Kim and Cross , 2010; 2011 ) , whereas they constituted ~27% of WT switchers here; conversely , ~65% of WT switchers recovered by Kim and Cross used VSG GC , as opposed to only ~10% here . The pattern of WT VSG switching described here is very comparable with that seen by Povelones et al . ( 2012 ) , who used the same constructs but generated the cells independently from us . Thus , it seems likely that small differences in the constructs , their expression levels or their integration into the active BES may result in the WT VSG switching profile differences discussed above . Irrespective of these differences , the increased contribution of VSG GC and telomere XO to VSG switching is a common effect of recq2-/- ( ~50% and ~40% , respectively ) , rmi1-/- ( 70% and 25% , respectively ) and topo3α-/- ( 70% and 23% , respectively ) mutation ( Kim and Cross , 2010; 2011 ) . Similarly , all three -/- mutants display an almost complete absence of events leading to ES loss ( either GC or deletion ) . As has been argued before ( Kim and Cross , 2010; 2011 ) , the increase in telomere XO is striking , as this effect is consistent with the action of the RTR complex in processing recombination intermediates to suppress chromosome crossover ( Cejka et al . , 2010 ) . In yeast , Sgs1 mutation leads to increased crossover recombination after the formation of a chromosome-internal DSB ( Ira et al . , 2003 ) . The absence of crossovers in HRES recq2-/- mutants after I-SceI induction provides further evidence that an induced DSB in the T . brucei active BES is tackled by a repair strategy that differs markedly from that which directs VSG switching . Yeast Sgs1 mutants also display increased repair by break-induced replication ( BIR ) at a telomere-proximal DSB ( Lydeard et al . , 2010 ) . The decreased survival of HRES recq2-/- mutants after I-SceI induction indicates that , perhaps surprisingly , an induced DSB in the active BES is inefficiently repaired by BIR , despite the potential for this reaction to direct VSG switching of telomeric genes ( Boothroyd et al . , 2009 ) . Inefficient engagement of a DSB in HR-mediated repair is consistent with the qPCR we describe after I-SceI-mediated cleavage of the BES ( see below ) . In contrast , RTR action on recombination intermediates , including Holliday junctions , that can arise following replication fork stalling ( Mankouri and Hickson , 2007 ) may more readily explain the T . brucei RTR phenotypes , leading to the suggestion that replication-associated instability initiates VSG switching ( below ) . Targeted formation of a DSB to elicit recombination and allow temporal analysis of repair has been most extensively described in S . cerevisiae ( Hicks et al . , 2011; Renkawitz et al . , 2013 ) . Such studies suggest that endonuclease cleavage , DNA DSB processing and recruitment of Rad51 occur rapidly , with homology search and functional engagement of a homologous DNA substrate being slower reactions that follow from the break . As has been stated elsewhere ( Glover et al . , 2013 ) , the main effect of an I-SceI-induced DSB in the active BES is cell death , to a larger extent than the same lesion in the interior of chromosome 11 . By monitoring the presence of an intact I-SceI site , we show that cleavage occurs in HRES cells at least as rapidly as in HR1 , with few cells retaining an intact site after ~8 hr ( 1 cell cycle ) . I-SceI cleavage therefore appears to be rapid in T . brucei also , with a timing largely consistent with formation of single-stranded DNA , cell cycle impairment and the detection of damage by formation of either RAD51 or γH2A foci ( all at around 12 hr ) ( Glover et al . , 2013; Glover and Horn , 2014 ) . However , whereas loss of PCR-amplifiable I-SceI sequence is maintained in HR1 cells from ~8–72 hr after I-SceI induction , intact I-SceI target sequence reforms at least twice in HRES cells over the 32 hr after induction . After the I-SceI site is maximally cleaved in HR1 cells , recombination product ( primarily gene conversion from the unbroken chromosome 11 homologue ) gradually forms over the next 72 hr ( Glover et al . , 2008 ) , consistent with induction of repair . In contrast , the abundance of the VSG gene ( VSG221 ) positioned downstream of the I-SceI site in HRES closely mirrors that of the I-SceI target sequence and , moreover , virtually all HRES cells continue to express VSG221 protein 24 hr after I-SceI induction . Collectively , these data reinforce the evidence for distinct responses to a DSB in the two locations examined ( Glover and Horn , 2014 ) . In addition , these data reveal that repair efficiency and profile is markedly different in the active BES compared with the interior of chromosome 11 . In fact , the data suggest that recovery of VSG switchers following induction of a DSB in the active BES may not result from rapid induction of efficient VSG-directed HR . Instead , switchers may be selected: i . e . most HRES cells that suffer a DSB die because repair is inefficient , and the population is gradually replaced thereafter , initially by cells in which the I-SceI site has not been cut and gradually by cells that have undergone a VSG switch that removes the I-SceI target . Given the slow kinetics of repair after Rad51 loading in yeast ( Hicks et al . , 2011 ) , it seems likely that the explanation for the high rate of death after DSB formation in the active T . brucei BES is that strand exchange ( homology search , invasion or resolution ) is very inefficient in this setting . Why this might be awaits further analysis , but it may reflect the limited length or level of sequence identity between VSG flanks ( including the 70 bp repeats ) , or the complexity in searching for a repair substrate throughout the VSG archive . What might explain VSG switch initiation , if not the direct formation of a DSB , such as by an endonuclease ? MFAseq mapping , validated by VSG-focused qPCR , provides the first evidence that initiation could be linked to DNA replication . Throughout the T . brucei genome , we reveal pronounced rigidity in core genome replication timing , with the same origins used with the same efficiency in both BSF and PCF cells . In this context , the actively transcribed BES is unique , being the single mapped telomeric site that is early replicating , and the only locus in the genome that displays different replication timing in PCF and BSF cells . As all BES are silenced and late replicating in PCF cells , the singular early replication of the active BES in mammal-infective T . brucei suggests a model for VSG switch initiation in which transcription of the BES allows the site to become accessible for replication and establishes conditions that generate instability , most likely through replication-transcription clashes within the BES ( Figure 9 ) . Replication stalling can generate the structures that the RTR complex acts upon ( Cejka et al . , 2010; Mankouri and Hickson , 2007 ) , explaining the differing contributions of TbRECQ2 to VSG switching and I-SceI-induced DSB repair . Moreover , stalling of replication by transcription can lead to DNA rearrangements ( Bermejo et al . , 2012 ) and pausing of the replication fork has been shown to induce targeted mating type switching in fission yeast ( Klar et al . , 2014 ) . How early replication of the active BES in T . brucei BSF cells occurs is currently unclear , in particular because replication direction cannot be determined from the MFAseq data . Thus , two possibilities can be considered . In one scenario ( Figure 9B ) , replication initiates from the BES promoter environment , which would be compatible with the close association between origin activity and TbORC1/CDC6 binding at transcription start sites in the genome core ( Tiengwe et al . , 2012 ) . Though replication and transcription would be co-directional in this model , clashes between the two processes can occur in this arrangement due to the different rates of the reactions ( Merrikh et al . , 2011 ) . In addition , it is possible that replication or transcription could encounter progression difficulties in traversing the 70 bp repeats , providing some localisation of the clashes , which may lead to DNA breaks , including DSBs . A complication in this model is uncertainty about whether control of BES transcription is exerted by differential promoter activity in the active versus the silent BES ( Kassem et al . , 2014; Nguyen et al . , 2014 ) . In a second model ( Figure 9C ) , replication initiates at the telomere of the BES , leading to head-on collision with transcription; variation in the location of the collisions might explain the uncertainty of DSB mapping in the BES ( Boothroyd et al . , 2009; Glover et al . , 2013; Jehi et al . , 2014 ) , and events that lead to gene loss or replacement upstream of the 70 bp repeats . Recently , TbORC1/CDC6 has been documented to bind telomeres in T . brucei , but whether this is selective for the active BES , and if it reflects a role in gene silencing or directing replication , is unknown ( Benmerzouga et al . , 2013 ) . In contrast with the data here , previous work has shown that the replicated copies of the active BES are segregated later than silent BES during mitosis ( Landeira et al . , 2009 ) . This observation is not incompatible with unique early replication of the active BES and could , indeed , be explained by a delay in chromatid separation due to the increased presence of unresolved recombination intermediates between sister chromatids at this telomeric locus . 10 . 7554/eLife . 12765 . 016Figure 9 . Two models for replication-directed VSG switching . A schematic of a bloodstream VSGexpression site is shown ( A; not to scale ) , detailing key features ( left to right ) : the promoter ( flag ) , a number of expression site-asscociated genes ( ESAGs; blue boxes ) , 70 bp repeats ( hatched box ) , the VSG gene ( red box ) and the telomere repeats ( white arrows ) . Transcription direction is detailed in B and C ( red arrow ) , which compare the effects of replication initating ( black arrow ) at the promoter or at the telomere , with replication fork movement proceeding left to right , or right to left , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 12765 . 016 T . brucei BSF cells , strain Lister 427were used throughout and maintained in HMI-9 medium supplemented with 10% ( v/v ) FBS ( Sigma-Aldrich , Missouri , USA ) and 1% ( v/v ) of penicillin-streptomycin solution ( Gibco ) , at 37°C and 5% CO2 in vented flasks . BES1 TetR blockade cells ( Glover et al . , 2007 ) were generously provided by David Horn and Lucy Glover . Other genetically modified cells were generated by transfection , as described previously ( Burkard et al . , 2007 ) , and clones selected using the following drug concentrations: 10 μg . mLl-1 blasticidin , 5 μg . mL-1 G418 ( Neomycin , NEO ) , 10 μg . mL-1 hygromycin , and 0 . 2 μg . mL-1 puromycin . For VSG switching analysis , the GFP221hygTK cell line was generated based on work and plasmids described in Povelones et al . , ( 2012 ) ( constructs 221GP1 and HYG-TK; generous gift , Gloria Rudenko ) . These cells were cultured in thymidine-free medium: Isocove’s Modified Dulbecco’s Medium ( Gibco ) supplemented with 20% FBS ( Sigma-Aldrich ) , 1 mM hypoxanthine , 0 . 05 mM bathocuproine disulphonic acid , 1 mM sodium pyruvate , 1 . 5 mM L-cysteine and 200 μM β-mercaptoethanol . Procyclic cell forms ( PCF ) cells , strain Lister 427 , were cultured in SDM-79 ( Gibco ) supplemented with 10% ( v/v ) FBS ( Sigma-Aldrich ) , 1% ( v/v ) penicillin-streptomycin solution ( Gibco ) , and 5 µg . mLl-1 of haemin ( Sigma-Aldrich ) , at 27°C , in non-vented flasks . Cell density was assessed using a Neubauer improved hemocytometer , as standard . Heterozygous ( +/- ) and homozygous ( -/- ) knockout mutants of TbRECQ2 mutants were generated by deleting most of replacing most of the gene’s the open reading frame ( ORF; Figure 1 ) with a selective drug marker gene . Two modified versions of the pmtl23 plasmid ( gift , Marshall Stark , University of Glasgow ) , containing either the blasticidin or neomycin resistance genes , were used . In this system , the 5’ and 3’ flanking non-translated regions of TbRECQ2 ORF were PCR-amplified ( 5’ region – GATCTTCAAGCTTGCGGCCGCTGTGTAAATCCGTTCCTTTCTTC , and GATCTTCTCTAGATACAACGACACAATACCAACCAC; 3’ region – GATCTTCGAGCTCACAGACAATCTCCATCAGCAACC , and GATCTTCATCGATGCGGCCGCATAAGACATCCACCAGAACCTGC ) and cloned in a four-way ligation into the modified pmtl23 plasmid , with each flank surrounding the drug resistance gene . The selective drug marker flanked by the TbRECQ2 5’ and 3’ non-translated regions was then excised using NotI and transfected into BSF cells , and clones selected using 10 μg . ml-1 blasticidin or 5 μg . mL-1 G418 . Tbrecq2 mutants were analysed by RT-PCR , amplifying a 232 bp region of the RECQ2 ORF with primers TTTGTGATAACTGCGCAAGC and ACCTTGGAGTGAGCTGAACC; a part of TbPIF6 was amplified using primers GGTGGGTGTACGATCCATTC and TCGCCAAGGAGAATAACCTG as a control . RNA was extracted from the cells using the Qiagen RNeasy kit , and cDNA synthesis was performed using random primers and the Primer Design Precision nanoScript Reverse Transcription kit ( Primer Design ) , according to manufacturer’s instructions . In order to N-terminally epitope tag TbRECQ2 with 12myc , a modified version of the construct pEnT6B ( Kelly et al . , 2007 ) was used . In this case , two fragments were PCR-amplified: a region of the TbRECQ2 ORF immediately downstream of the start codon , using the primers CAGACTAGTTCTGTCCACAGAATTCAT ( containing an SpeI restriction site ) and CAGGGTACCAGGACAAAACACTAAAAAATA ( containing a KpnI site ) ; and a section from the 5’ flanking un-translated region immediately upstream of the TbRECQ2 ORF , using the primers CAGGGTACCGACAAAGATTTAAGTTGCGTCT ( containing a KpnI site ) and CAGGGATCCTCGCCGCGGTAATAGTTG ( containing a BamHI site ) . The resulting plasmid was then linearized using KpnI prior to transfection into BSF cells , and transformants were selected with 10 µg . ml-1 blasticidin . For the VSG switching analysis , MITat1 . 2 BSF cells were first transformed with 221GP1 ( Sheader et al . , 2004 ) after digestion with NotI and XhoI; transformants were selected with 0 . 2 μg . mL-1 puromycin . These cells were then transformed with the construct HYG-TK , which was digested with NotI and HindIII prior to transformation . Significant difficulty was encountered in propagating the HYG-TK construct without rearrangement , and growth in E . coli XL 10 Gold Cells ( Stratagene ) and ZYM-5 medium appeared to provide greatest stability . Prior to transformation with HYG-TK , the eGFP-PUR cells were cultured in medium lacking thymidine , and transformants were selected in the same medium using 0 . 2 μg . mL-1 puromycin and 10 μg . mL-1 hygromycin . Integration of the 221GP1 construct was confirmed by PCR using primers GTGACCACCCTGACCTAC and GCAAACTGTGATGACCCGC . Integration of the HYG-TK construct was confirmed by PCR using primers TTTACGGGCTACTTGCCATT and CCTCATTTTGGATTTTGCTCCT . Expression of eGFP and VSG221 was confirmed by western blotting ( antisera below ) . Standard ( default settings ) protein-protein Basic Local Alignment Search Tool ( BLAST ) ( blastp ) was used to identify potential RecQ helicase-encoding genes in T . brucei . Searches were performed using 18 RecQ protein sequences from Homo sapiens , Mus musculus , Arabidopsis thaliana , Caenorhabditis elegans and Saccharomyces cerevisiae as queries and the T . brucei Lister 427 strain genome as target ( http://tritrypdb . org/tritrypdb/ ) , revealing two genes . Reciprocal blastp analysis was then performed against the non-redundant protein sequences database ( default settings ) ( http://blast . ncbi . nlm . nih . gov/Blast . cgi ) using the predicted protein sequences of both putative T . brucei RecQ-like genes; for TbRECQ2 the top hits ( lowest E values ) were all eukaryotic RecQ helicases . Protein domain analysis of TbRECQ2 predicted sequence was conducted using Pfam , version X ( http://pfam . xfam . org/ ) , and InterProScan sequence search ( http://www . ebi . ac . uk/interpro/search/sequence-search/ ) . Prior to setting up of the clonal survival assay , all cultures were passaged into drug free medium . The cell cultures were then diluted to 0 . 5 x 101 cells . mL-1 in media containing either 0 . 05 μg . mL-1 , 0 . 075 μg . mL-1 or 0 . 1 μg . mL-1 phleomycin , 0 . 02 mM , 0 . 03 mM or 0 . 04 mM hydroxyurea , or 0 . 0001% , 0 . 0002% , 0 . 0003% or 0 . 0004% MMS , or none as the untreated control . Each of these cultures were then distributed in 200 μl aliquots into three 96 well plates , and the number of surviving clones quantified after ~10 days growth . The mean survival of the treated samples was determined relative to untreated samples for each damaging agent concentration used and for each cell line , with each of the above experiments repeated at least three times . Assays were carried out as in ( Glover et al . , 2013 ) . Cells were cultured to mid-log phase ( 1 x 106 cells . mL-1 ) in Tet-free medium containing phleomycin , puromycin and hygromycin to maintain the I-SceI genetic components in the cells . For clonal survival , cultures were then diluted to sub-clonal dilutions ( HR1 cell lines: 0 . 15 x 101 cells . mL-1; HRES cell lines: 0 . 26 x 101 cells . mL-1 ) , divided into two aliquots , and Tet ( Calbiochem ) added to one ( final concentration 2 μg . mL-1 ) to induce I-SceI expression . Cultures were distributed in 200 μL aliquots into 96 well plates ( four plates each of uninduced and induced cells ) . After 7–10 days incubation the number of surviving clones was counted and survival was normalised to uninduced cultures . Mean survival in the induced cells was determined from multiple independent repeats the above experiments . Presence of the I-SceI target and VSG221 in each cell line prior to and after induction of I-SceI expression was evaluated by quantitative real-time PCR ( qPCR ) . To do this , ~1 x 106 cells were collected at various time points , and gDNA extracted using the Qiagen Blood and Tissue kit , which was then quantified using the Quant-iT PicoGreen dsDNA Assay Kit ( Life Technologies ) . Each DNA sample was diluted to 0 . 2 ng . ul-1 and 1 ng was analysed by qPCR using Precision qPCR MasterMix with SYBR Green and low ROX ( Primerdesign ) , and 6 pmol of primers ( Eurofins MWG Operon , Ebersberg , Germany ) , to a total of 25 μl per reaction . For each pair of primers ( below ) , triplicates of each sample were run per plate ( MicroAmp Optical 96-well Reaction Plate , Life Technologies ) , which were sealed with MicroAmp clear adhesive film ( Life Technologies ) . All experiments were run in a 7500 Real Time PCR system ( Applied Biosystems ) , using the following PCR cycling conditions: 95°C for 10 min , followed by 40 cycles of 95°C for 15 sec and 60°C for 1 min ( fluorescence intensity data collected at the end of the last step ) . Data was then analysed by relative quantification using the ΔΔCt method ( 7500 software version 2 . 3 , Applied Biosystems ) ( Livak and Schmittgen , 2001 ) . Abundance of the I-SceI site was evaluated using primer pairs CACAACGAGGACTACACCATC and CGGCCTATTACCCTGTTATCC ( HR1 cell line ) , or GTTGTGAGTGTGTGCTTACC and ATCTAGAGGATCTGGGACCC ( HRES ) . VSG221 abundance was assessed using primers AGCTAGACGACCAACCGAAGG , and GTTTCCTCTCGCCGTGGTCGC . The generation of HR product in the HR1 cells ( data not shown ) was determined using primers CACATTCACTTGACCATTCG and GATGCACTTCGAGAGCGTCAG , which recognise a chromosomal sequence deleted during insertion of the I-SceI target constructs , meaning abundance increases from 1n to 2n after gene conversion from that intact homologue . In all cases , product abundance was determined relative to one of two control loci , which were amplified with primer pairs TCTGAACCCGCGCACTTC and CCACTCACGGACTGCGTTT , or TTGTGACGACGAGAGCAAAC and GAAGTGGTTGAACGCCAAAT . Cells were harvested by centrifugation at 405 g for 10 min , and washed with 1x PBS supplemented with 15 . 7 g/L sucrose and 1 . 8 g/L glucose , pH 7 . 4 . ~2 x 106 cells were then loaded onto each well of a 12-well glass slide ( Menzel-Gläser ) , pre-treated with Poly-L-lysine ( Sigma-Aldrich ) , and allowed to settle for 5 min . The cells were then fixed with 3 . 7% paraformalydehyde ( PFA ) for four minutes , and permeabilised for 10 min with 0 . 2% Triton X-100 ( Promega , in 1x PBS ) . Next , 100 mM glycine was added and incubated for 5 min , twice . The cells were then washed with 1x PBS twice , 5 min each , and incubated with 1% BSA and 0 . 2% Tween-20 ( Sigma-Aldrich ) in 1x PBS , for 1 hr . Afterwards , the cells were incubated for 1 hr with mouse anti-myc antiserum conjugated with Alexa Fluor 488 ( Millipore ) diluted 1:2000 in 1% BSA and 0 . 2% Tween-20 in 1x PBS . The cells were washed twice with 1x PBS , after which Fluoromount G with DAPI mounting medium ( SouthernBiotech ) was added and incubated for 3 min . The slide was then covered with a coverslip and sealed with nail varnish . RAD51 was detected in cells treated in the same way , but using polyclonal anti-RAD51 antiserum as described previously ( Trenaman et al . , 2013 ) . VSG221 was detected using a rabbit α- VSG221 antiserum ( gift David Horn , University of Dundee ) diluted 1:10000 , and Alexa Fluor 594 conjugated goat α- rabbit antiserum ( Molecular Probes ) diluted 1:1000 . EP-procyclin was detected using mouse IgG1 α-EP procyclin antiserum ( clone TBRP1/247 , Cedarlane ) diluted 1:500 , and Alexa Fluor 488 conjugated goat α-mouse antiserum ( Molecular Probes ) diluted 1:1000 . Images were acquired and examined as described above . Images were acquired using a Zeiss Axioskop 2 fluorescent microscope attached to an HBO100 lamp and a digital ORCA-ER camara and camera controller ( Hamamatsu Photonics ) , using the Volocity 6 . 1 . 1 Cellular and Imaging Analysis software ( Perkin Elmer ) . Images were further analysed using Fiji ( http://fiji . sc/Fiji ) . A culture of the GFP221hygTK cell growing in thymidine-free medium supplemented with 0 . 2 μg . mL-1 puromycin and 10 μg . mL-1 hygromycin was passaged to a density of 1 x 104 cells . mL-1 in media lacking hygromycin and incubated for 48 hr to allow VSG switched variants to arise; in some experiments the cells were grown in the absence of puromycin , while in others puromycin ( 0 . 2 μg . mL-1 ) was retained . After 48 hr the cultures were diluted to 2 . 5 x 103 cells . mL-1 , 5 x 103 cells . mL-1 or 1 . 25 x 104 cells . mL-1in the presence of 4 μg . mL-1 ganciclovir ( Sigma-Aldrich ) and plated in 200 μl aliquots over three 96 well plates , resulting in 0 . 5 , 1 . 0 or 2 . 5 x 103 cells per well . After 7 days growth multiple surviving clones were randomly selected on a random basis and scaled-up in thymidine- and drug-free medium for further analysis ( below ) . The final number of surviving clones was only assessed after a further 7–10 days . VSG switching frequency was calculated in each experiment by dividing the total number of surviving clones was divided by number of cells plated , to obtain the number of switching events per cell . This number was then divided by the number of generations in the 48 hr incubation prior to plating the cells , to obtain the VSG switching rate ( switchers/cell/generation ) . The number of generations was calculated for each cell line using the cell density measured at the 48 hr time point . To analyse VSG switching events , the expanded clones were collected and both genomic DNA was extracted , for PCR analysis , and whole cell extracts ( 2 . 5 x 106 cells per sample ) for western blot analysis . VSG221 was detected using rabbit anti-VSG221 antiserum diluted 1:20 . 000 ( gift , David Horn ) , while eGFP was detected with a rabbit anti-GFP antiserum ( Abcam ) diluted 1:5000 . Both primary antisera were used in combination with the goat anti-rabbit IgG ( H+L ) horseradish peroxidase ( HRP ) conjugate antiserum ( Molecular Probes ) diluted 1:5000 . Ponceau staining of the membrane was carried out to confirm that protein was present on the blot for clones that were VSG221- and eGFP- . The presence of VSG221 and GFP genes was assessed by PCR using the primer pairs GCAAGTATATACGCTGAAATAAATCAC and TGTTTGGCTGTTCGCTACTGTGAC ( VSG221 ) , and CTTCTTCAAGTCCGCCAT and GCTCAGGTAGTGGTTGTC ( GFP ) . RNA polymerase I large subunit ( Tb427 . 08 . 5090 ) was PCR-amplified as a positive control using the primers CTGGATCCAGCGCCGTTCCACGCGAGA and GACTCGAGCTATCCCCAATCCGTGCCGTCCCG . For each sorting , 3 x 108 cells were collected from an exponentially growing BSF cell culture ( ~1 x 106 cells . ml-1 ) , and centrifuged for 10 min at 1000 g . Cells were then re-suspended in 25 ml of 1x PBS and centrifuged for 10 min at 1000 g . The pellet was then re-suspended in 500 μl of 1x PBS , and 9 . 5 ml of 1% formaldehyde ( methanol-free , Thermo Scientific , diluted in 1x PBS ) was added for 10 min at room temperature . The cells were then centrifuged for 10 min at 1000 g , washed once in 10 ml 1x PBS , and centrifuged again . The pellet was next re-suspended to a concentration of 2 . 5 x 107 cells . ml-1 in 1x PBS , and was stored protected from light at 4°C overnight . The fixed cells were then centrifuged for 10 min at 1000 g , and incubated in 20 ml of 0 . 01% Triton X-100 ( Promega ) in 1x PBS for 30 min at room temperature . Next , the cells were centrifuged for 10 min at 700 g , washed in 20 ml of 1x PBS , and centrifuged again . The resulting pellet was then re-suspended to a concentration of 2 . 5 x 107 cells . ml-1 in 1x PBS with 10 μg . ml-1 of propidium iodide ( PI , Sigma-Aldrich ) and 100 μg . ml-1 RNase A ( Sigma-Aldrich ) , and incubated for 1 hr at 37°C , protected from light . For PCF cells , 3 x 108 were collected from an exponentially growing PCF culture ( ~1 x 107 cells . ml-1 ) and centrifuged for 10 min at 1620 g . The pellet was then washed in 10 ml of 1x PBS supplemented with 5 mM EDTA ( Gibco ) , and centrifuged for 10 min at 1620 g . Next , the cells were re-suspended in 12 ml of 1x PBS supplemented with 5 mM EDTA , to which 28 ml of 100% ice cold-Methanol was added , in a drop-wise fashion while vortexing gently , so that the final fixing solution was 70% ( v/v ) Methanol , and the cell concentration was 2 . 5 x 107 cells . ml-1 . The cells were then kept at 4°C , protected from light , from overnight up to three weeks . For each FACS sorting session , four FACS tubes ( Becton Dickinson ) were prepared , each starting with ~1 x 108 fixed cells . The cells were collected and centrifuged for 10 min at 1000 g , at 4°C , washed in 1 ml of 1x PBS supplemented with 5 mM of EDTA , and centrifuged again for 10 min at 1000 g , at 4°C . The pellet was then re-suspended in 4 ml of 1x PBS supplemented with 5 mM EDTA , 10 μg . ml-1 PI and 10 μg . ml-1 RNase A , and incubated for 45 min at 37°C , in the dark . The cells ( either BSF or PCF ) were then transferred to a FACS tube through a cell strainer cap ( BD Biosciences ) , and sorted into G1 , early S , late S and G2 phases using a BD FACSAria I Cell Sorter ( BD Biosciences ) . The sorted cells were collected at 4°C into new FACS tubes containing 200 μl of lysis buffer ( 1 M NaCl , 10 mM EDTA , 50 mM Tris-HCL pH 8 . 0 , 0 . 5% SDS , 0 . 4 mg . ml-1 Proteinase K , and 0 . 8 μg . ml-1 of Glycogen ) . After the sorting has been completed , the collected cells were then incubated for 2 hr at 55°C , and the lysate was stored at -20°C . Genomic DNA was extracted using a Blood and Tissue DNA extraction kit ( Qiagen ) , by omitting the lysis steps of the manufacturer’s protocol . Sequencing was performed by Eurofins Genomics ( Germany ) ; the DNA library was prepared using the TruSeq DNA Sample Preparation kit ( Illumina ) , and sequenced using Illumina HiSeq paired- end 100 bp sequencing system ( Illumina ) . The samples were multiplexed , with each of the early S , late S , and G2 phase samples library DNA , both from BSF and PCF , being processed in the same run , for ease of comparison . Data from the sequencing was first analysed for quality using FastQC ( http://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) , and then trimmed using fastq-mcf ( http://code . google . com/p/ea-utils ) , to exclude the adapter sequences used during the library preparation and sequencing . The reads were then aligned to the reference genome ( T . brucei Lister 427 , retrieved from TriTrypDB version 8 . 0 ) , using Bowtie2 ( version 2 . 2 . 0 --very-sensitive-local -k1 ) ( Langmead and Salzberg , 2012 ) . The aligned reads were then compared using a method adapted from the one described previously ( Tiengwe et al . , 2012 ) , but simplified to facilitate inter-species comparisons . Briefly , the reads were binned in 2 . 5 Kbp sections along each chromosome , and the number of reads in each bin was then used to calculate the ratios between early S and G2 , as well as between late S and G2 samples , scaled for the total size of the read library ( reads per 2 . 5 Kbp per million reads mapped ) . These data were then represented in a graphical form using Prism 6 ( GraphPad software Inc . ) . Shell scripts used to generate these data are freely available in BitBucket ( https://bitbucket . org/WTCMPCPG/tb_antigenic_variation ) . Sequences are available at the European Nucleotide Archive: PRJEB11437 . MFA by qPCR was performed as described in ( Marques et al . , 2015 ) with primers targeting VSG221 ( AGCAGCCAAGAGGTAACAGC and CAACTGCAGCTTGCAAGGAA ) , VSG121 ( AGGAAGGCAAATACGACCAG and TTTGCGGGTAAAAGTCCTTG ) and selected origin ( TCCCAGAAACCAACTTCAGC and AGTTGGATTGCCATGTCCTC ) and non-origin regions ( GGCTGGATGATGAGAGGAAC and CCTCCAACCTCAAGATACGC ) in chromosome 5 . For normalization , a non-origin region in chromosome 2 was used ( CTCGCTCTCCGTACAGTTG and CACTCGTCGATGCAACCTC ) . For each sample , 0 . 18 ng of gDNA was used , and the data shown are averages of itriplicate experiments .	The African trypanosome , Trypanosoma brucei , is a parasite that is transmitted between mammals by the tsetse fly , and causes a disease known as sleeping sickness in humans . Like many other parasites , trypanosomes have evolved ways to avoid being killed by their hosts . One such survival strategy involves the parasites constantly changing the molecules that coat their surface , which are the main targets recognized by their hosts’ immune systems . Switching one coat protein for another similar protein , a process called antigenic variation , allows a parasite to evade an attack and establish a persistent infection . Antigenic variation also makes it almost impossible to develop a vaccine that will offer lasting protection against the parasite . Previous research suggested that a trypanosome might deliberately break its own DNA and then exploit a repair process to switch its current coat protein-encoding gene for another one located elsewhere within its genetic material . Devlin , Marques et al . now reveal that it is unlikely that trypanosomes use a specific enzyme to break DNA deliberately during coat switching . Instead , experiments using whole-genome sequencing suggest that coat-gene-switching might arise from the strategies trypanosomes use to copy their genetic material during cell division . These findings bring researchers closer to understanding how trypanosomes start antigenic variation in order to evade their hosts’ immune responses . In addition , the findings suggest a new model that could help researchers answer an important question: how does the timing of genome copying vary from cell to cell ? Nevertheless , the hypothesis proposed by Devlin , Marques et al . will now require rigorous testing . Future studies could also ask if other parasites use similar strategies to survive being attacked by their host’s immune systems .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "microbiology", "and", "infectious", "disease" ]	2016	Mapping replication dynamics in Trypanosoma brucei reveals a link with telomere transcription and antigenic variation
To maintain cellular structure and integrity during division , Gram-negative bacteria must carefully coordinate constriction of a tripartite cell envelope of inner membrane , peptidoglycan ( PG ) , and outer membrane ( OM ) . It has remained enigmatic how this is accomplished . Here , we show that envelope machines facilitating septal PG synthesis ( PBP1B-LpoB complex ) and OM constriction ( Tol system ) are physically and functionally coordinated via YbgF , renamed CpoB ( Coordinator of PG synthesis and OM constriction , associated with PBP1B ) . CpoB localizes to the septum concurrent with PBP1B-LpoB and Tol at the onset of constriction , interacts with both complexes , and regulates PBP1B activity in response to Tol energy state . This coordination links PG synthesis with OM invagination and imparts a unique mode of bifunctional PG synthase regulation by selectively modulating PBP1B cross-linking activity . Coordination of the PBP1B and Tol machines by CpoB contributes to effective PBP1B function in vivo and maintenance of cell envelope integrity during division . Cell shape and osmotic stability are maintained by the stress-bearing peptidoglycan ( PG ) sacculus ( cell wall ) in nearly all bacteria ( Vollmer et al . , 2008 ) . The sacculus , a continuous mesh-like structure of glycan strands cross-linked by short peptides , encases the inner ( cytoplasmic ) membrane ( IM ) and is essential for viability . Several prominent classes of antibiotics ( e . g . , β-lactams and glycopeptides ) inhibit PG synthesis , causing lysis and cell death ( Schneider and Sahl , 2010 ) . In Gram-negative bacteria , the outer membrane ( OM ) , an asymmetric bilayer of phospholipids and lipopolysaccharides , surrounds the mostly single-layered sacculus ( Gan et al . , 2008 ) and forms a vital permeability barrier ( Nikaido , 2003 ) . Covalent and non-covalent interactions between abundant OM proteins ( Lpp , Pal , OmpA ) and PG tether the OM to the sacculus , maintaining OM stability ( Hantke and Braun , 1973; Parsons et al . , 2006 ) . Together , the IM , PG , and OM comprise the tripartite Gram-negative cell envelope ( Figure 1A ) . Bacteria must synchronize growth and division of these layers , as imbalanced growth could lead to breaches that compromise the permeability barrier or even the structural integrity of the cell . However , we do not yet understand the mechanisms that accomplish this synchronization . 10 . 7554/eLife . 07118 . 003Figure 1 . Envelope constriction in Gram-negative bacteria and related protein machines . ( A ) Illustration of the tripartite Gram-negative cell envelope . In zoom ( right ) , processes involved in envelope constriction . ( B ) Major bifunctional PG synthases of E . coli , capable of both glycan strand elongation ( glycosyltransferase activity , GT ) and peptide cross-linking ( transpeptidase activity , TP ) . The IM-localized synthases are activated by cognate regulatory OM lipoproteins . ( C ) Proteins encoded by the tol-pal operons ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 003 Coordinating growth across the layers of the Gram-negative bacterial cell envelope is complex , particularly since all energy and precursors for assembling and constricting these layers must come from the cytoplasm . To overcome this challenge , bacteria utilize IM-associated multicomponent machineries that span the entire envelope . Two machineries , organized by distinct cytoskeletal elements , assemble and disassemble in a cell-cycle-regulated manner and mediate different phases of sacculus growth ( Typas et al . , 2012 ) : ( 1 ) the cell elongation machinery ( elongasome ) , organized by the actin homolog MreB , mediates lateral PG synthesis along the length of the cell , and ( 2 ) the cell division machinery ( divisome ) , organized by the tubulin homolog FtsZ , mediates new pole synthesis at the septum ( Egan and Vollmer , 2013 ) . These complex machineries are comprised of structural and regulatory subunits , components with distinct functions ( e . g . , DNA segregation , PG precursor synthesis and transport ) , and PG biosynthetic and modifying enzymes . Sacculus growth is orchestrated by a repertoire of PG synthases , including glycosyltransferases ( GTases ) that polymerize glycan strands from the precursor saccharide moiety lipid II , transpeptidases ( TPases ) that cross-link peptides between adjacent glycan strands , and bifunctional PG synthases that carry out both activities ( Typas et al . , 2012 ) . The two Escherichia coli monofunctional TPases , PBP2 , and PBP3 , are essential subunits of the elongasome and the divisome , respectively . Likewise , the two major bifunctional PG synthases , PBP1A and PBP1B , participate predominantly in elongation and division , respectively ( Bertsche et al . , 2006; Typas et al . , 2010; Banzhaf et al . , 2012 ) . However , in contrast to the monofunctional TPases , which are dedicated to their respective roles , the bifunctional synthases can partially substitute for each other , enabling cells to survive with only one of them ( Yousif et al . , 1985 ) . These IM-localized bifunctional synthases have obligate cognate regulatory OM lipoproteins , LpoA and LpoB , which are required for activity in vivo ( Paradis-Bleau et al . , 2010; Typas et al . , 2010 ) . The Lpo activators span most of the periplasm ( ∼210 Å in width; Matias et al . , 2003 ) and traverse the sacculus ( ∼40–60 Å pore size; Demchick and Koch , 1996; Vazquez-Laslop et al . , 2001 ) to interact with their partner PBPs ( Egan et al . , 2014; Jean et al . , 2014 ) , forming trans-envelope PG synthase complexes ( Figure 1B ) . Electron microscopy studies first indicated that distances between the OM , PG , and IM remain remarkably consistent throughout cell division , providing an early indication that envelope constriction processes occur in close proximity to each other and are tightly coordinated ( Weigand et al . , 1976; Fung et al . , 1978; MacAlister et al . , 1987; Bi and Lutkenhaus , 1991 ) . It is now clear that IM constriction , PG synthesis , and subsequent PG hydrolysis to separate daughter cells ( septal cleavage ) are coordinated via the divisome . FtsZ forms a ring-like structure in the cytoplasm that provides the membrane contractile force ( Osawa et al . , 2009 ) , and together with FtsA ( Szwedziak et al . , 2012; Osawa and Erickson , 2013; Loose and Mitchison , 2014; Szwedziak et al . , 2014 ) serves as a scaffold for divisome assembly , including recruitment of PG synthases and hydrolases ( Egan and Vollmer , 2013 ) . Septal PG synthesis , principally orchestrated by PBP3 and PBP1B ( Bertsche et al . , 2006 ) , occurs at the leading edge of the inward-moving septum , adjacent to the invaginating IM ( Figure 1A ) . Septal cleavage , controlled by tightly regulated periplasmic amidases ( Heidrich et al . , 2001; Uehara et al . , 2010 ) , follows closely after synthesis and adjacent to the invaginating OM . Both topological constraints and regulatory input from IM and/or OM proteins ensure tight spatial regulation of septal cleavage ( Uehara et al . , 2010; Yang et al . , 2011 ) . OM constriction is promoted by the energy-transducing Tol system , which localizes to mid-cell during the later stages of cell division in a divisome-dependent manner ( Gerding et al . , 2007 ) . IM proteins TolQ , TolR , and TolA , which form a complex ( Derouiche et al . , 1995; Lazzaroni et al . , 1995; Journet et al . , 1999 ) , periplasmic TolB , and OM lipoprotein Pal are all encoded in two adjacent operons ( Figure 1C , D ) . Loss of any of these components results in delayed OM constriction and defects in OM integrity , leading to OM blebbing , periplasmic leakage , and pleiotropic drug and stress sensitivities ( Bernadac et al . , 1998; Cascales et al . , 2002; Gerding et al . , 2007 ) . For function , Tol harnesses proton motive force ( PMF ) via TolQR , a homolog of the flagellar motor MotAB ( Cascales et al . , 2001 ) . This has been proposed to energize TolA , inducing it to adopt an extended conformation and interact with TolB and/or Pal ( Cascales et al . , 2000; Germon et al . , 2001; Lloubes et al . , 2001 ) ; cycles of Tol–Pal interaction and release are then thought to promote OM invagination ( Gerding et al . , 2007 ) . This interaction model has been challenged , however ( Bonsor et al . , 2009 ) , and the mechanism by which the Tol system promotes OM constriction remains to be fully elucidated . Further , how Tol-facilitated OM constriction is coordinated with septal PG synthesis and other envelope constriction processes has remained completely unknown . We previously identified a genetic link between PBP1B-LpoB and Tol ( Typas et al . , 2010 ) . Here , we report that physical and functional coordination of the two machines is required to properly synchronize PG synthesis and OM constriction during cell division . We implicate YbgF , previously of unknown function , in mediating this coordination , and therefore name it CpoB , or Coordinator of PG synthesis and OM constriction , associated with PBP1B . We show that CpoB , PBP1B-LpoB , and Tol localize concurrently to the septum during cell division , and interact to form a higher-order complex that spatially links PG synthesis and OM invagination . These physical interactions are dynamic and allow direct regulation of PBP1B activity in response to Tol assembly and cycles of PMF utilization . CpoB is required for proper PBP1B function in vivo , and loss of CpoB-mediated coordination between PBP1B and Tol leads to defects in OM integrity , illustrating the importance of mechanisms that ensure coordination of cell division processes across the envelope . To identify novel regulators of PG synthesis during cell division , we queried the E . coli chemical genomics database ( Nichols et al . , 2011 ) for deletion strains whose growth responses closely resembled those of a strain lacking PBP1B ( encoded by mrcB ) across a range of drug and environmental stress conditions . This approach previously identified LpoB , the OM lipoprotein activator of PBP1B ( Typas et al . , 2010 ) . ΔcpoB exhibited the second highest correlation with ΔmrcB ( Figure 2A; correlation coefficient = 0 . 47 , p < 10−18 ) , suggesting that CpoB may also be functionally associated with PBP1B . CpoB ( YbgF ) is encoded by the last gene in the tol-pal operon and binds TolA in vitro ( Krachler et al . , 2010 ) , but its deletion does not cause the severe OM integrity defects typically associated with tol-pal deletions ( Vianney et al . , 1996 ) . Thus , CpoB is not critical for Tol function , and its cellular role has remained enigmatic . 10 . 7554/eLife . 07118 . 004Figure 2 . Shared chemical sensitivities and genetic interactions implicate CpoB in PBP1B function . ( A ) Chemical genetic phenotype profiles for strains lacking PBP1B ( encoded by mrcB ) , LpoB , and CpoB , from hierarchical clustering of the growth fitnesses of 3979 gene deletions across 324 chemical stress conditions ( Nichols et al . , 2011 ) . Blue , low fitness ( sensitivity ) ; yellow , high fitness ( resistance ) . ΔmrcB and ΔcpoB ( ΔybgF ) showed significant similarity . Underlined regions correspond to strongest shared sensitivities and are magnified below the full profile . ( B ) Sensitivity of ΔcpoB to cefsulodin , predominantly an inhibitor of PBP1A , is alleviated by PBP1B overexpression . Growth fitness was assessed by pinning cells to agar plates in a 1536-spot format and quantifying colony size after 12–14 hr of growth . Error bars depict standard deviations ( n ≥ 44 ) . ( C ) CpoB-associated envelope defects visualized by fluorescence microscopy using periplasmic mCherry . Arrowheads , fluorescent foci indicating OM blebs or vesicles; outlines , loss of peripheral fluorescence due to periplasmic leakage . ( D ) and ( E ) . Quantification of microscopy phenotypes: ( D ) OM blebbing , and ( E ) loss of periplasmic fluorescence , indicative of periplasmic leakage . Threshold value was established based on quantification of cells with known severe OM defects ( ΔtolA ) . Cells lacking CpoB exhibited envelope defects that were severely exacerbated in the absence of LpoA and alleviated by PBP1B overexpression . ( F ) Quantification of growth fitness for indicated mutant strains . A strain lacking CpoB and LpoA exhibited a synthetic growth defect that was alleviated by PBP1B overexpression . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 00410 . 7554/eLife . 07118 . 005Figure 2—figure supplement 1 . Elevated cell lysis in cells lacking PBP1B or CpoB; loss of LpoA exacerbates lysis in the absence of CpoB . ( A ) and ( B ) Increased cell envelope lysis or permeability measured by a CPRG assay ( Paradis-Bleau et al . , 2014 ) at different salt concentrations . The colorless LacZ substrate CPRG cannot enter intact cells , where it would yield red CPR , unless the envelope is severely compromised or the cells lyse . Assay was performed on agar plates , on which mutants were arrayed robotically in a 384-colony format ( Paradis-Bleau et al . , 2014 ) . Box plots of rate of accumulation of CPR color are shown ( n > 32 ) ; rates are more sensitive and robust compared to end-point measurements . Box corresponds to 25–75% ( IQR ) . Both ΔmrcB and ΔcpoB mutants exhibit elevated cell lysis , but the effect is not additive . As the double ΔmrcB ΔcpoB mutant phenocopies the single ΔmrcB mutant , we deduce that PBP1B is epistatic to CpoB . In contrast , CpoB and LpoA appear to work in redundant pathways , as the double ΔlpoA ΔcpoB mutant exhibits much stronger cell lysis than either of the parental mutants . ( C ) A ΔlpoA ΔcpoB mutant exhibits periplasmic leakage and lysis during growth . Immunodetection of periplasmic ( mCherryperi , α-mCherry ) and cytoplasmic ( RNA polymerase β' subunit , α-RNAP β'; GroEL ) cell contents in filtered culture supernatants ( sup ) vs pelleted cells ( pel ) ; presence of mCherry in sup indicates periplasmic leakage; presence of RNAP β' and GroEL in sup indicates lysis . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 00510 . 7554/eLife . 07118 . 006Figure 2—figure supplement 2 . Complementation of ΔcpoB phenotypes . ( A ) Expression of CpoB in trans complements ΔcpoB cefsulodin sensitivity ( see Figure 2B ) . ( B ) Expression of CpoB in trans complements the ΔcpoB ΔlpoA double mutant growth defect ( see Figure 2F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 00610 . 7554/eLife . 07118 . 007Figure 2—figure supplement 3 . LpoA encodes a second function , and the LpoA TPR domain is dispensable for PBP1A activation . ( A ) and ( B ) Quantification of growth fitness for indicated mutant strains ( see Figure 2F ) . ( A ) A strain lacking CpoB and LpoA exhibited a synthetic growth defect , but a strain lacking CpoB and PBP1A ( encoded by mrcA ) did not . A triple ΔcpoB ΔlpoA ΔmrcA mutant also showed a growth defect , indicating that the growth defect of the ΔcpoB ΔlpoA double mutant is not caused by PBP1A malfunction . ( B ) The LpoA TPR domain is dispensable for PBP1A activation . Lack of PBP1B ( encoded by mrcB ) and LpoA is synthetic lethal because PBP1A requires activation by LpoA; however a strain lacking PBP1B and the LpoA TPR domain shows no fitness defect . ( C ) Summary of genetic relationships . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 00710 . 7554/eLife . 07118 . 008Figure 2—figure supplement 4 . The N-terminal TPR domain of LpoA encodes a novel function that compensates for loss of CpoB . ( A ) Structure of LpoA ( Jean et al . , 2014 ) and summary of domain functions . The C-terminal domain is required for PBP1A activation; the N-terminal TPR domain is dispensable for PBP1A activation ( Figure 2—figure supplement 3B ) but required for a second , CpoB-related function . ( B ) and ( C ) Quantification of microscopy phenotypes ( see Figure 2D , E ) . ( B ) OM blebbing . ( C ) Loss of periplasmic fluorescence , indicative of periplasmic leakage . ( D ) Quantification of growth fitness for indicated mutant strains ( see Figure 2F ) . CpoB/compensatory function: + , strains that possess either CpoB , the compensatory function of the LpoA TPR domain , or both; − , strains that lack both . PBP1A function: + , strains that possess both PBP1A and the LpoA C-terminal domain , which is required for PBP1A activation in vivo; − , strains that lack one or both . A ΔcpoB ΔmrcA lpoA ( ΔTPR ) triple mutant lacks both CpoB/compensatory function and PBP1A function , and thus recapitulates the envelope and growth defects of the ΔcpoB ΔlpoA mutant . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 008 ΔcpoB and ΔmrcB mutants shared increased sensitivities to multiple β-lactams ( Figure 2A ) , including cefsulodin , which preferentially inhibits PBP1A . As loss of both PBP1A and PBP1B function is synthetically lethal ( Yousif et al . , 1985 ) , the enhanced cefsulodin sensitivity of ΔcpoB likely derives from defects in PBP1B function . Consistent with this idea , overexpression of PBP1B abrogated the cefsulodin sensitivity of ΔcpoB ( Figure 2B ) . Additionally , both ΔcpoB and ΔmrcB mutants exhibited increased lysis in an envelope stability screen ( Paradis-Bleau et al . , 2014 ) . Importantly , the lysis phenotype of ΔcpoB ΔmrcB did not exceed that of the ΔmrcB single mutant ( Figure 2—figure supplement 1A , B ) . This epistatic relationship strongly suggests that CpoB acts via PBP1B . Expression of CpoB in trans complemented ΔcpoB phenotypes , confirming that they result from loss of CpoB rather than indirect effects on upstream tol-pal gene expression ( Figure 2—figure supplement 2 ) . Having established a functional relationship between CpoB and PBP1B , we asked whether they interact physically in vivo . Indeed , we identified CpoB in a screen for novel PBP1B interaction partners , using in vivo photo-cross-linking following incorporation of a non-natural amino acid ( pBpa ) at specific exposed sites ( Chin and Schultz , 2002 ) . Initially focusing on the PBP1B UB2H domain , we identified two sites ( T118 and E123 ) , of 19 examined , that cross-linked to a protein of approximately 30 kDa ( Figure 3A , B ) . These sites reside in a cleft between the UB2H and TPase domains . Testing opposing amino acids in the TPase domain ( T751 and T753 ) yielded cross-link products of the same size ( Figure 3A , B ) . Mass spectrometric analysis of the cross-link bands identified CpoB as a main constituent of the cross-linking product ( Table 1 ) . These results imply that PBP1B and CpoB physically interact in vivo . 10 . 7554/eLife . 07118 . 009Figure 3 . PBP1B interacts with CpoB in vivo . ( A ) Photo-cross-linkable amino acid ( pBpa ) substitutions at the indicated positions in His6-PBP1B , lining a cleft between the TPase and UB2H domains , formed similar UV cross-linking products in vivo . ( B ) SDS-PAGE and α-His6 immunoblot analysis revealed a cross-linked adduct ∼120 kDa in size ( MW of PBP1Bγ = 88 . 9 kDa ) . Analysis of the products by mass spectrometry revealed the presence of CpoB ( MW = 28 . 2 kDa ) cross-linked to PBP1B at each indicated position . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 00910 . 7554/eLife . 07118 . 010Table 1 . His6-PBP1B pBpa—CpoB cross-linking mass spectroscopy dataDOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 010MutantScoreCoverage†Peptides‡PSM§T118300 . 5219 . 77%712E12379 . 9911 . 41%34T75179 . 9911 . 41%34T753301 . 6419 . 77%712Addition of individual scores of the ion fragmentation spectra of the identified peptides . †Percentage of protein sequence covered by the identified peptides . ‡Number of unique peptides of the protein identified in the sample . §Number of individual spectra in which the peptides were identified . If CpoB contributes to PBP1B function , then absence of PBP1A or its required activator LpoA ( Paradis-Bleau et al . , 2010; Typas et al . , 2010 ) should result in an aggravated phenotype ( Figure 2—figure supplement 3C ) . Consistent with this expectation , a strain lacking both CpoB and LpoA showed a severe growth defect ( Figure 2F ) . To our surprise , however , a strain lacking both CpoB and PBP1A grew normally ( Figure 2—figure supplement 3A ) . We tracked down the reason for this discrepancy to a previously unknown function of LpoA . LpoA possesses two-domains , an N-terminal tetratricopeptide repeat ( TPR ) domain and a C-terminal domain ( Figure 2—figure supplement 4A ) . The C-terminal domain is sufficient to bind and activate PBP1A in vitro ( Typas et al . , 2010; Jean et al . , 2014 ) and , we show here , in vivo ( Figure 2—figure supplement 3B ) . We found that the N-terminal TPR domain encodes an additional function that can compensate for loss of CpoB . As evidence , though a ΔcpoB ΔmrcA mutant has no growth defect , a triple mutant also lacking the LpoA TPR domain , ΔcpoB ΔmrcA lpoA ( ΔTPR ) , recapitulates the growth defect of the ΔcpoB ΔlpoA mutant ( Figure 2—figure supplement 4D ) . Importantly , this CpoB compensatory function of LpoA is required only when PBP1B is the sole active bifunctional synthase , as a ΔcpoB lpoA ( ΔTPR ) mutant ( in which PBP1A is active ) grows normally ( Figure 2—figure supplement 4D ) . To further assess the importance of CpoB function for envelope integrity , we used periplasmic mCherry to fluorescently visualize envelope morphology ( Gerding et al . , 2007 ) . Both ΔmrcB and ΔcpoB strains exhibited minor OM defects , including elevated levels of OM blebbing ( Figure 2C , E ) . As with the growth phenotype , these ΔcpoB envelope defects were greatly exacerbated in the absence of LpoA , with extensive OM blebbing , periplasmic leakage and cell lysis ( Figure 2C , E and Figure 2—figure supplement 1A–C ) . All of these effects were recapitulated in a ΔcpoB ΔmrcA lpoA ( ΔTPR ) triple mutant ( Figure 2—figure supplement 4B–D ) , and ameliorated by overexpression of PBP1B ( Figure 2C–F and Figure 2—figure supplement 4B–D ) . Of note , the envelope defects of ΔcpoB ΔlpoA were distinct from those of tol-pal mutants ( Gerding et al . , 2007 ) , with more extensive lysis but less OM blebbing and periplasmic leakage ( Figure 2C–E and Figure 2—figure supplement 1C ) . Taken together , these results validate further that CpoB contributes to proper PBP1B function in vivo; indicate that this function is important for envelope integrity , and particularly important when cells must rely on PBP1B; and serendipitously reveal a partial redundancy of function between CpoB and the LpoA TPR domain . We next investigated whether CpoB , like PBP1B and Tol ( Bertsche et al . , 2006; Gerding et al . , 2007 ) , is recruited to the septum . Using a functional , endogenously expressed CpoB-mCherry fusion protein ( Figure 4—figure supplement 1A ) , we observed that CpoB localized at sites of visible envelope constriction in pre-divisional cells ( Figure 4A ) and was positioned preferentially at mid-cell with increasing prevalence as the cell cycle progressed ( Figure 4B , C ) . This pattern of localization was corroborated for the native protein , as visualized by immunolabeling cells with antibody specific to CpoB ( Figure 4—figure supplement 2 ) . Mid-cell positioning of CpoB-mCherry was coincident with envelope constriction ( Figure 4A–D ) , suggesting a role during that stage of cell division . 10 . 7554/eLife . 07118 . 011Figure 4 . CpoB localizes to mid-cell concurrent with PBP1B , TolA , and the onset of OM constriction . ( A ) Endogenously expressed CpoB-mCherry localizes to mid-cell during cell division . Left , phase contrast image; right , mCherry fluorescence; scale bar , 2 μm . ( B ) Localization of CpoB-mCherry as a function of cell cycle progression ( cell length ) , with comparison to other proteins . PBP1B , LpoB , FtsZ , PBP3 , and FtsN were visualized by immunofluorescence using specific antibodies . Cell membranes were stained with Bodipy-12 . Cell width ( diameter ) was measured from phase contrast images; constant brightness along the length of the cell indicates constant diameter , and darkening at mid-cell in longer cells indicates constriction . To generate profile maps for each protein , fluorescence intensity profiles ( integrated fluorescence as a function of cell length ) were derived for >1000 individual cell images and sorted vertically by cell length . Each horizontal line corresponds to a single cell . Y-axis scale indicates % cell-cycle progression based on cell length . Data for FtsZ , PBP3 , FtsN , and cell width have been previously published ( van der Ploeg et al . , 2013 ) and are used here for comparisons . ( C ) Average distribution of CpoB-mCherry for different cell-cycle progression age groups . Fluorescence intensity profiles were normalized by length , then averaged and normalized to the peak local brightness for each age group . ( D ) Relative timing of mid-cell recruitment for each examined protein . Initiation is based on the onset of enriched localization at mid-cell ( quantified in Figure 4—figure supplement 3A , B and van der Ploeg et al . , 2013 ) . Peak is the point in the division cycle when maximal localization is reached . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 01110 . 7554/eLife . 07118 . 012Figure 4—figure supplement 1 . Endogenously-encoded CpoB-mCherry and GFP-TolA fusion proteins are functional . ( A ) and ( B ) Quantification of growth fitness for indicated strains ( see Figure 2B , F ) . The native chromosomal copy of cpoB was replaced with cpoB-mCherry to allow endogenous expression of the mCherry fusion . A ΔcpoB ΔlpoA strain exhibits a severe growth defect , whereas a cpoB-mCherry ΔlpoA strain shows no defect , indicating that the CpoB-mCherry fusion is functional . ( B ) The native chromosomal copy of tolA was replaced with gfpmut2-tolA to allow endogenous expression of the GFP fusion . A ΔtolA strain was extremely sensitive to 0 . 5% SDS and 0 . 5 mM EDTA ( envelope stress ) , while the gfp-tolA strain showed no defect , indicating that the GFP-TolA fusion is functional . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 01210 . 7554/eLife . 07118 . 013Figure 4—figure supplement 2 . CpoB localizes to mid-cell . ( A ) WT cells immunolabeled with antibody to CpoB . ( B ) Average distribution of CpoB along the length of the cell ( see Figure 4C ) . ( C ) Localization of CpoB as a function of cell cycle progression ( see Figure 4B ) . ( D ) Antibody to CpoB is specific . Image and profile map of ΔcpoB cells immunolabeled with antibody to CpoB; no appreciable fluorescence signal is observed . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 01310 . 7554/eLife . 07118 . 014Figure 4—figure supplement 3 . Timing of localization to mid-cell for CpoB-mCherry , GFP-TolA , and PBP1B , with comparison to divisome proteins FtsZ , PBP3 , and FtsN . ( A ) Average distribution of indicated cell division proteins for different cell-cycle progression age groups ( see Figure 4C and key , bottom-right ) . Profiles of the 10 age classes for each protein have been offset equally to facilitate comparison and minimize overlap . Each age group contains 300–500 cells . Age class when increased localization at mid-cell was first observed . ( B ) Localization summary table . 1Ring fraction , the amount of mid-cell fluorescence in a rectangle of 0 . 8 μm divided by the total fluorescence in the cell at the time point corresponding to the moment . 2Initiation , the time point in the division cycle where the fluorescence at mid cell starts to increase as determined from the collective profiles in age classes of 10% ( Figure 4—figure supplement 3A ) . 3Moment , the time point in the cell cycle when mid-cell fluorescence is highest . 4Data from van der Ploeg et al . ( 2013 ) . CpoB-mCherry was used instead of immunolabeling so as to allow live-cell imaging; we have found that freely diffusing periplasmic proteins in cells grown in minimal medium do not retain their original localization following fixation , likely due to osmotic effects . **GFP-TolA was used because the quality of available TolA antiserum was not sufficient for reliable labeling . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 01410 . 7554/eLife . 07118 . 015Figure 4—figure supplement 4 . Localization of CpoB to mid-cell is dependent on divisome assembly and function . ( A ) – ( G ) Profile maps showing localization of CpoB as a function of cell cycle progression ( see Figure 4B ) , visualized by immunolabeling cells of the indicated strains with antibody specific to CpoB . ( A ) – ( E ) . For wild type and temperature-sensitive ( ts ) strains , localization at both the permissive ( 28°C , left ) and restrictive ( 42°C , right ) temperatures is shown . ( A ) Parental wild type strain ( LMC500 ) . ( B ) FtsZ ( ts ) strain LMC509 is not completely temperature sensitive; consequently , CpoB still localizes at mid-cell in shorter pre-divisional cells , but not in longer smooth filaments . ( E ) PBP3 ( ts ) strain LMC510 is already somewhat elongated at the permissive temperature because even at this temperature the PBP3 ( ts ) protein is partially unstable ( Fraipont et al . , 2011 ) . ( F ) CpoB localization over the course of FtsN depletion ( see ‘Methods’ ) . ( G ) CpoB localization following treatment with aztreonam , which inhibits PG synthesis by PBP3 . CpoB localization to mid-cell is thus dependent on FtsZ , FtsA , FtsW , PBP3 , ongoing PBP3-mediated PG synthesis , and FtsN . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 01510 . 7554/eLife . 07118 . 016Figure 4—figure supplement 5 . Localization of CpoB to mid-cell is independent of PBP1B and TolA . ( A ) and ( B ) Profile maps showing localization of CpoB as a function of cell-cycle progression ( see Figure 4B ) , visualized by immunolabeling cells of the indicated strains . Cell width ( diameter ) is based on phase contrast images; darkening indicates constriction at mid-cell . ( A ) CpoB localizes to mid-cell in the absence of PBP1B ( encoded by mrcB ) . ( B ) CpoB localizes to mid-cell in the absence of TolA . Loss of TolA causes mild cell chaining , observed here in the longer ΔtolA cells possessing multiple points of constriction ( ΔtolA , cell width plot ) . CpoB also localizes to these future division sites . Cells were grown under conditions that minimized non-specific accumulation of periplasmic proteins at the septum in ΔtolA cells ( glucose minimal medium ( GB1 ) pH 7 . 0 , 28°C; see 'Methods' ) . ( C ) Average distribution of CpoB in WT , ΔtolA and TolA-overexpressing complemented ( pTolA ) strains . Fluorescence intensity profiles were normalized by length and then averaged . In ΔtolA cells , CpoB abundance is increased , while in cells overexpressing TolA , CpoB abundance is decreased . A shorter exposure time was used to image ΔtolA cells in ( B ) to compensate for increased fluorescence; quantification in ( C ) accounts for this difference . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 016 Tol localizes at constriction sites during later stages of cell division ( Gerding et al . , 2007 ) . To better define the timing of CpoB recruitment , we compared the cell cycle dynamics of septal CpoB-mCherry localization with those of other divisome components ( Figure 4B , D ) . As cell length correlates with cell cycle progression , measuring fluorescence enrichment at mid-cell in cells sorted by length provides a quantitative metric for temporal comparison ( den Blaauwen et al . , 1999; Aarsman et al . , 2005; van der Ploeg et al . , 2013 ) . FtsZ initiates cell division and divisome assembly , with PBP3 and FtsN localizing at progressively later stages ( van der Ploeg et al . , 2013; Figure 4B , D; Figure 4—figure supplement 3 ) ; FtsN is the last essential component recruited in the divisome ( Aarsman et al . , 2005 ) . CpoB-mCherry , a functional TolA-GFP fusion ( Figure 4—figure supplement 1B ) , PBP1B , LpoB , and FtsN localized at approximately the same time ( Figure 4B , D ) . Thus , CpoB is recruited to the septum during the later stages of cell division , concurrent with Tol , PBP1B , and the onset of OM constriction . We next asked which proteins and cell division events are required for recruitment of CpoB to the septum . CpoB failed to localize in cells with temperature-sensitive ( ts ) FtsZ , FtsA , FtsW , or PBP3 under non-permissive conditions , and in cells depleted for FtsN , indicating that CpoB localization is dependent on divisome assembly ( Figure 4—figure supplement 4A–F ) . CpoB also failed to localize following inhibition of PBP3 with aztreonam ( Figure 4—figure supplement 4G ) , indicating that its localization also requires ongoing septal PG synthesis . However , CpoB localized normally in the absence of PBP1B ( Figure 4—figure supplement 5A ) . TolA dependence was more difficult to examine , since delayed OM constriction in the absence of TolA leads to the non-specific accumulation of periplasmic proteins at cell division sites ( Gerding et al . , 2007 ) . Under growth conditions that minimized this effect , CpoB continued to localize in the absence of TolA ( Figure 4—figure supplement 5B ) , suggesting that it can be recruited independently of TolA . Because a strain lacking PBP1B and TolA is only marginally viable ( Typas et al . , 2010 ) , we cannot distinguish whether CpoB localization is independent of both PBP1B and TolA , or whether each partner alone is sufficient to drive CpoB localization . CpoB localization is thus tied to cell division . Further , CpoB interacts and co-localizes with proteins involved in both OM constriction ( TolA; Krachler et al . , 2010; Gerding et al . , 2007 ) and PG septal synthesis ( PBP1B; Figure 3 ) , raising the possibility that CpoB participates in coupling these processes following recruitment to the septum during cell division . Interactions between TolA and CpoB ( Krachler et al . , 2010 ) and between PBP1B and LpoB ( Paradis-Bleau et al . , 2010; Typas et al . , 2010; Egan et al . , 2014 ) have been previously reported . We searched for associations between these complexes that might facilitate functional coordination by testing for further direct pairwise interactions between these proteins in vitro . As measured by surface plasmon resonance ( SPR ) , CpoB bound to immobilized PBP1B with an apparent dissociation constant ( KD ) of 102 ± 21 nM , but not to PBP1A or to a surface without protein ( Figure 5A , B ) , indicating a direct and specific interaction . CpoB is a trimer under the conditions used ( Krachler et al . , 2010 ) , but it is unclear whether CpoB binds as a monomer and/or trimer to PBP1B . PBP1B and TolA also interact directly , as untagged TolA was retained on Ni-NTA beads in the presence but not absence of His6-PBP1B ( Figure 5C ) . A purified soluble version of TolA lacking its transmembrane anchor ( domain I ) was not pulled down with His6-PBP1B ( Figure 5C ) , suggesting that domain I of TolA is important for the interaction . In contrast , LpoB does not bind either CpoB or TolA directly , as a His-tagged version of LpoB did not retain CpoB or TolA on Ni-NTA beads , while retaining PBP1B as expected ( Figure 5—figure supplement 1A ) . 10 . 7554/eLife . 07118 . 017Figure 5 . PBP1B interacts directly with CpoB and TolA . ( A ) CpoB interacts directly with PBP1B in vitro , assayed by SPR . Sensorgrams show binding of CpoB to a chip surface with immobilized PBP1B , but not PBP1A or an empty control surface . Concentrations of trimeric CpoB ( 0–0 . 167 µM ) are indicated . ( B ) The KD of the PBP1B-CpoB interaction was determined by non-linear regression using Sigma Plot 11 . 0 , and assuming that CpoB interacts in its trimeric form . ( C ) TolA interacts directly with PBP1B in vitro only when TolA contains its transmembrane domain I . Interaction between His6-PBP1B and either full length TolA or TolA ( sol ) , a soluble variant lacking domain I , was assessed using an in vitro cross-linking/pull-down approach . His-PBP1B specifically retained TolA , but not the soluble version lacking domain I , when pulled down by Ni-NTA beads . TolA or TolA ( sol ) alone showed no significant binding to Ni-NTA . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 01710 . 7554/eLife . 07118 . 018Figure 5—figure supplement 1 . Demonstration of ternary complexes in vitro . A ) Ternary complexes of LpoB-PBP1B-CpoB and LpoB-PBP1B-TolA detected by in vitro cross-linking/pull-down . Proteins were cross-linked and applied to Ni-NTA beads . Cross-linkage of bound proteins was cleaved and samples separated by SDS-PAGE . Gels were then stained with Coomassie blue . His-LpoB retained CpoB or TolA only in the presence of PBP1B , indicating the presence of His-LpoB-PBP1B-CpoB and His-LpoB-PBP1B-TolA complexes . ( B ) Ternary complex of PBP1B , CpoB , and TolA . Mixtures of proteins ( indicated above ) were cross-linked and complexes were resolved by SDS-PAGE . After western blot PBP1B , CpoB and TolA were detected by specific antibodies . The CpoB–TolA complex ( blue box ) and several bands containing all three proteins ( green , orange and red boxes ) were detected . The bottom part shows an experiment where PBP1B was replaced by PBP1A , showing the absence of the ternary complex bands . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 01810 . 7554/eLife . 07118 . 019Figure 5—figure supplement 2 . CpoB copy number determination . Representative western blot detection of CpoB from BW25113 whole cell lysate along with purified CpoB standards . Standards were loaded in BW25113 ΔcpoB lysate to ensure a similar transfer efficiency to the endogenous CpoB . Chemiluminescence signal was quantified using the ImageQuant LAS4000 software . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 019 To contextualize these pairwise interactions , we assessed higher order interactions . As CpoB and TolA each interact with PBP1B , we tested whether they also form a ternary complex . Indeed , chemically cross-linking a mixture of PBP1B , TolA , and CpoB identified a high-molecular weight species containing all three proteins ( Figure 5—figure supplement 1B ) . Using this technique , we did not detect analogous high molecular weight complexes of PBP1A with TolA or CpoB ( Figure 5—figure supplement 1B ) , confirming that the CpoB and TolA interactions and complex formation with PBP1B were specific . Since LpoB is required for PBP1B activation in vivo , we next tested whether CpoB and TolA interact with PBP1B in complex with LpoB . Indeed , in the presence but not absence of PBP1B , His-LpoB retained CpoB and TolA on Ni-NTA beads after chemical cross-linking , indicating that LpoB-PBP1B-CpoB and LpoB-PBP1B-TolA form ternary complexes ( Figure 5—figure supplement 1A ) . Given that all possible ternary complexes containing PBP1B can be formed , we propose that PBP1B , LpoB , CpoB , and TolA also form a quaternary complex . As CpoB can form a complex with both PBP1B-LpoB and TolA , its cellular abundance might influence which interactions occur in vivo . Using purified protein as a standard and an antibody specific to CpoB , we determined a CpoB cellular copy number of 4550 ± 540 ( n = 5 ) in growing cells ( Figure 5—figure supplement 2 ) . This is similar to an estimated protein synthesis rate for CpoB of ∼5200 molecules per generation , determined via ribosome profiling ( Li et al . , 2014 ) . Thus , the abundance of CpoB significantly exceeds the ∼520 molecules of PBP1B and ∼480 molecules of TolA synthesized per generation ( Li et al . , 2014 ) . CpoB forms trimers , which disassociate to the monomeric form when bound to TolA ( Krachler et al . , 2010 ) . It is unclear in what form ( s ) CpoB binds to PBP1B . Nevertheless , even in its trimeric form ( ∼1500 trimers ) , CpoB is in excess to both PBP1B and TolA and is therefore likely able to interact constitutively with both partners . Taken together , these data suggest that CpoB , PBP1B-LpoB , and Tol associate to form a higher-order complex , spatially linking PG synthesis and OM constriction during cell division . We tested the effects of TolA and CpoB on PBP1B GTase and TPase activities in vitro . As measured by consumption of fluorescently labeled lipid II substrate ( Figure 6A and Figure 6—figure supplement 1A ) , TolA increased the GTase rate of PBP1B 1 . 9 ± 0 . 5-fold , indicating that TolA is a novel regulator of PBP1B . Although weaker than the eightfold stimulation by LpoB ( Figure 6A; Egan et al . , 2014 ) , TolA and LpoB stimulations were additive , together yielding an 11 . 3 ± 0 . 5-fold increase in GTase activity . These results indicate that LpoB and TolA stimulate PBP1B GTase activity by compatible mechanisms . In contrast , CpoB had no effect on the GTase activity of PBP1B , alone or in combination with LpoB and/or TolA ( Figure 6A ) . On the other hand , neither CpoB nor TolA affected basal PBP1B TPase activity ( Figure 6B , orange ) , as quantified by percentage of peptides with cross-links in PG produced with radiolabelled lipid II as substrate . Note that , unlike for GTase activity , a continuous TPase assay is currently not available , and therefore we cannot rule out that one of the two proteins affects the TPase rate of PBP1B; this would in fact be expected for TolA as a consequence of GTase activation ( see below ) . 10 . 7554/eLife . 07118 . 020Figure 6 . CpoB modulates stimulation of PBP1B TPase activity by LpoB . GTase and TPase activities of PBP1B in the presence of LpoB , TolA , and/or CpoB . ( A ) GTase rate was assayed by consumption of fluorescently labelled lipid II by PBP1B in vitro . Change in GTase rate is relative to PBP1B alone and is shown as mean ± SD ( n = 3–4 ) . ( B ) TPase activity was determined by analyzing peptide cross-linkage following in vitro PG synthesis by PBP1B with radiolabelled lipid II substrate . TPase activity is shown as percentage of peptides present in cross-links in PG produced by PBP1B ( mean ± SD ) . TPase assays were performed at 150 mM NaCl ( n = 4–7 ) and at an increased salt concentration of 215 mM NaCl ( n = 2 , shown as the mean ± maximum and minimum values ) , at which the negative impact of CpoB on PBP1B TPase stimulation by LpoB is exacerbated . Statistical significance/difference was determined by Mann–Whitney U test , with p values of 0 . 00058 for both data sets . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 02010 . 7554/eLife . 07118 . 021Figure 6—figure supplement 1 . PBP1B GTase and TPase activity assays . ( A ) PBP1B GTase activity assay . Representative continuous fluorescence assay graphs for GTase data shown in Figure 6 . PBP1B concentrations were 0 . 5 µM ( graphs 2 and 4 ) or 1 µM ( graphs 1 and 3 ) , the differences in basal PBP1B activity are also due to different detergent concentration in the reactions . The reactions shown in the bottom graph were performed at a lower temperature of 25°C to slow the reaction and allow the measurement of initial rates . Proteins present in each reaction are indicated next to their corresponding curves in the same color . Each point is the mean ± SD of three or four independent experiments . ( B ) PBP1B TPase activity assay . Representative examples of HPLC chromatograms for TPase data shown in Figure 6B . PBP1B was incubated with radioactive lipid II and 150 mM NaCl ( standard condition , top chromatograms ) or 215 mM NaCl ( bottom chromatograms ) in the presence of the proteins indicated above the corresponding chromatogram . The resulting PG was digested with muramidase ( cellosyl ) to yield muropeptides , which were reduced with sodium borohydride and separated by HPLC . The percentage of peptides in cross-links was calculated as 100% minus the monomeric ( non-cross-linked ) disaccharide pentapeptide ( peak 1 ) . Peak 2 corresponds to bis-disaccharide tetrapentapeptide , and peak 3 to tris-disaccharide tetrapentapeptide . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 021 The GTase and TPase activities of PBP1B are coupled , as is generally true for bifunctional synthases ( Bertsche et al . , 2005; Born et al . , 2006; Lupoli et al . , 2014 ) . LpoB activates both the PBP1B GTase domain , increasing reaction rate , and the TPase domain , thereby producing hyper-cross-linked PG ( Figure 6B , blue; Typas et al . , 2010; Egan et al . , 2014 ) . Interestingly , although CpoB did not affect the cross-linking activity of PBP1B on its own , it partially prevented the formation of hyper-cross-linked PG in the presence of LpoB ( Figure 6B , blue ) . As CpoB affected neither LpoB binding to PBP1B ( Figure 5—figure supplement 1A ) nor GTase stimulation of PBP1B ( Figure 6A ) , CpoB must directly interfere with LpoB activation of the PBP1B TPase . Strikingly , TolA alleviated the inhibitory effect of CpoB , restoring the synthesis of highly cross-linked PG in the presence of LpoB ( Figure 6B , cyan ) . The effect of CpoB on TPase activity was further enhanced at a moderately increased salt concentration ( 215 mM NaCl ) , suggesting that its regulatory effect on PBP1B may be intensified under stress . In this condition , LpoB increased cross-linking from 50% to 60% . The addition of CpoB completely prevented this stimulation , whereas TolA again alleviated the CpoB effect ( Figure 6B , right graph ) . Though the magnitudes of these effects are moderate , they are likely significant in vivo given the essentiality of LpoB for PBP1B function under all conditions tested ( Paradis-Bleau et al . , 2010; Typas et al . , 2010 ) , including different salt concentrations . In summary , interaction of the PBP1B-LpoB synthase complex with TolA further stimulates PBP1B GTase activity , and interactions with CpoB and TolA modulate PBP1B TPase activity via reciprocal effects . To establish in vivo relevance and to understand how complex formation and PBP1B activity are regulated in vivo , we systematically characterized interactions between CpoB , TolA , PBP1B , and LpoB via DTSSP cross-linking and co-immunoprecipitation ( Figure 7A–C ) . In addition to the previously reported PBP1B-LpoB pair ( Typas et al . , 2010 ) , we observed specific pairwise interactions between CpoB , TolA , and PBP1B , validating that these interactions also occur in vivo . Interestingly , CpoB exhibited minimal or no cross-linking with LpoB in the presence of TolA , but a dramatic increase in cross-linking in its absence ( Figure 7A ) . Thus , TolA both prevents CpoB from associating with LpoB ( Figure 7A ) and from interfering with PBP1B TPase stimulation by LpoB ( Figure 6B ) , suggesting that these effects are linked . Since an interaction between CpoB and LpoB was not detected in vitro ( Figure 5—figure supplement 1A ) , their association in vivo likely occurs in the context of the CpoB-PBP1B-LpoB ternary complex ( Figure 8A ) . This possibility could not be tested directly because a strain lacking both PBP1B and TolA is only marginally viable ( Typas et al . , 2010 ) . TolA did not prevent PBP1B from interacting with CpoB ( Figure 7A ) ; thus , interaction with TolA may instead alter the conformation of CpoB in a manner that both prevents it from interacting with LpoB ( Figure 7A ) and , concurrently , from modulating PBP1B TPase activation by LpoB ( Figure 6B ) . 10 . 7554/eLife . 07118 . 022Figure 7 . Regulatory interactions between CpoB , TolA , and PBP1B-LpoB respond to Tol apparatus assembly and energy state . ( A ) and ( B ) . Interactions between CpoB , TolA , PBP1B , and LpoB were characterized in vivo by cross-linking and co-immunoprecipitation for wild type ( WT ) and indicated mutant strains . Antibodies specific for CpoB , TolA , and PBP1B were used to immunoprecipitate their antigens from Triton X-100-solublized E . coli membrane extracts derived from cells treated with DTSSP cross-linker . Interacting proteins were detected in the immuno-precipitates by western blot using specific antibodies . IP and input panels are from different blot exposures ( see ‘Methods’ ) . ( C ) Schematic of observed in vivo interactions . Black bars indicate interactions observed in WT; red bars indicate novel interactions in mutant strains . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 02210 . 7554/eLife . 07118 . 023Figure 8 . Model for physical and functional coordination of the PBPB-LpoB PG synthesis and Tol OM constriction machines by CpoB . ( A ) Data-driven docking of the PBP1B/LpoB/CpoB complex calculated with HADDOCK/CNS protocols ( de Vries et al . , 2010 ) and integrating experimental data . The lowest energy structure obtained is shown with PBP1B , LpoB , and CpoB colored in wheat , pale green , and blue , respectively . LpoB binds the PBP1B UB2H domain , while CpoB binds in an adjacent cleft between the UB2H and TPase domains ( Figure 3 ) . We propose that CpoB can contact LpoB in this conformation , preventing hyper-activation of the PBP1B TPase by LpoB . TolA counteracts this CpoB effect , and increases both PBP1B TPase activity ( by restoring stimulation by LpoB ) and GTase activity ( directly ) . ( B ) Schematic of PBP1B regulation by LpoB , CpoB , and TolA . TP: transpeptidase domain; GT: glycosyltransferase domain . Green circles indicate activity stimulation ( increase in glycan synthesis rate for GTase; formation of more highly cross-linked product for TPase ) . Top left: LpoB binding stimulates both PBP1B GTase and TPase activities; the latter produces highly cross-linked PG . Top right: CpoB modulates TPase stimulation without interfering with GTase stimulation . Bottom left and right: When energized , TolA reverses the effect of CpoB , restoring production of highly cross-linked PG . Regulation by TolA is thus dependent on TolQR–TolA PMF utilization , tying cycles of PBP1B TPase activation to cycles of Tol function . DOI: http://dx . doi . org/10 . 7554/eLife . 07118 . 023 PBP1B-CpoB-TolA complex formation was highly responsive to Tol complex status . In the absence of TolQ , abrogating formation of the TolQR–TolA complex , PBP1B-TolA and PBP1B-CpoB interactions were both substantially diminished . In contrast , CpoB–LpoB interaction and PBP1B-LpoB complex formation increased significantly ( Figure 7A ) . Together , these data suggested that CpoB regulates PBP1B-LpoB in response to Tol assembly state , energy state , or both . To distinguish between these possibilities , we compared the effects of a tolR deletion , which prevents TolQR–TolA complex assembly , to those of a tolR ( D23R ) point mutation that allows complex assembly but prevents TolQR–TolA from utilizing PMF ( Cascales et al . , 2001 ) . In the tolR ( D23R ) mutant , interaction between TolA and PBP1B was maintained ( Figure 7B ) , suggesting that physical association between PBP1B-LpoB and Tol depends on assembly of the TolQR–TolA complex in vivo but not on its energy state . In contrast , interaction between CpoB and LpoB was highly elevated ( Figure 7B ) . This strongly suggests that TolQR–TolA must be energized for TolA to prevent CpoB both from interacting with LpoB and , in conjunction , from inhibiting PBP1B TPase stimulation by LpoB . This implies that cycles of TolQR–TolA PMF utilization and energization can in turn modulate PBP1B TPase activity , allowing dynamic regulation of PBP1B activity in response to Tol energy state . Taken together , interactions between PBP1B-LpoB , TolA , and CpoB promote physical and spatial coordination of PBP1B-LpoB and the Tol apparatus , and provide a mechanism for direct regulation of septal PG synthesis in response to Tol function . In the absence of input from CpoB , LpoB binding to PBP1B stimulates both its GTase and TPase activities . TPase activation results in hyper-cross-linked PG ( Figure 8B , top left ) , containing ∼70% cross-linked peptides in vitro . CpoB binding interferes only with TPase stimulation ( Figure 8B , top right ) so that ∼60% of the peptides are cross-linked . Importantly , this level of cross-linking is similar to that observed in mini-cells , which are formed by abnormal polar divisions in vivo and have an exclusively polar PG ( Obermann and Höltje , 1994 ) . Therefore , CpoB reduces cross-linkage to a more physiological level . This may be the default cellular state , as CpoB is present in ∼10-fold excess over its PBP1B and TolA binding partners ( Figure 5—figure supplement 2; Li et al . , 2014 ) . CpoB is the first example of an endogenous regulator that can control the TPase activity of a bifunctional PG synthase independently of its GTase activity . It thus mimics the activity of penicillin G , which blocks PBP1B TPase activity without interfering with GTase stimulation by LpoB ( Lupoli et al . , 2014 ) . TolA further modulates PBP1B function by reversing CpoB inhibition , in a manner that is dependent on the state of the Tol apparatus ( Figure 8B , bottom panels ) . Fuelled by PMF-derived energy , TolQR–TolA cycles between at least two alternative states; we propose that each state has distinct functional consequences for PG synthesis . During the non-energized phase of Tol function , CpoB inhibits LpoB-mediated PBP1B TPase hyper-activation ( Figure 8B , bottom right ) . During the energized phase , however , TolA adopts a new conformation that allows it to alleviate CpoB inhibition , thereby restoring maximal TPase stimulation by LpoB ( Figure 8B , bottom left ) . Thus , CpoB provides the means to selectively regulate PBP1B catalytic activities , and TolA modulates this effect , linking cycles of PBP1B activity state and the resulting PG cross-linking with cycles of TolQR–TolA energy state while Tol promotes OM invagination . TolA also directly stimulates PBP1B GTase activity , independently of and additively with LpoB . We do not yet know whether this stimulation occurs during one or both phases of Tol function . Importantly , recruitment and assembly of the TolQR–TolA IM sub-complex is required for regulation of PBP1B by TolA in vivo: in the absence of TolQ or TolR , interaction between TolA and PBP1B is significantly decreased , whereas interaction between PBP1B and LpoB is enhanced . TolQ and TolA localize to the septum independently of each other ( Gerding et al . , 2007 ) , and their recruitment may incorporate different regulatory cues . We propose that selectively tuning the GTase and TPase activities of PBP1B offers functional flexibility that is particularly important during constrictive PG synthesis . PBP1B-LpoB has the capacity to produce highly cross-linked PG , which may be required at certain stages of new pole synthesis , but the TPase activity of PBP1B is normally down-regulated by CpoB . Up-regulation is allowed only during energized phases of Tol function . Thus , for PBP1B-LpoB at the septal leading edge , increased cross-linking ( TPase ) activity is permitted only when the OM is brought into close proximity—as facilitated by cycles of Tol function and energy utilization—allowing energized TolQR–TolA to counteract CpoB . The IM-bound septal leading edge can thereby sense and respond the status of the trailing edge of the constricting envelope ( i . e . the OM ) , and activity cycles of the two machines are synchronized , helping to ensure a constant distance between the OM and IM . Restricting TPase activity based on Tol status may also promote better coordination of septal PG synthesis and septal cleavage by OM-controlled amidases ( Uehara et al . , 2010; Yang et al . , 2011 ) . By making PBP1B activity responsive to the presence , assembly and energy state of Tol , CpoB enables potential feedback regulation of PG synthesis based on the status of OM invagination . The importance of such coordination can be seen when it is broken , as demonstrated by the phenotypes of a strain lacking CpoB . Multiple antibiotic sensitivities , envelope defects , and genetic interactions paralleling those of a strain lacking PBP1B suggested that CpoB contributes to proper PBP1B function in vivo , an interpretation bolstered by the facts that PBP1B overexpression alleviated these ΔcpoB phenotypes and a strain lacking both proteins phenocopied a PBP1B mutant . Envelope defects in the absence of CpoB were relatively mild , likely because redundant mechanisms compensate for loss of coordination by CpoB ( e . g . , LpoA TPR domain function , discussed below ) . Yet defects were greatly exacerbated under osmotic stress ( Figure 2—figure supplement 1A ) or in the absence of LpoA function ( Figure 2B–F and Figure 2—figure supplement 4B–D ) , indicating that coordination by CpoB becomes more crucial under stress and/or when cells must rely more heavily on PBP1B function . Apparent loss of PBP1B function in the absence of CpoB may result from loss of proper coordination with Tol that is required for efficient and productive utilization of PBP1B activity during septation . In addition , loss of TPase modulation may itself be deleterious , as uncontrolled PG synthase activity can be as detrimental to the cell as no activity ( Cho et al . , 2014 ) . Synthetic and hydrolytic enzymatic reactions have to be carefully balanced for seamless PG growth , a principle exploited by β-lactam antibiotics , which lyse cells predominantly via uncontrolled PG hydrolase activity ( Kohlrausch and Höltje , 1991 ) . In this scenario , overexpression of PBP1B may alleviate ΔcpoB phenotypes by ensuring that there is enough unbound ( hypoactive ) PBP1B in the cell to counter-balance hyperactive PBP1B-LpoB in septal PG synthesis . Coordination of the PBP1B and Tol machines by CpoB may also be important for proper Tol function . Defects in OM integrity in the absence of PBP1B or CpoB are consistent with such a role . The molecular mechanism by which Tol facilitates OM constriction remains poorly defined ( Gerding et al . , 2007; Egan and Vollmer , 2013 ) , and the energized system is currently not amenable to in vitro dissection . It is thus more difficult to determine specific regulatory effects of CpoB and PBP1B on Tol activity . However , coordination likely promotes synchronous OM constriction during cell division , and reciprocal regulation of Tol activity would allow bidirectional coordination of function . Since PBP1B-LpoB can partially compensate for loss of Tol ( Typas et al . , 2010 ) , it is possible that the two complexes contribute to OM constriction cooperatively . Finally , it is possible that both Tol and CpoB play additional roles in PG synthesis and/or other envelope processes , which may be elucidated as we continue to improve our understanding of Tol function in vivo . PBP1B and PBP1A retain some core functional redundancy , which allows cell survival when only one is present ( Yousif et al . , 1985 ) . This partial redundancy adds robustness to cell wall biosynthesis . However , PBP1B and PBP1A also have distinct roles . PBP1B is adapted to play a key role in cell division in E . coli , as illustrated by phenotypic , localization , and physical interaction data ( Bertsche et al . , 2006; Muller et al . , 2007; Typas et al . , 2010 ) , and by the fact that cells without PBP1B lyse from the mid-cell at a higher frequency ( Garcia del Portillo and de Pedro , 1990 ) . Here , we provide additional evidence for the specialized role of PBP1B-LpoB complex in cell division , as it interacts and physically coordinates its function with the Tol apparatus ( Gerding et al . , 2007 ) . Physical and regulatory interactions thus link PBP1B to septal PG synthesis ( PBP3 , FtsN , FtsW; Bertsche et al . , 2006; Muller et al . , 2007; Fraipont et al . , 2011 ) , PG hydrolysis ( MltA; Vollmer et al . , 1999 ) , and now OM constriction ( Tol and CpoB ) , integrating division processes that span the envelope . Multiple divisome-associated regulators ( FtsN , LpoB , CpoB , TolA ) either stimulate or inhibit PBP1B activity . Regulation by other factors may be synergistic or antagonistic with regulation by CpoB and TolA , or alternatively may be employed at different stages of septation , helping PBP1B to further address unique requirements of constrictive PG synthesis . PBPs and other PG-related enzymes ( e . g . , amidases; Uehara et al . , 2009 ) are often characterized by modular architecture , combining broadly conserved catalytic domains with evolutionarily confined non-catalytic domains . We have proposed that these non-catalytic domains provide specificity and diversification to the enzymes ( Typas et al . , 2012 ) . In the case of PBP1B , the non-catalytic UB2H domain shares an extensive binding interface with LpoB ( Egan et al . , 2014 ) , and as we show here is also involved in binding CpoB . Interestingly , LpoB and CpoB approach UB2H from different sides ( Figure 8A ) , consistent with their independent binding to PBP1B . CpoB and LpoB come close in this conformation , and this proximity could facilitate the LpoB–CpoB interaction , which is stabilized in the non-energized state of the Tol system . As the UB2H domain and LpoB are present only in enterobacteria and Vibrionaceae ( Typas et al . , 2010; Dörr et al . , 2014 ) , it remains to be determined whether and how the more conserved Tol system is integrated with septal PG synthesis in other bacteria . Interestingly , CpoB is not conserved in all bacteria with a Tol system , or even in some that contain LpoB ( e . g . , Pasteurellaceae ) . We postulate that other niche-specific factors will tightly coordinate PG septal synthesis and constriction of the different envelope layers in microbes that lack CpoB and/or LpoB . In this regard , our finding that LpoA has an additional function beyond activating PBP1A , one that is redundant with ( or otherwise exacerbates the need for ) CpoB and is independent of PBP1A , is particularly interesting . First , it extends for the first time the principle of modularity to PG enzyme regulators , as LpoA has two physically separated domains: one for binding and activating PBP1A ( C-terminal domain ) and one for its CpoB-related function ( N-terminal TPR domain ) . Second , as both CpoB and LpoA possess TPR domains , which are generally involved in protein–protein interactions , it raises the question of whether both their TPR domains are involved in additional binding and/or regulation with other PG-related enzymes . Such function ( s ) could be independent of PBP1A/B binding and regulation . An alternative scenario , which would explain the CpoB and LpoA redundancy , is that LpoA can directly link to the Tol system . In contrast to CpoB , we could not detect a direct interaction between LpoA and soluble TolA in vitro . However , the LpoA connection to TolA may be more complex ( e . g . , dependent on conformational state of TolA ) , or LpoA may achieve a CpoB-like function via connections to other Tol components . Such a connection would be consistent with the ability of PBP1A to partially substitute for PBP1B , and with the general role of the Tol system in envelope integrity . As all these components ( PBP1B/A , LpoB/A , CpoB , and Tol ) localize independently to division sites , multiple redundant or alternative paths may exist for coordinating PG synthesis and OM constriction , increasing the robustness of the process to function under different conditions and across different organisms . Physical and functional coordination of PBP1B with Tol , facilitated by CpoB , reveals an important additional layer in the regulatory circuitry that controls PBP1B function , and paints an increasingly intricate picture of PBP1B as a regulatory scaffold that is uniquely adapted for the coordination of cell division process in E . coli . This is the first mechanistic view of how different envelope layers talk to each other as they grow and constrict during cell division in Gram-negative bacteria . Importantly , the process retains a high degree of modularity , which would allow for individual components to be replaced across evolution , and presumably to some degree in E . coli itself . Continued dissection of the dynamic , multi-enzyme , membrane-spanning machines that mediate coordinated cell division processes—including investigation of which interactions occur simultaneously , when and where in the cell they occur , and what additional regulatory relationships they confer—will provide greater understanding of the molecular mechanisms that govern interconnected divisome functions and their synchronization . Looking across different organisms will also provide a better overview of which interconnections are broadly required and by what different means they can be achieved . Considering the many facets of cell envelope composition ( Radolf et al . , 2012 ) and cell division ( Leisch et al . , 2012 ) , these questions are bound to have different answers in different bacteria . Nevertheless , the geometric and topological changes inherent during cell division likely generally necessitate a high degree of feedback between the different envelope layers through physical connections that modulate biochemical activities ( Weiss , 2015 ) . [14C]GlcNAc-labelled lipid II and dansylated lipid II were prepared as published ( Breukink et al . , 2003; Bertsche et al . , 2005 ) . The following proteins were prepared as previously described: PBP1B ( Bertsche et al . , 2006 ) , His-LpoB ( sol ) , and LpoB ( sol ) ( Egan et al . , 2014 ) . Antisera against PBP1B , LpoB , CpoB , TolA , and TolB ( rabbit ) were obtained from Eurogentec ( Liege , Belgium ) and purified over an antigen column as described ( Bertsche et al . , 2006 ) . Cellosyl was provided by Hoechst AG ( Frankfurt , Germany ) . VIM-4 was a kind gift from Adeline Derouaux . BL21 ( DE3 ) strains harbouring plasmids pET28-His6-CpoB ( for the purification of CpoB; signal sequence replaced by a cleavable oligohistidine tag ) , pET28-TolA-His6 ( for the purification of full length TolA with a C-terminal oligohistidine tag ) , pET28-His6-TolA ( for the purification of full length TolA with a cleavable N-terminal oligohistidine tag ) or pET28-His6-TolA ( sol ) ( for purification of TolA ( 72-421 ) ) were grown in 3 L of LB medium with appropriate supplements at 30°C to an OD578 of 0 . 5–0 . 6 . Recombinant genes were overexpressed by adding 1 mM IPTG to the cell culture followed by a further incubation for 3 hr at 30°C . Cells were harvested by centrifugation ( 10 , 000×g , 15 min , 4°C ) and the pellet was resuspended in buffer I ( 25 mM Tris/HCl , 10 mM MgCl2 , 500 mM NaCl , 20 mM imidazole , 10% glycerol , pH 7 . 5 ) . A small amount of DNase , protease inhibitor cocktail ( Sigma-Aldrich , St . Louis , MO; 1/1000 dilution ) , and 100 μM phenylmethylsulfonylfluoride ( PMSF ) were added before cells were disrupted by sonication ( Branson digital ) . The lysate was centrifuged ( 130 , 000×g , 1 hr , 4°C ) . At this point , purification procedures of full length TolA constructs and of CpoB and TolA ( sol ) differ . For CpoB and TolA ( sol ) , the supernatant was applied to a 5-mL HisTrap HP column ( GE healthcare Bio-Sciences , Piscataway , NJ ) , attached to an ÄKTA Prime+ ( GE Healthcare Bio-Sciences ) at 1 mL/min . The column was washed with 4 volumes buffer I before step-wise elution of bound proteins with buffer II ( 25 mM Tris/HCl , 10 mM MgCl2 , 500 mM NaCl , 400 mM imidazole , 10% glycerol , pH 7 . 5 ) . To remove the oligohistidine tag from His-CpoB and His-TolA ( sol ) , 50 U/mL of restriction grade thrombin ( Merck Millipore , Darmstadt , Germany ) was added , and the protein was then dialyzed against 2 L of 25 mM Tris/HCl , 10 mM MgCl2 , 500 mM NaCl , 10% glycerol , pH 7 . 5 for 18 hr at 4°C . Proteins samples were then concentrated to 4–5 mL using a VivaSpin-6 column ( MW cut-off 6000 Da ) and applied to a Superdex200 HiLoad 16/600 column at 0 . 8 mL/min for size exclusion chromatography in 25 mM Tris/HCl , 1 M NaCl , 10% glycerol , pH 7 . 5 . Finally , proteins were dialyzed against storage buffer ( 25 mM Tris/HCl , 500 mM NaCl , 10% glycerol , pH 7 . 5 ) . For full length TolA or TolA-His , the membrane pellet resulting from the above ultracentrifugation was resuspended in extraction buffer ( 25 mM Tris/HCl , 10 mM MgCl2 , 1 M NaCl , 2% Triton X-100 , 10% glycerol , pH 7 . 5 ) and incubated overnight with mixing at 4°C . Samples were centrifuged ( 130 , 000×g , 1 hr , 4°C ) and the supernatant applied to a 5-mL HisTrap HP column ( GE Healthcare Bio-Sciences ) attached to an ÄKTA Prime+ ( GE Healthcare Bio-Sciences ) at 1 mL/min . The column was washed with 4 volumes extraction buffer , followed by 4 volumes of wash buffer I ( 25 mM Tris/HCl , 10 mM MgCl2 , 1 M NaCl , 20 mM imidazole , 2% Triton X-100 , 10% glycerol , pH 7 . 5 ) , followed by a final wash with 4 volumes of wash buffer II ( as wash buffer I , with 40 mM imidazole and 0 . 2% Triton X-100 ) . Bound protein was eluted step-wise with elution buffer ( 25 mM Tris/HCl , 10 mM MgCl2 , 500 mM NaCl , 400 mM imidazole , 0 . 2% Triton X-100 , 10% glycerol , pH 7 . 5 ) . At this point , TolA-His was dialyzed into storage buffer ( 25 mM Tris/HCl , 500 mM NaCl , 0 . 2% Triton X-100 , 10% glycerol , pH 7 . 5 ) . To remove the oligohistidine tag from His-TolA , 50 U/mL of restriction grade thrombin ( Merck Millipore ) was added to the protein , which was then dialyzed against 2 L of 25 mM HEPES/NaOH , 10 mM MgCl2 , 50 mM NaCl , 10% glycerol , pH 6 . 0 for 24 hr at 4°C . Sample was then applied to a 5-mL HiTrap HP SP column attached to an ÄKTA Prime+ ( GE Healthcare Bio-Sciences ) for ion exchange chromatography , at a flow rate of 0 . 5 mL/min in buffer A ( 25 mM HEPES/NaOH , 10 mM MgCl2 , 50 mM NaCl , 0 . 2% Triton X-100 , 10% glycerol , pH 6 . 0 ) . The column was washed with 6 volumes buffer A before stepwise elution of bound TolA with buffer B ( as A , with 500 mM NaCl ) . TolA sample was then dialyzed into storage buffer ( 25 mM HEPES/NaOH , 10 mM MgCl2 , 100 mM NaCl , 0 . 2% Triton X-100 , 10% glycerol , pH 7 . 5 ) . Recombinant PBP1B ( pDML924; the functional , short PBP1Bγ version , starting from amino acid 46 ) was purified as described previously ( Bertsche et al . , 2005 ) . Versions of proteins retaining their His-tags were purified as above , omitting the addition of thrombin . Bacterial strains and plasmids used in this work are listed in Supplementary File 1A . Primers used in this work are listed in Supplementary File 1B . For in vivo assays , cells were grown aerobically at 30°C or 37°C in Lennox Luria–Bertani ( LB ) medium ( 10 g/L tryptone , 5 g/L yeast extract , 5 g/L NaCl ) ( Fisher Scientific , Houston , TX ) unless otherwise indicated . Where appropriate , antibiotics or inducers were added: ampicillin ( 100 μg/mL ) , chloramphenicol ( 10–34 μg/mL ) , kanamycin ( 30 μg/mL ) , arabinose ( 0 . 2–1% , wt/vol ) , IPTG ( 1 mM ) . For protein production , cells were grown aerobically at 30°C or 37°C in Miller LB medium ( 10 g/L tryptone , 5 g/L yeast extract , 10 g/L NaCl ) . To generate and combine E . coli gene deletions , kan-marked alleles from the Keio E . coli single-gene knockout library ( Baba et al . , 2006 ) were transferred into relevant background strains using P1 phage transduction ( Thomason et al . , 2007 ) . The Keio pKD13-derived kan cassette is flanked by FRT sites , allowing removal of the kan marker via expression of FLP recombinase to generate unmarked ( kanamycin-sensitive ) deletions with a FRT-site scar sequence ( Datsenko and Wanner , 2000; Baba et al . , 2006 ) . Other chromosomal mutations and gene fusions ( lpoA ( Δ58-252 ) , tolQ ( D23R ) , cpoB-mCherry , and gfpmut2-tolA ) were generated via lambda Red recombinase-mediated oligonucleotide and/or PCR recombineering ( Thomason et al . , 2014 ) using recombineering plasmid pSIM19 ( Datta et al . , 2006 ) . Desired mutations were confirmed by PCR and sequencing . To generate the lpoA ( Δ58-252 ) strain , a sacB-kan cassette from pIB279 ( Blomfield et al . , 1991 ) was amplified using primers ANG014 x ANG015 and inserted into lpoA to generate precursor strain CAG70134 ( BW25113 lpoA ( Δ1-255:: ( sacB-kan ) ) . A PCR product containing the lpoA ( Δ58-252 ) allele was then generated via overlap extension ( ‘stitching’ ) PCR ( Heckman and Pease , 2007 ) , using PCR primers ANG235 x ANG239 and ANG241 x ANG242 with E . coli genomic DNA as template for the first round of PCR and ANG235 x ANG242 with first-round-PCR-products as template for the second round of PCR . The PCR product was then used to replace the sacB-kan cassette via counterselection on no-salt LB plates ( 10 g/L tryptone , 5 g/L yeast extract , 18 g/L agar ) containing 7% ( wt/vol ) sucrose . To generate the tolQ ( D23R ) strain , a sacB-cat cassette from plasmid pDS132 ( Philippe et al . , 2004 ) was amplified using primers ANG478 x ANG479 and inserted into tolQ to generate strain CAG70764 ( BW25113 tolR ( D23:: ( sacB-cat ) ) . The tolQ ( D23R ) mutation was then introduced by using oligonucleotide ANG484 to replace the sacB-cat cassette via sucrose counterselection . To generate the cpoB-mCherry strain , sacB-cat was inserted after cpoB using primers ANG143 x ANG144 and replaced with a GSGSGSGS linker followed by mCherry using primers ANG177 x ANG178 with pMG36 ( encodes pal-mCherry ) ( Gerding et al . , 2007 ) as template . To generate the gfpmut2-tolA strain , sacB-cat was inserted before tolA using primers ANG263 x ANG264 and replaced with gfpmut2 followed by an ATGTRT linker using primers ANG347 x ANG349 with pBAD24-gfp ( Nilsen et al . , 2004 ) ( encodes gfpmut2 ) as template . To transfer chromosomal mutations and gene fusions to other strain backgrounds , the sacB-cat or sacB-kan cassette was first transferred to the intended recipient strain by P1 phage transduction ( Thomason et al . , 2007 ) . The unmarked mutant allele was then transferred into the sacB-cat or sacB-kan-bearing recipient strain by P1 phage transduction with selection on no-salt LB plates containing 7% ( wt/vol ) sucrose and 10 mM potassium citrate . ( Potassium citrate is substituted for sodium citrate because the addition of sodium prevents sacB-mediated sucrose sensitivity . ) pBAD33-PBP1B was constructed by cloning mrcB ( encodes PBP1B ) along with 20 bp of upstream sequence so as to include the native ribosome binding site ( PCR primers: ANG093 x ANG094 ) into the SacI/XbaI restriction sites of pBAD33 ( Guzman et al . , 1995 ) . Similarly , pBAD33-CpoB was constructed by cloning cpoB with 25 bp upstream sequence into SacI/SphI pBAD33 ( PCR primers: ANG005 x ANG006 ) . Similarly , pTolA ( pBAD33-TolA ) was constructed by cloning tolA with 20 bp upstream sequence into SacI/XbaI pBAD33 ( PCR primers: ANG147 x ANG148 ) . For pET28-His6-CpoB , cpoB was cloned into NdeI/SacI pET28a ( EMD Millipore , Billerica , MA ) ( PCR primers: ANG182 x ANG184 ) . For pET28-His6-TolA , tolA was cloned into NdeI/SacI pET28a , replacing the native GTG tolA start codon with the NdeI site ATG . For pET28-His6-TolA ( sol ) , tolA ( 74-421 ) was cloned into NdeI/SacI pET28a . For pET28-TolA-His6 , tolA ( without stop codon ) was cloned into NcoI/XhoI . The forward primer used to amplify tolA contained a BsaI restriction site , GGTCTCTCATG , to allow cloning into the pET28 NcoI site while preserving the native tolA sequence after the start codon . Amber ( TAG ) mutations were introduced in pDML924 using the QuickChange PCR site-directed mutagenesis kit and protocol ( Stratagene/Agilent Technologies , Santa Clara , CA ) . PCR reactions were treated with DpnI ( Fermentas/Thermo Fisher Scientific , Waltham , MA ) to remove parent plasmid and then transformed into E . coli DH5α . Candidate plasmid derivatives containing amber mutations were purified from isolated transformants and confirmed by sequencing . E . coli gene deletion alleles were derived from the Keio collection ( Baba et al . , 2006 ) . Strains for overexpressing PBP1B carried arabinose-inducible plasmid pBAD33-PBP1B . Strains for expressing YbgF in trans ( for complementation of ΔcpoB ) carried arabinose-inducible plasmid pBAD33-CpoB . E . coli wild type and mutant strains were arrayed and grown in a 1536-colony format ( n ≥ 96 for each strain ) on LB Lennox plates ( 20 g/L agar ) at 37°C , then replica-pinned robotically ( Rotor HDA , Singer Instruments , Roadwater , UK ) to LB Lennox and/or indicated condition plates . Assay plates were incubated for 8–14 hr at 37°C and then imaged . Colony sizes were determined using in-house software . The CPRG assay was performed as described ( Paradis-Bleau et al . , 2014 ) with slight modifications . Instead of normal rectangular agar plates in which colonies are arrayed next to each other and color can diffuse , we used 384-well plates , filled with LB-agar-CPRG by a liquid handling robot ( Biomek FX; Beckman Coulter , Brea , CA ) . Mutants carrying pCB112 , a mobile plasmid encoding lacZ under control of the lactose promoter ( Plac ) , were arrayed robotically on each well ( ROTOR , Singer Instruments ) . Color development ( CPR-red; 570 nm ) was monitored for 48 hr ( every 30 min ) in a multi-well plate reader ( Tecan M1000 Pro with a stacker ) at room temperature . Absorption wavelength ( not to overlap with colony development ) , agar volume per well and a number of other technical parameters were optimized for the assay ( George Kritikos and Athanasios Typas; unpublished data ) . The accumulation rate of CPR color was calculated after fitting a linear curve on an absorption-time plot . Culture supernatants were collected and filtered using a 0 . 2 μm-pore SFCA syringe filter ( Fisher Scientific ) , then TCA-precipitated and prepared for analysis by SDS-PAGE and western blot as described ( Wagner et al . , 2009 ) . Amber mutations were introduced at indicated sites via site-directed mutagenesis of plasmid pDML924 , which encodes His6-tagged PBP1B ( Terrak et al . , 1999 ) . E . coli BL21 ( DE3 ) cells were co-transformed with pSup-BpaRS-6TRN ( Ryu and Schultz , 2006 ) and either pDML924 or amber mutant derivatives . Cells were grown to an OD600 of 0 . 5–0 . 6; protein production was then induced with 10 μM IPTG , and 1 mM freshly-prepared p-benzoyl-L-phenylalanine ( Bachem ) dissolved in 1 M NaOH was added . After 2 hr of protein production , cells were harvested by centrifugation , washed with PBS and resuspended in 3 mL PBS , then transferred to a petri dish and exposed to UV light ( 365 nm ) for 1 . 5 min ( photoMax Housing 200W , model 60100 , 30 cm distance to sample; Oriel Instruments ) . Cells were cooled on ice during UV illumination . Cell samples were separated by SDS-PAGE ( 8% ) , transferred to a nitrocellulose membrane ( Bio-Rad Laboratories , Hercules , CA ) , and analyzed by western blot using monoclonal anti-polyhistidine peroxidase-conjugated antibody ( 1:4000 dilution; Sigma-Aldrich ) . Excised protein bands were reduced with DTT , alkylated with iodoacetamide and in-gel digested with trypsin ( Shevchenko et al . , 2006 ) . Nanoflow liquid chromatography coupled to mass spectrometry was performed on an Agilent 1200 nanoflow system ( Agilent Technologies ) connected to a MS LTQ-Orbitrap XL ( Thermo Fisher Scientific ) . The samples were trapped on a 20 mm ReproSil-Pur C18-AQ ( Dr . Maisch GmbH , Ammerbuch , Germany ) trapping column ( packed in-house , i . d . , 100 μm; resin , 5 μm ) with a flow-rate of 5 μL/min . Sequential elution of peptides was accomplished using an analytical column ( Dr . Maisch GmbH; packed in-house , i . d . , 50 μm; resin , 3 μm ) with a 35 min gradient of 10–38% buffer B ( buffer A , 0 . 1 M acetic acid; buffer B , 0 . 1 M acetic acid , 80% [vol/vol] acetonitrile ) followed by 38–100% B in 3 min , 100% B for 2 min . The flow rate was passively split from 0 . 45 mL/min to 100 nL/min ( Nesvizhskii et al . , 2003 ) . Nanospray was achieved using a distally coated fused silica emitter ( made in-house , o . d . , 375 μm; i . d . , 20 μm ) biased to 1 . 7 kV . Mass spectrometer was operated in the data dependent mode to automatically switch between MS and MS/MS . The high resolution survey full scan was acquired in the orbitrap from m/z 350 to m/z 1500 with a resolution of 30 . 000 ( FHMW ) . The most intense ions at a threshold of above 500 were fragmented in the linear ion trap using collision-induced dissociation at a target value of 10 , 000 . Peak lists were generated from the raw data files using the Proteome Discoverer software package version 1 . 3 . 339 ( Thermo Fisher Scientific ) . Peptide identification was performed by searching the individual peak lists against a concatenated target-decoy database containing the E . coli sequences in the Uniprot database ( release 2012_06 ) supplemented with a common contaminants database using the Mascot search engine version 2 . 3 ( Matrix Science , London , United Kingdom ) via the Proteome Discoverer interface . The search parameters included the use of trypsin as proteolytic enzyme allowing up to a maximum of two missed cleavages . Carbamidomethylation of cysteines was set as a fixed modification , whereas oxidation of methionines was set as a variable modification . Precursor mass tolerance was initially set at 50 ppm , while fragment mass tolerance was set at 0 . 6 Da . Subsequently , the peptide identifications were filtered for an ion score of 20 . Cells were grown to steady state in glucose minimal medium ( GB1 ) pH 7 . 0 ( den Blaauwen et al . , 1999 ) supplemented with 50 μg/mL of required amino acids ( LMC500 and derived strains ) or 20 μg/mL thymine ( BW25113 and derived strains ) at 28°C , or in Lennox LB medium pH 7 . 0 at 28°C , as indicated . Absorbance was measured at 450 nm ( GB1 ) or 600 nm ( Lennox LB ) with a Biochrom Libra S70 . Generation times of the LMC500 strain grown in GB1 and Lennox LB medium at 28°C are 85 and 40 min , respectively , and for BW25113 are 92 and 40 min , respectively . SPR experiments were performed as previously described ( Egan et al . , 2014 ) . The concentration of CpoB injected ranged from 0 . 05 to 0 . 5 μM , or 0 . 017–0 . 167 μM assuming trimerization . Assays were performed in triplicate at 25°C , at a flow rate of 75 μL/min and with an injection time of 5 min . The dissociation constant ( KD ) was calculated by non-linear regression using SigmaPlot 11 software ( Systat Software Inc . ) . Continuous fluorescence GTase assays and measurement of TPase activity using radiolabelled lipid II substrate were performed as described previously ( Bertsche et al . , 2005; Banzhaf et al . , 2012 ) with slight modification . Triton X-100 concentration varied , depending on constituent proteins , from 0 . 04 to 0 . 075% . Fold-increases in GTase rates were calculated against the mean rate obtained with PBP1B alone at the same reaction conditions , at the fastest rate . Assays were performed as described ( Bertsche et al . , 2006; Typas et al . , 2010 ) with minor modifications: Cells were grown overnight , diluted 1:400 into 250 mL Lennox LB , and grown with shaking in baffled flasks at 30°C to OD600 = 0 . 3–0 . 4 . Cells were then pelleted by centrifugation and resuspended at density of ∼1010 cells/mL ( i . e . , equivalent to OD600 = 10 ) in 8 . 0-mL ice-cold CL Buffer I ( 50 mM NaH2PO4 20% [wt/vol] sucrose pH 7 . 4 ) with 100 μg/mL DTSSP ( freshly prepared as a 20 mg/mL stock in CL Buffer I ) . Cells were incubated at 4°C with mixing for 1 hr , then pelleted and frozen at −80°C . Cells were then thawed , resuspended at OD600 = 4 in ice-cold CL Buffer II ( 100 mM Tris–HCl 10 mM MgCl2 , 1 M NaCl pH 7 . 5 ) with 100 μM PMSF , 50 μg/mL protease inhibitor cocktail ( P8465 , Sigma-Aldrich ) and 50 μg/mL DNAse I , and lysed using a microfluidizer . Otherwise , method was as described in ( Bertsche et al . , 2006; Typas et al . , 2010 ) . Antibody amounts for immunoprecipitation were optimized to capture all detectible target protein , as assessed by western blot of pellet and supernatant fractions . Pelleted material was extracted in 1/10 original volume SDS-PAGE buffer to concentrate the final IP fraction . The docking models of CpoB/PBP1B/LpoB complex were built using HADDOCK2 . 1 data-driven docking protocols ( Dominguez et al . , 2003 ) and CNS1 . 2 ( Brunger et al . , 1998 ) for the structure calculations . The initial coordinates of the CpoB , PBP1B , and LpoB molecules were taken from the PBP1B crystal structure ( PDB code 3VMA ) ( Sung et al . , 2009 ) , the CpoB crystal structure from the N-terminal domain ( 2XDJ ) and the C-terminal TPR domain ( 2XEV ) ( Krachler et al . , 2010 ) , and from the LpoB NMR structure ( 2MII ) ( Egan et al . , 2014 ) . During the multi-body docking , ambiguous restraints were applied to drive the docking according to the different experimental data . For LpoB , ambiguous interaction restraints were defined based on the NMR chemical shift perturbation mapping recorded on the LpoB/UB2H complex and on the in vivo and in vitro activities of LpoB and PBP1B alleles as described in Egan et al . ( 2014 ) . For CpoB , in vivo cross-linking detected using pBpa substitutions at PBP1B residues 118 , 123 , 751 , and 753 were used as active restraints to drive the interaction between PBP1B and CpoB . Due to the absence of a unique structure of the disordered N-terminal domain of LpoB ( residues 1–60 ) and the linker between the N- and C-terminal domains of CpoB ( residues 91–108 ) , the two regions were treated as fully flexible . The docking was performed with default HADDOCK parameters including a clustering cutoff of 7 . 5 Å . The HADDOCK score was used to rank the generated models ( Lutje Hulsik et al . , 2013 ) . The generated docking models were analyzed within the Pymol software . BW25113 and BW25113 ΔcpoB were grown in 50 mL of LB to an OD578 of 0 . 3 . The culture was then back-diluted 1:50 into 50 mL fresh LB and grown to an OD578 of 0 . 3 . Cells were then cooled on ice for 10 min prior to harvesting 5 mL by centrifugation ( 10 , 000×g , 10 min , 4°C ) . At this point , a viable count was also performed to determine the number of cells per milliliter of culture ( see below ) . The pellets were resuspended in 100 µL TBS ( Tris-buffered saline ) buffer and lysed by addition of 100 µL SDS-PAGE loading buffer and boiling for 10 min . 3 × 20 µL samples of BW25113 lysate were resolved by SDS-PAGE along with purified CpoB standards ( 0 , 1 , 2 , 4 , 8 , and 16 ng ) loaded in 20 µL of BW25113 ΔcpoB lysate . CpoB was detected with specific antibody after western blot ( Figure 5—figure supplement 2 ) . Images were analyzed using ImageQuant LAS4000 software , giving the chemiluminescence signal derived from CpoB bands over the background ( the signal at the same position as CpoB in the 0 ng sample was used as the background ) . A standard curve was plotted using the known CpoB standards and the amount of CpoB in the BW25113 lysate samples was calculated . This was then converted to molecules per cell using the viable counts , which were performed as follows: Cells were serially diluted 10-fold in ice-cold LB . 100 μL of the 10−5 and 10−6 dilutions were plated on LB agar and grown overnight at 37°C; colonies were then counted and the number of cells per milliliter of culture was calculated .	All bacterial cells are surrounded by a membrane , which forms a protective barrier around the cell . Most bacteria also have a wall surrounding the membrane , which provides structural support . When a bacterial cell divides to produce two daughter cells , it produces a belt-like structure around the middle of the cell . This brings the membrane and cell wall on each side together to a ‘pinch-point’ until the two halves of the cell have been separated . This process must be carefully controlled to ensure that the cell does not burst open at any point . Some bacteria known as ‘Gram-negative’ bacteria have a second membrane on the other side of the cell wall . These cells divide in the same way as other bacteria , but the need to coordinate the movement of three structures instead of two makes it more complicated . Many proteins are known to be involved . For example , one group ( or ‘complex’ ) of proteins—which includes a protein called PBP1B—helps to produce new cell wall material . Another complex called the Tol system provides the energy needed for the outer membrane to be pulled inwards towards the pinch point . However , it has not been clear how these complexes work together to allow the cell to divide . Here , Gray , Egan et al . searched for proteins that can interact with PBP1B during cell division in the Gram-negative bacterium E . coli . The experiments found that a protein called CpoB interacts with both PBP1B and the Tol system . CpoB is found in a band around the middle of the cell , and it regulates the activity of PBP1B in response to signals from the Tol system . If the activity of CpoB is disrupted , cell wall production and the movement of the outer membrane are no longer coordinated , and the membrane falls apart , leading to the death of the bacteria . Gray , Egan et al . 's findings show how the production of new cell wall material can be linked to the inwards movement of the outer membrane during cell division . The next challenges are to understand the precise details of how these processes are coordinated by CpoB and to find out whether CpoB also plays the same role in other bacteria .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "Methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2015	Coordination of peptidoglycan synthesis and outer membrane constriction during Escherichia coli cell division
Phase-amplitude coupling between theta and multiple gamma sub-bands is a hallmark of hippocampal activity and believed to take part in information routing . More recently , theta and gamma oscillations were also reported to exhibit phase-phase coupling , or n:m phase-locking , suggesting an important mechanism of neuronal coding that has long received theoretical support . However , by analyzing simulated and actual LFPs , here we question the existence of theta-gamma phase-phase coupling in the rat hippocampus . We show that the quasi-linear phase shifts introduced by filtering lead to spurious coupling levels in both white noise and hippocampal LFPs , which highly depend on epoch length , and that significant coupling may be falsely detected when employing improper surrogate methods . We also show that waveform asymmetry and frequency harmonics may generate artifactual n:m phase-locking . Studies investigating phase-phase coupling should rely on appropriate statistical controls and be aware of confounding factors; otherwise , they could easily fall into analysis pitfalls . Local field potentials ( LFPs ) exhibit oscillations of different frequencies , which may co-occur and also interact with one another ( Jensen and Colgin , 2007; Tort et al . , 2010; Hyafil et al . , 2015 ) . Cross-frequency phase-amplitude coupling between theta and gamma oscillations has been well described in the hippocampus , whereby the instantaneous amplitude of gamma oscillations depends on the instantaneous phase of theta ( Scheffer-Teixeira et al . , 2012; Schomburg et al . , 2014 ) . More recently , hippocampal theta and gamma oscillations were also reported to exhibit n:m phase-phase coupling , in which multiple gamma cycles are consistently entrained within one cycle of theta ( Belluscio et al . , 2012; Zheng and Zhang , 2013; Xu et al . , 2013 , 2015; Zheng et al . , 2016 ) . The existence of different types of cross-frequency coupling suggests that the brain may use different coding strategies to transfer multiplexed information . Coherent oscillations are believed to take part in network communication by allowing opportunity windows for the exchange of information ( Varela et al . , 2001; Fries , 2005 ) . Standard phase coherence measures the constancy of the phase difference between two oscillations of the same frequency ( Lachaux et al . , 1999; Hurtado et al . , 2004 ) , and has been associated with cognitive processes such as decision-making ( DeCoteau et al . , 2007; Montgomery and Buzsáki , 2007; Nácher et al . , 2013 ) . Similarly to coherence , cross-frequency phase–phase coupling , or n:m phase-locking , also relies on assessing the constancy of the difference between two phase time series ( Tass et al . , 1998 ) . However , in this case the original phase time series are accelerated , so that their instantaneous frequencies can match . Formally , n:m phase-locking occurs when Δφnm ( t ) =n∗φB ( t ) −m∗φA ( t ) is non-uniform but centered around a preferred value , where nφB ( mφA ) denotes the phase of oscillation B ( A ) accelerated n ( m ) times ( Tass et al . , 1998 ) . For example , the instantaneous phase of theta oscillations at 8 Hz needs to be accelerated five times to match in frequency a 40 Hz gamma . A 1:5 phase-phase coupling is then said to occur if theta accelerated five times has a preferred phase lag ( i . e . , a non-uniform phase difference ) in relation to gamma; or , in other words , if five gamma cycles have a consistent phase relationship to one theta cycle . Cross-frequency phase-phase coupling has previously been hypothesized to take part in memory processes ( Lisman and Idiart , 1995; Jensen and Lisman , 2005; Lisman , 2005; Schack and Weiss , 2005; Sauseng et al . , 2008 , 2009; Holz et al . , 2010; Fell and Axmacher , 2011 ) . Recent findings suggest that the hippocampus indeed uses such a mechanism ( Belluscio et al . , 2012; Zheng and Zhang , 2013; Xu et al . , 2013 , 2015; Zheng et al . , 2016 ) . However , by analyzing simulated and actual hippocampal LFPs , in the present work we question the existence of theta-gamma phase-phase coupling . We first certified that we could reliably detect n:m phase-locking when present . To that end , we simulated a system of two Kuramoto oscillators – a ‘theta’ and a ‘gamma’ oscillator – exhibiting variability in instantaneous frequency ( see Materials and methods ) . The mean natural frequency of the theta oscillator was set to 8 Hz , while the mean natural frequency of the gamma oscillator was set to 43 Hz ( Figure 1A ) . When coupled , the mean frequencies aligned to a 1:5 factor by changing to 8 . 5 Hz and 42 . 5 Hz , respectively ( see Guevara and Glass , 1982; García-Alvarez et al . , 2008; Canavier et al . , 2009 ) . Figure 1B depicts three versions of accelerated theta phases ( m = 3 , 5 and 7 ) along with the instantaneous gamma phase ( n = 1 ) of the coupled oscillators ( see Figure 1—figure supplement 1 for the uncoupled case ) . Also shown are the time series of the difference between gamma and accelerated theta phases ( Δφnm ) . The instantaneous phase difference has a preferred lag only for m = 5; when m = 3 or 7 , Δφnmchanges over time , precessing forwards ( m = 3 ) or backwards ( m = 7 ) at an average rate of 17 Hz . Consequently , Δφnm distribution is uniform over 0 and 2π for m = 3 or 7 , but highly concentrated for m = 5 ( Figure 1C ) . The concentration ( or ‘constancy’ ) of the phase difference distribution is used as a metric of n:m phase-locking . This metric is defined as the length of the mean resultant vector ( Rn:m ) over unitary vectors whose angle is the instantaneous phase difference ( eiΔφnm ( t ) ) , and thereby it varies between 0 and 1 . For any pair of phase time series , an Rn:m‘curve’ can be calculated by varying m for n = 1 fixed . As shown in Figure 1D , the coupled – but not uncoupled – oscillators exhibited a prominent peak for n:m = 1:5 , which shows that Rn:m successfully detects n:m phase-locking . 10 . 7554/eLife . 20515 . 003Figure 1 . Measuring cross-frequency phase-phase coupling . ( A ) Traces show 500 ms of the instantaneous phase time series of two Kuramoto oscillators ( see Materials and methods ) . When uncoupled ( top panels ) , the mean natural frequencies of the ‘theta’ and ‘gamma’ oscillator are 8 Hz ( blue ) and 43 Hz ( red ) , respectively . When coupled ( bottom panels ) , the oscillators have mean frequencies of 8 . 5 Hz and 42 . 5 Hz . ( B ) Top blue traces show the instantaneous phase of the coupled theta oscillator for the same period as in A but accelerated m times , where m = 3 ( left ) , 5 ( middle ) and 7 ( right ) . Middle red traces reproduce the instantaneous phase of the coupled gamma oscillator ( i . e . , n = 1 ) . Bottom black traces show the instantaneous phase difference between gamma and accelerated theta phases ( Δφnm ) . Notice roughly constant Δφnm only when theta is accelerated m = 5 times , which indicates 1:5 phase-locking . See Figure 1—figure supplement 1 for the uncoupled case . ( C ) Δφnm distributions for the coupled case ( epoch length = 100 s ) . Notice uniform distributions for n:m = 1:3 and 1:7 , and a highly concentrated distribution for n:m = 1:5 . The black arrow represents the mean resultant vector for each case ( see Materials and methods ) . The length of this vector ( Rn:m ) measures the level of n:m phase-locking . See Figure 1—figure supplement 1 for the uncoupled case . ( D ) Phase-locking levels for a range of n:m ratios for the uncoupled ( left ) and coupled ( right ) oscillators ( epoch length = 100 s ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 00310 . 7554/eLife . 20515 . 004Figure 1—figure supplement 1 . Uncoupled oscillators display uniform Δφnm distribution . ( A , B ) Panels show the same as in Figure 1B , C , but for the uncoupled oscillators . Notice roughly uniform Δφnm distributions . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 004 We next analyzed white-noise signals , in which by definition there is no structured activity; in particular , the spectrum is flat and there is no true n:m phase-locking . Rn:m values measured from white noise should be regarded as chance levels . We band-pass filtered white-noise signals to extract the instantaneous phase of theta ( θ: 4–12 Hz ) and of multiple gamma bands ( Figure 2A ) : slow gamma ( γS: 30–50 Hz ) , middle gamma ( γM: 50–90 Hz ) , and fast gamma ( γF: 90–150 Hz ) . For each frequency pair , we constructed n:m phase-locking curves for epochs of 1 and 10 s , with n = 1 fixed and m varying from 1 to 25 ( Figure 2B ) . In each case , phase-phase coupling was high within the ratio of the analyzed frequency ranges: Rn:m peaked at m = 4–6 for θ−γS , at m = 7–11 for θ−γM , and at m = 12–20 for θ−γF . Therefore , the existence of a ‘bump’ in the Rn:m curve may merely reflect the ratio of the filtered bands and should not be considered as evidence for cross-frequency phase-phase coupling: even filtered white-noise signals exhibit such a pattern . 10 . 7554/eLife . 20515 . 005Figure 2 . Detection of spurious n:m phase-locking in white-noise signals due to inappropriate surrogate-based statistical testing . ( A ) Example white-noise signal ( black ) along with its theta- ( blue ) and gamma- ( red ) filtered components . The corresponding instantaneous phases are also shown . ( B ) n:m phase-locking levels for 1- ( left ) and 10 s ( right ) epochs , computed for noise filtered at theta ( θ; 4–12 Hz ) and at three gamma bands: slow gamma ( γS; 30–50 Hz ) , middle gamma ( γM; 50–90 Hz ) and fast gamma ( γF; 90–150 Hz ) . Notice Rn:m peaks in each case . ( C ) Boxplot distributions of θ−γS R1:5 values for different epoch lengths ( n = 2100 simulations per epoch length ) . The inset shows representative Δφnm distributions for 0 . 3- and 100 s epochs . ( D ) Overview of surrogate techniques . See text for details . ( E ) Top panels show representative Δφnm distributions for single surrogate runs ( Time Shift; 10 runs of 1 s epochs ) , along with the corresponding Rn:m values . The bottom panel shows the pooled Δφnm distribution; the Rn:m of the pooled distribution is lower than the Rn:m of single runs ( compare with values for 1- and 10 s epochs in panel C ) . ( F ) Top , n:m phase-locking levels computed for 1- ( left ) or 10 s ( right ) epochs using either the Original or five surrogate methods ( insets are a zoomed view of Rn:m peaks ) . Bottom , R1:5 values for white noise filtered at θ and γS . Original Rn:m values are not different from Rn:m values obtained from single surrogate runs of Random Permutation and Time Shift procedures . Less conservative surrogate techniques provide lower Rn:m values and lead to the spurious detection of θ−γS phase-phase coupling in white noise . p<0 . 01 , n = 2100 per distribution , one-way ANOVA with Bonferroni post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 00510 . 7554/eLife . 20515 . 006Figure 2—figure supplement 1 . Filtering induces quasi-linear phase shifts in white-noise signals . ( A ) Distribution of the phase difference between two consecutive samples for white noise band-pass filtered at theta ( 4–12 Hz , top ) and slow gamma ( 30–50 Hz , bottom ) . Epoch length = 100 s; sampling rate = 1000 Hz ( dt = 0 . 001 s ) . Notice that the top histogram peaks at ~0 . 05 , which corresponds to 23 . 1480 . 001 ( i . e . , 2πfcdt , where fc is the center frequency ) , and the bottom histogram peaks at ~0 . 25 =23 . 14400 . 001 . ( B ) Rn:m curves computed for theta- and slow gamma-filtered white-noise signals . The black curve was obtained using continuous 1 s long time series sampled at 1000 Hz . The red curve was obtained by also analyzing 1000 data points , but which were subsampled at 20 Hz ( subsampling was performed after filtering ) . Notice Rn:m peak at n:m = 1:5 only for the former case . See also Figure 5—figure supplement 7 for similar results in hippocampal LFPs . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 00610 . 7554/eLife . 20515 . 007Figure 2—figure supplement 2 . Filter bandwidth influences n:m phase-locking levels in white-noise signals . ( A ) Mean Rn:m curves computed for 1 s long white-noise signals filtered into different bands ( same color labels as in B; n = 2100 ) . Notice that the narrower the filter bandwidth , the higher the Rn:m peak . ( B ) Mean Rn:m peak values for different filter bandwidths and epoch lengths ( n = 2100 simulations per filter setting and epoch length ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 00710 . 7554/eLife . 20515 . 008Figure 2—figure supplement 3 . Uniform p-value distributions upon multiple testing of Original Rn:m values against Single Run Rn:m surrogates . The histograms show the distribution of p-values ( bin width = 0 . 02 ) for 10000 t-tests of Original Rn:m vs Single Run surrogate values ( n = 30 samples per group; epoch length = 1 s ) . The red dashed line marks p=0 . 05 . The p-value distributions do not statistically differ from the uniform distribution ( Kolmogorov-Smirnov test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 008 The bump in the Rn:m curve of filtered white noise is explained by the fact that neighboring data points are not independent . In fact , the phase shift between two consecutive data points follows a probability distribution highly concentrated around 2πfcdt , where fc is the filter center frequency and dt the sampling period ( Figure 2—figure supplement 1 ) . For instance , for dt = 1 ms ( 1000 Hz sampling rate ) , consecutive samples of white noise filtered between 4 and 12 Hz are likely to exhibit phase difference of 0 . 05 rad ( 8 Hz center frequency ) ; likewise , signals filtered between 30 and 50 Hz are likely to exhibit phase differences of 0 . 25 rad ( 40 Hz center frequency ) . In turn , the ‘sinusoidality’ imposed by filtering leads to non-zero Rn:m values , which peak at the ratio of the center frequencies , akin to the fact that perfect 8 Hz and 40 Hz sine waves have Rn:m = 1 at n:m = 1:5 . In accordance to this explanation , no Rn:m bump occurs when data points of the gamma phase time series are made independent by sub-sampling with a period longer than a gamma cycle ( Figure 2—figure supplement 1 ) , or when extracting phase values from different trials ( not shown ) . As expected , the effect of filtering-induced sinusoidality on Rn:m values is stronger for narrower frequency bands ( Figure 2—figure supplement 2 ) . Qualitatively similar results were found for 1- and 10 s epochs; however , Rn:m values were considerably lower for the latter ( Figure 2B ) . In fact , for any fixed n:m ratio and frequency pair , Rn:m decreased as a function of epoch length ( see Figure 2C for θ−γS and R1:5 ) : the longer the white-noise epoch the more the phase difference distribution becomes uniform . In other words , as standard phase coherence ( Vinck et al . , 2010 ) and phase-amplitude coupling ( Tort et al . , 2010 ) , phase-phase coupling has positive bias for shorter epochs . As a corollary , notice that false-positive coupling may be detected if control ( surrogate ) epochs are longer than the original epoch . We next investigated the reliability of surrogate methods for detecting n:m phase-locking ( Figure 2D ) . The ‘Original’ Rn:m value uses the same time window for extracting theta and gamma phases ( Figure 2D , upper panel ) . A ‘Time Shift’ procedure for creating surrogate epochs has been previously employed ( Belluscio et al . , 2012; Zheng et al . , 2016 ) , in which the time window for gamma phase is randomly shifted between 1 to 200 ms from the time window for theta phase ( Figure 2D , upper middle panel ) . A variant of this procedure is the ‘Random Permutation’ , in which the time window for gamma phase is randomly chosen ( Figure 2D , lower middle panel ) . Finally , in the ‘Phase Scramble’ procedure , the timestamps of the gamma phase time series are shuffled ( Figure 2D , lower panel ) ; clearly , the latter is the least conservative . For each surrogate procedure , Rn:m values were obtained by two approaches: ‘Single Run’ and ‘Pooled’ ( Figure 2E ) . In the first approach , each surrogate run ( e . g . , a time shift or a random selection of time windows ) produces one Rn:m value ( Figure 2E , top panels ) . In the second , Δφnm from several surrogate runs are first pooled , then a single Rn:m value is computed from the pooled distribution ( Figure 2E , bottom panel ) . As illustrated in Figure 2E , Rn:m computed from a pool of surrogate runs is much smaller than when computed for each individual run . This is due to the dependence of Rn:m on the epoch length: pooling instantaneous phase differences across 10 runs of 1 s surrogate epochs is equivalent to analyzing a single surrogate epoch of 10 s . And the longer the analyzed epoch , the more the noise is averaged out and the lower the Rn:m . Therefore , pooled surrogate epochs summing up to 10 s of total data have lower Rn:m than any individual 1 s surrogate epoch . No phase-phase coupling should be detected in white noise , and therefore Original Rn:m values should not differ from properly constructed surrogates . However , as shown in Figure 2F for θ−γS as an illustrative case ( similar results hold for any frequency pair ) , θ−γS phase-phase coupling in white noise was statistically significantly larger than in phase-scrambled surrogates ( for either Single Run or Pooled distributions ) . This was true for surrogate epochs of any length , although the longer the epoch , the lower the actual and the surrogate Rn:m values , as expected ( compare right and left panels of Figure 2F ) . Pooled R1:5 distributions derived from either time-shifted ( Figure 2F ) or randomly permutated epochs ( not shown ) also led to the detection of false positive θ−γS phase-phase coupling . On the other hand , Original Rn:m values were not statistically different from chance distributions when these were constructed from Single Run Rn:m values for either Time Shift and Random Permutation surrogate procedures ( Figure 2F; see also Figure 2—figure supplement 3 ) . We conclude that neither scrambling phases nor pooling individual surrogate epochs should be employed for statistically evaluating n:m phase-locking . Chance distributions should be derived from surrogate epochs of the same length as the original epoch and which preserve phase continuity . To check if Single Run surrogate distributions are capable of statistically detecting true n:m phase-locking , we next simulated noisy Kuramoto oscillators as in Figure 1 , but of mean natural frequencies set to 8 and 40 Hz . Original R1:5 values were much greater than the surrogate distribution for coupled – but not uncoupled – oscillators ( Figure 3A ) . This result illustrates that variability in the instantaneous frequency leads to low n:m phase-locking levels for independent oscillators even when their mean frequencies are perfect integer multiples . On the other hand , coupled oscillators have high Rn:m because variations of their instantaneous frequencies are mutually dependent . We then proceeded to analyze simulated LFPs from a previously published model network ( Kopell et al . , 2010 ) . The network has two inhibitory interneurons , called O and I cells , which spike at theta and gamma frequency , respectively ( for a motivation of this model , see Tort et al . , 2007 ) . Compared to Single Run surrogate distributions , the model LFP exhibited significant n:m phase-locking only when the interneurons were coupled; Rn:m levels did not differ from the surrogate distribution for the uncoupled network ( Figure 3B ) . ( Note that the Rn:m curve also exhibited a peak for both the uncoupled network and Single Run surrogate data , which is due to the low variability in the instantaneous spike frequency of the model cells; without this variability , however , all networks would display perfect n:m phase-locking ) . 10 . 7554/eLife . 20515 . 009Figure 3 . True n:m phase-locking leads to significant Rn:m values . ( A ) The left panels show mean Rn:m curves and distributions of R1:5 values for original and surrogate ( Random Permutation/Single Run ) data obtained from the simulation of two coupled Kuramoto oscillators ( n = 300; epoch length = 30 s; p<0 . 001 , t-test ) . The right panels show the same , but for uncoupled oscillators . In these simulations , each oscillator has instantaneous peak frequency determined by a Gaussian distribution; the mean natural frequencies of the theta and gamma oscillators were set to 8 Hz and 40 Hz , respectively ( coupling does not alter the mean frequencies since they already exhibit a 1:5 ratio; compare with Figure 1 ) . ( B ) Top panels show results from a simulation of a model network composed of two mutually connected interneurons , O and I cells , which emit spikes at theta and gamma frequency , respectively ( Tort et al . , 2007; Kopell et al . , 2010 ) . Original n:m phase-locking levels are significantly higher than chance ( n = 300; epoch length = 30 s; p<0 . 001 , t-test ) . The bottom panels show the same , but for unconnected interneurons . In this case , n:m phase-locking levels are not greater than chance . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 009 The simulations above show that Single Run surrogates can properly detect n:m phase-locking for oscillators exhibiting variable instantaneous frequency , which is the case of hippocampal theta and gamma oscillations . However , it should be noted that high asymmetry of the theta waveform may also lead to statistically significant Rn:m values per se . As illustrated in Figure 4A , a non-sinusoidal oscillation such as a theta sawtooth wave can be decomposed into a sum of sine waves at the fundamental and harmonic frequencies , which have decreasing amplitude ( i . e . , the higher the harmonic frequency , the lower the amplitude ) . Importantly , the harmonic frequency components are n:m phase-locked to each other: the first harmonic exhibits a fixed 1:2 phase relationship to the fundamental frequency , the second harmonic a 1:3 relationship , and so on ( Figure 4B ) . Of note , the higher frequency harmonics not only exhibit cross-frequency phase-phase coupling to the fundamental theta frequency but also phase-amplitude coupling , since they have higher amplitude at the theta phases where the sharp deflection occurs ( Figure 4C left and Figure 4—figure supplement 1; see also Kramer et al . , 2008 and Tort et al . , 2013 ) . 10 . 7554/eLife . 20515 . 010Figure 4 . Waveform asymmetry may lead to artifactual n:m phase-locking . ( A ) The top traces show a theta sawtooth wave along with its decomposition into a sum of sinusoids at the fundamental ( 7 Hz ) and harmonic ( 14 Hz , 21 Hz , 28 Hz , 35 Hz , etc ) frequencies . The bottom panel shows the power spectrum of the sawtooth wave . Notice power peaks at the fundamental and harmonic frequencies . ( B ) Phase-phase plots ( 2D histograms of phase counts ) for the sawooth wave in A filtered at theta ( 7 Hz; x-axis phases ) and harmonic frequencies ( 14 , 21 , 28 and 35 Hz; y-axis phases ) . ( C ) The left traces show 500 ms of a sawtooth wave along with its theta- and gamma-filtered components and corresponding phase time series . The sawtooth wave was set to have a variable peak frequency , with mean = 8 Hz; no gamma oscillation was added to the signal . Notice that the sharp deflections of the sawtooth wave give rise to artifactual gamma oscillations in the filtered signal ( Kramer et al . , 2008 ) , which have a consistent phase relationship to the theta cycle . The right panels show that artifactual n:m phase-coupling levels induced by the sharp deflections are significantly higher than the chance distribution ( n = 300; epoch length = 30 s; p<0 . 001 , t-test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 01010 . 7554/eLife . 20515 . 011Figure 4—figure supplement 1 . Waveform asymmetry may lead to spurious phase-amplitude coupling . ( A ) A theta sawtooth wave along with its theta- ( 7 Hz ) and gamma-filtered ( 35 Hz ) components . Notice that no gamma oscillations exist in the original sawtooth wave , but they spuriously appear when filtering sharp deflections ( Kramer et al . , 2008 ) . The amplitude of the spurious gamma waxes and wanes within theta cycles . ( B ) Mean gamma amplitude as a function of theta phase . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 01110 . 7554/eLife . 20515 . 012Figure 4—figure supplement 2 . The statistical significance of artifactual n:m phase-locking levels induced by waveform asymmetry depends on epoch length and peak frequency variability . Shown are the median R1:5 computed between theta and slow gamma for sawtooth waves simulated as in Figure 4C , but of different epoch lengths and peak frequency variability . Dashed area corresponds to the interquartile range ( n = 300 ) . Surrogate data were obtained either by Radom Permutation ( top row ) or Time Shift ( bottom row ) procedures . Notice that the longer the epoch or the peak frequency variability , the larger the difference between original and surrogate data , and that this difference is greater for randomly permutated than time-shifted surrogates . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 012 The gamma-filtered component of a theta sawtooth wave of variable peak frequency thus displays spurious gamma oscillations ( i . e . , theta harmonics ) that have a consistent phase relationship to the theta cycle irrespective of variations in cycle length . In randomly permutated data , however , the theta phases associated with spurious gamma differ from cycle to cycle due to the variability in instantaneous theta frequency . As a result , the spurious n:m phase-coupling induced by sharp signal deflections is significantly higher than the Random Permutation/Single Run surrogate distribution ( Figure 4C right and Figure 4—figure supplement 2 top row ) . Interestingly , the significance of this spurious effect is much lower when using the Time Shift procedure ( Figure 4—figure supplement 2 bottom row ) , probably due to the proximity between the original and the time-shifted time series ( 200 ms maximum distance ) . We next proceeded to analyze hippocampal CA1 recordings from seven rats , focusing on the periods of prominent theta activity ( active waking and REM sleep ) . We found similar results between white noise and actual LFP data . Namely , Rn:m curves peaked at n:m ratios according to the filtered bands , and Rn:m values were lower for longer epochs ( Figure 5A; compare with Figure 2B ) . As shown in Figure 5B , Original Rn:m values were not statistically different from a proper surrogate distribution ( Random Permutation/Single Run ) in epochs of up to 100 s ( but see Figure 10 ) . Noteworthy , as with white-noise data ( Figure 2F ) , false positive phase-phase coupling would be inferred if an inadequate surrogate method were employed ( Time Shift/Pooled ) ( Figure 5B ) . 10 . 7554/eLife . 20515 . 013Figure 5 . Spurious detection of theta-gamma phase-phase coupling in the hippocampus . ( A ) n:m phase-locking levels for actual hippocampal LFPs . Compare with Figure 2B . ( B ) Original and surrogate distributions of Rn:m values for slow ( R1:5; left ) and middle gamma ( R1:8; right ) for different epoch lengths . The original data is significantly higher than the pooled surrogate distribution , but indistinguishable from the distribution of surrogate values computed using single runs . Similar results hold for fast gamma . p<0 . 01 , n = 7 animals , Friedman’s test with Nemenyi post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 01310 . 7554/eLife . 20515 . 014Figure 5—figure supplement 1 . Lack of evidence for cross-frequency phase-phase coupling between theta and gamma oscillations using alternative phase-locking metrics . ( A ) The left plots show the mean radial distance ( R ) computed for gamma phases in different theta phase bins , as described in Sauseng et al . ( 2009 ) . The lines denote the mean ± SD over all channels across animals ( n = 16 channels per rat x seven rats ) ; 300 1 s long epochs were analyzed for each channel . Note that original and surrogate R values overlap . The variations of R values within a theta cycle are explained by the different number of theta phase bins ( right bar plot ) , which leads to different number of analyzed samples; the higher the number of analyzed samples , the lower the R ( see also Figure 2C ) . ( B ) The first column shows the mean pairwise phase consistency ( PPC ) between gamma and accelerated theta phases as a function of the number of Δφnm samples ( dashed lines denote SD over individual PPC estimates; n = 112 channels x 1000 PPC estimates per channel ) . Since PPC requires independent observations ( Vinck et al . , 2010 ) , Δφnm was randomly sampled to avoid the statistical dependence among neighboring data points imposed by the filter ( Figure 2—figure supplement 1; see also Figure 5—figure supplement 7 ) . The second column shows mean PPC as function of n:m ratio ( individual PPC estimates were computed using 1000 Δφnm samples ) ; the boxplot distributions show PPC values at selected n:m ratios , as labeled . PPC values are very low for all analyzed frequency pairs and not statistically different from zero . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 01410 . 7554/eLife . 20515 . 015Figure 5—figure supplement 2 . Spurious detection of theta-gamma phase-phase coupling when theta phase is estimated by interpolation . ( A ) n:m phase-locking levels for actual hippocampal LFPs ( same dataset as in Figure 5 ) . Theta phase was estimated by the interpolation method described in Belluscio et al . ( 2012 ) . ( B ) Original and surrogate distributions of Rn:m values . The original data are significantly higher than surrogate values obtained from pooled Δφnm , but indistinguishable from single run surrogates . p<0 . 01 , n = 7 animals , Friedman’s test with Nemenyi post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 01510 . 7554/eLife . 20515 . 016Figure 5—figure supplement 3 . Spurious detection of theta-gamma phase-phase coupling ( second dataset ) . ( A ) n:m phase-locking levels for actual hippocampal LFPs . ( B ) Original and surrogate distributions of Rn:m values . Results obtained for three rats recorded in an independent laboratory ( see Materials and methods ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 01610 . 7554/eLife . 20515 . 017Figure 5—figure supplement 4 . Lack of evidence for theta-gamma phase-phase coupling in all hippocampal layers . ( Left ) Example estimation of the anatomical location of a 16-channel silicon probe by the characteristic depth profile of sharp-wave ripples ( inter-electrode distance = 100 μm ) . ( Middle ) Original and surrogate ( Random Permutation/Single Run ) distributions of Rn:m values computed between theta phase and the phase of three gamma sub-bands ( 1 s long epochs ) . Different rows show results for different layers . ( Right ) Distribution of original and surrogate Rn:m values computed for current-source density ( CSD ) signals ( 1 s long epochs ) in three hippocampal layers: s . pyramidale ( top ) , s . radiatum ( middle ) , and s . lacunosum-moleculare ( bottom ) . Notice no difference between original and surrogate values . Similar results were found in all animals . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 01710 . 7554/eLife . 20515 . 018Figure 5—figure supplement 5 . Lack of theta-gamma phase-phase coupling in independent components of gamma activity . ( Left ) Average phase-amplitude comodulograms for three independent components ( IC ) that maximize coupling between theta phase and the amplitude of slow gamma ( top row ) , middle gamma ( middle row ) and fast gamma ( bottom row ) oscillations ( n = 4 animals ) . Each IC is a weighted sum of LFPs recorded in different hippocampal layers ( see Schomburg et al . , 2014 ) . ( Middle ) n:m phase-locking levels for theta phase and the phase of ICs filtered at the gamma band maximally coupled to theta in the phase-amplitude comodulogram . ( Right ) Original and surrogate distributions of Rn:m values . Rn:m values were computed for 1 s long epochs ( n = 4 animals ) ; surrogate gamma phases were obtained by Random Permutation/Single Run . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 01810 . 7554/eLife . 20515 . 019Figure 5—figure supplement 6 . Lack of theta-gamma phase-phase coupling during transient gamma bursts . ( A ) Examples of slow-gamma bursts . Top panels show raw LFPs , along with theta- ( thick blue line ) and slow gamma-filtered ( thin red line ) signals . The amplitude envelope of slow gamma is also shown ( thick red line ) . The bottom rows show gamma and accelerated theta phases ( m = 5 ) , along with their instantaneous phase difference ( Δφ1:5 ) . For each gamma sub-band , a ‘gamma burst’ was defined to occur when the gamma amplitude envelope was 2SD above the mean . In these examples , periods identified as slow-gamma bursts are marked with yellow in the amplitude envelope and phase difference time series . Notice variable �φ1:5 across different burst events . ( B ) The left panel shows n:m phase-locking levels for theta phase and the phase of different gamma sub-bands ( 1 s epochs ) ; for each gamma sub-band , Rn:m values were computed using only theta and gamma phases during periods of gamma bursts . The right panels show original and surrogate ( Random Permutation/Single Run ) distributions of Rn:m values ( n = 4 animals ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 01910 . 7554/eLife . 20515 . 020Figure 5—figure supplement 7 . The bump in the Rn:m curve of hippocampal LFPs highly depends on analyzing contiguous phase time series data . Average Rn:m curves computed for theta- and gamma-filtered hippocampal LFPs . The green curves were obtained using 1 s ( top ) or 10 s ( bottom ) continuous epochs of the phase time series , sampled at 1000 Hz ( same analysis as in Figure 5A ) . The blue curves were obtained by analyzing 1000 data points subsampled from the phase time series at 20 Hz ( i . e . , 50 ms sampling period , longer than a gamma cycle ) . The red curves were obtained by analyzing 1000 ( top ) or 10000 ( bottom ) data points randomly sampled from the phase time series . These plots show that the prominent bump in the Rn:m curve of actual LFPs only occurs for continuously sampled data ( 1000 Hz sampling rate ) , and therefore probably reflects the ‘sinusoidality’ imposed by the filter ( see also Figure 2—figure supplement 1 ) . But notice that a small Rn:m bump remains for θ−γS ( see Figure 10—figure supplement 2 ) . Due to limitation of total epoch length , we could not perform the 20 Hz subsampling analysis for 10000 points , but notice that the blue and red curves coincide for 1000 points . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 02010 . 7554/eLife . 20515 . 021Figure 5—figure supplement 8 . Different filter types give rise to similar results . ( A ) Original ( green ) and surrogate ( red ) n:m phase-locking levels for actual hippocampal LFPs ( same dataset as in Figure 5 ) filtered at theta and slow gamma ( 1 s epochs ) . Different rows show results for different types of filters . FIR corresponds to the same finite impulse response filter employed in all other figures . For the infinite impulse response filters ( Butterworth and Bessel ) , the digit on the right denotes the filter order . Wavelet filtering was achieved by convolution with a complex Morlet wavelet with a center frequency of 7 Hz . ( B ) Original and surrogate distributions of R1:5 values . For each filter type , the original data is significantly higher than surrogate values obtained from pooled Δφnm , but indistinguishable from single run surrogates . *p<0 . 01 , n = 7 animals , Friedman’s test with Nemenyi post-hoc test . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 021 We also found no difference between original and surrogate n:m phase-locking levels when employing the metric described in Sauseng et al . ( 2009 ) ( Figure 5—figure supplement 1 ) , and when estimating theta phase by interpolating phase values between 4 points of the theta cycle ( trough , ascending , peak and descending points ) as performed in Belluscio et al . ( 2012 ) ( Figure 5—figure supplement 2 ) . The latter was somewhat expected since the phase-phase coupling results in Belluscio et al . ( 2012 ) did not depend on this particular method of phase estimation ( see their Figure 6Ce ) . Moreover , coupling levels did not statistically differ from zero when using the pairwise phase consistency metric described in Vinck et al . ( 2010 ) ( Figure 5—figure supplement 1 ) . We further confirmed our results by analyzing data from three additional rats recorded in an independent laboratory ( Figure 5—figure supplement 3; see Materials and methods ) . In addition , we also found similar results in LFPs from other hippocampal layers than s . pyramidale ( Figure 5—figure supplement 4 ) , in neocortical LFPs ( not shown ) , in current-source density ( CSD ) signals ( Figure 5—figure supplement 4 ) , in independent components that isolate activity of specific gamma sub-bands ( Schomburg et al . , 2014 ) ( Figure 5—figure supplement 5 ) , and in transient gamma bursts ( Figure 5—figure supplement 6 ) . Since Original Rn:m values were not greater than Single Run surrogate distributions , we concluded that there is lack of convincing evidence for n:m phase-locking in the hippocampal LFPs analyzed here . However , as in previous reports ( Belluscio et al . , 2012; Zheng et al . , 2016 ) , phase-phase plots ( 2D histograms of theta phase vs gamma phase ) of actual LFPs displayed diagonal stripes ( Figure 6 ) , which seem to suggest phase-phase coupling . We next sought to investigate what causes the diagonal stripes in phase-phase plots . 10 . 7554/eLife . 20515 . 022Figure 6 . Phase-phase plots of hippocampal LFPs display diagonal stripes . Phase-phase plot for theta and slow gamma ( average over animals; n = 7 rats ) . Notice diagonal stripes suggesting phase-phase coupling . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 022 In Figure 7 we analyze a representative LFP with prominent theta oscillations at ~7 Hz recorded during REM sleep . Due to the non-sinusoidal shape of theta ( Belluscio et al . , 2012; Sheremet et al . , 2016 ) , the LFP also exhibited spectral peaks at harmonic frequencies ( Figure 7A ) . We constructed phase–phase plots using LFP components narrowly filtered at theta and its harmonics: 14 , 21 , 28 and 35 Hz . Similarly to the sawtooth wave ( Figure 4B ) , the phase-phase plots exhibited diagonal stripes whose number was determined by the harmonic order ( i . e . , the 1st harmonic exhibited two stripes , the second harmonic three stripes , the third , four stripes and the fourth , five stripes; Figure 7Bi–iv ) . Interestingly , when the LFP was filtered at a broad gamma band ( 30–90 Hz ) , we observed five diagonal stripes , the same number as when narrowly filtering at 35 Hz; moreover , both gamma and 35 Hz filtered signals exhibited the exact same phase lag ( Figure 7Biv–v ) . Therefore , these results indicate that the diagonal stripes in phase-phase plots may be influenced by theta harmonics . Under this interpretation , signals filtered at the gamma band would be likely to exhibit as many stripes as expected for the first theta harmonic falling within the filtered band . Consistent with this possibility , we found that the peak frequency of theta relates to the number of stripes ( Figure 8 ) . 10 . 7554/eLife . 20515 . 023Figure 7 . Phase–phase coupling between theta and gamma oscillations may be confounded by theta harmonics . ( A ) Top , representative LFP epoch exhibiting prominent theta activity ( ~7 Hz ) during REM sleep . Bottom , power spectral density . The inset shows power in dB scale . ( B ) Phase–phase plots for theta and LFP band-pass filtered at harmonic frequencies ( 14 , 21 , 28 and 35 Hz ) , computed using 20 min of concatenated REM sleep . Also shown are phase-phase plots for the conventional gamma band ( 30–90 Hz ) and for the average over individual surrogate runs . Notice that the former mirrors the phase-phase plot of the fourth theta harmonics ( 35 Hz ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 02310 . 7554/eLife . 20515 . 024Figure 7—figure supplement 1 . Histogram counts leading to diagonal stripes in phase-phase plots are statistically significant when compared to the distribution of surrogate counts . Panels show the significance of the phase-phase plots in Figure 7 when compared to the mean and standard deviation of pooled surrogate counts . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 02410 . 7554/eLife . 20515 . 025Figure 8 . The number of stripes in phase-phase plots is determined by the frequency of the first theta harmonic within the filtered gamma range . ( A ) Representative example in which theta has peak frequency of 7 . 1 Hz . The phase-phase plot between theta and slow gamma ( 30–50 Hz ) exhibits five stripes , since the fourth theta harmonic ( 35 . 5 Hz ) is the first to fall within 30 and 50 Hz . The rightmost panels show the average phase-phase plot computed over all time-shifted surrogate runs ( n = 1000 ) and the significance of the original plot when compared to the mean and standard deviation over individual surrogate counts , respectively . ( B ) Example in which theta has peak frequency of 8 . 4 Hz and the phase-phase plot exhibits four stripes , which correspond to the third theta harmonic ( 33 . 6 Hz ) . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 02510 . 7554/eLife . 20515 . 026Figure 8—figure supplement 1 . Individual time-shifted surrogate runs exhibit diagonal stripes in phase-phase plots . ( A ) The middle panels show phase-phase plots for theta and slow gamma computed for different time shifts of the example epoch analyzed in Figure 8A . Notice diagonal stripes in individual surrogate runs . The left- and the rightmost panels show the original and average surrogate phase-phase plots , respectively ( same panels as in Figure 8A ) . ( B ) Same as above , but for the example epoch analyzed in Figure 8B . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 026 As in previous studies ( Belluscio et al . , 2012; Zheng et al . , 2016 ) , phase-phase plots constructed using data averaged from individual time-shifted epochs exhibited no diagonal stripes ( Figure 7Bvi and Figure 8 ) . This is because different time shifts lead to different phase lags; the diagonal stripes of individual surrogate runs that could otherwise be apparent cancel each other out when combining data across multiple runs of different lags ( Figure 8—figure supplement 1 ) . Moreover , as in Belluscio et al . ( 2012 ) , the histogram counts that give rise to the diagonal stripes were deemed statistically significant when compared to the mean and standard deviation over individual counts from time-shifted surrogates ( Figure 7—figure supplement 1 and Figure 8 ) . To gain further insight into what generates the diagonal stripes , we next analyzed white-noise signals . As shown in Figure 9A , phase-phase plots constructed from filtered white-noise signals also displayed diagonal stripes . Since white noise has no harmonics , these results show that the sinusoidality induced by the filter can by itself lead to diagonal stripes in phase-phase plots , in the same way that it leads to a bump in the Rn:m curve ( Figure 2 and Figure 2—figure supplement 1 ) . Importantly , as in actual LFPs , bin counts in phase-phase plots of white-noise signals were also deemed statistically significant when compared to the distribution of bin counts from time-shifted surrogates ( Figure 9A ) . Since by definition white noise has no n:m phase-locking , we concluded that the statistical analysis of phase-phase plots as originally introduced in Belluscio et al . ( 2012 ) is too liberal . Nevertheless , we found that phase-phase plots of white noise were no longer statistically significant when using the same approach as in Belluscio et al . ( 2012 ) but corrected for multiple comparisons ( i . e . , the number of bins ) by the Holm-Bonferroni method ( the FDR correction still led to significant bins; not shown ) . This result was true for different epoch lengths and also when computing surrogate phase-phase plots using the Random Permutation procedure ( Figure 9A ) . Consistently , for all epoch lengths , Original Rn:m values fell inside the distribution of Single Run surrogate Rn:m values computed using either Time Shift and Random Permutation procedures ( Figure 9B ) . 10 . 7554/eLife . 20515 . 027Figure 9 . Phase-phase plots of white-noise signals display diagonal stripes . ( A ) Representative phase-phase plots computed for white-noise signals . Notice the presence of diagonal stripes for both 100 s ( left ) and 1200 s ( right ) epochs . The colormaps underneath show the p-values of the original bin counts when compared to the mean and standard deviation over bin counts of single time-shifted surrogate runs . Also shown are significance maps after correcting for multiple comparisons ( Holm-Bonferroni ) using either time-shifted ( top ) or randomly permutated ( bottom ) surrogate runs . No bin count was considered statistically significant after the correction . ( B ) The top panels show original Rn:m curves ( green ) plotted along with Single Run distributions of Rn:m curves of time-shifted ( orange ) and randomly permutated ( red ) surrogates for different epoch lengths ( shades denote the 2 . 5th–97 . 5th percentile interval; n = 2100 per distribution ) . The bottom panels show the same in a zoomed scale . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 027 The observations above suggest that the diagonal stripes in phase-phase plots of hippocampal LFPs may actually be caused by filtering-induced sinusoidality , as opposed to being an effect of theta harmonics as we first interpreted . To test this possibility , we next revisited the significance of phase-phase plots of actual LFPs . For epochs of up to 100 s , we found similar results as in white noise , namely , bin counts were no longer statistically significant after correcting for multiple comparisons ( Holm-Bonferroni method ) ; this was true when using either the Time Shift or Random Permutation procedures ( Figure 10A ) . Surprisingly , however , when analyzing much longer time series ( 10 or 20 min of concatenated periods of REM sleep ) , several bin counts became statistically significant when compared to randomly permutated , but not time-shifted , surrogates ( Figure 10A ) . Moreover , this result reflected in the Rn:m curves: the Original Rn:m curve fell within the distribution of Time Shift/Single Run surrogate Rn:m values for all analyzed lengths , but outside the distribution of Random Permutation/Single Run surrogates for the longer time series ( Figure 10B ) . We believe such a finding relates to what we observed for synthetic sawtooth waves , in which Random Permutation was more sensitive than Time Shift to detect the significance of the artifactual coupling caused by waveform asymmetry ( Figure 4—figure supplement 2 ) . In this sense , the n:m phase-locking between fundamental and harmonic frequencies would persist for small time shifts ( ±200 ms ) , albeit in different phase relations , while it would not resist the much larger time shifts obtained through random permutations . However , irrespective of this explanation , it should be noted that since the n:m phase-locking metrics cannot separate artifactual from true coupling , the possibility of the latter cannot be discarded . But if this is the case , we consider unlikely that the very low coupling level ( ~0 . 03 ) would have any physiological significance . 10 . 7554/eLife . 20515 . 028Figure 10 . Weak but statistically significant n:m phase-locking can be detected when analyzing long LFP epochs ( >100 s ) . ( A ) Panels show the same as in Figure 9A but for a representative hippocampal LFP . Notice that several bin counts of the 1200 s epoch remain statistically significant after correction for multiple comparisons ( Holm-Bonferroni ) when compared to randomly permutated , but not time-shifted , surrogates ( bottom right plot ) . ( B ) As in Figure 9B , but for actual LFPs ( n = 300 samples per animal; the number of analyzed animals is stated in each panel ) . For the very long epochs , notice that the original Rn:m curve falls within the distribution of time-shifted surrogates but outside the distribution of randomly permutated ones . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 02810 . 7554/eLife . 20515 . 029Figure 10—figure supplement 1 . Random Permutation leads to less visible diagonal stripes than Time Shift in phase-phase plots of long LFP epochs . Examples of phase-phase plots computed for single Time Shift ( top ) and Random Permutation ( bottom ) surrogate runs of 100 ( left ) and 1200 s ( right ) for the same hippocampal LFP as in Figure 10A . Notice that , for the longer LFP epoch , the randomly permutated surrogate exhibits less discernible stripes than the time-shifted one . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 02910 . 7554/eLife . 20515 . 030Figure 10—figure supplement 2 . Diagonal stripes in phase-phase plots depend on analyzing contiguous phase time series data . Phase-phase plots of a white noise ( left ) and an actual LFP ( right ) computed using 100 s of total data , but subsampled at 83 . 3 Hz ( we used a subsampling period of 12 ms because the total data length of the actual LFP was 1200 s ) . Notice no diagonal stripes in the phase-phase plot of white noise , while some striped-like pattern persists for the actual LFP data . DOI: http://dx . doi . org/10 . 7554/eLife . 20515 . 030 We conclude that the diagonal stripes in phase-phase plots of both white noise and actual LFPs are mainly caused by a temporary n:m alignment of the phase time-series secondary to the filtering-induced sinusoidality , and as such they are also apparent in surrogate data ( Figure 8—figure supplement 1 and Figure 10—figure supplement 1 ) . However , for actual LFPs there is a second influence , which can only be detected when analyzing very long epoch lengths , and which we believe is due to theta harmonics . When searching for phase-phase coupling between theta and gamma , we noticed that our Rn:m values differed from those reported in previous studies ( Belluscio et al . , 2012; Xu et al . , 2013 , 2015; Zheng et al . , 2016 ) . We suspected that this could be due to differences in the duration of the analyzed epochs . We then investigated the dependence of Rn:m on epoch length , and found a strong positive bias for shorter epochs . In addition , Rn:m values exhibit greater variability across samples as epoch length decreases for both white noise and actual data ( e . g . , compare in Figure 5B the data dispersion in Original R1:5 or R1:8 boxplots for different epoch lengths ) . Since theta and gamma peak frequencies are not constant in these signals , the longer the epoch , the more the theta and gamma peak frequencies are allowed to fluctuate and the more apparent the lack of coupling . On the other hand , Δφnm distribution becomes less uniform for shorter epochs . The dependence of n:m phase-coupling metrics on epoch length has important implications in designing surrogate epochs for testing the statistical significance of actual Rn:m values . Of note , methodological studies on 1:1 phase-synchrony have properly used single surrogate runs of the same length as the original signal ( Le Van Quyen et al . , 2001; Hurtado et al . , 2004 ) . As demonstrated here , spurious detection of phase-phase coupling may occur if surrogate epochs are longer than the original epoch . This is the case when one lumps together several surrogate epochs before computing Rn:m . When employing proper controls , our results show that Rn:m values of real data do not differ from surrogate values in theta epochs of up to 100 s . Moreover , the prominent bump in the Rn:m curve disappears when subsampling data at a lower frequency than gamma for both white noise and hippocampal LFPs ( see Figure 2—figure supplement 1 and Figure 5—figure supplement 7 ) , which suggests that it is due to the statistical dependence among contiguous data points introduced by the filter ( which we referred to as ‘filtering-induced sinusoidality’ ) . Therefore , even though the n:m phase-locking metric Rn:m is theoretically well-defined and varies between 0 and 1 , an estimated Rn:m value in isolation does not inform if two oscillations exhibit true phase-coupling or not . This can only be inferred after testing the statistical significance of the estimated Rn:m value against a proper surrogate distribution ( but notice that false-positive cases may occur due to waveform asymmetry; Figure 4C ) . While constructing surrogate data renders the metric computationally more expensive , such an issue is not specific for measuring n:m phase-locking but also happens for other metrics commonly used in the analysis of neurophysiological data , such as coherence , spike-field coupling , phase-amplitude coupling , mutual information and directionality measures , among many others ( Le Van Quyen et al . , 2001; Hurtado et al . , 2004; Pereda et al . , 2005; Tort et al . , 2010 ) . The recent studies assessing theta-gamma phase-phase coupling in hippocampal LFPs have not tested the significance of individual Rn:m values against chance ( Belluscio et al . , 2012; Zheng and Zhang , 2013; Xu et al . , 2013 , 2015; Zheng et al . , 2016 ) . Two studies ( Belluscio et al . , 2012; Zheng et al . , 2016 ) statistically inferred the existence of n:m phase-locking by comparing empirical phase-phase plots with those obtained from the average of 1000 time-shifted surrogate runs . Specifically , Belluscio et al . ( 2012 ) established a significance threshold for each phase-phase bin based on the mean and standard deviation of individual surrogate counts in that bin , and showed that the bin counts leading to diagonal stripes were statistically significant . Here we were able to replicate these results ( Figure 7—figure supplement 1 and Figure 8 ) . However , we note that a phase-phase bin count is not a metric of n:m phase-locking; it does not inform coupling strength and even coupled oscillators have bins with non-significant counts . A bin count would be analogous to a phase difference vector ( eiΔφnm ( t ) ) , which is also not a metric of n:m phase-locking per se , but used to compute one . That is , in the same way that the Rn:m considers all phase difference vectors , n:m phase-locking can only be inferred when considering all bin counts in a phase-phase plot . In this sense , by analyzing the phase-phase plot as a whole , it was assumed that the appearance diagonal stripes was due to theta-gamma coupling; no such stripes were apparent in phase-phase plots constructed from the average over all surrogate runs ( see Figure 6A in Belluscio et al . , 2012 ) . However , here we showed that single time-shifted surrogate runs do exhibit diagonal stripes ( Figure 8—figure supplement 1 and Figure 10—figure supplement 1 ) , that is , similar stripes exist at the level of a Single Run surrogate analysis , in the same way that Single Run surrogates also exhibit a bump in the Rn:m curve . Averaging 1000 surrogate phase-phase plots destroys the diagonal stripes since different time shifts lead to different phase lags . Moreover , since the average is the sum divided by a scaling factor ( the sample size ) , computing the average phase-phase plot is equivalent to computing a single phase-phase plot using the pool of all surrogate runs , which is akin to the issue of computing a single Rn:m value from a pooled surrogate distribution ( Figure 2 ) . Note that even bin counts in phase-phase plots of white noise are considered significant under the statistical analysis introduced in Belluscio et al . ( 2012 ) ( Figure 9A ) . Nevertheless , this was no longer the case when adapting their original framework to include a Holm-Bonferroni correction for multiple comparisons ( Figure 9A ) . Here we showed that the presence of diagonal stripes in phase-phase plots is not sufficient to conclude the existence of phase-phase coupling . The diagonal stripes are simply a visual manifestation of a maintained phase relationship , and as such they essentially reflect what Rn:m measures: that is , the ‘clearer’ the stripes , the higher the Rn:m . Therefore , in addition to true coupling , the same confounding factors that influence Rn:m also influence phase-phase plots , such as filtering-induced sinusoidality and frequency harmonics . Our results suggest that the former is a main factor , because white-noise signals have no harmonics but nevertheless display stripes in phase-phase plots ( Figure 9A ) . In accordance , no stripes are observed in phase-phase plots of white noise when subsampling the time series ( Figure 10—figure supplement 2; see also Figure 2—figure supplement 1 ) . However , in actual LFPs filtering is not the only influence: ( 1 ) for the same filtered gamma band ( 30–50 Hz ) , the number of stripes relates to theta frequency ( Figure 8 ) ; ( 2 ) for very long time series ( i . e . , 10–20 min of concatenated data ) , the stripes in phase-phase plots of actual data – but not of white noise – persist after correcting for multiple comparisons when employing Random Permutation/Single Run surrogates ( Figure 10A ) ; ( 3 ) a striped-like pattern remains in phase-phase plots of actual LFPs after subsampling the time series ( Figure 10—figure supplement 2 ) . Consistently , Rn:m values of actual LFPs are greater than those of white noise in 1200 s epochs ( ~0 . 03 vs ~0 . 005 , compare the bottom right panels of Figures 9B and 10B ) . Interestingly , Original Rn:m values of actual LFPs are not statistically different from the distribution of Time Shift/Single Run surrogates even for the very long epochs ( Figure 10B ) , which suggests that Random Permutation is more powerful than Time Shift and should therefore be preferred . Though a very weak but true coupling effect cannot be discarded , based on our analysis of sawtooth waves ( Figure 4 and Figure 4—figure supplement 2 ) , we believe these results can be explained by theta harmonics , which would remain phase-locked to the fundamental frequency under small time shifts . Sharp signal deflections have been previously recognized to generate artifactual phase-amplitude coupling ( Kramer et al . , 2008; Scheffer-Teixeira et al . , 2013; Tort et al . , 2013; Aru et al . , 2015; Lozano-Soldevilla et al . , 2016 ) . Interestingly , Hyafil ( 2015 ) recently suggested that the non-sinusoidality of alpha waves could underlie the 1:2 phase-locking between alpha and beta observed in human EEG ( Nikulin and Brismar , 2006; see also Palva et al . , 2005 ) . To the best of our knowledge , there is currently no metric capable of automatically distinguishing true cross-frequency coupling from waveform-induced artifacts in collective signals such as LFP , EEG and MEG signals . Ideally , learning how the signal is generated from the activity of different neuronal populations would answer whether true cross-frequency coupling exists or not ( Hyafil et al . , 2015 ) , but unfortunately this is methodologically challenging . One could argue that we did not analyze a proper dataset , or else that prominent phase-phase coupling would only occur during certain behavioral states not investigated here . We disagree with these arguments for the following reasons: ( 1 ) we could reproduce our results using a second dataset from an independent laboratory ( Figure 5—figure supplement 3 ) , and ( 2 ) we examined the same behavioral states in which n:m phase-locking was reported to occur ( active waking and REM sleep ) . One could also argue that there exists multiple gammas , and that different gamma types are most prominent in different hippocampal layers ( Colgin et al . , 2009; Scheffer-Teixeira et al . , 2012; Tort et al . , 2013; Schomburg et al . , 2014; Lasztóczi and Klausberger , 2014 ) ; therefore , prominent theta-gamma phase-phase coupling could exist in other hippocampal layers not investigated here . We also disagree with this possibility because: ( 1 ) we examined the same hippocampal layer in which theta-gamma phase-phase coupling was reported to occur ( Belluscio et al . , 2012 ) ; moreover , ( 2 ) we found similar results in all hippocampal layers ( we recorded LFPs using 16-channel silicon probes , see Materials and methods ) ( Figure 5—figure supplement 4 ) and ( 3 ) in parietal and entorhinal cortex recordings ( not shown ) . Furthermore , similar results hold when ( 4 ) filtering LFPs within any gamma sub-band ( Figure 5 and Figure 5—figure supplement 1 to 6 ) , ( 5 ) analyzing CSD signals ( Figure 5—figure supplement 4 ) , or ( 6 ) analyzing independent components that maximize activity within particular gamma sub-bands ( Schomburg et al . , 2014 ) ( Figure 5—figure supplement 5 ) . Finally , one could argue that gamma oscillations are not continuous but transient , and that assessing phase-phase coupling between theta and transient gamma bursts would require a different type of analysis than employed here . Regarding this argument , we once again stress that we used the exact same methodology as originally used to detect theta-gamma phase-phase coupling ( Belluscio et al . , 2012 ) . Nevertheless , we also ran analysis only taking into account periods in which gamma amplitude was >2 SD above the mean ( ‘gamma bursts’ ) and found no statistically significant phase-phase coupling ( Figure 5—figure supplement 6 ) . Following Belluscio et al . ( 2012 ) , other studies also reported theta-gamma phase-phase coupling in the rodent hippocampus ( Zheng and Zhang , 2013; Xu et al . , 2013 , 2015; Zheng et al . , 2016 ) and amygdala ( Stujenske et al . , 2014 ) . In addition , human studies had previously reported theta-gamma phase-phase coupling in scalp EEG ( Sauseng et al . , 2008 , 2009; Holz et al . , 2010 ) . Most of these studies , however , have not tested the statistical significance of coupling levels against chance ( Sauseng et al . , 2008 , 2009; Holz et al . , 2010; Zheng and Zhang , 2013; Xu et al . , 2013 , 2015; Stujenske et al . , 2014 ) , while Zheng et al . ( 2016 ) based their statistical inferences on the inspection of diagonal stripes in phase-phase plots as originally introduced in Belluscio et al . ( 2012 ) . We further note that epoch length was often not informed in the animal studies . Based on our results , we believe that differences in analyzed epoch length are likely to explain the high variability of Rn:m values across different studies , from ~0 . 4 ( Zheng et al . , 2016 ) down to 0 . 02 ( Xu et al . , 2013 ) . Since it is philosophically impossible to prove the absence of an effect , the burden of proof should be placed on demonstrating that a true effect exists . In this sense , and to the best of our knowledge , none of previous research investigating theta-gamma phase-phase coupling has properly tested Rn:m against chance . Many studies have focused on comparing changes in n:m phase-locking levels , but we believe these can be influenced by other variables such as changes in power , which affect the signal-to-noise ratio and consequently also the estimation of the phase time series . Interestingly , in their pioneer work , Tass and colleagues used filtered white noise to construct surrogate distributions and did not find significant n:m phase-locking among brain oscillations ( Tass et al . , 1998 , 2003 ) . On the other hand , it is theoretically possible that n:m phase-locking exists but can only be detected by other types of metrics yet to be devised . In any case , our work shows that there is currently no convincing evidence for genuine theta-gamma phase-phase coupling using the same phase-locking metric ( Rn:m ) as employed in previous studies ( Belluscio et al . , 2012; Zheng and Zhang , 2013; Xu et al . , 2013 , 2015; Stujenske et al . , 2014; Zheng et al . , 2016 ) , at least when examining LFP epochs of up to 100 s of prominent theta activity . For longer epoch lengths , though , we did find that Rn:m values of hippocampal LFPs may actually differ from those of randomly permuted , but not time-shifted , surrogates ( Figure 10B ) . While we tend to ascribe such result to the effect of theta harmonics , we note that the possibility of true coupling cannot be discarded . But we are particularly skeptical that the very low levels of coupling strength observed in long LFP epochs would be physiologically meaningful . Lisman and Idiart ( 1995 ) proposed an influential model in which theta and gamma oscillations would interact to produce a neural code . The theta-gamma coding model has since been improved ( Jensen and Lisman , 2005; Lisman , 2005; Lisman and Buzsáki , 2008 ) , but its essence remains the same ( Lisman and Jensen , 2013 ) : nested gamma cycles would constitute memory slots , which are parsed at each theta cycle . Accordingly , Lisman and Idiart ( 1995 ) hypothesized that working memory capacity ( 7 ± 2 ) is determined by the number of gamma cycles per theta cycle . Both phase-amplitude and phase-phase coupling between theta and gamma have been considered experimental evidence for such coding scheme ( Lisman and Buzsáki , 2008; Sauseng et al . , 2009; Axmacher et al . , 2010; Belluscio et al . , 2012; Lisman and Jensen , 2013; Hyafil et al . , 2015; Rajji et al . , 2016 ) . In the case of phase-amplitude coupling , the modulation of gamma amplitude within theta cycles would instruct a reader network when the string of items represented in different gamma cycles starts and terminates . On the other hand , the precise ordering of gamma cycles within theta cycles that is consistent across theta cycles would imply phase-phase coupling; indeed , n:m phase-locking is a main feature of computational models of sequence coding by theta-gamma coupling ( Lisman and Idiart , 1995; Jensen and Lisman , 1996; Jensen et al . , 1996 ) . In contrast to these models , however , our results show that the theta phases in which gamma cycles begin/end are not fixed across theta cycles , which is to say that gamma cycles are not precisely timed but rather drift; in other words , gamma is not a clock ( Burns et al . , 2011 ) . If theta-gamma neural coding exists , our results suggest that the precise location of gamma memory slots within a theta cycle is not required for such a code , and that the ordering of the represented items would be more important than the exact spike timing of the cell assemblies that represent the items ( Lisman and Jensen , 2013 ) . In summary , while absence of evidence is not evidence of absence , our results challenge the hypothesis that theta-gamma phase-phase coupling exists in the hippocampus . At best , we only found significant Rn:m values when examining long LFP epochs ( >100 s ) , but these had very low magnitude ( and we particularly attribute their statistical significance to the effects of harmonics ) . We believe that the evidence in favor of n:m phase-locking in other brain regions and signals could potentially also be explained by simpler effects ( e . g . , filtering-induced sinusoidality , asymmetrical waveform , and improper statistical tests ) . While no current technique can differentiate spurious from true phase-phase coupling , previous findings should be revisited and , whenever suitable , checked against the confounding factors and the more conservative surrogate procedures outlined here . All procedures were approved by our local institutional ethics committee ( Comissão de Ética no Uso de Animais - CEUA/UFRN , protocol number 060/2011 ) and were in accordance with the National Institutes of Health guidelines . We used seven male Wistar rats ( 2–3 months; 300–400 g ) from our breeding colony , kept under 12 hr/12 hr dark-light cycle . We recorded from the dorsal hippocampus through either multi-site linear probes ( n = 6 animals; 4 probes had 16 4320 μm2 contacts spaced by 100 μm; 1 probe had 16 703 μm2 contacts spaced by 100 μm; 1 probe had 16 177 μm2 contacts spaced by 50 μm; all probes from NeuroNexus ) or single wires ( n = 1 animal; 50 μm diameter ) inserted at AP −3 . 6 mm and ML 2 . 5 mm . Results shown in the main figures were obtained for LFP recordings from the CA1 pyramidal cell layer , identified by depth coordinate and characteristic electrophysiological benchmarks such as highest ripple power ( see Figure 5—figure supplement 4 for an example ) . Similar results were obtained for recordings from other hippocampal layers ( Figure 5—figure supplement 4 ) . We also analyzed data from three additional rats downloaded from the Collaborative Research in Computational Neuroscience data sharing website ( www . crcns . org ) ( Figure 5—figure supplement 3 ) . These recordings are a generous contribution by György Buzsáki’s laboratory ( HC3 dataset , Mizuseki et al . , 2013 , 2014 ) . Recording sessions were performed in an open field ( 1 m x 1 m ) and lasted 4–5 hr . Raw signals were amplified ( 200x ) , filtered between 1 Hz and 7 . 5 kHz ( third order Butterworth filter ) , and digitized at 25 kHz ( RHA2116 , IntanTech ) . The LFP was obtained by further filtering between 1–500 Hz and downsampling to 1000 Hz . Active waking and REM sleep periods were identified from spectral content ( high theta/delta power ratio ) and video recordings ( movements during active waking; clear sleep posture and preceding slow-wave sleep for REM ) . The results were identical for active waking and REM epochs; throughout this work we only show the latter . The analyzed REM sleep dataset is available at http://dx . doi . org/10 . 5061/dryad . 12t21 . MATLAB codes for reproducing our analyses are available at https://github . com/tortlab/phase_phase . We used built-in and custom-written MATLAB routines . Band-pass filtering was obtained using a least squares finite impulse response ( FIR ) filter by means of the ‘eegfilt’ function from the EEGLAB Toolbox ( Delorme and Makeig , 2004 ) . The filter order was three times the sampling rate divided by the low cutoff frequency . The eegfilt function calls the MATLAB ‘filtfilt’ function , which applies the filter forward and then again backwards to ensure no distortion of phase values . Similar results were obtained when employing other types of filters ( Figure 5—figure supplement 8 ) . The phase time series was estimated through the Hilbert transform . To estimate the instantaneous theta phase of actual data , we filtered the LFP between 4–20 Hz , a bandwidth large enough to capture theta wave asymmetry ( Belluscio et al . , 2012 ) . Estimating theta phase by the interpolation method described in Belluscio et al . ( 2012 ) led to similar results ( Figure 5—figure supplement 2 ) . The CSD signals analyzed in Figure 5—figure supplement 4 were obtained as −A +2B −C , where A , B and C denote LFP signals recorded from adjacent probe sites . In Figure 5—figure supplement 5 , the independent components were obtained as described in Schomburg et al . ( 2014 ) ; phase-amplitude comodulograms were computed as described in Tort et al . ( 2010 ) . We measured the consistency of the phase difference between accelerated time series ( Δφnm ( tj ) =n∗φγ ( tj ) −m∗φθ ( tj ) ) . To that end , we created unitary vectors whose angle is the instantaneous phase difference ( eiΔφnm ( tj ) ) , where j indexes the time sample , and then computed the length of the mean vector: Rn:m=‖1N∑j=1NeiΔφnm ( tj ) ‖ , where N is the total number of time samples ( epoch length in seconds x sampling frequency in Hz ) . Rn:m equals 1 when Δφnm is constant for all time samples tj , and 0 when Δφnm is uniformly distributed . This metric is also commonly referred to as ‘mean resultant length’ or ‘mean radial distance’ ( Belluscio et al . , 2012; Stujenske et al . , 2014; Zheng et al . , 2016 ) . Qualitatively similar results were obtained when employing the framework introduced in Sauseng et al . ( 2009 ) , which computes the mean radial distance using gamma phases in separated theta phase bins , or the pairwise phase consistency metric described in Vinck et al . ( 2010 ) ( Figure 5—figure supplement 1 ) . Phase-phase plots were obtained by first binning theta and gamma phases into 120 bins and next constructing 2D histograms of phase counts , which were smoothed using a Gaussian kernel of σ = 10 bins . In all cases , theta phase was kept intact while gamma phase was mocked in three different ways: ( 1 ) Time Shift: the gamma phase time series is randomly shifted between 1 and 200 ms; ( 2 ) Random Permutation: a contiguous gamma phase time series of the same length as the original is randomly extracted from the same session . ( 3 ) Phase Scrambling: the timestamps of the gamma phase time series are randomly shuffled ( thus not preserving phase continuity ) . For each case , Rn:m values were computed using either Δφnm distribution for single surrogate runs ( Single Run Distribution ) or the pooled distribution of Δφnm over 100 surrogate runs ( Pooled Distribution ) . For each animal , behavioral state ( active waking or REM sleep ) and epoch length , we computed 300 Original Rn:m values using different time windows along with 300 mock Rn:m values per surrogate method . Therefore , in all figures each boxplot was constructed using the same number of samples ( =300 x number of animals ) . For instance , in Figure 5B we used n = 7 animals x 300 samples per animal = 2100 samples ( but see Statistics below ) . In Figure 2 , boxplot distributions for the white-noise data were constructed using n = 2100 . Kuramoto oscillators displaying n:m phase-locking were modeled as described in Osipov et al . ( 2007 ) :φθ˙= ωθ+ εsin ( nφγ−mφθ ) φγ˙= ωγ+ εsin ( mφθ−nφγ ) , where ε is the coupling strength and ωθ and ωγ are the natural frequencies of theta and gamma , respectively , which followed a Gaussian probability ( σ = 5 Hz ) at each time step . We used ε = 10 , n = 1 , m = 5 , and dt = 0 . 001 s . The mean theta and gamma frequencies of each simulation are stated in the main text . For uncoupled oscillators , we set ε = 0 . For implementing the O-I cell network ( Figure 3B ) , we simulated the model previously described in Kopell et al . ( 2010 ) . We used the same parameters as in Figure 3A of Kopell et al . ( 2010 ) , with white noise ( σ = 0 . 001 ) added to the I cell drive to create variations in spike frequency . NEURON ( https://www . neuron . yale . edu/ ) codes for the model are available at ModelDB ( https://senselab . med . yale . edu/ ) . The sawtooth wave in Figure 4C was simulated using dt = 0 . 001 s . Its instantaneous frequency followed a Gaussian distribution with mean = 8 Hz and σ = 5 Hz; white noise ( σ = 0 . 1 ) was added to the signal . In Figures 3 and 4C , boxplot distributions for simulated data were constructed using n = 300 . For white noise data ( Figure 2F ) , given the large sample size ( n = 2100 ) and independence among samples , we used one-way ANOVA with Bonferroni post-hoc test . For statistical analysis of real data ( Figure 5B ) , we avoided nested design and inflation of power and used the mean Rn:m value per animal . In this case , due to the reduced sample size ( n = 7 ) and lack of evidence of normal distribution ( Shapiro-Wilk normality test ) , we used the Friedman’s test and Nemenyi post-hoc test . In Figures 3 and 4C , we tested if Rn:m values of simulated data were greater than the distribution of surrogate values using one-tailed t-tests .	Neuroscientists have long sought to understand how the brain works by analyzing its electrical activity . Placing electrodes on the scalp or lowering them into the brain itself reveals rhythmic waves of activity known as oscillations . These arise when large numbers of neurons fire in synchrony . Recordings reveal that the frequency of these oscillations – the number of cycles of a wave per second , measured in Hertz – can vary between brain regions , and within a single region over time . Moreover , oscillations with different frequencies can co-exist and interact with one another . Within the hippocampus , an area of the brain involved in memory , two types of oscillations dominate: theta waves and gamma waves . Theta waves are relatively slow waves , with a frequency between 5 and 10 Hertz . Gamma waves are faster , with a frequency of up to 100 Hertz . Recent work has suggested that gamma waves and theta waves show a phenomenon called phase-phase coupling . Since gamma waves are faster than theta waves , multiple cycles of gamma can occur during a single cycle of theta . Phase-phase coupling is the idea that gamma and theta waves align themselves , such that gamma waves always begin at the same relative position within a theta wave . This was thought to help the hippocampus to encode memories . Using computer simulations and recordings from the rat hippocampus , Scheffer-Teixeira and Tort have now reexamined the evidence for theta-gamma phase-phase coupling . The new results suggest that previous reports describing the phenomenon may have relied on inadequate statistical techniques . Using stringent control analyses , Scheffer-Teixeira and Tort find no evidence for prominent theta-gamma phase-phase coupling in the hippocampus . Instead , the simulations suggest that what appeared to be statistically significant coupling may in reality be an artifact of the previous analysis . Phase-phase coupling of theta and gamma waves has also been reported in the human hippocampus . The next step therefore is to apply these more robust analysis techniques to data from the human brain . While revisiting previously accepted findings may not always be popular , it will likely be essential if neuroscientists want to accurately understand how new memories are formed .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	On cross-frequency phase-phase coupling between theta and gamma oscillations in the hippocampus
PARP-7 ( TiPARP ) is a mono ( ADP-ribosyl ) transferase whose protein substrates and biological activities are poorly understood . We observed that PARP7 mRNA levels are lower in ovarian cancer patient samples compared to non-cancerous tissue , but PARP-7 protein nonetheless contributes to several cancer-related biological endpoints in ovarian cancer cells ( e . g . growth , migration ) . Global gene expression analyses in ovarian cancer cells subjected to PARP-7 depletion indicate biological roles for PARP-7 in cell-cell adhesion and gene regulation . To identify the MARylated substrates of PARP-7 in ovarian cancer cells , we developed an NAD+ analog-sensitive approach , which we coupled with mass spectrometry to identify the PARP-7 ADP-ribosylated proteome in ovarian cancer cells , including cell-cell adhesion and cytoskeletal proteins . Specifically , we found that PARP-7 MARylates α-tubulin to promote microtubule instability , which may regulate ovarian cancer cell growth and motility . In sum , we identified an extensive PARP-7 ADP-ribosylated proteome with important roles in cancer-related cellular phenotypes . Members of the poly ( ADP-ribose ) polymerase ( PARP ) family of enzymes catalyze ADP-ribosylation ( ADPRylation ) , a posttranslational modification of proteins , through covalent transfer of ADP-ribose ( ADPR ) from β-nicotinamide adenine dinucleotide ( NAD+ ) onto a variety of amino acid residues , including glutamate , aspartate , and serine ( Gibson and Kraus , 2012; Gupte et al . , 2017 ) . ADPRylation can occur as a single ADP-ribose ( i . e . mono ( ADP-ribose ) , [MAR] ) or multiple ADP-ribose moieties ( i . e . poly ( ADP-ribose ) , [PAR] ) ( Gibson and Kraus , 2012; O’Sullivan et al . , 2019 ) . Both MAR and PAR modifications are reversible and can be removed by a variety of ADPR hydrolases ( O’Sullivan et al . , 2019; Rack et al . , 2020 ) . Free and protein-linked ADP-ribose moieties can be bound by proteins ( ‘readers’ ) containing ADPR-binding domains ( e . g . macrodomains , WWE domains ) , allowing MAR or PAR to be interpreted or ‘read’ ( Gibson and Kraus , 2012; Teloni and Altmeyer , 2016 ) . ADPRylation can modulate target protein functions , including enzymatic activity , interactions with binding partners , and protein stability through both direct effects resulting from chemical modification and indirect effects mediated by ADPR-binding ‘reader’ proteins ( Gupte et al . , 2017; Kim et al . , 2020; Ryu et al . , 2015 ) . In humans , the PARP family includes 17 members , each with unique structural domains , expression patterns , targets , enzymatic activity , localization , and functions ( Amé et al . , 2004; Vyas et al . , 2013; Vyas et al . , 2014 ) . In spite of the moniker , PARP family members are mostly monoADPR transferases ( MARTs ) . PARP family MARTs include PARPs 3 , 4 , 6–12 , and 14–16 ( Vyas et al . , 2014 ) . Much of the research on the PARP family has focused on its founding member , PARP-1 , which plays roles in transcription and DNA damage repair ( Pascal , 2018; Ray Chaudhuri and Nussenzweig , 2017 ) , but the molecular and cellular functions of the PARP family are considerably more diverse ( Gibson and Kraus , 2012; Vyas et al . , 2013 ) . Recent efforts to understand how ADP-ribosylation by other PARPs modulates cellular biology have identified roles in stress responses , cellular metabolism , and immune function ( Gupte et al . , 2017; Kim et al . , 2020; Luo and Kraus , 2011; Luo and Kraus , 2012; Ryu et al . , 2015 ) . Further understanding of the diversity of functions of PARP family members will require more information about the catalytic activities of these enzymes , as well as identification of their substrate proteins . In this study , we addressed these questions for PARP-7 through an examination of its catalytic activity and identification of its substrates . PARP-7 is a MART that localizes to both the cytoplasm and nucleus ( MacPherson et al . , 2013; Vyas et al . , 2014 ) . PARP-7 is also known as 2 , 3 , 7 , 8-tetrachlorodibenzo-p-dioxin ( TCDD ) -inducible PARP ( TiPARP ) because it is induced by TCDD as a target gene of the aryl hydrocarbon receptor ( AHR ) ( Ma et al . , 2001 ) . PARP-7 contains a carboxyl-terminal catalytic domain , as well as central region containing a WWE ( tryptophan-tryptophan-glutamate ) motif thought to bind PAR and a single CCCH-type zinc-finger ( Aravind , 2001; Schreiber et al . , 2006; Wang et al . , 2012 ) . Unlike true PARPs , which contain a signature H-Y-E motif in their catalytic domains , PARP-7 contains H-Y-I ( H532 , Y564 , and I631 ) , which underlies its MART activity ( MacPherson et al . , 2013; Vyas et al . , 2014 ) . The histidine and tyrosine residues of the H-Y-E catalytic triad are required for the binding of NAD+ in the ADP-ribosyl transferase catalytic site , whereas the glutamate residue is required for catalysis of PARylation activity ( Marsischky et al . , 1995; Papini et al . , 1989 ) . Little is known about the catalytic activity and substrates of PARP-7 , although previous studies have identified a few targets , including PARP-7 itself through automodification , as well as histones , AHR , and liver X receptors ( LXRs ) ( Bindesbøll et al . , 2016; Gomez et al . , 2018; Ma et al . , 2001; MacPherson et al . , 2013 ) . Gomez et al . , 2018 showed that the sites of PARP-7 automodification are resistant to meta-iodobenzylguanidine ( MIBG ) , an inhibitor of arginine-specific MARTs ( Loesberg et al . , 1990 ) , but are sensitive to iodoacetamide and hydroxylamine ( Zhang et al . , 2013 ) , implicating cysteines and acidic side chain residues ( e . g . glutamate and aspartate ) , respectively , as target residues for MARylation . More work is needed in this area , with higher throughput substrate identification . Although some progress has been made , the biology of PARP-7 is poorly characterized and limited in scope . PARP-7 is involved in a negative feedback loop regulating AHR responses ( MacPherson et al . , 2013 ) , but also plays a role in other cellular processes , including stem cell pluripotency , transcriptional regulation , mitotic spindle formation , viral replication , immune responses , and neuronal function ( Grimaldi et al . , 2019; Kozaki et al . , 2017; Roper et al . , 2014; Vyas et al . , 2013; Yamada et al . , 2016 ) . PARP-7 has recently emerged as a potentially interesting target for cancer therapy , in part because breast cancers have altered expression of PARP7 mRNA ( Cheng et al . , 2019 ) . In addition to studies in breast cancer , the PARP7 gene ( located at 3q25 ) was identified in a susceptibility locus for ovarian cancer in a genome-wide association study ( Goode et al . , 2010 ) . Determining its role in cellular signaling remains an important step in determining the utility of PARP-7 as a therapeutic target for breast and ovarian cancers . Here , we describe a chemical genetics approach for the identification of PARP-7 substrates based on a previously described NAD+ analog-sensitive PARP ( asPARP ) approach developed for PARPs 1 , 2 , and 3 ( Gibson and Kraus , 2017; Gibson et al . , 2016 ) . In this asPARP approach , NAD+ with an alkyne-containing R group at position 8 of the adenine ring works in concert with PARP proteins mutated at gatekeeper residues within their active site to facilitate ‘click’-able ADP-ribosylation of asPARP-specific protein targets . This approach is conceptually similar to one described by Carter-O'Connell et al . , 2014 in which analog sensitivity is conferred through an addition to the nicotinamide moiety , with click functionality conferred by an alkyne moiety added at position 6 of adenine . We used the asPARP approach described in Gibson et al . , 2016 in combination with mass spectrometry to determine the PARP-7 MARylated proteome , as well as complementary genomic , biochemical , cell-based , and biological analyses . Our studies have revealed an expansive set of PARP-7 protein substrates , including α-tubulin , which reduces microtubule stability when MARylated by PARP-7 . Recent studies linking PARP-7 to cancer ( Cheng et al . , 2019; Goode et al . , 2010 ) prompted us to examine the expression of PARP7 mRNA in human cancer samples . We mined the Genotype-Tissue Expression ( GTEx; normal ) and The Cancer Genome Atlas ( TCGA; cancer ) databases to quantify PARP7 mRNA expression across different normal and cancerous human tissues , including ovary , breast , pancreas , and kidney . Of these four tissues , PARP7 mRNA was expressed to the highest levels in ovary and the lowest levels in the pancreas ( Figure 1A , Figure 1—figure supplement 1A ) . The levels of PARP7 mRNA in cancers compared to the cognate noncancerous tissues varied; ovarian cancer tissues had decreased levels of PARP7 mRNA , while pancreatic cancers had elevated levels of PARP7 mRNA , when compared to noncancerous tissues ( Figure 1B ) . However , analysis at a more granular level using single-cell RNA-seq data ( Izar et al . , 2020; Wagner et al . , 2020 ) revealed that ( 1 ) the cell types in normal ovarian tissue are different than those in cancerous tissue and ( 2 ) malignant cells from ovarian cancer have a higher average PARP7 expression level than any of the normal ovarian cell types ( Figure 1—figure supplement 1B ) . The latter is supported by data from the Pan-Cancer Atlas in the TCGA database ( Hoadley et al . , 2018 ) showing a high frequency of PARP7 gene gains and amplifications ( Figure 1—figure supplement 1C ) . Given the difference in cell types and the possibility that ovarian cancers actually arise from cells in the Fallopian tubes ( Erickson et al . , 2013; Labidi-Galy et al . , 2017 ) , direct comparisons between PARP7 expression levels in normal and cancerous ovarian tissues are difficult . But , the available data suggest the possibility that gene amplifications drive elevated PARP7 expression in malignant cells in ovarian cancers , which may portend a dependence of ovarian cancers on PARP-7 . PARP-7 plays multiple essential roles in biological processes , such as transcription , RNA metabolism , translation in response to viral infections , and AHR activation ( Atasheva et al . , 2014; Bindesbøll et al . , 2016; Kozaki et al . , 2017; MacPherson et al . , 2013 ) . To determine the biological role of PARP-7 in ovarian cancers , we performed cell growth , migration , and invasion assays following PARP-7 depletion . siRNA-mediated knockdown of PARP7 , as determined by western blotting and RT-qPCR ( Figure 1C and D ) , resulted in a reduction of OVCAR4 cell growth , migration , and invasion ( Figure 1E-H ) . Similar results were observed in three other cell lines: OVCAR3 ( ovarian cancer ) , HeLa ( cervical cancer ) , and A704 ( kidney cancer ) whose PARP-7 levels are comparable to OVCAR4 cells . In all three cell lines and all three assays , PARP7 knockdown caused an inhibition of the cancer-related phenotypes ( Figure 1—figure supplement 2 ) , similar to what we observed in OVCAR4 cells . Thus , the observed effects of PARP-7 on cancer-related endpoints are not restricted to OVCAR4 cells . To determine how PARP-7 might regulate ovarian cancer phenotypes , such as cell growth and migration , we performed RNA-sequencing ( RNA-seq ) on OVCAR4 human ovarian cancer cells subjected to siRNA-mediated PARP7 knockdown ( Figure 2A ) . We used the gene expression patterns determined by RNA-seq as an indicator of the biological state of the cells . The RNA-seq analysis revealed statistically significant changes in 834 genes , with both increased and decreased expression observed ( Figure 2B and Supplementary file 1 ) . Gene ontology ( GO ) analyses revealed the enrichment of genes encoding proteins with roles in cell-cell adhesion , cell cycle arrest , apoptosis , and gene regulation ( Figure 2C ) . The results from the RNA-seq analysis are consistent with previously described roles of PARP-7 in viral responses and transcription ( Bindesbøll et al . , 2016; Kozaki et al . , 2017; MacPherson et al . , 2013 ) , while identifying additional roles for PARP-7 in cell cycle regulation and apoptosis . Multiple GO terms associated with reduced PARP-7 levels , such as cell cycle arrest and cell adhesion , are consistent with observed decreases in cell growth and motility ( Figure 1 ) . To understand how loss of PARP-7-mediated MARylation results in the altered cellular state and gene expression found following PARP-7 knockdown , we sought to identify the direct protein substrates of PARP-7 . For these studies , we focused on cytoplasmic substrates because PARP-7 is predominantly localized in the cytoplasm of OVCAR4 cells ( Figure 3A ) . To identify proteins MARylated by PARP-7 , we re-engineered a chemical genetics ( ‘bump-hole’ , ‘analog-sensitive’ ) approach that we previously developed for nuclear PARPs ( PARPs 1 , 2 , and 3 ) ( Gibson and Kraus , 2017; Gibson et al . , 2016 ) to work with the active site of PARP-7 . In the asPARP approach , a ‘gatekeeper’ amino acid in the NAD+ binding pocket of a PARP protein is mutated to a smaller residue to create a void ( ‘hole’ ) , which typically reduces the affinity of the PARP for NAD+ and , hence , its catalytic activity . Catalytic activity can be restored , however , by adding a bulky moiety onto NAD+ ( ‘bump’ ) , which fills the void and restores binding ( Figure 3B ) . In our asPARP-7 approach , we focused on two NAD+ analog-sensitive gatekeeper residues that we previously identified in PARP-1 ( Gibson and Kraus , 2017; Gibson et al . , 2016 ) . Multiple sequence alignment ( Figure 3C ) and structural analysis ( Figure 3D ) of the PARP-1 and PARP-7 ADP-ribosyltransferase domains revealed Phe547 and Ser563 in PARP-7 as homologous to the 1° and 2° gatekeeper mutations previously identified in PARP-1 . We generated a series of ten single or double PARP-7 mutants at these two residues , expressed them in insect cells , and purified them along with wild-type PARP-7 ( WT ) ( Figure 3E ) . We then tested the automodification activity of wild-type and mutant PARP-7 proteins in vitro by incubating purified protein and NAD+ , followed by western blotting with a MAR detection reagent ( Gibson et al . , 2017 ) . As expected , wild-type PARP-7 exhibited robust activity , whereas all the mutants exhibited reduced activity , with some showing little activity ( Figure 3F and G ) . Next , we tested the catalytic activity of the wild-type and mutant PARP-7 with a set of four NAD+ analogs suitable for copper-catalyzed azide-alkyne cycloaddition ( ‘click’ ) reactions ( Haldón et al . , 2015; Lutz and Zarafshani , 2008 ) . Each NAD+ analog has different R groups at position 8 of the adenine ring of NAD+ . 8-BuT-6-Parg-NAD+ also contains a modification at position 6 of the adenine ring . Two of the analogs have alkyne-containing R groups ( analogs 1 and 2 ) and two have azide-containing R groups ( analogs 3 and 4 ) ( Figure 4A and Supplementary file 2 ) to facilitate subsequent click chemistry for purification and fluorescent labeling ( Gibson and Kraus , 2017; Gibson et al . , 2016 ) of MARylated proteins . We assayed wild-type PARP-7 and the ten gatekeeper mutants for NAD+ analog sensitivity in an automodification reaction with water , NAD+ , and the four NAD+ analogs ( Figure 4B ) . The different PARP-7 mutants exhibited a range of basal activities ( water ) , which likely results from ADP-ribosylation during recombinant protein expression and purification . Automodification activity ( i . e . normalized activity ) was determined by subtraction of this basal activity for each mutant from the ADP-ribosylation signals observed upon incubation with NAD+ or the four NAD+ analogs ( Figure 4C ) . Three PARP-7 mutants , F547V , S563G and S563T , exhibited some activity with the NAD+ analogs ( Figure 4B and C ) . Comparison of the activity of the PARP-7 mutants with natural NAD+ ( Figure 3F and G ) and the four NAD+ analogs ( Figure 4B and C ) revealed that the mutants exhibiting greater activity with NAD+ also showed greater activity with NAD+ analogs . Although this screen yielded three PARP-7 mutants with NAD+ analog sensitivity , we chose to focus on the S563G mutant with 8-Bu ( 3-yne ) T-NAD+ for our subsequent studies . PARP-7 S563G , referred to from here forward as asPARP-7 , had the highest activity with 8-Bu ( 3-yne ) T-NAD+ , a well-characterized and readily available ‘click’-able NAD+ analog ( Gibson and Kraus , 2017; Gibson et al . , 2016 ) , as seen in an in-gel fluorescence-based assay for ‘click’-able autoMARylation ( Figure 5 ) . Extract-based PARP-specific substrate identification using the asPARP approach relies on poor usage of 8-Bu ( 3-yne ) T-NAD+ by endogenous PARPs compared to the asPARP protein . To determine whether addition of asPARP-7 to a complex cellular extract results in substantive incorporation of ‘click’-able ADP-ribosylation , we incubated OVCAR4 cytoplasmic extract with 8-Bu ( 3-yne ) T-NAD+ and purified recombinant wild-type or asPARP-7 protein . 8-Bu ( 3-yne ) T-ADP-ribose-labeled proteins were ‘clicked’ to an azido-TAMRA fluorophore and imaged using in-gel fluorescence ( Figure 6A ) . We found that addition of asPARP-7 , but not wild-type PARP-7 , resulted in strong and specific incorporation of ‘click’-able ADP-ribose onto OVCAR4 cytoplasmic extract proteins . To identify PARP-7 target proteins by mass spectrometry , we incubated cytoplasmic extracts from HeLa or OVCAR4 cells with 8-Bu ( 3-yne ) T-NAD+ and asPARP-7 , and then clicked the 8-Bu ( 3-yne ) T-ADP-ribose–labeled proteins to azide-agarose . Following extensive washing of the agarose beads , we performed trypsin-based identification of the ADP-ribosylated proteins by mass spectrometry ( Supplementary files 3 and 4 ) . We identified ~500 high confidence protein substrates of PARP-7 common to both OVCAR4 and HeLa cells , as well as an additional ~500 high confidence protein substrates of PARP-7 each unique to OVCAR4 and HeLa cells ( Figure 6B ) . GO analyses of the common PARP-7 substrates revealed functions in translation , rRNA processing , cytoskeleton organization , and cell proliferation , among others ( Figure 6C ) . These results provide important clues to the direct biological effects of PARP-7 . Our asPARP-7 plus mass spectrometry approach identified α-tubulin as a target for PARP-7-mediated MARylation . Our results are consistent with a recent report indicating that PARP-7 colocalizes with and MARylates α-tubulin in cells ( Grimaldi et al . , 2019 ) . We first confirmed that α-tubulin was MARylated in OVCAR4 cells by immunoprecipitating MARylated proteins and assaying for α-tubulin by western blotting ( Figure 7A ) . Next , we immunoprecipitated MARylated proteins from OVCAR4 cells subjected to either control or PARP7 knockdown and assayed by western blotting for the presence of α-tubulin , confirming that PARP-7 was required for MARylation of α-tubulin ( Figure 7A and B ) . Microtubule architecture , mediated in part by α-tubulin , plays a key role in cell migration and mitotic progression ( Boggs et al . , 2015; Pillai et al . , 2015; Piperno et al . , 1987 ) . Given the observed effects of PARP-7 depletion on mRNAs and proteins involved in cell cycle regulation and cytoskeleton organization ( Figures 2C and 6C ) , as well the MARylation of α-tubulin by PARP-7 , we surmised that the functional interplay between PARP-7 and α-tubulin might underlie microtubule dynamics . To test this directly , we assayed how OVCAR4 cells with or without PARP7 knockdown recovered from microtubule depolymerization by performing immunostaining for α-tubulin with visualization by confocal fluorescence microscopy . The cells were first incubated at a cold temperature , which destabilizes the microtubules , and then moved to 37°C to allow the regrowth of microtubules . Depletion of PARP-7 by knockdown of PARP7 mRNA stabilized the cold-destabilized α-tubulin-containing microtubule structures under these conditions ( Figure 7C ) . Similar results were observed in a parallel experiment in cells treated with Nocodazole , a drug that depolymerizes microtubules ( Pillai et al . , 2015; Figure 7C ) . Similar results were observed in three other cell lines: OVCAR3 , HeLa , and A704 ( Figure 7—figure supplement 1 ) , indicating that the effects are not restricted to ovarian cancer cells . The levels of α-tubulin MARylation and PARP-7 increase after treatment with cold or Nocodazole ( Figure 7D ) , linking PARP-7 levels , α-tubulin MARylation , and microtubule depolymerization . To connect the catalytic activity of PARP-7 more directly to the function of α-tubulin in cells , we used two approaches . First , we used a catalytically dead PARP-7 mutant ( Y564A ) , which we expressed using a Flag-tagged mouse Parp7 cDNA to produce an siRNA-resistant mRNA that could be used in knockdown-addback experiments ( the PARP7 siRNAs that we used were designed against the human PARP7 mRNA ) . Ectopic expression of the catalytically dead PARP-7 mutant resulted in a loss of α-tubulin MARylation ( Figure 7—figure supplement 2A ) . In addition , expression of the catalytically dead PARP-7 mutant after depletion of endogenous PARP-7 phenocopied PARP7 knockdown in the microtubule stability assay ( Figure 7—figure supplement 2B; compare to Figure 7C ) and the cell migration assay ( Figure 7—figure supplement 2 , C and D; compare to Figure 1F and G ) ( i . e . it was impaired versus wild-type PARP-7 in both cases ) . These results demonstrate that PARP-7 catalytic activity is required for the observed effects of PARP-7 . Second , we identified the sites of PARP-7-mediated MARylation on α-tubulin using our asPARP-7 approach , mutated them , and performed functional analyses with the MARylation site mutant . To do so , we eluted trypsin-digested , 8-Bu ( 3-yne ) T-ADP-ribose-labeled proteins from the azide-agarose beads described above using hydroxylamine and analyzed them by mass spectrometry . Hydroxylamine-cleaved ADPR modifications produce a 15 . 019 m/z shift identifying the specific site of glutamate or aspartate modification ( Zhang et al . , 2013 ) . Our analysis identified multiple sites of PARP-7-mediated MARylation on α-tubulin ( Supplementary file 5 ) . We focused on three residues in the critical amino-terminal nucleotide-binding domain ( i . e . D69 , E71 , E77 ) , all of which were contained in a single peptide that was identified in all four replicates ( two each for OVCAR4 and HeLa cells ) . We mutated all three of these residues to similar , but non-modifiable , residues ( D → N and E → Q ) and then expressed the mutant protein in OVCAR4 cells . Mutation of these sites impaired MARylation of α-tubulin ( Figure 7E ) , and ectopic expression of the α-tubulin mutant phenocopied PARP7 knockdown and expression of a catalytically dead PARP-7 mutant in the microtubule stability assay ( Figure 7F and G; compare to Figure 7C and Figure 7—figure supplement 2B ) . These results show that direct MARylation of α-tubulin is required for the observed effects of PARP-7 . Collectively , these results indicate that PARP-7 reduces cellular microtubule content following recovery from depolymerization , likely in part through direct MARylation of α-tubulin . Moreover , the PARP-7-mediated microtubule control may play a role in the regulation of cancer cell growth and motility . Information on the protein substrates of PARP-7 has been limited , although previous studies have identified a few targets , including PARP-7 itself through automodification , as well as histones , AHR , and LXRs ( Bindesbøll et al . , 2016; Gomez et al . , 2018; Ma et al . , 2001; MacPherson et al . , 2013 ) . To identify the MARylated substrates that underlie PARP-7-mediated ovarian cancer cell phenotypes , we developed a chemical genetics approach for NAD+ analog-sensitivity comprising an NAD+ binding pocket PARP-7 mutant ( S563G ) paired with the NAD+ analog 8-Bu ( 3-yne ) T-NAD+ ( Figures 3–5 ) . We used this approach coupled with mass spectrometry to identify an extensive PARP-7 ADP-ribosylated proteome in OVCAR4 and HeLa cells , including cell-cell adhesion and cytoskeleton organization proteins ( Figure 6C ) . Interestingly , similar gene ontologies were enriched in our gene expression experiments in OVCAR4 cells after knockdown of PARP7 ( Figure 2C ) , suggesting a coordination of the regulatory functions of PARP-7 . One PARP-7 substrate that we identified , α-tubulin , has links to cancer , playing a role in cell migration and mitotic progression ( Boggs et al . , 2015; Pillai et al . , 2015; Piperno et al . , 1987 ) . A previous study demonstrated that MARylation of α-tubulin by PARP-7 supports proper organization of the mouse neuronal cortex , including the correct distribution and number of GABAergic neurons ( Grimaldi et al . , 2019 ) . In our studies , we observed that PARP-7 reduces cellular microtubule content following recovery from depolymerization , likely in part through direct MARylation of α-tubulin ( Figure 7 ) . This may result from changes in the polymerization rate , tubulin dimer stability , microtubule stability , etc . PARP-7-mediated microtubule control is likely to underlie the regulation of cell growth and motility that we observed in ovarian cancer cells ( Figure 1C-H ) . Why might PARP-7 MARylate α-tubulin to destabilize microtubules in ovarian cancer cells ? How does this benefit the cancer cells ? One possibility is that the ability to quickly depolymerize and disassemble microtubules is important for efficient cell proliferation and migration . Indeed , there is evidence for this in the literature . Taxanes and related molecules comprise a class of anticancer drugs that stabilize microtubules , leading to the arrest of proliferation and mitosis . Moreover , previous studies have shown that well-characterized tumor suppressors , such as RASSF1A and APC , act to stabilize microtubule polymerization ( Liu et al . , 2003; van Es et al . , 2001 ) . These observations provide a plausible explanation for the observed effects of PARP-7 on α-tubulin and microtubules in our assays . Rodriguez et al . ( co-submitted ) did not identify α-tubulin as a PARP-7 substrate in their analysis , but they did identify tubulin-specific chaperone E , a key chaperone for α-tubulin ( Tian and Cowan , 2013 ) , perhaps indicating a broader role for PARP-7 in regulating microtubule formation . Together , our results demonstrate a functional link between PARP-7 , its catalytic activity toward a specific substrate ( i . e . α-tubulin ) , a defined cellular process ( i . e . microtubule control ) , and a broader biological outcome ( i . e . cancer-related phenotypes ) . Among the PARP-7 substrates that we identified in our analyses , we found other PARP family members , including PARP-1 ( both replicates from OVCAR4 and HeLa cells ) , PARP-4 ( one replicate from OVCAR4 cells and both replicates from HeLa cells ) , and PARP-2 ( one replicate from HeLa cells ) ( Supplementary file 3 ) . PARPs 1 and 2 are nuclear proteins with functions in DNA repair and transcription ( Gupte et al . , 2017; Ryu et al . , 2015 ) , whereas PARP-4 ( a . k . a . vault PARP ) is a component of enigmatic cell structures known as vault particles ( Kickhoefer et al . , 1999 ) . In previous studies of PARPs 1 , 2 , and 3 , we observed extensive crosstalk between them , each acting as a substrate for the other ( Gibson et al . , 2016 ) . Such crosstalk may be a universal feature of PARPs . In this regard , Rodriguez et al . ( co-submitted ) identified PARP-13 as a substrate of PARP-7 using a similar chemical genetics approach , which also included MARylation site identifications . Their analysis identified cysteine as a major ADPR acceptor for PARP-7 , consistent with observations by Gomez et al . implicating cysteines and acidic residues as targets for MARylation ( Gomez et al . , 2018 ) . TIPARP ( the gene encoding PARP-7 , located at 3q25 ) was identified in a susceptibility locus for ovarian cancer in a recent genome-wide association study ( Goode et al . , 2010 ) . We have recently shown that ADP-ribosylation levels and patterns correlate with gene expression and clinical outcomes in ovarian cancers ( Conrad et al . , 2020 ) . In addition , we and others have observed that PARP7 mRNA expression is lower in ovarian cancers compared to normal ovarian epithelium , both in patient samples ( Figure 1B ) and cell lines ( Goode et al . , 2010 ) . However , a more detailed analysis using single cell RNA-seq data ( Izar et al . , 2020; Wagner et al . , 2020 ) revealed that malignant cells from ovarian cancer have a higher average PARP7 expression level than any of the normal ovarian cell types ( Figure 1—figure supplement 1B ) . The latter is supported by data from the Pan-Cancer Atlas in the TCGA database ( Hoadley et al . , 2018 ) showing a high frequency of PARP7 gene gains and amplifications ( Figure 1—figure supplement 1C ) . Although direct comparisons between PARP7 expression level in normal and cancerous ovarian tissues are difficult , the available data suggest the possibility that gene amplifications drive elevated PARP7 expression in malignant cells in ovarian cancers . In breast cancers , PARP7 mRNA expression is lower in tumor tissues compared to normal tissues , and higher PARP7 mRNA is associated with better survival outcomes ( Cheng et al . , 2019; Zhang et al . , 2020 ) . Moreover , in breast and colon cancer cells , PARP-7 acts to suppress multiple oncogenic transcription factors , including HIF-1α , to reduce tumorigenesis ( Zhang et al . , 2020 ) . Knockdown of PARP7 in breast and colon cancer xenografts promotes enhanced tumor formation ( Zhang et al . , 2020 ) . These observations stand in contrast to our results with ovarian cancer cells , in which PARP7 knockdown resulted in reduced cell growth , migration , and invasion ( Figure 1C-H ) , as well as increased microtubule content , which may reduce cancer-related outcomes . Thus , PARP-7 may have context-specific effects in cancer cells . In ovarian cancers and other cancers that respond similarly to PARP7 knockdown , inhibition of PARP-7 catalytic activity with small molecules would be expected to have positive therapeutic effects . In this regard , the first PARP-7 inhibitor ( RBN-2397 ) is now in Phase I clinical trials for solid tumors ( ClinicalTrials . gov identifier: NCT04053673 ) . In contrast to the current FDA-approved PARP1/2 inhibitors , RBN-2397 is the first inhibitor of a MART , representing a previously unexplored therapeutic target . Finally , treatment with carboplatin/paclitaxel is the primary initial treatment regimen for the vast majority of ovarian cancers ( Boyd and Muggia , 2018; Kampan et al . , 2015 ) . Interestingly , paclitaxel is a microtubule-stabilizing drug that prevents mitosis ( Orr et al . , 2003; Weaver , 2014 ) . Although paclitaxel binds specifically to β-tubulin ( Orr et al . , 2003 ) , overexpression of α-tubulin can increase the resistance of cancer cells to paclitaxel ( Han et al . , 2000 ) . These observations , viewed in light of our results , suggest an intriguing link between PARP-7 , α-tubulin , microtubules , and the sensitivity of cancer cells to paclitaxel . The custom recombinant antibody-like MAR binding reagent ( anti-MAR ) was generated and purified in-house ( now available from Millipore Sigma , MABE1076; RRID:AB_2665469 ) ( Gibson et al . , 2017 ) . The antibodies used were as follows: PARP-7 ( Invitrogen , PA5-40774; RRID:AB_2607074 ) , α-tubulin ( Santa Cruz , sc-8035; RRID:AB_628408 ) , β-tubulin ( Abcam , ab6046; RRID:AB_2210370 ) , SNRP70 ( Abcam , ab51266; RRID:AB_10673827 ) , Flag ( Sigma-Aldrich , F3165; RRID:AB_259529 ) , HA ( Sigma-Aldrich , H3663; RRID:AB_262051 ) , rabbit IgG fraction ( Invitrogen , 10500C; RRID:AB_2532981 ) , goat anti-rabbit HRP-conjugated IgG ( Pierce , 31460; RRID:AB_228341 ) , goat anti-mouse HRP-conjugated IgG ( Pierce , 31430; RRID:AB_228307 ) , and Alexa Fluor 488 goat anti-mouse IgG ( ThermoFisher , A-11001; RRID:AB_2534069 ) . OVCAR4 , OVCAR3 , 293T , and HeLa cells were purchased from the American Type Cell Culture ( ATCC ) . A704 cells were obtained from Dr . Laura Banaszynski . OVCAR4 and OVCAR3 cells were maintained in RPMI ( Sigma-Aldrich , R8758 ) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin . HeLa , 293T and A704 cells were maintained in DMEM ( Sigma-Aldrich , D5796 ) supplemented with 10% fetal bovine serum and 1% penicillin/streptomycin . Fresh cell stocks were regularly replenished from the original ATCC-verified stocks and confirmed as mycoplasma-free every three months using a commercial testing kit . The siRNAs targeting PARP-7 were from Dharmacon ( see below ) and the control siRNA ( SIC001 ) was from Sigma . The siRNA oligos were transfected into OVCAR4 cells at a final concentration of 30 nM using Lipofectamine RNAiMAX reagent ( Invitrogen , 13778150 ) according to the manufacturer’s instructions . The cells were used for various assays 24 hr after siRNA transfection . The siRNA oligos used to knockdown PARP7 mRNA were as follows: OVCAR4 cells were cultured as described above before the preparation of whole cell lysates , nuclear extracts , and cytoplasmic extracts . Cells were transfected with control or PARP7-specific siRNAs at a final concentration of 30 nM in 10 cm diameter culture dishes . Twenty-four hours later , the cells were collected and plated at a density of 20 , 000 cells per well in a 24-well plate . The cells were fixed with 4% paraformaldehyde at each of the indicated time points and washed with water . The plates were stored at 4°C until the end of the time course when all samples could be analyzed simultaneously . After all samples were collected , the fixed cells were stained with crystal violet ( 0 . 5% crystal violet in 20% methanol ) for 30 min with gentle shaking at room temperature . The stained cells were washed several times with water and air dried . The crystal violet in each well was re-dissolved in 10% acetic acid and the absorbance at 570 nm was read using a spectrophotometer . The absorbance of a blank well was subtracted from the samples and the values were normalized to the values at day 1 . Three independent experiments were performed with independent biological replicates to ensure reproducibility . Statistical differences were determined using Student’s t-tests at each time point . Boyden chamber assays were used to determine the migration and invasive capacity of cells as described below . The cells cultured in six-well plates were transfected with 30 nM control siRNA or two different siRNAs targeting PARP7 mRNA . Twenty-four hours later , the cells were trypsinized and were seeded at a density of 100 , 000 cells into migration chambers ( Corning , 353097 ) or invasion chambers ( Corning , 354480 ) following the manufacturer’s protocols . Briefly , the trypsinized cells were resuspended in serum-free RPMI media and collected by centrifugation at 300 x g for 3 min at room temperature . The cell pellets were resuspended in serum-free RPMI media and 500 μL of the media containing 100 , 000 cells were plated into the top chamber . The chambers were then incubated in 750 μL of RPMI media with 10% FBS . For rescue experiments in OVCAR4 cells expressing wild-type or catalytically dead mutant PARP-7 , the cells were plated as above and then treated with 1 µg/mL doxycycline ( Dox ) . After 24 hr of incubation at 37°C , the cells in the top chamber were scraped and removed . The chambers were stained with crystal violet ( 0 . 5% crystal violet in 20% methanol ) for 30 min with gentle shaking . The chambers were washed with water , air-dried and the images of cells at the bottom of the membrane were collected using an upright microscope . Three independent biological replicates were performed for each condition . Statistical differences were determined using Student’s t-test . OVCAR4 cells were plated in 6-well plates and transfected with 30 nM of PARP7 or control siRNAs as described above . Total RNA was isolated using the Qiagen RNAeasy Plus Mini kit ( Qiagen , 74136 ) according to the manufacturer's protocol . Total RNA was reverse transcribed using oligo ( dT ) primers and MMLV reverse transcriptase ( Promega , PR-M1705 ) to generate cDNA . The cDNA samples were subjected to RT-qPCR using gene-specific primers , as described below . Target gene expression was normalized to the expression of RPL19 mRNA . All experiments were repeated a minimum of three times with independent biological samples to ensure reproducibility and a statistical significance of at least p<0 . 05 . Statistical differences between control and experimental samples were determined using Student’s t-test . The primers used for RT-qPCR were as follows: Multiple sequence alignment of PARP-7 ( Q7Z3E1 ) , PARP-1 ( P09874 ) , and PARP-12 ( Q9H0J9 ) was performed using Clustal Omega ( Madeira et al . , 2019 ) . The catalytic domains of all three PARPs , especially the two examined homologous gatekeeper residues , are highly conserved . Available structures of PARP-1 ( PDBID: 3PAX ) and PARP-12 ( PDBID: 2PQF ) were downloaded from the RCSB Protein Data Bank for analysis . The molecular graphic was generated using PyMOL software ( PyMOL Molecular Graphics System , Version 2 . 3 . 2 ) ( Schrodinger LLC , 2010 ) . A close-up view of residues F547 and S563 in PARP-7 were illustrated by comparison of the available structures of PARP-1 and PARP-12 . The NAD+ analogs used in this study were either purchased from , or synthesized in collaboration with , the BIOLOG Life Science Institute , Bremen , Germany ( Gibson et al . , 2016 ) . All the NAD+ analogs are listed in Supplementary file 2 . PARP-7 automodification in the in vitro assays was determined by western blotting as described above . Immunoblot signals were detected by ECL and quantified using Image Lab 6 . 0 ( Bio-Rad ) . For determination of relative PARP-7 activity with NAD+ or each NAD+ analog , the signal from a ‘water only control’ run simultaneously with the experimental conditions was subtracted from itself and each PARP-7 automodification signal ( wild-type or mutant , NAD+ or NAD+ analog ) . The signal from wild-type PARP-7 with NAD+ was then set to 100 , The water control had a value of 0 because of the subtraction of the background signal from itself . Any negative values after the water background subtraction ( i . e . less than background ) were set to zero . The signals from the wild-type or mutant PARP-7 proteins with NAD+ or the NAD+ analogs were then expressed relative to the background-corrected signal from wild-type PARP-7 and NAD+ . This yielded values that ranged from 0 to more than 100 , with the highest values coming from pairings between PARP-7 mutants and NAD+ analogs that yielded more PARP-7 automodification than wild-type PARP-7 with NAD+ . The peptides identified from the mass spectrometry experiments were used for further analyses and the data were expressed in various formats , as described below . Immunoprecipitation ( IP ) of MARylated proteins from OVCAR4 cells was performed as follows . OVCAR4 cells plated on 15 cm diameter cell culture dishes were transfected with 30 nM control siRNA or two different siRNAs targeting PARP-7 . Two 15 cm dishes of cells were used for each siRNA transfection . Forty-eight hours after transfection , the cells were collected , washed twice with ice-cold PBS , and resuspended in IP Lysis Buffer ( 50 mM Tris-HCl pH7 . 5 , 0 . 5 M NaCl , 1 . 0 mM EDTA , 1% NP-40 and 10% glycerol , freshly supplemented with 1 mM DTT , 250 nM ADP-HPD , 10 μM PJ34 , 1x complete protease inhibitor cocktail ) , and incubated at 4°C for 30 min with gentle shaking . The cell debris was cleared by centrifugation for 15 min at 4°C at 15 , 000 g . The supernatants were collected and the protein concentrations were measured using a Bradford assays . Aliquots of cell lysates containing equal amount of protein were used for each IP condition . Five percent of each cell lysate was saved for input . The cell lysates for IP were incubated with 3 µg of MAR detection reagent or rabbit IgG ( Thermo Fisher Scientific , 10500C ) , and protein A agarose beads overnight at 4°C with gentle rotation . The beads were then washed five times with Lysis Buffer for 5 min each at 4°C with gentle mixing . The beads were then heated to 100°C for 10 min in 1x SDS-PAGE loading buffer to release the bound proteins . The immunoprecipitated material was subjected to western blotting as described above . The plasmid for expression of C-terminal HA-tagged α-tubulin was obtained from Sino Biologicals ( HG14201-CY ) . Mutations of the sites of MARylation ( i . e . D69 , E71 and E77 ) were introduced into the pCMV3-C-HA-TUBA3C plasmid using a protocol adapted from the Quickchange site-directed mutagenesis kit ( Agilent ) . D69N and E71Q were added first , then E77Q to the double mutant to generate the triple mutant using the primers listed below . Primers used for generatingα-tubulin MARylation site mutants: Immunoprecipitation ( IP ) of HA-tagged α-tubulin from OVCAR4 or 293 T cells was performed as follows: For Nocodazole or cold treatments , OVCAR4 cells were transiently transfected with an expression vector for HA-tagged α-tubulin . Forty-eight hours after transfection , the cells were either placed on ice for 45 min or treated with 6 μM nocodazole for 1 hr and then washed with warm medium three times . The cells were allowed to recover by incubation in normal growth medium at 37°C for 15 min , and then collected and extracted as described above . For the induction of PARP-7 expression in OVCAR4 cells , cells expressing wild-type or catalytically dead mutant PARP-7 were transiently transfected with HA-tagged α-tubulin and simultaneously treated with 1 µg/mL Dox . For expression of PARP-7 in 293 T cells , the cells were transiently transfected with Flag-tagged PARP-7 and wild-type or mutant α-tubulin constructs . Forty-eight hours after transfection , the cells were collected as described above . Cell lysates for IP were incubated with 1 . 5 µg of mouse monoclonal antibody against HA-tag and protein G agarose beads overnight at 4°C with gentle rotation . The beads were then washed three times with Lysis Buffer for 5 min each at 4°C with gentle mixing . The beads were then heated to 100°C for 5 min in 1x SDS-PAGE loading buffer to release the bound proteins . The immunoprecipitated material was subjected to western blotting as described above . The Parp7 cDNA was prepared by extracting total RNA from 3T3-L1 cells using Trizol ( Invitrogen , 15596026 ) , followed by reverse transcription using Superscript III reverse transcriptase ( Invitrogen , 18080051 ) and an oligo ( dT ) primer , according to the manufacturer’s instructions . The Parp7 cDNA was then amplified from the cDNA library with primers listed below and cloned into pCDNA3 using the primers listed below . A cDNA for the catalytically dead PARP-7 mutant ( Y564A ) was generated by site-directed mutagenesis using Pfu Turbo DNA polymerase ( Agilent , 600250 ) using the primers listed below . The PCR products were then amplified with primers encoding an N-terminal FLAG epitope tag and cloned into the pINDUCER20 lentiviral Dox-inducible expression vector ( Addgene , 44012 ) . Primers used for amplifying Parp7 cDNA: Primers used for cloning Parp7 cDNA into pCDNA3 vector: Primers used for cloning Parp7 cDNA into pINDUCER20 vector: Primers used for generating Parp7 catalytic dead mutant , Y564A: Cells were transduced with lentiviruses for Dox-inducible ectopic expression of PARP-7 ( wild-type and mutant ) . We generated lentiviruses by transfection of the pINDUCER20 constructs described above , together with: ( i ) an expression vector for the VSV-G envelope protein ( pCMV-VSV-G , Addgene 8454 ) , ( ii ) an expression vector for GAG-Pol-Rev ( psPAX2 , 12260 ) , and ( iii ) a vector to aid with translation initiation ( pAdVAntage , Promega ) into 293 T cells using Lipofectamine 3000 reagent ( Invitrogen , L3000015 ) according to the manufacturer’s instructions . The resulting viruses were collected in the culture medium , concentrated by using a Lenti-X concentrator ( Clontech , 631231 ) , and used to infect OVCAR4 cells . Stably transduced cells were selected with G418 sulfate ( Sigma , A1720; 1 mg/ml ) . The cells were treated with 1 µg/mL Dox for 24 hr to induce protein expression . Inducible ectopic expression of the cognate proteins was confirmed by western blotting . OVCAR4 cells grown in six-well plates were transfected with 30 nM control siRNA or two different siRNAs targeting PARP-7 . Twenty-four hours after transfection , the cells were trypsinized and seeded into eight-well chambered slides ( Thermo Fisher , 154534 ) . For rescue experiments in OVCAR4 cells expressing wild-type or catalytically dead mutant PARP-7 , the cells were plated as above and then treated with 1 µg/mL Dox . Twenty-four hours later , the cells were either placed on ice for 45 min or treated with 6 μM nocodazole for 1 hr and then washed with warm medium . The cells were allowed to recover by incubation in normal growth medium at 37°C for 15 min and were then washed twice with PBS , fixed with 4% paraformaldehyde for 15 min at room temperature , and washed three times with PBS . The cells were permeabilized for 5 min using Permeabilization Buffer ( PBS containing 0 . 01% Triton X-100 ) , washed three times with PBS , and incubated for 1 hr at room temperature in Blocking Solution ( PBS containing 1% BSA , 10% FBS , 0 . 3 M glycine and 0 . 1% Tween-20 ) . The fixed cells were incubated with α-tubulin antibody ( 1:1000 Santa Cruz , sc-8035 ) or HA antibody ( 1:200 Abcam , ab9110 ) in PBS overnight at 4°C . After washing three times with PBS , the cells were incubated with Alexa Fluor 488 goat anti-mouse IgG ( 1:500 , ThermoFisher , A-11001 ) in PBS for 1 hr at room temperature . After washing three times with PBS , coverslips were mounted with the VectaShield Antifade Mounting Medium with DAPI ( Vector Laboratories , H-1200 ) . All images were acquired using an inverted Zeiss LSM 780 confocal microscope and the fluorescence intensities were measured using Fiji ImageJ software .	Cancer is a complex illness where changes inside healthy cells causes them to grow and reproduce rapidly . Specialized proteins called enzymes – which regulate chemical reactions in the cell – often help cancer develop and spread through the body . One such enzyme called PARP-7 labels other proteins by attaching a chemical group which changes their behavior . However , it was unknown which proteins PARP-7 modifies and how this tag alters the actions of these proteins . To investigate this , Parsons , Challa , Gibson et al . developed a method to find and identify the proteins labelled by PARP-7 in ovarian cancer cells taken from patients and cultured in the laboratory . This revealed that PARP-7 labels hundreds of different proteins , including adhesion proteins which affect the connections between cells and cytoskeletal proteins which regulate a cell’s shape and how it moves . One of the cytoskeletal proteins modified by PARP-7 is α-tubulin , which joins together with other tubulins to form long , tube-like structures known as microtubules . Parsons et al . found that when α-tubulin is labelled by PARP-7 , it creates unstable microtubules that alter how the cancer cells grow and move . They discovered that depleting PARP-7 or mutating the sites where it modifies α-tubulin increased the stability of microtubules and slowed the growth of ovarian cancer cells . Ovarian cancer is the fifth leading cause of cancer-related deaths among women in the United States . A new drug which suppresses the activity of PARP-7 has recently been developed , and this drug could potentially be used to treat ovarian cancer patients with high levels of PARP-7 . Clinical trials are ongoing to see how this drug affects the behavior of cancer cells in patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cancer", "biology" ]	2021	Identification of PARP-7 substrates reveals a role for MARylation in microtubule control in ovarian cancer cells
Genetic drift is expected to remove polymorphism from populations over long periods of time , with the rate of polymorphism loss being accelerated when species experience strong reductions in population size . Adaptive forces that maintain genetic variation in populations , or balancing selection , might counteract this process . To understand the extent to which natural selection can drive the retention of genetic diversity , we document genomic variability after two parallel species-wide bottlenecks in the genus Capsella . We find that ancestral variation preferentially persists at immunity related loci , and that the same collection of alleles has been maintained in different lineages that have been separated for several million years . By reconstructing the evolution of the disease-related locus MLO2b , we find that divergence between ancient haplotypes can be obscured by referenced based re-sequencing methods , and that trans-specific alleles can encode substantially diverged protein sequences . Our data point to long-term balancing selection as an important factor shaping the genetics of immune systems in plants and as the predominant driver of genomic variability after a population bottleneck . Balancing selection describes the suite of adaptive forces that maintain genetic variation for longer than expected by random chance . It can have many causes , including heterozygous advantage , negative frequency-dependent selection , and environmental heterogeneity in space and time . The unifying characteristic of these situations is that the turnover of alleles is slowed , resulting in increased diversity at linked sites ( Charlesworth , 2006 ) . In principle , it should be simple to detect the resulting footprints of increased coalescence times surrounding balanced sites ( Tellier et al . , 2014 ) , and many candidates have been identified using diverse methodology ( Fijarczyk and Babik , 2015 ) . However , balanced alleles will be stochastically lost over long time spans , suggesting that most balanced polymorphism is short lived ( Fijarczyk and Babik , 2015 ) . The strongest evidence for balancing selection comes from systems in which alleles are maintained in lineages that are reproductively isolated and that have separated millions of years ago , resulting in trans-specific alleles with diagnostic trans-specific single nucleotide polymorphisms ( tsSNPs ) . A few , well known genes fit this paradigm: the self-incompatibility loci of plants ( Vekemans and Slatkin , 1994 ) , mating-type loci of fungi ( Wu et al . , 1998 ) , and the major histocompatibility complex ( MHC ) and ABO blood group loci in vertebrates ( McConnell et al . , 1988; Mayer et al . , 1988; Lawlor et al . , 1988; Watkins et al . , 1990; Ségurel et al . , 2012 ) . Additional candidates have been proposed by comparing genome sequences from populations of humans and chimpanzees , and from populations of multiple Arabidopsis species . These efforts have revealed six loci in primates ( Leffler et al . , 2013b; Teixeira et al . , 2015 ) and up to 129 loci , that were identified by at least two shared SNPs each , in Arabidopsis ( Novikova et al . , 2016; Bechsgaard et al . , 2017 ) , as potential targets of long-term balancing selection and/or introgression . In both systems , genes involved in host–pathogen interactions were enriched , which in Arabidopsis is consistent with previous findings that several disease resistance loci appear to be under balancing selection in this species , based on the analysis of individual genes ( Huard-Chauveau et al . , 2013; Botella , 1998; Caicedo et al . , 1999; Noel , 1999; Stahl et al . , 1999; Tian et al . , 2002; Bakker , 2006; Rose et al . , 2004; Todesco et al . , 2010 ) . However , even with the ability to conduct whole-genome scans for balancing selection in A . thaliana , the total number of examples with robust evidence across species remains small ( Cao et al . , 2011; 1001 Genomes Consortium , 2016 ) . One explanation for this paucity of evidence for pervasive and stable balancing selection is that cases of long-term maintenance of alleles are rare . However , there are good reasons to believe that many studies lacked the power to detect the expected effects ( Fijarczyk and Babik , 2015; DeGiorgio et al . , 2014 ) . If one requires that alleles have been maintained in species separated by millions of years , then only targets of outstandingly strong selective pressures that remain the same over many millennia can be identified . Furthermore , recombination between deeply coalescing alleles will typically reduce the size of the genomic footprint to very short sequence stretches , thus limiting the opportunity for distinguishing old alleles from recurrent mutations . We hypothesised that self-fertilizing species provide increased sensitivity to detect balancing selection based on two observations ( Wiuf et al . , 2004; Wright et al . , 2008 ) . First , self fertilisation greatly reduces the effective rate of recombination , thus potentially expanding the footprint of balancing selection . In addition , the transition to self fertilisation is generally associated with dramatic genome-wide reductions in polymorphism , potentially making it easier to detect outlier loci that retain variation from the outcrossing , more polymorphic ancestor . In this study we sought to assess how strongly selection acts to maintain genetic diversity in the context of repeated transitions to self fertilisation in the flowering plant genus Capsella . Like many plant lineages , the ancestral state of Capsella is outcrossing ( found in the extant diploid species C . grandiflora ) , but selfing has evolved independently in two diploid species , C . rubella and C . orientalis ( Figure 1A ) ( Foxe et al . , 2009; Guo et al . , 2009; Bachmann et al . , 2018 ) . The genomes of both species exhibit the drastic loss of genetic diversity typical for many selfers ( Figure 1B–C ) ( Guo et al . , 2009; Foxe et al . , 2009; St Onge et al . , 2011; Slotte et al . , 2013; Brandvain et al . , 2013; Slotte et al . , 2012 ) . In the younger species , C . rubella , loss of genetic diversity was initially thought to have occurred uniformly throughout the entire genome ( Foxe et al . , 2009; Guo et al . , 2009 ) , but subsequent reports already hinted at some loci having increased diversity ( Gos et al . , 2012; Brandvain et al . , 2013 ) , motivating the present study . The species Capsella rubella is young , only 30 , 000 to 200 , 000 years old , and was apparently founded when a small number of C . grandiflora individuals became self-compatible ( Foxe et al . , 2009; Guo et al . , 2009 ) . Previous studies had hinted at unequal retention of C . grandiflora alleles across the C . rubella genome ( Gos et al . , 2012; Brandvain et al . , 2013 ) , leading us to analyse this phenomenon systematically by comparing the genomes of 50 C . rubella and 13 C . grandiflora accessions from throughout each species’ range ( Figure 1—figure supplement 1 and Figure 1—source data 1 ) . Because the calling of trans-specific SNPs ( tsSNPs ) is particularly sensitive to mismapping errors in repetitive sequences , we applied a set of stringent filters , resulting in 74% of the C . rubella reference genome remaining accessible to base calling in both species , with almost half ( 47% ) of the masked sites in the repeat rich pericentromeric regions . After filtering , there were 5 , 784 , 607 SNPs and 883 , 837 indels . Unless otherwise stated , all subsequent analyses were performed using SNPs . Of these , only 27 , 852 were fixed between the two species , whereas 824 , 540 were found in both species ( tsCgCrSNPs ) , consistent with the expected sharing of variation between the two species . In addition , 4 , 291 , 959 SNPs segregated only in C . grandiflora ( species-specific SNPs; ssCgSNPs ) , and 640 , 256 only in C . rubella ( ssCrSNPs ) . Sample rarefaction by subsampling our sequenced accessions indicated that common ssCrSNP and tsCgCrSNP discovery was near saturation in our experiment , though additional sampling will continue to uncover rare alleles ( Figure 1D ) . The consequences of selfing are easily seen as a dramatic reduction in genetic diversity in C . rubella ( Figure 1—source data 2 ) , consistent with the previously suggested genetic bottleneck ( Foxe et al . , 2009; Guo et al . , 2009 ) . As expected from a predominantly selfing species , SNPs segregating in C . rubella were much less likely to be heterozygous than those segregating in C . grandiflora , though evidence for occasional outcrossing in C . rubella is observed in the form of a variable number of heterozygous calls ( Figure 1E ) . Selfing is also expected to reduce the effective rate of recombination between segregating polymorphisms . Linkage disequilibrium ( LD ) decayed , on average , to 0 . 1 within 5 kb in C . grandiflora , while it only reached this value at distances greater than 20 kb in C . rubella ( Figure 1F ) . Though C . rubella is a relatively young species , it exhibits characteristics typical of a predominantly ( but not exclusively ) self-fertilising species: reduced genetic diversity , reduced observed heterozygosity , and reduced effective recombination rate . This last effect could potentially increase the visibility of signals for balancing selection from linked sites ( Wiuf et al . , 2004 ) . The degree of trans-specific allele sharing depends upon the level of gene flow between species , the age of the speciation event , and the demographic history of each resultant species . We first sought to understand how these neutral processes have affected extant polymorphism in C . grandiflora and C . rubella . We searched for evidence of population structure in our dataset by fitting individual ancestries to different numbers of genetic clusters with ADMIXTURE ( Alexander et al . , 2009 ) ( Figure 2A and Figure 2—figure supplement 1A-B; k-values from 1 to 6 ) . The best fit as determined by the minimum cross-validation error was three clusters , with one including all C . grandiflora individuals , and C . rubella samples split into two clusters . Principal component ( PC ) analysis ( Price et al . , 2006 ) of genetic variation revealed a similar picture , with PC1 separating the two species and PC2 separating the C . rubella samples ( Figure 2A ) . C . rubella population structure was strongly associated with geography . Samples from western Europe and southeastern Greece were unambiguously assigned to separate groups , while samples from northern and western Greece , near the presumed site of speciation in the current range of C . grandiflora ( Hurka and Neuffer , 1997 ) , showed mixed ancestry ( or intermediate assignment to these groups , Figure 2A–B ) . A single C . rubella sample from western Europe showed some mixed ancestry . This sample was collected near Gargano National Park on the eastern coast of Italy . The source of its mixed ancestry is unclear , but its proximity to Greece suggests that it may result from ongoing migration across the Adriatic Sea . The general pattern of population structure is consistent with the centre of diversity for C . rubella being in northern Greece and a more recent rapid expansion into Western Europe , and agrees with predictions made based on previous , smaller datasets ( Brandvain et al . , 2013 ) . The observed structure is principally organised by a major geographic barrier , the Adriatic Sea . We therefore separated our samples into into two distinct groups to the west ( W ) and east ( E ) of the Adriatic Sea for subsequent analyses . Because their current ranges overlap , ongoing gene flow between sympatric C . rubella and C . grandiflora could be a potentially important source of allele sharing between the two species . While a previous study had not found any evidence for such a scenario ( Brandvain et al . , 2013 ) , one of our C . grandiflora samples was assigned partial ancestry to the otherwise C . rubella-specific clusters , and resided at an intermediate position along PC1 ( Figure 2A ) . Furthermore , eastern C . rubella individuals , many of which grew in sympatry with C . grandiflora , were less differentiated from C . grandiflora compared to western C . rubella samples along PC1 ( Figure 2A and Figure 2—figure supplement 1C-D ) . Gene flow between eastern C . rubella and C . grandiflora was supported by significant genome-wide D-statistics for C . rubella samples from the C . grandiflora range ( ABBA-BABA test; comparing each E individual with the W population ) ( Green et al . , 2010; Durand et al . , 2011 ) , with D decreasing as a function of distance from the centre of C . grandiflora’s range ( Figure 2—figure supplement 1 and Figure 2—source data 1 ) . Because D statistics can be sensitive to ancient population structure ( Durand et al . , 2011 ) , we further relied on identity-by-descent ( IBD ) segments as detected by BEAGLE ( Browning and Browning , 2013 ) to identify genomic regions of more recent co-ancestry across these species . The proportion of the genome shared in IBD segments between C . rubella and C . grandiflora also decreased as a function of distance between samples , and the strongest evidence for recent ancestry was found between C . grandiflora individuals and sympatric northern Greek C . rubella lines ( Figure 2C–D ) . These results indicate that gene flow is ongoing between the species , consistent with interspecific crosses often producing fertile offspring , specifically with C . rubella as the paternal parent ( Sicard et al . , 2011; Rebernig et al . , 2015 ) . To estimate the magnitude and direction of gene flow and other demographic events that have shaped genetic variation in the two species we used fastsimcoal2 ( Excoffier et al . , 2013 ) to compare the likelihood of a large number of demographic models given the observed joint site frequency spectrum ( Figure 2E , Figure 2—figure supplement 2 and Figure 2—source data 2 ) . The best fitting model estimated the split between C . rubella and C . grandiflora to have occurred 170 , 000 generations ago , associated with a strong reduction in C . rubella population size ( to only 2–14 effective chromosomes , or 1–7 individuals ) . Bidirectional gene flow at a relatively low rate apparently occurred until just over 10 , 000 generations ago , when C . rubella split into the W and E populations , after which gene flow continued only from E C . rubella to C . grandiflora ( Figure 2E ) . The close timing of the end of gene flow into C . rubella and the split into two populations suggests that westward expansion of the C . rubella range reduced the opportunity for gene flow from C . grandiflora , with potential genetic reinforcement by the development of hybrid incompatibilities ( Sicard et al . , 2015 ) . If we assume an average of 1 . 3 years per generation as found in the close relative , A . thaliana ( Falahati-Anbaran et al . , 2014 ) , which has similar life history and ecology , the population split and the end of introgression from C . grandiflora occurred around 13 , 500 years ago . This date is similar to the spread of agriculture and the end of the last glaciation in Europe ( Walker et al . , 2009 ) , suggesting that C . rubella’s success might have been facilitated by one or both of these events . Our analyses provide dates for the bottleneck and rapid colonisation events that have led to dramatically reduced genetic variation in C . rubella . Yet , over half of the segregating variants in C . rubella were also found in C . grandiflora ( Figure 1D ) . Such tsCgCrSNPs could originate from independent mutation in each species ( identity by state , IBS ) . Alternatively , they could be the result of introgression after speciation or they could reflect retention of the same alleles since the species split ( identity by descent , IBD ) . Older retained alleles are expected to be found at elevated frequencies relative to the genome-wide average , while younger , recurrent mutations are expected to be rare . We therefore identified ancestral and derived alleles by comparison with the related genus Arabidopsis , and then compared the derived allele frequency spectra of tsCgCrSNPs and ssSNPs in Capsella as a proxy for allele age . We found that tsCgCrSNPs are strongly enriched among high-frequency alleles in both Capsella species ( Figure 3A , p-value << 0 . 0001 in C . grandiflora and C . rubella , Mann-Whitney U-test ) . At allele frequencies greater than 0 . 25 in C . rubella , tsCgCrSNPs accounted for more than 80% of all variation . These results indicate that tsCgCrSNPs are predominantly older alleles that were already present in the common ancestral population of C . rubella and C . grandiflora or that were introgressed from C . grandiflora to C . rubella prior to its expansion into western Europe . The distribution of tsCgCrSNPs was uneven across the genome . When compared to ssCrSNPs drawn from the same allele frequency distribution , tsCgCrSNPs were less likely to result in nonsynonymous changes ( Figure 3B , p-value < 0 . 001 , from 1000 jackknife resamples from the same allele frequency distribution ) , but they were more likely to be in genes ( Figure 3C ) . As expected for transpecific haplotype sharing , eighty-three percent of all tsCgCrSNPs were in complete LD with at least one other tsCgCrSNP in C . rubella , and the density of tsSNPs along the genome was highly variable ( Figure 3D–G ) . tsCgCrSNP density was positively correlated with local genetic diversity in C . rubella ( and less strongly so with genetic diversity in C . grandiflora; Figure 3F–I and Figure 3—figure supplements 2–5 ) , and negatively correlated with differentiation between the species as measured by Fst ( Figure 3J and Figure 3—figure supplements 2–5 ) . The uneven pattern of diversity was similar in each C . rubella subpopulation ( Figure 3—figure supplements 6–9 ) , indicating that most of the retained polymorphism already segregated prior to colonisation . Thus , most common genetic variation in C . rubella is also retained in its outcrossing ancestor , and the rate of retention varies dramatically between genomic regions . The observed heterogeneity in shared diversity across the C . rubella genome could be a simple consequence of a bottleneck during the transition to selfing . In the simplest scenario , C . rubella was founded by a small number of closely related individuals , and stochastic processes during subsequent inbreeding caused random losses of population heterozygosity . A study of genetic variation in bottlenecked populations of the Catalina fox found this exact pattern ( Robinson et al . , 2016 ) . Alternatively , there may be selective maintenance of diversity in specific regions of the genome due to balanced polymorphisms , with contrasting activities of the different alleles . To explore this latter possibility , we tested whether the likelihood of allele sharing was dependent on annotated function of the affected genes . We found that tsCgCrSNPs were strongly biased towards genes involved in plant biotic interactions , including defense and immune responses , and also toward pollen-pistil interactions , though less strongly ( Supplementary file 1 , Figure 4A ) . Amongst the top ten enriched Gene Ontology ( GO ) categories for biological processes were apoptotic process , defense response , innate immune response , programmed cell death , and defensive secondary metabolite production ( specifically associated with terpenoids ) . Of genes annotated with apoptotic process , 87% were homologs of A . thaliana NLR genes , a class of genes best known for its involvement in perception and response to pathogen attack ( Jones and Dangl , 2006 ) . An even higher enrichment for tsCgCrSNPs was found when testing this class of genes specifically , with tsCgCrSNPs falling in NLR genes being more likely than those in other types of genes to result in nonsynonymous changes ( Figure 4A–B ) . These results indicate that despite a severe global loss of genetic diversity , genes involved in plant-pathogen interactions have maintained high levels of genetic variation in C . rubella . While the high density of tsCgCrSNPs near immunity genes was intriguing , NLR genes frequently occur in complex clusters , which could elevate error rates during SNP calling and thus potentially influence our analyses . Of particular concern is that sequencing reads derived from paralogs not found in the reference , but present in some accessions , could be mismapped against the reference , leading to false positive tsCgCrSNPs calls . We therefore examined whether tsCgCrSNPs showed more evidence of such errors than other SNPs . Mismapping should increase coverage and reduce concordance ( the fraction of reads supporting a particular call ) at a site . That the distributions of these two metrics were nearly identical at tsCgCrSNPs and ssSNP sites indicates , however , that mismapping is unlikely to have affected our SNP calls ( Figure 4—figure supplement 1 ) . Mismapping is also expected to cause pseudo-heterozygous calls , due to reads from different positions in the focal genome being mapped to the same target in the reference genome . However , tsCgCrSNPs were not more likely to be found in the heterozygous state as compared to ssSNPs ( Figure 4—figure supplement 1 ) . In addition , we asked whether the signal of increased tsCgCrSNPs density extended into sequences adjacent to NLRs and is detectable even when masking the NLR clusters themselves . For this purpose , we collapsed NLR genes within 10 kb of one another into a single region , and calculated tsCgCrSNPs rates and genetic diversity as a function of distance from these collapsed regions , ignoring SNPs within the focal cluster . We found that elevated tsCgCrSNPs sharing and genetic diversity extended over 100 kb from NLR genes . Thus , increased sharing is not an artefact of the internal structure of NLR clusters ( Figure 4B–C ) . Increased retention of genetic diversity near immunity loci suggests that these genes might be the targets of balancing selection in either C . rubella , C . grandiflora , or both species . However , neutral processes including random introgression and stochastic allele fixation can give rise to uneven distributions of genetic variation across the genome after genetic bottlenecks ( Robinson et al . , 2016 ) . We sought to identify regions that showed a pattern of allele sharing that was unlikely to have occured neutrally , as indicated by low values of the fixation index Fst , which quantifies genetic differentiation between populations . We compared the observed values of Fst between C . rubella and C . grandiflora to a distribution calculated from simulated sequences under our previously inferred neutral demographic model , which included gene flow between C . rubella and C . grandiflora . We simulated one million 20 kb DNA segments , or just over 7 , 000 C . rubella genome equivalents , under the neutral model and calculated the expected distribution of Fst values . Using this distribution , we assigned the probability of observing the Fst value for each non-overlapping 20 kb window throughout the genome . After Bonferroni correction and joining of adjacent significant segments , we identified 21 genomic regions that we designated as candidate targets of balancing selection ( Bal , Figure 4D and Figure 4—source data 1 ) . Bal regions showed several classical indications of balancing selection including substantially higher Tajima’s D and within-C . Rubella genetic diversity relative to the remainder of the genome ( Figure 4E–F; p<<0 . 001 Mann-Whitney U-test for both statistics ) . tsCgCrSNPs in Bal regions were also less likely to have been lost during colonisation of Western Europe than ssCrSNPs or tsCgCrSNPs in other parts of the genome , and allele frequencies in Bal regions showed elevated correlation across populations ( Figure 4—source data 2 ) . Like tsSNPs in general , Bal regions did not show evidence for increased heterozygosity that might indicate increased error rates in SNP calling ( Median Observed - Expected Heterozygosity in 20 kb windows was −0 . 021 inside of Bal regions , and −0 . 020 outside of these regions ) . Estimates of Fst were reduced in large segments surrounding NLR and other immune gene candidate clusters ( Figure 4G ) , consistent with allele sharing being the result of linkage to a nearby balanced polymorphism . Of the 21 candidate regions , nine overlapped with clusters of NLR genes , and five with clusters of RLK/RLP or CRK genes , two classes of genes with broad roles in innate immunity ( Yeh et al . , 2015; Zipfel , 2008 ) . Many of the specific regions we identified in Capsella have been directly demonstrated to function in disease resistance in A . thaliana ( McDowell et al . , 2000; McDowell , 1998; Goritschnig et al . , 2012; Holub , 1994; Gassmann et al . , 1999; Zhang et al . , 2014; Zhang et al . , 2013; Xu , 2006; Yeh et al . , 2015 ) . RPP1 and RPP8 have been previously suggested as candidate targets of balancing selection , and trans-specific polymorphism has been reported at the RPP8 locus in the genus Arabidopsis ( Bergelson et al . , 2001; Wang et al . , 2011 ) . It should be noted , however , that these genes are often members of larger linked NLR gene superclusters , with some of the regions our approach identified being sizeable and thus making it difficult to pinpoint a single focal gene . Indeed the strong signal found in these regions could result from multiple linked balanced sites . Furthermore , the strongest signals of balancing selection are mostly derived from linked sites , rather than the clusters themselves , because confident SNP calling is very difficult , if not impossible , with short reads in the most complex genomic regions ( Figure 4G ) . It is formally possible that the unusual pattern of diversity that we observe near Bal loci could result from historical balancing selection in the outcrossing ancestor C . grandiflora rather than ongoing selection in the selfing C . rubella . Population genetic indices such as Fst , nucleotide diversity pi , Tajima’s D , and allele sharing are not fully independent , and elevated diversity in the C . rubella founding population , driven by historical balancing selection , could also generate the observed patterns . Genetic diversity was only modestly elevated in these regions in C . grandiflora ( p<<0 . 001 Mann-Whitney U-test , Figure 4E ) , and Tajima’s D was not significantly different from other windows ( Figure 4F ) , suggesting that this is not very likely . If balancing selection is acting at these loci in the outcrosser , it is clear that its genomic footprint is small , perhaps due to the rapid decay of LD in this species relative to the selfing C . rubella . Still , it is possible that even a small elevation of genetic diversity in Bal regions in the founding populations might have considerable impact on subsequent C . rubella diversity . We approximated this situation using our simulated genetic data . We subsetted simulations by the level of genetic diversity in C . grandiflora , choosing the top 1% of simulated values . Even in the case of elevated founder diversity in these data , the observed Fst values in Bal regions remain exceptionally unlikely ( p<0 . 0001 ) . These observations point to ongoing balancing selection within C . rubella maintaining diversity in Bal regions . Adaptive retention of C . grandiflora diversity in Bal regions could be explained by two non-exclusive models . Allelic variation might have been present in the C . rubella founding population and maintained by balancing selection until the present . Alternatively , beneficial alleles may have been introgressed from C . grandiflora after the evolution of selfing , and retained by balancing selection . We searched for evidence of recent ancestry between the two species in Bal regions . A larger fraction of Bal region sequence was found to be IBD when compared to the genome-wide average ( Figure 4—figure supplement 2 ) , consistent with elevated retention of introgressed alleles in these regions . Shared segments in Bal regions were on average shorter than those found in other parts of the genome , suggesting that they are older and have been subjected to longer periods of recombination since the introgression event ( median within 3 , 503 bp , median outside 6 , 661 bp , Wilcoxon-rank sum test , p=1e-54 ) , although we cannot exclude the influence of differing patterns of recombination in these regions as a contributing factor to this observation . Elevated IBD rates in Bal regions might result from gene flow between the species in either direction , and our previous results suggested that most modern gene flow occurs through introgression of C . rubella alleles into C . grandiflora . We explored the geographic pattern of IBD between C . rubella and C . grandiflora in Bal regions to determine whether it differs from that of the genome-wide pattern . Within the East population , IBD decayed as a function of distance from the C . grandiflora range in a manner comparable to the observed genome-wide pattern , albeit with a more shallow slope ( Figure 4—figure supplement 2 ) . In contrast to the genome-wide pattern , high levels of IBD were observed between C . grandiflora and West population accessions . Thus , we find evidence for neutral gene flow throughout the genome , perhaps dominated by C . rubella to C . grandiflora introgression , as indicated by our demographic simulations . However , allele sharing appears to be older in Bal regions and introgressed alleles have been retained for longer periods even after colonisation of Western Europe . This latter observation is consistent with the hypothesis that alleles were introgressed prior to the most recent range expansions in C . rubella , and that variation was subsequently maintained by selection in Bal regions . Although evidence for balancing selection at immunity-related loci in C . rubella is much stronger than in C . grandiflora , it is difficult to completely exclude the effect of founder diversity at these loci on the observed patterns . We therefore sought to validate our findings in a related species that has been separated from C . grandiflora and C . rubella for a long time . The genus Capsella offers a unique opportunity to test the longevity of balancing selection , because selfing has evolved independently in C . orientalis , which diverged from C . grandiflora and C . rubella more than one million years ago and whose modern range no longer overlaps with the two other species , preventing ongoing introgression ( Hurka et al . , 2012; Douglas et al . , 2015 ) . We expected the evolution of selfing to have generated a similar bottleneck as in C . rubella ( Douglas et al . , 2015; Bachmann et al . , 2018 ) , and we therefore resequenced 16 C . orientalis genomes , to test whether there is evidence of balancing selection at similar types of loci . After alignment , SNP calling , and filtering , we identified a mere 71 , 454 segregating SNPs in C . orientalis . This is a surprisingly small amount of variation , corresponding to an almost 50-fold reduction in diversity relative to the outcrossing C . grandiflora ( Figure 5—source data 1 ) . Using our divergence and diversity measures , we estimated that C . orientalis diverged from C . grandiflora over 1 . 8 million generations ago ( calculated as in ref . Brandvain et al . , 2013 ) . The combination of long divergence times and low variability in C . orientalis makes it unlikely that alleles will have been maintained by random chance . Using estimates of Ne from nucleotide diversity at four-fold degenerate sites ( C . orientalis [14 , 643] and C . grandiflora [694 , 643] ) , the divergence time above , and the genome assembly size of 134 . 8 Mb , the probability of finding a single tsSNP is <4×10−19 using the methodology of Leffler and colleagues and Wiuf and colleagues ( Leffler et al . , 2013a; Wiuf et al . , 2004 ) , which assumes constant population size . It was therefore surprising that 8 , 408 C . orientalis variants were shared with either C . rubella or C . grandiflora ( ts2-waySNPs ) , and 3992 with both ( ts3-waySNPs , Figure 5A–B ) . In each of the three species , ts3-waySNPs were enriched at higher derived allele frequencies relative to ssSNPs and ts2-waySNPs , suggesting that they are on average the oldest SNPs ( Figure 5C ) . Because this large amount of trans-specific polymorphism was unexpected , we wanted to ensure that this was not due to more error-prone read mapping to a distant reference . We therefore also used an additional set of more stringent filters to identify high confidence ts3-waySNPs ( ts3-wayhqSNPs; see Materials and methods ) . Importantly , we required ts3-wayhqSNPs to be in LD with at least one other ts3-wayhqSNP in all three species ( r2 >0 . 2 in the same phase ) , to provide evidence that they represented the same ancestral haplotype . The aim was to improve the likelihood that such SNPs were true examples of identity by descent . Furthermore , we generated a draft assembly of the C . orientalis genome using Pacific Biosciences SMRT cell technology , and re-called ts3-waySNP sites . We identified 812 high quality transpecific SNPs segregating in all three species ( ts3-wayhqSNPs ) . The distributions of coverage and concordance values in this dataset were similar between ts3-waySNP sites and other C . orientalis sites , further supporting their authenticity ( Figure 5—figure supplement 1 ) . As discussed earlier , the presence of trans-specific polymorphism in diverged species could be driven by stable balancing selection or it could result from gene flow between the species . While C . grandiflora and C . rubella occur around the Mediterranean , C . orientalis is restricted to Central Asia ( Hurka et al . , 2012 ) and its current distribution is far from that of C . grandiflora and C . rubella . Modern gene flow between the C . orientalis and C . rubella/C . grandiflora lineages is therefore unlikely , but it is possible that the ranges of these species overlapped in the past . If alleles have been maintained since the split between the lineages , then the divergence between maintained alleles should meet or exceed the divergence between the species . On the other hand , if ts3-wayhqSNPs are the result of recent gene flow between the lineages , then divergence between species near these SNPs should be reduced compared to the genome-wide average divergence . We examined diversity and divergence at neutral ( four-fold degenerate ) sites surrounding ts3-wayhqSNPs ( Figure 5D ) . In all three species , diversity was high directly adjacent to ts3-wayhqSNPs , close to average levels for genome-wide divergence between the two Capsella lineages . This footprint of elevated diversity is much more discernible in the two selfing species than in C . grandiflora . No obvious reduction in divergence was observed near ts3-wayhqSNPs ( Figure 5D ) . We conclude that ts3-wayhqSNPs correspond predominantly to long-term maintained alleles that diverged on ancient time scales and that they are not the result of recent introgression . The finding of tsSNPs shared between two independent lineages , C . grandiflora/C . rubella and C . orientalis , for over a million generations in spite of strong geographic barriers suggests that they are targets of stable long-term balancing selection . If this selection pressure remains constant across species , ancient alleles are expected to evolve towards similar equilibrium intermediate frequencies . In comparison to ts2-waySNPs , the minor alleles at ts3-wayhqSNP sites are closer to intermediate frequencies in all three species ( Figure 5—figure supplement 2 ) . Furthermore , ts3-wayhqSNPs segregate at more similar allele frequencies in C . rubella and C . grandiflora than other two-way tsSNPs , as measured by Fst median values: 0 . 03 for ts3-wayhqSNPs and 0 . 16 for ts2-waySNPs , p<<0 . 001 Mann-Whitney test ) and correlation of derived allele frequencies ( Figure 4—source data 2 ) . These results suggest a conserved equilibrium maintained since the isolation of C . rubella and C . grandiflora over 10 , 000 generations ago . Derived allele frequencies for ts3-wayhqSNPs are not correlated between the two ancient Capsella lineages ( Spearman’s rho −0 . 08 C . orientalis to C . grandiflora and −0 . 04 to C . rubella ) . It is possible that demographic reduction or habitat shift in C . orientalis has disturbed this equilibrium . Like tsCgCrSNPs , ts3-waySNPs are strongly enriched in GO categories associated with immunity ( Supplementary file 2 ) . Our previously identified balanced regions strongly predicted the genomic distribution of ts3-waySNPs; 50% of ts3-wayhqSNPs fell into these regions , even though they encompass fewer than 10% of tsCgCrSNPs and fewer than 3% of all SNPs , resulting in an even more skewed and uneven distribution of genetic diversity along the genome ( Figure 5F–G ) . At least one ts3-wayhqSNP was found in each of 10 of the 21 original candidate regions under balanced selection . Six of these corresponded to NLR clusters , two to RLK/RLP clusters , and one to a TIR-X cluster . Only one region did not contain a clear immunity candidate , with the caveat that this conclusion is based on the single annotated C . rubella reference genome ( Slotte et al . , 2013 ) . Thus , even in a situation where a recent genetic bottleneck has wiped out almost all genetic diversity , there is very strong selection to maintain allelic diversity at specific immunity-related loci , consistent with these alleles having persisted already for very long evolutionary times . The balanced regions we identified contained very old tsSNPs , yet as mentioned , the immunity genes themselves are often not accessible to variant discovery based on mapping short reads to a single reference genome . Furthermore , it is possible , or even likely , that the strongest evidence for balancing selection comes from loci that include several linked targets of balancing selection . This combination of factors makes it difficult to pinpoint potential functional changes maintained by balancing selection in these regions . To discover functional changes , we therefore focused on ts3-wayhqSNPs that did not fall in our large balanced regions but were clustered in regions of the genome that were likely less complex . We selected genes that were well covered by reads in all three species ( >80% sites ) , contained at least six high quality tsSNPs , at least one non-synonymous ts3-wayhqSNP , were at least 100 kb from any of our candidate balanced regions , and had been functionally characterised in A . thaliana . These filters singled out a homolog of the A . thaliana MLO2 gene as a particularly good candidate for more detailed analysis ( Supplementary file 3 and Figure 6 ) . MLO2 encodes a seven-transmembrane domain protein with a conserved role in plant disease susceptibility ( Figure 6A ) ( Consonni et al . , 2006 ) . The C . rubella MLO2 locus has experienced a tandem duplication , resulting in two genes , MLO2a and MLO2b . Although both homologs are sufficiently diverged to be accessible to unambiguous read mapping , all six ts3-wayhqSNPs were in MLO2b ( Figure 6B–C ) . In C . rubella and C . orientalis , the ts3-wayhqSNPs were arranged in five different haplotypes , which we collapsed into three related haplogroups , A , B and C ( Figure 6B ) . The reference haplogroup A was most frequent in both species . Because several known targets of balancing selection in A . thaliana are the result of structural variation , or lesions larger than 1 kb ( Mauricio et al . , 2003; Stahl et al . , 1999 ) , we examined coverage patterns around the MLO2 locus to identify potential linked indels . We found that haplogroup B in both C . rubella and C . orientalis exhibited similar patterns of low read coverage at the 5’ end of MLO2b , suggesting a possible indel ( Figure 6C ) . To examine the exact sequence of each allele , we took advantage of the homozygous nature of sequence data from these two selfing species and performed local de novo assembly of the MLO2 locus from read pairs mapping to this region . We were able to reconstruct the locus for 15 C . orientalis samples ( 13 haplogroup A and two haplogroup B ) and 43 C . rubella samples ( 34 A , 2 B , and 7 C ) . Surprisingly , a comparison of the different haplotypes revealed that the pattern of low coverage observed for haplogroup B was not due to structural variation , but instead to extremely high divergence from the reference haplogroup A ( Figure 6—figure supplement 1 ) . Divergence between alleles within species was greater than 0 . 15 differences per bp , over three times higher than the genome-wide divergence between the species ( Figure 6—figure supplement 1 and Figure 5—source data 1 ) . This highly diverged region had therefore been originally inaccessible to reference-based read mapping in haplogroup B samples . De novo assembly allowed us to identify a total of 204 additional tsSNPs , nearly all of which mapped to the 5’ end of MLO2b ( Figure 6—figure supplement 1 ) . Neighbour-joining trees revealed the expected clustering of samples by species in regions adjacent to MLO2b , but clear clustering by haplogroup within the 5’ region , a pattern that is reproduced in phylogenetic analysis of the entire CDS ( Figure 6C and Figure 6—figure supplement 2 ) . Importantly , divergence within haplogroup across species was greater than , or similar to genome-wide averages for both A and B , demonstrating that recent introgression did not give rise to allele sharing ( Figure 6—figure supplement 1 ) . The high nucleotide divergence between haplogroups A and B translates into numerous amino acid differences in the N terminal half of the encoded proteins . In a 157 amino acid stretch , 31 amino acid differences are found in both species ( Figure 6A and Figure 6—figure supplement 3 ) , with an indel polymorphism accounting for another seven amino acid differences . The large number of differences between the two haplogroups makes it difficult to point to any specific change as the target of balancing selection , but it seems likely that the two alleles differ functionally , perhaps reinforced by additional differences in the promoter . In summary , the nucleotide divergence in this region suggests that the MLO2b haplogroups are much older than the split between the two species . While balancing selection has long been recognised as an important evolutionary force , its relevance as a major factor shaping genomic variation has remained unclear ( Charlesworth , 2006; Wiuf et al . , 2004; Asthana et al . , 2005 ) . We have taken advantage of unique demographic situations in two Capsella lineages to demonstrate not only that there is pervasive balancing selection at immunity-related loci in this genus , but also that the same alleles are maintained in species that are likely experiencing quite different pathogen pressures . We expect that balancing selection plays a similar role in other taxa , but that its effects are masked by a background of higher neutral genetic diversity and more frequent recombination between balanced sites and linked variants ( Wiuf et al . , 2004; Charlesworth , 2006 ) . In addition , the detection of long-term balancing selection is further compounded by very old alleles being less accessible to short read re-sequencing , the dominant mode of variant discovery today . In the two selfing Capsella species , the footprints of balancing selection extend for tens of kilobases , greatly impacting diversity of many other genes . While this makes it more difficult to pinpoint the actual selected variants , it greatly improves statistical power to identify regions under balancing selection . This is reminiscent of genome-wide association studies , where extended LD improves statistical power to detect causal regions of the genome but reduces the ability to identify the specific causal variants ( Atwell et al . , 2010 ) . The nature of balancing selection acting on the regions we have identified remains to be clarified . Stable balancing selection in self fertilising species is unlikely to derive from heterozygous advantage , pointing to negative frequency-dependent selection or fluctuating selection from variable pathogen pressures as possible factors . While the mode of selection cannot be determined from these static data , the strong signal that we observe in highly selfing lineages points to environmental heterogeneity or negative frequency dependent selection over heterozygote advantage . Based on the enrichment of immunity-related genes , it appears that biotic factors are the dominant drivers of long-term maintenance of polymorphism . This observation is consistent with a large body of work on intraspecific variation in A . thaliana . The signal of balancing selection has been observed for specific pairs of disease resistance alleles in A . thaliana ( Stahl et al . , 1999; Tian et al . , 2002; Tian et al . , 2003; Mauricio et al . , 2003; Bakker , 2006 ) , and in the case of the resistance gene RPS5 , alternative alleles have been shown to affect fitness in the field ( Karasov et al . , 2014 ) . It is possible , or perhaps even likely , that the signal of balancing selection is amplified by the fact that immunity-related loci occur in clusters ( Meyers , 2003 ) and that our strongest signal is the result of simultaneous selection on several genes in these regions in a situation analogous to the MHC in animals ( Hedrick , 1998 ) . Thus , biotic factors might not be quite as important as our analyses make them appear . On the other hand , it is also possible that the clustering of disease resistance genes itself is a product of selection , if selection was more effective when acting on groups of genes ( Charlesworth and Charlesworth , 1975 ) , or if evolution under a balanced regime was deleterious at other types of loci . Even if we accept that biotic factors predominate , the nature of the potential trade-offs that prevent individual alleles from becoming fixed is still a mystery , but it might involve conflicts between growth and defense ( Coley et al . , 1985; Walling , 2009; Herms and Mattson , 1992 ) , beneficial and harmful microbe interactions ( Walters and Heil , 2007 ) , or defense against different types of pathogens ( Kliebenstein and Rowe , 2008 ) . What is clear is that the trade-offs must be stable over very long periods of evolution . Our findings suggest a model in which the success of self fertilising populations may be buoyed by gene flow from outcrossing relatives in a situation analogous to evolutionary rescue strategies in conservation biology ( Whiteley et al . , 2015 ) . This model is a variation on the theme of adaptive introgressions , which have recently emerged as a major evolutionary force in a wide range of taxa ( Whitney et al . , 2006; Castric et al . , 2008; Pease et al . , 2016; Dasmahapatra et al . , 2012; Henning and Meyer , 2014; Hedrick , 2013; Huerta-Sánchez et al . , 2014; Racimo et al . , 2015; Castric et al . , 2008; Pease et al . , 2016; Dasmahapatra et al . , 2012; Whitney et al . , 2006; Huerta-Sánchez et al . , 2014; Hedrick , 2013 ) . The unique feature of self-fertilisation in comparison to these examples is that the amplified effects of linked selection and genetic drift lead to a steady loss of genetic variation over time . Constant replenishment via adaptive introgression from an outcrossing relative counters the loss of diversity at immunity-related loci , thereby preventing decreased fitness in competition with pathogens . Whether this model generally applies will require independent study of other lineages of related self-fertilising and outcrossing populations at various stages of speciation . Finally , we note that maintenance of ancient variants is most easily detectable in a background of low variation . Therefore , it could potentially be used to rapidly identify loci with meaningful functional variation . Typically , agricultural breeding panels seek to maximise surveyed diversity , but our results indicate that identification of useful immunity-related polymorphism with genomic data might be facilitated in otherwise homogeneous wild populations . Seeds were stratified for two weeks at 4°C and germinated in controlled environment chambers . Four to six rosette leaves were collected from each accession and frozen in liquid nitrogen for gDNA extraction . The methods available for extraction and sequencing varied as the project progressed , and 24 of the C . rubella and the 13 C . grandiflora samples were analysed independently in previous studies ( Agren et al . , 2014; Williamson et al . , 2014 ) . See Figure 1—source data 1 for a listing of DNA preparation , library construction , and sequencing technology by sample . In brief , DNA was extracted following an abbreviated nuclei enrichment protocol ( Becker et al . , 2011 ) or using the Qiagen Plant DNeasy Extraction kit . The recovered DNA was sheared to the desired length using a Covaris S220 instrument , and Illumina sequencing libraries were prepared using the NEBNext DNA Sample Prep Reagent Set 1 ( New England Biolabs ) or the Illumina TruSeq DNA Library Preparation Kit and sequenced on the instrument as listed in Figure 1—source data 1 . We aimed for a minimum genome coverage of 40x . We mapped reads to the C . rubella reference genome ( Slotte et al . , 2013 ) resulting in realised coverages of 30 – 126x . Initial sequence read processing , alignment , and variant calling were carried out using the SHORE ( v0 . 8 ) software package ( Ossowski et al . , 2008 ) . Read filtering , de-multiplexing , and trimming were accomplished using the import command discarding reads that had low complexity , contained more than 10% ambiguous bases , or were shorter than 75 bp after trimming . Reads were mapped to the C . rubella reference genome ( Phytozome v . 1 . 0 ) using the GenomeMapper aligner ( Schneeberger et al . , 2009 ) with a maximum edit distance ( gaps or mismatches ) of 10% . Alignments from each sample were then processed to generate raw whole genome reference and variant calls with qualities computed using an empirical scoring matrix approach ( Cao et al . , 2011 ) allowing heterozygous positions . Of the initial 53 C . rubella samples , two were removed because of low or uneven coverage , and one was removed as a misidentified C . bursa-pastoris sample ( C . rubella and its polyploid relative C . bursa-pastoris are not easily identified phenotypically , but they can be distinguished by the extreme number of pseudo-heterozygous calls in the latter ) . The per-sample raw consensus calls produced by SHORE were used to construct a whole genome matrix of finalised genotype calls for each species . Positions were considered only if covered by at least four reads and if overlapping reads mapped uniquely ( GenomeMapper applies a ‘best match’ approach , so unique means that only one best match exists ) ( Schneeberger et al . , 2009 ) . We simultaneously considered information from all samples within a species to make base pair calls . If no variant was called in any sample then the site was treated as reference . Individual sample calls were made if four reads supported the reference base , the computed quality was above 24 , and at least 80% of reads supported a reference call . A site was excluded if more than 30% of the samples from that species did not meet these criteria . If at least one sample reported a difference from the reference in the raw consensus , then variant ( indel or SNP ) or reference calls were considered . The SNP calling parameters were slightly different for the two selfing species as compared to the outcrossing C . grandiflora because variants should only rarely be found in the heterozygous state in the former ( and the frequency of heterozygous calls in a selfing species is a powerful filter to detect problems with mismapped reads ) . The general approach was to require at least one high quality variant call at a site and then to call genotypes in other samples with slightly reduced stringency . If no variant call met the more stringent threshold , then the site was reconsidered using the above reference criteria . Finally , the calls from each of the three species were combined into a master matrix . If a position was not called biallelic or invariant across the compared species , then it was not considered . To facilitate further analyses in PLINK ( v1 . 9 ) ( Chang et al . , 2015 ) and vcftools ( v0 . 1 . 12a ) ( Danecek et al . , 2011 ) , the genome matrix at biallelic SNP sites was also converted into a minimal vcf format . Regions of high repeat density near the centromeres of all chromosomes as well as two large , repeat-rich regions in chromosomes 1 and 7 were removed from genome scans . Coordinates for these regions are listed in Supplementary file 4 . We used the SnpEff ( v . 3 . 2a ) ( Cingolani et al . , 2012 ) software package to annotate variant and invariant sites for the whole genome . The annotation database was built using the C . rubella v1 . 0 Phytozome gff file . Sites were annotated using the table input function that includes annotation of fold degeneracy for each site in coding regions . Invariant sites were annotated using a table with dummy SNPs at each position . The SnpEff program outputs several annotations for some sites , and a primary annotation was selected by ranking the strength of effect of each annotation and reporting the annotation with the strongest effect ( the rankings are listed in Supplementary file 5 ) . To calculate derived allele frequency spectra we assigned ancestral state to each polymorphic site using three-way whole genome alignments between C . rubella , A . thaliana , and A . lyrata ( Slotte et al . , 2013 ) . Only biallelic sites identical between A . lyrata and A . thaliana ( indels were ignored ) were considered . For the two species analysis , only sites also fixed for the ancestral allele in C . orientalis were considered . To compare tsSNP and ssSNP annotations from similar allele frequency spectra , we binned 20 , 000 tsSNPs randomly drawn from throughout the genome by derived allele frequency ( 10 bins ) . We then drew an equivalent number of ssSNPs from each allele frequency bin and calculated the fraction of CDS SNPs that caused nonsynonymous changes and the fraction that fell in genes . This process was repeated 1000 times for both species to generate the plots shown in Figure 3B . Genotypes at four-fold degenerate SNP sites called in C . grandiflora and C . rubella were pruned in PLINK ( 50 kb windows , 5 kb step , and 0 . 2 r2 LD threshold ) and used as input for ADMIXTURE ( v . 1 . 23 ) ( Alexander et al . , 2009 ) and EIGENSTRAT ( v6 . 0 beta ) ( Price et al . , 2006 ) . For demographic modelling in Fastsimcoal ( v2 . 5 . 2 . 11 ) ( Excoffier et al . , 2013 ) , joint minor allele frequency spectra were generated at four-fold degenerate sites with complete information and ignoring heterozygous calls in selfing lineages ( counting only one allele from each individual ) . Demographic parameters for each tested model were then inferred in 50 runs of Fastsimcoal ( parameters: -l40 -L40 -n100000 -N100000 -M0 . 001 -C5 ) . The global maximum likelihood model was selected after correcting for number of estimated parameters using Akaike Information Criterion . Confidence intervals were set for estimated parameters using 100 bootstraps of identical inference runs on simulated data under the most likely model . To reduce computational times , global maximum likelihoods were calculated for bootstraps after 13 runs rather than 50 . The mutation rate assumed for this and other analyses was 7 × 10−9 mutations/generation/ bp based on mutation rate measurements in Arabidopsis thaliana ( Ossowski et al . , 2010 ) . Segments of IBD were identified using the phasing and segment identification in Beagle ( r1339 ) ( Browning and Browning , 2013 ) . For the analysis presented here , we considered only the first haplotype from each C . rubella sample and both haplotypes from each C . grandiflora sample . Segments were required to be larger than 1 kb to be considered in the analyses . D statistics were calculated as in Green et al . ( 2010 ) ; Patterson et al . ( 2012 ) ; Dasmahapatra et al . ( 2012 ) comparing each individual genotype from the eastern C . rubella population to allele frequencies from western C . rubella and C . grandiflora . The outgroup species for these analyses was C . orientalis . Population genetic diversity statistics for genome scans were calculated for each species by transforming variant calls from the genome matrix into FASTA files and inputting these files into the compute function from the libsequence analysis package ( Thornton , 2003 ) . Heterozygous bases were randomly assigned as reference or variant to generate a single haplotype for each sample . Weir and Cockerham’s Fst was calculated using vcftools ( v . 0 . 1 . 12a ) on biallelic SNP sites . To identify regions of the genome with unusually low Fst after speciation , we generated a null distribution of Fst values by simulating one million 20 kb segments under our inferred best demographic model using Fastsimcoal2 . The output of each simulation was transformed to vcf format and Fst between C . grandiflora and each C . rubella subpopulation was calculated using vcftools . The probability of a particular Fst value in the observed data was then assigned based on its rank in these simulations ( independently for the two subpopulations; one sided test ) . Multiple testing was accounted for using Bonferroni correction . Significant outlier windows ( adjusted p-value<0 . 05 ) identified for each subpopulation were collapsed into regions using a two state hidden markov-model as implemented in the Rhmm package . The HMM approach has the advantage of joining windows of high coverage separated by a low coverage window . Only regions significant in both subpopulations were considered for further analysis . Windows overlapping the pericentromeric regions were removed from the analysis . LD was calculated in 30 kb windows in C . grandiflora and C . rubella using PLINK ( v . 1 . 9 ) . The decay of LD is the mean value at each position up to 30 kb from a focal SNP . Because the C . rubella annotation is sparse , we used annotations from nucleotide blast best hit matches ( e < 1e-10 ) to CDS sequences from its close relative , A . thaliana , for our GO analysis . Enrichment tests were performed with the SNP2GO R library ( Szkiba et al . , 2014 ) using tsCgCrSNPs as the test set and all SNP sites called in either C . rubella or C . grandiflora as the background set . We chose this approach because it is less sensitive to gene length ( which should similarly affect tsSNP and non-tsSNP distributions across genes ) . A corresponding analysis was performed in the three-way comparisons using a background set of all SNP sites called in all three species . Significant enrichments were considered at a q-value threshold of q < 0 . 01 after false discovery correction . A gene was considered as belonging to the NLR family in C . rubella if its best blast hit in A . thaliana was annotated as such ( Supplementary file 6 ) . To generate a list of high quality ts3-waySNPs , we applied a series of empirical filters . First , all ts3-waySNPs were required to have an r2 >0 . 2 with another ts3-wayhqSNP in the same phase in all three species . We excluded SNPs overlapping pericentromeric or annotated repeat sequences ( Slotte et al . , 2013 ) . We also required that the coverage of SNPs was no more than two standard deviations above the mean coverage of all SNPs for that species , to have an average concordance greater than 0 . 98 , and to be identified in more than one individual . These criteria were selected to increase our confidence in identified tsSNPs; it is likely that our inferences are conservative . To validate our trans-specific SNPs we aligned the C . orientalis samples against the draft C . orientalis assembly using the bwa ( v . 0 . 7 . 12 ) mem command with default parameters . The output bam format file was sorted using samtools ( v . 1 . 6 ) and multisample variant calls were made with freebayes ( v . 1 . 1 . 0 ) using the parameter settings -z . 1–0 w . The resulting vcf file was filtered using vcftools ( v . 0 . 1 . 13 ) using the settings --remove-indels --minQ 50 --max-missing 0 . 8 --max-alleles two and further filtered to remove sites that were called as heterozygous in more than 5% of the samples . The sites overlapping with the original call set were extracted from this vcf and used for validation . Coordinate transforms between the two genomes were necessary to validate tsSNPs . The draft assembly of C . orientalis and the C . rubella reference genome were aligned using the LAST ( v . 923 ) aligner . The C . rubella reference database was built with the lastdb command with the parameter settings -uMAM8 -cR11 , and then the two genomes were aligned with the lastal command with the settings -m50 -E0 . 05 . Equivalent sites were considered if they were present in alignments at least 500 bp long and contained only one C . orientalis and one C . rubella sequence . To reconstruct alleles from the MLO2 locus , we used an iterative assembly approach . Reads were first mapped to the entire reference genome using bwa ( v . 0 . 7 . 8 ) ( Li and Durbin , 2009 ) using the bwa-mem alignment algorithm for each sample . Reads that mapped to the MLO2 locus were then extracted and assembled de novo using SPAdes ( v . 3 . 5 . 0 ) ( Nurk et al . , 2013 ) . Assemblies were filtered to be longer than 2 , 000 bp with a coverage greater than 5 , and then used to create an index for a second round of read mapping . Reads that mapped to the assembly without mismatches were collected together with their mates ( regardless of the mate’s mapping quality ) , and were again de novo assembled . This process was iterated six times until scaffolds covering both coding regions were achieved . Format conversions and file handling made use of the software samtools ( v . 0 . 1 . 19 ) ( Li et al . , 2009 ) and bamutil ( v . 1 . 0 . 13 ) . Assemblies were filtered for appropriate length , and aligned using MAFFT ( Katoh and Standley , 2013 ) . Alignments were visualised using AliView ( Larsson , 2014 ) , and manually edited where appropriate . The protein encoded by MLO2b annotated in the C . rubella reference was truncated relative to A . thaliana MLO2 . We aligned the genomic and coding regions from both species and found that the premature stop in MLO2b is likely due to a mis-annotated splice junction . The A . thaliana junction is conserved in C . rubella and alternative annotations on phytozome identify the A . thaliana-like splice variant . We therefore used the full-length version derived from manual alignments for our analysis . The phylogeny of Capsella MLO2 CDS sequences was produced using the optim . pml command from the R package phangorn using Jukes-Cantor distances . 1000 bootstrap iterations were run to estimate support for nodes in the tree . To determine where amino acid substitutions had occurred , we aligned the proteins encoded by each allele against the barley mlo protein and annotated domains ( UniProtKB P3766 ) . The draft genome from the C . orientalis accession 2007–03 ( Figure 1—source data 1 ) was assembled from long reads generated by PacBio single-molecule real-time sequencing . Long reads were assembled with Falcon ( Chin et al . , 2016 ) ( version 0 . 5 . 4 , max_diff = 150 , max_cov = 150 , min_cov = 2 ) . The resulting primary contig set was iteratively polished with Quiver again using long reads ( Chin et al . , 2013 ) ( version 2 . 0 . 0 ) and with Pilon ( Walker et al . , 2014 ) ( version 1 . 16 ) using short reads from a single Illumina TruSeq DNA PCR-free library . The draft genome of C . orientalis comprises 135 Mb distributed over 423 gap-free contigs and covers 60% of the C . rubella reference with non-ambiguous 1-to-1 whole genome alignments . Its completeness is comparable to that of the C . rubella reference .	Capsella rubella is a small plant that is found in southern and western Europe . This plant is young in evolutionary terms: it is thought to have emerged less than 200 , 000 years ago from a small group of plants belonging to an older species known as Capsella grandiflora . Individuals of the same species may carry alternative versions of the same genes – known as alleles – and the total number of alleles present in a population is referred to as genetic diversity . When a few individuals form a new species , the gene pool and the genetic diversity in the new species is initially much lower than in the ancestral species , which may make the new species less robust to fluctuations in the environment . For example , alternative versions of a gene might be preferable in hot or cold climates , and loss of one of these versions would limit the species’ ability to survive in both climates . A mechanism known as balancing selection can maintain various alleles in a species , even if the population is very small . However , it was not clear how common long-lasting balancing selection was after a species had split . To address this question , Koenig et al . assembled collections of wild C . rubella and C . grandiflora plants and sequenced their genomes in search of alleles that were shared between individuals of the two species . The analysis found not just a few , but thousands of examples where the same genetic differences had been maintained in both C . rubella and C . grandiflora . Some of these allele pairs were also shared with individuals of a third species of Capsella that had split from C . rubella and C . grandiflora over a million years ago . The shared alleles did not occur randomly in the genome; genes involved in immune responses were far more likely to be targets of balancing selection than other types of genes . These findings indicate that there is strong balancing selection to maintain different alleles of immunity genes in wild populations of plants , and that some of this diversity can be maintained over hundreds of thousands , if not millions of years . The strategy developed by Koenig et al . may help to identify new versions of immunity genes from wild relatives of crop plants that could be used to combat crop diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "genetics", "and", "genomics" ]	2019	Long-term balancing selection drives evolution of immunity genes in Capsella
During nervous system development , commissural axons cross the midline despite the presence of repellant ligands . In Drosophila , commissural axons avoid premature responsiveness to the midline repellant Slit by expressing the endosomal sorting receptor Commissureless , which reduces surface expression of the Slit receptor Roundabout1 ( Robo1 ) . In this study , we describe a distinct mechanism to inhibit Robo1 repulsion and promote midline crossing , in which Roundabout2 ( Robo2 ) binds to and prevents Robo1 signaling . Unexpectedly , we find that Robo2 is expressed in midline cells during the early stages of commissural axon guidance , and that over-expression of Robo2 can rescue robo2-dependent midline crossing defects non-cell autonomously . We show that the extracellular domains required for binding to Robo1 are also required for Robo2's ability to promote midline crossing , in both gain-of-function and rescue assays . These findings indicate that at least two independent mechanisms to overcome Slit-Robo1 repulsion in pre-crossing commissural axons have evolved in Drosophila . The secreted Slit repellents and their Roundabout ( Robo ) receptors constitute a repulsive axon guidance system whose function is conserved across a wide range of animal taxa including vertebrates , planarians , nematodes , and insects ( Brose and Tessier-Lavigne , 2000; Evans and Bashaw , 2012 ) . Slits are normally expressed at the midline of the central nervous system ( CNS ) , and axons expressing Robo receptors are thus repelled from the midline ( Battye et al . , 1999; Brose et al . , 1999; Kidd et al . , 1999 ) . Prior to crossing the midline , commissural neurons in vertebrates and insects prevent premature responsiveness to Slit by regulating the expression and activity of Robo receptors through a variety of mechanisms ( Evans and Bashaw , 2010a; Neuhaus-Follini and Bashaw , 2015 ) . For example , the divergent Robo receptor Robo3/Rig-1 in vertebrates negatively regulates the activity of the Robo1 ( Roundabout1 ) and Robo2 ( Roundabout2 ) receptors in pre-crossing commissural axons in the spinal cord , thereby allowing midline crossing ( Sabatier et al . , 2004 ) . In Drosophila , Commissureless ( Comm ) antagonizes Slit-Robo1 repulsion by preventing the trafficking of the Robo1 receptor to the growth cone , instead diverting newly synthesized Robo1 into the endocytic pathway ( Kidd et al . , 1998b; Keleman et al . , 2002 , 2005 ) . As commissural axons approach the midline , Comm expression is high , allowing axons to cross the midline ( Keleman et al . , 2002 ) . Once the midline is reached , Comm is down regulated , restoring Robo1-dependent Slit sensitivity and ensuring that commissural axons do not re-cross the midline . Accordingly , loss of Robos or Slits can cause axons to ectopically cross the midline , while loss of Comm or Robo3/Rig1 prevents commissural axons from crossing ( Tear et al . , 1996; Kidd et al . , 1998a , 1999; Long et al . , 2004; Sabatier et al . , 2004 ) . In Drosophila , the three members of the Robo receptor family ( Robo1 , Robo2 , and Robo3 ) cooperate to control multiple aspects of axon guidance during embryonic development , including midline repulsion of axons and the formation of longitudinal axon pathways at specific mediolateral positions within the nerve cord . Although Robo2 contributes to promoting midline repulsion , gain-of-function genetic experiments suggest that in some contexts Robo2 can also promote midline crossing ( Rajagopalan et al . , 2000; Simpson et al . , 2000b ) . More recently , endogenous roles for robo2 in promoting midline crossing were identified during the guidance of foreleg gustatory neurons in the adult , as well as during the guidance of interneurons in the embryonic CNS ( Mellert et al . , 2010; Spitzweck et al . , 2010 ) . Robo2's pro-crossing role in the embryo is highlighted in frazzled and Netrin mutant backgrounds , in which midline attraction is partially compromised ( Spitzweck et al . , 2010 ) . In the absence of Netrin-dependent midline axon attraction , loss of robo2 ( but not robo1 or robo3 ) leads to a dramatic disruption in midline crossing that is far more severe than the complete loss of Netrins , indicating that robo2 likely acts in parallel to Netrin-Fra to promote midline crossing ( Spitzweck et al . , 2010 ) . In a complementary series of gain-of-function experiments using a panel of chimeric receptors comprising different regions of Robo1 and Robo2 fused together , we have previously shown that Robo2's ability to promote ectopic midline crossing correlates with the presence of the first and second immunoglobulin-like domains ( Ig1 and Ig2 ) within its extracellular domain ( Evans and Bashaw , 2010b ) . Consistent with these observations , replacing endogenous Robo2 by homologous recombination with chimeric receptors , in which the cytoplasmic domains of the Robo1 and Robo2 receptors were exchanged , reveals that the Robo2-1 chimeric receptor ( containing the extracellular region of the Robo2 receptor ) can rescue the commissural guidance defect observed in Netrin , robo2 mutants more effectively than the reciprocal chimeric receptor ( Spitzweck et al . , 2010 ) . However , the mechanism by which Robo2 promotes midline crossing remains unclear . We can envision two alternative models that could account for Robo2's role in promoting midline crossing of commissural axons . First , Robo2 may act as an attractive receptor to signal midline attraction in response to a ligand produced by midline glia , analogous to Frazzled/Deleted in Colorectal Cancer ( DCC ) 's role in Netrin-dependent midline attraction . Indeed , a role for Robo2 in mediating attractive responses to Slit has been described in the context of muscle cell migration ( Kramer et al . , 2001 ) . Alternatively , Robo2 may antagonize Slit-Robo1 repulsion by preventing Robo1 from signaling in response to midline-derived Slit , similar to the proposed role of Robo3/Rig-1 in pre-crossing commissural axons in the vertebrate spinal cord ( Figure 1 ) . Although Comm is an essential regulator of Robo1 activity in Drosophila , low levels of Robo1 escape Comm-dependent sorting and can be detected on commissural axons , raising the question of whether and how the activity of these Robo1 receptors is regulated ( Kidd et al . , 1998a ) . 10 . 7554/eLife . 08407 . 003Figure 1 . Robo2 commissural guidance defects are rescued by a Robo2 BAC transgene . ( A–E ) Stage 17 Drosophila embryos of the indicated genotypes stained with anti-HRP antibodies to label all CNS axons . ( F–J ) Stage 15–16 embryos of the indicated genotypes carrying eg-GAL4 and UAS-TauMycGFP transgenes , stained with anti-HRP and anti-GFP antibodies . Anti-GFP labels cell bodies and axons of the eagle neurons ( EG and EW ) in these embryos . ( A and F ) Embryos heterozygous for both frazzled ( fra ) and robo2 display a wild-type arrangement of longitudinal and commissural axon pathways , and axons of the EW neurons cross the midline in the posterior commissure in 100% of segments ( arrowhead ) . ( B and G ) robo2 mutants ( robo2123/robo233 ) display a mildly disorganized axon scaffold , but no detectable defects in EW crossing . ( C and H ) : fra mutants ( fra3/fra4 ) display thinning commissures indicative of decreased midline crossing , and the EW axons fail to cross the midline in 30% of abdominal segments ( arrowhead with asterisk ) . ( D and I ) Simultaneous removal of robo2 and fra ( robo2123 , fra3/robo2135 , fra4 ) strongly enhances the midline crossing defects seen in fra single mutants . ( E and J ) Midline crossing is partially restored in robo2 , fra double mutants carrying one copy of an 83 . 9-kb robo2 BAC transgene . The overall organization of the axon scaffold approaches that seen in fra single mutants , and EW axon crossing defects are significantly rescued , although not completely restored to the level seen in fra single mutants . Histogram quantifies EW midline crossing defects in the genotypes shown in ( F–J ) . Error bars represent s . e . m . n , number of embryos scored for each genotype ( p<0 . 0001 ) . Bottom right: Two possible models for how Robo2 might promote midline crossing of commissural axons . Left , Robo2 may act as a midline attractive receptor to promote midline crossing in response to an unidentified ligand , analogous to Fra's role as an attractive Netrin receptor . Right , Robo2 may antagonize canonical Slit-Robo1 repulsive signaling to down-regulate midline repulsion and thus allow Robo1-expressing axons to cross the midline . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 003 Here , we show that in addition to its cell-autonomous role in midline repulsion , Robo2 acts non-autonomously to promote midline crossing by inhibiting canonical Slit-Robo1 repulsion and offer insights into the molecular and cellular mechanisms underlying this activity of Robo2 . We find that the cytoplasmic domain of Robo2 is dispensable for its pro-crossing role , suggesting that Robo2 does not transduce a midline attractive signal , and that Robo2 over-expression can suppress comm mutants , supporting a model in which Robo2 antagonizes Slit-Robo1 repulsion . Moreover , Robo2 can bind to Robo1 in Drosophila embryonic neurons , and this biochemical interaction , like Robo2's pro-crossing role , correlates with the presence of Ig1 and Ig2 . Surprisingly , we observe that Robo2 is able to promote midline crossing of axons non-cell autonomously when mis-expressed in midline cells , and we further show that Robo2 is expressed in midline glia and neurons during the early stages of commissure formation . Finally , we find that restoring Robo2 expression in midline cells can rescue midline crossing of axons in robo2 , fra double mutants and that this rescue activity is dependent on Ig1 and Ig2 . Together , our results indicate that Robo2 acts non-autonomously to bind to Robo1 and prevent Slit-Robo1 repulsion in pre-crossing commissural axons . This model accounts for Robo2's seemingly paradoxical roles in both promoting and inhibiting midline crossing and explains how the small amount of Robo1 present on pre-crossing commissural axons might be prevented from responding to Slit . If Robo2 were to act as a midline attractive receptor ( Figure 1 , model 1 ) , its cytoplasmic domain would likely be required for midline attraction . To test whether the Robo2 cytoplasmic domain contributes to its pro-crossing activity , we tested whether a truncated Robo2 receptor lacking its cytoplasmic domain ( Robo2∆C ) could promote midline crossing when mis-expressed in embryonic neurons . We found that , as with full-length Robo2 , pan-neural mis-expression of Robo2∆C ( with elav-GAL4 ) produced strong ectopic crossing of FasII-positive axons in the embryonic CNS ( Figure 2 ) . Indeed , the Robo2∆C mis-expression phenotype was stronger than full-length Robo2 . In contrast , pan-neural over-expression of Robo1 did not generate ectopic crossing ( Figure 2 ) . In these experiments , all UAS-Robo transgenes are expressed from the same genomic insertion site in order to ensure that they are expressed at similar levels . Importantly , the pro-crossing activity of Robo2∆C is unlikely to be caused solely by a dominant-negative effect of the truncated receptor , as a similarly truncated form of Robo1 ( Robo1∆C ) has a qualitatively weaker ectopic crossing phenotype when combined with elav-GAL4 ( Figure 2 ) . Robo2∆C expression , unlike Robo1∆C , leads to ectopic crossing of all of the ipsilateral FasII axon bundles and also results in many segments exhibiting a slit-like phenotype ( Figure 2 ) . Due to the strong phenotypic effects of the targeted insertion lines of Robo1∆C and Robo2∆C , we also compared the phenotypes generated by lower levels of expression of the two truncated receptors using standard UAS inserts and observed that Robo2∆C is significantly more potent at driving ectopic midline crossing than comparable levels of the Robo1∆C receptor ( Figure 3 ) . Together these observations indicate that the pro-crossing activity of Robo2 is independent of the cytoplasmic domain and argue against the idea that Robo2 promotes midline crossing by signaling attraction . 10 . 7554/eLife . 08407 . 004Figure 2 . Robo2 can promote midline crossing independent of its cytoplasmic domain . ( A–E ) Stage 17 embryos carrying elav-GAL4 and the indicated UAS-Robo transgenes , stained with anti-HRP ( magenta ) and the longitudinal pathway marker anti-FasciclinII ( FasII; green ) . ( A ) Embryos carrying elav-GAL4 alone exhibit a wild-type arrangement of axon pathways , including distinct anterior and posterior commissures and three FasII-positive longitudinal pathways that do not cross the midline . ( B ) In elav-GAL4/UAS-Robo1 embryos , commissure formation is strongly impaired , and no ectopic midline crossing of FasII-positive axons is observed . ( C ) Mis-expression of Robo2 with elav-GAL4 produces a biphasic phenotype , where some segments appear nearly commissureless ( arrowhead with asterisk ) while others exhibit ectopic crossing reminiscent of robo1 mutants ( arrow ) . See Figure 5 for quantification of ectopic crossing in elav-GAL4/UAS-Robo2 embryos . ( D and E ) Mis-expression of truncated forms of Robo1 ( Robo1∆C ) or Robo2 ( Robo2∆C ) with elav-GAL4 induces ectopic crossing in 100% of segments , although the Robo2∆C mis-expression phenotype is qualitatively more severe than Robo1∆C . In elav-GAL4/UAS-Robo1∆C embryos ( D ) only the medial FasII pathway crosses the midline and the axon scaffold overall exhibits a robo1-like appearance , while in elav-GAL4/UAS-Robo2∆C embryos ( E ) all three FasII-positive pathways collapse at the midline in nearly every segment and the axon scaffold appears slit-like . All UAS-Robo transgenes shown here were inserted into the same genomic location ( 86FB ) to ensure equivalent expression levels . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 00410 . 7554/eLife . 08407 . 005Figure 3 . Comparison of Robo1∆C and Robo2∆C gain of function activities . Since the effects of expressing ∆C transgenes in the 86Fb insertion site are too potent to allow quantitative comparison , we used traditional UAS insertion lines that are expressed at lower and comparable levels ( right panels , anti-Myc is shown in green and anti-FasII in magenta ) to compare activities of Robo1∆C and Robo2∆C . In embryos expressing only an elav-GAL4 transgene ( top left ) , FasII axons appear wild-type and remain ipsilateral . Mis-expression of Robo2 leads to a high level of ectopic crossing . Robo2∆C expression results in a much greater degree of ectopic midline crossing than does Robo1∆C . Segments with ectopic midline crossing of FasII axons are quantified on the right . Significance was assessed by multiple comparisons using the Student's t-test and a Bonferonni correction ( p < 0 . 001 ) . Error bars represent s . e . m . n , number of embryos scored for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 005 Slit is the canonical ligand for Robo family receptors , and all three Drosophila Robos can bind to the single Drosophila Slit ( Howitt et al . , 2004 ) . To test whether Robo2's pro-crossing activity depends on its ability to bind Slit , we deleted the canonical Slit-binding domain ( the first immunoglobulin-like domain: Ig1 ) from Robo2 . As predicted by previous in vitro binding studies using Drosophila Robo1 ( Brose et al . , 1999; Fukuhara et al . , 2008 ) , we found that deleting Ig1 from Robo2 prevented Slit binding in cultured Drosophila cells ( Figure 4 and Figure 4—figure supplement 1 ) . Pan-neural over-expression of Robo2 produces a phenotype in which some axons are repelled from the midline , and some axons ectopically cross the midline , reflecting Robo2's two opposing activities in regulating midline crossing . As expected , deleting the Ig1 domain prevents Robo2 from signaling midline repulsion in vivo , both broadly in all neurons ( when expressed pan-neurally with elav-GAL4 ) ( Figure 5 ) and in a subset of commissural neurons ( the EW neurons , labeled by eg-GAL4 ) ( Figure 4 ) , confirming that Slit binding is required for Robo2-mediated repulsion . In contrast , we found that the Robo2 receptor lacking Ig1 retained a partial ability to promote ectopic midline crossing of FasII-positive axons , indicating that the pro-crossing activity of Robo2 does not strictly depend on its ability to bind Slit ( Figure 5 ) . Notably , the ectopic crossing phenotype produced by Robo2∆Ig1 mis-expression was significantly weaker than that caused by mis-expression of full-length Robo2 ( Figure 5 ) . This result suggests that the Ig1 domain contributes to , but is not strictly required for , promotion of midline crossing by Robo2 . 10 . 7554/eLife . 08407 . 006Figure 4 . Slit binding and Robo gain of function . ( A–E ) Slit-conditioned media was collected and used to treat cells expressing the indicated HA-tagged receptors . Receptor expression is shown with anti-HA in the top panels ( magenta ) and anti-Slit staining is shown in the bottom panels ( green , A′–E′ ) . Robo1 ( A ) , Robo2 ( B ) , and Robo2∆Ig2 ( D ) bind efficiently to Slit , while little to no binding is detected in cells expressing Robo2∆Ig1 ( C ) or Robo2∆Ig1+2 ( E ) . ( F–J ) Stage 16 embryos expressing the indicated transgene in the Eg commissural interneurons . HRP labels the axon scaffold ( magenta ) and anti-GFP labels the Eg neurons . The percentages under each panel indicate the percentage of EW axons that fail to cross the midline in each condition . Expression of Robo1 ( F ) , Robo2 ( G ) , and Robo2∆Ig2 ( I ) all lead to strong disruption of midline crossing , while expression of Robo2∆Ig1 ( H ) , and Robo2∆Ig1+2 ( J ) result in little to no crossing defects . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 00610 . 7554/eLife . 08407 . 007Figure 4—figure supplement 1 . Quantification of relative fluorescence intensity of HA and Slit antibody staining in S2R+ cells transfected with HA-tagged Robo2 proteins , and treated with Slit-conditioned media . Fluorescence intensities are normalized to the average pixel intensity of untransfected cells from the same slide . The values from three images per Robo2 variant were averaged; error bars represent the standard error of the mean . * denotes p < 0 . 03 and ** denotes p < 0 . 01 from a Student's t-test . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 00710 . 7554/eLife . 08407 . 008Figure 4—figure supplement 2 . Robo2 transgenes are localized to axons and expressed at equivalent levels in vivo and are present at the surface of S2R+ cells in vitro . ( A–E ) Embryos carrying elav-GAL4 and the indicated UAS-Robo2 transgenes were stained with anti-HA antibodies and imaged via confocal microscopy . Staining and imaging conditions were identical for all samples . ( A–D ) Representative images of embryos expressing each transgene and stained with anti-HA . All of the Robo2 variants are localized to axons when expressed pan-neurally with elav-GAL4 . ( E ) Quantification of pixel intensity for each transgenic line . Confocal max projections through the entire neuropile were collected for three stage 16 embryos for each line , and average pixel intensity was measured across five 25-pixel regions within the longitudinal axon pathways for each embryo . Bar graph shows average pixel intensity across the three embryos for each line . Error bars indicate standard deviation . Average pixel intensity values were not significantly different for any of the four transgenic lines by Student's t-test ( p > 0 . 2 for all comparisons ) . ( F ) S2R+ cells transfected with the indicated Robo2 constructs were permeabilized and stained with anti-HA and anti-tubulin antibodies . No differences were observed in the localization or expression of the different HA-Robo2 variants . Staining and imaging conditions were identical for all samples . ( G ) S2R+ cells transfected with the indicated Robo2 constructs were incubated with anti-HA and anti-tubulin antibodies at 4°C for 30 min , in the absence of detergent . All Robo2 proteins were robustly detected at the cell surface by this method , with no noticeable differences in localization or staining intensity; no tubulin signal was detected , confirming that cells were not permeabilized . Staining and imaging conditions were identical for all samples . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 00810 . 7554/eLife . 08407 . 009Figure 5 . Robo2's pro-crossing activity depends on its Ig2 domain . ( A–F ) Stage 17 embryos carrying elav-GAL4 and the indicated UAS-Robo transgenes , stained with anti-HRP and anti-FasII . ( A ) Embryos carrying elav-GAL4 alone exhibit a wild-type arrangement of axon pathways , including three FasII-positive longitudinal pathways that do not cross the midline . ( B ) Robo1 does not promote midline crossing of FasII-positive axons when misexpressed in all neurons with elav-GAL4 . ( C ) Misexpression of full-length Robo2 induces ectopic midline crossing in over 80% of segments ( arrow ) . ( D ) Deleting the Ig1 domain ( Robo2∆Ig1 ) disrupts Slit binding but does not completely prevent Robo2 from promoting midline crossing . ( E and F ) Robo2 receptors lacking the Ig2 domain ( Robo2∆Ig2 ) or both the Ig1 and Ig2 domains ( Robo2∆Ig1+2 ) are unable to promote ectopic midline crossing above background levels ( both are comparable to Robo3; see histogram ) . Schematics show domain composition of receptors shown in ( A–F ) . All UAS-Robo transgenes shown here were inserted into the same genomic location ( 86FB ) to ensure equivalent expression levels . Histogram quantifies ectopic midline crossing in the indicated genotypes . Significance was assessed by multiple comparisons using the Student's t-test and a Bonferonni correction ( p < 0 . 01 ) . Error bars represent s . e . m . n , number of embryos scored for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 009 We have previously shown that Robo2's pro-crossing activity is conferred at least in part by its Ig2 domain: replacing the Ig1–Ig2 region of Robo1 with the equivalent region from Robo2 ( Robo1R2Ig1+2 ) confers Robo2-like pro-crossing activity to Robo1 ( Evans and Bashaw , 2010b ) . Further , replacing Ig1–Ig2 of Robo2 with Robo1 Ig1–Ig2 ( Robo2R1Ig1+2 ) abolishes its pro-crossing activity ( Evans and Bashaw , 2010b ) . To directly test whether Ig2 is necessary for Robo2 to promote midline crossing , we generated a Robo2 receptor lacking Ig2 but with all other Ig domains intact ( Robo2∆Ig2 ) . We found that deleting Robo2's Ig2 domain did not interfere with Slit binding ( Figure 4 ) , nor did it affect Robo2's ability to signal repulsion in commissural neurons ( Figure 4 ) . However , deletion of Robo2's Ig2 domain strongly disrupted its ability to promote ectopic midline crossing ( Figure 5 ) . The low level of ectopic crossing induced by Robo2∆Ig2 was indistinguishable from that caused by Robo3 , a related receptor that does not share Robo2's pro-midline crossing activity ( Figure 5 ) . These results contrast with those observed with Robo2∆Ig1 , which lacks Slit-dependent midline repulsive activity but retains some pro-midline crossing activity ( Figures 4 , 5 ) . These data indicate that Robo2's Ig2 domain is essential for promoting midline crossing when Robo2 is mis-expressed in all neurons . In these experiments , all UAS-Robo transgenes are expressed from the same genomic insertion site in order to ensure that they are expressed at similar levels . In addition , we assayed the protein localization and expression levels of Robo2 and its deletion variants and observed comparable surface expression in cultured S2R+ cells in vitro , as well as comparable expression levels and localization in CNS axons in vivo ( Figure 4—figure supplement 2 ) . Midline crossing is strongly reduced in robo2 , fra double mutants , and pan-neural mis-expression of Robo2 can promote ectopic midline crossing . However , it is unclear whether Robo2 acts autonomously or non-autonomously to promote midline crossing . The ectodomain-dependent nature of Robo2's pro-crossing activity and our pan-neural mis-expression assays do not distinguish between these possibilities . Indeed , elav-GAL4 is transiently expressed in midline glia as well as post-mitotic neurons , preventing us from ruling out a non-neuronal contribution to the observed mis-expression phenotypes . Notably , we have never observed a clearly cell-autonomous pro-crossing phenotype caused by Robo2 . We have previously described a series of experiments in which we mis-expressed full-length , truncated , or chimeric receptor variants of Robo1 and Robo2 in a subset of ipsilateral neurons ( the apterous neurons , labeled by ap-GAL4 ) ( Evans and Bashaw , 2010b ) . In contrast to the very different phenotypes caused by pan-neural mis-expression of these two truncated receptors ( where Robo2∆C is much more potent at inducing midline crossing than Robo1∆C ) , Robo1∆C and Robo2∆C induce similar low levels of ectopic crossing when expressed in the apterous neurons ( Figure 6A ) . We interpret this as a cell-autonomous dominant-negative effect of these truncated receptors . 10 . 7554/eLife . 08407 . 010Figure 6 . Robo2 acts cell non-autonomously to promote midline crossing in ipsilateral neurons . ( A ) Stage 17 embryos stained with anti-HRP ( magenta ) and anti-GFP ( green ) antibodies . Anti-GFP labels the apterous ( ap ) cell bodies and axons , which normally project ipsilaterally . Mis-expression of Robo2ΔC in ap neurons results in a mild ectopic crossing phenotype , which is similar to the effect of Robo1ΔC ( arrowheads with asterisks ) . Segments with ectopic crossing of ap axons are quantified in the histogram . ( B ) Stage 17 embryos stained with anti-FasII ( magenta ) and anti-GFP ( green ) antibodies . Anti-GFP labels the axons of hb9-GAL4 expressing cells . Mis-expression of Robo2 with hb9-GAL4 results in a lateral shift of hb9-Gal4+ axons , and causes FasII+ axons that do not express hb9-GAL4 to ectopically cross the midline . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 010 Full-length Robo2 is unable to autonomously promote midline crossing of the apterous axons ( Evans and Bashaw , 2010b ) . Instead , Robo2 mis-expression redirects apterous axons to lateral regions of the neuropile . In the course of examining this lateral positioning activity of Robo2 , we mis-expressed Robo2 in a second class of longitudinal interneurons: those labeled by Hb9-GAL4 . Intriguingly , we observed two distinct phenotypes in embryos where Hb9-GAL4 drives Robo2 expression . First , Hb9-positive axons were shifted to more lateral positions within the neuropile . Second , Hb9-negative FasII-positive axons ectopically crossed the midline ( Figure 6B ) . These results suggest that Robo2 can autonomously specify the lateral position of Hb9-positive axons while non-autonomously instructing FasII axons to cross the midline . We note that Hb9-GAL4 expression initiates earlier than ap-GAL4 and includes a larger number of neurons , including some located near the CNS midline ( such as the RP motor neurons ) , suggesting the possibility that early midline-proximal expression of Robo2 accounts for the non-autonomous effect observed with Hb9-GAL4 . To more explicitly test whether Robo2 can promote midline crossing non-autonomously , we used slit-GAL4 to drive Robo2 expression in midline glia and neurons . We found that mis-expression of Robo2 or Robo2∆C in midline cells caused many FasII-positive axons which do not express slit-GAL4 to ectopically cross the midline , confirming that Robo2 can act non-autonomously to promote midline crossing of axons , and that this effect does not depend on the cytoplasmic domain ( Figure 7 ) . We observed a significantly milder effect with mis-expression of Robo1 , suggesting that Robo2's non-cell autonomous activity is not solely a consequence of Slit titration ( Figure 7 ) . Moreover , this non-cell autonomous activity of Robo2 appears to be Ig1/Ig2-dependent , as Robo1R2Ig1+2 but not Robo2R1Ig1+2 promoted strong ectopic midline crossing when expressed using slit-GAL4 ( Figure 7 ) . In addition , expression of Robo2 variants missing either Ig1 or Ig2 with slit-Gal4 did not result in any ectopic midline crossing ( Figure 7 ) . The requirement for both Ig1 and Ig2 in this context contrasts with our findings with pan-neural mis-expression , in which Robo2∆Ig1 retained some pro-crossing activity . However , it is worth noting that the phenotype generated by elav-GAL4 mis-expression of Robo2 is stronger than that generated by slit-GAL4 , perhaps because slit-GAL4 is expressed in a much smaller number of cells . 10 . 7554/eLife . 08407 . 011Figure 7 . Robo2 can promote crossing non cell-autonomously . ( A–D ) Stage 17 embryos stained with anti-HRP ( magenta ) and anti-FasII ( green ) . ( A and B ) Mis-expression of Robo1 ( A ) in midline cells using slit-GAL4 results in a mild ectopic crossing phenotype . In contrast , mis-expression of Robo2 ( B ) produces a much stronger effect , as indicated by quantification of ectopic FasII crossing in the histogram ( I ) . ( C and D ) Mis-expression of either Robo2∆Ig1 ( C ) or Robo2∆Ig2 ( D ) with slit-GAL4 does not produce ectopic crossing of FasII axons . ( E and F ) Consistent with requirement of Robo2's first two IG domains , the chimeric protein Robo1R2IG ( 1+2 ) produces an ectopic crossing phenotype ( F ) , whereas Robo2R1 ( IG1+2 ) has no effect ( E ) . ( G and H ) Mis-expression of Robo2∆C with slit-GAL4 also results in severe ectopic crossing defects ( H ) that are much stronger than those observed with Robo1∆C ( G ) , as indicated by quantification of ectopic FasII crossing ( I ) and fused commissures observed in anti-HRP stained embryos ( J ) . All UAS-Robo transgenes were inserted into the same genomic location ( 86FB ) . Significance was assessed by multiple comparisons using the Student's t-test and a Bonferonni correction ( p < 0 . 001 ) . Error bars represent s . e . m . n , number of embryos scored for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 011 Robo2 can promote midline crossing when expressed in a subset of embryonic neurons and glia , and endogenous robo2 contributes to midline crossing of commissural axons . During embryogenesis , robo2 expression is dynamically regulated: it is broadly expressed in neurons during early stages of CNS development , including transient expression in a number of ipsilateral pioneer neurons , and later becomes restricted to neurons whose axons form longitudinal pathways in the lateral regions of the neuropile ( Simpson et al . , 2000a ) . To gain additional insight into Robo2's role in promoting midline crossing of commissural neurons , we examined robo2 mRNA and protein expression in embryos during the early stages of axon path finding , when the first commissural axons are crossing the midline ( stages 12–13 ) . Using fluorescent mRNA in situ hybridization , we were able to detect robo2 mRNA expression in cells labeled by slit-GAL4 in late stage 12 embryos , around the time that pioneer commissural axons are crossing the midline ( Figure 8B ) . robo2 mRNA expression persists through the end of stage 13 , but is no longer detectable by stage 14; thus , midline expression of robo2 coincides with the time when most commissural axons are crossing the midline ( Figure 8 , Figure 8—figure supplement 1 ) . Moreover , a robo2-GAL4 enhancer-trap insertion is expressed in midline glia at this time , as detected by anti-GFP staining in robo2-GAL4 , UAS-TauMycGFP embryos ( Figure 8A ) . Expression of UAS-HARobo2 with robo2-Gal4 and detection of transgenic Robo2 with anti-HA reveals an expression pattern that closely resembles the endogenous pattern of Robo2 protein , suggesting that the robo2-GAL4 faithfully reports Robo2 expression ( data not shown ) . In addition , we could detect weak expression of Robo2 protein produced by an HA-tagged knock-in allele of robo2 ( Spitzweck et al . , 2010 ) in a subset of slit-GAL4 expressing cells at stage 12 , confirming that Robo2 protein is produced in midline cells during the stages of commissural axon path finding , and raising the possibility that Robo2 endogenously acts in these cells to promote midline crossing of commissural axons ( Figure 8B ) . 10 . 7554/eLife . 08407 . 012Figure 8 . robo2 is expressed in midline cells during commissural axon path finding , and over-expressing robo2 with slit-GAL4 restores midline crossing in robo2 , fra double mutants . ( A ) A robo2-GAL4 enhancer trap that recapitulates robo2's endogenous expression pattern drives UAS-TauMycGFP expression ( green ) in midline cells at stages 12–13 , when many commissural axons cross the midline . Midline glia are labeled by an anti-wrapper antibody ( magenta ) . ( B ) Top: Fluorescent in situ for robo2 mRNA ( green ) . robo2 is transiently expressed in midline glia and neurons ( magenta ) during stage 12 ( arrows ) . This in situ signal is not observed in robo2 mutant embryos confirming the specificity of the mRNA detection ( right ) . ( B ) Bottom: Robo2 protein is expressed in midline cells during the stages of commissural axon path finding , as shown by the expression pattern of a HA-tagged robo2 cDNA knock-in allele ( robo2HArobo2 ) . Stage 12 embryos carrying robo2HArobo2 , slit-GAL4 , and UAS-TauMycGFP show HARobo2 expression in slitGAL4-expressing cells ( arrows ) , whereas this signal is not detected in control embryos ( right ) . ( C–G ) Stage 14 embryos of the indicated genotypes stained with anti-HRP antibodies to label all CNS axons . The absence of PC was scored in abdominal segments A1–A8 ( arrows indicate examples of missing commissures ) . The PC defects of robo2 , fra double mutants ( D ) are significantly rescued by over-expressing UAS-Robo2 with slit-GAL4 ( E ) , whereas over-expression of UAS-Robo2ΔIg1 ( F ) or UAS-Robo2ΔIg2 ( G ) has no effect . Embryos were scored blind to genotype . Significance was assessed by one-way ANOVA followed by multiple comparisons using the Student's t-test and a Bonferonni correction ( p < 0 . 01 ) . Error bars represent s . e . m . n , number of embryos scored for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 01210 . 7554/eLife . 08407 . 013Figure 8—figure supplement 1 . robo2 mRNA is transiently expressed in midline cells . Fluorescent in situ for robo2 mRNA ( green ) . robo2 is transiently expressed in midline glia ( magenta ) during Stages 12 and 13 but is no longer detected there by Stage 14 . Midline glia are labeled with an antibody to Wrapper . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 013 Robo2 can promote midline crossing non-autonomously , and endogenous robo2 expression can be detected in slit-GAL4-expressing cells as well as in contralateral and ipsilateral neurons during the initial stages of commissure formation ( Figure 8 and data not shown ) . Our ability to partially rescue midline crossing in robo2 , fra double mutants with the robo2 BAC confirms that this is a robo2-specific phenotype , but does not address in which cells robo2 acts to instruct commissural axons to cross the midline . To address this question , we attempted to rescue robo2's endogenous pro-crossing activity by restoring robo2 expression in restricted subsets of cells in robo2 , fra double mutants . We first expressed Robo2 in the commissural EW neurons ( using eg-GAL4 ) in robo2 , fra double mutants . We found that neither full-length Robo2 nor Robo2ΔC can rescue midline crossing when expressed autonomously in the EW neurons ( Figure 9 ) , suggesting that Robo2 may not act cell autonomously to promote midline crossing . We next attempted to rescue midline crossing in robo2 , fra double mutants by expressing Robo2 using slit-GAL4 . Strikingly , we found that driving Robo2 expression in these cells significantly restores posterior commissure formation , as measured by anti-HRP staining ( Figure 8C–G ) . Furthermore , this effect is dependent on Ig1 and Ig2 ( Figure 8C–G ) . Cell-type specific loss of function experiments will be necessary to confirm the site of Robo2's endogenous activity , and our attempts to recapitulate the robo2 , fra phenotype by over-expression of RNAi transgenes have so far been unsuccessful , likely because of the well known difficulty of achieving sufficient knockdown in embryonic stages . Nevertheless , our results suggest that Robo2 promotes midline crossing non-cell autonomously , and may act in midline glia and neurons , where it is expressed during the stages of commissural axon path finding , to promote midline crossing of commissural axons . 10 . 7554/eLife . 08407 . 014Figure 9 . Robo2 cannot rescue midline crossing cell autonomously . ( A–D ) Stage 16 embryos of the indicated genotypes stained with anti-HRP ( magenta ) and anti-GFP ( green ) antibodies . Anti-GFP labels the EG and EW cell bodies and axons . EW crossing defects in robo2 , fra double mutants ( A ) are not rescued by eg-GAL4 mediated over-expression of UAS-Robo2 ( B ) , UAS-Robo2ΔIG1 ( C ) , or UAS-Robo2ΔC ( D ) , suggesting that Robo2 cannot act cell autonomously to promote midline crossing . Segments with non-crossing EW axons are indicated by arrowheads with asterisks . Significance was assessed by multiple comparisons using the Student's t-test and a Bonferonni correction . No significant differences between any of the genotypes were observed ( p > 0 . 3 ) . Error bars represent s . e . m . n , number of embryos scored for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 014 Robo2 can promote midline crossing of axons independently of its cytoplasmic domain and its Slit-binding Ig1 domain , suggesting that Robo2 does not promote crossing by acting as an attractive signaling receptor or solely by titrating Slit . Does Robo2 antagonize Slit-Robo1 repulsion through another mechanism ? In order to test this hypothesis , we took advantage of comm mutants , which provide a genetic background in which hyperactive Slit-Robo1 signaling completely prevents midline crossing . In comm mutants , endogenous Robo1 is inappropriately trafficked to the growth cone plasma membrane in pre-crossing commissural axons , triggering premature Slit repulsion and preventing commissure formation . We reasoned that if Robo2 antagonizes Slit-Robo1 repulsion , then Robo2 mis-expression might restore midline crossing in comm mutant embryos . Indeed , pan-neural mis-expression of Robo2 with elav-GAL4 significantly restored commissure formation in comm mutant embryos ( Figure 10 ) . To specifically test the Ig1/Ig2-dependence of Robo2's pro-crossing activity in this assay , we mis-expressed our Ig1+2 chimeric receptors ( Robo1R2Ig1+2 and Robo2R1Ig1+2 ) with elav-GAL4 in comm mutant embryos . We found that pan-neural mis-expression of Robo1R2Ig1+2 in comm mutant embryos strongly suppressed the commissureless phenotype and restored midline crossing of many axons , as assayed by anti-HRP antibody staining , while mis-expression of Robo2R1Ig1+2 had a much milder effect ( Figure 10 ) . These results suggest that Robo2 promotes midline crossing in an Ig1/Ig2-dependent manner by antagonizing canonical Slit-Robo1 repulsion . 10 . 7554/eLife . 08407 . 015Figure 10 . Robo2 receptors that promote midline crossing suppress comm mutants . ( A ) Schematic diagram of the two chimeric receptors shown in ( B and C ) . Robo1 sequences are depicted in blue and Robo2 sequences are depicted in yellow . ( B and C ) Stage 16 embryos of the indicated genotype stained with anti-HRP to visualize CNS axons and anti-HA to visualize the epitope tagged chimeric receptor . Single channel images of HRP and HA are presented to the right of the color panels . Expression of the HA-Robo2R1Ig1−2 chimeric receptor in a comm mutant background ( B ) does not restore commissure formation , while expression of the reciprocal HA-Robo1R2Ig1+2 chimeric receptor ( C ) strongly suppresses the comm mutant phenotype . ( D and E ) Quantification of the average number of commissures per embryo in comm mutants expressing the indicated HA-tagged receptor transgenes in either all neurons using elav-GAL4 ( D ) or in midline cells using slit-GAL4 ( E ) . Significance was assessed by one-way ANOVA followed by multiple comparisons using the Student's t-test and a Bonferonni correction ( p < 0 . 0001 ) ( p < 1 . 0 e−10 ) . Error bars represent s . e . m . n , number of embryos scored for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 015 We were also able to suppress the comm mutant phenotype by over-expressing Robo2 using slit-GAL4 , and this effect was fully dependent on both Ig1 and Ig2 of Robo2 ( Figure 10 ) . This is consistent with our observations that Robo2 can act non-cell autonomously to promote ectopic midline crossing ( Figure 7 ) and rescue midline crossing defects ( Figure 8 ) in an Ig1/2-dependent manner . Of note , the suppressive effect of Robo2 expression in comm mutants is much greater when expressed in midline cells than when expressed pan-neurally ( Figure 10 ) . This is likely because when expressed pan-neurally , in addition to its pro-crossing activity , full-length Robo2 also has repulsive activity . In contrast , when expressed in midline cells , Robo2 would be unable to act as a repulsive receptor . As shown above , Robo2 is able to antagonize Slit-Robo1 repulsion in an Ig1/Ig2-dependent manner . One possibility is that Robo2 may form an inhibitory receptor–receptor complex with Robo1 to prevent it from signaling midline repulsion in response to Slit . If this is the case , we reasoned that we might be able to detect a physical interaction between Robo2 and Robo1 in embryonic protein extracts . To test this idea , we mis-expressed epitope-tagged forms of Robo1 and Robo2 in Drosophila embryonic neurons with elav-GAL4 and looked for physical interactions by co-immunoprecipitation ( Figure 11 ) . We found that Robo1-myc and HA-Robo2 co-immunoprecipitated from embryonic lysates when both were expressed in embryonic neurons ( Figure 11A ) . Interactions were also observed between Robo1 and the closely related Robo3 receptor but not with a similarly tagged and structurally related Fra receptor ( Figure 11A ) . As we would predict from our gain of function experiments , Robo2's ability to bind to Robo1 is independent of its cytoplasmic domain ( Figure 11 , Figure 11—figure supplement 1 ) . Importantly , Robo2's ability to bind Robo1 depends on the Ig1–Ig2 region of Robo2 , as Robo1R2Ig1+2 was readily co-immunoprecipitated with Robo1 , while binding between Robo1 and the reciprocal receptor Robo2R1Ig1+2 was only weakly detected ( Figure 11B ) . Consistently , deleting both of the Ig1 and Ig2 domains from Robo2 results in a diminished interaction with Robo1 in vivo and in vitro ( Figure 11 , Figure 11—figure supplement 1 ) . The fact that not all binding is eliminated when both Ig1 and Ig2 are deleted suggests that other regions of the Robo2 receptor may contribute to the interaction with Robo1 . Thus , we see a correlation between the presence of the Ig1 and Ig2 domains , a biochemical interaction with Robo1 , and pro-crossing activity in the Robo2 receptor . These observations suggest that Robo2 may promote midline crossing through inhibitory interactions with Robo1 , likely mediated at least in part by the Robo2 Ig1 and Ig2 domain . Of note , the Ig2 domain is essential for Robo2's pro-crossing activity but is not required for the interaction with Robo1 or Slit , suggesting the existence of an Ig2-specific activity that is distinct from the ability to bind Robo1 or Slit . One possible mechanism that could explain how receptor–receptor interactions could prevent Robo1 signaling is through blocking the access of Slit to the Ig1 region of Robo1 . Deleting Robo1's Ig1 domain does not significantly attenuate the interaction with Robo2 , but further experiments will be necessary to determine if Robo2 interferes with Robo1's interaction with Slit ( Figure 11 , Figure 11—figure supplement 1 ) . 10 . 7554/eLife . 08407 . 016Figure 11 . Robo2 binds to the Robo1 receptor in vitro and in vivo . ( A–C ) Protein extracts from embryos expressing Robo1-Myc and various HA-tagged receptors in all neurons were immunoprecipitated with anti-Myc antibodies and analyzed by western blot . Immunoprecipitates were probed with anti-HA ( top blots ) and total lysates were compared for HA expression and Myc expression to ensure that equal inputs were analyzed . Representative western blots from multiple experiments are shown . ( A ) Robo1-Myc binds to HARobo1 , HARobo2 , and HARobo3 but not to a HA-tagged Fra receptor ( two exposures are shown ) . Total lysate blots reveal comparable loading with the exception of the Fra negative control in which there is substantially more HA-tagged receptor . ( B ) Robo1-Myc binds efficiently to HARobo2 , HARobo2∆Ig1 , and the HARobo1Robo2 ( IG1-2 ) chimera but not to the reciprocal chimera that has Ig1 and Ig2 domains from Robo1 ( asterisk ) . ( C ) Deletion of either Robo2 Ig1 or Ig2 alone does not substantially affect Robo1 binding , while deleting both domains results in reduced binding ( asterisk ) . See Figure 11—figure supplement 1 for an additional example . ( D ) Cell lysates of S2R+ cells separately transfected for Robo1-Myc or HA-tagged Robo2 variants were mixed , immunoprecipitated with anti-Myc , and analyzed by western blot . In this assay , Robo1-Myc binds efficiently to HARobo2 , HARobo2∆Ig1 , and HARobo2∆Ig2 , and less well to HARobo2∆Ig1+2 ( asterisks ) . In lanes 1–4 , cells were untreated; in lanes 5–8 , cells were treated with Slit-conditioned media before lysing . We note that in addition to detection of the predicted full-length Robo2 receptor with anti-HA , we also routinely detect a smaller ∼80 kD fragment that corresponds to an extracellular domain cleavage product . The size of this fragment is shifted to predictably smaller sizes when Ig1 , Ig2 , or both Ig1 and Ig2 are deleted . We do not currently know , whether this cleavage event is required for Robo2 function in any context , since we have not been able to generate an uncleavable version of the receptor . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 01610 . 7554/eLife . 08407 . 017Figure 11—figure supplement 1 . Robo2 binding to Robo1 does not depend on its cytoplasmic domain or on Robo1's Ig1 domain . ( A–C ) Protein extracts from embryos expressing Robo1-Myc and various HA-tagged receptors in all neurons were immunoprecipitated with anti-Myc antibodies and analyzed by western blot . Immunoprecipitates were probed with anti-HA ( top blot ) and total lysates were compared for HA expression and Myc expression to ensure that equal inputs were analyzed . Representative western blots from multiple experiments are shown . ( A ) Deletion of either Robo2 Ig1 or Ig2 alone does not substantially affect Robo1 binding , while deleting both domains results in reduced binding . ( B ) Extracts from embryos co-expressing either HARobo1 or HARobo2 , and either Robo1∆C-Myc or Robo2∆CMyc were analyzed for interactions . Both of the C-terminal truncation receptor variants can efficiently pull down both HARobo1 and HARobo2 indicating that binding is independent of the cytoplasmic domain . ( C ) Similar experiments to those described above and in the legend to the main Figure 11 indicate that Robo2 does not bind to the Slit-binding Ig1 region of Robo1 . ( D ) Lysates of S2R+ cells expressing HA-tagged Robo2 variants were mixed with lysates of untransfected cells , and immunoprecipitated with anti-Myc as a negative control for the experiment in Figure 11D . Very little Robo2 protein was detected in the immunoprecipitates . ( E ) S2R+ cells were co-transfected with Robo1-Myc and HA-Robo2 and immunoprecipitated with anti-Myc ( middle panel ) or anti-HA ( right panel ) . A strong interaction was detected between Robo1 and Robo2 when the pull-down was performed in either direction . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 017 Our biochemical experiments examining receptor–receptor interactions when the Robo receptors are expressed in all neurons do not distinguish between cis and trans interactions . As we observed that Robo2 is able to non-cell autonomously inhibit Robo1 repulsion and promote midline crossing , we reasoned that we might be able to detect physical interactions between Robo1 and Robo2 receptors when they are presented in trans . We tested this prediction by transfecting Drosophila cultured S2R+ cells with either Robo1-myc or HA-tagged Robo2 and assaying for physical interactions by co-immunoprecipitation . Although we detected strong interactions between Robo1 and Robo2 in co-transfected cells , we could not detect interactions in cells that were transfected separately and mixed together ( Figure 11 , Figure 11—figure supplement 1 and data not shown ) . However , when we mixed the membrane lysates of cells that were transfected separately , we observed that Robo1 readily co-immunoprecipitated Robo2 , in an Ig1-2 dependent manner ( Figure 11 ) . These data suggest that physical interactions can occur between Robo1 and Robo2 receptors that are expressed in different cells and are consistent with the possibility of a physical interaction occurring across cell membranes in vivo . It is important to recognize that binding detected with mixed cell lysates could occur in either cis or trans . Future work should more rigorously evaluate the potential for trans interactions . Our data are consistent with a non-autonomous requirement for Ig1 and Ig2 of Robo2 in antagonizing Robo1 to promote midline crossing . However , the genetic data supporting this model arise from gain of function and rescue experiments using GAL4/UAS over-expression . In order to more rigorously address the endogenous requirement for Robo2 in promoting midline crossing , we generated modified BACs and evaluated the ability of either wild-type Robo2 or Robo2∆Ig2 to restore midline crossing in robo2 , fra double mutants , when expressed under robo2's endogenous control elements . As Ig1 is required for both Robo2's pro-crossing activity and for its repulsive signaling output , the Robo2ΔIg2 variant provides a more specific reagent for testing our model . Therefore , we modified the original Robo2 BAC by recombineering to insert wild-type Robo2 cDNA or Robo2∆Ig2 cDNA , and introduced these BAC transgenes into robo2 , fra double mutants . We determined the rescuing activity of each BAC through two assays: first , by scoring midline crossing of EW axons labeled by eg-GAL4 , and second , by analyzing commissure formation in embryos stained with anti-HRP to label all axons ( Figure 12 ) . 10 . 7554/eLife . 08407 . 018Figure 12 . Robo2's endogenous activity in promoting midline crossing depends on Ig2 . ( A–D ) Stage 16 embryos stained with anti-HRP ( magenta ) and anti-GFP ( green ) antibodies . Anti-GFP labels the EG and EW cell bodies and axons . Arrowheads indicate EW axons that have crossed the midline and arrowheads with asterisks indicate non-crossing EW axons . ( A ) Almost all EW axons cross the midline in robo2 , fra/+ , + double heterozygotes . ( B ) EW crossing defects are observed in 85% of segments in robo2 , fra double mutants . ( C–D ) The FL Robo2 cDNA BAC transgene ( C ) significantly rescues EW crossing , to 66% of segments with defects ( Student's t-test , p < 0 . 001 ) whereas the Robo2ΔIG2 transgene ( D ) does not significantly rescue . Right: Removing one copy of robo2 significantly enhances midline crossing defects in fra mutants . ( E–H ) Stage 14 embryos of the indicated genotypes stained with anti-HRP . Posterior commissures were scored in abdominal segments A1–A8 . Missing posterior commissures are indicated by arrowheads with asterisks . ( E–G ) The posterior commissure defects of robo2 , fra double mutants are significantly rescued by a full-length ( FL ) Robo2 cDNA BAC transgene ( Student's t-test , *p < 0 . 001 ) ( F ) , as well as by a Robo2ΔIG2 BAC ( p < 0 . 0167 ) ( G ) . The Robo2ΔIG2 BAC does not rescue as well as FL Robo2 ( *p < 0 . 0167 ) . All embryos were scored blind to genotype . Significance was assessed by one-way ANOVA followed by multiple comparisons using the Student's t-test and a Bonferonni correction . Error bars represent s . e . m . n , number of embryos scored for each genotype . DOI: http://dx . doi . org/10 . 7554/eLife . 08407 . 018 We found that the ability of the Robo2 BAC to rescue midline crossing defects in robo2 , fra double mutants was strongly impaired by deleting the Ig2 domain . In the EW crossing assay , one copy of the Robo2 FL cDNA BAC provides a significant rescue of robo2 , fra double mutants at stage 16 , whereas one copy of the Robo2ΔIg2 BAC has no effect ( Figure 12A–D ) . Of note , removing one allele of robo2 enhances midline crossing defects in fra mutants , explaining in part the incomplete rescue ( Figure 12 ) . In addition , it is likely that the Robo2 BAC does not contain all of the regulatory elements required for robo2's pro-crossing function , as one copy of the BAC does not restore EW crossing back to the levels of fra mutants heterozygous for robo2 ( Figure 12 ) . We also assessed the ability of the BAC transgenes to rescue midline crossing defects when analyzing all axons using anti-HRP . By this method , we see a robust rescue in posterior commissure ( PC ) formation in robo2 , fra double mutant embryos with one copy of the Robo2 cDNA BAC compared to controls ( Figure 12E–H ) . In contrast , the Robo2ΔIg2 BAC provides a much weaker rescue ( Figure 12G ) . The partial rescue by the Robo2ΔIg2 BAC in this assay suggests that the severe fra , robo2 phenotype is due to the combined requirement for multiple activities of Robo2 , including one that is Ig2-independent . Nevertheless , these data unambiguously reveal an endogenous requirement for Robo2's Ig2 domain during commissural axon guidance . Importantly , the Robo2∆Ig2 BAC fully rescues Robo2's repulsive activity at the midline ( data not shown ) , further demonstrating that the Ig2 domain is specifically required for Robo2 to successfully promote midline crossing , but not for other known activities of the Robo2 receptor . Taken together , these results demonstrate a requirement for Robo2's Ig2 domain in promoting midline crossing when expressed under its endogenous control elements and strongly support the model that Robo2 promotes midline crossing of commissural axons by antagonizing repulsion through an Ig1/Ig2-mediated inhibitory interaction with Robo1 . Given the prominent role that Comm plays in regulating Robo1 receptor expression to prevent premature responses to Slit , it is fair to ask why it is necessary to invoke a second mechanism to down-regulate Robo1 receptor signaling . Indeed , in wild-type animals , there is no obvious requirement for Robo2's pro-crossing activity , at least not at the embryonic midline in the populations of neurons that we have assayed . A requirement for Robo2 in promoting midline crossing in otherwise wild-type animals has been described for the guidance of foreleg gustatory neurons in the adult nervous system , although it is not clear in this context if the same mechanism that we have described is at work ( Mellert et al . , 2010 ) . Nevertheless , a clear endogenous contribution for Robo2 at the embryonic midline can be demonstrated in conditions where attractive guidance cues , such as Netrin , are compromised . One probable explanation for the existence of this second regulatory mechanism is that it confers robustness on the essential process of midline circuit formation , and that this is important to the animal when developmental conditions are not optimal . While Comm is an efficient and potent negative regulator of Robo1 trafficking to the growth cone surface , it is clear that not all Robo1 is prevented from reaching the surface in the presence of Comm . Low levels of Robo1 can be detected on commissural axons by immunostaining and immunoelectron microscopy ( Kidd et al . , 1998a ) . Data from surface labeling experiments indicate that Comm acts on newly synthesized Robo1 , and the question of how Robo1 receptors already present on the plasma membrane prior to the initiation of comm expression might be regulated remains unresolved ( Keleman et al . , 2002 ) . The role of Robo2 may thus be to negatively regulate the low levels of Robo1 that escape Comm-dependent sorting . In addition to the complementary actions of Comm , a cell autonomous regulator of Robo1 trafficking ( Keleman et al . , 2002 , 2005 ) , and Robo2 , a cell non-autonomous inhibitor of Robo1 signaling ( this study ) , it is likely that there are additional levels of regulation that contribute to preventing premature response to midline Slit . In particular , a recent study shows quite convincingly that Comm's role in sorting Robo1 is insufficient to explain how Robo1 activity is limited in pre-crossing commissural axons . Specifically , embryos in which the endogenous Robo1 receptor is replaced with a variant of Robo1 that is insensitive to the sorting activity of Comm by homologous recombination show no defects in midline crossing ( Gilestro , 2008 ) . This observation is in marked contrast to the prediction of the sorting model , in which embryos carrying a Comm-resistant Robo1 receptor would be expected to resemble comm mutants . It will be of great interest to obtain an explanation for this paradoxical finding and to determine what additional targets of Comm could also regulate Slit-dependent repulsion . Our results suggest that Robo2 can inhibit Robo1 activity and that this effect is mediated by receptor–receptor interactions between the Robo2 and Robo1 extracellular domains . Cis-inhibitory interactions , such as those that occur between the transmembrane protein Kekkon 1 and the epidermal growth factor receptor ( EGFr ) ( Ghiglione et al . , 2003 ) , and between ligand and receptor pairs , as in the cases of Ephs/ephrins and Notch/Delta , have been well documented ( del Alamo et al . , 2011; Kao and Kania , 2011; Yaron and Sprinzak , 2012 ) . While we were not able to detect trans interactions by co-immunoprecipitation or by an S2 cell aggregation assay ( data not shown ) , our genetic data strongly suggest that Robo2 acts in trans to inhibit Robo1 signaling . A recent in vitro screen for trans interactions among Drosophila cell surface receptors did not report a direct interaction between Robo1 and Robo2 , suggesting that if trans interactions do occur , they might be mediated by a cofactor ( Ökzan et al . , 2013 ) . Indeed , Slit-dependent trans interactions between Robo1 and Robo2 have been proposed to play a role in the migration of sensory neurons in the Drosophila peripheral nervous system , although in this case Robo2 is thought to promote Slit-Robo repulsive signaling by presenting Slit to Robo receptors expressed in trans ( Kraut and Zinn , 2004 ) . Previous studies have defined growth factor and morphogen receptor regulatory mechanisms that bear some resemblance to the mechanism that we have described here . For example , EGFr signaling and Bone Morphogenetic Protein receptor ( BMPr ) signaling can be attenuated cell non-autonomously by various inhibitory factors , such as Argos for EGFrs and Noggin for BMPrs ( Klein et al . , 2004; Walsh et al . , 2010 ) . In the case of EGFr , receptor signaling is blocked because the soluble inhibitory factor Argos binds to and sequesters the EGF ligand , thereby preventing receptor activation ( Klein et al . , 2008 ) . The mechanism through which Robo2 regulates Robo1 is similar in that it acts cell non-autonomously and that it depends on extracellular interactions , but distinct , since Robo2 does not appear to act solely by binding and sequestering Slit , as the Robo2∆Ig2 receptor can still bind Slit , but is completely unable to inhibit Robo1 activity . It remains to be determined , but a closer analogy may exist with the way Dickkopf ( DKK ) family proteins antagonize Wnt receptor signaling ( Niehrs , 2006 ) . In this case , secreted DKK binds to the lipoprotein related proteins ( LRP5 and 6 ) , which are co-receptors for Wnt , and prevents LRP interaction with the Frizzled/Wnt ligand receptor complex ( Ahn et al . , 2011; Chen et al . , 2011 ) . While Robo2 is not secreted , there is evidence that Robo1 receptor extracellular domains can be cleaved and shed into the extracellular space ( Coleman et al . , 2010 ) , and we have observed that the Robo2 ectodomain can also be shed in vitro and in vivo ( Evans and Bashaw , unpublished ) . In the future , it will be interesting to investigate whether Robo2 binding prevents Robo1 from interacting with Slit in vivo , and whether Robo2 receptor cleavage is important for its ability to promote midline crossing . Alternatively , Robo2 could prevent the recruitment of Robo1's downstream signaling molecules such as Enabled , Nck/Dock and Son of Sevenless ( Bashaw et al . , 2000; Fan et al . , 2003; Yang and Bashaw , 2006 ) . In addition to its Ig1/Ig2 dependent role in inhibiting Slit-Robo1 repulsion that we have described here , Robo2 has at least three other distinct axon guidance activities that can be attributed to different structural elements of the receptor . In the context of midline axon repulsion , Robo2 binds Slit through its extracellular Ig1 domain and cooperates with Robo1 to prevent abnormal midline crossing . It is not known how Robo2 signals repulsion , but based on receptor swap experiments that demonstrate that Robo1 can substitute for Robo2's midline repulsive activity , it seems likely that a common cytoplasmic signaling output shared by Robo1 and Robo2 ( perhaps mediated by the shared CC0 or CC1 motifs ) is important for repulsion ( Spitzweck et al . , 2010 ) . Robo2 also directs the mediolateral position of axons in the CNS , an activity conferred by a combination of its extracellular Ig1 and Ig3 domains ( Evans and Bashaw , 2010b ) . In this context , distinct biochemical properties conferred by Ig3 appear to direct Robo2 receptor multimerization , and this property correlates with the ability to regulate lateral position in vivo ( Evans and Bashaw , 2010b ) . Finally , in addition to these activities , we have recently discovered a new function for Robo2 in regulating the guidance of specific populations of motor axons to their appropriate muscle targets . In this case , Robo2's guidance activity depends on unique features of its cytoplasmic domain ( Santiago et al . , 2014 ) . A major challenge for the future will be to understand how these diverse guidance activities are selectively deployed at the right time and place to allow for coordinated guidance responses . One important factor that is likely to contribute to the coordination of these activities is the regulation of the spatial and temporal expression of Robo2 . For example , in late stage embryos , Robo2 protein expression is restricted to the lateral most regions of the longitudinal connectives where it is presumably acting to control lateral positioning , while in younger embryos robo2 mRNA can be detected in ipsilateral pioneer neurons where it is likely contributing to midline repulsion . Robo2 is also detected in midline glia and neurons , where we propose it may act to prevent premature responses to Slit . At present , little is known about how these patterns of expression are established and temporally regulated , although we have recently shown that the homeodomain transcription factors dHb9 and nkx6 are required for robo2 expression in a subset of motor neurons ( Santiago et al . , 2014 ) . While controlling the time and place of Robo2 expression is no doubt part of the explanation for how Robo2's diverse and sometimes opposing activities are coordinated , we expect that the distinct biochemical features of Robo2's different activities , as well as the potential interaction with context-specific cofactors will also play an important role . Here , we note that Robo2 does not appear to be able to promote midline crossing cell-autonomously , either in subsets of commissural neurons in rescue experiments , or in the apterous ipsilateral interneurons in gain-of-function experiments . This could be because Robo2 is unable to bind to Robo1 in cis in vivo , or alternatively because Robo1–Robo2 cis interactions confer a distinct outcome from the inhibitory effect of Robo2 presented from other cells . This is reminiscent of the different responses produced by cis and trans interactions between receptors and their ligands ( Yaron and Sprinzak , 2012 ) . How distinct signaling responses are triggered by the different structural conformations resulting from cis vs trans interactions remains poorly understood . Future experiments to define the mechanisms that control the specific expression domains and biochemical activities of Robo2 promise to continue to offer new insights into the molecular biology of axon guidance . For statistical analysis , comparisons were made between genotypes using the Student's t-test . For multiple comparisons , significance was assessed by using a Bonferroni correction .	When an animal embryo is developing , nerve pathways grow to connect the left and right halves of the nervous system . These pathways allow coordination between the two sides of the body—which is important for tasks such as walking and swimming . However , in order for these pathways to be properly established , the activity of certain genes must determine whether each nerve fiber ( or axon ) will stay on one side of the body , or cross the midline to the other side . It is not fully understood how genes , and the proteins they encode , interact with each other to regulate the crossing of the body's midline , but the process is known to involve proteins called Robo receptors . Robo receptors are a class of proteins found on nerve fibers . Most Robo receptors work to prevent nerve fibers from crossing the midline , but some proteins in this family—including Robo2—can also promote midline crossing . Based on previous studies , it was not clear how Robo2 could have such opposing effects on different sets of nerve cells . Evans , Santiago et al . have now explored how Robo2 regulates the development of the nervous system of fruit fly embryos , and found that Robo2 promotes midline crossing by inhibiting the activity of a closely related protein called Robo1 . Further experiments unexpectedly showed that Robo2 does not promote midline crossing in the cells in which it is produced . Instead , the Robo2 receptor instructs other Robo1-producing nerve cells to cross the midline . These findings reconcile the previous , seemingly paradoxical , observations about the activity of Robo2 . Following on from these findings , one important next step will be to determine exactly how Robo2 can inhibit the activity of Robo1 , such that it no longer prevents nerve fibers from crossing the midline . Determining whether similar inhibitory interactions between Robo receptors are important for the development of other tissues in the fruit fly , or in other animals , is another challenge for the future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "neuroscience" ]	2015	Robo2 acts in trans to inhibit Slit-Robo1 repulsion in pre-crossing commissural axons
Continuous contact with self-major histocompatibility complex ligands is essential for the survival of naive CD4 T cells . We have previously shown that the resulting tonic TCR signaling also influences their fate upon activation by increasing their ability to differentiate into induced/peripheral regulatory T cells . To decipher the molecular mechanisms governing this process , we here focus on the TCR signaling cascade and demonstrate that a rise in intracellular calcium levels is sufficient to modulate the phenotype of mouse naive CD4 T cells and to increase their sensitivity to regulatory T-cell polarization signals , both processes relying on calcineurin activation . Accordingly , in vivo calcineurin inhibition leads the most self-reactive naive CD4 T cells to adopt the phenotype of their less self-reactive cell-counterparts . Collectively , our findings demonstrate that calcium-mediated activation of the calcineurin pathway acts as a rheostat to shape both the phenotype and effector potential of naive CD4 T cells in the steady-state . T-cell precursors originate in the bone-marrow and are educated in the thymus through processes called positive and negative selections , which result in MHC-restriction and self-tolerance , respectively ( Stritesky et al . , 2012 ) . Only those T cells that bear an αβT-cell receptor ( TCR ) recognizing self-MHC with a relatively low affinity will differentiate and exit into the systemic circulation as self-MHC restricted T cells . T cells carrying an αβ TCR that reacts with self-MHC with very low affinity die by neglect , whereas those recognizing self-MHC with high affinity are mostly deleted by apoptosis or differentiate into regulatory T cells called ‘natural’ or thymically derived ( tTreg ) in order to prevent autoimmunity ( Bautista et al . , 2009; Leung et al . , 2009 ) . Therefore , self-MHC and the associated self-reactivity of T cells influence both T-cell production and phenotype in the thymus . In the periphery , the pre-immune repertoire of T cells is composed of almost 70% of naive T cells . The remaining 30% are divided between recent thymic emigrants with a comparable phenotype , regulatory T cells ( Foxp3+ ) and cells with an activated/memory phenotype . Naive T cells are kept alive through continuous TCR interactions with MHC molecules complexed with various self-peptides . Such TCR/MHC interactions plus contacts with IL-7 cause low-level signaling , which promotes long-term survival of naive T cells in interphase through the synthesis of anti-apoptotic molecules such as Bcl-2 ( Martin et al . , 2006; Takada and Jameson , 2009 ) . The degree of TCR self-reactivity of a given T-cell clone has been correlated with its expression of CD5 and Nur77 ( Azzam et al . , 1998; Moran et al . , 2011 ) . We have recently identified the cell surface GPI-anchored protein , Ly-6C , as an additional and complementary sensor of T-cell self-reactivity ( Martin et al . , 2013 ) . Indeed , we have shown that , in contrast to CD5 and Nur77 which expression directly correlates with self-reactivity , the expression of Ly-6C by peripheral naive CD4 T cells ( CD4 TN cells ) inversely correlates with their ability to interact with self-MHC ( Martin et al . , 2013 ) . Ly-6C- CD4 TN cells were therefore identified as more self-reactive than their Ly-6C+-cell counterparts . In the absence of foreign antigen , peripheral naive T cells continuously recirculate between lymphoid organs ( Gowans , 1959 ) , in which they migrate along the fibroblastic reticular cells network ( Bajénoff et al . , 2006 ) and interact frequently and briefly with dendritic cells ( DC ) ( Bajénoff et al . , 2006; Mempel et al . , 2004 ) . It is generally accepted that these frequent DC-T-cell interactions increase the probability of contacts between very rare antigen-specific naive T cells and the few DCs presenting their cognate antigen during the initial course of an infection . Experimental evidences indicate that self-MHC recognition in the periphery is also required to maintain T cells in a state of responsiveness toward foreign antigen ( Persaud et al . , 2014; Stefanová et al . , 2002; Wülfing et al . , 2002 ) , suggesting a crucial role for self-MHC mediated ‘education’ and TCR self-reactivity in determining the intrinsic functional attributes of CD4 TN cells . Altogether , this steady-state tonic TCR signaling was shown to influence CD4 TN-cell effector fate by increasing the magnitude of their response toward their cognate antigens . Following activation by antigen-presenting cells ( APCs ) in the periphery , the bulk of CD4 TN cells can differentiate into a variety of well documented T-helper ( TH ) cell subsets , such as TH1 , TH2 , TH17 or peripherally induced regulatory T cells ( pTreg cells ) , characterized by their cytokine production profiles , specific effector functions and lineage-specific transcription factors ( T-bet for TH1 cells , GATA-3 for TH2 cells , RORγt/RORα for TH17 cells and Foxp3 for pTreg cells ) ( Abbas et al . , 1996; Bilate and Lafaille , 2012; Fontenot et al . , 2003; Hori et al . , 2003; Ivanov et al . , 2006; Liang et al . , 2006; Mosmann et al . , 1986; Szabo et al . , 2000; Ye et al . , 2001; Zheng and Flavell , 1997 ) . Among these effector CD4 T cells , pTreg cells produce TGF-β and share phenotypic and functional characteristics with tTreg cells ( Bilate and Lafaille , 2012 ) . The immunological context in which CD4 TN cells are immersed at the time of their activation is known to drive lineage commitment . The strength of the activating TCR signals received by a CD4 TN cell also influences its subsequent polarization toward particular differentiation pathways ( Corse et al . , 2011 ) . Indeed , in weakly polarizing conditions , low TCR signals favor TH2- and pTreg-cell differentiation , whereas TH1- and TFH-cell differentiation arises from stronger signals ( Gottschalk et al . , 2010; Rogers and Croft , 1999; Turner et al . , 2009 ) . Most of these data were obtained in vitro by modulating signal strength with graded dose of peptide-MHC ligands of varying potency . However , only relatively high-affinity TCR–MHC interactions were shown to facilitate the induction of persistent Foxp3+ T cells in vivo ( Gottschalk et al . , 2010 ) . Our recent work has reinforced the link between the tonic TCR signaling received by CD4 TN cells in the steady-state and their fate in the effector phase . Indeed , we have demonstrated that TCR/self-MHC interactions not only increase quantitatively but also shape qualitatively the response of CD4 TN cells to their cognate antigens in the effector phase ( Martin et al . , 2013 ) . More precisely , by taking advantage of our data showing that Ly-6C expression can be considered as a new sensor of CD4 TN-cell self-reactivity , we have demonstrated that CD4 TN cells with the highest avidity for self-MHC ( Ly-6C- ) have a biased commitment toward the iTreg/pTreg-cell lineage ( Martin et al . , 2013 ) . The binding of antigen/MHC complexes to the TCR triggers the recruitment of a series of signaling molecules and adaptors to the TCR/CD3 complex that ultimately results in the phosphorylation and activation of phospholipase C-γ ( PLCγ ) . PLCγ then cleaves the phospholipid phosphatidylinositol 4 , 5-bisphosphate ( PIP2 ) in the plasma membrane to generate diacylglycerol , which activates protein kinase C ( PKC ) and Ras-dependent pathways , as well as 1 , 4 , 5-inositol trisphosphate ( IP3 ) , which induces the release of calcium ( Ca2+ ) from intracellular stores ( the endoplasmic reticulum ( ER ) ) . This Ca2+ store release only transiently elevates intracellular Ca2+ concentrations but this transient rise induces in turn a massive and sustained Ca2+ entry from the extracellular space ( Hogan et al . , 2010 ) . With the aim of deciphering the molecular mechanisms involved in the tonic TCR-signaling-mediated shaping of the CD4 TN-cell compartment , we have focused on the TCR signaling cascade . By using transcriptomic and phenotypic approaches as well as in vitro and in vivo assays , we have identified the Ca2+ signaling pathway as key for the acquisition of both the phenotype of the most self-reactive CD4 TN cells and their enhanced cell-intrinsic ability to commit into regulatory T cells upon activation in vitro ( iTreg ) and in vivo ( pTreg ) . We have recently shown that CD4 TN cells with the highest avidity for self-MHC ( Ly-6C- CD4 TN cells ) have a biased commitment toward the iTreg/pTreg-cell lineage ( Martin et al . , 2013 ) . As TH1- and TH2-cell-derived cytokines are known to inhibit iTreg-cell induction in vitro ( Henderson et al . , 2015 ) , we first wondered whether Ly-6C- and Ly-6C+ CD4 TN cells had the same ability to produce such cytokines after stimulation . Ly-6C- and Ly-6C+ CD4 TN cells were thus stimulated with αCD3- and αCD28-coated antibodies in the presence or absence of TGFβ . Interferon-gamma ( IFN-γ ) and interleukins ( IL ) -4 , -17 and -10 were assayed in the supernatants collected 24 hr after the beginning of the culture . We found that , whatever the presence or absence of TGFβ in the culture medium , Ly-6C- and Ly-6C+ CD4 TN cells produced similar amounts of these cytokines ( Figure 1—figure supplement 1A , B ) . To further characterize the enhanced ability of Ly-6C- CD4 TN cells to commit into iTregs in vitro , we asked whether this feature was cell-intrinsic . To this end , Ly-6C- and Ly-6C+ CD4 TN cells were isolated from peripheral LNs of C57BL/6 Foxp3-GFP mice by flow cytometry sorting , barcoded with CTv or CTv and CTfr proliferation dyes , and stimulated with αCD3- and αCD28-coated antibodies in the presence of graded doses of TGFβ . These cells were cultured separately or together ( Figure 1A , B ) . The percentages of Foxp3+ cells among the progeny of both naive cell-subsets were assessed on day 4 . For suboptimal doses of exogenous TGFβ , Ly-6C- CD4 TN cells gave rise to a twofold higher proportion of iTreg cells than their Ly-6C+-cell counterparts in both culture conditions ( Figure 1C , D ) . The concentration of TGFβ needed to obtain 50% of the maximal percentage of iTreg cells ( effective concentration , EC50 ) was calculated by fitting the dose-response curves of both CD4 TN-cell subsets in the different culture conditions ( Figure 1D , E ) . EC50 values for TGFβ were statistically different between the 2 CD4 TN-cell subsets whether they were cultured separately or together . Of note , and in line with their similar ability to produce TH1- and TH2-cell-derived cytokines , blocking IFN-γ and IL-4 during in vitro iTreg-cell polarization did not abolish the difference in the ability of Ly-6C- and Ly-6C+ CD4 TN cells to differentiate into iTreg cells . ( Figure 1—figure supplement 1C–E ) . These results suggest strongly that the greater sensibility of Ly-6C- CD4 TN cells to iTreg-cell polarization signals is cell-intrinsic . To further compare Ly-6C- and Ly-6C+ CD4 TN cells , we obtained Affymetrix gene expression profiles from both CD4 TN-cell subsets directly isolated from peripheral LNs of C57BL/6 Foxp3-GFP mice by flow cytometry sorting ( Figure 2 ) . Only few genes were significantly differentially expressed between the two types of CD4 TN cells ( at a 1 . 3-fold cutoff , 167 genes over-expressed and 164 under-expressed in Ly-6C- CD4 TN cells when compared to Ly-6C+ CD4 TN cells; Figure 2A ) . This set of differentially expressed genes between Ly-6C- and Ly-6C+ CD4 TN cells was compiled into a comprehensive signature that we named 6CSign ( Figure 2B , C ) . The differential expression of several genes by Ly-6C- and Ly-6C+ CD4 TN cells was then validated at the protein level by flow-cytometry ( Figure 2—figure supplement 1 ) . In line with our microarray analysis , Ly-6C- CD4 TN cells were expressing higher amounts of CD5 , CD73 , CD122 , CD200 , Ikzf3 and Izumo1r and lower levels of Sca-1 and IL18Rα than their Ly-6C+ CD4 TN-cell counterparts ( Figure 2—figure supplement 1A , B ) . We have previously shown that Ly-6C- CD4 TN cells were more self-reactive than Ly-6C+ CD4 TN cells ( Martin et al . , 2013 ) . Accordingly , among the 6CSign , several genes such as Ctla4 , Cd5 , Tnfrsf4 , Tnfrsf9 and Nr4a1 were previously shown to belong to activation-induced or -repressed gene families ( Figure 2C; [Wakamatsu et al . , 2013] ) . We thus compared more precisely our signature , the 6CSign , with several public Geo Datasets comparing various ‘activated’ CD4 TN cells to their non-activated cell counterparts ( Figure 3A , B ) . CD5 expression levels on CD4 TN cells are actively maintained by interactions with self-MHC and rapidly decline in their absence ( for example in the blood , [Stefanová et al . , 2002] ) . In agreement with a greater self-reactivity of Ly-6C- CD4 TN cells , the 6CSign correlated significantly with the CD5hi versus CD5lo CD4 TN-cell signature ( Richards et al . , 2015 ) . Interestingly , whereas the 6CSign genes also correlated with the transcriptional signature of αCD3-activated CD4 TN cells ( compared to unstimulated cells ) ( Wakamatsu et al . , 2013 ) , there was no significant correlation with the signature of Phorbol 12-Myristate 13-Acetate ( PMA ) -activated CD4 TN cells ( Bevington et al . , 2016 ) . Interestingly , the 6CSign contained several genes characteristically expressed in Treg cells such as Ctla4 , Izumo1r , Cd200 , Lag3 or Il2rb . All these genes were upregulated in Ly-6C- CD4 TN cells when compared to Ly-6C+ CD4 TN cells ( Figure 2C ) . By comparing CD4 T-cell effectors with naive CD4 T cells , Wei et al . ( 2009 ) have recently defined the transcriptional signature of the main CD4 TH-cell subsets such as in-vitro-induced Treg cells , TH1 , TH2 and TH17 cells . Comparison of the 6CSign with these cell signatures revealed that the differences in gene expression observed between Ly-6C- and Ly-6C+ CD4 TN cells correlated significantly with the in-vitro-induced Treg-cell signature , and to a lesser extent with the TH17 one but not with the TH1 or TH2 transcriptional signatures ( Figure 3C ) . Similarities were also observed between the transcriptional profiles of Ly-6C- CD4 TN cells and ex vivo purified peripheral Treg cells ( Figure 3C ) ( Wei et al . , 2009 ) . One common characteristic shared by Ly-6C- CD4 TN cells and CD4 Treg cells is their high degree of self-reactivity . Recent studies have highlighted a continuous requirement of self-MHC recognition and of the associated TCR-mediated signaling for maintaining both the function and transcriptional signature of CD4 Treg cells ( Delpoux et al . , 2012; Levine et al . , 2014; Vahl et al . , 2014 ) . Whereas self-deprivation or TCR-ablation did not impair the expression of the transcription factor Foxp3 , they induced major transcriptional changes ( Delpoux et al . , 2012; Levine et al . , 2014; Vahl et al . , 2014 ) . Interestingly , 6CSign genes strongly correlated with the transcriptional signature of TCR+ CD4 Treg cells ( compared to TCR- CD4 Treg cells , Figure 3D ) ( Vahl et al . , 2014 ) . More precisely , most genes upregulated in Ly-6C- CD4 TN cells when compared to their Ly-6C+-cell counterparts were positively regulated by steady-state TCR signaling in CD4 Treg cells ( such as Cd5 , Cd200 , Il2rb , Itih5 , Maf and Myb; Figure 3E ) . Conversely , an important proportion of the genes downregulated in Ly-6C- CD4 TN cells were also down-regulated by steady-state interactions with self-MHC in CD4 Treg cells ( Figure 3E ) . Altogether , these data point to a role for the TCR signaling pathway in the installation and maintenance of the 6CSign . The transcriptional signature of Ly-6C- CD4 TN cells revealed some similarities between these cells and αCD3-stimulated CD4 TN cells . We therefore decided to analyze the effect of TCR signaling on Ly-6C expression . Ly-6C+ CD4 TN cells were isolated from peripheral LNs of C57BL/6 Foxp3-GFP mice by flow cytometry sorting and incubated with various stimulating agents mimicking all or part of TCR-induced signals ( Figure 4A ) . As expected from our transcriptomic analysis and previous work ( Martin et al . , 2013 ) , Ly-6C expression was clearly downregulated when cells were stimulated with αCD3 and αCD28-coated antibodies for 5 days ( Figure 4A , B ) . To decipher which TCR-induced signals led to Ly-6C down-regulation , we roughly dichotomized the TCR signaling cascade into its two main components , for example the Ca2+ signaling pathway that can be elicited by Thapsigargin ( TG ) and the PKC and ERK signaling pathways activated by PMA . When combined , PMA and TG , induced complete Ly-6C down-regulation , whereas , when separated , each drug had an opposite effect on Ly-6C expression . Indeed , whereas PMA alone upregulated Ly-6C expression , TG alone induced a near-complete disappearance of Ly-6C protein at the surface of Ly-6C+ CD4 TN cells . Interestingly , while in all other conditions , CD4 TN cells were proliferating , this phenotypic conversion of Ly-6C+ CD4 TN cells into Ly-6C- CD4 TN cells induced by TG alone occurred without any proliferation ( Figure 4A , B ) . Importantly , to avoid TG-induced cell death , a sub-optimal dose ( 4 nM ) was used in these culture conditions . 4 nM TG induced a reproducible increase in intracellular calcium levels , although to a lesser extent than the classical dose of 200 nM ( Figure 4C ) . Accordingly , by analyzing basal Ca2+ contents at the end of the culture period ( 5 days ) , we observed that 4 nM TG treated Ly-6C+ CD4 TN cells exhibited higher cytoplasmic Ca2+ levels than control cells cultured in IL-7 alone ( Figure 4D , E ) . To further characterize the long-term effect of this low-dose TG , subcellular localization of the nuclear factor of activated T-cell protein 1 ( NFAT1 ) was assessed in Ly-6C+ CD4 TN cells in the presence or absence of 4 nM TG at various time points along the culture . Indeed , increases in intracellular Ca2+ levels result in the activation of calcineurin that dephosphorylates members of the NFAT family , leading to their translocation into the nucleus . NFAT1 localization was quantified by high-resolution imaging flow-cytometry using the ImageStreamX technology ( Figure 4F ) . In line with the Ca2+ increase induced by 4 nM TG treatment , NFAT was translocated into the nucleus of Ly-6C+ CD4 TN cells in the presence of TG while it remained cytoplasmic in its absence . NFAT translocation into the nucleus peaked on day 1 and remained significantly higher in TG-treated cells than in control cells throughout the culture . Finally , in agreement with their resting status , Ly-6C+ CD4 TN cells cultured in TG alone for 5 days maintained a naive phenotype according to their low forward scatter profile and expression of CD44 and CD62L ( Figure 4—figure supplement 1 ) . We then studied the kinetic aspect of the TG-mediated conversion of Ly-6C+ CD4 TN cells into Ly-6C- CD4 TN cells and found that it occurred in 3–4 days of culture ( Figure 4G ) . Altogether , our results suggest that the Ca2+ signaling pathway is sufficient to induce Ly-6C down-regulation on CD4 TN cells in vitro . We therefore hypothesized that the Ca2+ signaling pathway might be involved as part of the self-mediated tonic TCR signaling in the generation/maintenance of Ly-6C- CD4 TN cells in the periphery of a normal mouse in the steady-state . To evaluate the activation status of the Ca2+ signaling pathway within Ly-6C- and Ly-6C+ CD4 TN cells in vivo in the steady-state , we analyzed NFAT1 subcellular localization in both cell types . To this aim , LN cells were directly fixed after recovery and NFAT1 or NFAT2 localization was imaged by confocal microscopy ( Figure 4H ) and quantified by high-resolution imaging flow-cytometry using the ImageStreamX technology ( Figure 4I and Figure 4—figure supplement 2 ) . In line with our hypothesis , NFAT localization is more nuclear in Ly-6C- CD4 TN cells than in their Ly-6C+-cell counterparts . As a control , CD4 TN cells were rested in vitro for 30 min before fixation and staining . As expected , differences in NFAT localization between Ly-6C- and Ly-6C+ CD4 TN cells were abolished in these conditions . 200 nM TG treatment induced the nuclear translocation of NFAT in both CD4 TN-cell subsets . Of note , in this latter condition , differences in the localization of NFAT1 and NFAT2 between Ly-6C- and Ly-6C+ CD4 TN cells were diminished but not completely abolished ( Figure 4I and Figure 4—figure supplement 2 ) . Altogether , our data demonstrate that increasing in vitro intracellular Ca2+ levels is sufficient to down-modulate Ly-6C expression at the cell surface of Ly-6C+ CD4 TN cells without inducing any significant proliferation or activation . Accordingly , the greater nuclear-cytoplasmic ratio of NFAT proteins observed in Ly-6C- CD4 TN cells , when compared to their Ly-6C+ CD4 TN-cell counterparts , might reflect differences in the intensity of the Ca2+/Calcineurin signaling induced in vivo in these cells . Such differences could result from the differential ability of these cells to regularly interact with self-MHC/self-peptide complexes in the steady-state . We have identified several proteins differentially expressed between Ly-6C- and Ly-6C+ CD4 TN cells ( Figure 2—figure supplement 1 ) and have showed that TG induced Ly-6C downregulation at the cell surface of Ly-6C+ CD4 TN cells . We next studied whether proteins of the 6CSign other than Ly-6C , were also modulated by an increase in intracellular Ca2+ . To go further , we examined in parallel the involvement of the calcineurin phosphatase in these processes . To this aim , Ly-6C+ CD4 TN cells were isolated from peripheral LNs of C57BL/6 Foxp3-GFP mice by flow cytometry sorting and cultured with IL-7 in the presence or absence of TG and calcineurin-inhibitors ( Cyclosporin A , CsA and Tacrolimus , FK506 , FK ) . Ly-6C- CD4 TN cells cultured in IL-7 were added as control . After 5 days of culture in these conditions , the expression of Ly-6C , CD5 , CD73 , CD122 , CD200 and Izumo1r was analyzed by flow-cytometry ( Figure 5A ) . For all these proteins , TG induced changes in their expression at the cell surface of Ly-6C+ CD4 TN cells . More precisely , their level of expression reached those observed in Ly-6C- CD4 TN cells . Blocking calcineurin activation with either CsA or FK506 led to the complete inhibition of this phenotypic conversion of Ly-6C+ ( CD5lo , CD73int , CD122lo , CD200lo , Izumo1rlo ) CD4 TN cells into Ly-6C- ( CD5hi , CD73hi , CD122int , CD200int , Izumo1rhi ) CD4 TN cells . This Ca2+-induced phenotypic conversion thus depends on the activity of the canonical Ca2+-calcineurin signaling pathway . We then investigated whether this in vitro observation could be mimicked in vivo . We first confirmed that the Ca2+-calcineurin signaling cascade was active in vivo in Ly-6C- CD4 TN cells by showing that blocking calcineurin activation for 18 hr with FK506 was sufficient to abrogate the nuclear localization of NFAT in these cells ( Figure 5—figure supplement 1 ) . We then wondered whether a longer treatment with this calcineurin inhibitor would affect the phenotype of CD4 TN cells in vivo . To this aim , C57BL/6 Foxp3-GFP mice were injected daily with FK506 or PBS for 2 weeks ( Figure 5B ) . After 14 days , CD4 TN cells from peripheral LNs and the spleen were analyzed for their expression of Ly-6C , Izumo1r and CD200 . In line with our in vitro experiments , both the percentage of Ly-6C+ cells among CD4 TN cells and the MFI of Ly-6C at the cell surface of Ly-6C+ CD4 TN cells increased in treated mice when compared to control mice ( Figure 5C–E ) . Moreover , FK506 induced significant decreases of Izumo1r and CD200 surface levels in CD4 TN cells ( Figure 5C , F ) . Such changes in the phenotype of the bulk of CD4 TN cells could result from either the conversion of Ly-6C- CD4 TN cells into Ly-6C+ CD4 TN cells or the disappearance of the Ly-6C--cell subset . We therefore decided to compare the behavior of adoptively transferred Ly-6C- CD4 TN cells in FK506 or PBS-treated mice ( Figure 5G ) . 106 Ly-6C- CD4 TN cells purified from LNs of CD45 . 1+ Foxp3-GFP donor mice were adoptively transferred into CD45 . 2+ Foxp3-GFP-recipient mice . Host mice were then daily injected with FK506 or PBS for 2 weeks ( Figure 5G ) . After 14 days , donor-derived CD4 TN cells from peripheral LNs and the spleen were analyzed . Although similar numbers of donor-derived CD4 TN cells were recovered from both FK506 and PBS -treated mice ( Figure 5H ) , these cells were still greatly enriched in Ly-6C-expressing cells in FK506-treated recipients ( Figure 5I ) . Altogether , our data demonstrate that the activation of the Ca2+-calcineurin signaling pathway drives the phenotypic conversion of Ly-6C+ CD4 TN cells into Ly-6C- CD4 TN cells both in vitro and in vivo . As a rise in intracellular Ca2+ level converts phenotypically Ly-6C+ CD4 TN cells into Ly-6C- CD4 TN cells , we then tested whether the in vitro iTreg-cell polarization potential of these ex-Ly-6C+ CD4 TN cells ( referred thereafter as ‘Ca2+-converted’ Ly-6C+ CD4 TN cells ) was also modified . Ly-6C- and Ly-6C+ CD4 TN cells were recovered from C57BL/6 Foxp3-GFP mice and cultured in vitro with or without TG . After 5 days of culture , viable cells were FACS-sorted and stimulated with αCD3- and αCD28-coated antibodies in the presence of graded doses of TGFβ for 4 days ( Figure 6A ) . Of note , even after 5 days of resting in the presence of IL-7 , Ly-6C- CD4 TN cells were keeping a greater sensitivity to iTreg-cell polarization signals , than Ly-6C+ CD4 TN cells cultured in the same conditions . Importantly , the iTreg-cell polarization potential of Ly-6C+ CD4 TN cells rose up when these cells were pre-incubated in the presence of TG and became similar to the one observed for Ly-6C- CD4 TN cells ( Figure 6B , C ) . In agreement with the fact that calcineurin inhibitors blocked the TG-mediated phenotypic conversion of Ly-6C+ CD4 TN cells into Ly-6C- CD4 TN cells ( Figure 5A ) , adding CsA at the time of TG pre-incubation also abrogated the sensitization of Ly-6C+ CD4 TN cells to iTreg-cell polarization signals ( Figure 6B ) . EC50 values for TGFβ were calculated in these conditions and were statistically different between the 2 CD4 TN-cell subsets when cells were pre-incubated in IL-7 alone but dropped to similar levels when TG was added in the pre-culture medium ( Figure 6C ) . Of note , pre-incubating Ly-6C+ CD4 TN cells with TG and CsA further limit their ability to commit into iTreg cells as reflected by a significant increase in EC50 ( Figure 6C ) . Altogether , our data demonstrate that an increase in intracellular Ca2+ levels not only shapes the phenotype of the CD4 TN-cell compartment but also sensitizes in vitro these cells to iTreg-cell polarization signals , both processes occurring through a calcineurin-dependent pathway . To confirm these data in vivo , we used the well-known model of antigen-specific pTreg-cell development induced by oral tolerance ( Coombes et al . , 2007; Sun et al . , 2007 ) . This protocol studies the behavior of CD4 TN cells from ovalbumin-specific TCR transgenic OT-II mice adoptively transferred into wild-type mice fed with ovalbumin ( OVA ) . Indeed , in these conditions , a significant proportion of OT-II cells rapidly differentiate into pTreg cells in the mesenteric lymph nodes and Peyer Patches of recipient mice . Most OT-II CD4 TN cells expressed Ly-6C ex vivo ( Figure 7—figure supplement 1A ) . FACS-sorted CD45 . 1/2+ OT-II CD4 TN cells were first cultured in IL-7 in the presence or absence of TG ( Figure 7A ) . After 5 days of culture , TG led to a marked downregulation of Ly-6C ( Figure 7—figure supplement 1B ) . Living cells were then FACS-sorted and 0 . 5–1 . 106 cells were adoptively transferred into CD45 . 1 Foxp3-GFP mice . Finally , recipient mice were fed or not for 7 days with OVA in their drinking water . As expected , OVA administration led to the activation of OT-II cells , as reflected by a significant CD44 upregulation at their cell surface ( Figure 7—figure supplement 1C ) . Similar numbers of OT-II CD4 T cells were recovered from the secondary lymphoid organs of OVA-fed mice whether they were initially injected with ‘Ca2+-converted’ or not OT-II CD4 TN cells ( Figure 7B ) . In all secondary lymphoid organs , Ca2+-converted OT-II CD4 TN cells gave rise to greater proportions and absolute numbers of Foxp3-expressing cells than OT-II CD4 TN cells cultured with IL-7 alone prior to injection ( Figure 7C–E ) . Specifically , a total of 1 . 16 ± 0 . 22×104 pTreg ( Foxp3-expressing ) cells were recovered from the whole periphery of recipient mice injected with Ly-6C+ OT-II CD4 TN cells compared to 2 . 26 ± 0 . 32×104 pTreg cells when mice were injected with Ca2+-converted OT-II CD4 TN cells ( p<0 . 05 ) . These latter results were confirmed by using a second protocol of oral administration of OVA . In this setting , ‘Ca2+-converted’ OT-II CD4 TN cells were co-transferred with OT-II CD4 TN cells cultured in IL-7 alone in order to compare their ability to convert into pTreg in the same recipient mice . FACS-sorted CD45 . 2+ and CD45 . 1/2+ OT-II CD4 TN cells were first cultured in IL-7 in the absence or presence of TG , respectively ( Figure 7—figure supplement 1A ) . After 5 days of culture , living cells were FACS-sorted , mixed at a 1:1 ratio and 1 . 106 cells were adoptively transferred into CD45 . 1 Foxp3-GFP mice . Finally , recipient mice were fed with OVA by gavage ( 4 and 24 hr after the transfer of OT-II cells ) . Nine days later , secondary lymphoid organs were recovered and the phenotype of donor-derived OT-II T cells was analyzed . In this setting , ‘Ca2+-converted’ OT-II CD4 TN cells were also giving rise to greater absolute numbers of Foxp3-expressing cells than OT-II CD4 TN cells cultured with IL-7 alone prior to injection ( Figure 7—figure supplement 1B , C ) . More precisely , more than three quarters of the Foxp3+ OT-II cells arising in these conditions derived from ‘Ca2+-converted’ cells ( Figure 7—figure supplement 1D ) . These latter results validate our in vitro data showing that a rise in intracellular Ca2+ leads to an enhanced sensitivity of CD4 TN cells to iTreg-cell polarization signals . In the steady-state , naive T cells continually recirculate between the blood , lymph and secondary lymphoid organs , scanning dendritic cells ( DCs ) for the presence of foreign antigens . In the course of their journey , naive T cells also make weak , but functional , interactions with self-peptides presented by self-MHC molecules ( self-MHC ) . Such contacts with self-MHC are required for the long-term survival of peripheral naive T cells ( Martin et al . , 2006 , 2003; Stritesky et al . , 2012; Tanchot et al . , 1997 ) . The signals derived from the recognition of self-MHC by TCRs also allow maintaining naive T cells in a state of greater sensitivity for responses to foreign antigens ( Dorfman et al . , 2000; Stefanová et al . , 2002 ) . The seminal work of Štefanová et al . showed a rapid decline in the ability of CD4 TN cells to respond to their cognate antigen once contacts with self-MHC were disrupted ( Stefanová et al . , 2002 ) . These findings were confirmed by several groups using various elegant experimental models ( Hochweller et al . , 2010; Lo et al . , 2009; Mandl et al . , 2013; Persaud et al . , 2014 ) . Beside these works , we have recently demonstrated that CD4 TN-cell self-reactivity not only increases quantitatively but also shapes qualitatively their response toward their cognate antigens in the effector phase by increasing their ability to commit toward the iTreg/pTreg-cell lineage ( Martin et al . , 2013 ) . In the present paper , we first wondered whether the enhanced ability of the most self-reactive CD4 TN cells to convert into iTreg/pTreg cells upon appropriate stimulation was a cell-intrinsic property . The unchanged ability of both Ly-6C- and Ly-6C+ CD4 TN cells to polarize into Foxp3-expressing iTreg cells in vitro whether they were cultured together or separately demonstrate that the biased commitment of the most self-reactive CD4 TN cells toward the iTreg-cell lineage is cell-intrinsic . We have recently described the cell surface GPI-anchored protein , Ly-6C , as an additional and complementary sensor of T-cell self-reactivity ( Martin et al . , 2013 ) . However , significant differences may be noticed between CD5 and Ly-6C . First , whereas CD5 surface levels directly correlate with self-reactivity , Ly-6C expression by peripheral CD4 TN cells inversely correlates with their ability to interact with self-MHC . Second , in contrast to CD5 , Ly-6C expression at the cell surface of CD4 TN cells is stable over time in homeostatic conditions as its up-regulation after self-MHC deprivation takes several days ( Martin et al . , 2013 ) . Notwithstanding these differences , Ly-6C- CD4 TN cells express higher protein and mRNA levels of CD5 than their Ly-6C+-cell counterparts . Two recent papers by the group of Daniel Hawiger have highlighted a crucial role of CD5 in promoting the conversion of CD4 TN cells into iTreg/pTreg cells ( Henderson et al . , 2015; Jones et al . , 2016 ) . CD5 would block the activation of the mammalian target of rapamycin ( mTOR ) and would allow activated CD4 TN cells to resist to the inhibition of iTreg-cell induction induced by TH1- and TH2-cell-derived cytokines . Accordingly , in the absence of effector-differentiating cytokines , CD5hi and CD5lo CD4 TN cells were shown to differentiate with a similar efficiency into iTreg/pTreg cells ( Henderson et al . , 2015 ) . Such a phenomenon is unlikely to account for the greater ability of Ly-6C- CD4 TN cells to commit to the iTreg/pTreg-cell lineage . Indeed , Ly-6C- and Ly-6C+ CD4 TN cells produced similar amounts of these cytokines after stimulation and Ly-6C- CD4 TN cells still differentiated more efficiently than their Ly-6C+-cell counterparts into iTreg cells in vitro in the presence of anti-cytokine ( IL-4 and IFN-γ ) blocking antibodies ( Figure 1—figure supplement 1C–E ) . Moreover , whereas rapamycin drastically diminished the difference in the ability of CD5hi and CD5lo CD4 TN cells to convert to iTreg/pTreg cells in the presence of cytokines known as restraining this effector fate ( Henderson et al . , 2015 ) , this mTOR inhibitor similarly enhanced the generation of iTreg cells from both Ly-6C- and Ly-6C+ CD4 TN cells and thus preserved the difference between these two cell subsets ( data not shown ) . In the present study , we have identified the Ca2+ signaling pathway as sufficient to induce Ly-6C down-regulation at the cell surface of CD4 TN cells in vitro . Indeed , incubation of Ly-6C+ CD4 TN cells with the sarco/endoplasmic reticulum calcium ATPase inhibitor , thapsigargin , led to multiple phenotypic changes including not only Ly-6C down-regulation but also variations in the expression of many other genes of the 6CSign ( such as CD5 , CD73 , CD122 , CD200 and Izumo1r ) . This phenotypic conversion of Ly-6C+ CD4 TN cells into Ly-6C- CD4 TN cells takes four days to occur in vitro and relies on the activity of Calcineurin , as shown by its complete blocking in the presence of Cyclosporin A or FK506 . Interestingly , calcium- and PKC/Ras-dependent signaling pathways had divergent effects on the expression of Ly-6C . Indeed , whereas TG induced Ly-6C down-regulation , PMA led to its upregulation ( Figure 4A ) . In line with this observation , the 6CSign does not correlate with the changes in gene expression induced by PMA ( Figure 3B ) . These opposite effects of TG and PMA may reflect the well-documented and complex interplay between the PKC and Ca2+ signaling pathways . For example , PKC translocation to the plasma membrane is strictly Ca2+ dependent ( Reither et al . , 2006 ) and calcineurin is phosphorylated and inhibited by PKC ( Hashimoto and Soderling , 1989 ) . Altogether , our results suggest that interactions with self-MHC in the steady-state result in a dominant Ca2+ signaling ( when compared to PKC and Ras-dependent pathways ) leading to down-regulation of Ly-6C expression . This hypothesis is consistent with our results showing that in vivo Calcineurin inhibition leads to an increase in Ly-6C expression at even higher levels than those observed at the cell surface of Ly-6C+ CD4 TN cells from untreated mice . This hypothesis is reinforced by the work of Dong et al . ( Dong TX et al . , co-published with the present article ) showing that Ca2+ fluxes can be measured in mouse total lymph node T cells in the steady-state and that anti-MHC blocking antibodies significantly reduced their frequency . Continuous interactions with self-MHC in the steady-state may thus induce calcium waves that shape both the phenotype of CD4 TN cells and their behavior in the effector phase by favoring their differentiation into pTreg cells . tTreg and pTreg cells have complementary roles in immune-mediated tolerance ( Haribhai et al . , 2011 ) . An attractive hypothesis would be that tTreg cells would be responsible for tolerance to self-antigens , whereas pTreg cells would be in charge of restraining deleterious immune responses to non-self-antigens . In particular , pTreg cells are involved in the control of the responses to non-self-antigens leading to allergy and asthma ( Josefowicz et al . , 2012 ) as well as to commensal organism- ( Lathrop et al . , 2011 ) and food-derived antigens ( Mucida et al . , 2005 ) in the gut . Foetus-derived and allograft-derived antigens represent other obvious examples of acute exposure to non-self-antigens arising in the adults and requiring the establishment of a tolerance . In both cases , pTreg cells are generated against non-self antigens ( either conceptus-male-derived [Samstein et al . , 2012] or allograft-derived [Francis et al . , 2011; Wood et al . , 2012] ) . These cells are needed to establish an efficient tolerance toward the foetus ( Samstein et al . , 2012 ) . However , there is still a lack of evidence to definitely implicate pTreg cells in the induction of an efficient tolerance toward allograft , in part because of the difficulties to achieve such a state . We have previously demonstrated that self-reactivity in the steady-state increases the ability of CD4 TN cells to differentiate into iTreg/pTreg cells ( Martin et al . , 2013 ) . Accordingly , the most self-reactive CD4 TN cells ( i . e . Ly-6C- CD4 TN cells ) should contribute predominantly to the pTreg-cell pool generated under physiologic and pathologic conditions . In the present study , our data suggest strongly that this tonic TCR-signaling-mediated shaping of the CD4 TN-cell compartment is calcineurin-dependent . In particular , chronic treatment with a calcineurin inhibitor leads to the disappearance of Ly-6C- CD4 TN cells . Cyclosporin A and Tacrolimus treatments could thus interfere with the neoconversion of CD4 TN cells into pTreg cells and limit the development of tolerance in transplant patients . This may explain the difficulty to safely interrupt these immunosuppressive therapies even after years . Thus , besides their obvious clinical utility , calcineurin inhibitors may have potentially harmful side effects that should be studied to better assess and adapt their use . C57BL/6 mice ( CD45 . 2 ) were obtained from Charles River Laboratories . C57BL/6 CD45 . 1 mice were maintained in our own animal facilities , under specific pathogen-free conditions . C57BL/6 Foxp3-GFP CD45 . 2 mice ( Wang et al . , 2008 ) , initially obtained from Dr Bernard Malissen , Centre d’Immunologie de Marseille-Luminy , France , were crossed with C57BL/6 CD45 . 1 mice to generate C57BL/6 Foxp3-GFP CD45 . 1 and CD45 . 1/ . 2 mice . C57BL/6 OT-II mice were obtained from Charles River Laboratories and crossed with C57BL/6 Foxp3-GFP CD45 . 1 ( or CD45 . 2 ) mice to generate C57BL/6 Foxp3-GFP CD45 . 1/ . 2 ( or CD45 . 2 ) OT-II mice . Four- to 12-week-old mice were used for all experiments . Experiments were carried out in accordance with the guidelines of the French Veterinary Department . All procedures performed were approved by the Paris-Descartes Ethical Committee for Animal Experimentation ( decision CEEA34 . CA . 080 . 12 ) . Sample sizes were chosen to ensure the reproducibility of the experiments and according to the 3Rs of animal ethics regulation . Peripheral Lymph Nodes ( pLNs ) , mesenteric Lymph Nodes ( mLNs ) , Peyer’s patches , spleen and thymus were homogenized and passed through a nylon cell strainer ( BD Falcon ) in PBS supplemented with 10% FCS ( Biochrom ) for adoptive transfer or cell culture ( LNs only ) , or in 5% FCS and 0 . 1% NaN3 ( Merck-Sigma-Aldrich , Lyon , France ) in PBS for flow cytometry . CD4 T cells were purified from LNs ( pooled superficial cervical , axillary , brachial , inguinal and mLNs ) of C57BL/6 Foxp3-GFP CD45 . 1 mice by incubating cell suspensions on ice for 15 min with a mixture of anti-CD8 ( 53–6 . 7 ) , anti-CD19 ( 1D3 ) and anti-Ter-119 antibodies ( Abs ) obtained from hybridoma supernatants , and then with magnetic beads coupled to anti-rat immunoglobulins ( Invitrogen , Cergy-Pontoise , France ) . Ly-6C- CD4 TN cells were sorted as Foxp3-GFP- Lineage ( CD25 , TCRγδ , CD8β , CD11b , CD11c ) -PE- CD44-/lo Ly-6C- cells using a FACS-ARIA3 flow cytometer ( BD Biosciences , Le Pont de Claix , France ) and injected intravenously into sex-matched recipient mice whose then were injected intraperitoneally every day for two weeks with 2 . 5 mg/kg of Prograf ( Tacrolimus; Astellas Pharma Inc . , Tokyo , Japan ) . CD4 T cells were purified from LNs of C57BL/6 Foxp3-GFP OT-II CD45 . 2 or CD45 . 1/ . 2 mice by using Dynabeads Untouched Mouse CD4 Cells Kit ( Invitrogen ) and cultivated with recombinant mouse IL-7 ( 10 ng/ml; R and D Systems , Minneapolis , MN ) with or without Thapsigargin ( 4 nM; Merck-Sigma-Aldrich ) into 96-well round-bottom treated cell culture microplate ( Corning; 1 × 105 cells per well ) . After 5 days of culture , cells were recovered and labelled with PE-conjugated anti-TCRγδ ( GL3 ) , anti-CD8 . b2 ( 53–5 . 8 ) , anti-NK-1 . 1 ( PK136 ) and APC-conjugated anti-CD44 ( IM7 ) , all from BD Biosciences . OT-II CD4 TN cells were sorted as GFP- Lineage-PE- CD44-/lo cells using a FACS-ARIA3 flow cytometer ( BD Biosciences ) and 0 . 5 to 1 × 106 cells were injected intravenously into sex-matched C57BL/6 Foxp3-GFP CD45 . 1 mice . Recipient mice were then continuously fed with Albumin from chicken egg white ( OVA; 1 . 5% w/v; Merck-Sigma-Aldrich ) in the drinking water or not . LNs and spleens were collected at day seven and CD45 . 2+ CD4 T cells analyzed . In a second protocol , sorted CD4 TN cells from CD45 . 2+ and CD45 . 1/ . 2+ C57BL/6 Foxp3-GFP OT-II mice were cultured in IL-7 ( 10 ng/ml ) without or with TG ( 4 nM ) , respectively . After 5 days live CD4 TN ( CD44lo CD25lo CD8β- CD11b- CD11c- NK1 . 1- TCRγδ- Foxp3-GFP- ) cells were flow-cytometry sorted , mixed at a 1:1 ratio and injected intravenously ( 0 . 5−1 × 106 cells ) into sex-matched CD45 . 1+ C57BL/6 Foxp3-GFP recipient mice gavaged with Ovalbumin ( OVA; 50 mg ) 4 and 24 hr later . LNs and spleens were collected at day 10 and donor-derived CD4 T cells were analyzed . Cell suspensions were collected and dispensed into 96-well round-bottom microtiter plates ( Greiner Bioscience; 6 × 106 cells per well ) . Surface staining was performed by incubating the cells on ice , for 15 min per step , with Abs in 5% FCS and 0 . 1% NaN3 in PBS . Each cell-staining reaction was preceded by a 15 min incubation with a purified anti-mouse CD16/32 Abs ( FcγRII/III block; 2 . 4G2 ) obtained from hybridoma supernatants . Alexa Fluor 700-conjugated anti CD45 . 2 ( 104 ) , Allophycocyanin ( APC ) -conjugated anti-CD25 ( PC61 ) and anti-CD44 ( IM7 ) , Brilliant Violet ( BV ) 421-conjugated anti Ly-6C ( AL-21 ) , BV 510-conjugated anti-CD4 ( RM4-5 ) , BV 786-conjugated anti-CD25 ( PC61 ) , Phycoerythrin ( PE ) -conjugated anti-CD25 ( PC61 ) , anti-CD69 ( H1 . 2F3 ) , anti-Izumo1r ( TH6 ) , anti-TCRγδ ( GL3 ) and anti-Vβ5 . 1/5 . 2 ( MR9-4 ) , PE-Cy7-conjugated anti-CD44 ( IM7 ) and anti-CD45 . 1 ( A20 ) , biotinylated anti-CD5 ( 53–7 . 3 ) , anti-CD62L ( MEL14 ) , anti-Ly-6C ( AL-21 ) and anti-Sca1 ( E13-161 . 7 ) were obtained from BD Biosciences . Alexa Fluor 647-conjugated anti-IL18rα ( BG/IL18rα ) , APC-conjugated streptavidin , BV 421-conjugated anti-Ly-6C ( HK1 . 4 ) and PE-conjugated anti-Ly-6C ( HK1 . 4 ) were obtained from BioLegend ( London , United Kingdom ) . PE-conjugated anti-CD200 ( OX-90 ) , anti-Ikzf3 ( 8B2 ) and anti-Nur77 ( 12 . 14 ) , PerCP-Cy5 . 5-conjugated anti-TCRβ ( H57-597 ) and biotinylated anti-CD73 ( eBioTY/11 . 8 ) were obtained from eBioscience ( Montrouge , France ) . Pacific Blue-conjugated streptavidin was obtained from Invitrogen . APC-Vio770-conjugated anti-CD8α ( 53–6 . 7 ) and PE-conjugated anti-CD122 ( TM-β1 ) were obtained from Miltenyi Biotec . Multi-colour immunofluorescence was analyzed using BD-LSR2 and BD-FORTESSA ( BD Biosciences ) flow-cytometers . List-mode data files were analyzed using Diva software ( BD Biosciences ) . Data acquisition and cell sorting were performed on the Cochin Immunobiology facility . Ex vivo purified CD4 T cells or cells recovered after 5 days of culture were loaded for 30 min at 37°C with the membrane-permeable fluorescent Ca2+ indicator dye Indo-1 AM ( Invitrogen ) at a concentration of 1 µM . Cells were stained either in HBSS ( for ex-vivo-purified CD4 T cells ) or directly in the culture medium ( cultured cells ) . Thereafter , ex-vivo-purified CD4 T cells were stained for surface markers and kept on ice . Before acquisition , cell aliquots were allowed to equilibrate to 37°C for 5 min and then were analyzed by flow cytometry . After acquisition of background intracellular Ca2+ concentrations for 2 min , cells were stimulated with Thapsigargin ( at a concentration of 4 or 200 nM ) . Flow-cytometry sorted Ly-6C- and Ly-6C+ CD4 TN cells from LNs of C57BL/6 Foxp3-GFP mice were stained with CellTrace Violet ( CTv; 5 µM; Life Technologies ) and cultured with IL-7 ( 10 ng/ml ) alone or in combination with Thapsigargin ( TG; 4 nM ) , Phorbol 12-myristate 13-acetate ( PMA; 1 . 25 ng/ml ) , PMA +TG ( 1 . 25 ng/ml and 4 nM , respectively ) and immobilized anti-CD3 ( clone 145 . 2C11; 4 µg/ml; obtained from hybridoma supernatants ) and anti-CD28 ( clone 37 . 51; eBioscience; 4 µg/ml ) Abs . For in vitro polarization assays Ly-6C+ CD4 TN cells were additionally stained with CellTrace Far Red ( CTfr; 1 . 25 µM; Life Technologies ) . Cells were then stimulated separately or together for 4 days with coated anti-CD3 and anti-CD28 Abs , in the presence of graded concentrations of exogenous recombinant human TGFβ1 ( Invitrogen ) . In some experiments , anti-IFN-γ ( clone R4-6A2; 10 µg/mL ) and anti-IL-4 ( clone 11B11; 10 µg/mL ) blocking antibodies were added in the culture . The concentration of TGFβ needed to obtain 50% of the maximal percentage of iTreg cells ( Effective Concentration , EC50 ) was calculated by fitting the dose-response curves of CD4 TN-cell subsets in the different culture conditions . To this end , the means of 3 to 5 independent experiments were used to build dose response curves using nonlinear least-squares regression to the Hill equation . The model used for this function was Y=[B+ ( T–B ) ] / [1 + 10 ( [LogEC50-X]*HillSlope ) ] , where ‘Y’ represents Foxp3+ cells as a percentage among CD4+ cells , ‘T’ and ‘B’ represent the plateaus at the beginning and end of the curve , respectively , and ‘X’ represents the concentration of TGFβ added at the beginning of the culture . The absolute EC50 was calculated to interpolate X at 50% with 95% confidence intervals . Flow-cytometry sorted Ly-6C- and Ly-6C+ CD4 TN cells from LNs of C57BL/6 Foxp3-GFP mice were stimulated as described above with immobilized anti-CD3 and anti-CD28 Abs in the presence or absence of exogenous recombinant human TGFβ1 ( Invitrogen , 4 µg/mL ) . Supernatants were recovered 24 hr later and cytokines were quantified by MSD multi-array U-PLEX assays ( IFN-γ , IL-4 , IL-17A/F and IL-10; Meso Scale Discovery , Rockville , MD ) according to the manufacturer’s instructions . LNs cells of C57BL/6 mice were harvested and fixed in 4% paraformaldehyde , immediately or after 30 min of resting or stimulation with 200 nM of Thapsigargin in RPMI 1640 Glutamax ( Gibco ) . Cells were washed in 1% FCS and 0 . 1% NaN3 in PBS and incubated in glycine ( 0 . 1M ) for 10 min . Cell surface was stained with biotinylated anti-Ly-6C ( AL-21 ) , BV 510-conjugated anti-CD4 ( RM4-5 ) , PE-conjugated anti-CD25 ( PC61 ) , anti-TCRγδ ( GL3 ) , anti-CD8 . β2 ( 53–5 . 8 ) , anti-NK-1 . 1 ( PK136 ) , anti-CD11b ( M1/70 ) , PE-Cy7-conjugated anti-CD44 ( IM7 ) and PerCp-Cy5 . 5-conjugated streptavidin , all from BD Biosciences . Intracellular stainings were performed using Foxp3 Staining kit ( eBioscience ) and Alexa 448-conjugated anti-NFAT1 ( D43B1; Cell Signaling , Leiden , The Netherlands ) or anti-NFAT2 ( 7A6; BioLegend ) and APC-conjugated anti-Foxp3 ( FJK-165; eBioscience ) Abs were used . Ly-6C- and Ly-6C+ CD4 TN cells were sorted as CD4-BV510+ Lineage-PE- CD44-/lo Foxp3-APC- Ly-6C+/- cells using a FACS-ARIA3 flow cytometer ( BD Biosciences ) . After sort , DRAQ5 ( Cell Signaling ) was used to stain nuclei . Cells were acquired with ImageStreamX ( Amnis; EMD Millipore ) and analyzed with IDEAS software . NFAT1 and NFAT2 nuclear localization was calculated as the similarity score between NFAT and DRAQ5 intensities . CD4 T cells from LNs of C57BL/6 Foxp3-GFP mice were enriched as described above . Then , Ly-6C- and Ly-6C+ CD4 TN cells were flow-cytometry sorted as CD4+ CD8α- TCRβ+ GFP- CD25- CD44-/lo cells using a FACS-ARIA3 flow cytometer . Total RNA was extracted using the RNeasy Mini kit ( QIAGEN , Courtaboeuf , France ) . RNA quality was validated with Bioanalyzer 2100 ( using Agilent RNA6000 nano chip kit ) . Experimental and analytical part of the microarray analysis was performed according to the MIAME standards . Amplified , fragmented and biotinylated sense-strand DNA targets were synthesized from 50 ng total RNA according to the manufacturer’s protocol ( Ovation PicoSL WTA System V2 and Encore Biotin Module kit ( Nugen , Leek , The Netherlands ) ) and hybridized to a mouse gene 2 . 0 ST array ( Affymetrix , Paris , France ) . The stained chips were read and analysed with a GeneChip Scanner 3000 7G and Expression Console software ( Affymetrix ) . Raw data ( . cel files ) were then processed and normalized using the quantile normalization method in RMA with R package ( Bioconductor ) . Statistical analysis was then performed with Partek Genomics Suite software ( Partek ) . Gene expression was z-transformed , for visualization , using the following formula: z= ( X-μ ) /τ , with X = normalized intensity , μ = mean of the normalized intensity across replicates and τ = s . d . of mean of the normalized intensity across replicates . Experimental and analytical part of the microarray was performed on the Cochin Genomic facility . Raw and processed data microarray data are provided in the Gene Expression Omnibus ( GEO ) under accession number GSE97477 . Normalized microarray datasets ( GSE14308 , GSE42276 , GSE67464 , GSE70154 and GSE62532 ) were recovered from NCBIs Gene Expression Omnibus ( GEO , http://www . ncbi . nlm . nih . gov/geo/ ) . For each datasets the values mean of Probset with the same Gene-id was performed to generate a file ( . xlsx ) with a unique value per Gene-id for each sample . These files were then statistical analyzed as described above . The newly created public GEO Datasets were then aligned with our microarray data by keeping only the commons Gene-id . Finally , these alignment files were filtered on our data for a p-value<0 . 05 and a fold change >1 . 3 and the differential expression of genes was compared between our and public GEO microarray .	To help protect the body from disease , small immune cells called T lymphocytes move rapidly , searching for signs of infection . These signs are antigens – processed pieces of proteins from invading microbes – that are displayed on the surface of so-called antigen-presenting cells . Before it encounters its specific antigen , a T cell is called naive . After encountering its antigen , the naive T cell activates and then develops into a variety of immune cells , each with a specific activity . These immune cells include so-called peripherally induced regulatory T cells ( or “pTreg cells” for short ) , which , as the name suggests , help to regulate the immune response . In addition to foreign antigens from microbes , antigen-presenting cells display fragments of the body’s own proteins too . All naive T cells recognize some “Self-antigens” , but not as strongly as they recognize foreign antigens . As a naive T cell travels around the body , it repeatedly interacts with antigen-presenting cells that display Self-antigens , which triggers a low level of signaling in the T cell . While this background signaling was known to help the T cell survive , in 2013 , researchers reported that: it also makes the T cell more responsive to foreign antigens; and it shapes how these cells will respond when activated . For example , the naive T cells that respond the most to Self-antigens were seen to be much more likely to become pTreg cells when activated than other T cells . Guichard et al . – who include several of the researchers involved in the 2013 work – set out to understand why the most Self-reactive T cells show this bias toward becoming pTreg cells . The experiments used a range of approaches with T cells both in the laboratory and in mice . By looking at which genes were active in the most Self-reactive T cells , Guichard et al . narrowed in on a signaling pathway that involves calcium ions and an enzyme called Calcineurin . Blocking this pathway caused the most Self-reactive T cells to lose their bias , and instead develop in the same way as the least Self-reactive T cells . Guichard et al . propose that the continuous interactions with Self-antigens trigger waves of calcium ions in a naive T cell that shapes its behavior and future development . In a related study , Dong , Othy et al . also conclude that contact with antigen-presenting cells causes calcium signals that shape how the T cells behave . In addition to providing more detail about the inner workings of immune cells , these findings may also have implications in a clinical setting . Calcineurin inhibitors are often used to suppress the immune system in transplant patients to prevent rejection of the transplanted organ . However , it has proved difficult to safely interrupt these therapies even after many years . These new findings may provide a possible explanation for this , by suggesting that the inhibitors may also interfere with the generation of pTreg cells . Without these cells’ regulatory influence , the immune system is unlikely to ever become tolerant of the transplant .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "immunology", "and", "inflammation" ]	2017	Calcium-mediated shaping of naive CD4 T-cell phenotype and function
Non-segmented , ( - ) RNA viruses cause serious human diseases . Human metapneumovirus ( HMPV ) , an emerging pathogen of this order of viruses ( Mononegavirales ) is one of the main causes of respiratory tract illness in children . To help elucidate the assembly mechanism of the nucleocapsid ( the viral RNA genome packaged by the nucleoprotein N ) we present crystallographic structures of HMPV N in its assembled RNA-bound state and in a monomeric state , bound to the polymerase cofactor P . Our structures reveal molecular details of how P inhibits the self-assembly of N and how N transitions between the RNA-free and RNA-bound conformational state . Notably , we observe a role for the C-terminal extension of N in directly preventing premature uptake of RNA by folding into the RNA-binding cleft . Our structures suggest a common mechanism of how the growth of the nucleocapsid is orchestrated , and highlight an interaction site representing an important target for antivirals . Viruses possessing a non-segmented , single-strand , negative-sense RNA genome are the causative agents of many serious human illnesses . Notable members belonging to this group of viruses ( Mononegavirales ) include measles , rabies , Ebola , respiratory syncytial virus ( RSV ) and Human metapneumovirus ( HMPV ) . HMPV ( Paramyxoviridae , subfamily Pneumovirinae ) is a leading cause of serious respiratory tract infections in children , the elderly , and immunocompromised individuals ( Boivin et al . , 2003; Osterhaus and Fouchier , 2003; van den Hoogen et al . , 2001 ) . In all members of the Mononegavirales , the RNA genome is packaged in the form of a nucleocapsid , a ribonucleoprotein complex consisting of polymerized viral nucleoproteins ( N ) and RNA ( Ruigrok et al . , 2011 ) . Besides protecting the viral genome from host nucleases , the nucleocapsid serves as the template for transcription by the viral RNA-dependent RNA polymerase L . Nucleocapsid assembly necessitates a pool of monomeric , RNA-free N , termed N0 , which is kept in an unassembled state through an interaction with an N-terminal portion of the polymerase cofactor P , until delivered to the sites of viral RNA synthesis ( Ruigrok et al . , 2011; Curran et al . , 1995; Mavrakis et al . , 2006 ) . The P protein is a multifunctional , modular protein containing large intrinsically disordered regions and is found to be tetrameric in HMPV ( Leyrat et al . , 2013 ) . In addition , P binds to the nucleocapsid via its C–terminus , and mediates the attachment of the RNA-dependent RNA polymerase L . Furthermore in pneumoviruses , P recruits the processivity factor M2-1 ( Leyrat et al . , 2014 ) . A great deal of effort has been spent on understanding the functions of P and recent crystal structures of P bound to N proteins ( N0-P ) from vesicular stomatitis virus ( VSV ) , Ebola virus , Nipah virus , and measles virus have highlighted its role in preventing assembly of N by blocking the C-terminal and N-terminal extensions of N ( CTD-arm and NTD-arm ) which facilitate N oligomerization ( Leyrat et al . , 2011; Guryanov et al . , 2015; Leung et al . , 2015; Yabukarski et al . , 2014 ) . However , there is still paucity in our understanding of the molecular details behind the proposed mechanisms , specifically regarding how P-bound N is released , attaches to the nucleocapsid and is loaded with RNA . To address these questions in the mechanism of N-chaperoning by P and nucleocapsid assembly we performed a structural analysis of assembled and unassembled N from HMPV . Our structure of N0-P reveals a conformational change , in which the negatively charged CTD-arm of N occupies the positively charged RNA binding site via specific and conserved interactions . Together with our RNA-bound structure of N these data imply a mechanism of how the growth of nucleocapsid filaments is coordinated in HMPV and related viruses . Biochemical studies of the nucleocapsid building block N are complicated by the fact that N proteins have a strong tendency to irreversibly oligomerize and bind host nucleic acids immediately upon recombinant expression ( Gutsche et al . , 2015; Tawar et al . , 2009 ) . One technique to mitigate this problem is to truncate regions of N that facilitate oligomerization ( Yabukarski et al . , 2014 ) . To stabilize monomeric full-length N0 we fused the N-terminal domain of P to N , a strategy that has seen success with nucleoproteins from other viruses ( Guryanov et al . , 2015; Kirchdoerfer et al . , 2015 ) . We obtained crystals of RNA-free HMPV N in a monomeric state and bound to a P peptide at 1 . 9 Å resolution by adding trace amounts of trypsin ( Dong et al . , 2007 ) to prune flexible loops and promote crystallization ( Figure 1—figure supplement 1 and Table 1 ) . In the structure , the P peptide is firmly nestled into a hydrophobic surface of the C-terminal domain of N ( CTD ) primarily composed of α-helices αC1 and αC2 ( Figure 1A , B ) . Ile9 , Leu10 and Phe11 of P occupy key positions and insert into this hydrophobic groove ( Figure 1B ) . Unlike the N0-P structure recently reported for measles ( Guryanov et al . , 2015 ) , we find that the linker connecting N and P in our chimeric construct has been cleaved prior to crystal growth . The P peptide wraps around the CTD and residues 12–28 form an alpha helix that lies atop N ( Figure 1B ) . This helix is initiated at Gly12 and pinned to the CTD through an aromatic side-to-face interaction of Phe23 with Tyr354 of N , both residues belonging to the so-called mir motif which is conserved within Pneumovirinae ( Karlin and Belshaw , 2012 ) . This result is consistent with an earlier study , in which alanine mutations of the corresponding residues in respiratory syncytial virus resulted in a drop of polymerase activity by more than 75% in a minireplicon system ( Galloux et al . , 2015 ) . 10 . 7554/eLife . 12627 . 003Figure 1 . Structure of the HMPV N0-P complex . ( A ) Crystal structure of RNA-free HMPV N0 bound to P1-28 . The C-terminal domain ( CTD ) of N is colored in light blue and the N-terminal domain ( NTD ) in dark blue . Secondary structure elements involved in the interaction with P are indicated . The P peptide is colored in orange . ( B ) Residues that are important in facilitating the interaction between P and N are shown in stick representation . Conserved hydrophobic residues of the P binding site are colored in yellow . ( C ) Multiple sequence alignment of N proteins from Paramyxoviridae members . Conserved residues of the P-binding site are highlighted in yellow and correspond to those in B . Virus name abbreviations are given in Methods . ( D ) N0-P complexes throughout Mononegavirales . Surface representations of N-CTDs of HMPV , Nipah virus ( PDB ID:4CO6 ) , Ebola virus ( PDB ID:4YPI ) and Vesicular stomatitis virus ( PDB ID:3PMK ) , colored by electrostatics . CTDs are shown in the same orientation . Bound P proteins ( VP35 , in the case of Ebola virus ) are colored in orange . The red dotted circle indicates a P-binding sub-region which is shared in all structures . Arrows are explained in the accompanying text . DOI: http://dx . doi . org/10 . 7554/eLife . 12627 . 00310 . 7554/eLife . 12627 . 004Figure 1—figure supplement 1 . Construct design and purification of the HMPV N0-P hybrid . ( A ) Schematic of the N0-P hybrid construct . The N-terminal ( NTD ) and C-terminal ( CTD ) domains of N are coloured in dark and light blue , respectively . The N-terminal and C-terminal arms are indicated . The first 40 residues of the HMPV P protein ( shown in orange ) were cloned at the C-terminus of N , immediately following the CTD-arm . ( B ) size exclusion chromatogram ( Superdex 75 ) and ( C ) accompanying SDS-PAGE analysis of the last purification step of N0-P . Protein elution was monitored by the absorbance at 280 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 12627 . 00410 . 7554/eLife . 12627 . 005Table 1 . Data collection and refinement statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 12627 . 005N0-PN-RNAData collectionSpace groupP 1C 2 2 21Cell dimensionsa , b , c ( Å ) 40 . 9 , 62 . 8 , 86 . 7202 . 0 , 233 . 2 , 203 . 6α , β , γ ( ° ) 91 . 0 , 96 . 4 , 109 . 090 , 90 , 90Wavelength ( Å ) 0 . 9790 . 917Resolution ( Å ) 28 . 42-1 . 86 ( 1 . 91-1 . 86 ) 101 . 19-4 . 17 ( 4 . 28-4 . 17 ) CC ( 1/2 ) 1 . 00 ( 0 . 47 ) 1 . 00 ( 0 . 38 ) Rmerge0 . 055 ( 0 . 590 ) 0 . 220 ( 2 . 924 ) I / σI9 . 2 ( 1 . 1 ) 9 . 2 ( 1 . 0 ) Completeness ( % ) 94 . 8 ( 75 . 0 ) 99 . 9 ( 100 ) Redundancy1 . 7 ( 1 . 6 ) 13 . 5 ( 13 . 8 ) RefinementResolution ( Å ) 28 . 42-1 . 86101 . 19-4 . 17No . reflections64451 ( 3743 ) 36125 ( 2617 ) Rwork / Rfree17 . 1/20 . 5319 . 1/23 . 0No . atomsProtein570727957Non-protein5881400B-factorsProtein34 . 54216 . 06Non-protein42 . 56215 . 08R . m . s . deviationsBond lengths ( Å ) 0 . 0070 . 010Bond angles ( ° ) 1 . 0001 . 120Ramachandran plot qualityFavoured ( % ) 99 . 7295 . 01Allowed ( % ) 0 . 284 . 96Outliers ( % ) 0 . 000 . 03Numbers in parentheses refer to the highest resolution shell . Rfree was calculated as per Rwork for a 5% subset of reflections that was not used in the crystallographic refinement . Molprobity scores are included in the Methods section . Alignment of Paramyxoviridae N sequences revealed that many hydrophobic residues lining the P-binding surface of αC1 through αC2 are shared within the family ( Figure 1C ) . For all known N0-P complexes ( Leyrat et al . , 2011; Leung et al . , 2015; Yabukarski et al . , 2014 ) , P binds to the CTD of N ( Figure 1D ) . Interestingly , although the specific interaction sites diverge ( Figure 1D , indicated by white and black arrows ) , a sub-region of the CTD ( Figure 1D , indicated by dotted circle ) is bound by P in all structures , indicating that it is widely conserved throughout Mononegavirales . To provide a rationale for the molecular switching between the monomeric , P-bound state and the assembled , RNA-bound state , a direct comparison at the atomic level is necessary . To this end , we purified and crystallized assembled HMPV N in the form of a decameric N-RNA ring ( Figure 2—figure supplement 1 and Table 1 ) . By exploiting the ten-fold non-crystallographic symmetry in the rings , we were able to obtain excellent electron density maps at 4 . 2-Å resolution ( Figure 2—figure supplement 2A–C ) and build a reliable model ( Karplus and Diederichs , 2012 ) ( Figure 2—figure supplement 2D ) . Assembled HMPV decameric N-RNA rings are ~0 . 5 MDa in molecular mass and 160 Å in diameter and 70 Å in height ( Figure 2A ) . The observed RNA binding mode is similar to that seen in the related RSV N-RNA structure ( Tawar et al . , 2009 ) . The RNA wraps around the N ring and wedges tightly in the cleft between the NTD and CTD of N , which is lined by positively charged residues ( Figure 2A and Figure 2—figure supplement 3 ) . In members of the Paramyxovirinae , the number of nucleotides in the viral genome is required to be a multiple of six ( Calain and Roux , 1993 ) and the structural basis for this so-called rule of six has been elucidated recently ( Gutsche et al . , 2015 ) . In members of the Pneumovirinae , however , this rule is not observed ( Tawar et al . , 2009 ) . Our structure further highlights this difference; with each N subunit contacting seven RNA nucleotides ( Figure 2—figure supplement 2C and Figure 2—figure supplement 3B ) . 10 . 7554/eLife . 12627 . 006Figure 2 . Comparison of N in assembled RNA-bound and monomeric RNA-free states . ( A ) top- and side-views of RNA-bound HMPV subnucleocapsid rings . N protomers and RNA are shown as surfaces with RNA rendered in brown . The diameter and height of the ring are indicated . ( B ) Three adjacent protomers of assembled RNA-bound N are shown viewed looking outwards from the centre of the ring , with the middle subunit rendered as surface . The exchange subdomains ( NTD- and CTD-arm ) that facilitate assembly of N are indicated . ( C ) The overlay with P1-28 ( orange ) bound to the middle protomer shows that the P-binding site overlaps with that of the NTD- and CTD-arms and that binding is mutually exclusive . ( D ) Hinge-motion of NTD and CTD of N . Monomeric N0 is superposed onto a single protomer of assembled , RNA-bound N ( N-RNA , shown in grey ) . The NTD pivots by 10 degrees relative to the CTD ( indicated ) . For clarity , only the NTD and CTD of the two states are shown . ( E ) showing N0 , and ( F ) showing N-RNA , close-up of the pivot point facilitating the hinge-motion of N . The white arrow in F indicates where the hinge region uncoils , allowing pivoting . For clarity , the P-peptide and the CTD-arm are omitted in E and F . DOI: http://dx . doi . org/10 . 7554/eLife . 12627 . 00610 . 7554/eLife . 12627 . 007Figure 2—figure supplement 1 . Purification of HMPV N-RNA and characterisation of oligomeric state . ( A ) Size exclusion chromatogram ( Superose 6 ) of N-RNA after purification from E . coli . The fractions containing N-RNA are indicated by a red bar . The broad peak centred around 10 mL ( indicated with white arrow ) constitutes nucleic acid co-purified from the expression host as indicated by the ratio of absorption at 260 and 280 nm . ( B ) SDS-PAGE analysis of the fractions marked with red bar in a . ( c ) Purified N-RNA was analysed by transmission electron cryomicroscopy , showing oligomeric rings . ( D ) 2D-class averages of N-RNA rings reveal three oligomeric states: 9-mers , 10-mers and 11-mers ( as indicated with white labels ) . The population distribution of the different oligomeric states is indicated in the accompanying pie chart . DOI: http://dx . doi . org/10 . 7554/eLife . 12627 . 00710 . 7554/eLife . 12627 . 008Figure 2—figure supplement 2 . Electron density maps of N-RNA . ( A–C ) Samples of electron density of the N-RNA crystal at 4 . 2 Å . A 2Fo-Fc map contoured at 1 . 0 σ after density modification with Parrot and B-factor map sharpening is shown . ( A ) zoomed-out overview of the density , ( B ) close-up view of two consecutive helices and ( C ) density for the bound RNA . ( D ) Data and model quality . Comparison of the correlation of the true signal CC* with CCfree and CCwork . The CC* plot shows that there is useful information up to a resolution of 4 . 2 Å . CCwork and CCfree values below CC* show that the model is not overfitting the data . DOI: http://dx . doi . org/10 . 7554/eLife . 12627 . 00810 . 7554/eLife . 12627 . 009Figure 2—figure supplement 3 . RNA-binding cleft of HMPV N . ( A ) External side view showing RNA inserted into three neighbouring protomers of assembled N . The two outer protomers are shown as surface representation coloured by electrostatics , highlighting the basic nature of the RNA-binding cavity . The central N subunit is shown in cartoon representation with the NTD coloured in dark blue and the CTD in light blue . RNA is shown in stick representation and coloured in brown . ( B ) close-up of the RNA-binding site of one N protomer . NTD and CTD are coloured as in A . The seven bound nucleotides are numbered , counting from 3’-end to 5’-end . Important residues interacting with RNA are shown in stick representation . DOI: http://dx . doi . org/10 . 7554/eLife . 12627 . 00910 . 7554/eLife . 12627 . 010Figure 2—figure supplement 4 . Role of a conserved aromatic residue in N hinge motion . ( A ) Monomeric N0 ( blue ) is superposed onto RNA-bound N ( grey ) . The dotted arrow indicates the tilting of α-helix αC3 during the transition from N0 to N-RNA . Tyr252 is thereby pushed upwards , facilitating the hinge motion . ( B–G ) panel of N proteins throughout Mononegavirales for which an aromatic residue ( shown in red ) can be observed at the same position and orientation , indicating a conserved function despite low overall sequence identity . ( H and I ) in Paramyxovirinae the aromatic residue located before the hinge region is flipped in the opposite direction in respect to other mononegaviruses . DOI: http://dx . doi . org/10 . 7554/eLife . 12627 . 010 Similar to N proteins from other members of Mononegavirales ( Tawar et al . , 2009; Alayyoubi et al . , 2015; Albertini et al . , 2006; Green et al . , 2006 ) , the NTD- and CTD-arms grasp the neighbouring protomers , thus facilitating assembly of polymeric N ( Figure 2B ) . The NTD-arm packs against the flank of the previous protomer ( Figure 2B , the NTD-arm of Ni+1 packs against Ni ) . The CTD-arm in turn latches onto the top of the CTD of the next protomer ( Figure 2B , CTD-arm of Ni-1 latches onto CTD of Ni ) . We observed that the binding site of the P peptide overlaps with the binding sites of the NTD- and CTD-arms ( Figure 2C ) . Our structures thus provide conclusive evidence that P hampers subdomain exchange between adjacent proteins in Pneumovirinae . This mechanism has also been proposed for a range of viruses thoughout Mononegavirales ( Leyrat et al . , 2011; Guryanov et al . , 2015; Yabukarski et al . , 2014; Alayyoubi et al . , 2015 ) and there is mounting evidence that it may be universal throughout the entire viral order . A hinge-like motion has been proposed by which N alternates between an open , RNA-free conformation ( N0 ) and a closed RNA-bound ( N-RNA ) conformation ( Guryanov et al . , 2015; Yabukarski et al . , 2014 ) . Comparison of these two states for HMPV reveals a rigid body movement of the NTD relative to the CTD ( Figure 2D ) . The conformational change rotates the NTD towards the CTD by 10° , the interface between the two domains acting as a hinge . At the interface , hinge residues Thr257 and Ala254 play a particularly crucial role . In the open , RNA-free state the hinge is maintained in a helical conformation by stabilization of Ala254 through the side chain of Thr257 and an additional backbone interaction with Thr175 ( Figure 2E ) . Upon RNA binding , Thr257 contacts the backbone of a nucleotide instead of stabilizing Ala254 ( Figure 2F ) . In addition , the loop containing Thr157 retracts to sterically accommodate the RNA chain . Having lost the stabilizing contacts of Thr257 and Thr157 , the helical hinge region around Ala254 unravels and becomes flexible ( Figure 2F , indicated by white arrow ) , allowing the relative domain motions of NTD and CTD . Furthermore , we propose that Tyr252 is important in facilitating the hinge motion . Tyr252 is positioned just before the pivot point and packs tightly against αC3 ( Figure 2—figure supplement 4A ) . An aromatic residue at this position is found packing against the same helix in most known structures of N ( Figure 2—figure supplement 4B–G ) . Transition from the RNA-free to RNA-bound state induces a rotation of αC3 , exerting upwards pressure on Tyr252 that is conferred onto the NTD ( Figure 2—figure supplement 4A ) . Intriguingly , in structures of Paramyxovirinae N , which obey the rule-of-six , this aromatic is flipped in the opposite direction ( Figure 2—figure supplement 4H , I ) and contacts RNA ( Gutsche et al . , 2015 ) , suggesting a similar coupling of RNA-binding and hinge-motion in these viruses . The most profound changes between assembled and unassembled states , however , involve the CTD-arm of N , a region that has been little characterized in pneumoviruses . In the polymeric , RNA-bound state of N ( N-RNA ) the CTD-arm flips upwards and latches onto the next protomer , whilst in the monomeric state ( N0 ) it packs down against the core of N ( Figure 3A ) . The downward , monomeric conformation is stabilized by specific salt-bridges linking the CTD-arm with the core of N ( Figure 3B ) . In this position the negatively charged CTD-arm folds into the positively charged RNA binding cleft , occupying it and directly blocking the binding of RNA ( Figure 3A ) . It is interesting to note , that whilst the CTD-arm blocks the RNA site in HMPV , it is the P peptide that inserts itself there in VSV ( Leyrat et al . , 2011 ) . Because this is not observed in paramyxoviral N0-P complexes ( Guryanov et al . , 2015; Yabukarski et al . , 2014 ) we hypothesize that , in Rhabdoviridae , a different strategy has evolved to block off the RNA binding cleft . The question arises how the interactions that hold the downwards-positioned CTD-arm in place are broken when assembly of N-RNA necessitates it flipping into the upwards position . In the RNA-free state , Arg260 and Trp261 contact Glu375 , while Arg186 forms a salt-bridge with Asp373 of the CTD-arm ( Figure 3B ) . In the assembled , RNA-bound state these interactions are broken , with Arg186 and Trp261 now positioning RNA nucleotides in the cleft , whilst Arg260 instead fastens onto the NTD-arm of the neighbouring Ni+1 ( Figure 3C ) . The shift from initial stabilization of the inhibitory ( downwards ) CTD-arm conformation to stabilization of bound RNA and neighbouring N subunit implies that attachment of a new N protomer and insertion of nascent RNA occur concomitantly . This makes sense in the context of viral replication sites , where tetrameric P proteins act as molecular chaperones attaching to the nucleocapsid template , polymerase and free N0 , leading to high local concentrations of nucleoprotein and RNA . 10 . 7554/eLife . 12627 . 011Figure 3 . Role of the CTD-arm in inhibiting premature RNA uptake . ( A ) Conformational switch of the CTD-arm . The CTD-arm ( red ) is shown in a upward conformation assumed in the N-RNA state and downward conformation of the N0 state ( indicated ) . ( B ) Polar interactions fastening the CTD-arm ( red ) in the downward conformation . Involved residues are shown as sticks . ( C ) in the assembled state , the CTD-arm is displaced by RNA ( shown in brown ) . The NTD-arm of the neighboring Ni+1 protomer is colored in green . ( D ) Schematic model of nucleocapsid filament growth . Nascent RNA and the active RdRP complex are indicated . Binding of emerging RNA to Ni primes the displacement of P ( colored in orange ) and attachment of incoming Ni+1 by liberating the CTD-arm ( colored in red ) . The dotted arrows indicate that CTD-arms switch to the upward conformation and latch onto incoming N during attachment of the next N protomer . ( E ) Multiple sequence alignment of CTD-arms from Paramyxoviridae family members . Residues are colored using the ClustalX color scheme . The consensus secondary structure is indicated below the alignment . Virus name abbreviations are given in Materials and methods . DOI: http://dx . doi . org/10 . 7554/eLife . 12627 . 01110 . 7554/eLife . 12627 . 012Figure 3—figure supplement 1 . CTD-arms in other Mononegavirales family members . ( A ) Conformational change of the CTD-arm in HMPV as in Figure 3A . ( B and C ) negative charges within the CTD-arm are topologically conserved in Borna virus ( Bornaviridae ) Rudolph et al . , 2003 and Ebola virus ( Filoviridae ) . Downwards motion of the CTD-arms could position them into the RNA-binding cleft ( indicated by dotted arrows ) , analogous to what is observed in HMPV . DOI: http://dx . doi . org/10 . 7554/eLife . 12627 . 012 Based on the comparison of our N-RNA and N0-P structures we suggest a model for nucleocapsid growth ( Figure 3D ) . Upon delivery of fresh N0-P to the growth site , addition of the next N protomer ( Ni+1 ) to the filament necessitates that the CTD-arm of the terminal Ni unbinds and flips upwards ( Figure 3D , indicated by dotted arrow ) , latching onto Ni+1 and displacing P . In our model , this is driven by the formation of new interactions to the NTD- and CTD arms and , importantly , the concerted insertion of nascent RNA into the RNA binding cleft of Ni , with the CTD-arm switching into the upward conformation . In this model the growth of the filament is reminiscent of a zipper closing up with one row of teeth corresponding to nascent viral RNA and the other to newly delivered N subunits which interdigitate in a fluid , concerted motion . The notion that concerted RNA insertion is required for the hand-over of N subunits from P lends additional specificity to the nucleocapsid polymerization reaction . We hypothesized that the role of the CTD-arm in inhibiting premature RNA binding may be conserved and therefore compared sequences throughout Paramyxoviridae ( Figure 3E ) . We find a semi-conserved LGLT-motif within the CTD-arms which is followed by a stretch of residues with helical propensity . The beginning of this stretch preferentially features negatively charged residues at positions equivalent to HMPV which may in turn pack against the complementary charges of the RNA binding cleft . Indeed , analysis of structures of more distantly related members of Mononegavirales shows that these negatively charged residues are topologically conserved and that a switch to the downward conformation would position these residues into the RNA binding cleft ( Figure 3—figure supplement 1 ) . In conclusion , the reported structures of a paramyxoviral N protein reveal two distinct conformational states , N bound either to the polymerase cofactor P or to RNA . A direct comparison of these two structures provides a molecular level rationale for how nucleocapsid assembly is controlled through P by sterically blocking the binding sites of the NTD- and CTD-arms . In addition , this work elucidates a key role of the CTD-arm in hindering premature RNA insertion into the binding cleft , thus presenting a mechanistic explanation of how premature RNA uptake is directly inhibited in Paramyxoviridae . Peptides of the N0-binding region of P have previously been shown to inhibit replication activity in RSV ( Galloux et al . , 2015 ) , Nipah virus ( Yabukarski et al . , 2014 ) , and rabies virus ( Castel et al . , 2009 ) . The characterization of P-binding surfaces on N proteins is therefore of biomedical importance as these surfaces constitute genuine targets for the development of antivirals . The full-length N gene from human metapneumovirus ( strain NL1-00 , A1 ) was cloned into the pOPINE expression vector , which includes a C-terminal His-tag , using the In-Fusion system ( Takara Clontech , Mountain View , CA ) following standard procedures . The construct was verified by sequencing . Rosetta2 E . coli cells harboring the expression plasmid were grown at 37°C in terrific broth containing appropriate antibiotics and expression was induced at an OD600 of 0 . 8 by adding isopropyl β-D-1-thiogalactopyranoside to 1 mM . The temperature was then lowered to 18°C and after further 18 hrs the cells were harvested by centrifugation ( 18°C , 20 min , 4000 x g ) . Cell pellets were resuspended in 40 mL of 25 mM Tris , pH 8 , 1 M NaCl per L of culture and lysed by sonication . The lysate was centrifuged ( 4°C , 45 min , 50000 x g ) and the supernatant was filtered and loaded on a column containing pre-equilibrated Ni2+-nitrilotriacetic ( NTA ) agarose ( Qiagen , Netherlands ) . The column was washed and the protein was eluted in 25 mM Tris , pH 8 , 1 M NaCl , 400 mM imidazole . The eluate was further purified by size exclusion chromatography using a Superose6 10/300 column ( GE Healthcare , United Kingdom ) equilibrated in 25 mM Tris , pH 8 , 1 M NaCl . The protein was buffer exchanged into 25 mM Tris , pH 8 , 150 mM NaCl , 500 mM NDSB201 , 50 mM Arginine using a PD10 column ( GE Healthcare ) and then concentrated to ~4 mg/mL for crystallization . The N0-P hybrid gene was generated by fusing the sequence corresponding to the first 40 residues of HMPV P ( strain NL1-00 , A1 ) to the 3’ end of the full-length N gene using overlapping primer PCR . The resulting hybrid construct was cloned into POPINE as described above and verified by sequencing . Protein expression was carried out as described for N , above . Cell pellets were resuspended in 20 mM Tris , pH 7 , 1M NaCl , lysed by sonication and the lysate was subsequently centrifuged ( 4°C , 45 min , 50000 x g ) . The supernatant was purified using a column containing pre-equilibrated Ni2+-NTA agarose and elution was carried out using 20 mM Tris , pH 7 , 1M NaCl , 300 mM imidazole . The protein was then buffer exchanged into 20 mM Tris , pH 7 , 100 mM NaCl and loaded onto a HiTrap Heparin HP column ( GE Healthcare ) for further purification using a stepwise NaCl gradient . Finally , the N0-P hybrid was gel-filtrated using a Superdex 75 column ( GE Healthcare ) equilibrated with 20 mM Tris , pH 7 , 100 mM NaCl , and concentrated to ~7 mg/mL for crystallization . Sitting drop , vapor diffusion crystallization trials were set up in 96-well Greiner plates using a Cartesian Technologies robot ( Walter et al . , 2005 ) . A diamond-like , diffraction quality N-RNA crystal was obtained after 132 days in mother liquor containing 100 mM Tris/Bicine , pH 8 . 5 , 90 mM NPS ( NaN03 , Na2HPO4 , ( NH4 ) 2SO4 ) , 37 . 5% methyl-2 4-pentanediol , polyethylene glycol 1000 and polyethylene glycol 3350 of the MORPHEUS crystal screen . The crystal was frozen in liquid nitrogen and diffraction data up to 4 . 2 Å were recorded at 100 K on the I04-1 beamline at Diamond Light Source , Didcot , UK . For the N0-P hybrid , crystals were obtained via in-situ proteolysis ( Dong et al . , 2007 ) using 1 µg of trypsin per 1000 µg of sample . The trypsin was added to the concentrated N0-P preparation just before setting up the crystallization trials . Initial crystals formed in mother liquor containing 100 mM PCB System , pH 7 , 25% polyethylene glycol 1500 and improved crystals could be grown with additives of the Hampton Silver Bullet screen ( 9 mM 1 , 2-diaminocyclohexane sulfate , 6 mM diloxanide furoate , 17 mM fumaric acid , 10 mM spermine , 9 mM sulfaguanidine and 20 mM HEPES , pH 6 . 8 ) . The crystals were cryoprotected in 25% glycerol and frozen in liquid nitrogen . Diffraction data up to 1 . 9 Å were recorded at 100 K on the I04 beamline at Diamond Light Source , Didcot , UK . All data were processed and scaled with XIA2 ( Winter , 2010 ) . The structure of N0-P was solved by molecular replacement using PHASER ( McCoy et al . , 2007 ) with the structure of RSV N ( Tawar et al . , 2009 ) as a search model . Iterative rounds of refinement using PHENIX ( Adams et al . , 2010 ) with TLS parameters and manual building in COOT ( Emsley and Cowtan , 2004 ) resulted in a model for HMPV N starting at residue 30 and ending at residue 383 of the total 394 . Residues 101 to 111 were found to be disordered and were not included in the model . Of the 40 P residues contained in our N0-P construct the first 28 were well-resolved . The structure of the RNA-bound subnucleocapsid ring was solved with PHASER ( McCoy et al . , 2007 ) using a decameric model of our high-resolution HMPV N structure as a search model . Initially , we performed iterative rounds of manual building with COOT ( Emsley and Cowtan , 2004 ) and refinement using PHENIX ( Adams et al . , 2010 ) with non-crystallographic symmetry ( NCS ) constraints to lower the parameter to observations ratio . To aid model building we made use of density modified maps obtained with PHENIX RESOLVE ( Adams et al . , 2010 ) and Parrot of the CCP4 suite ( Winn et al . , 2011 ) in combination with B-factor sharpening . Later stages of refinement were performed with autoBuster ( Smart et al . , 2012 ) , applying NCS restraints , TLS parameters and using our high-resolution N0-P structure to generate reference model restraints . Structures were validated with MolProbity ( Chen et al . , 2010 ) resulting in overall MolProbity scores of 0 . 95 and 2 . 22 for N0-P ( at 1 . 9 Å ) and N-RNA ( at 4 . 2 Å ) , respectively . Refinement and geometry statistics are given in Table 1 . Multiple sequence alignments ( MSA ) were carried out with PROMALS3D ( Pei and Grishin , 2014 ) and figures were prepared with Jalview . Nucleoprotein sequences of the following viruses were used: HMPV , Human metapneumovirus , AMPV , Avian metapneumovirus , RSV , Respiratory syncytial virus , MPV , Murine pneumonia virus , BRSV , Bovine respiratory syncytial virus , CPV , Canine pneumonia virus , MeV , Measles virus , MuV , Mumps virus , RPV , Rinderpest virus , HPIV5 , Human parainfluenza virus 5 , SeV , Sendai virus , HPIV2 , Human parainfluenza virus 2 , SV41 , Simian virus 41 , NiV , Nipah virus , HeV , Hendra virus , CDV , Canine distemper virus , MENV , Menangle virus . N-RNA rings were analysed via electron cryomicroscopy ( cryo-EM ) . Aliquots ( 3 µl ) of N-RNA preparations were pipetted onto glow-discharged Cflat holey carbon grids ( Protochips , Raleigh , NC ) and excess liquid was blotted with filter paper for 3 s . Grids were then plunge-frozen in an ethane-propane mixture at liquid nitrogen temperature using a CP3 plunging device ( Gatan ) . Cryo-EM data were acquired using a 300-kV Polara transmission electron microscope ( FEI ) equipped with a K2 Summit direct electron detector ( Gatan ) and using defocus values ranging from -2 . 0 to -6 . 0 μm at a calibrated magnification of 37 , 000x , resulting in a pixel size of 1 . 35 Å . The contrast transfer function ( CTF ) parameters were determined using CTFFIND3 ( Mindell and Grigorieff , 2003 ) and 2D-classification was carried out with RELION ( Scheres , 2012 ) .	Human metapneumovirus ( HMPV for short ) is a major cause of infections of the airways and lungs , particularly in children , elderly individuals and people with weakened immune systems . As for all viruses , HMPV cannot survive on its own . Instead , it must invade and hijack cells in order to replicate its own genetic material and form new viruses . In HMPV , this genetic information is in the form of a strand of RNA , and is protected by a shell-like structure called a nucleocapsid . Drugs that disrupt the nucleocapsid may therefore help to kill the viruses and treat the illnesses that they cause . Nucleocapsids are built out of many copies of a protein called nucleoprotein , which binds to a strand of RNA . However , viral nucleocapsids can only be built from nucleoproteins that are bound to viral RNA . Potentially , nucleoproteins could instead bind to RNA belonging to the cells that HMPV infects and they would then be trapped in a dead-end state . To prevent this type of unproductive binding , before the nucleocapsid is formed the nucleoprotein is kept unassembled with the help of another protein called the polymerase cofactor . However , it was not clear exactly how the polymerase cofactor helps to maintain this unassembled state . Using techniques called cryo-electron microscopy and X-ray crystallography , Renner et al . studied the structures formed when nucleoproteins are either bound to RNA or are unassembled and bind to the polymerase cofactor . Comparing these structures revealed that RNA normally binds to a specific cleft in the nucleoprotein . However , when nucleoprotein is bound to the polymerase cofactor a portion of the nucleoprotein folds into this cleft instead , blocking the insertion of RNA . This prevents the nucleoprotein from associating with the wrong RNA , allowing the nucleoprotein to remain in an unassembled state until it is needed for the virus . Renner et al . also found that the interactions between the nucleoprotein and the polymerase cofactor of HMPV occur at sites that are also found in several other related viruses , such as Ebola . Targeting this common region could therefore be a good strategy for developing new antiviral drugs .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "short", "report", "structural", "biology", "and", "molecular", "biophysics", "microbiology", "and", "infectious", "disease" ]	2016	Nucleocapsid assembly in pneumoviruses is regulated by conformational switching of the N protein
Evolutionary adaptations of temporo-parietal cortex are considered to be a critical specialization of the human brain . Cortical adaptations , however , can affect different aspects of brain architecture , including local expansion of the cortical sheet or changes in connectivity between cortical areas . We distinguish different types of changes in brain architecture using a computational neuroanatomy approach . We investigate the extent to which between-species alignment , based on cortical myelin , can predict changes in connectivity patterns across macaque , chimpanzee , and human . We show that expansion and relocation of brain areas can predict terminations of several white matter tracts in temporo-parietal cortex , including the middle and superior longitudinal fasciculus , but not the arcuate fasciculus . This demonstrates that the arcuate fasciculus underwent additional evolutionary modifications affecting the temporal lobe connectivity pattern . This approach can flexibly be extended to include other features of cortical organization and other species , allowing direct tests of comparative hypotheses of brain organization . The temporal lobe is a morphological adaptation of the brain that is unique to primates ( Bryant and Preuss , 2018 ) . Its origins likely include expansion of higher-order visual areas to accompany the primate reliance on vision ( Allman , 1982 ) . Temporal association cortex contains areas devoted to higher-level visual processing and social information processing ( Rushworth et al . , 2013; Sallet et al . , 2011 ) that , in turn , rely strongly on visual information in primates ( Perrett et al . , 1992 ) . The expanded temporal cortex in apes and humans contains several multimodal areas and areas associated with semantics and language ( Dronkers et al . , 2004; Hickok and Poeppel , 2007; Price , 2000 ) . As such , understanding the evolution of temporal cortex across the primate order is a vital step to understanding primate behavioral adaptations . Two lines of evidence are often brought to bear on differences in temporal lobe organization across humans and other primates . The first line emphasizes selective local expansions of temporal cortex and subsequent relocation of areas . Morphologically , great apes possess an extra sulcus in the temporal cortex , suggesting at the very least expansion of this part of cortex . Mars et al . ( 2013 ) reported a region in the middle part of the superior temporal sulcus of the macaque that shares anatomical features of the human temporo-parietal junction area located at the caudal end of the temporal cortex , suggesting a major relocation of this area . In a similar vein , Patel et al . ( 2019 ) suggest that expansion of the temporo-parietal junction and superior temporal sulcus gave rise to a modified ventral visual processing stream to support increased social abilities in humans . The second line of evidence emphasizes changes in the connectivity of the temporal lobe . Rilling et al . ( 2008 ) first suggested dramatic expansion of the arcuate fasciculus temporal cortex projections in the human , but more recent studies also emphasize increased projections of the middle longitudinal and inferior fronto-occipital fasciculi and their role in language-related processes in the human ( Catani and Bambini , 2014; Makris et al . , 2013; Makris and Pandya , 2009; Saur et al . , 2008 ) . These different schools place different emphasis on what happened to temporal cortex across different primate lineages . Their results , however , should be interpreted in relation to one another as species differences in brain organization can come in many forms that can interact in unpredictable ways ( Krubitzer and Kaas , 2005; Mars et al . , 2018a; Mars et al . , 2017 ) . Dissociating such different types of species differences is challenging ( Figure 1A , B ) . For instance , given an ancestral or reference state , local expansions of the cortical sheet can lead to the relocation of homologous areas between two species . As a case in point , human MT+ complex is located much more ventrally in posterior temporal cortex than its macaque homolog ( Huk et al . , 2002 ) . Such cortical relocations also affect the location of connections of these areas , but this situation is distinct from the scenario in which a tract extends into new cortical territory . These two scenarios are illustrated in Figure 1B . The red area in the top panel has expanded , leading to a relocation of the blue area with respect to the purple and yellow area . The connections of the areas do not change in this scenario , resulting in a relocation of the connections of the blue area . In the bottom panel , a yellow area’s tract terminations have invaded the neighboring purple territory , but this change is independent from cortical expansion . Thus , in both cases connections are located in a different place from those in the reference state , but the causes are different . In this study we investigate to which extent species differences in temporal lobe organization are due to cortical relocation and tract extension . To be able to do this , we propose a framework to test among different forms of cortical reorganization by registering brains together into a single shared coordinate system ( Figure 1C , D ) . Such an approach allows us to place different brains into a common space based on one feature and then compare the results to registration based on another feature . This ‘common space’ concept proved feasible in a previous study testing whether the extension of the human arcuate fasciculus ( AF ) compared with the macaque AF could be accounted for by differential cortical expansion between the two brains ( Eichert et al . , 2019 ) . In the present study , we generalize this approach to develop a cross-species registration based on a multimodal surface matching algorithm ( MSM , Robinson et al . , 2018; Robinson et al . , 2013; Robinson et al . , 2014 ) to derive a cortical registration between different species . We based our registration framework on whole brain neuroimaging data of macaque , chimpanzee and human brains . Neuroimaging allows one to acquire high-resolution data from the same brains using different modalities within a short time . The digital nature of the data allows easy manipulation , making it ideal for the present purposes ( Le et al . , 1985; Grannell and Mansfield , 1975; Lauterbur , 1973; Le Bihan et al . , 1986; Thiebaut de Schotten et al . , 2019 ) . As primary modality we use surface maps derived from the cortical ribbon of T1- and T2-weighted scans , which have been shown to correlate well with cortical myelinization and which are available for all three species ( Glasser et al . , 2014; Glasser and Van Essen , 2011 ) . Such ‘myelin maps’ can be used to identify homologous areas across brains and species , such as primary sensory and motor cortex , which is high in myelin , and association cortex , which is low in myelin ( Glasser et al . , 2014; Large et al . , 2016 ) . As second modality , we use diffusion MRI tractography to reconstruct long range white matter fibers of the temporal and parietal lobes to establish its connections ( Bryant et al . , 2019; Mars et al . , 2018c ) . Given data from these two modalities , we developed an approach to reveal different types of cortical reorganization . We argue this approach is particularly suitable to study the temporal lobe , as it has well described myelin markers and cortical connections ( Glasser et al . , 2014; Large et al . , 2016; Mars et al . , 2013; Ruschel et al . , 2014 ) . First , we register the cortical surfaces of the different species to one another based on myelin maps ( Figure 1C ) . This cortical alignment uses the distinction of primary and higher order areas in myelin maps as anchor points across all three species . Next , we apply this registration to overlay homologous parts of the cortex and to calculate the underlying distortions of the cortical sheet . This registration field effectively models areal expansion or contraction underlying cortical relocation of homologous areas . We then apply this registration to the cortical projection maps of temporal and inferior parietal lobe white matter tracts to assess how well the myelin-based registration can predict changes in tract projection patterns across species ( Figure 1D ) . A good prediction , i . e . a high overlap of the tract maps , indicates that cortical expansion and relocation of targets zones alone can predict tact projections ( Figure 1D , top ) , a poor prediction indicates that the tract is reaching new cortical territory ( Figure 1D , bottom ) . Here , we distinguish different scenarios of cortical evolution for a set of temporal and parietal white matter tracts . By applying a cross-species registration we can infer if only cortical relocation was affecting a tract’s connectivity profile or if a tract is reaching into new cortical territory . A deeper understanding of species differences in brain reorganization is essential for our understanding how evolutionary specializations of the temporal lobe underlie uniquely human cognitive functions . We developed a surface registration between species based on myelin maps using multimodal surface matching ( MSM , Robinson et al . , 2018 ) . Figure 2 shows the final results of chimpanzee-to-human , macaque-to-chimpanzee , and macaque-to-human brain registrations . The cross-species registration aligns the myelin maps well , with the predicted human maps showing most of the distinctive features of the actual human myelin map ( Figure 2 , top row ) . Posterior areas such as V1 are well aligned , with the highest myelin evident on the medial part of the occipital cortex , having relocated quite substantially from a more lateral orientation in the macaque . The prominent myelin hot spot in the location of the MT+ complex is also noticeable . Areas where the myelin maps showed fewer distinctive features to guide the registration , such as in the prefrontal cortex , showed some differentiation between the predicted and actual human maps . Spatial correlation maps of the human myelin maps and the predicted myelin maps as well as the deformation fields underlying the registrations are provided in Appendix 2—figure 1 . We constructed the cortical projection maps of the following tracts in all three species: Middle longitudinal fasciculus ( MDLF ) , inferior longitudinal fasciculus ( ILF ) , the third branch of the superior longitudinal fasciculus ( SLF3 ) , the inferior fronto-occipital fasciculus ( IFO ) , and the arcuate fasciculus ( AF ) ( Figure 3A , C , E ) . The human and macaque tract maps resemble those obtained in previous studies ( Mars et al . , 2018c; Schmahmann and Pandya , 2009 ) and the chimpanzee SLF3 and AF are similar to previous reports ( Hecht et al . , 2015; Rilling et al . , 2008 ) . The other chimpanzee tracts are reported here for the first time , apart from a previous exploratory study ( Mars et al . , 2019 ) . We applied the myelin-based surface registration to assess whether the cortical relocation demonstrated in the myelin registration above fully explains the changes in tracts . Figure 3 shows actual and predicted tract maps . For visual assessment , a thresholded overlay of actual human and predicted tract maps is shown in Figure 4A . As described above , we focus on a description of temporo-parietal cortex given the multiple competing theories of its reorganization in different primate lineages . We assessed the success of the myelin registration in predicting the tract projections in a number of ways . First , weighted correlation maps provide a visualization of the local quality of the prediction ( Figure 4B , C ) . A high value means that the myelin registration alone is sufficient to predict a tract’s projection in this part of the brain . A low correlation value indicates that reorganization of a tract’s connectivity pattern took place in addition to cortical relocation modelled by the myelin registration . Second , the Dice coefficient of similarity provides a more general measure of similarity between the predicted and actual tract maps , where a Dice coefficient of ‘1’ indicates perfect overlap and thus no tract extension into areas other than would be predicted by cortical relocation assessed using the myelin map registration . Finally , we calculated a ‘tract extension ratio’ that indicates how much of the actual human tract projections extends into parts of the surface not predicted , where a value of >1 indicates a tract extension into novel territory . Both the Dice coefficients and tract extension ratios were computed for thresholded tract maps defined by the human tract map covering 40% of the brain’s surface , but the resulting pattern of values is robust across a range of thresholds ( see Appendix 3—figure 2 ) . In general , it can be observed that the myelin-based registration can predict the tract maps well in both hemispheres , with the notable exception of AF and to a lesser extent ILF and SLF3 ( Figure 4 ) . AF in particular shows the lowest Dice coefficient and the highest extension ratio ( Figure 5A , B ) , indicating that this tract’s differential projections in the human brain are not merely due to relocation of areas . The maps for macaque and chimpanzee are overall predicted to a similar degree , as can be seen in the overlay and correlation maps , with the exception of AF ( Figure 4 ) . The effect for AF is captured in the Dice coefficients and tract extension measures ( Figure 5A , B ) . A two-way statistical analysis was performed in both hemispheres to assess the effect of species and tract on the extension ratios . In the left hemisphere , there was no significant main effect of species ( F ( 1 , 179 ) =1 . 38 , p=0 . 47 ) , but a highly significant main effect of tract ( F ( 4 , 792 ) =565 . 00 , p<0 . 001 ) and a highly significant interaction effect of species and tract ( F ( 4 , 792 ) =207 . 73 , p<0 . 001 ) . In the right hemisphere , we found a significant main effect of species ( F ( 1 , 179 ) =16 . 76 , p<0 . 001 ) as well as a highly significant effect of tract ( F ( 4 , 792 ) =261 . 94 , p<0 . 001 ) and a highly significant interaction effect ( F ( 4 , 792 ) =225 . 70 , p<0 . 001 ) . We will discuss the various tracts in more detail below . The myelin-based registration results in good prediction for tract projections in the temporo-parietal cortex . The actual human tract terminations of MDLF span the superior temporal gyrus and reach the inferior parietal cortex ( Figure 4A ) . In the macaque and chimpanzee , the actual MDLF terminates in superior temporal gyrus but reaches only to a small part of the inferior parietal cortex . When applying the myelin-based registration , macaque and chimpanzee MDLF are both predicted to reach a comparable portion of the human temporal lobe and parts of both angular and supramarginal gyri of the inferior parietal lobe ( Makris et al . , 2013 ) . This overlap is captured in the weighted correlation maps , which have high values in the temporal lobe ( Figure 4B , C ) . The Dice coefficients for the chimpanzee and macaque MDLF are high and the extension ratio is close to one indicating no tract extension in addition to cortical expansion ( Figure 5A , B ) . A similar observation can be made for the posterior terminations of SLF3 and IFO . The myelin-based registration can predict the parietal cortical projections to a large degree . The predicted cortical terminations of the tract show a strong overlap with the actual human tract terminations . In line with the overlay maps and the weighted correlation maps , the Dice coefficients are relatively high . Taken together , this suggests that expansion and relocation of brain areas are largely sufficient to model the posterior cortical terminations of SLF3 and IFO , while extension of the tract’s connectivity pattern plays a minor role in explaining the species differences . Predicted ILF terminations show that the expected occipito-temporal connection can be modelled well ( Catani and Thiebaut de Schotten , 2008 ) . The extension ratio for ILF is elevated indicating that there is some remaining tract extension that has not been modelled by cortical relocation . This is also reflected in the tract overlay ( Figure 4A ) , which shows that human ILF has more extended posterior projections than predicted by the myelin registration . Thus , although the overall architecture of the tracts is well predicted , this tract seems to have extended into new cortical territory in the human lineage . The clearest case of a tract extension in the human brain was presented by AF . Human AF reaches anteriorly to inferior and dorsal frontal areas . The posterior projections of human AF reach into middle and inferior temporal cortex . The chimpanzee posterior terminations are in inferior parietal lobe and superior temporal cortex and in the macaque , the temporal projections reach superior temporal areas . For AF , the myelin-based registration does not provide a good prediction of the tract map across species , especially for the macaque . The correlation map for the chimpanzee shows low correlation along the temporal lobe and for the macaque , correlation values in temporal lobe are extremely low . AF has the lowest Dice coefficient and the extension ratio is high , especially in the macaque , which is in line with overlay and correlation maps . The ‘failure’ of the myelin registration in the temporal lobe indicates that extension and relocation of cortical areas is not sufficient to explain the posterior tract projections of AF , but that the tract extended into new cortical territory in the temporal lobe . To further characterize the effects in the predicted tract maps , we obtained connectivity fingerprints at two representative vertices on the left brain surface: One in inferior parietal lobe , where most tracts are predicted well and one in the middle temporal gyrus , where we observed strong species differences , in particular for AF . The connectivity fingerprints were derived based on the intensity values in the actual and predicted tract maps using an extended set of seven tracts to give a more detailed picture ( see Appendix 3—figure 1 ) . Figure 5C demonstrates that in the inferior parietal lobe , the connectivity fingerprint of actual human and predicted chimpanzee and macaque tract maps are highly similar , except for a small increase in the intensity of ILF indicating tract extension as discussed above . This indicates that the myelin-derived registration can predict this area’s connectivity profile well , despite the local expansion of the cortical sheet . For the temporal vertex , however , there is a strong mismatch regarding the connectivity profile , in particular regarding the intensities for AF . This indicates that the connectional fingerprint of this temporal area is different in the human than would be predicted purely based on cortical relocation . Thus , the connectivity fingerprints of the two representative areas match the pattern of species differences that emerged from the results above . The goal of this study was to study brain specializations of the temporal lobe across multiple primate species in the context of two forms of cortical reorganization: cortical relocation due to local expansions and extensions of tracts into new cortical territory ( Figure 1 ) . For this reason , we developed a cross-species surface registration method based on cortical myelin content , which gives us an index of how cortical areas have relocated during evolution . In a subsequent step , we tested if cortical relocation can predict the connectivity patterns of a set of tracts across species . We showed that cortical expansion and resulting relocation of brain areas alone provide a good prediction of several tracts’ terminations in posterior temporal and parietal cortex . In the case of AF in particular , we showed that an additional change in brain architecture was extension of the tract into new cortical areas independent of cortical expansion and resulting relocation . As pointed out in the introduction , different lines of evidence on temporal reorganization across primate species have emphasized either the expansion and subsequent relocation of brain areas or changes in temporal lobe connectivity . Both approaches are valuable and we here demonstrate that both are applicable to different parts of the temporal lobe . Previous work has suggested expansion and relocation of areas in posterior temporal cortex and adjacent parietal cortex ( Mars et al . , 2013; Patel et al . , 2019 ) , which is consistent with the ventral location of MT+ complex in humans compared to other primates . Similar reorganizations have been suggested by Haak et al . ( 2018 ) , who proposed modifications of visual-temporal pathways in the human . These major relocations were captured by our myelin registration and demonstrated to predict certain features of white matter tracts , such as the posterior projections of MDLF . The aim of our research is not to find the ‘best’ species registration , but to shed light on brain evolution by studying where registrations based on different modalities disagree . Many aspects of brain architecture can be modified during the course evolution and each individual aspect would provide us with a unique registration . The present study represents only one example of how cortical specializations can be studied by comparing cross-species registrations of different modalities . Testing the effect of a myelin registration on connectivity is not meant to imply pre-eminence of myelin over connectivity . In fact , the reversed approach , i . e . to derive a registration based on individual tract maps and then test their differential effect on transformed myelin maps , would be highly informative . We chose myelin here as the primary feature , because we wanted to test the specific hypothesis that some tracts expanded while others simply followed relocation of areas across the cortex . Thus , we used a measure that could index the relocation ( myelin ) and test it on our measure of interest ( the tracts ) . Separate from cortical relocation , changes in connectivity between humans and other species have been described for a variety of association tracts , including the temporal projections of AF into middle temporal gyrus ( Eichert et al . , 2019; Rilling et al . , 2008 ) , the frontal projections of SLF3 ( Hecht et al . , 2015 ) , and expansion of the ventral route consisting of MDLF and IFO ( Forkel et al . , 2014; Makris et al . , 2013; Makris et al . , 2009 ) . Both of these latter tracts have been suggested to play a role in language functions in the human brain ( Catani and Bambini , 2014; Hagoort , 2016; Makris et al . , 2009; Makris and Pandya , 2009; Saur et al . , 2008 ) . In the case of MDLF , the human tract projections can be predicted well by the cortical relocation model . This indicates that the pattern of cortical terminations changed according to the general expansion and relocation of brain areas , without additional extension into new regions of the brain . For other tracts reaching to temporo-parietal areas , such as ILF , IFO and SLF3 , the posterior tract projections can also be modelled well , despite the large distortions of target areas within the cortical sheet . These tracts seem to follow the evolutionary scenario described in the upper row in Figure 1B , D , where a tract’s extension in the human brain can be explained by relocation of areas along the cortical sheet . This does not necessarily mean that the tracts have not been recruited for new functions , but the type of change is different from that of tracts such as AF . AF showed the lowest consistency across species when applying the myelin-based registration . Dice ratio and extension ratio reflect the increased tract termination especially into the temporal lobes , which can be observed in the tract maps . This points to further evolutionary adaptations that specifically affected the connectivity pattern of AF independent of cortical relocation , the scenario described in the bottom figure in Figure 1B , D . Our result is consistent with previous accounts in the literature ( Ardesch et al . , 2019; Eichert et al . , 2019; Rilling et al . , 2008 ) , but the approach described here enabled us to formally test this hypothesis in the wider context of cortical reorganization across three primate species and to quantify the species differences . Importantly , extension of a tract into new cortical territories alters the unique connectivity fingerprint of the innervated areas , which profoundly changes the computational capabilities that area supports ( Mars et al . , 2018b ) . Being able to dissociate different modifications of brain architecture can inform us about how temporal lobe specializations link to uniquely human higher cognition ( Qi et al . , 2019; Roelofs , 2014; Schomers et al . , 2017 ) . Apart from modifications of AF , we also noted some minor extensions of ILF into temporo-parietal cortex . ILF’s extension in the human brain is consistent with reports that that showed a split of this tract into multiple subtracts due to the expansion of parts of the temporal cortex , including the fusiform gyrus ( Latini et al . , 2017; Roumazeilles et al . , 2019 ) . This extensions could be related to the increase of cortical territory related to processing social information , such as social networks and faces ( Noonan et al . , 2018; Sallet et al . , 2011 ) . Previously , it has been shown that parietal SLF3 projections are most prominent in the human brain , which has been linked to our unique capacity of social learning ( Hecht et al . , 2013 ) . We show that species differences in the posterior projections of SLF3 can be mostly explained by local expansion of the posterior temporal and parietal cortex . Similarly , we show that cortical expansion can model the terminations of MDLF , a tract , which has been linked to visuospatial and integrative audiovisual function ( Makris et al . , 2013 ) . Our results thus suggest that SLF3 and MDLF didn’t undergo additional evolutionary modifications that affected their posterior terminations . The MSM framework we adopted is ideally suited to work with multimodal descriptors of the cerebral cortex . It has become a vital tool for human surface registration ( Abdollahi et al . , 2014; Garcia et al . , 2018; Glasser et al . , 2016 ) and here we demonstrated its utility for cross-species research . With the presented surface matching method , we showed that a registration based on T1w/T2w MRI data can match critical landmarks across species . We have referred to these maps as ‘myelin maps’ in accordance with other studies in the literature ( Glasser et al . , 2014; Large et al . , 2016 ) but it should be noted that this is a heuristic . T1w/T2w maps are sensitive to other features than myelin and other sequences are sensitive to aspects of cortical myelin ( Lutti et al . , 2014 ) . The crucial point is that the maps we employed here are similar across species , allowing us to compare like with like ( Glasser et al . , 2014 ) . Projecting data of different modalities to a surface representation is a useful tool for comparative neuroscience . It allows us to visualize the different modalities within the same cortical sheet and to compare topologies on this 2D surface , which opens a wide array of mathematical tools . A similar approach has been taken to investigate the relationship between gene expression and myelin content of the cortex ( Burt et al . , 2018 ) and gradients of change across multiple modalities of brain organization ( Blazquez Freches et al . , 2020; Huntenburg et al . , 2018 ) . The presented approach can be flexibly modified to include a variety of cortical features , which can be compared across species . Myelin does not provide high contrast in the large human frontal cortex and , as such , it is difficult to provide a good registration in frontal areas . Furthermore , the effects we report can only be reliably interpreted within the spatial resolution of brain areas . More fine scale species differences and homology assignments are not possible with the data shown here . However , the current method can be generalized to any modality of cortical organization , so future studies can incorporate modalities that have greater contrast in this part of the brain such as neurite orientation dispersion and density imaging ( NODDI ) measures ( Zhang et al . , 2012 ) and resting state fMRI networks ( Vincent et al . , 2007 ) . In sum , here we present a framework for analyzing structural reorganization of the temporo-parietal cortex across different primate brains . We dissociated cortical relocation of areas due to local expansion and modifications of white matter tract connectivity . Future work will expand this approach not only to different modalities , but also to a much wider range of species , which is now becoming increasingly possible due to the availability of multi-species datasets ( Heuer et al . , 2019; Milham et al . , 2018 ) . This provides a crucial step towards the understanding of phylogenetic diversity across the primate brain . Human data were acquired in 20 subjects ( 12 females , 18–40 years ) on a 3T Siemens Prisma scanner with a 32-channel head coil . The study was approved by the Central University ( of Oxford ) Research Ethics Committee ( CUREC , R55787/RE001 ) in accordance with the regulatory standards of the Code of Ethics of the World Medical Association ( Declaration of Helsinki ) . All participants gave informed consent to their participation and were monetarily compensated for their participation . High-resolution structural images were acquired using a ( MPRAGE ) T1w sequence ( TR = 1900 ms; TE = 3 . 97 ms; flip angle = 8°; 192 mm FoV; voxel size 1 mm isotropic ) and ( SPC ) T2w sequence ( TR = 3200 ms; TE = 451 ms; 256 mm FoV; voxel size 1 mm isotropic; Grappa factor = 2 ) . Diffusion-weighted ( DW ) MRI data were acquired in the same subjects using a sequence from the UK Biobank Project ( Miller et al . , 2016 ) . In brief , we used a monopolar Stejskal-Tanner diffusion encoding scheme ( Stejskal and Tanner , 1965 ) . Sampling in q-space included two shells at b = 1000 and 2000 s/mm2 ( voxel size 2 mm , MB = 3 ) . For each shell , 50 distinct diffusion-encoding directions were acquired ( covering 100 distinct directions over the two b-values ) . Five b = 0 images were obtained together with additional three b = 0 images with the phase-encoding direction reversed . T1w and T2w scans were pre-processed using the HCP-pipeline ( Glasser et al . , 2013 ) cloned from the ‘OxfordStructural’ - fork ( https://github . com/lennartverhagen/Pipelines ) . The processing pipeline includes automatic anatomical surface reconstruction using FreeSurfer and provides measures of sulcal depth and surface maps of cortical myelin content ( Fischl , 2012; Jenkinson et al . , 2012 ) . The mean image of the T1w scans was divided by the mean image of the T2w scans to create a T1w/T2w image . The bias corrected T1w/T2w-ratio was mapped onto the mid-thickness surface using Connectome Workbench command-line tools . We refer to this surface map as T1w/T2w ‘myelin map’ ( Glasser et al . , 2014; Glasser and Van Essen , 2011 ) . In order to create a human average myelin map , the subject’s individual myelin maps were aligned using MSM . The myelin alignment was initialized using alignment based on maps of sulcal depth ( a table with parameters is provided in Supplementary file 1 ) . To create the species average maps , we used an implementation of MSM that optimizes based on a first-order ( pairwise ) cost function to penalize against distortions , given that no excessive distortions were expected . Human volume data were registered to the Montreal Neurological Institute standard space ( MNI152 ) and surface data was transformed to a surface template space ( fs_LR ) . In vivo chimpanzee structural MRI and DW-MRI data were obtained from the National Chimpanzee Brain Resource ( www . chimpanzeebrain . org ) . Data were acquired at the Yerkes National Primate Research Center ( YNPRC ) at Emory University through separate studies covered by animal research protocols approved by YNPRC and the Emory University Institutional Animal Care and Use Committee ( approval no . YER-2001206 ) . Both structural MRI and DWI-MRI data were collected on a Siemens 3T Trio Scanner ( Siemens Medical System , Malvern , PA , USA ) . These chimpanzee MRI scans were obtained from a data archive of scans obtained prior to the 2015 implementation of U . S . Fish and Wildlife Service and National Institutes of Health regulations governing research with chimpanzees . All the scans reported in this publication were completed by the end of 2012 . T1w/T2w myelin maps were obtained from a group of 29 adult chimpanzees ( all female ) , scanned at 0 . 8 mm isotropic resolution ( Donahue et al . , 2018; Glasser et al . , 2014; Glasser et al . , 2012 ) . T1w and T2w scans were processed using a modified version of the HCP-pipeline ( Glasser et al . , 2013 ) . DW-MRI data were obtained in a subset of five individuals . Acquisition and pre-processing was previously described ( Li et al . , 2013; Chen et al . , 2013; Mars et al . , 2019 ) . Two DW images ( TR = 5900 ms; TE = 86 ms; 41 slices; 1 . 8 mm isotropic resolution ) were acquired using a single-shot spin-echo echo planar sequence for each of 60 diffusion directions ( b = 1000 s/mm2 ) , each with one of the possible left–right phase-encoding directions and four repeats , allowing for correction of susceptibility-related distortion . For each repeat of diffusion-weighted images , five images without diffusion weighting ( b = 0 s/mm2 ) were also acquired with matching imaging parameters . Chimpanzee volume and surface data were registered to a standard space template based on 29 chimpanzee scans acquired at the YNPRC ( Donahue et al . , 2018 ) . A species average myelin map from the 29 chimpanzees was derived using MSM as described for the human . Ex vivo DW-MRI data were obtained from four rhesus macaques ( one female , age at death: range 4–14 years ) using a 7T magnet with Agilent Directive ( Agilent Technologies , Santa Clara , CA , USA ) . Data acquisition and DW-MRI pre-processing have been previously described in detail ( Eichert et al . , 2019; Folloni et al . , 2019 ) . Data were acquired using a 2D diffusion-weighted spin echo multi slice protocol with single line readout ( DW-SEMS; TE = 25 ms; TR = 10 s; matrix size: 128 × 128; resolution 0 . 6 mm; number of slices: 128; slice thickness: 0 . 6 mm ) . Nine non-diffusion-weighted ( b = 0 s/mm2 ) and 131 diffusion-weighted ( b = 4000 s/mm2 ) volumes were acquired with diffusion encoding directions evenly distributed over the whole sphere , except in one monkey were seven non-diffusion-weighted images and 128 diffusion directions were collected . This protocol and similar ones have previously shown to be sufficient for comparison with in vivo human data ( see for example: D’Arceuil et al . , 2007; Dyrby et al . , 2011; Eichert et al . , 2019; Mars et al . , 2016 ) . Additionally , ex vivo data from one male macaque were obtained ( de Crespigny et al . , 2005 ) and pre-processed as described previously ( Jbabdi et al . , 2013 ) . Relevant imaging parameters for DW-MRI data were: 4 . 7T Oxford magnet equipped with BGA 12 gradients; 3D segmented spin-echo EPI 430 μm isotropic resolution , eight shots , TE = 33 ms , TR = 350 ms , 120 isotropically distributed diffusion directions , b = 8000 s/mm2 . Despite the different scanning parameters , data quality was appropriate to allow pooling of the ex vivo data sets . In vivo data from the same macaque subjects was not available . To obtain macaque T1w/T2w myelin maps , in vivo T1w and T2w scans data were obtained from a previous study on five separate rhesus macaques ( four females , age range 3 . 4 years - 11 . 75 years ) . Data acquisition and pre-processing of the macaque data have been described previously ( Bridge et al . , 2019; Large et al . , 2016 ) . Procedures of the in vivo macaque data acquisition were carried out in accordance with Home Office ( UK ) Regulations and European Union guidelines ( EU directive 86/609/EEC; EU Directive 2010/63/EU ) . Macaque surface reconstruction and average myelin maps were derived as described for the human . Macaque volume and surface data were registered to a standard space , which is based on data from 19 macaques acquired at YNPRC ( Donahue et al . , 2018; Donahue et al . , 2016 ) . Our aim was to derive a cross-species registration to model expansion and relocation of cortical brain areas . Therefore , we performed registration based on average surface myelin maps in the three species using MSM with higher-order smoothness constraints ( Ishikawa , 2014; Robinson et al . , 2018 ) . We derived a transformation of the cortical surface so that homologous myelin landmarks across species matched . The general processing steps were as follows , but a more detailed description of the methodology and an explanatory figure are provided in Appendix 1 . We obtained a ‘chimpanzee-to-human’ and a ‘macaque-to-chimp’ registration . A ‘macaque-to-human’ registration was derived as a concatenation of both registration stages to minimize the between-species distortions needed . As input for the registration we used the species average myelin maps and we performed the registration for both hemispheres separately . In general , the registration was derived using two stages . The first stage was based on three regions-of-interest ( ROIs ) to handle the gross distortions that are involved in matching myelin landmarks across species . Two ROIs captured the highly myelinated precentral motor cortex ( MC ) and MT+ complex and a third ROI covered the medial wall ( MW ) . We used MSM to obtain a registration so that the ROIs are roughly matched across species . In the second stage , the ROI-based registration was used as initialization for the subsequent alignment of the whole-hemisphere myelin maps . To derive a macaque-to-human registration , we resampled the average macaque myelin map to chimpanzee space using the MSM-derived macaque-to-chimpanzee registration . Then we aligned the resampled macaque map in chimpanzee space with that of the human and used the chimpanzee-to-human registration as initialization . The quality of the registration performance was assessed by computing a local spatial correlation between the human myelin map and the result of the chimpanzee and macaque registration . Furthermore , we visualized the deformations underlying the registration in form of a surface distortion map . The methods and results for these two analyses are provided in Appendix 2 . . Human and chimpanzee DW-MRI data were pre-processed using tools from FDT ( FMRIB's Diffusion Toolbox , part of FSL 5 . 0 [Smith et al . , 2004] ) . We applied the TOPUP distortion correction tool followed by eddy-current distortion and motion correction ( Andersson et al . , 2003; Andersson and Sotiropoulos , 2016 ) as implemented in FSL . Macaque ex vivo DW-MRI data were processed using tools from FSL as implemented in an in-house MR Comparative Anatomy Toolbox ( Mr Cat , www . neuroecologylab . org ) . Pre-processed DW-MRI images were processed by fitting diffusion tensors ( FSL's DTIFIT [Behrens et al . , 2003] ) and by fitting a model of local fiber orientations including crossing fibers ( FSL's BedpostX; Behrens et al . , 2007; Jbabdi et al . , 2012 ) . Up to three fiber orientations per voxel were allowed . Tractography was performed using FSL’s probtrackx2 . Registration warp-fields between each subject's native space and standard space were created using FSL's FNIRT ( Andersson et al . , 2007 ) . We performed tractography of the following tracts: Middle longitudinal fasciculus ( MDLF ) , inferior longitudinal fasciculus ( ILF ) , the third branch of the superior longitudinal fasciculus ( SLF3 ) , the inferior fronto-occipital fasciculus ( IFO ) , and the arcuate fasciculus ( AF ) . Placement of seed , waypoint , and exclusion masks was based on previous studies , in order to reconstruct known pathways for these tracts in all three species ( human and macaque: de Groot et al . , 2013; Mars et al . , 2018c , protocols for AF: Eichert et al . ( 2019 ) ; chimpanzee: Bryant et al . ( 2018 ) . Masks were drawn in standard space and warped to native subject diffusion MRI space for probabilistic tractography . The resulting tractograms were normalized by dividing each voxel’s value by the total number of streamlines that successfully traced the required route ( ‘waytotal’ ) . To decrease computational load for further processing all tractograms were down-sampled ( human: 2 mm , chimpanzee: 1 . 5 mm , macaque: 1 mm ) . In addition , tractography and surface-based analysis was performed for cortico-spinal tract ( CST ) and vertical occipital fasciculus ( VOF ) . Results for all tracts are reported in Appendix 3—figure 1 . To assess which part of the cortical grey matter might be reached by the tracts , we derived the surface representation of each individual tractogram using a matrix multiplication method described in Mars et al . ( 2018b ) ; Appendix 1—figure 1B ( 2 ) . We calculated whole-hemisphere vertex-wise connectivity matrices , tracking from the 20k-vertices mid-thickness surface to all voxels in the brain . These matrices were computed for both hemispheres and each subject individually in the three species . In the macaque we used the five subject’s average mid-thickness in standard space as input for the computation instead of individual surfaces . To rebalance the weights in the tracts to be more homogenous , connectivity values were weighted by the distance between vertex and voxel . A distance matrix across all vertices of the mid-thickness surface and all brain voxels was computed using MATLAB’s pdist2-function resulting in a matrix of the same size as the connectivity matrix . Each element in the connectivity matrix was then divided by the corresponding value in the vertex-to-voxel distance matrix . To decrease data storage load ( approximately 10 GB per matrix ) the weighted connectivity matrices of the five subjects were averaged for each hemisphere and species . To visualize a tract's surface representation , we multiplied the averaged connectivity matrix with a tract’s tractogram ( ‘fdt_paths’ ) . We refer to the tract surface representation here as ‘tract map’ . The approach described above decreases gyral bias in the resulting tract map notably when compared to surface-based tractography or surface projections of the tractogram . However , the method introduced spurious effects on the medial wall and insular cortex , which are generally not well captured in the tract map . Given that both areas are not of interest in this study , they were masked out for further analysis . Tract maps were derived for each subject and both hemisphere separately . Individual surface maps were smoothed on the mid-thickness surface ( human: 4 mm kernel ( sigma for the gaussian kernel function ) , smoothing on individual surface; chimpanzee: 3 mm kernel , smoothing on average surface; macaque: 2 mm kernel , smoothing on average surface ) , logNorm-transformed and averaged across subjects . Next , we tested if our myelin-based registration can be used to predict the tract maps across species . We resampled individual chimpanzee and macaque tract maps to human space using the macaque-to-human and the chimpanzee-to-human registration ( Appendix 1—figure 1B ( 3 ) ) . Intensity values in actual and predicted tract maps ranged from 0 to 1 . We averaged all predicted tract maps and displayed the average map onto a human average surface ( Q1-Q6_R440 ) . For visual inspection we also assessed and showed thresholded tract maps . Thresholds were chosen different for each tract , ranging from 0 . 6 to 0 . 85 , so that the most characteristic termination is visible . To visualize and quantify the prediction of macaque and chimpanzee tracts in human space , we derived weighted whole-hemisphere local correlation maps of the human map and the map predicted by macaque or chimpanzee . The local correlation map was computed using a sliding window around every vertex on the sphere ( diameter 10 cm for all three species ) using MATLAB’s corrcoef-function ( Mathworks , Natick , MA ) . We used a search kernel of 40° that corresponds to a circular search window with a radius of approximately 7 cm . The correlation map was modified to up-weight the brain areas where the tract is represented on the surface . A weighting mask was derived by multiplication of the intensities in the actual human tract map and the other species’ predicted map . The values for the weighted correlation map are thus high in parts of the brain where both actual human and predicted tract show a termination , and where the spatial patterns of intensity values correlate . Weighted correlation maps were derived for each pair of 20 human subjects and five subjects of the other species . As result figure we display the averaged correlation map onto the human average surface . In order to quantify how well a tract is predicted , we computed Dice coefficients of similarity ( Dice , 1945 ) , which quantifies the amount of overlap of the tract maps . The metric was derived for each pair of 20 human subjects and five subjects of the other species . The Dice coefficient was computed for the binarized and thresholded actual human tract map and the map predicted by the other species . The threshold was chosen for each tract individually so that 40% of surface vertices were covered by the human tract map . The same threshold was applied to the macaque and chimpanzee map . As a quantification of tract extension , we computed the ratio of the number of vertices covered by the thresholded human tract map and the number of vertices covered by both the human and the other tract map . To confirm that the pattern of values is robust , both Dice coefficients and tract extension ratios were computed for a range of percentages of surface coverage and data for a coverage of 20% , 30% and 50% is provided in Appendix 3—figure 2 . The differences in tract extension ratios at a surface coverage of 40% were assessed in a non-parametric permutation test implemented in PALM ( Winkler et al . , 2014 ) using 5000 permutations . We constructed a mixed-effects model matrix using R software ( R Development Core Team , 2015 Core p-values were corrected for family-wise error over multiple contrasts . We performed additional control analyses to assess if the observed effects of tract expansion correlate with potential sources of confounds arising from our connectivity measures and the myelin-driven registration . The methodology and results of these control analyses are reported in Appendix 4 . We characterized the effect of cortical expansion on brain connectivity using the concept of connectivity fingerprints ( Passingham et al . , 2002 ) . In brain areas where cortical expansion can explain the human connectivity pattern , actual and predicted tract maps will have similar intensity values . In brain areas where the connectivity profile was further modified due to tract extensions , the intensity values of actual and predicted tract maps will show a discrepancy . By computing the intensity values of multiple tract maps in a brain area , we can derive a characteristic profile of values that can be understood as connectivity fingerprint of this area . In brain areas , where tract extension happened in addition to cortical expansion , we expect to observe a difference between actual human connectivity profile and the predicted connectivity profile . We manually selected two representative vertices and derived their actual and predicted connectivity profile: One vertex in the inferior parietal lobe , where we expect the intensity values of actual and predicted tract maps to be similar and one in the middle temporal gyrus , where we expect to find differences in actual and predicted tract maps . The whole set of tracts investigated ( CST , MDLF , VOF , IFO , ILF , SLF3 and AF ) was included to give a more detailed estimate of the connectivity fingerprint . Availability of software used in the present study is provided in the Key Resources Table . Processing code is openly available from the Wellcome Centre for Integrative Neuroimaging’s GitLab at https://git . fmrib . ox . ac . uk/neichert/project_MSM ( Eichert , 2020; copy archived at https://github . com/elifesciences-publications/project_msm ) . Data setReference for original data paperAvailabilityHuman in-vivo diffusion MRI data and myelin mapspresent studyAnonymised raw data is openly available for download via OpenNeuro . Accession code: ds002634 ( version 1 . 0 . 1 ) , project_larynx ( https://openneuro . org/datasets/ds002634 ) Chimpanzee in-vivo diffusion MRI data ( Chen et al . , 2013 ) Available from the National Chimpanzee Brain Resource ( http://www . chimpanzeebrain . org/ ) . Data from the following subjects were used: Bo , Cheetah , Lulu , Wenka , Foxy . Chimpanzee in-vivo myelin maps ( Glasser et al . , 2014 ) Raw data available from the National Chimpanzee Brain Resource ( http://www . chimpanzeebrain . org/ ) . Data from all 29 subjects were used . Macaque ex-vivo diffusion MRI data ( 4 macaques ) ( Folloni et al . , 2019 ) Source data available from the PRIMatE Data Exchange ( PRIME-DE ) resource ( http://fcon_1000 . projects . nitrc . org/indi/indiPRIME . html4 ) . Dataset: University of Oxford WIN Macaque PMMacaque in-vivo myelin maps ( Bridge et al . , 2019; Large et al . , 2016 ) Data of four monkeys freely available at: https://gin . g-node . org/hbridge_oxford/brainwithoutv1 . Data of the fifth monkey available upon request .	How did language evolve ? Since the human lineage diverged from that of the other great apes millions of years ago , changes in the brain have given rise to behaviors that are unique to humans , such as language . Some of these changes involved alterations in the size and relative positions of brain areas , while others required changes in the connections between those regions . But did these changes occur independently , or can the changes observed in one actually explain the changes we see in the other ? One way to answer this question is to use neuroimaging to compare the brains of related species , using different techniques to examine different aspects of brain structure . Imaging a fatty substance called myelin , for example , can produce maps showing the size and position of brain areas . Measuring how easily water molecules diffuse through brain tissue , by contrast , provides information about connections between areas . Eichert et al . performed both types of imaging in macaques and healthy human volunteers , and compared the results to existing data from chimpanzees . Computer simulations were used to manipulate the myelin-based images so that equivalent brain areas in each species occupied the same positions . In most cases , the distortions – or 'warping' – needed to superimpose brain regions on top of one another also predicted the differences between species in the connections between those regions . This suggests that movement of brain regions over the course of evolution explain the differences previously observed in brain connectivity . But there was one notable exception , namely a bundle of fibers with a key role in language called the arcuate fasciculus . This structure follows a slightly different route through the brain in humans compared to chimpanzees and macaques . Eichert et al . show that this difference cannot be explained solely by changes in the positions of brain regions . Instead , the arcuate fasciculus underwent additional changes in its course , which may have contributed to the evolution of language . The framework developed by Eichert et al . can be used to study evolution in many different species . Interspecies comparisons can provide clues to how brain structure and activity relate to each other and to behavior , and this knowledge could ultimately help to understand and treat brain disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2020	Cross-species cortical alignment identifies different types of anatomical reorganization in the primate temporal lobe
In cancer cells , loss of G1/S control is often accompanied by p53 pathway inactivation , the latter usually rationalized as a necessity for suppressing cell cycle arrest and apoptosis . However , we found an unanticipated effect of p53 loss in mouse and human G1-checkpoint-deficient cells: reduction of DNA damage . We show that abrogation of the G1/S-checkpoint allowed cells to enter S-phase under growth-restricting conditions at the expense of severe replication stress manifesting as decelerated DNA replication , reduced origin firing and accumulation of DNA double-strand breaks . In this system , loss of p53 allowed mitogen-independent proliferation , not by suppressing apoptosis , but rather by restoring origin firing and reducing DNA breakage . Loss of G1/S control also caused DNA damage and activation of p53 in an in vivo retinoblastoma model . Moreover , in a teratoma model , loss of p53 reduced DNA breakage . Thus , loss of p53 may promote growth of incipient cancer cells by reducing replication-stress-induced DNA damage . To prevent cells become cancerous , different cell-cycle checkpoints can be activated to halt cell cycle progression . The G1/S checkpoint is responsible for controlling S phase entry and key effectors of this checkpoint are the retinoblastoma ( Rb ) proteins pRB , p107 and p130 . Anti-proliferative conditions , such as lack of growth factors , suppress the activity of the D-type cyclin-dependent kinases ( CDKs ) CDK4 and CDK6 . This results in hypo-phosphorylation of the Rb proteins , which can then bind E2F transcription factors thereby inhibiting the transcription of E2F target genes required for S-phase entry ( Bertoli et al . , 2013; Burkhart and Sage , 2008 ) . In a majority of human tumors , the G1/S checkpoint is lost , for example by loss of pRB or the CDK inhibitor p16INK4A , or by overexpression of Cyclin D1 ( Ho and Dowdy , 2002; Weinberg , 2007 ) and insensitivity to antigrowth signals is an hallmark of tumor cells ( Hanahan and Weinberg , 2000 ) . Cells lacking the G1/S phase checkpoint can start synthesizing DNA under non-permissive conditions which may lead to DNA damage . To deal with DNA damage , cells have evolved another cell cycle checkpoint that is part of the DNA-damage response ( DDR ) ( Jackson and Bartek , 2009 ) . Activation of the DNA damage checkpoint triggers cellular senescence or cell death , thereby providing an intrinsic biological barrier against tumor progression ( Bartkova et al . , 2006; Gorgoulis et al . , 2005 ) . It is often rationalized that inactivation of Trp53 , a central player in the DDR and the most frequently mutated gene in human cancer ( Olivier et al . , 2010 ) , promotes tumorigenesis by counteracting apoptosis and senescence induced by a defective G1/S checkpoint ( Sherr and McCormick , 2002; Reinhardt and Schumacher , 2012; Polager and Ginsberg , 2009; Bunz et al . , 1998; Bieging et al . , 2014 ) . However , here we present an unanticipated effect of p53 loss in cells that lack G1/S control . To study the consequences of G1/S checkpoint loss in a well-defined system , we used primary mouse embryonic fibroblasts ( MEFs ) in which the three retinoblastoma ( Rb ) genes were inactivated . Previously , we and others demonstrated that these so-called triple knockout ( TKO ) MEFs can enter S phase without mitogenic signaling ( Dannenberg et al . , 2000; Sage et al . , 2000 ) . However , proliferation of TKO MEFs was still mitogen dependent: without mitogens , most cells became apoptotic whereas surviving cells arrested in a G2-like state . Suppression of apoptosis by ectopic expression of Bcl2 ( TKO-Bcl2 MEFs ) revealed that G2 arrest resulted from induction of p27Kip1 and p21Cip1 that inhibit Cyclin A- and B1-dependent kinase activity ( Foijer et al . , 2005 ) . Induction of p21Cip1 upon mitogen deprivation may be indicative for DNA damage ( Karimian et al . , 2016 ) . Intriguingly , we previously showed that RNAi-mediated suppression of p53 and thereby reduction of p21Cip1 levels revitalized CDK activity and supported mitogen-independent proliferation of Rb-protein-deficient cells ( Foijer et al . , 2005 ) . In the present study , we provide mechanistic insight into the relief of proliferative arrest in mitogen-deprived TKO cells by p53 loss . We show that the DNA DSBs observed in mouse and human cells lacking G1/S phase control are caused by replication stress reflected by decreased replication speed and reduced origin firing . Inactivation of p53 allowed for mitogen-independent proliferation , not only by suppressing apoptosis but also by restoring the levels of origin firing and reducing DSB formation . Similarly , in an in vivo model and in Rb-protein-deficient human cells , DNA breakage was reduced by loss of p53 . Consistent with our previous observations ( Foijer et al . , 2005 ) , mouse embryonic fibroblast ( MEFs ) lacking the three retinoblastoma proteins and overexpressing the anti-apoptotic gene Bcl2 ( TKO-Bcl2 MEFs ) ceased proliferation upon mitogen deprivation ( Figure 1A , black line ) and arrested in a G2-like state ( Figure 1C , upper panel ) . We also reported that proliferation was rescued by RNAi-mediated knockdown of Trp53 , the gene that encodes the p53 protein ( TKO-p53RNAi MEFs ) ( Foijer et al . , 2005 ) . However , in recent experiments , proliferation of mitogen-starved TKO-p53RNAi MEFs appeared transient and was followed by severe cell loss ( Figure 1A , green line ) , possibly as a result of residual p53 activity ( Figure 1—figure supplement 1A ) . We therefore exploited CRISPR/Cas9 technology to create full Trp53 knockout ( KO ) TKO MEFs ( Figure 1—figure supplement 1A ) . Disruption of p53 clearly rescued proliferation of mitogen-starved TKO MEFs ( TKO-p53KO ) and this effect was even greater in TKO MEFs expressing Bcl2 ( TKO-Bcl2-p53KO ) , which reached 100% confluency ( Figure 1A , blue and red lines ) . The improved proliferative capacity was accompanied by reduced apoptosis ( Figure 1B ) and the absence of G2 arrest ( Figure 1C , lower panel , Figure 1—figure supplement 1B ) . Mitogen-deprived TKO-Bcl2-p53KO cells maintained a cell cycle profile similar to cells cultured in the presence of mitogens ( Figure 1C , lower panel ) and , unlike TKO-Bcl2 cells , continued to incorporate high levels of nucleotides ( Figure 1D ) . Not only loss of p53 , but also disruption of its downstream target Cdkn1a , the gene that encodes the p21Cip1 protein ( Figure 1—figure supplement 1C ) , rescued proliferation of mitogen-deprived TKO-Bcl2 cells ( Figure 1E ) . Apparently , the induction of p21Cip1 , which we previously found to inhibit Cyclin A- and B1-dependent kinases ( Foijer et al . , 2005 ) , was critical for G2-like arrest of mitogen-deprived TKO cells . The p53/p21Cip1 axis is part of the DNA damage response ( DDR ) and its activation is consistent with the high levels of DNA double-strand breaks ( DSBs ) that accumulated in arrested TKO-Bcl2 cells ( van Harn et al . , 2010 ) . To understand how disruption of p53/p21Cip1 rescued proliferation , we investigated the mechanism of DSB formation . We studied cell cycle progression of individual cells using the Fucci system , in which fluorescent proteins fused to the degradation motifs of Cdt1 and Geminin mark G1 and S/G2 cells , respectively ( Sakaue-Sawano et al . , 2008 ) . In the presence of mitogens , TKO-Bcl2 and TKO-Bcl2-p53KO MEFs proliferated with a cell-cycle duration of 10 to 15 hr ( Figure 2A , B , left ) . In the absence of mitogens , TKO-Bcl2 MEFs arrested in S/G2 phase , either immediately or after one cell cycle ( Figure 2A , right ) . In contrast , mitogen-deprived TKO-Bcl2-p53KO MEFs could be followed for two or three cell divisions ( Figure 2B , right ) , although G1 and S/G2 phase durations were increased , together encompassing 15 to 30 hr . These tracking experiments confirm that TKO-Bcl2-p53KO MEFs can proliferate in the absence of mitogens albeit at slower pace . Cell cycle delay may be caused by DSBs that accumulate in mitogen-deprived TKO-Bcl2 MEFs ( van Harn et al . , 2010 ) . This level was comparable to irradiation with 20 Gy , which is expected to severely impair mitosis resulting in cell death ( Zachos et al . , 2003 ) . Nonetheless , TKO-Bcl2-p53KO and TKO-Bcl2-p21KO MEFs were able to proliferate mitogen-independently . We therefore investigated whether Trp53 or Cdkn1a inactivation affected DSB formation as a consequence of mitogen deprivation by performing neutral comet assays ( Olive and Banáth , 2006 ) . Mitogen restriction of TKO-Bcl2 MEFs caused a clear increase in tail moment , an indicator of the level of DSBs ( Figure 3A , B ) . In contrast , the tail moments in TKO-Bcl2-p53KO and TKO-Bcl2-p21KO MEFs were not significantly increased by mitogen depletion ( Figure 3B ) although the basal levels of DSBs ( i . e . , in the presence of mitogens ) were somewhat higher compared to TKO-Bcl2 cells . Possibly , MEFs accumulated some DNA damage under optimal culture conditions that was tolerated or not adequately repaired in the absence of p53/p21Cip1 activity ( Levine and Oren , 2009; Williams and Schumacher , 2016 ) . Nevertheless , the critical observation here is that the induction of DNA breakage due to mitogen deprivation was suppressed in the absence of p53/p21Cip1 . It is known that p53 modulates different DNA repair pathways ( Bieging et al . , 2014; Williams and Schumacher , 2016 ) . Could the level of DSBs in mitogen-deprived p53KO MEFs be reduced by passage through M phase and subsequent repair in G1 ? To examine this possibility , we blocked cell cycle progression towards G1 by culturing cells in medium without mitogens , but containing nocodazole ( Figure 3—figure supplement 1 ) . This allowed us to measure the level of DSBs in TKO-Bcl2 and TKO-Bcl2-p53KO cells in comparable cell cycle phases , between S and M phase . In the presence of nocodazole , the same results were obtained: mitogen-deprived TKO-Bcl2 MEFs showed the expected increase in tail moment , while the tail moments of TKO-Bcl2-p53KO MEFs were still not increased ( Figure 3C ) . To directly investigate whether p53 status affected repair of replication-stress-induced DSBs , we treated mitogen-stimulated TKO-Bcl2 MEFs with 2 mM hydroxyurea ( HU ) for 1 hr in order to induce and alleviate replication stress instantaneously . HU depletes the cells of nucleotides , which results in stalling and collapsing of replication forks and hence DNA breakage ( Bianchi et al . , 1986; Koç et al . , 2004 ) . When comparing cells harvested immediately after HU treatment and cells harvested 30 min after HU treatment , we observed an equally strong decrease in tail moment in TKO-Bcl2 and in TKO-Bcl2-p53KO MEFs ( Figure 3D ) . This indicates that the repair of DSBs induced by HU treatment was independent of p53 status . Assuming that repair of replication-stress-induced DSBs under mitogen-deprived conditions follows similar rules , these results suggest that reduced levels of DSBs in mitogen-deprived TKO-Bcl2-p53KO cells resulted from suppressed formation rather than increased repair of DSBs . To study the mechanism of DNA breakage , we assessed the quality of DNA replication in mitogen-deprived TKO-Bcl2 MEFs by looking at co-localization of the thymidine analogue chloro-deoxyuridine ( CldU , marking DNA replication ) and γ-H2AX ( marking DNA damage ) . While the number of cells containing CldU foci gradually decreased in mitogen-starved TKO-Bcl2 MEFs , virtually all CldU foci that were still present after 4 and 7 days co-localized with γ-H2AX foci ( Figure 4A , B ) . Furthermore , the gradual increase of phosphorylated Chk1 ( pChk1 ) , a target of ataxia telangiectasia related ( ATR ) , is indicative for accumulation of single-stranded DNA ( Figure 4C ) . Taken together , these results are indicative for perturbed replication in mitogen-deprived TKO-Bcl2 MEFs . We next visualized the progression of individual replication forks using a DNA fiber assay ( Tuduri et al . , 2010 ) . Sequential pulse-labeling of newly synthesized DNA strands with the thymidine analogs CldU ( red tracks ) and iodo-deoxyuridine ( IdU , green tracks ) identifies ongoing replication forks and new origin firing ( Figure 5A ) . The length of double-labelled tracks in TKO-Bcl2 MEFs cultured with FCS indicated an average fork speed of 1 . 66 kb/min ( Figure 5B ) . In the absence of p53 the average fork speed was somewhat lower , 1 . 37 kb/min , consistent with a previous study ( Klusmann et al . , 2016 ) . Mitogen deprivation caused a progressive decline in replication speed , somewhat unexpectedly not only in arresting TKO-Bcl2 MEFs but also in proliferating TKO-Bcl2-p53KO MEFs ( Figure 5B ) . Prolonged S-phase and decelerated DNA synthesis indicate that mitogen-deprived TKO-Bcl2-p53KO MEFs were able to proliferate despite sustained replication stress . Disruption of the nucleotide pool can contribute to replication stress ( Bester et al . , 2011; Poli et al . , 2012 ) and may therefore be the underlying cause of reduced replication speed , DSB formation and G2-like arrest in mitogen-deprived TKO-Bcl2 MEFs . Mitogen deprivation is a strong anti-proliferative signal that may inhibit MYC transcription factors and therefore repress genes involved in nucleotide synthesis ( Gassmann et al . , 1999; Perna et al . , 2012 ) . Indeed , we found that mitogen-deprivation of TKO-Bcl2 cells reduced transcript levels of phosphoribosyl pyrophosphate amidotransferase ( Ppat ) and inosine monophosphate dehydrogenase 1 and 2 ( Impdh1 and Impdh2 ) , genes involved in purine metabolism , 2-fold ( Figure 5—figure supplement 1A ) . Reduced levels of nucleotide synthesis enzymes could impair DNA replication by disturbing the balance in the dNTP pool . Indeed , RNAi-mediated suppression of Ppat expression ( Figure 5—figure supplement 1B ) reduced the replication speed in mitogen-stimulated TKO-Bcl2 MEFs from 1 . 25 kb/min to 0 . 84 ( shRNA #1 ) and 0 . 86 ( shRNA #2 ) kb/min ( Figure 5—figure supplement 1C ) . Conversely , replication speed in mitogen-deprived TKO-Bcl2 MEFs could be partially rescued by the exogenous supply of nucleosides . Similar to previous experiments , one day of mitogen deprivation decreased the average fork speed by ±30% , in this experiment from 0 . 94 kb/min to 0 . 65 kb/min . In contrast , when cells were supplemented with nucleosides , reduction of replication speed was less pronounced ( from 1 . 03 kb/min to 0 . 83 kb/min ) ( Figure 5—figure supplement 1D ) . However , daily nucleosides supplementation did not alleviate the proliferation defect of mitogen-deprived TKO-Bcl2 MEFs: G2 accumulation was hardly affected ( Figure 5—figure supplement 1E ) and also Chk1 phosphorylation and p21Cip1 induction were not reduced ( Figure 5—figure supplement 1F ) . As we observed that mitogen-independent proliferation upon p53 loss did not require restoration of replication speed ( Figure 5B ) , this indicates that rather than decelerated DNA replication another factor was causal to G2 arrest upon mitogen deprivation . Previously , we showed that inhibition of CDK activity by p27Kip1 and p21Cip1 was critical for arrest of mitogen-deprived TKO-Bcl2 MEFs ( Foijer et al . , 2005 ) . Since CDK activity is required to activate origins of replication ( Fragkos et al . , 2015; Méndez and Stillman , 2003 ) , origin firing may be perturbed in mitogen-deprived TKO-Bcl2 MEFs . Indeed , among CldU/IdU-labelled DNA fibers from mitogen-deprived TKO-Bcl2 MEFs , staining patterns indicative for new origin firing were significantly reduced ( Figure 5C ) . In contrast , in TKO-Bcl2-p53KO MEFs , origin firing was not disturbed during the first days of mitogen deprivation and maintained levels similar as in mitogen-stimulated cells ( Figure 5C ) . Only after 5 days of mitogen deprivation , origin firing was reduced , which may be related to the state of confluency that was reached by that time ( Figure 1A ) . Similar to TKO-Bcl2-p53KO MEFs , also TKO-Bcl2-p21KO MEFs maintained normal origin firing during the first days of mitogen deprivation ( Figure 5C ) . The increased level of origin firing upon loss of p53/p21Cip1 contrasts to a recent publication by Roy et al . who identified a transcription-independent function of p53 in balancing replication fork homeostasis and , in contrast to our findings , observed a decrease in the level of origin firing upon loss of p53 ( Beroukhim et al . , 2010 ) . An explanation for this seeming discrepancy may be found in comparing the different replication stress conditions . Roy et al . studied the role of p53 in conditions with a low dose of HU that did not induce DSBs . In contrast , by serum starvation we induced severe replication stress as observed by the drastic decrease of replication fork progression and induction of DNA breaks ( Figure 5B ) . We therefore compared the consequences of low versus high doses of HU . A low dose of HU ( 300 µM ) did not induce DNA DSBs whereas a high dose of HU ( 2 mM ) did ( Figure 5—figure supplement 2A ) . Consistent with Roy et al . , loss of p53 reduced the level of origin firing upon treatment with 300 µM HU . However , loss of p53 did not change the level of origin firing after treatment with the high dose of HU ( 2 mM ) ( Figure 5—figure supplement 2B ) . Collectively , our results suggest that under conditions of severe replication stress , restoration of the level of origin firing upon p53 loss prevents DNA breakage , allowing mitogen-independent proliferation of TKO-Bcl2-p53KO MEFs . To investigate whether p53 affects DNA breakage under replication stress conditions in human cells , we used the human retinal pigment epithelial cell line RPE-1 . The G1/S phase checkpoint was perturbed either by inactivating all three retinoblastoma genes , RB , RBL1 and RBL2 ( TKO; Figure 6A ) or by overexpressing a non-degradable form of human Cyclin D1 ( CyclinD1; Figure 6B ) . Overexpression of Cyclin D1 is biologically relevant since the gene encoding Cyclin D1 represents the second most frequently amplified locus in the human cancer genome ( Beroukhim et al . , 2010 ) . In addition , in many human tumors overexpression of D type cyclins takes place in the absence of detectable genomic alterations ( Hosokawa and Arnold , 1998 ) . In the presence of mitogens , TKO and CyclinD1 RPE-1 cells proliferated faster than wild type RPE-1s ( Figure 6—figure supplement 1A ) . 24 hr after mitogen-deprivation , wild type RPE-1s arrested in the G1 phase of the cell cycle ( Figure 6C ) , whereas both , TKO and CyclinD1 cultures maintained a normal cell cycle profile up to 72 hr ( Figure 6D and E , respectively ) . Upon prolonged mitogen starvation for more than 4 days , TKO and CyclinD1 cells started to die ( Figure 6—figure supplement 1B ) . Cell death could not be avoided by overexpression of Bcl2 ( Figure 6—figure supplement 1C , D and E ) nor by additional inactivation of TP53 ( Figure 6F , G and Figure 6—figure supplement 1F ) . Apparently , RPE-1 cells lacking the G1/S phase checkpoint were very sensitive to apoptosis in the absence of mitogenic stimulation , which could not easily be suppressed . Nonetheless , we could follow the behavior of these cells during the first days of mitogen deprivation . Similar to TKO-Bcl2 MEFs , p53-proficient TKO-Bcl2 RPE-1s showed induction of DNA DSBs after one day of mitogen starvation . In contrast , no DSB induction was seen in TKO-Bcl2-p53KO RPE-1s ( Figure 6H ) . Similarly , mitogen starvation hardly induced DSBs in CyclinD1-Bcl2-p53KO RPE-1s compared to CyclinD1-Bcl2 RPE-1s ( Figure 6I ) . Mitogen-deprived TKO-Bcl2 and CyclinD1-Bcl2 RPE-1s showed a decrease in the level origin firing ( Figure 6J , K ) . In contrast , TKO-Bcl2-p53KO and CyclinD1-Bcl2-p53KO RPE1-s maintained normal levels of origin firing after mitogen deprivation ( Figure 6J , K ) , although there is no difference in replication fork speed upon mitogen-deprivation ( Figure 6—figure supplement 1G and F ) . These results show that also in human cells inactivation of p53 in G1/S phase checkpoint defective cells reduced the accumulation of DNA DSBs following mitogen deprivation , possible by rescuing the level of origin firing . To investigate whether the accumulation of DNA DSBs also operates in vivo to impede tumor growth , we studied retinoblastoma development in chimeric mice generated by blastocyst injection of Rb-/-Rbl2-/-embryonic stem cells ( ESCs ) ( Dannenberg et al . , 2000 ) . Remarkably , murine retinoblastomas showed pronounced p53 and γ-H2AX staining ( Figure 7A ) . However , by sequencing , p53 appeared wild-type in a separate series of seven tumors , indicating that in this model retinoblastomas did activate the DDR but could still colonize the entire eyeball . To study if p53 inactivation accelerates tumorigenesis , we inactivated p53 in Rb-/- Rbl2-/- ESCs using CRISPR/Cas9-mediated gene disruption . However , no chimeric animals were obtained from Rb-/-Rbl2-/-p53-/- ESCs , likely indicating that combined ablation of the Rb and p53 pathways is incompatible with embryonic development . As an alternative in vivo readout , we injected Rb-/-Rbl2-/- and Rb-/-Rbl2-/-p53-/- mouse ESCs under the skin of nude mice . Rb-/-Rbl2-/- ESCs developed a teratoma in 4 out of 6 mice; in contrast , Rb-/-Rbl2-/-p53-/- ESCs developed a tumor in 6 out of 6 injected mice . On average the Rb-/-Rbl2-/-p53-/- tumors were larger than Rb-/-Rbl2-/- tumors ( Figure 7B ) , although there is a 11% chance that the difference is accidental ( p=0 . 1116 , unpaired t-test ) . Teratomas of both genotypes mainly showed early neuronal differentiation and stained positive for the replication stress marker γ-H2AX ( Figure 7C ) , suggesting that all tumors were suffering from replication stress . To assess the presence of DSBs , we performed a neutral comet assay on teratoma tissues . Three of the four Rb-/-Rbl2-/- teratomas showed an increase in tail moment compared to the tail moments of Rb-/-Rbl2-/-p53-/- teratomas ( Figure 7D ) . Of note , unlike the other tumors , the largest Rb-/-Rbl2-/- tumor ( marked with asterisk in Figure 7B , D ) had high levels of infiltrating neutrophils , which possibly explains its bigger size as well as the low level of DNA DSBs . Although the number of tumors was small , p53 knockout teratomas showed a trend towards lower levels of DSBs and accelerated tumor growth . Therefore , both our in vitro as well as in vivo data suggest that inactivation of p53 in G1/S checkpoint deficient cells contributes to tumorigenesis by reducing DNA DSBs ( Figure 7E ) . We have previously shown that apoptosis-resistant MEFs that lack the G1/S phase checkpoint ( TKO-Bcl2 MEFs ) can undergo unscheduled S-phase entry . Here we show that they do so at the expense of severe replication stress and the accumulation of DNA DSBs , which ultimately causes G2-like cell cycle arrest . Inactivation of p53 allowed mitogen-independent proliferation , which remarkably was not only associated with alleviated G2 arrest but also with reduced DNA breakage and restored origin firing . The firing of origins requires Cyclin-CDK activity ( Fragkos et al . , 2015; Méndez and Stillman , 2003 ) . In mitogen-deprived TKO-Bcl2 MEFs , CDK activity was low due to the high levels of p21Cip1 and p27Kip1 explaining the low level of origin firing . It is therefore likely that upon p53 inactivation and therefore reduction of p21Cip1 , CDK activity increased ( Foijer et al . , 2005 ) and hence promoted origin firing . Consistently , we found that also genetic inactivation of p21Cip1 restored origin firing and promoted proliferation of mitogen-deprived TKO-Bcl2 MEFs . Importantly , this phenomenon was not restricted to murine cells , but also observed in human RPE-1 cells: mitogen deprivation restricted origin firing and induced DNA breakage in G1/S-checkpoint-defective RPE-1s , and both could be reverted by inactivation of p53 . However , for as yet unknown reasons , RPE-1 cells appeared highly sensitive to apoptosis and therefore the damage-reducing effect of p53 loss did not translate into mitogen-independent proliferation . While restored origin firing upon ablation of the p53/p21Cip1 axis is mechanistically plausible , we have not directly proven that DNA breakage as a consequence of replication stress was prevented by increased origin firing . Related to this , an important question is whether p53 inactivation reduced the formation of DNA breaks or stimulated DSB repair . Apart from its role as transcription factor , p53 has many transcription-independent functions , among which inhibition of DNA DSB repair by both non-homologous end joining ( NHEJ ) and homologous recombination ( HR ) ( Menon and Povirk , 2014; Sengupta and Harris , 2005; Akyüz et al . , 2002; Dudenhöffer et al . , 1998 ) . However , we show that KO of the p53 transcription target p21Cip1 phenocopied the effects of KO of p53 in TKO-Bcl2 MEFs , arguing against a transcription-independent role of p53 in suppressing DNA repair . Furthermore , increased DNA repair by NHEJ in G1 phase is unlikely as the levels of DSBs in serum-starved TKO-Bcl2-p53KO MEFs remained low when G1 entry was prevented by artificially arresting cells in M-phase . In this experiment it remains possible that an increase in HR activity during S/G2 phase contributed to less DNA breaks in mitogen-deprived p53KO MEFs . Also this possibility seems unlikely as the repair of HU-induced DSBs was not affected by p53 status . However , this experiment does not exclude the possibility that under mitogen-deprived conditions p53 suppressed DSB repair . With this restriction , we hypothesize that abrogation of p53 reduced the formation of DNA breaks , rather than facilitated repair . Novel roles of pRB and p53 are emerging but it is unclear to which extent they are implicated in suppression of cancer . Apart from its well documented role in cell cycle control , pRB has emerged as a multi-functional protein involved in a wide range of biological processes including chromatin architecture , cohesion , chromosome condensation during mitosis , DNA replication via interaction with replication components and involvement in DNA repair processes such as HR and NHEJ ( Vélez-Cruz and Johnson , 2017; Dick and Rubin , 2013; Huang et al . , 2015 ) . We suggest that these other functions of pRB do not play a role in the accumulation of DNA damage in Rb-deficient cells since we observed the same phenotype in Rb-proficient Cyclin D1 overexpressing RPE-1s . Thus , the accumulation of DNA DSBs in mitogen-deprived conditions can be attributed to loss of the G1/S phase checkpoint and not to other functions of the pRB protein or its family members . It has been described that some ribosomal proteins have a function in the DNA damage response that is activated upon intrinsic replication stress and mediated through the Mdm2-p53 axis ( Xu et al . , 2016 ) . In addition , some ribosomal proteins act as a sensor for DNA damage and directly participate in the process of DNA repair . In this study , we cannot exclude an effect of p53 loss on the extra-ribosomal functions of these proteins . Also p53 has functions outside its canonical role in the DDR . Recently , a novel transcription-independent role for p53 in balancing replication homeostasis was reported . The p53 protein can bind to replication forks and facilitate replication fork restart in replication stress conditions ( Roy et al . , 2018 ) . Since we found a transcription-dependent role of p53 in suppressing origin firing , we hypothesize that the two different effects of p53 loss reflect different functions of p53 that operate side by side: dependent on the severity of replication stress , p53 facilitates replication fork restart and suppresses the firing of new origins . Others have shown that disruption of the nucleotide pool can contribute to replication stress , DNA breakage and cell death in oncogene-expressing cells ( Bester et al . , 2011; Poli et al . , 2012; Beck et al . , 2012; Pfister et al . , 2015 ) . We were able to partially rescue replication speed in mitogen-deprived TKO-Bcl2 MEFs by exogenous supply of nucleosides . However , increased replication speed was not sufficient to overcome G2 arrest and to support normal cell cycle progression . Furthermore , we found that despite the capacity of TKO-Bcl2-p53KO MEFs to proliferate mitogen-independently , replication speed was still reduced . These observations indicate that another factor rather than the speed of DNA synthesis was critical for DNA breakage and G2 arrest in mitogen-deprived TKO-Bcl2 cells . As only origin firing but not replication speed was affected by p53 status , we favor a scenario where restoration of origin firing upon inactivation of p53 , given the involvement of p21Cip1 likely as a result of restored CDK activity , suppressed DNA breakage and allowed mitogen-independent proliferation . Loss of the Rb and p53 pathways frequently occur and co-occur in human tumors ( Polager and Ginsberg , 2009 ) . The p53 gene is mutated in more than 50% of human cancer , and mutations in other genes that affect p53 function occur in many , if not all , tumors that retain a normal p53 gene ( Perri et al . , 2016 ) . In addition , most human tumors lack the G1/S phase checkpoint . For example many human tumors overexpress D-type cyclins and CCND1 represents the second most frequently amplified locus in the human cancer genome ( Hosokawa and Arnold , 1998; Menon and Povirk , 2014 ) . Furthermore , evading apoptosis is one of the hallmarks of cancer and the anti-apoptotic gene Bcl2 , which is used in this study to suppress apoptosis , is commonly overexpressed in many types of cancer , including renal , prostate , gastric , lung and colorectal cancer , neuroblastoma , non-Hodgkin’s lymphoma and acute and chronic leukemia ( Frenzel et al . , 2009; Kirkin et al . , 2004 ) . Thus , most tumor cells harbor the type of mutations used in this study . Whereas we can only speculate about the precise number of cancer types that harbor the exact combination of Rb , p53 and Bcl2 aberrations as used in this study , there are examples known . For example , approximately 90% of small cell lung cancer tumors have lost both p53 and Rb ( Sekido et al . , 2003 ) . Beside this , small cell lung tumors are also characterized by expression of Bcl2 ( Kaiser et al . , 1996 ) . Furthermore , human retinoblastoma originates from an intrinsic death-resistant precursor cell ( Xu et al . , 2009 ) , is characterized by mutations in the Rb gene and it is suggested that the p53 pathway is inactivated ( Xu et al . , 2009; Laurie et al . , 2006 ) . Although p53 mutations are infrequent in human retinoblastomas , the p53 pathway may be intrinsically attenuated upon RB1 loss by miR-24-mediated downregulation of p14ARF ( To et al . , 2012 ) and by NANOS-mediated suppression of p53-activating kinases ( Miles et al . , 2014 ) . Other studies suggested that RB1-deficient retinal cells achieve attenuation of the p53 pathway by high expression of MDM2 and MDMX ( Xu et al . , 2009; Laurie et al . , 2006 ) , although a recent paper revealed critical p53-independent functions of high MDM2 expression ( Qi and Cobrinik , 2017 ) . In our chimeric mouse model of retinoblastoma , we found evidence for DNA damage , but loss of p53 was not a requirement for development of eye-filling tumors . Unfortunately , we could not study the effect of p53 loss , but using a hereditary retinoblastoma model , others reported a dramatic effect of p53 inactivation . When Rb was conditionally inactivated in retinal progenitor cells in a p107-/- background , non-invasive retinoblastomas developed with a penetrance of 60% . Upon additional inactivation of p53 , aggressive , invasive bilateral retinoblastomas developed with 100% penetrance and reduced latency ( Dyer et al . , 2005; Zhang et al . , 2004 ) . Importantly , evidence has been obtained that murine retinoblastomas originate from an intrinsically death-resistant cell of origin ( Chen et al . , 2004 ) . We therefore propose that the tumor promoting effect of attenuated p53 activity was not due to abrogation of an apoptotic response but rather required for maintaining sufficient CDK activity to counteract the deleterious effects of replication stress . We obtained support for such tumor-promoting effect of p53 ablation in an Rb-/-Rbl2-/- teratoma model: tumor size was inversely correlated with the level of DNA breaks and Rb-/-Rbl2-/-p53-/- teratomas generally showed lower levels of DNA breaks than Rb-/-Rbl2-/- teratomas . Finally , our results are likely related to intriguing observations that at least for some tumor types the outgrowth of early cancerous lesions is prohibited by activation of the DDR ( Bartkova et al . , 2006 ) . It has been suggested that oncogene activation can directly cause replication stress by hyper-stimulating DNA replication , which activates the ATR-Chk1 axis ( Hills and Diffley , 2014; Di Micco et al . , 2006; Halazonetis et al . , 2008 ) . Furthermore , frequent DNA breakage associated with replication stress activates the complementary ATM-Chk2-p53 module that provides a strong barrier to proliferation by inducing apoptosis or permanent cell cycle arrest . It has therefore been suggested that activation of the DDR may explain the strong selective pressure for loss of p53 in human cancer ( Halazonetis et al . , 2008 ) . Rather than a direct consequence of oncogene activation , replication stress in our system was the consequence of the combination of Rb-protein deficiency ( hyper-activating E2F transcription factors ) and growth-restricting conditions ( the absence of mitogenic signaling ) , leading to DNA breakage and activation of the DDR . Furthermore , we found loss of p53 not only abrogated cell cycle arrest and apoptosis , but also suppressed the induction of DNA damage itself , providing a novel mechanistic explanation for the frequent co-occurrence of p53 and pRb pathway inactivation in cancer . MEFs were isolated from chimeric embryos as previously described ( Dannenberg et al . , 2000 ) and cultured in GMEM ( Invitrogen ) , supplemented with 10% fetal calf serum ( FCS ) , 0 . 1 mM nonessential amino acids ( Invitrogen ) , 1 mM sodium pyruvate ( Invitrogen ) , 100 µg/ml penicillin , 100 µg/ml streptomycin ( Invitrogen ) and 0 . 1 mM β-mercaptoeethanol ( Merck ) in the absence or presence of nucleoside ( 200 nM of Cytidine , Guanosine , Adenosine and Thymidine ) . TKO-Bcl2 overexpressing MEFs and TKO-p53RNAi were generated as described previously ( Foijer et al . , 2005 ) . CRISPR/Cas9 technology was used to inactive Trp53 and Cdkn1a . RPE-1 cells were kindly provided by J . Raaijmakers , who purchased the cells from ATCC . RPE-1 cells were cultured in DMEM/F12+GlutaMAX ( Invitrogen ) , supplemented with 10% FCS , 100 µg/ml penicillin and 100 µg/ml streptomycin ( Invitrogen ) . CRISPR/Cas9 technology was used to inactivate Rb , Rbl1 , Rbl2 and TP53 . Bcl2 cDNA and a non-degradable form of CylcinD1 cDNA was overexpressed using retroviral transfection . For serum starvation experiments , cells were trypsinized and allowed to attach in the presence of serum for 4 hr . Subsequently , cells were washed with PBS and supplemented with serum free medium . To block progression into mitosis , cells were cultured in the presence of 250 ng/ml nocodazole . All cell lines have been tested for mycoplasma ( PCR ) . The FUCCI constructs CSII-EF-MCS-mKO-hCdt1 ( 30/120 ) and CSII-EF-MCS-mAG-hGem ( 1/110 ) were kindly provided by A . Miyawaki ( Sakaue-Sawano et al . , 2008 ) . The 19-mer Trp53 targeting sequence in pRetroSuper-RNAi-p53 is GTACATGTGTAATAGCTCC ( Foijer et al . , 2005 ) . Gene specific guideRNAs ( mouse Trp53: TACCTCTCTTTGCGCTCCCT ( Platt et al . , 2014 ) ; human RB TGAACGACATCTCATCT , human RBL1 TTTCGTGAACGTATAGAA , human RBL2 CGAGGTTGCTCCTCTTGA and human TP53 GACGCTAGGATCTGACTG ) were annealed to generate short double-strand DNA fragments with four base pairs overhang ( CACC and AAAC ) compatible with ligation into the BbsI digested Cas9/CRISPR px330-puro plasmid . The px330-p53 guideRNA vector was transfected into MEFs using Polyethylenimine ( PEI ) transfections . The px330-Rb- , px330-Rbl1- , px330-Rbl2 and px330-p53 guideRNA vectors were transfected into RPE-1 cells using Lipofectamine 2000 ( Invitrogen ) . Afterwards , RPE-1 cells were selected with 10 ug/ml puromycin for two days . Two specific guideRNAs targeting the mouse Cdkn1a gene were AGCGCAGATTGGTCTTCT and CCCGCAGCCGTGACGACC with four base pairs overhang ( CACC and AAAC ) compatible with ligation into the in BmsbI digested pLentiCRISPR v2 vector . The 21-mer oligos in pLKO . 1 targeting Ppat were: #1: CCACATGCTTATGTATGTATA and #2: CCGGAGAAATTGTAGAAATAT . Corresponding empty vector ( EV ) was used as control . Lentiviral plasmids were co-transfected with the helper plasmids pMDLgpRRE , VSV-G and pRSV-Rec into HEK293T cells by PEI transfection . A pBABE-puro retroviral vector encoding a non-degradable form of Cyclin D1 ( T286A ) was kindly provided by R . Agami ( Agami and Bernards , 2000 ) . This retroviral vector was co-transfected with the helper plasmids puMCV-Gag pol MMLV and pCMV VSVG into HEK293T cells by PEI transfection . Both for lentiviral and retroviral transfections , forty-eight and sixty-two hours post transfection viral supernatants were filtered through 0 . 45 µm filter and used to infect MEFs in the presence of 4 µg/ml polybrene three times for 8–12 hr . The IncuCyte ZOOM instrument ( Essen Bioscience ) live cell imaging system was used to monitor cell growth . Cells were plated in a 96 Greiner micro clear plate and imaged every 4 hr . The default software parameters for a 96 well plate with a 10x objective were used for imaging . The IncuCyte software was used to calculate mean confluence from two non-overlapping bright phase images of each well . The IncuCyte ZOOM instrument in combination with the Cell Player 96-well kinetic caspase-3/7 reagent ( Essen Bioscience ) were used to identify apoptosis by caspase 3/7 activity . The software was used to calculate mean green fluorescence from two non-overlapping fluorescent images of each well . Green fluorescent confluency was normalized to phase contrast confluency to determine apoptosis . Cells were harvested and subsequently lysed for 30 min in RIPA ( 25 mM Tris-HCl pH 7 . 6; 150 mM NaCl; 1% NP-40; 1% Sodiumdeoxycholate and 0 . 1% SDS ) or ELB ( 150 mM NaCl; 50 mM Hepes pH7 . 5; 5 mM EDTA; 0 . 1% NP-40 ) containing protease inhibitors ( Complete , Roche ) . Protein concentrations were measured using the BCA protein assay kit ( Pierce ) . The primary antibodies used were rabbit polyclonal phospho-Chk1 Ser317 ( Bethyl ) , mouse monoclonal Chk1 ( G4; Santa Cruz ) , goat polyclonal CDK4 ( C22; Santa Cruz ) , rabbit polyclonal p21 ( C19; Santa Cruz ) , mouse monoclonal p27 ( BD Transduction Laboratory ) , goat polyclonal Rb ( C15; Santa Cruz ) ; rabbit polyclonal Rbl1 ( C18; Santa Cruz ) , mouse monoclonal Rbl2 ( CAS14; Lab Vision ) , mouse monoclonal p53 ( IMX25; monosan; for detection of mouse p53 ) , mouse monoclonal p53 ( DO-1; BD Biosciences; for detection of human p53 ) , γ-tubulin ( GTU-88; Sigma ) , rabbit polyclonal Cyclin D1 ( Santa Cruz; H296 ) and goat polyclonal CDK4 ( C22; Santa Cruz ) . Secondary antibodies used were IR Dye 800CW Goat anti-Mouse IgG , Goat anti-Rabbit IgG and Donkey anti-Goat IgG ( Licor ) and HRP-conjugated Goat anti-Mouse and Goat anti-Rabbit ( Dako ) . For Rad51 and γ-H2AX immunofluorescence staining , cells were cultured on cover slides , washed with PBS and fixed for 5 min using 4% paraformaldehyde ( Merck ) . Cells were permeabilized by 0 . 1% Triton-X100 ( sigma ) in PBS for 5 min . Subsequently , cells were washed three times using staining buffer ( 0 . 15% glycine ( Merck ) , 0 . 5% Bovine Serum Albumine ( BSA , Sigma ) in PBS ) and incubated for 1 hr at room temperature in staining buffer . Cells were incubated for 4 hr and 1 hr with primary and secondary antibodies , respectively . For CldU and γ-H2AX immunofluorescence , cells were cultured on cover slides , incubated with CldU ( 100 mM ) for 30 min , washed with PBS and fixed for 10 min using 70% EtOH . Cells were treated with MeOH for 5 min and incubated with 1 . 5 M HCl for 20 min . Subsequently , cells were blocked using PBS , 0 . 5% Tween , 0 . 25% BSA , 5% FCS for 30 min . Cells were incubated with primary and secondary antibodies for 2 hr and 1 hr , respectively in PBS , 0 . 5% Tween , 0 . 25 BSA . Bleaching was prevented by Vectashield ( Vetcor laboratories ) . The primary antibodies used were rat-anti-BrdU ( Clone BU1/75 , Novus Biologicals ) , rabbit polyclonal Rad51 ( a gift from Prof . Roland Kanaar ) and mouse monoclonal phosphorylated H2AX ( Upstate ) in 1:20 , 1:2500 and 1:100 dilutions , respectively . Secondary antibodies used were Alexa 488-labeled Chicken-anti-Mouse , Alexa 568-labeled Goat-anti-Rabbit and Alexa 568-labeled Goat-anti-Rat antibodies ( Molecular probes ) and these were used in a 1:100 dilution . DNA was stained using To-Pro3 dye ( Molecular probes ) . Cells were pulse-labelled with 25 μM CldU followed by 250 μM IdU for 20–40 min each . After labelling , cells were trypsinized and lysed in spreading buffer ( 200 mM Tris-Hcl pH 7 . 4 , 50 mM EDTA and 0 . 5% SDS ) before spreading on a microscope slide ( Menzel-Gläser , Superfrost ) . Slides were fixed in methanol: acidic acid 3:1 . Before immunodetection , slides were treated with 2 . 5 M HCl for 1 hr and 15 min . To detect CldU and IdU labelled tracts slide were incubated for 1 hr with rat-anti-Brdu ( Clone BU1/75 , Novus Biologicals; 1:500 ) and mouse-anti-BrdU ( clone B44 , Becton Diskinson; 1:750 ) , respectively . Subsequently , slides were fixed with 4% paraformaldehyde for 10 min and incubated with Alexa 488-labeled goat-anti-mouse and Alexa 555-labeled goat-ant-rat ( Molecular probes; 1:500 ) for 1 hr and 30 min . Pictures were taken with a Zeiss AxioObserver Z1 inverted microscope using a 63x lens equipped with a cooled Hamamatsu ORCA AG Black and White CCD camera and track lengths were analyzed with ImageJ software . Replication track lengths were calculated using the conversion factor 1 μm = 2 . 59 kb ( Jackson and Pombo , 1998 ) . The 1-way ANOVA ( nonparametric Kruskal-Wallis test ) was used for statistical analyses . Culture dishes were transferred to a heated stage ( 37°C ) on a Zeiss Axiovert 200M inverted microscope . PhC ( phase contrast ) images ( 59 ms exposure ) and fluorescent images ( red: 500 ms and green 300 ms exposure ) were captured with a 20x/0 . 25 Ph1 Achroplan objective in combination with 1 . 6 optovar every 30 min using a cooled Hamamatsu ORCA R2 Black and White CCD-camera and appropriate filter blocks to select specific fluorescence . Images were taken in 2 × 2 binning mode ( 672 × 512 pixels ) and processed using AxioVision Rel . 4 . 7 . 2 . software . MEFs cultured in the presence or absence of 10% FCS were labeled with BrdU ( 10 μM ) for 1 hr , fixed in 70% EtOH and stained with Propidium Iodide ( PI ) . Data acquisition was performed on a Beckman Coulter Cyan ADP and data analysis ( cell cycle ) was performed using FlowJo software version 7 . 6 . 1 ( Tree Star , Ashland , OR , USA ) . Neutral comet assays were performed as described by Olive et al . ( Olive and Banáth , 2006 ) . Pictures of individual cells were taken with a Zeiss AxioObserver Z1 inverted microscope equipped with a cooled Hamamatsu ORCA AG Black and White CCD camera and analyzed with CASP software ( http://www . casp . of . pl ) . The p-value was determined using 1-way ANOVA ( nonparametric Kruskal-Wallis test ) . All experiments involving animals comply with local and international regulations and ethical guidelines ( protocol 12026 ) and have been authorized by our local experimental animal committee at the Netherlands Cancer Institute ( DEC-NKI ) . Rb-/-Rbl2-/- ESCs were generated previously ( Dannenberg et al . , 2004 ) . These cells were injected into C57Bl/6 blastocysts ( 6 cells per blastocyst ) to generate chimeric mice , which were monitored weekly for retinoblastoma development . Rb-/-Rbl2-/-p53-/- ESCs were generated using CRISPR/Cas9 technology . One million cells of both cell lines were injected into the flank of Balb/c nude mice and tumors were harvested 20 days later . Eyes and teratomas were removed immediately after euthanasia and fixed in 4% formaldehyde for at least 24 hr . For histological analysis , formaldehyde fixed tissues were embedded in paraffin , cut into 5 μm sections and stained with Hematoxilin and Eosin . The antibodies used were α-p53 ( VectorLabs ) , α-p-ATM ( Cell signaling ) , α−γ−H2AX ( Cell signaling ) , α-p-CHK2 ( Cell signaling ) and α-p-ATM ( Genetex ) .	Healthy cells go through a strictly regulated process called the cell cycle in order to divide . During this cycle the cell’s DNA is duplicated and the two copies are equally distributed between the two newly formed cells . Duplicating DNA is a complex procedure that can go wrong and damage the DNA . This damage , in turn , can cause cells to stop growing or even die . Normal cells only start replicating their DNA when there are substances known as growth factors in the environment . Without growth factors cells remain in the first phase of the cell cycle , known as G1 . Most cancer cells , however , lack this ‘G1 checkpoint’ and enter the cell cycle even when growth factors are absent . This leads to DNA replication problems and damage that should cause the cells to die . Yet a characteristic of cancer cells is that they overcome these problems to grow and divide uncontrollably . Cancer cells also often lack a protein called p53 . Previous studies demonstrated that the lack of p53 helps tumor cells to survive by maintaining cell growth and reducing the likelihood of cell death . By growing cells in culture without growth factors , Benedict , van Harn et al . now show that p53 also helps cells that lack the G1 checkpoint to continue dividing . In the experiments , cells that lacked the G1 checkpoint but still contained the p53 protein suffered from DNA replication problems and DNA damage , and subsequently died . Deleting p53 from these cells stimulated DNA replication , stopped cells from dying and helped to prevent the DNA from getting damaged . Cells could thus grow and proliferate under unfavorable conditions . Benedict , van Harn et al . also deleted p53 in tumor cells growing under the skin of mice and observed less DNA damage in these cells than in tumor cells that still have p53 . Despite reduced levels of DNA damage , the cells still had severe DNA replication problems . It is possible that these cells rely on mechanisms that allow just enough DNA replication to occur to support their proliferation . Cancer cells may therefore be highly vulnerable to drugs that interfere with these mechanisms , since they are already using them as a last resort . Future experiments will be needed to identify these mechanisms .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2018	Loss of p53 suppresses replication-stress-induced DNA breakage in G1/S checkpoint deficient cells
Serine recombinases are often tightly controlled by elaborate , topologically-defined , nucleoprotein complexes . Hin is a member of the DNA invertase subclass of serine recombinases that are regulated by a remote recombinational enhancer element containing two binding sites for the protein Fis . Two Hin dimers bound to specific recombination sites associate with the Fis-bound enhancer by DNA looping where they are remodeled into a synaptic tetramer competent for DNA chemistry and exchange . Here we show that the flexible beta-hairpin arms of the Fis dimers contact the DNA binding domain of one subunit of each Hin dimer . These contacts sandwich the Hin dimers to promote remodeling into the tetramer . A basic region on the Hin catalytic domain then contacts enhancer DNA to complete assembly of the active Hin tetramer . Our results reveal how the enhancer generates the recombination complex that specifies DNA inversion and regulates DNA exchange by the subunit rotation mechanism . Site-specific recombination reactions have evolved as a relatively simple solution to a myriad of biological problems including gene regulation , viral integration , DNA transposition , chromosome segregation , and the programmed creation of genetic diversity ( Craig , 2002 ) . Most site-specific recombinases can be classified into two structurally and mechanistically unrelated groups that are named for the use of either a serine or tyrosine as the active site residue ( Grindley et al . , 2006 ) . Some reactions , such as those mediated by the tyrosine recombinases Cre and FLP , only require the recombinase and its cognate DNA binding sites , whereas others involve additional accessory proteins and assemble elaborate synaptic complexes that provide tight control over chemical and mechanical steps of the reaction . The synaptic complexes formed by serine recombinases contain the two recombining DNA segments on the outside of a tetrameric protein core ( Dhar et al . , 2004; Nollmann et al . , 2004; Li et al . , 2005 ) . All four DNA strands are cleaved by near simultaneous attack on the DNA backbone by the catalytic serines to form 5′-phosphoserine linkages , thereby generating double-strand breaks at both recombination sites . DNA strands are then exchanged by a subunit rotation mechanism where one pair of synapsed subunits , together with their covalently-bound DNA strands , rotates 180° relative to the other pair ( Stark et al . , 1989; Dhar et al . , 2004 , 2009a , 2009b; Li et al . , 2005; Bai et al . , 2011 ) . Attack of the phosphoserine by the free 3′ OH ligates the DNA in the recombinant configuration . Hin is a member of the DNA invertase subclass of serine recombinases that utilize a remote enhancer element to control early and late steps of the reaction ( Johnson , 2002 ) . Hin inverts a ∼1 kb segment of chromosomal DNA between two 26 bp hix recombination sites in Salmonella enterica ( Zieg et al . , 1977; Zieg and Simon , 1980 ) . Inversion switches the orientation of a promoter , resulting in alternative expression of flagellin genes . Flagellar phase variation is one way that Salmonella evades the host immune system . DNA inversion by Hin occurs upon assembly of the invertasome , a tripartite nucleoprotein complex made up of four Hin subunits bound to two hix sites and the recombinational enhancer element ( Figure 1A ) ( Heichman and Johnson , 1990 ) . The enhancer is a 65 bp DNA sequence that has recognition sites for the bacterial nucleoid-associated DNA binding and bending protein Fis on each end . Whereas the hin enhancer is normally positioned about 100 bp from one of the hix sites , it can efficiently activate DNA inversion many kb away from the closest hix site by DNA looping ( Johnson and Simon , 1985; Kahmann et al . , 1985 ) . Hin dimers bound to each hix site are remodeled within the invertasome into a chemically-active tetramer that is also competent for subunit rotation . Crystal structure snapshots of several serine recombinases highlight the large conformational changes required for formation of the tetramer and reveal a flat and hydrophobic interface between rotating subunit pairs ( Yang and Steitz , 1995; Li et al . , 2005; Kamtekar et al . , 2006; Yuan et al . , 2008; Keenholtz et al . , 2011; Ritacco et al . , 2013 ) . 10 . 7554/eLife . 01211 . 003Figure 1 . Hin-catalyzed DNA inversion reaction and outline of Fis-Hin crosslinking experiments . ( A ) The Hin reaction pathway proceeds by Hin dimers binding to hixL and hixR and two Fis dimers binding to the enhancer element ( brown ) ( ii ) . The DNA bending protein HU aids in formation of the small ( ∼100 bp ) loop during assembly of the invertasome ( iii ) . Each of the four Hin subunits cleaves the hix DNA by forming phosphoserine linkages with the 5′ ends , and the DNA strands are exchanged by rotation of the purple and yellow subunits relative to the green and blue subunits ( iv ) . The DNA is ligated by Hin in the recombinant orientation ( v ) . ( B ) Fis-DNA crystal structure ( PDB ID: 3IV5 ) highlighting the sole cysteine at the crosslinking target residue 21 ( green spheres ) in the protruding β-hairpin arms where surrounding residues that contact Hin are shown as red sticks . ( C ) Crosslinking is performed after incubating Fis and Hin under conditions stabilizing DNA-cleaved invertasomes . pRJ2372 ( Figure 1—figure supplement 1 ) contains EcoRI sites flanking the hix sites such that each Hin subunit can be 32P-labeled by DNA polymerase and radiolabeled dNTPs after digestion with EcoR1 , and the crosslinked Fis-Hin- ( 32P ) DNA product is then detected by SDS-PAGE . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 00310 . 7554/eLife . 01211 . 004Figure 1—figure supplement 1 . Plasmid substrate design . Wild type ( wt ) shows the spacing and layout of the elements in the invertible segment on the Salmonella enterica serovar Typhimurium chromosome . All plasmids used for Hin reactions in this work contain two copies of the Hin recombination site hixL called hix1 and hix2 . For crosslinking substrates , EcoRI sites are shown as RI and are located either 14 bp or 50 bp from the position of Hin cleavage at the center of the hix site . pMS551 , pMS614 , and pMS634 are from Johnson and Simon ( 1985 ) , and pRJ2372 is from Dhar et al . ( 2009a ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 004 Formation of the supercoiling-dependent invertasome intermediate containing the Fis/enhancer element is the critical regulatory step in the Hin-catalyzed recombination reaction . It functions to ensure that intramolecular DNA inversion is the exclusive outcome of the reaction , promotes the Hin conformational changes required for DNA chemistry and exchange , and limits most subunit rotation reactions to a single 180° step ( Kanaar et al . , 1990; Heichman et al . , 1991; Moskowitz et al . , 1991; Merickel and Johnson , 2004 ) ( Figure 1A ) . Whereas wild-type Hin is chemically inactive without the Fis/enhancer system , mutants that no longer require the Fis/enhancer element or DNA supercoiling have been isolated ( Klippel et al . , 1988; Sanders and Johnson , 2004; Heiss et al . , 2011 ) . These hyperactive mutants catalyze recombination promiscuously , being no longer restricted to promoting inversion between recombination sites on the same DNA molecule and no longer confined to a single round of subunit rotation . Three residues on one of the mobile β-hairpin arms of the Fis dimer are required to activate Hin-catalyzed DNA inversion through direct contact with Hin ( Figure 1B ) ( Koch et al . , 1991; Osuna et al . , 1991; Safo et al . , 1997; Dhar et al . , 2009a ) . In the present work we first identify residues within the DNA binding domain ( DBD ) of Hin that are contacted by the β-hairpin arms of Fis and identify which two of the four Hin subunits within the tetramer are associated with Fis in the active invertasome . We show that Fis contacts to inactive Hin dimers at an early step of invertasome formation lead to assembly of catalytically-active tetramers . Unexpectedly , we find that a localized basic surface on the Hin catalytic domain is also required for enhancer-dependent assembly of active tetramers and demonstrate using tethered chemical nucleases that it is associated with enhancer DNA between the two Fis dimers . These contacts enable us to construct a molecular model for the assembly of the invertasome structure at a plectonemic DNA branch that explains how the Fis/enhancer system controls orientation-specific synapsis , promotes the quaternary changes in Hin required for DNA cleavage and exchange , and prevents multiple subunit rotations . We first used site-directed crosslinking approaches to identify regions on Hin that are required for association with the Fis/enhancer element . For these experiments , a cysteine ( Q21C ) was introduced into the tip of the Fis β-hairpin arms that are critical for activating Hin ( Figure 1B ) . Invertasomes trapped in a DNA-cleaved state were assembled using Hin , Fis-Q21C , and a supercoiled plasmid substrate and subjected to crosslinking using the heterobifunctional agent AMAS ( N- ( α-maleimidoacetoxy ) succinimide ester ) . AMAS chemically links Cys21 on Fis via the maleimide group to a lysine residue on Hin via the succinimidyl group . After quenching the crosslinking reaction , the hix recombination sites were separated from vector sequences using EcoRI , the EcoR1 ends were 32P-labeled , and the products were analyzed by SDS-PAGE ( Figure 1C ) ( Dhar et al . , 2009a ) . Crosslinking of Fis-Q21C to a lysine on Hin generates a radiolabeled Fis-Hin-DNA product , in addition to the covalently-linked Hin-DNA product from cleaved invertasomes ( Figure 2C , Hin-wt panel; Figure 2—figure supplement 1 ) . Because the two functional groups of AMAS are separated by only a 4 . 4 Å spacer , the crosslinked Hin lysine ( s ) are expected to be close to Cys21 on Fis . Experiments employing crosslinkers with different spacer lengths and with Fis-Q19C , where the cysteine on Fis appears less optimally positioned than at residue 21 , are shown in Figure 2—figure supplement 1 . 10 . 7554/eLife . 01211 . 005Figure 2 . Identification of the Fis contact region on the Hin DBD . ( A ) DNA-cleaved tetramer model of Hin with the locations of eight lysines on each subunit shown as red spheres . ( B ) Fis-Hin AMAS or BMOE crosslinking efficiencies ( mean and standard deviation from ≥three experiments ) between Fis-Q21C and eight lysine to alanine ( AMAS , 4 . 4 Å spacer , gray bars ) or cysteine ( BMOE , 8 Å spacer , dark bars ) Hin mutants . Percent Fis-Hin crosslinks are relative to the Hin- ( 32P ) DNA cleavage product . ( C and D ) Representative phosphorimages of AMAS or BMOE crosslinking experiments , respectively . A Hin- and crosslinker-independent contaminant band , which sometimes originates from an incompletely digested DNA fragment , is marked with a ( + ) . ( E ) Hin DBD—hixL crystal structure ( PDB ID: 1IJW ) with residues evaluated for crosslinking or Fis-activated DNA inversion ( Table 1 ) shown as spheres ( Cβ atoms ) . Asterisks signify residues exhibiting BMOE crosslinking with Fis-Q21C when replaced with cysteine; red highlights the two key residues proposed to directly contact Fis . ( F ) DNA inversion activity of Hin-L155A in the absence or presence of the H107Y hyperactivating mutation enabling Fis/enhancer-independent inversion . Hin was added to reactions containing pMS551 , HU , and Fis , as designated , and aliquots were taken at 0 , 1 , 2 , and 5 min . Digestion with HindIII and PstI allows the inverted and parental DNA orientations to be distinguished . Hin-wt does not generate detectable inversions under these no-Fis conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 00510 . 7554/eLife . 01211 . 006Figure 2—figure supplement 1 . Heterobifunctional crosslinking using different length crosslinkers targeting either Cys21 or Cys19 on Fis and a primary amine on Hin . Fis-Q21C exhibits efficient crosslinking with Hin beginning at 4 . 4 Å , whereas Fis-Q19C exhibits less efficient crosslinking beginning at 7 . 3 Å . The additional product at 16 . 3 Å using Fis-Q21C is absent with Hin-K146A , suggesting this product represents a crosslink between these two residues . Fis-wt and Fis-R71C ( cysteine near DNA binding domain ) do not generate Hin crosslinks ( Dhar et al . , 2009a ) . Crosslinkers are: SIA ( 1 . 5 Å ) , AMAS ( 4 . 4 Å ) , GMBS ( 7 . 3 Å ) , EMCS ( 9 . 4 Å ) , SMPH ( 14 . 2 Å ) , and KMUS ( 16 . 3 Å ) ( Pierce-Thermo Scientific ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 00610 . 7554/eLife . 01211 . 007Figure 2—figure supplement 2 . Complete set of AMAS crosslinking data between Cys21 on Fis and a primary amine on Hin . Hin-wt and the eight lysine to alanine mutants are from one experiment and the C-terminal deletion ( Δ ( 186–190 ) , which includes Lys186 and Lys187 , ) and Hin-R48A are from separate experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 00710 . 7554/eLife . 01211 . 008Figure 2—figure supplement 3 . BMOE crosslinking data between Cys21 on Fis and cysteines introduced at eight lysine residues in Hin . Hin-wt contains a single cysteine at residue 28 , which does not crosslink with Fis-Q21C ( left panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 008 Hin contains 10 lysines per subunit; six are located in the catalytic domain and four in the DBD ( Figure 2A ) . Two of the four lysines in the DBD are located in the unstructured C-terminal segment ( Chiu et al . , 2002 ) , which is not present in other DNA invertases and can be deleted from Hin without affecting Fis-activated DNA inversion . The remaining eight lysines were each individually mutated to alanine and tested for crosslinking with Fis-Q21C; a truncation mutant missing the two C-terminal lysines was also evaluated . Each of the mutants accumulated substantial amounts of cleaved invertasomes at the time of crosslinking ( 10 min incubation ) , as also shown by the levels of covalent Hin- ( 32P-DNA ) products ( Figure 2C , Figure 2—figure supplement 2 ) . Hin-K51A and K158A exhibited reduced amounts of Fis-Hin- ( 32P-DNA ) crosslinked products relative to Hin-wt ( Figure 2B , C , Figure 2—figure supplement 2 ) . These results provide an initial indication that regions important for Hin association with the Fis/enhancer segment are located around Lys158 in the DBD and around Lys51 in the catalytic domain ( Figure 2A ) . To more directly probe the region on Hin that is positioned close to the Fis β-hairpin arms in the invertasome , cysteine-cysteine crosslinking between Fis-Q21C and Hin proteins containing cysteine substitutions at the eight native lysines was performed using BMOE ( bis-maleimidoethane , 8 Å spacer ) . Hin-K146C at the N-terminus of helix 1 within the DBD generated ∼15% crosslinked Fis-Hin products ( Figure 2B , D ) , a level similar to that of wild-type Hin crosslinked to Fis-Q21C by AMAS . Hin-K158C at the C-terminus of the DBD helix 1 generated low levels of BMOE-mediated crosslinks with Fis-Q21C . No cysteine substitutions within the catalytic domain generated crosslinks ( Figure 2—figure supplement 3 ) , suggesting that the alanine mutations ( e . g . , Hin-K51A in Figure 2B , C ) may have disrupted the AMAS-mediated crosslinking via an indirect mechanism ( see below ) . Additional solvent-exposed residues within the DBD helix 1 region were converted to cysteine and those that generated substantial levels of DNA-cleaved invertasomes were tested for BMOE crosslinking with Fis-Q21C ( Figure 2D , E ) . Hin-H147C , Hin-E150C , and Hin-Q151C within the α-helix formed crosslinks with Fis-Q21C , but H160C in the loop connecting to DBD helix 2 did not generate crosslinks . The 8 Å crosslinks between Fis-Q21C and five positions within helix 1 of the Hin DBD demonstrate that the Hin DBD is in close proximity to the Fis β-hairpin arms in the DNA-cleaved invertasome . An alanine scan of the DBD helix 1 was performed to determine the specific residues required for Fis/enhancer activation . Of the 11 solvent-exposed residues tested , Q151A , R154A , and L155A , exhibited strong reductions of Fis-activated DNA inversion in vivo and in vitro ( Table 1 ) . The most severe mutant , Hin-L155A , exhibits a >50-fold decrease of Fis-activated DNA inversion rates in vitro , comparable to the effects of the strongest Fis β-hairpin arm point mutations ( Safo et al . , 1997 ) ( see also Figure 7—figure supplement 1 ) . Although Hin-Q151A exhibits low activity , particularly in vitro , the role of Gln151 may be at least partially indirect because Hin-Q151C exhibits substantial Fis-activation and forms crosslinks with Fis-Q21C ( Figure 2D ) ( ‘Discussion’ ) . Additional changes were made at Hin-Arg154 and Hin-Leu155 and evaluated for DNA inversion ( Table 1 ) . The substitutions tested at Hin-Leu155 nearly inactivate recombination . Substitutions of Hin-Arg154 with polar residues ( asparagine and serine ) exhibit intermediate rates of DNA inversion , but alanine , cysteine , and aspartic acid are nearly inactive . 10 . 7554/eLife . 01211 . 009Table 1 . Fis-activated DNA inversion activities of Hin DBD helix 1 mutantsDOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 009MutantDNA inversion in vivoDNA inversion in vitro†WT+++0 . 31 ± 0 . 02WT ( −Fis ) –<0 . 001K146A+++0 . 30 ± 0 . 03H147A+++0 . 38 ± 0 . 02E148A+++0 . 19 ± 0 . 03Q149A+++0 . 37 ± 0 . 04E150A+++0 . 30 ± 0 . 05Q151A++0 . 08 ± 0 . 01S153A+++0 . 17 ± 0 . 01R154A+0 . 08 ± 0 . 02L155A±<0 . 005E157A+++0 . 29 ± 0 . 03K158A+++0 . 17 ± 0 . 02R154S+0 . 13 ± 0 . 03R154N++0 . 17 ± 0 . 02R154C+0 . 06 ± 0 . 01R154D+0 . 02 ± 0 . 01L155V±0 . 03 ± 0 . 01L155G±<0 . 01L155S±<0 . 02L155T±<0 . 01L155K±<0 . 01L155R±<0 . 02In vivo DNA inversion rates as measured by color development on lactose MacConkey media . +++ indicates red colonies developed within 24 hr , ++ red colonies between 25–29 hr , + red colonies between 30–34 hr , ± some red or papillations after 36 hr , and -no evidence of inversion ( red ) after 48 hr , as observed for no Hin or no Fis experiments . †In vitro recombination rates ( DNA inversions/molecule/minute ) obtained with purified proteins ( mean and standard deviation from at least three determinations ) . None of the above residues make DNA contacts in the Hin DBD co-crystal structures ( Feng et al . , 1994; Chiu et al . , 2002 ) ( Figure 2E ) , and mutants containing alanine substitutions at Arg154 or Leu155 bind hix DNA indistinguishably from Hin-wt ( data not shown ) . In order to confirm that mutations of these residues specifically affect Fis-activation and not protein misfolding or other aspects of the recombination reaction such as site synapsis or DNA chemistry , rescue experiments were performed . Hin-R154A and Hin-L155A were each coupled to the gain-of-function mutation Hin-H107Y , which can catalyze recombination without the Fis/enhancer system ( Sanders and Johnson , 2004; Heiss et al . , 2011 ) . As shown in Figure 2F , wild-type Hin recombines the plasmid substrate to a near equilibrium mixture of parental and inverted products within 5 min in the presence of Fis , whereas Hin-L155A inverts less than 2% of the substrate in the same amount of time . In contrast , both Hin-H107Y and the double mutant Hin-H107Y/L155A catalyze recombination in the absence of Fis at similar rates . Hin-R154A also efficiently promoted inversion when coupled with H107Y ( data not shown ) . Taken together , we conclude that Fis specifically interacts with Hin residues Arg154 and Leu155 and that these contacts are required for DNA inversion catalyzed by wild-type Hin . We next asked which of the four Hin subunits were positioned proximal to the two Fis dimers in the invertasome . Site-directed crosslinking was performed on DNA substrates that enable a single Hin subunit to be labeled by having a restriction site flanking only one of the hix sites positioned appropriately for radiolabeling ( Figure 3A , Figure 1—figure supplement 1 ) . Crosslinking reactions were performed with Fis-Q21C and Hin-K146C using BMOE ( 8 Å ) . Fis-Hin crosslinks were only observed when Hin was bound to hix1L or to hix2R; no Fis-Hin crosslinks formed at hix1R or hix2L . Similar crosslinking experiments employing AMAS ( cysteine-lysine ) with wild-type Hin and Fis-Q21C also show that Fis only crosslinks with subunits bound to hix1L or hix2R ( Figure 3—figure supplement 1 ) . These crosslinking results demonstrate that Fis contacts the DBDs of only the bottom two Hin subunits within the cleaved invertasome as drawn in Figure 3A . 10 . 7554/eLife . 01211 . 010Figure 3 . Fis contacts the two Hin subunits bound to half-sites hix1L and hix2R that are positioned at the base of the invertasome . ( A ) Plasmid substrates used to determine which Hin subunits contact Fis ( see also Figure 1—figure supplement 1 ) . The 3′ DNA ends from each of the four Hin-DNA covalent complexes are labeled using the standard substrate pRJ2372 upon digestion with EcoRI and DNA polymerase fill-in ( all hix ) . Four additional substrates were employed in which the DNA from only one Hin-DNA cleavage complex can be labeled due to the locations of the EcoR1 sites . DNA-labeled crosslinks between Fis-Q21C and the Hin protomers bound to the left half-site of hix1 ( hix1L ) and the right half-site of hix2 ( hix2R ) are observed ( red outlines ) . ( B ) Fis-Hin crosslinking between Fis-Q21C and Hin-K146C on the different DNA substrates using BMOE . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 01010 . 7554/eLife . 01211 . 011Figure 3—figure supplement 1 . Heterobifunctional crosslinking ( cysteine-lysine ) between Fis-Q21C and Hin-wt was performed using N-succinimidyl iodoacetate ( 1 . 5 Å , no crosslinks obtained ) or AMAS ( 4 . 4 Å ) . As observed with the BMOE ( cysteine-cysteine ) crosslinking , Fis-Hin crosslinks by AMAS only occurred with Hin subunits bound to hix1L and hix2R . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 011 As described above , Hin-K51A exhibits very low AMAS ( Lys ) crosslinking with Fis-Q21C , but crosslinking experiments with Hin-K51C provide no evidence for a location proximal to Fis ( Figure 2A–C , Figure 2—figure supplements 2 and 3 ) . Lys51 is located within the predicted helix-B of the Hin catalytic domain , which contains two other basic residues , Arg48 and Lys47 ( Figure 4A ) . Hin-R48A failed to form detectable AMAS crosslinks under standard conditions ( Figure 2—figure supplement 2 ) . Hin-K47A associates unstably with the enhancer as demonstrated by relatively efficient Fis-Hin crosslinking at early times but undetectable crosslinking after 20 min incubation ( Figure 4—figure supplement 1 ) . Fis-activated DNA inversion rates of Hin-R47A , R48A , and K51A are moderately decreased with R47A and K51A exhibiting about threefold reductions in vitro ( Figure 4B ) . However , the combinations of K47A/R48A and R48A/K51A severely impair Fis-activated Hin inversion both in vivo and in vitro ( Figure 4B , C ) . These mutations have no effect on Hin binding to hix ( data not shown ) . To confirm that these residues are directly functioning in Fis/enhancer-activation , Hin-R48A/K51A was coupled to the gain-of-function mutation Hin-H107Y . Hin-R48A/K51A/H107Y promotes Fis-independent inversion at rates that are indistinguishable from the single H107Y mutant ( Figure 4C ) , demonstrating that these mutations are not disturbing Hin catalytic properties . We conclude that basic residues in the Hin helix-B region function in the Fis/enhancer-dependent activation step of the Hin-catalyzed DNA inversion reaction . 10 . 7554/eLife . 01211 . 012Figure 4 . Residues within Hin helix-B function in Fis/enhancer-dependent Hin activation and control of subunit rotation . ( A ) DNA-cleaved Hin tetramer model with the locations of basic residues in the helix-B region highlighted with red spheres . ( B ) Fis-activated DNA inversion rates of helix-B Hin mutants reported as inversions/molecule/minute ( mean and standard deviation from ≥three experiments ) . In vivo rates are given in parentheses; see Table 1 for legend . ( C ) DNA inversion kinetics of a Hin helix-B double mutant . Inversion reactions were performed in the presence and absence of Fis for 0 , 1 , 2 , and 5 min . Hin-R48A/K51A exhibits 10-fold slower kinetics than Hin-wt under Fis-activating conditions , but is fully competent for Fis-independent inversion when coupled to the hyperactivating mutation Hin-H107Y . ( D ) Schematic representation of topological changes during Hin recombination . Normally Hin ligates the DNA after a single DNA exchange by subunit rotation due to the small loop between the hix site and the enhancer , resulting in an unknotted inverted product . If the loop is large or the enhancer is released ( as shown ) , multiple rounds of subunit rotations can occur resulting in DNA knots with increasing numbers of nodes . ( E ) Quantitation of DNA knotted forms relative to the amount of initially Hin-cleaved plasmid ( mean and standard deviation from ≥three experiments ) from single-round knotting experiments ( see Figure 4—figure supplement 3 ) . Hin-wt efficiently forms knots only on pMS634 containing the long spacer , whereas Hin-R48A and K51A are insensitive to spacer length , reflecting an unstable interaction with the enhancer . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 01210 . 7554/eLife . 01211 . 013Figure 4—figure supplement 1 . Hin helix-B mutation K47A destabilizes association of the Fis/enhancer with the active Hin tetramer . Invertasomes were assembled with Fis-Q21C and Hin-wt or Hin-K47A in Mg2+-free ethylene glycol buffer . At 2 . 5 , 10 , and 20 min after addition of proteins , the reaction was subjected to AMAS crosslinking for 30 s . Fis-Hin crosslinked products were quantified relative to the total Hin- ( 32P-DNA ) products at that time point and plotted with respect to time of incubation . Note that the percent of Hin-K47A—Fis crosslinked products relative to Hin-cleaved DNA are relatively high initially but decrease with time , whereas Hin-wt—Fis crosslinks remain high over the 20 min time period . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 01310 . 7554/eLife . 01211 . 014Figure 4—figure supplement 2 . Partial disruption of Fis-Hin interactions leads to DNA knotting by processive subunit rotations . Fis , HU , and Hin-wt or Hin-R154N were incubated for 10 min under inversion conditions with pMS614 , which contains AT dinucleotides at the hix2 crossover region that prevents ligation in the recombinant orientation ( Figure 1—figure supplement 1; Merickel and Johnson , 2004 ) . The reaction was quenched , ethanol precipitated , digested with the nicking enzyme Nt . BsmAI , and the product topologies were resolved on an agarose gel . Hin R154N generates 50% knotted products whereas Hin-wt generates 2% , despite a twofold decrease in inversion rate by Hin-R154N . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 01410 . 7554/eLife . 01211 . 015Figure 4—figure supplement 3 . Primary data for single round knotting experiments summarized in Figure 4E . ( A ) pMS551 or pMS634 , which have 99 bp or 696 bp between the enhancer and closest hix site , respectively , were incubated with Hin , Fis , and HU for 5 min under ethylene glycol Mg2+-free cleavage conditions , and an aliquot was quenched and electrophoresed in an agarose gel to assess the amount of DNA cleavage . Vector and invertible segment bands represent double strand cleavages at both hix sites . ( B ) The remainder of the reaction from panel A was diluted with buffer containing no ethylene glycol and 10 mM Mg2+ to enable ligation . The DNA was then nicked and electrophoresed in an agarose gel to resolve knotted forms . The identities of the bands , including the number of nodes in each knotted form for pMS551 and pMS634 are shown on the left and right , respectively . The percent of knotted ligation products per Hin-cleaved substrate is given in Figure 4E . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 015 Fis/enhancer-association with the activated Hin tetramer inhibits multiple subunit rotations ( processive recombination ) during the DNA exchange reaction because the small ( ∼100 bp ) loop between hixL and the enhancer element prevents multiple windings of DNA ( Figure 4D ) ( Heichman et al . , 1991; Merickel and Johnson , 2004; Dhar et al . , 2009a ) . Therefore , DNA exchange is normally limited to a single 180° rotation step , resulting in an unknotted inversion product . Additional subunit rotations , which generate DNA knots of increasing complexity , can occur when the enhancer is released from the invertasome complex . Reactions employing Fis mutants that have weakened interactions with Hin ( Merickel and Johnson , 2004 ) , or Hin DBD mutants that have weakened interactions with Fis ( Figure 4—figure supplement 2 ) lead to processive recombination . Substrates containing long DNA segments between the enhancer and each hix site also exhibit increased processive recombination because multiple DNA windings in the context of large DNA loops do not restrict subunit rotation ( Heichman et al . , 1991; Merickel and Johnson , 2004; Dhar et al . , 2009a ) . Single-round knotting experiments were performed to test whether mutations within the Hin helix-B region increase processive DNA exchanges . Hin was incubated with Fis and the plasmid substrate for 5 min under conditions that accumulate DNA-cleaved invertasomes . A portion of the reaction was then quenched with SDS and evaluated for the amount of substrate reacted ( Figure 4—figure supplement 3A ) . Under these conditions wild-type Hin cleaves 70–80% and helix-B mutants cleave 30–50% of the plasmid DNA . The remainder of the reaction was briefly switched to conditions allowing DNA ligation and the amount of knotted products were measured ( Figure 4E , Figure 4—figure supplement 3B ) . Two substrates were used: pMS551 , which contains the native ( 99 bp ) spacing between the hix1 site and the enhancer , and pMS634 , which has a long ( 868 bp ) spacer that does not restrict subunit rotation ( Figure 1—figure supplement 1 ) . Only ∼5% of the ligation products generated by Hin-wt on the short spacer substrate contained knots , reflecting a very low amount of processive DNA exchange even under reaction conditions where the invertasome is held in a DNA-cleaved structure for an extended time ( Figure 4E ) . On the other hand , wild-type Hin knotted 30% of the reacted substrates with a long spacer between the hix site and the enhancer . Hin-R48A and Hin-K51A , however , are insensitive to the length of the loop between the hix1 site and the enhancer; both mutants knotted about 40% of the reacted short spacer substrates corresponding to an eightfold increase in processive DNA exchange compared with wild-type Hin . The increased processive DNA exchange by Hin-R48A and Hin-R51A provide further evidence that these mutations destabilize the association of the enhancer with the Hin synaptic complex . To examine whether the Hin helix-B region is contacting enhancer DNA , the chemical nuclease FeBABE was covalently attached to cysteines introduced into the region ( Meares et al . , 2003 ) . Invertasomes were assembled , and the enhancer region probed for DNA scission after activation of FeBABE with H2O2 and ascorbate ( Figure 5A ) . Many of the tested cysteine mutants coupled to FeBABE exhibited low Hin-catalyzed DNA cleavage activity , but derivatives at residues 52–54 , located near the C-terminal end of helix-B exhibited relatively high activity ( Figure 5—figure supplement 1A ) . Hin-N54C-FeBABE , and to a lesser extent Hin-Y52C-FeBABE , generated two prominent scission sites within the center of the enhancer segment ( Figure 5—figure supplement 1B ) . These are located between the Fis dimer binding sites , as demarked by DNA scission induced by Fis-N98C coupled with FeBABE ( Figure 5—figure supplement 1B ) ; DNA scission by chemical nucleases coupled to Fis residue 98 has been used previously to map Fis binding site locations ( Pan et al . , 1994 ) . Electrophoresis on DNA sequencing gels ( Figure 5B , C ) identified the precise locations of the scission sites by Hin-N54C-FeBABE and Fis-98C-FeBABE on the enhancer DNA sequence ( Figure 5D ) . Figure 5E presents a molecular model showing the locations of the relevant Hin and Fis subunits on the enhancer DNA ( ‘Discussion’ ) together with the scission data . 10 . 7554/eLife . 01211 . 016Figure 5 . Localization of proteins on the hin enhancer DNA by site-specifically tethered FeBABE-mediated DNA scission . ( A ) Experimental approach for mapping enhancer DNA contacts by the Hin helix-B region . Red stars denote sites of FeBABE coupling . ( B and C ) Sequencing gels resolving primer extension products of DNA scission by FeBABE coupled to Hin-N54C , Hin-wt ( control ) , and Fis-N98C next to dideoxy sequencing reactions . Top strand primer in B; bottom strand primer in C . ( D ) Sequence of enhancer DNA with primary Fis-N98C-FeBABE and Hin-N54C-FeBABE scission sites denoted . Thickness of arrows approximates scission efficiency based on several gels . Gray lines designate the 15 bp core Fis binding sites . Below are the sequences of two mutant enhancers used in panel F; underlined sequences are changes from the wild type . ( E ) Molecular model of enhancer segment within the invertasome structure ( ‘Discussion’ , ‘Material and methods’ , and Figure 7C ) . Shown are the helix-turn-helix regions of the Fis dimers ( orange ) and the catalytic domains of the two enhancer-proximal Hin subunits ( green and blue ) . The Hin domains are rotated 40° about the y-axis relative to Figure 2A . Residues 98 on Fis and 54 on Hin have been replaced with cysteine and the Sγ atoms where FeBABE ( 12 Å to the Fe that generates hydroxyl radicals ) is coupled are highlighted by magenta spheres . DNA scission sites generated by Hin-N54C-FeBABE are in red and by Fis-N98C-FeBABE are in cyan . ( F ) Inversion rates on mutant enhancers contained on pRJ765 and pRJ2943 relative to their wild-type enhancer parent substrates pMS551 and pRJ2372 , respectively ( Figure 1—figure supplement 1 ) . Inversion rates are reduced by the mutations but are much greater than no-Fis reactions ( <0 . 001 inversions/molecule/min ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 01610 . 7554/eLife . 01211 . 017Figure 5—figure supplement 1 . Activities and scission over the enhancer region by Hin mutants coupled to FeBABE . ( A ) Invertasome were assembled on pRJ2372 with Hin-wt ( lane 3 ) or Hin mutants containing cysteine substitutions at the designated positions ( lanes 4–8 ) that had been subjected to FeBABE coupling . Formation of DNA-cleaved invertasomes was measured by addition of SDS after a 10 min incubation and electrophoresis on an agarose gel after digestion with proteinase K . Lane 1 is the plasmid control , and lane 2 is incubation with Fis-N98C that had been coupled with FeBABE . Hin-catalyzed DNA cleavage generates the vector backbone and the invertible segment ( inv seg ) . ( B ) Na ascorbate and H2O2 were added to the reactions in panel A to initiate DNA scission by FeBABE for 30 s . Scission over the enhancer segment was probed by primer extension using a 32P-labeled oligonucleotide that hybridizes 66 nts downstream from the closest Fis binding site ( see Figure 5A ) , and the products were electrophoresed on an 8% denaturing polyacrylamide-urea gel . Size standards ( nts ) are on the left . High resolution ( sequencing ) gels of the scission products by Hin-N54C-FeBABE and Fis-N98C-FeBABE are shown in Figures 5B , C . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 017 We asked if the identity of the sequence between the Fis binding sites where the Hin helix-B region contacts in the invertasome is important for enhancer activity . When most of the DNA between the two Fis binding sites is replaced by non-native sequence ( pRJ765 , Figure 5D ) , Hin inversion rates are reduced about fivefold ( Figure 5F ) . When only the two A/T-rich segments identified to be in proximity to the Hin helix-B region by the FeBABE experiments are replaced by G/C-rich sequences ( pRJ2943 , Figure 5D ) , Hin inversion rates are reduced 2 . 5-fold ( Figure 5F ) . We conclude that there is a modest effect of sequence identity within the enhancer DNA segments that are contacted by the helix-B region of Hin . The experiment outlined in Figure 6A demonstrates that Fis productively interacts with Hin dimers at an early step to promote formation of chemically-active Hin tetramers . Disulfide-linked Hin-M101C dimers ( Figure 6B ) bind normally to hix sites but are locked in an inactive conformation ( Figure 6C , lane 4 ) ( Haykinson et al . , 1996 ) . Upon reduction of the disulfide bond and in the presence of Fis , Hin-M101C can proceed to generate DNA-cleaved invertasomes ( Figure 6C , lane 5 ) and ligated inversion products ( Haykinson et al . , 1996 ) . Fis-Q21C and disulfide-linked Hin-M101C dimers were incubated with the DNA substrate and subjected to Fis-Hin crosslinking for 15 s . The crosslinking reaction was quenched with DTT ( plus free lysine ) , which also breaks the disulfide linkage and enables the reduced M101C dimers to then generate active tetramers and cleave the hix DNA . SDS-PAGE after EcoR1 digestion and radiolabeling of the EcoR1 ends revealed a substantial amount of Fis-Hin crosslinked products ( Figure 6D , lane 6 ) , albeit less than the amount generated with DNA-cleaved Hin-M101CSH invertasomes ( Figure 6D , lane 4 ) . Extended crosslinking times only slightly improved the yield of Fis-Hin dimer products ( Figure 6D , lanes 10–12 ) . Control experiments where DTT and lysine were added immediately before the crosslinker gave no detectable Fis-Hin crosslinked product ( Figure 6D , lane 7 ) , indicating that the Fis-Hin crosslinks could not have been formed after reduction of the disulfide-linked Hin-M101C dimer . We conclude that Hin dimers that are covalently crosslinked with Fis on the enhancer can be remodeled into catalytically-active tetramers . 10 . 7554/eLife . 01211 . 018Figure 6 . Fis-Hin connections at early and late steps in the DNA inversion reaction . ( A–D ) Hin dimers covalently linked to Fis proceed to tetramers active for DNA cleavage . ( A ) Outline of the experiment . ( B ) Hin dimer model highlighting Met101 . Hin-M101C forms disulfide-linked dimers; insert shows non-reducing SDS-PAGE of reduced Hin-M101C ( lane 1 ) and purified disulfide-linked dimeric Hin-M101C ( lane 2 ) . ( C ) Hin cleavage reactions with Fis-Q21C ( 10 min ) displayed on an agarose gel . Lane 1 , unreacted DNA; lane 2 , Hin-wt reaction; lane 3 , reduced Hin-M101C reaction; lane 4 , disulfide-linked Hin-M101Cs–s reaction; lane 5 , disulfide-linked Hin-M101Cs–s reaction then incubated 10 min with 10 mM DTT; lane 6 , disulfide-linked Hin-M101Cs–s reaction , crosslinked with GMBS ( 7 . 3 Å spacer ) , then incubated 10 min with 10 mM DTT , which inactivates the crosslinker and reduces the disulfide bond . ( D ) Fis-Hin crosslinking products displayed on an SDS gel . Lanes 1–6 are Hin-wt and reduced or disulfide-linked M101C crosslinked with Fis-Q21C for 30 s with GMBS as designated ( + ) , the crosslinker was quenched , and the reaction incubated an additional 10 min under reducing conditions to form DNA-cleaved invertasomes . The presence of a Fis-Hin- ( 32P ) DNA crosslinked product in lane 6 , as well as in lanes 10–12 where crosslinking times ( s ) were varied , demonstrates that covalently crosslinked Fis-Hin dimers can transition into Hin tetramers competent to cleave DNA . In lane 7 , the DTT + lysine quench was added immediately prior to the crosslinker , demonstrating that the crosslinking in lanes 6 and 10–12 could not have occurred after reduction of the dimer . ( E and F ) DNA ligation by Hin proceeds when Hin is covalently crosslinked to Fis . ( E ) Outline of the experiment . ( F ) Control and crosslinked DNA cleavage reactions were either quenched directly ( − ) or chased by addition of 10 mM Mg2+ and dilution of the ethylene glycol to induce DNA ligation . The amount of cleaved DNA remaining after switching to ligation conditions is assessed by the levels of Fis-Hin- ( 32P ) DNA or Hin- ( 32P ) DNA complex . Over 85% of the Hin-DNA and >95% of the Fis-crosslinked Hin-DNA covalent product is lost , demonstrating Fis-Hin association does not inhibit DNA ligation . The band designated # is a labeled DNA fragment from the substrate . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 018 We tested whether Hin tetramers that are covalently crosslinked to Fis within DNA-cleaved invertasomes remain active for DNA ligation , the last chemical step of the reaction ( Figure 6E ) . An earlier study with the related Gin DNA invertase concluded that the Fis-bound enhancer is normally released prior to the DNA ligation step ( Kanaar et al . , 1990 ) . Stabilized DNA-cleaved invertasomes were assembled with Fis-Q21C and Hin-wt and crosslinked with AMAS ( 4 . 4 Å spacer ) or GMBS ( 7 . 3 Å spacer ) . Uncrosslinked and Fis-Hin crosslinked invertasomes were then switched to reaction conditions allowing for ligation . As recognized by the loss of the Hin serine-DNA linkage in the Fis-Hin- ( 32P-DNA ) and Hin- ( 32P-DNA ) bands , both the Fis-Hin crosslinked ( Figure 6F , lanes 3 , 4 and 5 , 6 ) and uncrosslinked ( lanes 1 , 2 ) complexes were able to efficiently promote DNA ligation . We conclude that ligation can efficiently occur when Hin is covalently linked to Fis , reinforcing earlier topological data indicating that the enhancer normally remains associated with the Hin complex throughout the reaction ( Heichman et al . , 1991; Merickel and Johnson , 2004; Dhar et al . , 2009a ) . The hin enhancer contains two Fis dimer binding sites separated by 47 bp between their centers ( Figure 5D ) ( Johnson and Simon , 1985; Bruist et al . , 1987; Johnson et al . , 1987 ) . A structural model of the Fis-bound enhancer ( ‘Materials and methods’ ) results in an S-shaped DNA structure due to Fis-induced bending ( see Figure 7F; Video 1 ) . Previous helical phasing experiments provide strong evidence that the shape of the enhancer segment is critical for function ( Johnson et al . , 1987 ) . The Hin tetramer model , based on the crystal structures of the catalytic domain of the homologous serine recombinase γδ resolvase ( PDB ID: 1ZR4 ) ( Li et al . , 2005 ) and Hin DBD ( PDB ID: 1IJW ) ( Chiu et al . , 2002 ) , has been described previously ( Dhar et al . , 2009a ) . Extensive site-directed crosslinking data supports the validity of this model for the DNA-cleaved Hin synaptic complex , which is in a structure competent for DNA exchange by subunit rotation ( Li et al . , 2005; Dhar et al . , 2009a , 2009b ) . 10 . 7554/eLife . 01211 . 019Figure 7 . Assembly of the Hin invertasome . ( A ) Two Hin dimers bound to hix sites docked onto the enhancer ( brown ) with helix-B of the catalytic domain and helix-1 of the DBD highlighted in red . Fis dimers are gold with their mobile β-hairpin arms colored magenta . The hix DNA segments cross the enhancer to form 2 ( − ) nodes , consistent with a branch on negatively supercoiled DNA . ( B ) The pre-cleaved Hin tetramer model docked onto the enhancer . ( C ) DNA-cleaved Hin tetramer model docked onto the enhancer . In this conformation , basic residues within helix-B of the Hin catalytic domain ( Lys47 , Arg48 , and Lys51 , displayed as sticks ) are close to the enhancer DNA , and the flat interface enabling subunit rotation has formed . ( D ) Rotation ( 50° ) of the yellow and purple subunits relative to the green and blue subunits that remain bound to the Fis/enhancer element . ( E ) Complete subunit exchange positions the cleaved DNA ends into the recombinant configuration . See Videos 1 and 2 . ( F ) DNA-cleaved Hin tetramer model rotated 90° about the x-axis relative to panel C with the electrostatic surface potential ( ± 4 kT e−1 ) displayed . Two distinct basic regions ( blue ) surrounding helix-B on the bottom two Hin subunits are positioned adjacent to the enhancer DNA . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 01910 . 7554/eLife . 01211 . 020Figure 7—figure supplement 1 . Chemical properties of Fis residues proposed to contact the Hin DBD . A previous cysteine scan of residues throughout the β-hairpin arm of Fis identified the prime importance of Val16 , Asp20 , and Val22 located near the tip ( see Figure 1B ) for activation of Hin-catalyzed DNA inversion ( Safo et al . , 1997 ) . The results of further mutagenesis of these residues are presented here . Fis mutants were evaluated for activation of Hin-catalyzed DNA inversion in vivo using the same lacZ reporter system as utilized for the Hin mutants , except that the Fis mutant gene on a plasmid was introduced into the cell containing Hin-wt ( see Safo et al . , 1997 for experimental details ) . For each residue the order of amino acid residues ( top to bottom ) reflects their relative in vivo activities . ( ++++ ) indicates wild-type activity where red colonies developed on lactose MacConkey agar plates within 27 hr . ( +++ ) indicates red colonies developed between 28–30 hr; ( ++ ) indicates red colonies between 32–34 hr , ( + ) indicates red colonies after 36–38 hr , ( ± ) indicates evidence of very low inversion after 40 hr , ( − ) indicates no detectable inversion after 48 hr like the no Fis control . Selected Fis mutants were partially purified and assayed for activation of Hin-catalyzed inversion in vitro . Numbers in parentheses reflect inversion rates relative to Fis-wt ( set to 1 . 0 ) . The data indicate that Fis residues 16 and 22 are most active when hydrophobic ( Val , Ile , Leu ) . Polar substitutions ( Ser , Thr , Glu ) at Asp20 exhibit moderate activity , and a leucine substitution has surprisingly high activity . These Fis residues are proposed to specifically interact with Hin DBD residues Arg154 , Leu155 , and possibly Gln151 during invertasome assembly . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 02010 . 7554/eLife . 01211 . 021Figure 7—figure supplement 2 . Electrostatic surface potentials of Hin and γδ resolvase ( ± 4 . 0 kT e−1 for all images ) . ( A ) Hin invertasome model in a view similar to Figure 7F , except that the enhancer DNA is partially transparent to more easily visualize the basic regions ( blue ) surrounding the B-helices . ( B ) γδ resolvase tetramer ( PDB ID: 1ZR4 ) in a similar orientation as the Hin tetramer in panel A . Note the absence of the prominent basic patches that are present in Hin . ( C ) View of the Hin invertasome in a similar orientation as in Figure 7C . Arrows denote the basic regions surrounding the B-helices . ( D ) γδ resolvase tetramer in a similar orientation as the Hin tetramer in panel C . Arrows point to a basic region containing resolvase residues Lys29 , Arg32 , and Lys54; in wild-type resolvase Arg2 would also be located within the basic patch . Arg2 , Arg32 , Lys54 , and Glu56 comprise the 2–3′ crystallographic interface between dimers that is important for assembly of the active synaptic complex ( Hughes et al . , 1990; Murley and Grindley , 1998; Burke et al . , 2004; Olorunniji et al . , 2008 ) . The Sin resolvase regulatory residues F52 and R54 , which also participate in a crystallographic dimer-dimer interface , are located within this region as well ( Mouw et al . , 2008 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 02110 . 7554/eLife . 01211 . 022Video 1 . Different views of the Hin invertasome model shown in Figure 7C . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 022 Manual rigid-body docking positioned the Hin tetramer model onto the Fis-bound enhancer such that the DNA strands at the base of the hix sites crossed the enhancer at the base of a plectonemic branch on ( − ) supercoiled DNA ( shown schematically in Figure 1A ) , consistent with previous topological and electron microscopy studies ( Kanaar et al . , 1988; Heichman and Johnson , 1990; Heichman et al . , 1991 ) . The Fis-bound enhancer and Hin tetramer units fit remarkably well such that the critical amino acid triad on either of the flexible β-hairpin arms on both Fis dimers can be positioned proximal to Hin residues Gln151 , Arg154 , and Leu155 on the subunits bound to the hix1L and hix2R half sites ( Figure 7C; Video 1 ) . Previous experiments employing Fis heterodimers have shown that only one of the β-hairpin arms from each of the Fis dimers is sufficient to activate Hin inversion ( Merickel et al . , 1998 ) . Our data are consistent with the primary interaction surface between these proteins being composed of Val16 and Val22 on Fis and Leu155 on Hin together with Asp20 on Fis and Arg154 on Hin ( Figure 7—figure supplement 1 ) . Significantly , Hin residues Arg154 , Leu155 , and Gln151 , which also may contribute to the Fis contact patch , are conserved among Fis/enhancer-dependent DNA invertases , but not the related resolvases ( Figure 8A ) , further supporting the function of these residues in the Fis-activation step . 10 . 7554/eLife . 01211 . 023Figure 8 . Sequence alignments of select serine recombinases over the regions that Hin contacts the Fis/enhancer . ( A ) Sequence alignment over the DBD helix-1 region of members of the Fis/enhancer-dependent DNA invertase family ( Hin , Gin , Cin ) and Fis/enhancer-independent resolvase family ( γδ , Sin ) . Secondary structure is from Hin ( PDB ID: 1IJW ) ; the lengths of helix-1 in resolvases vary . Solvent-exposed residues Gln151 , Arg154 , and Leu155 , which are proposed to directly or indirectly interact with Fis , are uniquely conserved among DNA invertases . Sin residues Val163 and Ile164 ( underlined ) mediate synapsis of regulatory subunits during formation of the Sin synaptic complex ( Mouw et al . , 2008 ) . ( B ) Sequence alignment over the helix-B region of DNA invertases ( Hin , Gin , Cin ) , resolvases ( γδ , Sin ) , and a large serine recombinase ( TP901 integrase ) . Secondary structure is from the Hin model; the helix-B boundaries are similar in most resolvase and Sin structures . Hin residue Asn54 , the site of FeBABE coupling , is marked with an asterisk . The basic character of Hin residues 47 , 48 , and 51 are uniquely conserved among DNA invertases . Mutations at the conserved residue Arg43 , which functions in folding or catalysis in other serine recombinases ( Olorunniji and Stark , 2009; Keenholz et al . , 2011 ) , also inactivate Hin . γδ resolvase residues Lys54 and Glu56 and Sin residues Phe52 and Arg54 ( underlined ) participate in protein-protein interactions between regulatory and catalytic subunits in these resolution reactions ( Hughes et al . , 1990; Murley and Grindley , 1998; Burke et al . , 2004; Mouw et al . , 2008; Olorunniji et al . , 2008; Rowland et al . , 2009 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 023 Our crosslinking and mutagenesis data unexpectedly identified an important region modeled to be on helix-B of the Hin catalytic domain that also plays a critical role in Fis/enhancer-activated DNA inversion . Like the helix-1 region of the DBD , the helix-B region is specifically required for Fis/enhancer activation because mutations within this region have no effect on Fis-independent recombination by hyperactive Hin mutants . This region forms a localized basic surface , which is positioned against the DNA segment connecting the two Fis binding sites of the enhancer in our model ( Figure 7C , F , Figure 7—figure supplement 2A , C ) , as demonstrated by scission by the chemical nuclease FeBABE coupled to residue 54 located adjacent to helix-B . Arginines and lysines from the helix-B region protrude towards AT-rich ( ATTTA and TAATG ) minor grooves in a manner reminiscent of histone-DNA interactions in nucleosomes ( Figure 7C ) ( Luger et al . , 1997 ) . The increased electronegative potential of the AT-rich , narrowed , minor groove may enhance interactions by these basic residues on Hin ( Rohs et al . , 2009 ) . Indeed , enhancer sequences that are G/C-rich rather than A/T-rich exhibit moderately reduced activation rates ( Figure 5F ) . The mutations could also be causing small changes in DNA curvature of the enhancer that impact invertasome assembly . The basic character of key residues within the helix-B region is also uniquely conserved among the DNA invertase subclass of serine recombinases ( Figure 8B ) . Of the 15 DNA invertases we have aligned , residue 47 is always a lysine , and residues 48 and 51 are equally represented by lysine and arginine , suggesting they are not involved in base-specific interactions . Members of the resolvase or integrase subclasses of serine recombinases , which are not regulated by a Fis/enhancer system , tend to not have lysines and arginines at the same positions , and the electrostatic surface over this region in the resolvase tetramer is much less basic ( Figure 7—figure supplement 2 ) . However , as discussed further below , both γδ/Tn3 and Sin resolvases contain nearby regulatory residues that mediate critical protein-protein contacts during assembly of their respective synaptic complexes ( Hughes et al . , 1990; Murley and Grindley , 1998; Burke et al . , 2004; Mouw et al . , 2008; Olorunniji et al . , 2008; Rowland et al . , 2009 ) . The invertasome model depicted in Figure 7C represents a late complex in the activation pathway in which all four DNA strands are cleaved and poised for exchange by subunit rotation . However , the Fis/enhancer system initially functions much earlier in the pathway to promote synapsis and remodeling of Hin dimers into the active tetramer , as wild-type Hin is unable to form active tetramers without the Fis/enhancer element ( Dhar et al . , 2004; Sanders and Johnson , 2004 ) . We demonstrate here that Fis is able to productively contact inactive Hin dimers and that Hin dimers covalently crosslinked to Fis can transition into an active tetramer competent for DNA cleavage ( Figure 6A–D ) . In Figure 7A DNA-bound Hin dimer models are docked to the Fis dimers on the enhancer in a manner approximating the Fis-Hin subunit contacts in the invertasome model . The two Hin dimers are sandwiched between the Fis β-hairpin arms such that the catalytic domains are adjacent to each other and thus optimally positioned to isomerize into a tetramer . Remodeling into the active tetramer may occur through an intermediate captured by the crystal structures of TP901 integrase ( PDB ID: 3BVP ) and γδ resolvase ( PDB ID: 2GM5 ) . In Figure 7B we model this pre-activated Hin tetramer intermediate based on the TP901 integrase tetramer ( Yuan et al . , 2008; Heiss et al . , 2011 ) . The quaternary changes accompanying the remodeling of the dimers into the pre-activated tetramer result in a compact structure that more readily fits between the Fis dimers on the enhancer , similar to the DNA-cleaved tetramer model ( see also Video 2 ) . 10 . 7554/eLife . 01211 . 024Video 2 . Assembly of the Hin invertasome and DNA exchange by subunit rotation within the invertasome structure ( Figure 7A–E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01211 . 024 During the modeled conformational changes from pre-activated to DNA-cleaved tetramer the position of the Hin DBDs relative to the Fis dimers undergo relatively small changes ( Figure 7B , C; Video 2 ) . However , the position of Hin helix-B changes substantially relative to the enhancer . Initially , helix-B is at a ∼50° angle with respect to the enhancer DNA with its C-terminal end ( Lys51 ) near the phosphate backbone and the remainder of the helix oriented away ( Figure 7B ) . Further movement of the catalytic domains driven by coulombic forces between the helix-B region and the enhancer DNA will clamp the basic helix-B region onto the enhancer DNA , locking the subunits into the active conformation competent for DNA cleavage and subunit rotation ( Figure 7C ) . These DNA interactions by the catalytic domain of Hin subunits bound to hix1L and hix2R augment the relatively weak interactions between the DBDs of the same subunits and Fis to stabilize the active invertasome structure . Our model for Hin invertasome assembly reveals how the Fis/enhancer element functions to regulate recombination at multiple levels . Initially , Fis-Hin interactions , together with the propensity of plectonemically supercoiled DNA to form branched DNA structures containing 2 ( − ) nodes , localize the two hix-bound Hin dimers at the enhancer . As Fis-Hin interactions have not been observed by standard solution assays , conformational energy from DNA supercoiling appears essential to promote synapsis . Formation of the active Hin tetramer could then be driven by mass action forces . For example , dynamic scissor-like movements between subunits of each dimer may transiently expose the hydrophobic surfaces of the apposing dimer interfaces to initiate remodeling into the tetramer . Conformational energy from the Fis/enhancer segment may also be harvested to drive tetramer assembly forward . This energy could be transduced from the twist deficit present in the negatively supercoiled DNA and/or by the mobile Fis β-hairpin arms to facilitate compaction of the two dimers into the tetramer . Spring-like mechanisms contributing to assembly of active recombination complexes also have been proposed for the Tn10 transpososome ( Chalmers et al . , 1998 ) . Finally , attractive electrostatic forces between the helix-B region and the enhancer DNA promote the final conformational change into the active tetramer structure . In our invertasome model the supercoiling energy in the looped DNA between the enhancer and hix sites drives the clockwise rotation of the top pair of subunits relative to the static bottom subunit pair that is fixed onto the enhancer ( Figure 7D , E; Video 2 ) . A single clockwise rotation of subunits together with their linked DNA strands will result in inversion of the DNA between the hix sites with an accompanying loss of four supercoils , as shown for Hin- and Gin-mediated DNA inversion ( Kanaar et al . , 1988; Merickel and Johnson , 2004 ) . When the distance between a hix site and the enhancer is short , as in the native configuration in the Salmonella chromosome ( ∼100 bp between hixL and the proximal Fis binding site ) , the small loop will inhibit additional subunit rotations because of torsional strain generated from multiple DNA windings ( illustrated in Figure 4D ) . However , additional rotations can occur under conditions where the loop is expanded by release of the Fis/enhancer from the Hin tetramer during the reaction or with substrates containing long segments of DNA between hixL and the enhancer . The structures of the resulting knotted DNA products are fully consistent with DNA exchange initiating within our invertasome model ( Kanaar et al . , 1990; Heichman et al . , 1991; Crisona et al . , 1994; Merickel and Johnson , 2004 ) . Thus , the multiple connections stabilizing the Hin tetramer onto the Fis/enhancer element function to restrict processive subunit exchanges . We experimentally demonstrate in this work that the final chemical step in the reaction , Hin-catalyzed DNA ligation , can efficiently occur when Fis and Hin remain physically connected ( Figure 6F ) , implying that the invertasome normally remains intact over the course of the entire reaction . To summarize , we propose a multistep pathway for assembly of the recombinationally-competent Hin invertasome . Step 1 begins by the association of inactive Hin dimers bound to each of the hix sites with Fis dimers bound at the ends of the enhancer segment at the base of a supercoiled DNA branch ( Figure 7A ) . In step 2 the localized Hin dimers are reconfigured into a pre-activated tetramer ( Figure 7B ) . We propose this occurs via simultaneous opening of the dimer interfaces , which then transition into the tetramer in a process that may be facilitated by energetic forces transmitted by the Fis/enhancer segment . Formation of the initial tetrameric structure positions the C-terminal ends of helix-B towards enhancer DNA . In step 3 , attractive electrostatic forces then ‘pull’ the basic helix-B region of the catalytic domains towards the enhancer DNA , clamping it against the enhancer segment between the two Fis dimers and completing the assembly of the tetramer that is competent for DNA cleavage and exchange ( Figure 7C–E ) ( Video 2 ) . A hallmark of reactions promoted by the invertase/resolvase subfamily of serine recombinases is the formation of supercoiling-dependent , topologically-defined , higher-order synaptic complexes that are responsible for uniquely specifying the recombinant product . Whereas the invertasome structure utilizes a remote enhancer to align the recombination sites appropriately for inversion , resolvases assemble tightly interwrapped synaptosomes containing non-catalytic resolvase subunits bound to extended recombination sites , sometimes together with auxiliary DNA bending proteins , to align recombination sites for deletion ( Grindley et al . , 2006 ) . Even though the molecular architectures of the complexes are very different , there are some striking similarities in the regulatory protein interfaces , which provide general insights into the control of this family of recombinases . In the Sin-catalyzed deletion reaction , synapsis is initiated by an interaction between regulatory dimers bound to each recombination site via residues in their DBD helix 1 ( Figure 8A ) ( Mouw et al . , 2008 ) . This interface can be related to the Fis-Hin interaction involving helix I of the Hin DBD that initiates invertasome formation . As with Hin , the DBD interaction by the Sin regulatory dimers is dispensable in the context of hyperactive mutants that promote recombination in the absence of the regulatory sub-sites . A second protein interface required for recombination by Sin , as well as the γδ/Tn3 resolvases , occurs between regulatory and catalytic subunits and can be related to the helix-B—enhancer DNA interface in the Hin invertasome . For Sin this interaction involves residues Phe52 and Arg54 that are located immediately adjacent to helix-B ( Figure 8B ) ( Mouw et al . , 2008; Rowland et al . , 2009 ) . For γδ/Tn3 resolvases a cluster of residues ( Arg2 , Arg32 , Lys54 , and Glu56; Figure 8B ) comprising the so-called 2–3′ interface between regulatory and catalytic dimers is important for assembly of the active synaptosome ( Hughes et al . , 1990; Murley and Grindley , 1998; Burke et al . , 2004; Olorunniji et al . , 2008 ) . The surface of the γδ resolvase tetramer exhibits a basic patch that contains some of the 2–3′ interface residues but whose location is shifted from the basic region on Hin ( see Figure 7—figure supplement 2 ) . Like Hin , the interaction between regulatory and catalytic subunits in the resolvase reactions is dispensable in the context of strong hyperactive mutants that promote indiscriminant recombination ( Burke et al . , 2004; Olorunniji et al . , 2008; Rowland et al . , 2009 ) . Taken together , the Hin , Sin , and γδ/Tn3 results imply that protein-protein or DNA-protein forces acting on this region of the catalytic domain may not only stabilize the active recombination complex but promote remodeling of this subfamily of serine recombinases into their active tetrameric conformation . Site-directed mutagenesis of the hin gene cloned into pET11a ( Merickel et al . , 1998 ) was performed using the QuikChange method . Native wild-type and mutant Hin preparations were obtained as described ( Heiss et al . , 2011 ) . Homogeneous disulfide-linked Hin-M101C dimers ( Figure 6B ) were prepared by incubation with 10 mM oxidized DTT overnight at 4°C , followed by passage through a Thiopropyl Sepharose 6B column ( GE Healthcare Life Sciences , Pittsburg , PA , USA ) . Fis purification was described in Stella et al . ( 2010 ) . In vivo inversion rates were evaluated as described using inversion tester strain RJ3635 ( Heiss et al . , 2011 ) , except that the strain also contained additional lacIqs-D274N copies on a pACYC184-derived plasmid to reduce basal Hin expression . In vitro DNA cleavage and inversion reactions were performed as previously described ( Haykinson et al . , 1996 ) using pMS551 ( unless otherwise stated ) , which contains the native hixL-enhancer spacing ( Figure 1—figure supplement 1 ) . For single-round knotting experiments , Hin and Fis were incubated with pMS551 and pMS634 under 30% ethylene glycol , Mg2+-free conditions for 5 min to accumulate cleaved synaptic complexes , and an aliquot representing the DNA-cleaved sample was quenched with 1% SDS . The remainder of the reaction was then diluted ≥threefold in 37°C buffer containing no ethylene glycol and 10 mM MgCl2 and incubated for 1 min to allow for DNA ligation . DNA knots were resolved in 0 . 84% agarose gels in Tris-phosphate-EDTA buffer after nicking with Nt . BsmA1 or DNase I in the presence of 200 µg/ml ethidium bromide . ‘Knots/cleavage reaction’ was calculated by dividing the percent knotted molecules by the percent cleaved molecules prior to ligation . Fis-Hin crosslinking was performed essentially as described in Dhar et al . ( 2009a ) . Typically , Hin and Fis were incubated with pRJ2372 under ethylene glycol , Mg2+-free DNA-cleavage conditions for 10 min prior to crosslinking . Crosslinking with 0 . 4 mM AMAS ( or GMBS ) or BMOE ( Pierce-Thermo Scientific , Rockford , IL , USA , dissolved at 10 mM in DMSO ) was performed for 30 s and then quenched with 20 mM lysine pH 7 . 5/20 mM DTT/0 . 4% diethyl pyrocarbonate ( DEPC ) or 20 mM DTT/0 . 4% DEPC , respectively . After precipitation with ethanol , the DNA was digested with EcoR1 and BamH1 ( BamH1 removes an interfering DNA band ) and radiolabeled with α-32P-ATP using Klenow . The products were then subjected to SDS-PAGE and phosphorimaging . Plasmid substrates were also used where a single Hin protomer is labeled after DNA cleavage by placing the EcoR1 site immediately adjacent to three of the four hix sites ( see Figure 1—figure supplement 1 ) . Placement of the EcoR1 site next to the hix site prevents radiolabeling by Klenow because of interference by the covalently-bound Hin ( Dhar et al . , 2009a ) . Reduced Hin or Fis cysteine mutants ( 100 µg ) were batch chromatographed on Heparin-Sepharose ( GE Healthcare Life Sciences ) to remove reducing agent ( 20 mM TCEP ) and incubated with a 10-fold excess of FeBABE ( Pierce-Thermo Scientific or Dojindo Molecular Technologies , Inc . , Rockville , MD , USA ) overnight at 4°C in 20 mM HEPES ( pH 7 . 5 ) , 1 M NaCl , 0 . 1 mM EDTA , and 20% glycerol . Free FeBABE was removed by passage through a Zeba 7K MW spin column ( Pierce-Thermo Scientific ) . The FeBABE-coupled proteins were added to standard Hin cleavage reactions , and DNA scission activated after 10 min incubation at 37°C by addition of 4 mM ascorbic acid for 5 s followed by 4 mM H2O2 for 30 s . After ethanol precipitation , DNA scission sites were detected by primer extension using 5′-32P-labeled primers ( top strand beginning 49 bp upstream or bottom strand beginning 62 bp downstream of the Fis core sites ) and Vent ( exo− ) DNA polymerase ( New England Biolabs , Ipswich , MA , USA ) ( Miller et al . , 1996 ) . Primer extension products were precisely mapped on 8% polyacrylamide-7M urea gels , alongside sequencing ladders generated with the same labeled primers using the Sequenase Quick Denature Plasmid Sequencing Kit ( USB-Affymetrix , Santa Clara , CA , USA ) . Hin structural models of the DNA-bound dimers ( based on PDB ID: 1GDT , Yang and Steitz , 1995 ) , pre-cleaved tetramer ( based on PDB ID: 3BVP , Yuan et al . , 2008 ) , and cleaved tetramer ( based on PDB ID: 1ZR4 , Li et al . , 2005 ) combined with the Hin DBD-DNA structure ( PDB ID: 1IJW ) have been described previously ( Dhar et al . , 2009a , 2009b; Heiss et al . , 2011 ) . The enhancer DNA model was generated using the DNA rebuild module in 3DNA ( Lu and Olson , 2003 ) . DNA parameter files were compiled from structures of the central 21 bp of the Fis-DNA co-crystal ( PDB ID: 3IV5 , Stella et al . , 2010 ) for the two flanking Fis binding sites together with the native sequence for the intervening and flanking DNA generated by using mean DNA parameters from the protein-bound DNA library , which has an average helical twist value of 34 . 2° ( Olson et al . , 1998 ) . The Fis dimers were then aligned onto their binding sites based on the co-crystal structure . The Hin DNA-cleaved and pre-cleaved tetramer models were manually docked onto the Fis-bound enhancer such that the Fis β-hairpin arms were optimally positioned with Arg154 and Leu155 on the Hin DBDs and consistent with the ( −2 ) topology of a branched DNA on negatively supercoiled DNA . The model of Hin dimers associated with the enhancer was generated by independently docking Hin dimer-hix models onto the enhancer using the DNA-cleaved tetramer model as a guide to position the Hin DBD relative to Fis . Morphed intermediates ( Video 2 ) were generated using the Yale Morphing Server ( Krebs and Gerstein , 2000 ) . All structural figures were generated in PyMOL ( DeLano , 2008 ) ; surface electrostatic calculations utilized the APBS plug-in with a monovalent ion concentration of 0 . 15 M ( Baker et al . , 2001 ) .	Many processes in biology rely on enzymes that break both the strands in a DNA molecule , then rearrange the strands , and finally join them back together in a new configuration . These recombination reactions can , for example , change the positions of genetic elements such as enhancers and promoters within the DNA molecule and , therefore , influence how a given gene is expressed as a protein . Cells need to be able to control recombination reactions because they can lead to leukemia and lymphomas if they go wrong . The enzymes that catalyze these recombination reactions are called recombinases . One type of recombinase binds to specific sequences of DNA bases and uses an amino acid in the enzyme–usually serine or tyrosine–to break and rejoin the DNA strands . Recombination reactions require the assembly of complexes containing many proteins bound to DNA . Tyrosine recombinases form relatively simple protein-DNA complexes , and these have been studied in detail . Serine recombinases , on the other hand , form more elaborate protein-DNA complexes , and much less is known about these . Now McLean et al . have unraveled the mechanism that a serine recombinase called Hin uses to reverse the direction of a stretch of chromosomal DNA in the bacteria Salmonella enterica . Inverting this stretch of DNA–which contains about 1000 base pairs–changes the position of a gene promoter that is responsible for the production of flagellin , which is the protein that enables the bacterium to move . This is one of the tricks that Salmonella uses to evade the immune system of its host . Previous research has established that four Hin subunits and two copies of a protein called Fis are needed to invert this stretch of DNA: two Hin subunits bind to each of the two hix recombination sites , and the Fis proteins ( which are dimers ) bind to each end of an enhancer that is located between the hix sites . A protein called HU then causes the DNA to bend and form a loop , and the four Hin subunits and the two Fis dimers all come together at the enhancer to form a structure called the invertasome where the recombination reaction occurs . All four DNA strands at the crossover point are broken as a result of a near simultaneous attack by the catalytic serine amino acids in the Hin subunits . One pair of Hin subunits–and the two DNA strands attached to them–then rotate by 180 degrees around the other pair of Hin subunits . This means that the stretch of DNA between the hix sites is inverted when the DNA strands are rejoined at the end of the reaction . Enhancers often regulate transcription and other reactions from a distance . McLean et al . reveal how an enhancer of a DNA recombination reaction works . The pairs of Hin subunits that initially bind to the DNA are not catalytically active , but when they are brought together by the enhancer and form a tetramer , they become active . Two of the Hin subunits are clamped onto the enhancer by the Fis dimers and by directly interacting with the enhancer DNA , but the other two ( and the DNA strands attached to them ) are free to rotate within the tetramer . In the Salmonella chromosome the enhancer is located close to one of the hix sites ( ∼100 base pairs away from it ) , so the length of the DNA between the enhancer and hix site physically limits the number of Hin subunit rotations to just one .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "biochemistry", "and", "chemical", "biology" ]	2013	Multiple interfaces between a serine recombinase and an enhancer control site-specific DNA inversion
Intravital microscopy can provide unique insights into the function of biological processes in a native context . However , physiological motion caused by peristalsis , respiration and the heartbeat can present a significant challenge , particularly for functional readouts such as fluorescence lifetime imaging ( FLIM ) , which require longer acquisition times to obtain a quantitative readout . Here , we present and benchmark Galene , a versatile multi-platform software tool for image-based correction of sample motion blurring in both time resolved and conventional laser scanning fluorescence microscopy data in two and three dimensions . We show that Galene is able to resolve intravital FLIM-FRET images of intra-abdominal organs in murine models and NADH autofluorescence of human dermal tissue imaging subject to a wide range of physiological motions . Thus , Galene can enable FLIM imaging in situations where a stable imaging platform is not always possible and rescue previously discarded quantitative imaging data . In recent years , a number of fluorescence imaging techniques such as fluorescence lifetime imaging microscopy ( FLIM ) have allowed researchers to visualize not only the structure but also the activity and function of molecules in living cells and tissues . Genetically-expressed Förster resonant energy transfer ( FRET ) -based biosensors enable researchers to probe signalling events in native tissues ( Conway et al . , 2017; Ellenbroek and van Rheenen , 2014; Nobis et al . , 2018 ) where they can provide spatio-temporal information about drug target response in a tumour ( Conway et al . , 2018 , 2014; Hirata et al . , 2015; Nobis et al . , 2017 , 2013 ) or dynamic signalling events in migrating cells ( Mizuno et al . , 2016 ) . Time resolved imaging of NADH autofluorescence ( Blacker and Duchen , 2016; Skala et al . , 2007 ) can be used to probe the metabolic state of cells and multispectral imaging ( Patalay et al . , 2012 ) has been investigated for the diagnosis of dermatitis and malignant melanoma ( König , 2012 ) , among other applications . Hyperspectral imaging , time resolved imaging in multiple spectral channels , can be used to extract microenvironmental information from autofluorescence ( Cutrale et al . , 2017 ) . These techniques depend on measuring small changes in the fluorescence signal , such as a small change in lifetime or change in spectral properties . Consequently , more signal is required to determine the parameters of interest , often necessitating relatively long integration times . This requirement has proved to be a significant constraint to the uptake of FLIM in intravital microscopy , where physiological motion due to , for example peristalsis , respiration or the heartbeat can induce significant motion during the image acquisition . This motion can often be tolerated in intensity-based imaging where acquisition times are short . However , when an image must be integrated over tens or even hundreds of seconds , sample motion rapidly renders the image unintelligible . While physical restraints such as tissue clamping or the application of negative pressure , may be used to limit the sample motion to a degree , this approach is not always effective to the extent required for high resolution microscopy and , in some cases , can compromise the sample integrity . Given the increasingly wide application of both intravital and FLIM imaging , there is a growing need to enable the functional readouts provided by FLIM even in the presence of physiological motion ( Conway et al . , 2014 ) . Here , we describe a motion blurring compensation approach using image-based realignment that can be applied directly to data acquired on existing commercial and clinical FLIM and conventional fluorescence microscopy systems in two and three dimensions in post processing . FLIM is most commonly implemented using laser scanning microscopy ( LSM ) . To acquire sufficient number of photons , an image is constructed by integrating the photon signal over many frames ( passes over the scan area ) . A typical FLIM image may take several minutes to acquire and so is susceptible to motion blurring from physiological motion . Several image-based approaches have been used to correct for sample motion in intensity-based time lapse data without additional measurements in the context of LSM time series acquisitions . ( i ) Where the motion is slow relative to the frame rate , ( e . g . a slow drift ) , rigid body image registration ( Thévenaz et al . , 1998 ) or feature-based registration such as Scale Invariant Feature Transform ( SIFT ) -based algorithms have been applied to correct for the motion , ( ii ) When the motion is intermittent , frames captured during the motion may be automatically detected and excluded from the time series ( Soulet et al . , 2013 ) , and ( iii ) When the motion is fast relative to the frame rate , each frame will appear distorted as the sample moves while the laser is scanned across the field of view . In this case , more advanced approaches that model the intra-frame motion using methods such as Hidden-Markov-Models ( Dombeck et al . , 2007 ) , the Lucas–Kanade framework ( Greenberg and Kerr , 2009 ) , or algorithms based on Lie groups ( Vercauteren et al . , 2006 ) have been employed . These techniques have not , to date , been applied to FLIM data . This is because during FLIM acquisition , histograms of photon arrival times are typically accumulated to produce a single image . In this approach , blurring due to sample motion is ‘baked into’ the data and thus cannot be compensated . More recently , however , improved device-computer bandwidth and storage have enabled the recording of individual photon arrival times and markers associated with the scan frame and line clocks ( Becker et al . , 2006 ) . Most modern commercial time-correlated single photon counting ( TCSPC ) FLIM systems support this mode , often by default . Here , we describe an approach whereby we reconstruct each frame from this time-tagged FLIM data and determine the motion both between and within each frame using an approach based on the Lucas-Kanade framework ( Baker and Matthews , 2004 ) . We have implemented these algorithms in a new open source package , Galene , which can be used to correct for motion in two- and three-dimensional FLIM data collected using widely deployed commercial systems . We first evaluate the range of motions that can be effectively compensated using simulated data and compare the performance of Galene’s core motion correction algorithms with open source and commercially available motion correction tools . We then validate our approach using intravital imaging of a number of FRET biosensors in vivo in a murine system and in clinical applications by imaging autofluorescence of human skin . While the main focus of this manuscript is motion correction of time resolved data , we show that Galene may also be used to correct conventional fluorescence microscopy data using intravital 3D multispectral imaging data of labelled immune cells in the murine lymph node highlighting its wider application for intravital imaging applications . The motion correction procedure is illustrated schematically in Figure 1 and outlined in Video 1 . We acquire FLIM data in a time-tagged time-resolved ( TTTR ) mode whereby each photon arrival time is recorded alongside frame and line markers , which allow us to locate each photon within the image . From these data , we first reconstruct the intensity of each constituent frame , or , in the 3D case , stack of frames ( Figure 1A–C ) , that make up the image . We use these intensity data to determine the sample motion during the image acquisition relative to a reference stack ( Figure 1D–H ) . For each stack , we first perform a fast , rigid realignment using a 3D generalisation of the phase correlation method ( Foroosh et al . , 2002 ) . This corrects for coarse sample displacements but cannot correct for sample motion during the stack , which leads to distortions , rather than simply displacement of the stack . To estimate the displacement of the sample during the stack acquisition , we use a fitting approach where we account explicitly for the raster scan pattern used to acquire the data following the approach of Greenberg and Kerr ( 2009 ) . The microscope takes a finite time to scan over the stack . We assume that the sample moves linearly between a series of initially unknown two- or three-dimensional displacements spaced equally through the scan duration . For a given set of displacements , we know where the sample was when the microscope acquired each pixel , and we can thus reconstruct the undistorted stack by three-dimensional interpolation and compare this reconstruction to the reference stack . By estimating the motion at a number of points across the image , we can account for motion in both the fast-axis , which appears as a ‘wave-like’ pattern in the data , and in the slow axis , which appears as compression or expansion of sections of the image . We can then determine the sample motion during the stack acquisition by finding the set of displacements which minimises the difference between the corrected stack and the reference stack using a trust-region non-linear optimisation algorithm . This approach requires that we compute the Jacobian of this error function , that is the gradient of the difference between the two stacks at each pixel with respect to the unknown displacement parameters . To perform this optimisation efficiently we use a variant of the inverse-compositional Lucas-Kanade algorithm ( Lucas and Kanade , 1981 ) , used extensively in image registration applications ( Baker and Matthews , 2004 ) ; its key insight is that , rather than calculating the gradient of the interpolated stack at every iteration , a computationally expensive procedure , we can instead use the gradient of the reference stack , which is of course invariant ( Baker and Matthews , 2004 ) . This optimisation yields an estimate of the sample motion during the scan . With this sample displacement information , the FLIM image can be reconstructed accounting for the motion , reassigning each photon arrival to the correct pixel producing a distortion-free image ( Figure 1I–L ) . We use the displacement information to determine the effective dwell time in each pixel ( which will vary across the image due to the sample motion ) . This information is stored alongside the corrected FLIM data and may be used when displaying intensity-merged FLIM data . To determine the range of amplitudes and frequencies of sample motion that can be reliably corrected , we used Monte Carlo simulations of FLIM data in the presence of sample motion ( Figure 2A ) . We compare these results with reference to average values for physiological events known to impair image acquisition; the heartrate ( red arrow , 350 bpm , ~6 Hz ) and respiratory rate ( blue arrow , 60 bpm , 1 Hz ) of an adult mouse anaesthetized under ~1% isoflurane ( Ewald et al . , 2011 ) are shown . Figure 2B shows the average correlation coefficient , a measure of how well the correction has performed , for a range of frequencies and amplitudes of motion relative to the field of view ( FOV ) size . We observed that the system could compensate for motion parallel to the fast axis ( 0° ) more effectively than motion parallel to the slow axis ( 90° ) ; this is unsurprising as motion in the slow axis will result in whole lines of the sample being sampled either twice or not at all . Figure 2C–G illustrate several example alignment results with different motion conditions; the red dots indicate the location on the frequency-magnitude plot shown in Figure 2B for each condition . Figure 2C and D show , respectively , slow ( 1 . 5 Hz ) and fast ( 6 Hz ) motion with the same magnitude of ~10% of the FOV aligned with scanner fast axis . Here , we can effectively correct for the motion as shown in the realigned images and estimated displacements plotted with the simulated displacements . Figure 2E shows fast motion at 45° to the fast axis; again , we can effectively compensate for this motion although the resultant realigned image is marginally degraded . Figure 2F shows a larger ( higher magnitude ) motion , approximately ~20% of the field of view at 45° to the fast axis . In this case , we are not able to effectively compensate for the motion , and the realigned image is significantly degraded . We note that a similar motion along the fast axis could be corrected ( see matching point on Figure 2Bi ) . Figure 2G shows a very large ( ~30% of the field of view ) motion along the slow axis , which we are unable to correct . We note that the range of amplitudes and frequencies that Galene is able to correct covers a broad range of physiologically relevant motions commonly observed during functional intravital imaging and will therefore be of use in a range of in vivo imaging applications . We compared the core motion correction algorithm used by Galene with three open source motion correction packages using intensity-only data . We used two ImageJ plugins , StackReg , implementing a rigid registration algorithm ( Thévenaz et al . , 1998 ) and ‘Linear Stack Alignment with SIFT’ , an approach based on the Scale Invariant Feature Transform ( Lowe , 2004 ) . We also evaluated the python package SIMA ( Kaifosh et al . , 2014 ) , which uses a Hidden Markov Model ( HMM ) -based approach ( Dombeck et al . , 2007 ) . We generated simulated time lapse intensity data with sample motion at 45° to the fast axis over a range of amplitudes and frequencies . We performed motion correction of these data with each software package and plotted the average correlation between each motion corrected frame and the reference frame as a function of frequency and amplitude of motion ( Figure 2—figure supplement 1 ) . StackReg and SIFT both correct for rigid transformations between each frame and so are unable to cope with the distortion produced by LSM with a moving sample; consequently , these packages are only able to correct for low frequency motion ( as , for example , encountered during slow sample drifts where the observed motion artefact can be approximated by a linear transformation ( for example , Figure 2—figure supplement 1B ) . SIMA uses a HMM model which uses information about the laser scan pattern and so is able to cope with a larger range of motion ( for example , Figure 2—figure supplement 1C , equivalent to a small displacement caused by the heartbeat ) . However , Galene is able to correct for significantly larger sample motions than SIMA ( compare Figure 2—figure supplement 1Aiv , SIMA and v , Galene ) , for example the larger motion shown in Figure 2—figure supplement 1D , corresponding to a larger motion induced by the heartbeat , and so will be useful in a broader range of intravital experimental conditions . We went on to evaluate the use of Galene in an intravital imaging setting . As demonstrated using the simulated data , the speed and direction of the motion has a significant effect on the extent to which we are able to correct the data . The critical parameter is , in fact , the relative speed of the motion with respect to the scan rate; a 5 Hz motion acquired with a frame rate of 1 Hz will appear to oscillate five times during each frame acquisition while the same speed acquired with a frame rate of 5 Hz will only appear to oscillate once per frame . Since the scan rate is generally a user controllable parameter , we imaged the same region at different scan rates to determine how this parameter affects the motion correction performance . We acquired images through surgically implanted titanium windows ( see schematic in Figure 3 ) that enable longitudinal imaging of abdominal organs ( Ritsma et al . , 2014 ) . Imaging abdominal organs through an optical window is challenging as they experience considerable motion as a result of physiological activity in nearby organs such as respiration and the heartbeat . Simply finding areas sufficiently stable to acquire FLIM images significantly limits the usable area of the window , and , in some cases , can render mice completely unusable . Using these windows , we imaged the pancreas in a genetically engineered mouse expressing a FRET biosensor for the small GTPase Rac1 ( Itoh et al . , 2002; Johnsson et al . , 2014 ) . We analyzed the FLIM data before and after motion correction by fitting each pixel to a single-exponential model as previously demonstrated ( Johnsson et al . , 2014 ) . We acquired FLIM-FRET images in two locations with different motion patterns , ( i ) where motion was dominated by the heartbeat and ( ii ) where motion is dominated by respiration and peristalsis ( see displacement traces shown in Figure 3Ai , ii ) at 700 , 1000 and 1400 Hz line rates . We performed an amplitude-spectrum analysis on the estimated displacements ( Figure 3B , C ) and found motion of ( i ) a frequency of 5 . 4 Hz and peak-peak amplitude of 27 μm ( approximately 10% of the FOV ) , corresponding to the heartbeat and ( ii ) slower motions with a range of frequencies from 0 . 2 to 2 . 2 Hz with an average peak-peak amplitude of 42 μm ( approximately 15% of the field of view ) , corresponding to respiration . We found that for the faster but smaller motion due to the heartbeat we were able to correct for motion equally well at all scan rates ( see Figure 3D , average correlation quantified in Figure 3Div ) . Note that for the same correction performance the correlation is reduced slightly at higher scan rates due to the lower signal to noise in each image . For the larger due to respiration and peristalsis , we found that we were only able to successfully correct for motion at 1400 Hz scan rates; partial correction was obtained at 1000 Hz and effectively no correction at 700 Hz ( see Figure 3E ) . Underscoring the simulated data results , these data indicate that for larger motions it may be helpful to acquire at a faster frame rate where possible . We went on to evaluate the effect of the relative alignment between the dominant direction of motion and the fast scanner axis . We imaged the pancreas with the motion aligned with the slow axis ( red ) and fast axis ( blue ) at 700 and 1400 Hz ( see Figure 3F and G respectively ) by rotating the microscope scan field . Figure 3F , Gv shows an angular histogram of the displacements showing the alignment with the scanner axes for the two cases and Figure 3F , Giv shows the average correlation between the realigned frames . At 700 Hz the realignment is significantly improved when the motion is aligned with the fast axis . At 1400 Hz , when the motion is slower relative to the scan rate , the motion is corrected equally well in either case . This highlights both the importance of using a fast scan rate where possible and , where there is a clear direction of motion , approximately aligning it with the scanner fast axis by rotating the scanner field of view . There are a number of user controllable options for performing the realignment . The first is the number of realignment points used across each frame of the image . The motion is interpolated linearly between these points across the image . In principle , using a larger number of points allows correction of higher frequency motions . We acquired 256 × 256 images of the pancreas and realigned the data using either coarse translation-only information determined using phase correlation or realignment with 3 , 5 , 10 , 20 and 40 points per image ( Figure 3A , quantified in Figure 3B ) . For images of this size , there was a slight improvement in the average correlation with increasing number of realignment points up to 20 points . Using 40 points , however , the correlation was slightly reduced . Using a very large number of realignment points can be detrimental; as the number of points increases , the number of pixels which constrain each point is reduced and eventually there is not enough information to accurately determine the motion at each point . In general , we found that using between 5–20 realignment points for 256 × 256 images and 10–20 points for 512 × 512 images was sufficient to obtain good correction over a broad range of conditions . As the signal to noise level in each frame of a FLIM image is often low due to the restricted count rate requirements of TCSPC imaging , a Gaussian smoothing kernel may optionally be applied to each frame before realignment to improve the realignment . This smoothing is only applied in the x ( fast- ) axis so that pixels are only convolved with those acquired immediately before and after . To evaluate the effect of the degree of smoothing , we realigned an image with low signal to noise with a range of smoothing kernel widths between 0 and 10 pixels ( Figure 3C , quantified in Figure 3D ) . We normally use the correlation between the smoothed images to reduce the effect of noise on the realignment result quantification . Here , however , we use the correlation between the unsmoothed images to allow us to compare the correlation between different smoothing kernels . We found a noticeable improvement in the realignment result using a kernel of with 2 or 4 pixels compared to no smoothing . At larger kernel sizes , the correlation gradually reduced; using excessively large smoothing kernels can reduce the quality of the realignment as the contrast in the image is reduced . We have found that using a smoothing kernel of 2–4 pixels works well over a broad range of conditions . We went on to use Galene to image Rac1 activity in the intestinal crypts . Rac1 regulates a diverse array of cellular events including the cell cycle , cell-cell adhesion , motility and differentiation ( Heasman and Ridley , 2008 ) and has been shown to be a key driver of Wnt-induced stem cell activation within the intestinal crypt ( Myant et al . , 2013 ) . The Rac1 biosensor contains an ECFP donor , which has a complex decay profile , dominated by contributions from two conformations with similar spectral profiles . Here , we have fitted the data to a complex-donor FRET model previously described ( Warren et al . , 2013 ) consisting of two contributions with different levels of FRET . Using global analysis , we determined the FRET efficiencies of the Rac1 GTP ( active ) and GDP bound ( inactive ) states to be E = 0 . 65 and E = 0 . 02 . By fitting the contributions of each component , we can estimate the fraction of active Rac1 biosensor in each pixel as shown in Figure 4A . Imaging of the intestine is further complicated by motion induced by peristalsis , wave-like contractions of the digestive tract which propel food through the intestine . The gut can be attached only gently to the window ( Ritsma et al . , 2012 ) as immobilizing a tract of the intestine can obstruct the bowel . The movement caused by peristalsis almost completely obscures the sample structure when imaging for even a few seconds ( Figure 4Ai ) . Video 2 shows the individual frames from the acquisition with and without correction and the accumulated time resolved image . When imaging crypts we noted that , alongside persistent smaller motion , occasional large , transient displacements occur where the crypt under observation moves completely out of the field of view ( shown in the displacement estimates , Figure 4Aiii ) ; this , of course , cannot be compensated . We automatically identify and remove these frames by applying a threshold to the correlation between the reference image and the best estimate of the corrected frame , in this case 0 . 8 ( Figure 4Aiv ) , discarding frames with lower correlation values . Figure 4A shows examples of frames which were successfully corrected ( vi before correction , vii after correction ) and frames which were excluded from the reconstructed image ( viii , ix ) . Using this procedure , we can successfully recover an undistorted image of the live crypt ( Figure 4Aii ) , and , by fitting to a complex-donor FRET model , estimate the fraction of the active biosensor in each pixel . Peristaltic motion continues even ex vivo when the intestine is maintained appropriately for live cell imaging . We imaged intestinal crypts in freshly excised tissue from the Rac1 and observed significant peristaltic motion which we could effectively correct using Galene , as illustrated in Figure 4B and Video 3 . To inhibit peristalsis for imaging , researchers often use scopolamine , a small molecule muscarinic antagonist that inhibits the contraction of the smooth muscle layer surrounding the intestine ( Wang et al . , 2008 ) . To compare this approach to image based correction , we imaged tissue with and without pre-treatment with scopolamine . To quantify the data , we first used phasor analysis ( Figure 4C , phasor analysis of image shown in Figure 4B ) to separate the biosensor fluorescence ( blue gate ) from the tissue autofluorescence ( red gate ) . We then manually segmented single cells and computed the average fraction of active Rac1 biosensor . Treatment with scopolamine effectively inhibited peristalsis; however , unexpectedly we also observed a significant activation of Rac1 in tissue treated with scopolamine compared to untreated tissue ( Figure 4D , quantified in E ) This activation was of a similar magnitude to that of tissue treated with phorbol myristate acetate ( PMA ) ( Johnsson et al . , 2014 ) , a potent small molecule activator of Rac1 . This effect was also observed when fixed tissue was stained with a Rac1-GTP specific antibody ( Myant et al . , 2013 ) ( representative images and quantification shown in Figure 4—figure supplement 1 ) . This interference with Rac1 signalling highlights the need to ensure pharmacological approaches to reducing sample motion do not affect the process under observation . In contrast , by using image based correction , these artefacts can be avoided when looking , for example , at subtle changes in Rac1 GTPase regulation which is known to drive stem cell activity in intestinal crypts ( Myant et al . , 2013 ) . Without motion correction , it is extremely difficult to identify subcellular compartments in a majority of the intestinal crypt data . After correcting for motion , however , we are able to robustly identify subcellular regions and structures in the data . We performed sub-cellular analysis of Rac1 activity in the basal and apical membranes of intestinal crypts with and without drug treatment ( Figure 4F ) . We observed a lower level of Rac1 activation the basal membrane compared to the apical membrane after application of PMA or scoloplamine , suggesting a potential negative regulation of Rac1 at the basal membrane . This may be consistent with the critical role of Rac1 in intestinal crypt patterning and differentiation ( de Santa Barbara et al . , 2003 ) . To benchmark the core motion correction algorithm used in Galene using real data , we exported the intensity of each frame from the intestinal crypt FLIM data shown in Figure 4 as time series data . Each frame has relatively low signal to noise and we found that the SIFT algorithm was unable to reliably extract feature points from the data to use for realignment . We therefore assessed the performance of StackReg , SIMA and Galene . Figure 4—figure supplement 2 shows the results of the realignment of intensity only versions of the intestinal crypt data acquired ( A ) in vivo through an optical window and ( B ) ex vivo . Due to the rapid motion observed in this image , the linear transformation used by StackReg is not able to adequately correct for the motion and the correlation between each frame and the reference frame is not improved compared to the unaligned data . SIMA provides an improvement in the image quality compared to the unaligned data and an increase in the average correlation between frames; however , a substantial motion artefact is still visible in the integrated image . Galene produces a significant improvement in the image quality over SIMA and StackReg . In line with the results of our simulations and in addition to enabling correction of time resolved data , Galene demonstrates a significant improvement in the realignment of both in vivo and ex vivo imaging data . In a recent study , we demonstrated that transient ‘priming’ using the pharmaceutical Rho kinase inhibitor Fasudil in a model of pancreatic cancer ( PC ) improved response to chemotherapy and impaired metastasis to the liver in an intrasplenic model ( Vennin et al . , 2017 ) . Src kinase has been shown to play a critical role in cell adhesion and proliferation in cancer ( Brunton and Frame , 2008 ) and is potentially an anti-invasive target in PC ( Evans et al . , 2012; Morton et al . , 2010b; Nobis et al . , 2013 ) . We previously observed a reduction in Src kinase activity in in vitro models of invasion and in end-point xenograft models of a primary tumour imaged using a skin flap technique . We therefore hypothesised that the reduction in the number of metastases after priming with Fasudil could be , in part , a consequence of a reduction in early adhesion events caused by disruption of Src activation . However , without correction for sample motion , the assessment of such early transient events using FRET ( illustrated schematically in Figure 5B ) is not possible in vivo , limiting our ability to quantify colonisation efficiency at this important stage . We used Galene to track Src activity in early adhesion events and so directly assess the effects of Fasudil during early attachment . We injected KPC cells expressing the Src-FRET biosensor ( Wang et al . , 2005 ) into mice implanted with abdominal imaging windows implanted on top of the liver ( see Figure 5A ) and imaged cells arriving in the liver 4 , 8 , 16 and 24 hr ( see timeline in Figure 5C ) after injection . We used a multispectral FLIM system to allow us to record the lifetime of the Src biosensor alongside microenvironmental context using a variant of the hyperspectral unmixing approach recently demonstrated ( Cutrale et al . , 2017 ) . Figure 5Di shows an example motion corrected merged intensity image in the three spectral channels used and Figure 5Dii and iii show the temporal phasor of the 525/50 nm channel and the spectral phasor respectively . Phasor gates associated with Src-FRET , hepatocytes , the vasculature and collagen were identified as shown and used to identify the associated region in the image ( Figure 5Div ) . Using the data from these regions , we created a pattern associated with each component and performed non-negative least squares to unmix the autofluorescence signal from nearby liver cells , blood vessels and the collagen network which have distinct hyperspectral signatures . To determine the lifetime of the Src biosensor , we identified regions containing the biosensor using phasor analysis and fitted the data to a single exponential model in the donor channel . We manually segmented single cells to determine the average lifetime per cell . The liver shows significant motion when imaged behind an optical window due to its proximity to the lungs and heart and attachment to the diaphragm . This frustrates attempts to acquire data for lifetime and hyperspectral unmixing as integration over even a few frames leads to significant image blurring as illustrated in Figure 5Ei . Figure 5Eii shows the same image after correction for motion with liver cells shown in grey , blood vessels in red , collagen in magenta . The lifetime of the biosensor is color-coded from blue ( low lifetime , low Src activity ) to red ( high lifetime , high Src activity , see schematic ) . Using Galene , we were able to reliably correct for motion in this context and so probe the activation of Src in relation to the true attachment state or spreading phenotype of cells during these early adhesion events in the liver . Here , we saw a significant increase in Src activity ( increase in Src biosensor lifetime ) after 8 hr , which was maintained up to 16 hr before plateauing after 24 hr upon spreading ( see Figure 5E , control situation ) . In line with our previous study ( Vennin et al . , 2017 ) , to mimic systematic ROCK inhibition or adjuvant therapy in the presence of circulating tumour cells , we treated mice with Fasudil at 12 hr intervals with three treatments before intrasplenic injection ( see timeline in Figure 5C ) . Mice treated with Fasudil exhibited a significantly reduced and delayed Src activity , in line with a delayed spreading phenotype , compared to those treated with the vehicle ( see Figure 5Fi , blue-green shift at 8 hr and Figure 5Fii , blue-green shift at 24 hr , quantified in Figure 5G ) , indicating that Src-dependent spreading and activation during the first attachment events in the liver are indeed impaired by treatment with Fasudil . These results demonstrate a new role of Fasudil priming in altering adhesion efficiency in secondary sites ( Rath et al . , 2017; Vennin et al . , 2017 ) . There is an urgent need to develop anti-metastatic treatments in PC and other metastatic cancer types ( Steeg , 2016 ) and functional intravital microscopy combined with Galene may help development of new strategies to monitor agents that affect this critical event preceding colonisation ( Ritsma et al . , 2012; Steeg et al . , 2011 ) . We note that aside from reducing the image quality , motion during acquisition can have a number of more subtle effects: ( 1 ) blurring of the biosensor fluorescence with background autofluorescence which may have a very different lifetime can artificially change the apparent lifetime of the biosensor , giving a misleading result and ( 2 ) motion can significantly distort the apparent shape of the cells . Both artefacts are observed here; we see a reduction in the apparent lifetime of the biosensor due to blurring with the low autofluorescence lifetime of surrounding liver cells and an artefactual elongation and apparent spreading of the cancer cell ( compare Figure 5Di–ii ) which may lead to an incorrect assumption about their attachment state ( see Figure 5B ) . Video 4 shows the accumulated frames before and after correction ( note that Video 4 shows the mean arrival time of all channels acquired , not just the Src biosensor donor ) . To evaluate the impact of motion correction on our ability to quantify Src activity in dataset , we analysed the lifetime of the uncorrected data in the same way ( Figure 5H ) . The blurring of the biosensor lifetime with the background autofluorescence leads to an overall reduction in the lifetime in all treatment conditions . As the magnitude of this effect varies greatly from cell to cell depending on the motion and the environment of each cell , we found a significantly higher degree of variance within each condition . This increased variability abolishes our ability to statistically distinguish the conditions , highlighting the importance of motion correction to obtain robust results in this context ( compare in Figure 5G–H ) . The lifetime of NADH and FAD autofluorescence can be used as a readout of metabolic activity ( Blacker and Duchen , 2016; Lakowicz et al . , 1992; Skala et al . , 2007 ) with potential applications in the detection of precancerous tissue . This autofluorescence signal has been investigated in a clinical context for diagnostic purposes ( König , 2012 ) using static ( Patalay et al . , 2012 ) , flexible ( König et al . , 2008 ) and handheld multiphoton ( Sherlock et al . , 2015 ) microscopes . Unlike optical window experiments , where the inverted imaging configuration and weight of the mouse largely constrains the sample motion to two dimensions , the movement observed imaging human skin in an upright configuration occurs isotropically in three dimensions and so correction for lateral motion alone is insufficient . We therefore demonstrate two approaches for handling motion in three dimensions: ( 1 ) real-time detection and compensation for axial sample motion when imaging a single plane and ( 2 ) 3D motion correction within a z-stack . We recently demonstrated ( Sherlock et al . , 2015 ) a handheld multiphoton microscope system that incorporates an active ( online ) axial motion compensation system that corrects for motion in z-axis in real time . The optical coherence tomography ( OCT ) -based correction system tracks sample motion perpendicular to the imaging plane in real time and adjusts the objective position to keep the selected plane in focus . We applied this system to collect short FLIM images of human epidermis; Figure 6Ai–iii shows the autofluorescence FLIM images ( i ) without motion compensation , ( ii ) with axial motion compensation alone and ( iii ) both axial and lateral motion compensation with Galene . The combination of active axial motion compensation and software-based lateral motion compensation is able to effectively remove the motion artefact observed . In a clinical setting , it is often desirable to obtain a 3D map of the autofluorescence lifetime to build up structural and functional information resolved into the strata of the skin , for example to quantify drug penetrance and delivery ( Roberts et al . , 2011 ) . We acquired 3D FLIM images of autofluorescence from the dorsal forearm of a volunteer using a commercial clinical FLIM instrument ( Leite-Silva et al . , 2016 ) . Capturing time-resolved depth stacks with sufficient photon counts can be a time-consuming process; a 30–50 μm stack may take 10–15 min to acquire . A certain degree of motion during this period is inevitable in live subjects and , since conventionally each frame is accumulated consecutively ( Figure 6Bi ) , this can lead to both blurring of individual images in the stack and displacement between images in the stack . To overcome this issue , we accumulated a number of scans over the entire stack ( Figure 6Bii ) . We then apply the same motion estimation approach as used in 2D data , extending the displacement points to 3D . Each stack is aligned to a reference stack . Motion in three dimensions during each stack acquisition can then be estimated and corrected . To enable correction of these large volumes , we used GPU computation and several algorithmic optimisations to reduce the processing time ( see Methods , Figure 7—figure supplement 1 ) . We acquired 50 μm stacks with 36 images , accumulating 10 stacks in total . Figure 6C shows the autofluorescence FLIM images before and after image-based 3D motion compensation . We see that we are able to track and correct for motion in three dimensions during the stack acquisition , obtaining undistorted deep-tissue data free from motion artefacts . These approaches may enable the use of autofluorescence imaging of other parts of the body more susceptible to sample motion in three dimensions such as the chest . Galene can also be used to correct for motion in time lapse fluorescence microscopy data , supporting data import and export from a number of common microscopy formats , OME-TIFF ( Goldberg et al . , 2005 ) and Imaris data formats . We applied Galene to intravital multi-channel 3D imaging of immune cells in the inguinal lymph node ( Suan et al . , 2015 ) and benchmarked its performance against drift correction in Imaris . Intravital imaging is a crucial tool in immunology , providing unique spatiotemporal information about the localisation , function and interactions between immune cells in their native environment . The organisation and migration of different classes of immune cells within the lymph nodes has been shown to play a critical role in the adaptive immune response ( Kastenmüller et al . , 2012 ) . For example , the production of antibodies in response to antigen re-exposure after vaccination depends on the interaction between CD4+ T cells and B cells at a number of specific locations in the lymph node . Tracking of the migration of these cells is therefore critical to understanding this process and its dysregulation . We imaged tdTomato labelled B cells ( red ) , Kaede CD4+ OT2 T cells ( green ) and subcapsular sinus macrophages labelled with Alexa680 ( magenta ) in a 150 μm z-stack through the inguinal lymph node ( SHG signal from fibrillar capsule , blue ) over 30 min ( Figure 7A ) . Over the time series , the macrophages and capsule are essentially static while there is significant migration of the CD4+ T cells . Over this period , substantial sample motion is observed; there is a slow drift due to slight shifts in the immersion liquid meniscus and faster displacements due to physiological motion , primarily respiration . These displacements are visible in the temporally colour-coded projections shown in Figure 7B , where early time points are shown in blue and late time points in red . We used the Imaris ‘spot tracking’ function to track the stationary macrophages and used their trajectory to correct for drift . This provides a reduction in the motion artefact , however the correction is not complete ( see Figure 7C inset , Video 5 ) , and there is a gradual loss in the correlation between the nominally static capsule signal over time ( Figure 7E and F ) . We then used Galene to correct for the motion based on the static capsule and macrophages and observed a significant improvement in the quality of the correction ( see Figure 7D , Video 5 ) . This will allow more accurate quantification of in vivo cell movement within this organ over long imaging periods even in the presence of physiological motion and sample drift . We developed Galene , an open source tool to correct for sample motion in two and three-dimensional intravital functional imaging data where many frames are integrated to provide sufficient signal to noise to accurately determine , for example , the activity of FRET reporters using FLIM . The source code for Galene is freely available alongside compiled executable for Windows and Mac at http://galene . flimfit . org/ . Galene uses a fitting approach that explicitly accounting for the raster scan pattern performed by laser scanning microscopes to determine sample motion both between frames and during each frame . We use the Lucas-Kanade framework to efficiently estimate the sample displacements following the approach of Greenberg and Kerr ( 2009 ) . Using the motion estimate , both FLIM and intensity-based data can be reconstructed , allowing quantitative analysis free from artefacts introduced by motion . To scale this approach to motion correction across 3D volumes where there may be thousands of displacement points , we implement two modifications . First , we take advantage of graphical co-processors to accelerate computationally intensive sections of the motion correction algorithm using to take advantage of graphical co-processors . We then exploit the structure of the optimisation problem to radically reduce the computational burden of estimating each optimisation step . These modifications reduce the time required to perform motion correction by a factor of 30 for typical volumetric datasets and open the door to online motion correction in the future . Using simulations of FLIM data , we explored the range of motions that can be effectively compensated with Galene . As expected , we found that motion perpendicular to the scanner fast axis – a user controllable parameter on most modern confocal microscopes – can be more effectively compensated than those along the slow axis . We found that motions covering a broad range of physiologically relevant motions , from respiration to the heartbeat can be effectively compensated when the magnitude of the motion was ~10% of the FOV , while significantly large motions , up to 30% of the FOV could be corrected at lower frequencies . We note that faster motions still may be compensated by using faster scan rate , for example by employing a resonant scanner . The difference in quality of correction of motions in the fast and slow axis is due to the frequency with which different positions are sampled during the acquisition . Acquiring a 256 × 256 image at 1000 Hz means that every x-position will be sampled every millisecond ( albeit at a different y position ) , while every y-position is sampled every 256 milliseconds . A motion purely in the fast scan ( x- ) direction will appear as a wave-like motion while a motion in the slow axis will appear to compress and stretch the image and , for sufficiently large motion , whole lines may be missed . This means that the data constrain the estimate of motion in the fast axis more strongly than in the slow axis . We benchmarked the core motion correction algorithm against existing open source packages by generating simulated fluorescence intensity time series data and found that Galene can correct for a significantly larger range of motions than these packages . We confirmed these results using intravital FLIM-FRET biosensor imaging data and multispectral intensity-based data to demonstrate that we can compensate for a range of physiological motions . We showed that we can correct for motion occurring both between frames and within frames which otherwise render the image unintelligible and thereby retrieve both cellular and subcellular resolution . This will enable researchers to apply functional imaging modalities in contexts previously inaccessible due to excessive motion and therefore extend acquisition times , providing higher signal to noise ( or using a lower excitation power to prevent tissue damage or photo-bleaching ) . Other recent applications of intravital FRET investigating , for example , RhoA ( Nobis et al . , 2017 ) , intricate Erk signalling propagation events from in the skin ( Hiratsuka et al . , 2015 ) or cancer stemness ( Kumagai et al . , 2015 ) , PKA in vascular permeability ( Yamauchi et al . , 2016 ) and stromal targeting in melanoma ( Hirata et al . , 2015 ) could benefit from this approach . We contrasted our approach with pharmacological inhibition of motion using scopolamine in sections of ex vivo intestine . We observed that , although scopolamine effectively inhibited sample motion , it induces a significant activation of Rac1 . This would compromise any study of Rac1 signalling in this context , illustrating the potential pitfalls of pharmacological approaches to inhibiting sample motion . We used Galene to image the activation of Src , a key regulator of metastasis in PC ( Erami et al . , 2016; Evans et al . , 2012; Nobis et al . , 2013; Vennin et al . , 2017 ) , via FLIM-FRET in cancer cells arriving and attaching in the liver in an intrasplenic metastasis model of pancreatic cancer using live longitudinal imaging with titanium windows . We previously demonstrated that ROCK inhibition with Fasudil improved response to standard-of-care chemotherapy and reduced metastasis of pancreatic cancer cells using intrasplenic and orthotopic models and have shown that Src is a key regulator of metastasis in PC ( Erami et al . , 2016; Evans et al . , 2012; Nobis et al . , 2013; Vennin et al . , 2017 ) . To date , our ability to use FLIM-FRET imaging to observe these early , first attachment events has been hindered by sample motion . The extensive physiological motion observed in the liver not only significantly degrades the image but also , by mixing the FRET signal with the autofluorescence background , introduces a significant artefact into the FLIM readout of the Src-FRET activity . Using Galene , we observe a significant delay in Src activity after priming with Fasudil at early time points , providing a potential mechanism for the significant reduction in metastatic colonisation observed in endpoint metastatic burden experiments ( Vennin et al . , 2017 ) . As highlighted above , this may have wider applications in other metastatic cancers where early adhesion events are difficult to study . Imaging human skin autofluorescence using two clinical multiphoton systems we found that , unlike murine optical window experiments , where the weight of the mouse tends to constrain sample motion laterally within the image plane , in a clinical context sample motion occurs isotropically in the x , y and z-axis . Motion in the z-axis cannot be corrected when imaging at a fixed focal plane as the sample physically moves out of the scan area . We demonstrated two methods for motion correction in three dimensions; we first combined Galene with a custom handheld multiphoton system incorporating axial motion correction ( Sherlock et al . , 2015 ) while imaging human skin . By combining online axial motion compensation with lateral motion using Galene we can effectively compensate for motion in three dimensions . In an alternative approach , we acquired z-stacks of human skin , frequently used , for example , in skin penetration studies ( Labouta et al . , 2011 ) . Here , integration times on the order of tens of minutes are not uncommon due to weak autofluorescence signal and excitation powers limited by safety considerations . We showed that , by generalising the 2D motion correction approach to estimate motion in a three dimensional raster scan , we can effectively correct for isotropic motion in 3D . This will enable longer and more accurate FLIM volume acquisitions and imaging in locations more susceptible to physiological motion such as the chest where respiration is a significant challenge . We have demonstrated that it is possible to robustly analyse motion corrected FLIM data using several methods which require a high signal noise levels and sample stability using data from two widely available commercial FLIM systems . Galene may be applied directly to data acquired on systems post capture and may even be applied retrospectively to existing data . This approach could be readily applied to confocal endoscopic systems ( Kennedy et al . , 2010; Siegel et al . , 2003; Sparks et al . , in press; Sun et al . , 2013 ) to allow TCSPC FLIM acquisitions in an intraoperative context , for example for tumour margin assessment ( Gorpas et al . , 2015; Wang et al . , 2017 ) . For such applications , this approach could readily be extended to enable real time correction of FLIM data , as the processing time is typically significantly shorter than data acquisition , even for high speed 3D applications . In principle , this system could be combined with a hardware-based lateral motion correction approach ( Sherlock et al . , 2018 ) , that can track large displacements to enable high fidelity correction when the microscope itself is moving . This would allow , for example , a macro scale FLIM map to be constructed by freely moving the imaging system across the sample . In addition to correction of time resolved fluorescence data , we also demonstrated that Galene can effectively correct for motion in fluorescence intensity time series data in both two and three dimensions using multispectral intravital imaging data . Galene will therefore allow researchers to apply time resolved functional imaging in a broader range of contexts , relaxing previous restrictions on sample stability and imaging duration and make use of data which would have previously been discarded , in vitro , in vivo and in the clinic . After computing the displacement estimates , the TTTR data are reconstructed into histogrammed FLIM data taking the estimated sample motion into account . Each photon arrival is assigned to a pixel coordinate ( x , y , z ) based on the frame , line and ( if they exist ) pixel markers in the dataset . The photon is then reassigned to the coordinate ( x-Dxt , y-Dyt , z-Dz ( t ) ) using the final displacement estimates for the current frame , where t is the macro time relative to the start of the frame . If a correlation threshold has been set , photons arriving during frames where the correlation coefficient is less than the threshold value will be discarded . The sample motion leads to an effective variable integration time across the image . To correct for this , we calculate the integration time by integrating the dwell time in each pixel across the image . This integration time image is saved alongside the data and used to correct the intensity merged FLIM images . This corrects for the variable integration time without altering the photon statistics in the data used for analysis . OME-TIFF data ( Goldberg et al . , 2005 ) is supported using the OME files C++ implementation ( https://github . com/ome/ome-files-cpp ) , a number of standard microscopy data formats are supported using libbioimage ( https://bitbucket . org/dimin/bioimageconvert/ ) . Imaris data are supported using a custom reader implemented in C++ using the HDF5 library . Motion corrected data can be saved to an OME-TIFF or Imaris file respectively . Monte Carlo simulations of 2D TTTR data distorted by sample motion were performed using a subset of a high SNR , motion-free intensity image Is of ex vivo pancreas ( shown in Figure 1 ) . The sample motion was set to a sinusoidal motion such thatDxt=Acos⁡θsin⁡2πftDyt=Asin⁡θsin⁡2πftwhere A and f are the sample amplitude and frequency respectively while θ is the angle of the motion with respect to the scanner fast axis . A TTTR event stream was simulated according to Algorithm 2 . All simulations shown in Figure 1 were performed using a frame size of 256 × 256 pixels . The pixel time and interline time were set such that line rate was 1 kHz with a duty cycle of 0 . 33 and the interframe time was set equal to the interline time , approximately matching the scan pattern of the Leica SP8 scanner . The intensity was scaled to produce an average count rate of 1 MHz and 50 frames were generated per image . Animals were kept in conventional animal facilities on a 12 hr day-night cycle and fed ad libitum . All experiments were carried out in compliance with the Australian code for the care and use of animals for scientific purposes and in compliance with Garvan Institute of Medical Research/St . Vincent’s Hospital Animal Ethics Committee guidelines ( ARA 13/17 , 16/13 , 15/29 ) . For in vivo Rac1 activity experiments , mice ubiquitously expressing the Raichu-1011X ECFP-SEYFP Rac1 biosensor ( Itoh et al . , 2002 ) , generated previously ( Johnsson et al . , 2014 ) , were used . Experiments conducted on healthy human subjects using the DermaInspect were performed with informed consent and approval from the University of Queensland Human Research Ethics Committee ( approval number 2007/197–2008001342 ) . Experiments conducted on healthy human subjects using the hand held multiphoton system were performed with informed consent and approval from Imperial College London ( approval number 14IC2364 ) . Primary KPC cancer cells isolated from Pdx1-Cre; LSL-KRasG12D/+; LSL-Trp53R172H/+ tumours ( Morton et al . , 2010a ) were engineered to express a Src-FRET biosensor ( Wang et al . , 2005 ) modified to replace ECFP with mTurquoise2 using a transposon system ( Vennin et al . , 2017; Wilson et al . , 2007 ) . KPC cells were cultured in DMEM ( Gibco ) supplemented with 10% FBS and 1% glutamine , penicillin/streptomycin in 5% CO2 . Sections of freshly excised and flushed duodenal tissue were treated with 1 μM ( - ) -Scopolamine-N-butylbromide ( scopolamine , Sigma-Aldrich , S7882 ) in PBS for 30 mins at 37°C or 200 nM phorbol myristate acetate ( PMA , Sigma-Aldrich , P8139 ) for 15 mins . The tissue was fixed in 10% buffered formalin solution overnight and embedded in paraffin using the swiss roll method ( Bialkowska et al . , 2016 ) . Cut sections were de-paraffinised using xylene and rehydrated in graded ethanol washes . Antigen retrieval was performed in citrate buffer ( S1699 , pH = 6 ) for 30 min at 99°C and allowed to cool to RT for another 30 min . Endogenous peroxidase activity was subsequently quenched in 1 . 5% H2O2 before the application of 10% normal goat serum ( NGS ) in protein block ( Dako ) for 1 hr at RT . Slides were incubated overnight at 4°C with primary antibody ( active Rac1-GTP , 1:400 , NewEast Biosciences ) in 10% NGS in protein block prior to applying secondary HRP-coupled anti-mouse antibody ( Envision ) . Detection was performed with diaminobenzidine ( DAB ) for 5 min and slides counterstained with haematoxylin . Slides were digitalised at 20 × magnification using a slide scanner ( AperioCS2 , Leica Biosystems ) . Data were analysed using QuPath ( Bankhead et al . , 2017 ) . DAB and haematoxylin optical densities were computed for each pixel using colour deconvolution and regions containing crypts were manually identified . Nuclei within these regions were identified automatically using watershed cell detection based on the haematoxylin counterstain . Cell regions were then segmented by dilating the nuclear detections by 5 μm . The average DAB optical density was computed for each cell . These values were then averaged across each sample for n = 3 mice . Galene is provided as an open source package with a graphical user interface and is available for download at https://galene . flimfit . org/ alongside user documentation , and is integrated directly into the FLIMfit analysis software ( Warren et al . , 2013 ) . This software may be directly applied to data acquired on commercial microscope systems . The source code is available under the GPLv2 license at https://github . com/flimfit/Galene ( copy archived at https://github . com/elifesciences-publications/Galene ) . The executables , manual and source code used in this manuscript are attached as Supplementary files . A tutorial screencast documenting the use of Galene is shown in Video 6 .	Understanding how molecules and cells behave in living animals can give researchers key insights into what goes wrong in diseases such as cancer , and how well potential treatments for these diseases work . A number of tools help us to see these processes . For example , fluorescent ‘biosensors’ change colour to tell us how active a particular protein is . This can indicate how well a drug works in different parts of a tumour . High resolution microscopy makes it possible to image events happening in single cells , or even specific parts of a cell . However , small movements like those due to the heartbeat or breathing can blur the images , making it difficult to study living animals . This is particularly problematic for images that take several minutes to capture . Warren et al . have now developed a new open source software tool called Galene . The tool can correct for small movements in images collected by a technique called fluorescence lifetime imaging microscopy ( FLIM ) . As a result , clear images can be captured in situations that were not previously possible . For example , Warren et al . watched cancer cells migrating to the liver of a mouse from the spleen over 24 hours , and , using a fluorescent biosensor , showed that a repurposed drug interferes with how well the cells can attach to the liver . In addition , Warren et al . used the software to take steady 3D images of human skin in a volunteer’s arm , which could be used to study drug penetration . Galene could help researchers to study a wide range of biological processes in living animals . The software can also be applied to existing data to clean up blurred images . In the future Galene could be further developed to work with the imaging techniques used during surgery . For example , surgeons could use it to help them find the edges of tumours .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "cell", "biology", "tools", "and", "resources" ]	2018	Removing physiological motion from intravital and clinical functional imaging data
Antibodies are critical components of the human adaptive immune system , providing versatile scaffolds to display diverse antigen-binding surfaces . Nevertheless , most antibodies have similar architectures , with the variable immunoglobulin domains of the heavy and light chain each providing three hypervariable loops , which are varied to generate diversity . The recent identification of a novel class of antibody in humans from malaria endemic regions of Africa was therefore surprising as one hypervariable loop contains the entire collagen-binding domain of human LAIR1 . Here , we present the structure of the Fab fragment of such an antibody . We show that its antigen-binding site has adopted an architecture that positions LAIR1 , while itself being occluded . This therefore represents a novel means of antigen recognition , in which the Fab fragment of an antibody acts as an adaptor , linking a human protein insert with antigen-binding potential to the constant antibody regions which mediate immune cell recruitment . The antigen-binding sites of human antibodies commonly adopt similar structures , with the light and heavy chains each providing three hypervariable loops that combine to form a surface that is complementary to the epitope . While the sequences of these complementarity determining regions ( CDRs ) are highly variable , five of the six CDRs ( L1 , L2 , L3 , H1 and H2 ) can be classified into a number of relatively small sets , with similar lengths and architectures , and their structures are predictable from sequence ( Chothia et al . , 1989; North et al . , 2011 ) . In contrast , the third CDR loop of the heavy chain ( CDR H3 ) is more structurally diverse , most likely due to its location close to the V ( D ) J recombination site ( Weitzner et al . , 2015 ) . Human antibodies typically have CDR H3 lengths of 8–16 residues ( Zemlin et al . , 2008 ) while mouse antibodies have CDR H3 lengths of 5–26 residues ( Zemlin et al . , 2003 ) . However , recent years have seen the discovery of antibodies with major differences from the norm , in particular due to changes in the length of the third CDR of the heavy chain . A set of antibodies with broadly neutralizing potential against HIV is one such example . Here , the third CDR loop of the heavy chain is elongated , allowing it to reach through the glycan shield that surrounds the gp120 protein to bind an otherwise concealed epitope ( McLellan et al . , 2011; Pancera et al . , 2013; Pejchal et al . , 2010 ) . Such antibodies are rare , making the induction of a broadly inhibitory response against HIV a major challenge ( Corti and Lanzavecchia , 2013 ) . In a more extreme example , while the majority of bovine antibodies have CDR H3 loops of around 23 residues , around 10% contain a highly elongated third CDR loop of up to 69 residues , containing a small disulphide rich domain ( Saini et al . , 1999; Wang et al . , 2013 ) . These domains adopt a conserved β-sheet structure that displays variable loops and are each presented on an elongated , but rigid β-hairpin ( Stanfield et al . , 2016; Wang et al . , 2013 ) . While it is clear that the additional domains play an important role in ligand binding , the remaining five CDR loops are also exposed and further studies are needed to see the contribution that they make ( Wang et al . , 2013 ) . A recent study identified a group of even more unusual human antibodies in malaria endemic regions of Africa ( Tan et al . , 2016 ) . These antibodies were discovered through their capacity to agglutinate human erythrocytes infected with different strains of Plasmodium falciparum , and they bind to a subset of RIFIN proteins . These RIFINs are displayed by the parasite on infected erythrocyte surfaces and are of uncertain function ( Chan et al . , 2014; Gardner et al . , 1998; Kyes et al . , 1999 ) . The antibodies show a remarkable adaptation with an intact 96 residue protein , LAIR1 , inserted into the third CDR loop of the antibody heavy chain . Indeed , LAIR1 was shown to be essential for the antibody to interact with RIFINs ( Tan et al . , 2016 ) . In this study , we reveal the structure of the Fab fragment of one of these antibodies , showing how LAIR1 is presented on the antibody surface and drawing conclusions about how this class of antibody can recognize its ligand . We expressed the two chains that make up the Fab fragment of antibody MGD21 ( Tan et al . , 2016 ) in a secreted form from HEK293 cells . This antibody has a kappa light chain ( VK1-8/JK5 ) and a heavy chain in which LAIR1 has been inserted into CDR H3 . This fragment was purified and crystallised , allowing a dataset to be collected to 2 . 52 Å resolution . The structure was determined by molecular replacement using LAIR1 ( Brondijk et al . , 2010 ) and the Fab fragment of antibody OX117 ( Nettleship et al . , 2008 ) as search models . This identified two copies of the MGD21 Fab fragment in the asymmetric unit of the crystal . A model was built for residues 2–211 of the light chain and 1–351 ( with 214–219 and 264–270 disordered ) of the heavy chain ( Figure 1 , Figure 1—figure supplement 1 , Figure 1—figure supplement 2 , Table 1 ) . The two Fab fragments adopt the same structure with a root mean square deviation of 0 . 26 Å ( calculated over 475 Cα atoms ) suggesting a highly ordered linkage between the variable domains of the antibody and the LAIR1 insert ( Figure 1—figure supplement 3 ) . The antibody sequence has three putative N-linked glycosylation sites , but of these ( light chain N30; heavy chain N61 and N242 ) only N242 shows electron density corresponding to an Asn-linked N-acetyl glucosamine , in a position distant from the LAIR1 insert . 10 . 7554/eLife . 27311 . 003Figure 1 . Structure of a LAIR1-containing antibody Fab fragment . ( A ) The structure of the Fab fragment . LAIR1 ( red ) is inserted into the third CDR loop of the heavy chain ( yellow ) through two extended linkers ( orange ) . The light chain is blue . The dashed orange link represents protein disordered in the structure . ( B ) The organization of the CDRs . The three CDR loops of the light chain and remaining two CDR loops of the heavy chain directly contact the LAIR1 insert or the linkers . Each of the CDR loops and its corresponding label is a shown in a different colour . ( C ) A disulphide bond between C93 of the light chain and C223 of the heavy chain stabilizes the interface ( cysteine residues are shown as sticks ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27311 . 00310 . 7554/eLife . 27311 . 004Figure 1—figure supplement 1 . Annotated sequence of antibody MGD21 and its alignment to germ line LAIR1 . ( A ) The sequence of the light chain of MGD21 with the CDR loops indicated by blue bars . ( B ) The sequence of the heavy chain of MGD21 aligned to that of germ line LAIR1 . Hexagons represent putative glycosyation sites with the red hexagon representing a site lost in MGD21 . Yellow circles mark sites in germ line LAIR1 known to affect collagen binding . Red stars represent cysteine residues that make a cross-chain disulphide bond . Residues in the heavy chain are labeled according to whether they derive from the V , J or LAIR1 genes . In both cases , Kabat numbers ( Dunbar and Deane , 2016 ) of residues are given above the sequence . In addition , all the CDRs , with the exception of the third CDR of the heavy chain , are labeled with their canonical class ( Martin and Thornton , 1996 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 27311 . 00410 . 7554/eLife . 27311 . 005Figure 1—figure supplement 2 . Electron density . ( A ) A view of the electron density , showing the 2Fo-Fc map in blue , contoured at 1 . 0σ . ( B ) A view of the electron density in the region in which the CDR loops of the heavy chain ( dark blue ) contact the LAIR1 insert ( red ) . ( C ) A view of the electron density where the CDR loops of the light chain ( magenta ) contact linker 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 27311 . 00510 . 7554/eLife . 27311 . 006Figure 1—figure supplement 3 . Crystal packing and order . ( A ) A superimposition of the two antibody molecules in the asymmetric unit , with chains A and B in yellow and C and D in red . ( B ) The structure of chains A and B shown in putty representation with putty thickness determined by B factor . The colour scale for blue to red is from B factors of 30 to 100 . DOI: http://dx . doi . org/10 . 7554/eLife . 27311 . 00610 . 7554/eLife . 27311 . 007Table 1 . Data collection and refinement statistics . The structure was determined from a single crystal . Values in parentheses are for highest-resolution shell . Rfree was determined using 1968 reflections ( 4 . 8% ) The structure is deposited with pdb code 5NST . DOI: http://dx . doi . org/10 . 7554/eLife . 27311 . 007Fab-MGD21Data collection Space groupC121 Cell dimensions a , b , c ( Å ) 169 . 8 , 86 . 5 , 104 . 0 α β γ ( ° ) 90 . 0 , 126 . 7 , 90 . 0 Wavelength0 . 92819 Resolution ( Å ) 81 . 90–2 . 52 ( 2 . 56–2 . 52 ) Total Observations131833 ( 5451 ) Total Unique40946 ( 2031 ) Rpim ( % ) 5 . 4 ( 67 . 8 ) Rmerge ( % ) 8 . 3 ( 88 . 5 ) Rmeas ( % ) 9 . 9 ( 112 . 1 ) CC1/20 . 992 ( 0 . 571 ) I/σ ( I ) 7 . 4 ( 1 . 0 ) Completeness ( % ) 99 . 8 ( 98 . 3 ) Multiplicity3 . 2 ( 2 . 7 ) Wilson B factor55 . 216Refinement Number of reflections40946 Rwork / Rfree21 . 9/26 . 7 Number of residues Protein1076 R . m . s deviations Bond lengths ( Å ) 0 . 01 Bond angles ( ° ) 1 . 25 All Atom clash score5 B factors All atoms71 . 53 Solvent63 . 12 Variable domains65 . 17 Constant domains74 . 29 LAIR1 insert73 . 70 Linkers94 . 71 Ramachandran plot Favored ( % ) 95 . 2% Allowed ( % ) 4 . 8% Disallowed ( % ) 0 . 0% The structure shows LAIR1 emerging from the CDR3 loop of the heavy chain and lying across the antigen-binding surface of the variable domains of the Fab fragment ( Figure 1A ) . The long axis of the LAIR1 insert is positioned with the β-strands aligned approximately perpendicular to the groove between the heavy and light chain CDRs and the insertion and linkers interact with , and occlude all five of the remaining CDR loops . The N- and C-termini of LAIR1 lie at opposite ends of its structure , necessitating long linkers between the sites from which CDR3 emerges from the antibody heavy chain and each terminus of the LAIR1 insert ( Figure 1A ) . The N-terminal linker ( linker 1 ) is 10 residues long and adopts a simple loop structure that joins the antibody variable domain to the N-terminus of the LAIR1 insert . The C-terminal linker ( linker 2 ) is longer at 34 residues and is more complex in structure . It extends out from the C-terminus of the LAIR1 insert before zigzagging back towards the insertion site in the heavy chain variable domain . It is stabilized by hydrogen bonds to the LAIR1 insert and to the antibody heavy chain as well as by a disulphide bond to C93 of the antibody light chain . The linkers of the LAIR1-containing antibodies sequenced to date are variable both in length and content , involving different parts of the intronic regions of the LAIR1 gene , or intergenic sequences of chromosome 13 ( Tan et al . , 2016 ) . The arrangement of these linkers , which radiate away from the remainder of the antibody , will in theory accommodate almost limitless variation in both length and sequence without disturbing the packing of LAIR1 against the variable domains of the antibody . The five CDR loops lacking the LAIR1 insertion are representatives of previously identified canonical classes ( Figure 1—figure supplement 2 ) ( Martin and Thornton , 1996 ) . However , a search using the Abcheck server ( Martin , 1996 ) identified seven unusual residues within the antibody structure; C91 , C93 , D97 and I106 from the light chain and Y28 , R34 and Q54 from the heavy chain , all within the CDR loops . In particular , C91 , C93 and D97 all lie in CDR3 of the light chain , perhaps facilitating its interaction with linker 2 . Indeed , the most unusual residue is C93 , which is found in only 0 . 096% of light chains , and is the residue that forms a disulphide bond with linker 2 ( Figure 1B ) . The heavy chain CDR H3 loop has a base that adopts the ‘kinked’ conformation ( Shirai et al . , 1999 ) , with the loop rapidly spreading to the two termini of LAIR1 . In previous structures of antibodies with extended heavy chain CDR3 loops , the remaining five CDRs of the antibody are exposed , with the potential to engage in antigen binding ( McLellan et al . , 2011; Stanfield et al . , 2016 , Wang et al . , 2013 ) . One of the remarkable features of the LAIR1-containing antibody is therefore the occlusion of large parts of each of the remaining five CDRs , with these loops each interacting directly with the LAIR1 insert and/or linkers ( Figure 1B , Table 2 ) . The degree of occlusion of the CDRs by LAIR1 was determined by accessibility to a 1 . 4 Å probe in the presence and absence of LAIR1 and the linkers . Each of these five CDR loops was partly occluded by the presence of LAIR1 or the linkers ( occluding 12 . 7% of the accessible surface area of CDR L1 , 18 . 3% of CDR L2 , 47 . 3% of CDR L3 , 34 . 7% of CDR H1 and 16 . 0% of CDR H2 ) . Indeed , both the first and second CDRs of the heavy chain directly contact the LAIR1 insert ( Table 2 ) . In addition , each of the three CDR loops of the light chain interacts with one of the two linkers , with interactions including a disulphide bond between C93 of the light chain and C223 of linker 2 ( Figure 1C , Table 2 ) . These interactions are replicated in both copies of the molecule in the asymmetric unit of the crystal . 10 . 7554/eLife . 27311 . 008Table 2 . A list of interactions between the LAIR1 insert and linkers that occupies the heavy chain CDR3 loop and the other five CDR loops of the antibody . DOI: http://dx . doi . org/10 . 7554/eLife . 27311 . 008CDR loopResidueGroup LAIR1 regionResidueGroupInteraction Light chain CDR1Q27Side chainLinker 2A222Main ChainHydrogen Bond Light chain CDR2Y49Side chainLinker 1L102Side chainHydrophobic Packing Light chain CDR2N53Side chainLinker 1S104Side chainHydrogen Bond Light chain CDR3C93Side chainLinker 2C223Side chainDisulphide Bond Light chain CDR3F94Main ChainLinker 2E227Side ChainHydrogen Bond Heavy chain CDR1N32Side chainLAIR1R134Side chainHydrogen Bond Heavy chain CDR2R57Side ChainLAIR1P109Main ChainHydrogen Bond The structure of MGD21 argues for a rigid association of the LAIR1 insert with the remainder of the antibody . Firstly , the structures of the two molecules of the antibody in the asymmetric unit of the crystal superimpose closely ( Figure 1—figure supplement 2 ) . It is unlikely that this is due solely to constraints from crystal packing as LAIR1 is anchored to the variable domains of the antibody through three fixed positions: the attachment sites of the two linkers , and the disulphide bond between light chain C93 and heavy chain C223 ( Figure 1C ) . In addition , each of the five CDR loops not baring a LAIR1 insertion makes direct interactions with either LAIR1 or the linker , through contacts found in both copies of the antibody in the asymmetric unit of the crystal . This will stabilize a tight association between LAIR1 and the antibody variable domains . As these antibodies can include multiple different light chains , and very different linkers ( Tan et al . , 2016 ) , these interaction will not be replicated precisely across the antibody family , but some variant of interaction between light chain CDRs and the linkers is likely . A comparison of the LAIR1 insert with that of the chromosomal copy of LAIR1 ( referred to below as germ line ) ( Brondijk et al . , 2010 ) reveals that no global structural changes have taken place ( root mean square deviation 0 . 43 Å for the 82 Cα residues ) ( Figure 2A , F ) . Indeed , the LAIR1 insertion in the MDG21 antibody differs in only 13 positions relative to the germ line sequence . Mapping these sites onto the structure reveals that they do not alter residues through which the LAIR1 insert interacts with the rest of the antibody ( Figure 2B ) . Presumably , the stable interaction between LAIR1 and the antibody has therefore come instead from adaptations to the CDR loops . In contrast , polymorphisms are mostly located on the surface of LAIR1 distal to the rest of the antibody and have the potential to alter its interaction with its original ligand , collagen , and with the RIFIN proteins which are the target of these antibodies . 10 . 7554/eLife . 27311 . 009Figure 2 . Structure and polymorphism in the LAIR1 insertion . ( A ) An alignment of germ line LAIR1 ( cyan ) with the antibody LAIR1 insertion ( red ) . ( B ) The residues that differ between the LAIR1 insertion in antibody MGD21 and germ line LAIR1 are shown as red sticks . ( C ) A surface representation of the structure of LAIR1 ( grey ) with residues whose mutation has a major ( red ) or minor ( yellow ) effect on collagen binding highlighted ( Brondijk et al . , 2010 ) . ( D ) A surface view of the LAIR1 insert in antibody MGD21 ( grey ) with residues that differ from germ line LAIR1 highlighted ( red ) . ( E ) A surface view of the LAIR1 insert ( grey ) with residues that differ from germ line LAIR1 in all 27 antibodies tested to date ( Tan et al . , 2016 ) highlighted ( red ) . ( F ) A sequence alignment of germ line LAIR1 and the LAIR1 insert in the MGD21 antibody . Yellow circles are sites residues shown to play a role in collagen binding while a red hexagon represents a potential N-linked glycosylation site mutated in the LAIR1 insert . DOI: http://dx . doi . org/10 . 7554/eLife . 27311 . 009 The normal function of LAIR1 is to interact with collagen ( Meyaard , 2008 ) . The structure of germ line LAIR1 , together with NMR analysis and mutagenesis , allowed the mapping of residues critical for the collagen interaction onto a LAIR1 crystal structure ( Brondijk et al . , 2010 ) . In particular , mutations in residues R59 , E61 and R65 have a significant impact on collagen binding ( Brondijk et al . , 2010 ) . These residues map onto the surface of LAIR1 ( Figure 2C ) that is most exposed in the context of the antibody ( Figure 2D ) . Indeed , mapping of the polymorphisms found in the 27 LAIR1-containing antibodies sequenced to date shows that large parts of this surface are mutable ( Figure 2E ) . The polymorphisms in LAIR1 include R149N , which is in the position equivalent to R65 in germ line LAIR1 and this change may impact collagen binding . A second polymorphism , found in 7/27 of the antibodies ( including MGD21 ) alters the N-linked glycosyation site at residue 69 of LAIR1 ( Wollscheid et al . , 2009 ) , which may alter collagen binding and/or increase RIFIN binding , but is not conserved across the antibody family . Indeed 11 of the 27 sequenced antibodies have mutations in at least one of the residues implicated in collagen binding , or other polymorphisms that reduce the interaction ( Tan et al . , 2016 ) . The LAIR1-containing antibodies are a remarkable variant of the standard immunoglobulin fold . While the majority of mammalian antibodies have predicable and short CDRs , the third CDR of the heavy chain can accommodate usual diversity ( Figure 3 ) ( McLellan et al . , 2011; Wang et al . , 2013; Weitzner et al . , 2015 ) . This is seen in the elongated CDR3 of the broadly neutralizing antibodies that interact with HIV surface proteins and in the insertion of a β-hairpin and disulphide-rich domain in a fraction of bovine antibodies . However , in both of these cases , only the heavy chain CDR3 is altered and the remaining CDR loops remain exposed for antigen binding . The LAIR1-containing antibodies are an exception to this , with the LAIR1-insert interacting with , and partly occluding , all five of the remaining CDR loops . In many ways , the structure resembles an antibody with CDR loops adapted for LAIR1 binding , into which LAIR1 has also been inserted . 10 . 7554/eLife . 27311 . 010Figure 3 . Comparison of the LAIR1-containing antibody with other unusual antibodies . The structure of the LAIR1-containing monoclonal antibody is compared with a classical mouse monoclonal antibody ( 9AD4; PDB code 4U0R ) , a human monoclonal antibody with broadly neutralizing potential against HIV ( PG16; PDB code 4DQ0 ) and a bovine monoclonal antibody ( BLV5B8; PDB code 4K3E ) . In each case , the light chain is blue and the two immunoglobulin domains of the heavy chain are yellow . Inserted domains are shown in red with linker regions in orange . DOI: http://dx . doi . org/10 . 7554/eLife . 27311 . 010 This occlusion of large parts of the CDR loops by the LAIR1 insert has major consequences for its role in antigen recognition , as the majority of the antigen-binding surface will be contributed by LAIR1 . Indeed , it has been shown that the LAIR1 insert alone can bind to infected erythrocytes , as can a LAIR1-containing antibody with the heavy and light chain regions exchanged ( Tan et al . , 2016 ) . Surprisingly an antibody in which the LAIR1 insert has been exchanged for the unaltered germ line LAIR1 did not bind to erythrocytes , although the folding of this chimera was not tested ( Tan et al . , 2016 ) . In addition , the capacity of RIFINs to bind to unaltered LAIR1 alone has not yet been reported . Indeed , it seems most likely that LAIR1 , or a highly related homologue , is the physiological ligand of the group of RIFINs that are recognized by these antibodies and that its insertion into the Fab fragment of an antibody allows it to be affinity matured to mobilise it for immune recognition and recruitment of immune cells . This remarkable LAIR1-containing antibody therefore uses the classical hypervariable loops for a novel function: to position an inserted auxillary domain for antigen recognition . The classical Fab fragment therefore now acts as a link between a ligand for a pathogen surface receptor and the Fc region of the antibody with its immune recruitment capability . It will be fascinating to see if this is a paradigm that is repeated in other novel antibodies , as yet undiscovered . Two synthetic complementary DNA clones based on MGD21 ( Tan et al . , 2016 ) were obtained from GeneArt ( ThermoFisher , UK ) . The heavy chain variable region was amplified using primers VH-F: 5’-GATGGGTTGCGTAGCTGAAGTGCAGCTGGTGGAAACAGGC-3’ and VH-R: 5’-GGGTGTCGTTTTGGCGCTAGACACTGTCACGGTGGTGCC-3’ . The light chain variable region was amplified using primers VL-F: 5’-GATGGGTTGCGTAGCTGCCATCAGAATGACCCAGAGCCCC-3’ and VL-R: 5’-GTGCAGCATCAGCCCGCTTGATTTCCAGCCGGGTGCCC-3’ . The resulting PCR products were cloned into pOPINVH ( heavy chain variable region ) and pOPINVL ( light chain variable region ) by In-Fusion cloning ( Clontech , Mountain View , CA ) ( Nettleship et al . , 2008 ) . Therefore the variable domains from MGD21 were fused to the constant domains derived from the pOPINVH and pOPINVL vectors . DNA constructs expressing heavy and light chains were mixed into a 1 to 1 ratio and co-transfected in HEK293T cells ( ThermoFisher Scientific , UK ) with polyethyleneimine in the presence of 5 μM kifunensine ( Aricescu et al . , 2006 ) . After five days , conditioned media was dialysed against phosphate-buffered saline and purified by immobilised metal ion affinity chromatography using TALON resin ( Clontech , Mountain View , CA ) . The Fab heterodimer was further purified by size-exclusion chromatography using a Superdex 200 16/600 column ( GE Healthcare Life Sciences ) in 10 mM HEPES , pH 7 . 5 and 150 mM NaCl . Concentrated protein ( 10 mg/ml ) was incubated with Flavobacterium meningosepticum endoglycosidase-F1 for in situ deglycosylation ( Hsieh et al . , 2016 ) . The protein samples were then subjected to sitting drop vapour diffusion crystallisation trials in SwisSci 96-well plates by mixing 100 nl protein with 100 nl reservoir solution . The protein crystals were obtained in 20% ( w/v ) PEG4000 , 0 . 1 M sodium citrate , pH 4 . 5 at 18°C . Crystals were transferred into mother liquor containing 25% ( w/v ) glycerol and were then cryo-cooled in liquid nitrogen for storage and data collection . Data were collected on beamline I04-1 at the Diamond Light Source and were indexed and scaled using XDS ( Kabsch , 2010 ) . Phaser ( McCoy et al . , 2007 ) was used to determine a molecular replacement model , using the known structures of LAIR1 ( pdb: 3KGR ( Brondijk et al . , 2010 ) ) and a human monoclonal antibody Fab fragment similar to MGD21 ( pdb: 3DIF , ( Nettleship et al . , 2008 ) ) separated into two files containing the variable and the constant regions , as search models . This identified two copies of the LAIR1-containing Fab fragment in the asymmetric unit of the antibody . Refinement and rebuilding was completed using Buster ( Blanc et al . , 2004 ) and Coot ( Emsley et al . , 2010 ) respectively . To determine the effect of the LAIR1 insert on the accessible surface area of the CDR loops , we used AREAIMOL from the CCP4 suite ( Winn et al . , 2011 ) to determine the accessible surface area of each CDR loop both in the presence and absence of LAIR1 and the linkers .	When bacteria , viruses or parasites invade the human body , the immune system responds by producing proteins called antibodies . Antibodies recognize and bind to molecules ( known as antigens ) on the surface of the invaders . This binding can either neutralize the invader directly or trigger signals that cause other parts of the immune system to destroy it . Our blood contains a huge range of different antibody molecules that each bind to a different antigen . This is despite most human antibodies having the same basic shape and structure . Six loops , known as complementarity determining regions ( CDRs ) , emerge from the surface of the antibody to form the surface that recognizes the antigen . However , variations in the structure of the loops alter this surface enough to allow different antibodies to recognize completely different molecules . In 2016 , a new class of antibodies was identified . Unlike previously identified antibodies , these molecules had an entire human protein , called LAIR1 , inserted into one of their CDR loops . Members of this group of antibodies bind to a molecule , known as a RIFIN , that is found on the surface of human red blood cells that are infected with the parasite that causes malaria . How do LAIR1-containing antibodies bind to their RIFIN targets ? Hsieh and Higgins investigated this question by using a technique called X-ray crystallography to determine the structure of the antibody . This revealed that instead of binding directly to an antigen , all of the six CDR loops in the LAIR1-containing antibody bind to the LAIR1 insert . By doing so , LAIR1 is oriented in a manner that enables it to bind to the RIFIN molecule from the parasite . This is the first known example of an antibody that recruits another protein to bind to an antigen rather than binding directly to the pathogen itself . A future challenge will be to see if other antibodies exist that use this mechanism and whether it can be employed to design new therapeutic antibodies .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "short", "report", "immunology", "and", "inflammation" ]	2017	The structure of a LAIR1-containing human antibody reveals a novel mechanism of antigen recognition
The bacterial flagellum is a self-assembling nanomachine . The external flagellar filament , several times longer than a bacterial cell body , is made of a few tens of thousands subunits of a single protein: flagellin . A fundamental problem concerns the molecular mechanism of how the flagellum grows outside the cell , where no discernible energy source is available . Here , we monitored the dynamic assembly of individual flagella using in situ labelling and real-time immunostaining of elongating flagellar filaments . We report that the rate of flagellum growth , initially ∼1 , 700 amino acids per second , decreases with length and that the previously proposed chain mechanism does not contribute to the filament elongation dynamics . Inhibition of the proton motive force-dependent export apparatus revealed a major contribution of substrate injection in driving filament elongation . The combination of experimental and mathematical evidence demonstrates that a simple , injection-diffusion mechanism controls bacterial flagella growth outside the cell . Many bacteria move by rotation of a helical organelle , the flagellum . The external flagellar filament is several times longer than a bacterial cell body and is made out of up to 20 , 000 flagellin subunits ( Berg and Anderson , 1973; Chevance and Hughes , 2008; Macnab , 2003; Silverman and Simon , 1974 ) ( Figure 1a ) . A type III export apparatus located at the base of the flagellum utilizes the proton motive force ( pmf ) as the primary energy source to translocate axial components of the flagellum across the inner membrane ( Minamino and Namba , 2008; Paul et al . , 2008; Minamino et al . , 2011; Erhardt et al . , 2014 ) . Exported substrates travel through a narrow 2 nm channel within the structure and self-assemble at the tip of the growing flagellum . It has been a mystery how bacteria manage to self-assemble several tens of thousands protein subunits outside the cell , where no discernible energy source is available . Previous reports in the literature concerning the mechanism of flagellum growth have been conflicting ( Iino , 1974; Aizawa and Kubori , 1998; Turner et al . , 2012; Evans et al . , 2013 ) . An exponential decay of filament elongation with length was observed using electron microscopic measurements , which was proposed to be a result of decreased translocation efficiency ( Iino , 1974; Tanner et al . , 2011 ) . A recent study used dual-colour fluorescent labelling of flagellar filaments to distinguish basal from apical filament growth and found that the rate of polymerization was independent of filament length ( Turner et al . , 2012; Stern and Berg , 2013 ) . A model based on the pulling force of a filament-spanning chain of flagellin subunits was proposed to explain the apparent length-independent growth ( Evans et al . , 2013 ) . 10 . 7554/eLife . 23136 . 003Figure 1 . Flagellin protein export and flagella growth rate decrease with filament length . ( a ) Schematic depiction of the bacterial flagellum and proposed models to explain the filament elongation dynamics ( Iino , 1974; Turner et al . , 2012; Evans et al . , 2013 ) . OM=outer membrane , IM=inner membrane . ( b ) Top: Electron micrograph images of mutants deficient in the hook-filament junction protein FlgK or the flagellin-specific chaperone FliS . Bottom: Immunoblotting of cellular and Coomassie-staining of secreted flagellin ( FliC ) in ∆flgK and ∆fliS mutant strains ( relative secreted flagellin levels report mean ± s . d . , n = 3 ) . ( c ) Representative images of a population-based flagellin immunostaining time-course . Time in minutes after induction of flagellin synthesis is indicated . ( d ) Continuous in situ flagellin immunostaining reveals elongation kinetics of individual filaments in real time . Exemplary movie frames are shown and elapsed time in minutes after start of imaging is indicated . ( e ) Quantification of the population immunostaining . Measured filaments per group: t15’ ( n = 187 ) , t30’ ( n = 206 ) , t45’ ( n = 480 ) , t60’ ( n = 648 ) , t90’ ( n = 700 ) , t120’ ( n = 827 ) , t180’ ( n = 302 ) , t240’ ( n = 172 ) . The box plot reports the median ( in red ) , the 25th and 75th quartiles and the 1 . 5 interquartile range . ( f ) Quantification of the continuous in situ flagellin immunostaining . The dark line represents the global , averaged fit of 8 individual filaments . Raw data shown as coloured dots excluding measurement incidents as explained in Figure 1—figure supplement 2 . The initial velocity ( Vi ) was measured on the initial , linear part of the growth curve . Scale bars 2 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 00310 . 7554/eLife . 23136 . 004Figure 1—figure supplement 1 . Quantitative measurements of flagellin leakage during filament formation . ( a ) Schematic overview of experimental setup to determine flagellin leakage during filament assembly . The black elements represent the quantified fractions . ( b ) Coomassie-staining of total extracellular flagellin ( filaments attached to cell bodies and flagellin secreted into culture supernatant ) , polymerized flagellin ( filaments attached to cell bodies ) , detached and secreted flagellin ( FliC detected in culture supernatant ) , detached flagellin ( filaments detached from cell bodies as collected by ultracentrifugation of cleared culture supernatant ) and secreted flagellin ( FliC monomers detected in the culture supernatant after ultracentrifugation ) of wild-type ( WT ) and ∆flgK mutant strains . Relative extracellular flagellin levels in the wild-type were 1 . 000 ± 0 . 175 for total extracellular flagellin , 0 . 756 ± 0 . 046 for polymerized flagellin , 0 . 239 ± 0 . 037 for detached and secreted flagellin , 0 . 057 ± 0 . 036 for detached flagellin and 0 . 085 ± 0 . 021 for secreted flagellin ( mean ± s . d . , n = 4 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 00410 . 7554/eLife . 23136 . 005Figure 1—figure supplement 2 . Growth of individual filaments monitored by continuous flow real-time immunostaining . Raw data measurements of the real-time growth of individual filaments and corresponding fits to the saturated diffusion model are shown . The dashed grey line represents the global fit depicted in Figure 1f and the solid black line represents the fit computed for the individual filament . Arrows denote growth or measurement incidents ( e . g . filament diffused out of focus , the filament stopped growing until the end of the experiment or the cell body rotated thereby preventing accurate length measurements , see Video 1 ) . Only the growth rate data before the arrows were used to fit the model . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 005 In order to test whether filament length itself affects the export rate of flagellin subunits during filament formation , we constructed a flagella-assembly mutant deleted for the first hook-filament junction protein ( ΔflgK ) . This resulted in direct secretion of flagellin monomers into the culture media without transport through the elongated filament . The total amount of extracellular flagellin was analysed in the wild-type and the ΔflgK mutant by de-polymerizing flagellar filaments into flagellin monomers using heat treatment at 65°C . The amount of extracellular flagellin was approximately 1 . 6-fold higher in the ΔflgK mutant compared to wild-type cells . Consistently , cytoplasmic flagellin was substantially more abundant in the wild-type than in the ΔflgK mutant ( Figure 1b ) . Measurements of flagellin leakage during filament formation revealed that only a small fraction of the total flagellin is leaked in monomeric form by wild-type cells during filament formation ( Figure 1—figure supplement 1 ) , demonstrating that the majority of exported flagellin subunits are incorporated into the growing filament under our experimental conditions . These results indicate that the presence of an assembled filament decreases the rate of flagellin transport , which is consistent with the decreased rates of FlgE and FliK export in a long hook mutant ( Koroyasu et al . , 1998; Erhardt et al . , 2011 ) . A similar filament length-dependent effect on flagellin transport was also observed in a mutant of the flagellin-specific cytoplasmic chaperone FliS ( Figure 1b ) . FliS promotes docking and subsequent unfolding of flagellin at the export apparatus ( Kinoshita et al . , 2013; Furukawa et al . , 2016 ) , suggesting that the flagellin injection rate at the export apparatus substantially contributes to the flagellum growth dynamics . We next measured the growth kinetics of flagellar filaments to determine the relation between growth rate and filament length . We engineered a Salmonella strain where the production of flagellar basal bodies ( using the flhDC flagellar master regulatory operon under control of a anhydrotetracycline inducible promoter ) is uncoupled from the expression of chromosomally-encoded flagellin ( using the flagellin gene fliC under control of an arabinose inducible promoter ) . This well-established setup allowed for synchronization of flagella production ( Erhardt et al . , 2011; Karlinsey et al . , 2000 ) by first assembling basal bodies before initiating filament synthesis . The flagella of the synchronized culture were immunostained after increasing growth times ( Figure 1c ) . The initial filament growth rate was ~83 nm∙min−1 , which decreased over time ( Figure 1e ) . In a complementary approach , we monitored , in real-time , the dynamic assembly of individual filaments by employing a continuous in situ immunostaining approach ( Berk et al . , 2012 ) to visualize growing flagella ( Figure 1d , Video 1 ) . A Salmonella strain harbouring a functional , hemagglutinin-epitope tagged flagellin variant under its physiological promoter was grown in a microfluidic device in the presence of labelled , primary antibodies . We observed an initial filament growth rate of ~100 nm∙min−1 , which decreased over time similar as for the population-wide approach described above ( Figure 1f , Figure 1—figure supplement 2 ) . 10 . 7554/eLife . 23136 . 006Video 1 . Real-time flagellum growth observed using in situ continuous flow immunostaining . The animation represents the raw data of the filament length measurements of five representative flagella as a function of time . The inset depicts a 400× time-lapse movie of the corresponding microcolony grown in a CellASIC microfluidic device in the presence of 10 nM anti-HA fluorochrome-coupled primary antibodies . Elapsed time is depicted in min’sec’’ . Coloured circles highlight the onset of filament assembly of the respective length measurement data . Arrows denote growth or measurement incidents ( e . g . filament flipped out of focus or broke off ) . Scale bar 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 006 In a previous study , Turner et al . ( 2012 ) measured the growth kinetics of individual filaments in Escherichia coli by site-specific labelling of flagellin subunits containing an exposed cysteine residue using sulfhydryl-specific ( maleimide ) fluorochromes and reported a length independent growth rate of ~13 nm∙min−1 . We optimized this method to exchange dyes multiple ( three to six ) times in situ during normal culture growth with minimal perturbation of bacterial growth ( Figure 2 , Figure 2—figure supplement 1 , Figure 2—figure supplement 2 , Figure 3 , Figure 3—figure supplement 1 ) . The labelling of successive fragments of the flagellum with maleimide fluorochromes in situ allows observation of the filament growth dynamics at the end of the experiment . Triple labelling ( exchange of dyes three times ) demonstrated that the extension length of the filament ( apical fragment ) is inversely proportional to its initial length ( basal fragment ) , until the growth rate for long filaments decreases to a point where it becomes minimal ( Figure 2 ) . Using this setup , the dynamic range of basal fragment lengths was increased by performing the experiment with varying growth durations ( 15 to 180 min ) . 10 . 7554/eLife . 23136 . 007Figure 2 . In situ filament labelling reveals a negative correlation between filament length and elongation rate . ( a ) Experimental design of the in situ triple-colour labelling time-course . Basal ( F1 ) and apical ( F2 ) fragments were grown for 15–180 min and 30 min , respectively . The growth duration of basal fragments is indicated in the legend . Coloured arrows indicate the coordinates of the representative example images . The fit represents the injection-diffusion model with parameters kon ≈ 33 . 35 s−1and D ≈ 5 . 90 × 10−13 m2 ⋅ s−1 . Scale bar 2 µm . ( b ) Average size of the individual fragments for different durations of elongation of the first fragment . Error bars represent the 95% confidence interval of mean estimation . ( c ) Relation between the size of the second and third fragment . 93 . 4% of the filaments have F3 fragments shorter than the F2 fragment with the difference increasing with the length of F2 . ( d ) Flagella labelled in panel a were measured and sorted according to the length of F1 , which reveals the inverted relationship between the initial length and extension length of the filament . Each vertical line represents an individual filament ( n = 1254 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 00710 . 7554/eLife . 23136 . 008Figure 2—figure supplement 1 . In situ labelling of flagella using maleimide fluorochromes . ( a ) Localization of the T237 residue in FliC , mutated to cysteine for the fluorochrome-maleimide labelling . The inset shows the structure of the polymerized filament with the localization of T237 in red . ( b ) Filament length distribution of the FliCT237C mutant strain EM2046 after 1 hr of growth in presence of the maleimide fluorochrome ( Mal488 ) is identical to the wild-type ( TH15801 ) grown in the same conditions or in absence of the maleimide fluorochrome in the medium . The box plot reports the median ( in red ) , the 25th and 75th quartiles and the 1 . 5 interquartile range . ( c ) Representative images of the triple labelling of strain EM2400 with constant induction of FlhDC . Variations in F1 growth time ( 30 min and 60 min ) demonstrate length-dependent decrease in filament growth rate . Scale bar 2 µm . ( d ) Triple filament labelling after constant ( left , compare also panel c ) or transient ( right ) induction of FlhDC led to comparable fragments size ( n = 417 filaments for constant AnTc , n = 1254 filaments for AnTc pulse ) . Variations in F1 growth time ( 30 min and 60 min ) highlight the length-dependent filament elongation rate . Error bars represent the 95% confidence interval of mean estimation . ( e ) Representative images for multiple labelling of strain EM2400 with constant ( left ) or transient ( right ) induction of FlhDC . Transient induction of FlhDC decreases the number of flagella per cell without changing the growth kinetics ( see panel d ) , and accordingly facilitated quantitative analysis of filament length . Scale bar 2 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 00810 . 7554/eLife . 23136 . 009Figure 2—figure supplement 2 . Triple-colour labelling time course of second fragment F2 . ( a ) Triple labelling with time course of the second fragment ( F2 ) , the first and third fragment were grown for 30 min . The dotted lines , shown for comparison , are the fit lines of the 30 and 60 min curves of the multiple labelling in Figure 3c . ( b ) Average length of the fragments for the different time points . Error bars represent the 95% confidence interval of mean estimation . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 00910 . 7554/eLife . 23136 . 010Figure 3 . Growth kinetics of individual flagella revealed by in situ , multicolour labelling . ( a ) Left: Experimental design of the in situ , multicolour labelling . Right: Representative fluorescent microscopy image for multiple labelling of flagellar filaments with a series of maleimide dyes . TB: tryptone broth without dye , AnTc: anhydrotetracyline induction of flagella genes . Scale bar 2 µm . ( b ) Basal/apical length coordinates were obtained by varying the duration of basal growth and successive fragments were combined to generate a total of 1276 basal/apical coordinates from 291 filaments . The growth duration of the apical fragment was 30 min . Average speeds are calculated from the average elongation per 30 min ( <1 µm or >8 µm ) . The fit represents the injection-diffusion model with parameters kon ≈ 27 . 09 s−1and D ≈ 5 . 41 × 10−13 m2 ⋅ s−1 . ( c ) Basal/apical length coordinates were obtained for various durations of apical growth ( 30–150 min ) from the multiple labelling data shown in panel b . ( n = 1276 for 30 min , n = 986 for 60 min , n = 697 for 90 min , n = 422 for 120 min , n = 169 for 150 min ) . The fit for various durations of apical growth represents the injection-diffusion model with parameters kon and D ( 60 min: kon ≈ 27 . 72 s−1 , D ≈ 5 . 56 × 10−13 m2 ⋅ s−1; 90 min: kon ≈ 28 . 06 s−1 , D ≈ 5 . 63 × 10−13 m2 ⋅ s−1; 120 min: kon ≈ 27 . 03 s−1 , D ≈ 5 . 42 × 10−13 m2 ⋅ s−1; 150 min: kon ≈ 26 . 36 s−1 , D ≈ 5 . 29 × 10−13 m2 ⋅ s−1 ) . Average growth rates were estimated from the Y-intercept of the fit curve . The inset presents the average growth plotted against time . ( d ) Filament length as a function of time of individual flagella from the multiple labelling data . Each grey line represents the growth curve of an individual filament . The average growth rates estimated in panel c are plotted for comparison . ( e ) Quality of multiple labelling data . Only a minor fraction of the filaments were broken or stalled ( highlighted as red dots , Figure 3—figure supplement 1a ) , which has limited effect on the parameter fit . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 01010 . 7554/eLife . 23136 . 011Figure 3—figure supplement 1 . Quality of multiple labelling data . ( a ) Raw data of the multiple labelling . Each vertical line represents an individual filament ( n = 291 ) . Broken or strongly stalled filaments are denoted with a white arrow . Basal/apical length coordinates resulting from those filaments are highlighted in red in Figure 3e . ( b ) Basal/apical couples , as calculated in Figure 3c are shown in green/red , respectively . The elongation time of the apical fragment is indicated on top of each graph . Smoothing was applied on the apical data using exponentially weighted moving averages ( span = 50 ) to remove the individual-based variability . The smoothed data is represented for each elongation time as a dotted line . ( c ) Comparison of the triple labelling data ( 1254 fragments ) and the multiple labelling data for 30 min elongation ( 1276 fragments derived from 291 flagella ) . The multiple labelling allows obtaining greater resolution with about four times less flagella . ( d ) Basal/apical coordinates were obtained by smoothing of the raw data as described in panel b . The growth curves from Figure 3c are plotted for comparison . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 01110 . 7554/eLife . 23136 . 012Figure 3—figure supplement 2 . Filament breaking/stalling events and heterogeneous injection rates decrease the quality of the data required to fit the injection-diffusion model . ( a ) The combined multiple-labelling approach and the in vitro labelling protocol ensured minimal filament breaks in our dataset ( Figure 3—figure supplement 1a , compared to breaks induced by shearing in Figure 4 ) . Each filament of the dataset was subjected to a virtual breakage/stalling event with a probability pbreak ( pbreak = 0 , 0 . 1 , 0 . 25 , 0 . 5 , 0 . 75 or 1 ) . The position of the breakage/stalling event was chosen randomly along the length of the filaments with a uniform probability . ( b ) Simulation of random filament breakage ( of probability p ) using the multiple labelling dataset presented in Figure 3 . The random filament breakage simulation demonstrates that data points accumulate on the x-axis and below the fit curve ( in red ) . It is crucial to note that a high fraction of broken filaments masks the obvious length-dependency of filament growth and prevents an accurate fit on the complete set of points ( linear fit in blue; compare with Figure 3 of Turner et al . [2012] ) . ( c ) Combination of the data obtained in Figure 6a for the 0 , 10 , and 20 µM CCCP concentrations to simulate a heterogeneous injection rate . A variable injection rate within the population would mask the length-dependency of filament growth ( compare with Figure 3A of Turner et al . [2012] ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 012 Next , multiple labelling ( exchange of dyes six times ) of flagellar filaments allowed us to compute various basal/apical couples and increased the dynamic range of the growth rate data for individual flagella . The multiple labelling of flagellar filaments confirmed the length-dependent elongation mechanism with an elongation speed decreasing gradually from ~100 nm∙min−1 to ~20 nm∙min−1 ( Figure 3 , Figure 3—figure supplement 1 ) . Alternative combination of the fragments allowed us to determine the filament elongation kinetics for various growth durations and in fine to derive a growth curve ( Figure 3c–d ) . Our method further allowed us to exclude stalled or broken filaments and study the filament elongation dynamics under normal cultivation conditions for a wide range of fragment lengths . We note that we only observed a minor fraction of flagella that broke or stopped growing during the experiment ( Figure 3e ) . The solid curves in Figure 2 and Figure 3 represent the best fit of the data to a growth curve for which the growth rate is a function of the length L of the form ab+L , where the parameter a has units of a diffusion coefficient , and b has units of length . Derivation of this formula is based on an injection-diffusion model where flagellin monomers , which are at least partially α-helical inside the channel ( Shibata et al . , 2007 ) , are pushed by a pmf-driven export apparatus into the channel and move diffusively in one dimension through the length of the flagellum ( Stern and Berg , 2013; Keener , 2006 ) . An analytical expression for the flagellum length dependent growth rate is based on a continuum injection-diffusion model for the growth of flagellar filaments . Monomers ( each of length l ) in the growing filament are assumed to move diffusively . Because the filaments are long narrow tubes , monomers are partially unfolded and diffusion is constrained to be strictly one-dimensional , i . e . no passing allowed . In the corresponding continuum model , we define u ( x , t ) l as the density of monomers per unit length at position x and time t . Then u satisfies the diffusion equation ( 1 ) ut=Duxx . Here , D is the diffusion coefficient of the monomers . We assume that all end-to-end collisions between monomers are ballistic , with no end-to-end binding . For this , Fickian diffusion is the appropriate description of diffusion , even at high densities . We assume that at the growing end X=L , monomers are quickly removed by folding/polymerization so that effectively u ( L , t ) =0 . The details of the mechanism by which monomers are secreted at the basal end X=0 is not known , but it is known to be related to the pmf ( Paul et al . , 2008 ) . We assume that the rate of secretion ( number of monomers per unit time ) into an empty filament is Kon , but if it is not empty , then the rate of secretion is decreased because of the no-passing restriction . Consequently , the flux J0 ( number of monomers per unit time at the basal end ) is taken to be ( 2 ) J0=−Dlux ( 0 , t ) =Kon ( 1−u ( 0 , t ) ) . Finally , the rate of growth of the filament is given by ( 3 ) dLdt=βJL=−Dβlux ( L , t ) , where β is the length increment of the filament due to polymerization of a single monomer . Since the filament growth rate is small compared to the average velocity of monomers , it is reasonable to take the monomer diffusion to be in quasisteady state , i . e . uxx=0 . Thus , the monomer density in the filament is a linearly decreasing function and ux is the constant −u ( 0 ) L . It follows that the filament growth rate is ( 4 ) dLdt=βDl1Dkonl+L=ab+L , where a=βDl , with units of diffusion , and b=Dkonl , with units of length . This is readily solved to find the filament length as a function of time ( 5 ) L ( t ) =−b+b2+2at . We can estimate the diffusion coefficient using a=βDl , so that ( 6 ) D=alβ . From all the datasets presented above , we determined a ≈ 0 . 2 µm2 ⋅ min−1 . Using β = 0 . 47 nm ( a flagellar filament of 1 µm length is composed of approximately 2130 flagellin subunits [Yonekura et al . , 2003] ) , l = 74 nm ( assuming an extended , α-helical flagellin molecule ) this leads to an estimate of D ≈ 5 . 25 × 10−13 m2 ⋅ s−1 . Stern and Berg ( Stern and Berg , 2013 ) estimated D ≈ 5 . 78 × 10−11 m2 ⋅ s−1 for freely moving α-helical flagellin in water . The actual diffusion coefficient for movement in the narrow 2 nm channel would be substantially smaller , however . Stern and Berg ( Stern and Berg , 2013 ) ( their Figure 2 ) used a 480 times smaller diffusion coefficient ( D ≈ 1 . 25 × 10−13 m2 ⋅ s−1 ) for numerical simulations that resulted in a declining growth curve , which closely resembled the filament growth kinetics presented above . Our triple and multiple labelling experiments demonstrated that the growth of a new part of the filament ( apical fragment ) shows a strong inverse dependence on its initial length ( basal fragment ) for short filaments , while the growth rate for long filaments decreases to a point where this dependence becomes minimal ( Figure 2 , Figure 3 , Figure 3—figure supplement 1 ) . We note that several differences in the experimental setup of Turner et al . ( 2012 ) from ours might have affected the injection rate and frequency of filament breakage . As described in detail in Appendix 1 , the possibility of broken/stalled filaments and possible perturbations of the injection rate reconcile our data with the reported filament growth data of Turner et al . ( 2012 ) and explains why we observed a length-dependent decrease in growth rate . In support , we simulated in Figure 3—figure supplement 2 the effects of filament breaking/stalling events and heterogeneous injection rates . The simulated broken/stalled filaments accumulate on the x-axis , which results in a quasi-linear fit of the complete filament growth rate data , similar to the linear filament growth observed by Turner et al . ( 2012 ) . We further note that a length-dependent decrease in filament growth rate would explain why flagellar filaments do not growth indefinitely . However , flagellar filaments broken by mechanical shearing forces can re-grow ( Turner et al . , 2012; Rosu and Hughes , 2006; Vogler et al . , 1991 ) . The injection-diffusion model predicts that the elongation rate of re-growing filaments would increase compared to unbroken filaments . We performed multiple labelling in situ to determine the growth rate of individual filaments that had been broken using mechanical shearing forces . Consistent with the injection-diffusion mechanism , the elongation rate of re-growing , previously broken filaments was substantially faster than the elongation rate of unbroken filaments and was dependent on the length of the basal filament segment , which remained attached to the bacterial cell surface ( Figure 4 ) . 10 . 7554/eLife . 23136 . 013Figure 4 . Elongation rate of re-growing filaments increases after mechanical shearing . ( a ) Experimental design to determine filament elongation rate after mechanical shearing using multicolour labelling . ( b ) A successful shearing event removed fragment F3 and partially or completely fragment F2 . ( c ) Representative example images of control filaments and filaments broken using mechanical shearing forces . Flagellar filaments were sheared by passing the bacterial culture five times ( mild shearing ) or up to 30 times ( strong shearing ) in and out of a 22-gauge needle . Scale bar 2 µm . ( d ) Left panel: length of the basal , cell-attached filament after mechanical shearing demonstrating successful filament breakage . Right panel: length of apical , re-growing filament fragments demonstrating a length-dependent increase in filament elongation rate . The box plots reports the median , the 25th and 75th quartiles and the 1 . 5 interquartile range . Data points represent individual filament fragments . Statistical significance according to a two-tailed Student’s t-test is indicated . F4 strong vs . control: p-value=0 . 000026 ( * ) ; F5 strong vs . control: p-value=0 . 002452 ( ) ; F6 strong vs . control: p-value=0 . 034514 ( * ) ; F7 strong vs . control: not significant ( n . s . ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 01310 . 7554/eLife . 23136 . 014Figure 4—figure supplement 1 . Basal/apical coordinates of sheared and control filaments showing the dispersion of the filament growth data . Basal/apical coordinates of the data shown in Figure 4d of control filaments and filaments broken using mechanical shearing forces . The length of the apical F4 fragment ( post-shearing ) is shown in relation to the length of the basal fragment pre-shearing ( F1→F3 ) and highlights the increased elongation rate of short filaments after shearing . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 014 Based on the observations of Turner et al . ( 2012 ) , Evans et al . ( 2013 ) developed a model where folding of newly arriving subunits at the tip of the flagellum would provide energy to pull successive subunits through the channel at a constant rate . Evans et al . demonstrated that N-terminal regions of flagellar substrates ( FlgD , FlgE , FlgG and FliK ) can bind to the C-terminal cytoplasmic domain of FlhB , which is a component of the pmf-driven transmembrane export gate complex . Further , site-specific cysteine-cysteine crosslinking showed that the N- and C-terminal regions of hook ( FlgE ) and flagellin ( FliC ) can interact to form head-to-tail dimers . They hypothesized that formation of inter-subunit chains resulting from those interactions could enable their transport at a length-independent speed , as the folding of the exported molecules at the filament tip would provide a continuous pulling force . While the N- and C-terminal interactions of flagellar substrates might play an important role during substrate docking and in the final fold of assembled hook and filament subunits , the proposed inter-subunit chain mechanism for flagellin transport and filament assembly raises several issues that are incompatible with the known biophysical properties of flagellum assembly ( Yonekura et al . , 2003; Samatey et al . , 2001 ) . A flagellum-spanning chain requires interactions of the N- and C-terminal α-helical domains of flagellin , but the 2 nm wide filament channel ( Yonekura et al . , 2003 ) is too narrow to accommodate the secretion of much more than one folded α-helix ( Figure 5a ) . The chain mechanism hypothesizes that folding of a flagellin subunit at the tip of the flagellum can pull a chain of succeeding subunits , but the N- and C-termini of successive flagellin molecules are anti-parallel and far apart in the polymerized filament structure ( ∼17 Å on average ) ( Yonekura et al . , 2003; Samatey et al . , 2001 ) ( Figure 5b ) . Further , the chain mechanism is not compatible with simultaneous secretion of non-chaining substrates ( Figure 5d ) . Flagellar substrates such as FlgM or excess hook-associated proteins ( FlgK , FlgL , FliD ) are constantly exported during flagellum growth ( Komoriya et al . , 1999 ) and do not interact with flagellin ( Furukawa et al . , 2002 ) . Also , premature termination of translation is occurring frequently ( ~1∙10−4 to ~5∙10−4 events per codon ) ( Sin et al . , 2016 ) . Thus , a high proportion of 5–20% newly synthesized flagellin might be truncated for the C-terminal domain needed for head-to-tail chain formation . We estimate that secretion of as little as one non-chaining substrate every 10 , 000 full-length flagellin molecules would prevent filament elongation if a chain mechanism drives flagellum growth ( Figure 5d–g ) . 10 . 7554/eLife . 23136 . 015Figure 5 . The contribution of inter-subunit chains for filament elongation rate . ( a ) The 2 nm wide filament channel only accommodates one folded α-helix . ( b ) The N- and C-termini of successive flagellin molecules are anti-parallel and far apart in the polymerized filament structure . ( c ) Top: Structure , domains , and secondary structures of flagellin FliC ( PDB: 1UCU ) . Mutant flagellins lacking combinations of the N- and C-terminal domains required for head-to-tail coiled-coil chaining ( ΔN , ΔCS , ΔCL ) were co-expressed together with endogenous flagellin ( FliC ) to disrupt chain formation . Bottom: Flagellin immunoblotting on cellular and secreted fractions ( relative full-length flagellin levels report mean ± s . d . , n = 3 ) . ( d ) Simultaneous secretion of non-chaining substrates breaks a filament-spanning chain of flagellin molecules . A strict chain model requires the chain to span the entire filament and does not allow for disruptions of the chain . A chain model with the possibility of recovery by diffusion of broken chains is discussed in Figure 5—figure supplement 1 . ( e ) In situ , multicolour labelling of flagellar filaments during competitive co-expression of chain-disrupting mutant flagellins . The average growth of fits computed from basal/apical coordinates presented in Figure 5—figure supplement 3c and as described in Figure 3c is shown as a function of time . Basal/apical coordinates were derived from multiple labelling data of individual filaments: n = 399 from 89 filaments ( − ) , n = 271 from 58 filaments ( WT ) , n = 278 from 62 filaments ( ∆CL ) , n = 412 from 93 filaments ( ∆N ∆CL ) , n = 209 from 46 filaments ( ∆CS ) , n = 312 from 73 filaments ( ∆N ∆CS ) . The fits represent the injection-diffusion model and parameters kon and D are given in Figure 5—source data 1 . ( f ) Probability of existence of n-long chains defined by the binomial law . Long chains are highly improbable for a 15% proportion of competing substrates ( i . e . formation of a more than 2 . 4 µm long chain ( n > 33 ) has a probability of 1% ) . The bars show the individual probability of existence , the dotted blue line the cumulated probability of existence of chains longer than the number on the x-axis . The grey curve indicates the chain length in µm , which reflects that filaments cannot grow longer than a few hundred nanometres with a chain-based mechanism . ( g ) Simulation of filament growth dependent on inter-subunit chains or the injection-diffusion model in presence of random proportion of competing substrate . The injection-diffusion model fit represents the mean of the multi-labelling data set of Figure 3 with parameters kon ≈ 27 . 25 s−1and D ≈ 5 . 46 × 10−13 m2 ⋅ s−1 . Dashed white line: median length of the filament for chain-model dependent growth . Grey box: expression range of chain-disrupting mutant flagellins used in panel e and Figure 5—figure supplement 1a . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 01510 . 7554/eLife . 23136 . 016Figure 5—source data 1 . Parameters kon and D of the injection-diffusion model fits of Figure 5—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 01610 . 7554/eLife . 23136 . 017Figure 5—figure supplement 1 . Filament growth dynamics in the presence of competing non-chaining substrate . ( a ) Effect of deletion of fliS on secretion of flagellin truncation mutants . Anti-flagellin immunoblotting of secreted wild-type flagellin and flagellin truncation mutants . The flagellin truncation mutants additionally missing the FliS binding site ( ΔN = ∆29–32 , ΔCS = ∆450–495 , ΔCL = ∆310–495 ) were secreted in similar quantity in both the wild-type and ∆fliS mutant . Full-length flagellin was expressed from its native promoter . ( b ) Schematic illustration of chain-model dependent filament growth . Unlike the simulation presented in Figure 5 , we assume here that filament growth can resume after the basal part of the chain diffused to the tip . The diffusion coefficient depends on the number of flagellin subunits in the chain . ( c ) Simulation ( n = 100 ) of filament growth in the presence of 10–20% competing substrate . At this proportion of competing substrates , chains would constantly be broken and have an average size of 6–11 subunits . This would induce significant delays in growth , which is not observed experimentally ( Figures 1f and 5e ) . The dynamics of filament growth dependent on the injection-diffusion mechanism would not be perceptibly altered by this proportion of competing substrates and is consistent with the experimental observations ( Figure 5e ) . ( d ) Simulation of chain-model dependent filament growth . In contrast to the simulation presented in Figure 5g , here we allow for recovery upon a chain-breaking event by diffusion of the basal chain fragment to the tip , where it resumes growth . This simulation reveals two ranges of possible chain model dependent filament growth . The first one ( proportion of competing substrate p<10−4 ) is identical to the model without possibility to recover upon chain-breakage and requires a non-physiological low proportion of competing substrates . In this case chains are so long that it is virtually impossible to diffuse to the tip within the time frame of the simulation , thus effectively arresting filament elongation . The second range where a chain model dependent growth is valid ( p>0 . 3 ) postulates frequent chain-breakage events . Here , short chains ( 1–5 flagellin subunits ) are able to diffuse to the tip in a reasonable time frame , however , the contribution of the pulling force of such short chains to drive elongation of the filament is negligible and secretion of substrates is almost entirely driven by diffusion . The simulation of Figure 5g is shown in blue for comparison . The simulation with the injection-diffusion model is shown as a red dashed line . In all models , export starts to be affected by competition for injection when the number of competing substrates is significantly higher than the number of secreted flagellin molecules . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 01710 . 7554/eLife . 23136 . 018Figure 5—figure supplement 2 . Schematic illustration of the chain-model dependent simulation of filament growth . The simulation generates a random succession of substrates with a certain probability of non-chaining substrates incorporated into the chain of substrates that arrive at the export gate . If flagellin substrates arrive , they participate in filament elongation with the observed rate of growth . If , however , a non-chaining substrate arrives at the gate , then the simulation does either not allow for recovery of growth ( top and bottom left ) or introduces an elongation delay based on the time required for the newly formed basal chain to reach the tip through diffusion , where it then can resume rapid growth ( middle and bottom right ) . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 01810 . 7554/eLife . 23136 . 019Figure 5—figure supplement 3 . Characterization of chain-disrupting flagellin truncation mutants . ( a ) Schematic of flagellin FliC truncation constructs used to disrupt subunit chain formation . Overview of various flagellin mutants lacking parts and/or combinations of N- and C-terminal domains required for head-to-tail chain formation . Qualitative assessments of the ability to complement a fliC mutant strain ( labelled ‘complementation’ ) , the ability to form functional flagellar filaments ( labelled ‘assembly’ ) and the secretion levels are indicated . Truncations in the D0 domain were expected to prevent flagellin polymerization and thus the truncation mutants do not assemble filaments and are unable to complement a fliC mutant strain . Overproduction of N-terminal domain truncation mutants that retain the C-terminal chaperone-binding domain competes with wild-type flagellin for the available pool of flagellin-specific chaperone FliS ( Figure Supplement S6a ) . The absence of FliS impairs filament formation ( Figure 1b ) and thus overproduction of ∆29–33 , ∆11–18 and ∆11–18 + ∆29–33 results in a dominant-negative motility phenotype . N . D . = not determined . ( b ) Expression and secretion profile of flagellin truncation mutants . Western blot analysis of cellular and supernatant fractions of strain TM113 ( ∆fliC ) or strain NH001 ( ∆flhA ) expressing various flagellin truncation variant as outlined in panel a . All flagellin mutants are secreted in the presence of a functional flagellar export apparatus . ( c ) Mutant flagellins lacking parts or combinations of the N- and C-terminal domains required for head-to-tail chain formation ( ΔN = ∆29–32 , ΔCS = ∆450–495 , ΔCL = ∆310–495 ) or wild-type flagellin ( WT ) were co-expressed together with endogenous flagellin in strain EM2400 as shown in Figure 5 . Multicolour labelling of flagellar filaments was performed and basal/apical coordinates were computed for various durations of apical growth and plotted as outlined in Figure 3c . Successive fragments of individual filaments were combined to obtain n basal/apical coordinates as follows: n = 399 from 89 filaments ( − ) , n = 271 from 58 filaments ( WT ) , n = 278 from 62 filaments ( ∆CL ) , n = 412 from 93 filaments ( ∆N ∆CL ) , n = 209 from 46 filaments ( ∆CS ) , n = 312 from 73 filaments ( ∆N ∆CS ) . The fits represent the injection-diffusion model . Parameters kon and D are given in Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 019 To test the requirement of subunit chain formation for flagellum growth in more detail , we generated flagellin mutants truncated for the N- and C-termini that render head-to-tail linkage impossible ( Figure 5c ) . All flagellin truncation mutants were secreted , but were deficient in flagellum assembly due to deletions in the D0 and D1 domains needed for filament polymerization and FliS chaperone binding ( Yonekura et al . , 2003 ) ( Figure 5—figure supplement 1a , Figure 5—figure supplement 3 ) . We expressed those non-chaining , but secreted flagellin mutants in trans to disrupt formation of a chain of wild-type flagellin molecules ( Figure 5d ) . Competitive secretion of the flagellin truncation mutants did not affect endogenous flagellin transport during filament formation ( Figure 5c ) . Filament extension kinetics were determined using multiple labelling of individual flagellar filaments and , similarly , no significant difference was observed when chain-disrupting flagellin mutants were co-expressed ( Figure 5e , Figure 5—figure supplement 3c ) . Mathematical modelling of the chain model-dependent filament elongation dynamics predicted a linear growth up to a very long flagellum ( >0 . 1 mm ) , which is in clear contradiction with the experimental observations ( Appendix 2 ) . Our high-resolution filament growth rate data and the previous observations by Stern and Berg ( 2013 ) suggested that two major components drive flagellin export: pmf-dependent injection of subunits by the type III export apparatus at the base of the flagellum and diffusion of subunits along the length of the flagellum . We used carbonyl cyanide m-chlorophenyl hydrazone ( CCCP ) to disrupt the pmf , which is required for substrate translocation via the export apparatus into the central channel of the growing flagellar structure ( Minamino and Namba , 2008; Paul et al . , 2008 ) . The injection-diffusion model predicts that a decrease in the injection rate Kon results in slow , quasi-linear growth for sufficiently small Kon . As expected , CCCP treatment resulted in impaired filament extension in a dose-dependent manner , which recovered upon removal of the uncoupler ( Figure 6a , Figure 6—figure supplement 1 ) . We hypothesized that in presence of high concentration of CCCP , the injection of substrate would be strongly reduced and result in low-speed growth . As shown in Figure 6c , the filament elongation rate for the highest CCCP concentration ( ~18 nm∙min−1 ) was virtually independent of the length of the filament as predicted by the model . Interestingly , some filaments were unaffected by the CCCP treatment , likely due to the action of multidrug transporters ( Lomovskaya and Lewis , 1992 ) , and displayed kinetics similar to the untreated population ( Figure 6—figure supplement 1d ) , highlighting the major contribution of the pmf in energizing export . 10 . 7554/eLife . 23136 . 020Figure 6 . The effect of pmf on flagellin injection and filament growth rate . ( a ) Top: Experimental design . Carbonyl cyanide m-chlorophenyl hydrazone ( CCCP ) reduces the proton motive force ( pmf ) and was present during growth of the second fragment ( 60 min ) and removed during growth of the third fragment , which allowed the pmf to regenerate . TB: tryptone broth without dye , AnTc: anhydrotetracyline induction of flagella genes . Bottom: Fragment lengths represented as basal/apical ( F1/F2 ) coordinates ( n = 255 for 0 µM CCCP , n = 395 for 10 µM CCCP , n = 371 for 20 µM CCCP , n = 353 for 30 µM CCCP ) . The fits represent the injection-diffusion model with parameters D ≈ 5 . 25 × 10−13 m2 ⋅ s−1 , and kon ≈ 26 . 10 , 3 . 19 , 1 . 19 , 0 . 70 s−1 for 0 µM , 10 , 20 , 30 µM CCCP respectively . Representative fluorescent microscopy images of labelled flagella and matching coordinates are highlighted by coloured frames and arrows . Scale bar 2 µm . ( b ) Filament length as a function of time for decreasing values of kon . For small values of kon , the injection rate but not flagellin transport is rate-limiting , which results in a quasi-linear growth . ( c ) Growth curves for the CCCP raw data of panel a determined by fitting the data to the injection-diffusion model with a fixed parameter a . The values for kon decrease by a factor of 8 ( 10 µM CCCP ) , 22 ( 20 µM CCCP ) , and 38 ( 30 µM CCCP ) , compared to the untreated control . ( d ) Equation and biophysical parameters of the injection-diffusion model . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 02010 . 7554/eLife . 23136 . 021Figure 6—figure supplement 1 . Supporting data for effect of CCCP inhibition of the pmf-dependent protein export on flagella growth rate . ( a ) Raw data of triple-labelled flagella after inhibition of the pmf-dependent export apparatus by treatment with increasing concentrations of CCCP were measured and sorted according to the length of F1 . Each vertical line represents an individual filament ( n = 255 for 0 µM CCCP , n = 395 for 10 µM CCCP , n = 371 for 20 µM CCCP , n = 353 for 30 µM CCCP ) . ( b ) Basal/apical relationship ( 30 min elongation ) for decreasing values of kon , which result in linear-like growth rate . ( c ) Fragment lengths from the triple labelling data of Figure 6a presented as the average individual fragment size . Error bars represent the 95% confidence interval of mean estimation . ( d ) Minor population of CCCP-resistant flagella display filament elongation kinetics similar to the untreated control . Left panel: distribution of the second ( F2 ) fragments for the untreated and 30 µM CCCP conditions . The presence of long filaments ( F2 >2 µM ) is characteristic for resistance to the uncoupler . Middle panel: the filaments with long F2 fragments follow a kinetic similar to the untreated population ( blue curve ) . The fit curve for the 30 µM CCCP treatment is shown in orange for comparison . Right panel: the CCCP-sensitive population undergoes an increase in elongation speed after recovery from the CCCP treatment ( F3 > F2 ) while the elongation speed of the CCCP-resistant population follows the normal kinetic ( F3 < F2 , see Figure 2c ) . The dotted lines mark a threshold of 2 µm for F2 . The points above this threshold are shown in purple . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 021 The bacterial flagellum is a remarkably complex nanomachine . Here , we present the first real-time visualization and experimentally supported biophysical model of the dynamic self-assembly process of this large , widely conserved nanomachine . We propose that bacterial flagella grow through an injection-diffusion mechanism ( Figure 6d ) , which provides a simple explanation why the flagellar filament does not grow infinitely in the absence of any other length-control mechanism . It appears likely that similar biophysical principles are conserved for effector protein secretion in the evolutionary related , virulence-associated injectisome with important implications for the rational design of novel anti-infectives targeted against type III secretion systems . Salmonella enterica serovar Typhimurium strains and plasmids used in this study are listed in Table 1 . Lysogeny broth ( LB ) contained 10 g of Bacto-Tryptone ( Difco ) , 5 g of yeast extract , 5 g of NaCl and 0 . 2 ml of 5N NaOH per litre . Soft agar plates used for motility assays contained 10 g of Bacto-Tryptone , 5 g of NaCl , 3 . 5 g of Bacto-Agar ( Difco ) and 0 . 2 ml of 5N NaOH per liter . Tryptone broth ( TB ) contained 10 g of Bacto-Tryptone and 5 g of NaCl . Ampicillin was added to the medium at a final concentration of 100 µg/ml , L-arabinose at a final concentration of 0 . 2% and anhydrotetracyline at a final concentration of 100 ng/ml if required . 10 . 7554/eLife . 23136 . 022Table 1 . Strains and plasmids used in this study . DOI: http://dx . doi . org/10 . 7554/eLife . 23136 . 022StrainRelevant characteristicsSource or referenceSJW1103Salmonella enterica serovar Typhimurium wild-type strain SJW1103 for motility and chemotaxis ( Yamaguchi et al . , 1984 ) TM113SJW1103 ∆fliCT . Miyata , unpublishedNH001SJW1103 ∆flhA ( Hara et al . , 2011 ) MM1103iSSJW1103 ∆fliS::km ( Furukawa et al . , 2016 ) MM1103gKSJW1103 flgK::Tn10This studyMM1103gKiSSJW1103 ∆fliS::km flgK::Tn10This studyMM1103CPOPSJW1103 ∆PfliC::tetRA-62This studyTH437Salmonella enterica serovar Typhimurium wild-type strain LT2lab collectionTH15801LT2 PflhDC5451::Tn10dTc[del-25] ∆hin-5717::FCFlab collectionEM1237LT2 ∆araBAD1026::fliC ∆fliC7861::FRT ∆hin-5717::FCF PflhDC5451::Tn10dTc[del-25]This studyEM2046LT2 ∆hin-5717::FRT fliC6500 ( T237C ) PflhDC5451::Tn10dTc[del-25]This studyEM2400LT2 ∆hin-5717::FRT fliC6500 ( T237C ) ∆araBAD1005::FRT PflhDC5451::Tn10dTc[del-25]This studyEM4076LT2 ∆hin-5717::FRT fliC7746::3xHA ( ∆aa201-213::3xHA ) motA5461::MudJ PflhDC5451::Tn10dTc[del-25] ∆sseA-ssaU::FCF ( deletes Spi-2 ) This studyPlasmidsRelevant characteristicsSource or referencepBAD24Expression vectorInvitrogenpAOA001pBAD24/FliCThis studypAOA002pBAD24/FliC ( ∆29–32 ) This studypAOA003pBAD24/FliC ( ∆11–18 ) This studypAOA004pBAD24/FliC ( ∆11–18/∆29–32 ) This studypAOA005pBAD24/FliC ( ∆310–495 ) This studypAOA006pBAD24/FliC ( ∆29–32/∆310–495 ) This studypAOA007pBAD24/FliC ( ∆450–495 ) This studypAOA008pBAD24/FliC ( ∆29–32/∆450–495 ) This study DNA manipulations were carried out as described before ( Hara et al . , 2011 ) . Site-directed mutagenesis was carried out using QuickChange site-directed mutagenesis method as described by Agilent Technologies , Santa Clara , CA , USA . DNA sequencing reactions were carried out using BigDye v3 . 1 as described in the manufacturer’s instructions ( Applied Biosystems , Foster City , CA , USA ) , and then the reaction mixtures were analysed by a 3130 Genetic Analyzer ( Applied Biosystems ) . To check motility of the Salmonella SJW1103 ( wild-type ) and TM113 ( ∆fliC ) cells carrying a pBAD24-based plasmid encoding wild-type or FliC deletion variants , motility assays were performed in soft agar plates . Single colonies of the cells were inoculated into soft agar plates containing ampicillin and 0 . 2% arabinose . Plates were then incubated at 30°C for the required period of time . Their motility was observed as a ring of migrating cells emanating from the point of inoculation . Salmonella cells were grown with shaking in 5 ml of LB at 30°C until the cell density had reached an OD600nm of approximately 1 . 0–1 . 2 . To see the effect of the flagellar filament on flagellin transport during filament assembly , the cultures were heated at 65°C for 5 min to depolymerize the filaments into flagellin monomers and were centrifuged to obtain cell pellets and culture supernatants , which contains the cytoplasmic flagellin subunits and flagellins transported by the flagellar type III export apparatus , respectively . To test the effect of flagellin subunit linkage on the flagellar growth rate ( compare Figure 5c ) , strain MM1103CPOP carrying a pBAD24-based plasmid encoding FliC ( ∆310–495 ) , FliC ( ∆29–32/∆310–495 ) , FliC ( ∆450–495 ) or FliC ( ∆29–32/∆450–495 ) was grown with shaking in 5 ml of LB containing ampicillin at 30°C until the cell density had reached an OD600 of approximately 0 . 6–0 . 8 . To induce the expression of chromosomally encoded wild-type FliC ( from a tetracycline-inducible promoter in the native fliC locus ) and its deletion variant ( from an arabinose-inducible promoter encoded on pBAD24 ) , we added tetracycline and L-arabinose at the final concentrations of 15 µg/ml and 0 . 2% , respectively , and the incubation was continued for another hour . The cultures were directly heated at 65°C for 5 min , followed by centrifugation to obtain cell pellets and culture supernatants . Cell pellets were resuspended in the SDS-loading buffer , normalized to a cell density to give a constant amount of cells . Proteins in the culture supernatants were precipitated by 10% trichloroacetic acid , suspended in the Tris/SDS loading buffer and heated at 95°C for 3 min . After SDS-PAGE , both CBB-staining and immunoblotting with polyclonal anti-FliC antibodies were carried out as described before ( Minamino and Macnab , 1999 ) . Detection was performed with an ECL plus immunoblotting detection kit ( GE Healthcare , Tampa , FL , USA ) . At least six independent experiments were performed . Salmonella cells were grown with gentle shaking in 5 ml of LB at 30°C until the cell density had reached an OD600 of approximately 1 . 0 . After centrifugation , the cell pellets and the culture supernatants were collected separately . The culture supernatants were ultracentrifuged at 85 , 000 × g for 1 hr at 4°C and the pellets and the supernatants , which contain flagellar filaments detached from the cell bodies during shaking culture and flagellin monomers leaked out the culture media during filament formation , respectively , were collected separately . The cell pellets were suspended in 5 ml PBS and then were heated at 65°C for 5 min , followed by centrifugation to obtain the cell pellets and supernatants , which contained the cytoplasmic flagellin molecules and depolymerized flagellin monomers , respectively . The cell pellets and the pellet fractions after ultracentrifugation were resuspended in the SDS-loading buffer , normalized to the cell density to give a constant amount of cells . Proteins in the supernatants were precipitated by 10% trichloroacetic acid , suspended in Tris/SDS loading buffer and heated at 95°C for 3 min . After SDS-PAGE , both CBB-stating and immunoblotting with polyclonal anti-FliC antibodies were carried out . At least six independent experiments were performed . Salmonella cells were exponentially grown with gentle shaking in 5 ml LB at 30°C . 5 µl of the cell culture were applied to carbon-coated copper grids and negatively stained with 0 . 5% ( W/V ) phosphotungstic acid . Micrographs were recorded at a magnification of 1200× with a JEM-1010 transmission electron microscope ( JEOL , Tokyo , Japan ) operating at 100 kV . For immunolabelling of flagellar filaments , polyclonal anti-FliC and anti-rabbit IgG antibodies conjugated with Alexa Fluor 488 and 594 ( Invitrogen , Carlsbad , CA , USA ) were used as described ( Erhardt et al . , 2011; Minamino et al . , 2014 ) . For in situ labelling of flagellar filaments of the FliCT237C cysteine replacement mutant , an overnight culture was diluted 1:100 into 10 ml fresh TB in a 125 ml flask and grown at 30°C for 2 hr until OD600nm of 0 . 6 . Production of the flagellar master regulatory operon flhDC was induced by addition of 100 ng/ml anhydrotetracycline ( AnTc ) for 30 min . Afterwards , the culture was centrifuged for 3 min at 2500 × g , resuspended in 10 ml fresh TB and grown at 30°C for 30 min . An aliquot was transferred to a 2 ml Eppendorf tube and grown with shaking at 30°C for 30 min in the presence of 10–25 µM Alexa or DyLight-coupled maleimide dye ( ThermoFisher , Tampa , FL , USA ) . After the incubation , the dye was removed by centrifugation for 2 min at 2500 × g . The culture was resuspended in 1 mL fresh TB and incubated for additional 30 min in the presence of 10–25 µM Alexa or DyLight-coupled maleimide dye at 30°C . Dye removal and incubation with DyLight-coupled maleimide dye was repeated to label up to six flagellar filament fragments . After the final labelling period , the bacteria were resuspended in PBS and an aliquot was applied to a custom-made flow cell ( Wozniak et al . , 2010 ) with the modification of using Polysine microscope slides ( ThermoFisher ) . Non-adhering cells were flushed by addition of PBS and bacteria were fixed by addition of 2% formaldehyde , 0 . 2% glutaraldehyde in PBS for 5 min , followed by a washing step with PBS . Fluoroshield mounting medium ( Sigma-Aldrich , St . Louis , MO , USA ) was added and the cells were observed by fluorescent microscopy using a Zeiss ( Oberkochen , Germany ) Axio Observer microscope at 100× magnification . Fluorescence images were analysed using ImageJ software version 1 . 48 ( National Institutes of Health ) . Continuous flow in situ immunostaining of 3× hemagglutinin epitope tagged FliC filaments was performed as described by Berk et al . ( 2012 ) with the following adaptions . Strain EM4076 expressing mCherry from pZS*12-mCherry ( mCherry under control of Plac [Lutz and Bujard , 1997] ) was grown to mid-log phase in M9-glucose minimal medium supplemented with 0 . 2% casamino acids and 0 . 1% bovine serum albumin ( BSA ) and induced for 30 min with 100 ng/ml AnTc . Bacteria were diluted 10-fold , and applied to a continuous flow CellASIC microfluidic plate ( B04A; Merk Millipore , Billerica , MA , USA ) . Approximately 10 nM anti-HA Alexa Fluor488 fluorochrome-coupled primary antibodies ( Thermo Fisher A-21287 , final concentration 1 µg/ml ) were added to the flow medium , which was identical to the above mentioned growth medium without addition of AnTc . Cells were imaged at 30°C with a temperature-controlled Olympus total internal reflection fluorescence microscope equipped with a water-cooled Hamamatsu ( Hamamatsu City , Japan ) ImageEM C9100-13 with a pixel size of 160 µm using a NA1 . 4 100× objective and an additional 1 . 6× tubular lens at a highly-inclined above-critical angle . To image anti-HA Alexa Fluor488 decorated flagellin and mCherry , a 488 nm diode laser set to 0 . 25 mW and a 561 nm solid-state laser set to 0 . 85 mW were used . Images were taken every 10 s with exposure times of 15 msec for 488 nm and 8 msec for 561 nm at low camera gain settings . No statistical methods were used to predetermine sample size . The experiments were not randomized and the investigators were not blinded to allocation during experiments and outcome assessment . Biochemistry experiments were performed at least three times and representative experiments are reported in the figures . Where indicated , mean values and standard deviations were obtained from at least three independent biological replicates . All microscopy experiments were performed at least twice and the figures present individual data points of a representative experiment . Box plots report the median ( in red ) , the 25th and 75th quartiles and the 1 . 5 interquartile range . Error bars of bar graphs represent the 95% confidence interval of mean estimation . To compare the model with data , we need to find a best fit for the parameters a and b using the growth function ( Equation 4 ) . Accordingly , note that if F1 is the amount of filament growth in time ∆T following an initial growth of length F0 , then ( 7 ) ∫F0F0+F1 ( b+L ) dL=aΔT , which reduces to the equation ( 8 ) L ( L+2b ) \|F0+F1F0=2aΔT , or ( 9 ) bF1+12 ( 2F0F1+F12 ) =aΔT . This could be solved for F0 as a function of F1 and then fitted by standard regression to find parameters a and b . However , to do so would ignore the important fact that there is measurement error in both of the measurements F0 and F1 . Consequently , a different method of fitting this curve is needed . The method used here is to seek numbers W0 and W1 , which are approximations to F0 and F1 and satisfy the relationship ( 10 ) bW1+12 ( 2W0W1+W12 ) =aΔT This can be done by minimizing the function ( 11 ) E=∑N ( ( F0−W0 ) 2+ ( F1−W1 ) 2+λ ( bW1+12 ( 2W0W1+W12 ) −aΔT ) 2 ) , where λ is a fixed constant . In this way , both F0 and F1 are treated as noisy data values which need to be fitted . However , for this analysis , we found it better to introduce the change of variables L=bU1−U=g ( U ) , U=LL+b and then to find numbers U0 and U1 , α=aΔTb2 and b so that ( 12 ) E=∑N ( ( F0−bg ( U0 ) ) 2+ ( F0+F1−bg ( U1 ) ) 2+λ ( f ( U1 ) ) −f ( U0 ) −α ) 2 ) is minimized , where f ( U ) =1b2 ( bL+12L2 ) ≡12U ( 2−U ) 2 ( 1−U ) 2 . The minimization of E is equivalent to finding the solution of the system of 2N + 2 nonlinear algebraic equations ( 13 ) ∂∂α:∑N ( f ( U1 ) −f ( U0 ) −α ) =0 , ( 14 ) ∂∂b:∑N ( F0−bg ( U0 ) ) g ( U0 ) +∑N ( F0+F1−bg ( U1 ) ) g ( U1 ) =0 , ( 15 ) ∂∂U0:b ( F0−bg ( U0 ) ) g′ ( U0 ) +λ ( fU1 ) −f ( U0 ) −α ( f′ ( U0 ) =0 , ( 16 ) ∂∂U1:−b ( F0+F1−bg ( U1 ) ) g′ ( U1 ) +λ ( f ( U1 ) −f ( U0 ) −α ) f′ ( U1 ) =0 . This system of equations is readily solved with an iterative solution method such as Newton’s Method , details of which are not described here . Once U0 and U1 are known , so also are W0=bU01−U0 and W1=bU11−U1−W0 . From this we can estimate the time at which the F0 phase of growth ended to be ( 17 ) t0=1a ( bW0+W022 ) , and the time at which the F1 growth phase ended is t1=t0+ΔT . This information enables us to plot the growth curve and plot the F0 and F1 measurements at the appropriate times . Substrates arriving at the export gate were randomly chosen with a probability p = r / ( 1+r ) to be a competing substrate ( i . e . , non-chaining or not incorporated in the filament ) , where r is the ratio of competing molecules relative to flagellin . The following rules were used: In Figure 5 panel g and Figure 5—figure supplement 1 panel d , we simulated filament growth over 250 min for 20 , 000 filaments and assumed a mechanism based on the chain model ( strict in blue , with recovery in yellow ) or the injection-diffusion model ( in red ) , in the presence of a random proportion of competing substrates ( r ) between 10−9 and 1 , 000 . The simulation of chain model-dependent filament growth is illustrated in Figure 5—figure supplement 2 . The range of competing substrates compatible with a chain-driven elongation is very low ( <10−4–10−5 ) , while the injection-diffusion model allows for robust filament growth over a much broader range of competing substrate ( up to about a 10-fold excess of competing substrates ) . Complementary to the simulation , the median length of the filament under chain model-dependent growth and in presence of competing substrates can be calculated as follows: The probability of sequentially forming a chain of exact length n is Pn=pn ( 1−p ) . The expected number of molecules in the chain is: ( 18 ) E ( p ) = ( 1−p ) ∑nnpn=p1−p=1x Thus , the median length of a filament grown from a continuous chain is kβ , where β = 0 . 47 nm and k can be determined by:12=∑nkPn=∑nk ( 1−p ) pn=1−pk+1 , which leads to: ( 20 ) k=ln2ln ( 1+x ) −1 .	Most bacteria are able to move in a directed manner towards nutrients or other locations of interest . Many move by rotating long tail-like filaments called flagella that stick out from the cell . The flagellum is a remarkably complex nanomachine . It is several times longer than the main body of the bacterial cell body and its external filament is made of thousands of building blocks of a single protein called flagellin . This protein is made inside the cell and a structure at the base of the flagellum known as a type III secretion system uses chemical energy to pump it out of the cell so that it can be incorporated into the growing flagellum . The exported building blocks travel through a narrow channel within the flagellum and self-assemble at the tip . It has been a mystery for several decades how bacteria manage to assemble the building blocks of flagella outside of the cell , where no discernible energy source is available . Renault et al . used mathematical modeling , biochemical and microscopy techniques to observe how the flagella of a bacterium called Salmonella enterica assemble in real time . The experiments demonstrate that simple biophysical principles regulate the assembly of the flagellum . The building blocks are pumped into the channel of the flagellum by the type III secretion system and then diffuse to the tip of the filament . Accordingly , the longer the flagellum gets , the slower it grows . This molecular mechanism also explains why the growth of bacterial flagella will eventually stop even without any other control mechanisms in place . Further work will be needed to understand how the type III secretion system harnesses chemical energy to drive the movement of flagellin out of the cell into the growing flagellum . A molecular understanding of these processes will aid the design of new antibiotics targeted against type III secretion systems .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics", "microbiology", "and", "infectious", "disease" ]	2017	Bacterial flagella grow through an injection-diffusion mechanism
Value-based decisions could rely either on the selection of desired economic goods or on the selection of the actions that will obtain the goods . We investigated this question by recording from the supplementary eye field ( SEF ) of monkeys during a gambling task that allowed us to distinguish chosen good from chosen action signals . Analysis of the individual neuron activity , as well as of the population state-space dynamic , showed that SEF encodes first the chosen gamble option ( the desired economic good ) and only ~100 ms later the saccade that will obtain it ( the chosen action ) . The action selection is likely driven by inhibitory interactions between different SEF neurons . Our results suggest that during value-based decisions , the selection of economic goods precedes and guides the selection of actions . The two selection steps serve different functions and can therefore not compensate for each other , even when information guiding both processes is given simultaneously . Value-based decision-making requires the ability to select the reward option with the highest available value , as well as the appropriate action necessary to obtain the desired option . Currently it is still unclear how the brain compares value signals and uses them to select an action ( Gold and Shadlen , 2007; Cisek , 2012 ) . The goods-based model of decision-making ( Padoa-Schioppa , 2011 ) suggests that the brain computes the subjective value of each offer , selects one of these option value signals , and then prepares the appropriate action plan ( Figure 1A ) . Support for this model comes from recording studies in orbitofrontal cortex ( OFC ) during an economic choice task ( Padoa-Schioppa and Assad , 2006; Cai and Padoa-Schioppa , 2012 ) . In contrast , the action-based model of decision making ( Tosoni et al . , 2008; Cisek and Kalaska , 2010; Christopoulos et al . , 2015a ) suggests that all potential actions are represented in the brain in parallel and compete with each other ( Figure 1B ) . This competition is influenced by a variety of factors including the value of each actions’ outcome . According to this model , option value signals should not predict the chosen option , since these signals only serve as input into the decision process , which is determined by the competition among the potential actions . Support for this model comes primarily from recording studies in parietal and premotor cortex ( Platt and Glimcher , 1999; Sugrue et al . , 2004; Shadlen et al . , 2008; Cisek and Kalaska , 2010; Christopoulos et al . , 2015b ) . 10 . 7554/eLife . 09418 . 003Figure 1 . Architecture of different decision models . ( A , B ) Goods- and action-based models envision the important selection step during value-based decisions to be either at the value ( A ) or action ( B ) representation stage . ( C , D ) The other two models presume that important selection processes occur at both the value and the action representation stage . However , they differ in their underlying architecture and in the resulting pattern of activity across the network as it unfolds in time . ( C ) The distributed consensus model assumes reciprocal interactions between the value and the action representation . These reciprocal interactions allow the action selection to influence the simultaneous ongoing value selection . The selection of the chosen good and action proceeds therefore in parallel . ( D ) In contrast , the sequential model assumes that there are no meaningful functional reciprocal connections from the action to the value representation . Because of this the action value representations cannot influence the value selection process , which has to finish first , before the action selection can begin . Thus , this decision architecture by necessity implies a sequential decision process . Red arrows indicate excitatory connections . Green buttons indicate inhibitory connections . Thickness of the connection indicates relative strength of the neural activity . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 003 As there is evidence supporting both theories , it is unlikely that either the goods-based or the action-based model in their pure form are correct . However , the exact role of goods- and action-based selection processes in decision making is not known . The distributed consensus model ( Cisek , 2012 ) combines elements of the goods-based and the action-based model ( Figure 1C ) . It is characterized by strong reciprocal interactions between the goods and the action representation levels that allow the action selection to influence the simultaneous ongoing value selection and vice versa . This model predicts therefore that the selection of the chosen good and action are closely integrated and proceed in parallel . Here , we test these different models by recording neuronal activity in the supplementary eye field ( SEF ) . Previous research indicates that neurons in the SEF participate in the use of value signals to select eye movements ( So and Stuphorn , 2010 ) . Its anatomical connections make the SEF ideally suited for this role . It receives input from areas that represent option value , such as OFC , ACC , and the amygdala ( Huerta and Kaas , 1990; Matsumoto et al . , 2003; Ghashghaei et al . , 2007 ) , and projects to oculomotor areas , such as frontal eye field ( FEF ) and superior colliculus ( Huerta and Kaas , 1990 ) . We designed an oculomotor gamble task , in which the monkey had to choose between two gamble options indicated by visual cues . The monkeys indicated their choice by making a saccade to the cue indicating the desired gamble option . Across different trials , the visual cues were presented in different locations and required saccades in different directions to be chosen . This allowed us to distinguish the selection of gamble options or economic goods from the selection of actions . We found that the activity of SEF neurons predicted the monkey’s choice . Importantly , this decision process unfolded sequentially . First , the chosen gamble option was selected and only then the chosen action . The saccade selection process seemed to be driven by competition between directionally tuned SEF neurons . Our findings are not in agreement with any of the previously suggested models ( Figure 1A–C ) . Instead , they support a new sequential decision model ( Figure 1D ) . According to this model , at the beginning of the decision two selection processes start independently on the goods and action level . Our data indicate that the SEF activity is part of the action selection process . The action selection process receives input from the goods selection . However , due to the absence of recurrent feedback , the goods selection process does not receive input from the action selection process . Once the competition on the goods level is resolved , the value signals for the chosen gamble option increase in strength and the ones for the non-chosen one decrease in strength . This activity difference cascades downward to the action level and determines the outcome of the action selection . Two monkeys ( A and I ) were trained to perform a gambling task in which they chose between two different gamble options with different maximum reward and/or reward probability ( Figure 2A , B ) . The maximum and minimum reward amounts were indicated by the color of the target . The portion of a color within the target corresponded to the probability of receiving the reward amount ( see experimental procedures ) . We estimated the subjective value for each target based on the choice preference of the monkeys for all combinations of options ( Figure 2C ) ( Maloney and Yang , 2003; Kingdom and Prins , 2010 ) . The subjective value estimate ( referred to in the rest of the paper as ‘value’ ) is measured on a relative scale , with 0 and 1 being the least and most preferred option in our set . Consistent with earlier findings ( So and Stuphorn , 2010 ) , the mean saccade reaction times during no-choice trials were significantly negatively correlated with the value of the target ( Figure 2D , monkey B: t ( 5 ) = 8 . 40 , p = 0 . 03; monkey I: t ( 5 ) = 27 . 35 , p = 0 . 003 ) . On choice trials , the mean reaction times were significantly correlated with the signed value difference between chosen and non-chosen targets ( Figure 2E , monkey A: t ( 40 ) = 159 . 23 , p<10 –14; monkey I: t ( 38 ) = 16 . 18 , p<10–4 ) . Please note that there were a small number of trials with very negative value differences indicating that during these trials the monkey chose a normally non-preferred option . The unusually short reaction time in these trials suggests that the choices were not driven by the normal value-based decision process . These other mechanisms may include history effects , spatial selection bias , express saccades , lapse of attention to the task , and other factors . For this reason , these trials were excluded from the analysis and are marked by a separate color . 10 . 7554/eLife . 09418 . 004Figure 2 . Oculomotor gambling task and behavioral results . ( A ) Visual cues used in the gambling task . Four different colors ( cyan , red , blue , and green ) indicated four different reward amounts ( increasing from 1 , 3 , 5 to 9 units of water , where 1 unit equaled 30 µl ) . The expected value of the gamble targets along the diagonal axis was the same . For example , the expected value of the bottom right green/cyan target is: 9 units ( maximum reward ) x 0 . 2 ( maximum reward probability ) + 1 unit ( minimum reward ) x 0 . 8 ( minimum reward probability ) = 2 . 6 units . ( B ) Sequence of events during choice trials ( top ) and no-choice trials ( bottom ) . The lines below indicate the duration of various time periods in the gambling task . The black arrow is not part of the visual display; it indicates the monkeys' choices . ( C–E ) Behavioral results for monkey A ( top ) and monkey I ( bottom ) . ( C ) The mean subjective value of the seven gamble options is plotted as a function of expected value . Different colors indicate different amounts of maximum reward . ( D ) The mean reaction times in no-choice trial as a function of subjective value . ( E ) The mean reaction times in choice trial as a function of subjective value differences between chosen and non-chosen targets . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 00410 . 7554/eLife . 09418 . 005Figure 2—figure supplement 1 . Recording locations in SEF . Red dots indicate the locations in which neurons showed task related activity before saccade onset . Blue dots indicate the locations in which neurons were not modulated by task before saccade onset . ( A ) Recording sites in monkey A . ( B ) Recording sites in monkey I . SEF , supplementary eye field . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 005 We recorded 516 neurons in SEF ( 329 from monkey A , 187 from monkey I , Figure 2—figure supplement 1 ) . In the following analysis , we concentrated on a subset of 128 SEF neurons , whose activity was tuned for saccade direction ( see experimental procedures ) . First , we asked whether SEF activity predicted the chosen gamble option or saccade direction . We performed a trial by trial analysis using linear classification to decode the chosen direction and chosen value from the spike density function with 1 ms temporal resolution . Figure 3A shows the classification accuracy across all 128 directionally-tuned SEF neurons . Single neuron activity clearly predicted both chosen gamble option and chosen direction better than chance , but sequentially , not simultaneously . The activity began to predict chosen gamble option around 160 ms before saccade onset and reached a peak around 120 ms before saccade onset , after which it gradually decreased . The activity started to predict saccade direction only around 60 ms before saccade onset . The same pattern is shown by the number of neurons showing significant classification accuracy as a function of time ( Figure 3B ) . 10 . 7554/eLife . 09418 . 006Figure 3 . Time course of chosen gamble option and saccade direction representation in SEF . ( A ) Significant classification accuracy for chosen gamble option ( red ) and chosen saccade direction ( black ) across 128 neurons . We excluded values that were not significantly different from chance ( permutation test; p≤0 . 05 ) . ( B ) Number of neurons showing significant classification accuracy for chosen gamble option ( red ) and chosen saccade direction ( black ) . ( C ) Average mutual information between SEF activity and chosen and non-chosen gamble option ( top panel; dark and light red ) and saccade direction ( bottom panel; dark and light grey ) . The time period when the amount of information about chosen and non-chosen option/direction was significantly different ( paired t-test adjusted for multiple comparisons , p≤0 . 05 ) are indicated by the thick black line at the bottom of the plots . The onset of a significant difference is indicated by the vertical dashed line . SEF , supplementary eye field . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 00610 . 7554/eLife . 09418 . 007Figure 3—figure supplement 1 . Time course of value and saccade direction representation in SEF aligned on target onset . In our gamble task , the monkey was free to indicate his choice as soon as he was ready . Because of this design feature , the saccade onset is likely to be closer aligned with the conclusion of the decision process than target onset . The fact that reaction time reflected chosen value and non-chosen value difference ( i . e . choice difficulty ) , as indicated in Figure 2 , further confirms this idea . In the main text , we analyzed therefore the neural activity aligned on movement onset , because it likely reflects the dynamic of the decision process more accurately . The analysis of the neural activity aligned on target onset further confirms this conclusion . ( A ) Average significant classification accuracy for chosen value ( red ) and chosen direction ( black ) across 128 neurons . The decoding analysis shows that the SEF activity predicts chosen value early in the trial , but only weakly and over a widely spread out time period . Only late in the trial did the prediction accuracy of the chosen value increase . At this stage , the prediction of the saccade direction had already begun to increase . ( B ) Number of neurons showing significant classification accuracy for chosen value ( red ) and chosen direction ( black ) . The time course of the number of neurons with significant predictions shows a similar pattern , which likely reflects the variable relationship between target onset and decision making . ( C ) Average mutual information between SEF activity and chosen and non-chosen value ( top panel; dark and light red ) and saccade direction ( bottom panel; dark and light grey ) . Note that the onset of significant levels of chosen value and direction signals in the population are both late relative to target onset and nearly simultaneous . The time period when the amount of information about chosen and non-chosen direction/value was significantly different are indicated by the thick black line at the bottom of the plots . The onset of a significant difference is indicated by the vertical dashed line . The shaded areas represent SEM . Compare this figure with Figure 3 in the main text . SEF , supplementary eye field . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 007 This result indicates a sequence of decisions , whereby first an economic good , here a gamble option , is chosen and only later the action that will bring it about . To confirm this finding , we employed an independent information theoretic analysis to study how SEF activity encoded the chosen and non-chosen gamble option , as well as the chosen and non-chosen direction throughout the decision process ( Figure 3C ) . We used 106 neurons which were tested with at least 8 out of 12 possible target position combinations . We assumed that the onset of significantly more information about the chosen than the non-chosen variable in the neural firing rate indicated the moment at which the selection process had finished and the choice could be predicted . This moment was reached 66 ms earlier for gamble option information ( 113 ms before saccade onset; permutation test adjusted for multiple comparison , p≤0 . 05 ) than for saccade direction information ( 47 ms before saccade onset; permutation test adjusted for multiple comparisons , p≤0 . 05 ) . Thus , the onset and timing of the information representation in SEF is consistent with the results of the classification analysis and also indicates a sequential decision process . In our gamble task , the monkey was free to indicate his choice as soon as he was ready . Because of this design feature , the saccade onset is likely to be closer aligned with the conclusion of the decision process than target onset . The fact that reaction time reflected chosen value and value difference ( i . e . choice difficulty ) , as indicated in Figure 2 , also supports this idea . We analyzed therefore the neural activity aligned on movement onset , because it likely reflects the dynamic of the decision process more accurately . The analysis of the neural activity aligned on target onset further confirms this conclusion ( Figure 3—figure supplement 1 ) . Our findings indicated that SEF neurons show signs of a sequential decision process , whereby first a desired economic good is chosen and only then the action that is necessary to obtain the good . Next , we investigated the neural activity in the SEF neurons more closely to test if the SEF only reflects the outcome of the decision , or whether it also reflects one or both of the selection steps . Specifically , we searched for opposing contributions of the two choice options to the activity of SEF neurons , which would indicate a competitive network that could select a winning option from a set of possibilities . The directionally tuned SEF neurons represented the value of targets in the preferred direction ( PD ) ( Table 1 ) . The PD is the saccade direction for which a neuron is maximally active , irrespective of reward value obtained by the saccade . We estimated each neuron’s PD using a non-linear regression analysis of activity for saccades to all four possible target locations . We defined here PD as the target direction that is closest to the estimated PD . Figure 4A , B shows the activity of the SEF neurons during no-choice trials , that is , when only one target appears on the screen . Although the SEF neurons are strongly active for PD targets and show value-related modulations ( Figure 4A ) , they are not active for saccades into the non-preferred direction ( NPD ) , independent of their value ( Figure 4B; regression coefficient = 0 . 013 , t ( 5 ) = 1 . 324 , p=0 . 243 ) . The SEF neurons encode , therefore , the value of saccades to the PD target , confirming previous results ( So and Stuphorn , 2010 ) . 10 . 7554/eLife . 09418 . 008Figure 4 . SEF neurons represent the difference in action value associated with targets in the preferred and non-preferred direction . The neural activity of 128 directionally tuned SEF neurons was normalized and compared across trials with different values of targets in the preferred or non-preferred direction . ( A ) The neural activity in no-choice trials , when the target was in the preferred direction . ( B ) The neural activity in no-choice trials , when the target was in the non-preferred direction . ( C , D ) The neural activity in choice trials . To visualize the contrasting effect of targets in the preferred or non-preferred direction on neural activity , the value of one of the targets was held constant , while the value of the other target was varied . Activity was sorted by target value , but not by saccade choice . ( C ) The neural activity , when the value of the target in the preferred direction varied , while the value of the target in the non-preferred direction was held constant at a medium value . ( D ) The neural activity , when the value of the target in the non-preferred direction varied , while the value of the target in the preferred direction was held constant at a medium value . The color of the spike density histograms indicates the target value [high value = 6–7 units ( red line ) ; medium value = 3–5 units ( orange line ) ; low value = 1–2 units ( yellow line ) ] . ( E-H ) The regression analysis corresponding to ( A-D ) . A t-test was used to determine whether the regression coefficients were significantly different from 0 . The regression coefficients , confidence intervals , t-values , and p-values are listed in Table 1 . SEF , supplementary eye field . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 00810 . 7554/eLife . 09418 . 009Figure 4—figure supplement 1 . SEF neurons represent the difference in value of targets in the preferred and non-preferred direction . All activity is aligned on saccade onset . ( A ) The neural activity in no-choice trials , when the target was in the preferred direction . ( B ) The neural activity in no-choice trials , when the target was in the non-preferred direction . ( C , D ) The neural activity in choice trials . To visualize the contrasting effect of targets in the preferred or non-preferred direction on neural activity , the value of one of the targets was held constant , while the value of the other target was varied . Activity was sorted by target value , but not by saccade choice . ( C ) The neural activity , when the value of the target in the preferred direction varied , while the value of the target in the non-preferred direction was held constant at a medium value . ( D ) The neural activity , when the value of the target in the non-preferred direction varied , while the value of the target in the preferred direction was held constant at a medium value . The color of the spike density histograms indicates the target value [high value = 6–7 units ( red line ) ; medium value = 3–5 units ( orange line ) ; low value = 1–2 units ( yellow line ) ] . SEF , supplementary eye field . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 00910 . 7554/eLife . 09418 . 010Figure 4—figure supplement 2 . SEF neurons represent the difference in value of targets in the preferred ( PD ) and non-preferred direction ( NPD ) independent of the chosen saccade direction . This figure shows the normalized activity of SEF neurons sorted by chosen saccade direction ( PD: right; NPD: left ) and value of the chosen or non-chosen target . We grouped the subjective value of the reward options into three groups ( red: high value; orange: medium value; yellow: low value ) . The inset above the histograms indicates the location and value of the targets in the trials shown in the histograms below . The grey oval indicates the location of the preferred direction of the neuron , while the arrow indicates the chosen saccade direction . The p-values indicate the significance of a regression using all seven individual target values without grouping . The shaded areas represent standard error of the mean ( SEM ) . ( A ) The neural activity in no-choice trials . The color of the spike density histograms indicates the target value . As shown in Figure 4A , B , the neurons reflect the value of the PD target , but not of the NPD target . ( B ) The neural activity in those choice trials , in which the chosen target had the highest possible value ( 7 units ) . We chose this reference point , instead of a medium value ( as in Figure 4D ) , because it allowed us to a comparison with the widest possible range of non-chosen target values ( indicated by the color of the spike density histograms: high value = 5–6 units; medium value = 3–4 units; low value = 1–2 units ) . ( C ) The neural activity in those choice trials , in which the non-chosen target had the lowest possible value ( 1 unit ) . Again , this reference point allowed for the widest possible range of chosen target values ( indicated by the color of the spike density histograms: high value = 6–7 units; medium value = 4–5 units; low value = 2–3 units ) . Consistent with Figure 4 in the main text , we can observe that increasing PD target value increases neuronal activity ( B: left panel; C: right panel ) , while increasing NPD target value decreases neuronal activity ( C: left panel ) . This is true , whether the PD or the NPD target is chosen . However , the value of the chosen target has a stronger influence than the non-chosen one , and in the extreme case , when the PD target has the highest possible value and is chosen , the NPD value has no longer a significant effect on the neural activity ( B: right panel ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 01010 . 7554/eLife . 09418 . 011Figure 4—figure supplement 3 . SEF neurons represent the difference in action value associated with targets in the preferred and non-preferred direction . This analysis is the exact equivalent of the one shown in Figure 4 , but here we analyzed the neural activity of all 353 task-related SEF neurons . The neuronal activity was normalized and compared across trials with different values of targets in the preferred or non-preferred direction . ( A ) The neural activity in no-choice trials , when the target was in the preferred direction . ( B ) The neural activity in no-choice trials , when the target was in the non-preferred direction . ( C ) The neural activity , when the value of the target in the preferred direction varied , while the value of the target in the non-preferred direction was held constant at a medium value . ( D ) The neural activity , when the value of the target in the non-preferred direction varied , while the value of the target in the preferred direction was held constant at a medium value . The color of the spike density histograms indicates the target value [high value = 6–7 units ( red line ) ; medium value = 3–5 units ( orange line ) ; low value = 1–2 units ( yellow line ) ] . ( E-H ) The regression analysis corresponding to ( A-D ) . A t-test was used to determine whether the regression coefficients were significantly different from 0 . SEF , supplementary eye field . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 01110 . 7554/eLife . 09418 . 012Figure 4—figure supplement 4 . Neural activity modulated by the relative angle and position of the two targets . ( A ) The activity of a representative neuron aligned on target ( two left columns ) and saccade onset ( two right columns ) averaged over all possible pairs of target values . The first column and third column show the conditions when the non-preferred direction was chosen , the second and forth column show the conditions when the preferred direction was chosen . The color of the lines indicate the angle between the targets and their relative position in the visual field . Independent of whether the chosen saccade is into the preferred or non-preferred direction , the neural activity is least affected by targets that are far away ( light gray; 180° apart ) . However , for alternative targets that are closer to the location of the chosen target , the neuron shows reduced activity ( medium gray; 90° apart , contralateral hemifield ) , in particular when the target is in the same hemi-field ( black; 90° apart , ipsilateral hemi-field ) . This indicates that spatially distinct targets evoke inhibitory interactions among SEF neurons that become stronger the closer the targets are to each other , independent of their value . ( B ) Average neural activity across 128 neurons . The plots show the same inhibitory effect as the representative neuron . The shaded areas represent SEM . SEF , supplementary eye field; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 01210 . 7554/eLife . 09418 . 013Table 1 . Average value effect on neural activity across all directional SEF neurons . The upper two rows show the effect of preferred and non-preferred direction target value on normalized neuronal activity in no-choice trials , and the lower two rows show their effect in choice trials . Within each set , the upper row ( VPD ) shows the effect of the preferred direction target value on normalized neural activity , whereas the lower row shows the effect of the non-preferred direction target value on normalized neural activity . Significance was calculated using a t-test , which shows whether the regression coefficient is significant difference from zero . The analysis corresponds to the results presented in Figure 4 . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 013All neurons ( n = 128 ) Regression coefficientLower confidence boundUpper confidence boundt ( 5 ) pNo-choiceVPD0 . 0570 . 0200 . 0953 . 9450 . 011VNPD0 . 013-0 . 0120 . 0391 . 3240 . 243ChoiceVPD0 . 1190 . 0900 . 14810 . 629<0 . 001VNPD-0 . 058-0 . 093-0 . 0238-4 . 3450 . 007SEF , supplementary eye field . There are a number of subtypes of value signals that are associated with actions , such as saccades . These signals are related to different stages of the decision process ( Schultz , 2015; Stuphorn , 2015 ) . First , there are signals that represent the value of the alternative actions irrespective of the choice . These signals represent the decision variables on which the decision process is based and are commonly referred to as ‘action value’ signals . Second , there are signals that encode the central step in the decision process , namely the comparison between the values of the alternative actions . Such ‘relative action value’ signals represent a combination of different ‘action value’ signals . They should be positively correlated with the action value of one alternative and negatively correlated with the action value of the other alternatives . Third , there are signals that indicate the value of the chosen action . These ‘chosen action value’ signals represent the output of the decision process . Single target trials do not allow to distinguish between these functionally very different signals , but choice trials do . On their own , NPD targets did not evoke neural activity . However , the value of NPD targets clearly modulated the response of the SEF neurons to the PD targets in choice trials ( Figure 4D; Table 1 ) . To isolate the effect that targets in the two directions have on the neural activity , we first held the value of the NPD target constant at a medium amount and compared the SEF population activity across trials with PD targets of varying value ( Figure 4B ) . The neural activity clearly increased with the value of the PD target ( regression coefficient = 0 . 119 , t ( 5 ) = 10 . 629 , p<0 . 001; Figure 4E ) . Next , we held the value of the PD target constant at a medium value and compared the population activity across trials with NPD targets of varying value ( Figure 4C ) . The neural activity clearly decreased with the value of the NPD target ( regression coefficient=-0 . 058 , t ( 5 ) =-4 . 345 , p=0 . 007; Figure 4F ) . Thus , the SEF neurons represented a relative action value signal . The value of the PD target influences the SEF activity about twice as large as the value of the NPD target . This means that the SEF neurons do not encode the exact value difference . Nevertheless , the opposing influence of the targets indicates that SEF represents the essential step in decision-making , namely a comparison of the relative value of the available actions . All these effects were present well before saccade onset ( Figure 4—figure supplement 1 ) and did not depend on the chosen saccade direction ( Figure 4—figure supplement 2 ) . Similar activity pattern can also be observed if pooling together all task related neurons ( N=353 , Figure 4—figure supplement 3 ) . In contrast , the neurons were not significantly influenced by the relative spatial location of each target ( Figure 4—figure supplement 4 ) . We used a regression analysis to further quantify the relative contribution of the chosen and non-chosen gamble option and saccade to the neural activity during the decision . In modeling the neural activity in choice trials , we used each neurons’ activity in no-choice trials as a predictor of its response in choice trials . Specifically , we modeled the neural activity as a weighted sum of the activity in no-choice trials for saccades to targets with the same gamble option or direction as the chosen and non-chosen targets in the choice trials . The strength of the coefficients is a measure of the relative influence that each target has on the neural activity in a particular time period during the decision process . Figure 5 shows the time course of the coefficient strength for the chosen and non-chosen target when we sorted trials either by gamble option or saccade direction . In both cases , the correlation coefficients for the two targets were initially of equal value , indicating that the SEF neurons reflected each target equally during this time period . However , 110 ms before saccade onset ( permutation test adjusted for multiple comparisons , p≤0 . 05 ) the strength of the chosen gamble option coefficient started to rise , while the non-chosen gamble coefficient strength stayed the same . Later in the trial , 60 ms before saccade onset ( permutation test adjusted for multiple comparisons , p≤0 . 05 ) the coefficient strength for the chosen saccade direction increased . Simultaneously , the coefficient strength for the non-chosen saccade direction decreased . The results of the regression analysis allow a number of conclusions about the mechanism underlying decision-making and the role of SEF in it . First , the results confirm the findings of the decoding and encoding analysis ( Figure 3 ) and indicate that the decision process involves a sequence of two different selection processes . Second , the opposing pattern of influence on neural activity in the case of saccade direction suggests that the latter action selection step could involve at least partially the SEF . The mechanism underlying this selection involves competition between the action value signals associated with the two saccade targets . Thus , the increasing influence of one target leads simultaneously to a decreasing influence of the other target . However , the neural activity associated with the earlier gamble option selection step does not show such a pattern of competing influences . Instead , the influence of the chosen option just increases . That indicates that the choice of an economic good does not involve competitive interactions between SEF neurons . Instead , only the output of the gamble option selection is represented in SEF . This signal could reflect input from other brain regions . 10 . 7554/eLife . 09418 . 014Figure 5 . Relative influence of chosen and non-chosen target on SEF activity . ( A ) Regression coefficients for chosen and non-chosen gamble options ( dark and light red ) . ( B ) Regression coefficients for chosen and non-chosen saccade directions ( dark and light grey ) . Time periods in which the regression coefficients for chosen and non-chosen option/direction are significantly different ( paired t-test adjusted for multiple comparisons , p≤0 . 05 ) are indicated by a thick black line . The onset of a significant difference is indicated by a vertical dashed line . All panels are aligned on saccade onset . The shaded areas represent SEM . SEF , supplementary eye field; SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 014 Each directionally tuned SEF neuron represents the relative action value of saccades directed toward its PD . Together these neurons form a map encoding the relative value of all possible saccades during the decision process . Our analysis of the activity pattern in individual neurons suggested that the action selection relied on competition between different relative action value signals . In that case , the relative action value map in SEF should contain different groups of neurons that represent the competing relative action values of the two saccade choices . If the activities of these two groups of neurons indeed reflect inhibitory competition , the selection of a particular action should lead to increased activity of the neurons representing the chosen and decreased activity in the neurons representing the non-chosen saccade . Furthermore , the inhibition that the winning neurons can exert on the losing neurons should depend on their relative strength . We should therefore see differences in the dynamic of the neural activity within the relative action value map if we compare trials with small or large value differences . To reconstruct the SEF relative action value map , we combined the activity of all directionally tuned neurons in both monkeys ( Figure 6 ) . We sorted each SEF neuron according to its PD and normalized their activity across all trial types ( choice , no-choice trials ) . We then smoothed the matrix by linear interpolation at a bin size of 7 . 2° and plotted the activity . For each successive moment in time , the resulting vector represented the relative action value of all possible saccade directions , because all task-relevant saccades were equidistant to the fixation point . The succession of states of the map across time represented the development of relative action value-related activity in SEF over the course of decision making . 10 . 7554/eLife . 09418 . 015Figure 6 . Action value maps showing population activity in SEF during decision making . Each neuron’s activity was normalized across all trial conditions . The maps in the left column are aligned on target onset and the panels in the right column on saccade onset . In each map , horizontal rows represent the average activity of cells whose preferred direction lies at a given angle relative to the chosen target ( red circle on left ) . Color indicates change in normalized firing rate relative to the background firing rate ( scale on the right ) . ( A ) Population activity during no-choice ( A ) and choice trials ( B ) . Population activity in choice trials divided into trials with small ( C ) and ( D ) large value differences between the reward options . The subplots above the action value maps show the time course of the neural activity associated with the chosen ( 45–135° ) and non-chosen ( 225–315° ) target . The brown lines underneath show the time when population activities were significantly different than the baseline ( permutation test adjusted for multiple comparisons ) . The blue lines underneath show the time when the neural activities associated with the chosen target were significantly different from those associated with the non-chosen target ( permutation test adjusted for multiple comparisons ) . SEF , supplementary eye field . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 01510 . 7554/eLife . 09418 . 016Figure 6—figure supplement 1 . Action value maps for trials with variable choices . SEF population activity for the option pairs in which the monkeys chose different options across trials . The circle on top indicates the position of the target with the higher average value , while the circle below indicates the position of the target with the lower average value . The red circle indicates the chosen target , while the black circle indicates the non-chosen target . The time-direction maps are aligned on target ( left panels ) and saccade ( right panels ) onset . ( A ) SEF population activity on trials , in which the monkey chose the more valuable target . ( B ) SEF population activity on trials , in which the monkey chose the less valuable target . The SEF neurons whose preferred direction coincided with the chosen target location strongly increased their activity , irrespective of the target’s value . This indicates that SEF neurons not only encode the relative action value of a saccade in their preferred direction , but also the saccade choice . SEF , supplementary eye field . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 016 In choice trials , activity started to rise in two sets of neurons ( Figure 6B ) . One was centered on the chosen target ( indicated by the red dot ) , while the other one was centered on the non-chosen target ( indicated by the black dot ) . The initial rise in activity was not significantly different between choice and no-choice trials ( onset time on no-choice trial: 44 ms , choice trial: 40 ms; permutation test adjusted for multiple comparisons , p≤0 . 05 ) . However , there was a longer delay between the initial rise in activity and saccade onset ( onset time on no-choice trial: 141 ms before saccade onset , choice trial: 185 ms before saccade onset; permutation test adjusted for multiple comparisons , p≤0 . 05 ) , in keeping with the fact that reaction times were longer when the monkey had to choose between two response options ( Figure 2E ) . At the beginning , the activity associated with both possible targets was of similar strength , but 70 ms before saccade onset ( permutation test adjusted for multiple comparisons , p≤0 . 05 ) , a significant activity difference developed between the two sets of cells that predicted which saccade would be chosen . The activity centered on the chosen saccade became much stronger than the one centered on the non-chosen saccade . This differentiation reflected the decision outcome within the SEF relative action value map . The chosen saccade was often also the one with the larger value . We therefore performed a separate analysis of those choice trials were the monkeys made different choices for the same pair of gambles , which allowed us to differentiate between the representation of action value and choice . A comparison of the trials when the larger ( ‘correct’ ) or smaller ( ‘error’ ) value targets were chosen shows a strong increase of neural activity for the chosen target regardless of its average value ( Figure 6—figure supplement 1 ) . This confirmed that neural activity in SEF represented not only the relative action value of the competing saccades , but also the final choice . We hypothesized that the action selection process is driven by competition between the action values of the two targets . If that were true , we would expect the reduction in non-chosen saccade related activity to be less pronounced and to occur later because of weaker inhibition , when the differences in action values were less pronounced . We divided therefore the choice trials into two groups with small ( value difference smaller than 0 . 4 ) and large value differences ( value difference larger or equal to 0 . 4 ) , while controlling that the mean chosen value in both conditions was the same ( Figure 6C , D ) . As predicted , the neural activity associated with the non-chosen target was stronger and lasted longer , when the value differences were small ( onset of significant activity difference for chosen target and non-chosen target was 68 ms before saccade onset for small value difference trial and 42 ms before saccade onset for large value difference trials; permutation test adjusted for multiple comparisons , p≤0 . 05 , Table 2 ) . This longer lasting activity was consistent with the longer reaction time for smaller value difference trials ( Figure 2E ) . In contrast , the activity for the chosen target was weaker , when the value differences were small , especially early on ( 100–150 ms after target onset ) . The stronger activity associated with the non-chosen target in these trials was likely better able to withstand the competition of the activity associated with the chosen target and in turn reduced this activity more strongly . 10 . 7554/eLife . 09418 . 017Table 2 . The onset times in time-direction maps . The first main column shows the onset time calculated from trials aligned on target onset and the second main column shows the onset time calculated from trials aligned on saccade onset . Within each main column , the first minor column shows the time when the neural activity was significantly different from background activity ( -20 to 0 before target onset ) . The second minor column shows the time when the neural activity represented the choice . In no-choice trials , this corresponds to the time when the activity of neurons with a preferred direction within ± 30° of the target was significantly different from the activity of neurons where no target was presented ( the neurons with preferred direction within 240–300° ) . For choice trial , it corresponds to the time when the activity for the chosen target was significant different form the activity for the non-chosen target ( in both cases the neurons with preferred direction within ± 30° of their respective target ) . A permutation test with multiple comparison correction was used to calculate the onset times . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 017Time from target onsetTime from saccade onsetActivity vs backgroundChosen vs non-chosenActivity vs backgroundChosen vs non-chosenNo-choice44 ms-141 msChoice40 ms105 ms-185 ms-70 msChoice ( dV>=0 . 4 ) 44 ms92 ms-129 ms-68 msChoice ( dV<0 . 4 ) 41 ms139 ms-169 ms-42 ms On choice trials , there is a simultaneous onset of activity in the two areas of the relative action value map that correspond with the location of the two target options ( Figure 6B ) . Throughout the task , there is a robust representation of both options maintained in SEF , even after the divergence of activity that indicates the chosen option and action ( Figure 4C , D; Figure 6B , C ) . However , during the initial rise in activity the SEF population does not indicate the value of the target in its PD . At the time , when the neurons start to differentiate their activity according to the value of the target in the preferred direction they also reflect the value of the target in the NPD . This can be seen very clearly in Figure 4D showing the activity of SEF neurons for PD targets of medium value . Depending on the value of the NPD target , the activity starts to change ~110–120 ms after target onset . However , it took this much time for the SEF neurons to indicate value even during no-choice trials , when there was no competing target . It seems therefore that the SEF neurons always indicate relative action value . There is no time period in which two populations of SEF neurons represent the absolute action value of a target independent of the value of any competing target . Nevertheless , there is clear evidence of a succession of an initial undifferentiated state to a more and more differentiated state in which the influence of the chosen action value on the neuronal activity increases and the influence of the non-chosen one decreases ( Figure 4D; Figure 6B , C ) . This indicates a dynamic process as would be expected by a decision mechanism driven by competition via inhibitory interactions . Our results therefore support the idea that an ongoing process of inhibitory competition underlies the action selection . SEF neurons might directly participate in this action selection process , or at least reflect it . So far , all analysis have been performed using individual SEF neurons or comparisons of specific subsets of neurons . However , the decision process should also manifest itself in the dynamic changes in the instantaneous activity distribution across the entire SEF population . To study how the SEF population dynamically encodes the task variables underlying the monkeys’ behavior , we analyzed the average population responses as trajectories in the neural state space ( Yu et al . , 2009; Shenoy et al . , 2011 ) . As the previous analyses show , the activity pattern of individual SEF neurons during the decision process is not completely independent from each other , but follows particular pattern . The movement of SEF activity state trajectories in a lower-dimensional sub-space captured therefore most of the relationship between the SEF activity state and the behaviorally relevant task variables ( Figure 7 ) . We estimated this task-related subspace by using linear regression to define three orthogonal axis: chosen saccade direction along the horizontal and vertical dimension , and value of the chosen option ( Mante et al . , 2013 ) . 10 . 7554/eLife . 09418 . 018Figure 7 . Dynamics of SEF population activity trajectories in state space during decision making . The average population response for a given condition and time period ( 10 ms ) is represented as a point in state space . Responses are shown from 200 ms before to 10 after saccade onset . The time of saccade initiation is indicated by the larger dot . The four different chosen saccade directions are indicated by different colors ( up right: red; down right: orange: up left: black; down left: blue ) and the value of the chosen target by line style ( high value ( value>=0 . 7 ) : solid line , medium value ( value<0 . 7 and value >0 . 3 ) : dashed line , low value dotted line ) . ( A ) Trajectories of up-left and down-right movement in value and horizontal ( left/right ) subspace for three different values . ( B ) Trajectories of movements in value and action subspace . ( C ) The effect of the chosen option value on the state space trajectory at saccade onset . The subjective value of each chosen option was measured relative to the option with the smallest chosen value . The Euclidian distance in 3-D task space between the state vectors of each pair of chosen options increased as a function of their difference in subjective value . The significance of the relationship between difference in Euclidian distance and value was tested using a regression analysis ( t-test; the p-value indicates the probability that the regression slope is significantly different from zero ) . ( D ) The effect of the non-chosen option value on the state space trajectory at saccade onset . For trajectories with fixed saccade direction and chosen option value , the difference in Euclidian distance increased as a function of difference in subjective value of each non-chosen option relative to the option with the largest non-chosen value . SEF , supplementary eye field . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 01810 . 7554/eLife . 09418 . 019Figure 7—figure supplement 1 . Dynamics of SEF population activity trajectories in state space for choices of the down-left target . The trajectories show the population neuronal activity when down-left targets of varying value were chosen . The average population response for a given condition and time period ( 10 ms ) is represented as a point in state space . Responses are shown from 200 ms before to 10 after saccade onset . The time of saccade initiation is indicated by the larger dot . The trajectories are grouped according to both chosen and non-chosen values . The red , orange and blue colors indicate the large ( L ) , medium ( M ) , and small ( S ) chosen values in choice trials . Solid , dash , and dotted lines indicate the small , medium , and large non-chosen values , repectively . There are fewer trajectories when the chosen target was less valuable , because it was chosen less often . The trajectories for choice trials were influenced by both chosen and non-chosen values . Specifically , given a chosen value ( indicating by color ) , the trajectories were slightly lower along the value axis if the non-chosen value was larger ( indicating by line patterns ) . Thus , the neuronal population as a whole represents a relative action value signal . For comparison , purple , black , and green colors indicate the large , medium , and small chosen value in no-choice trials , repectively . If no-choice trials simply represent a situation that is similar to the choice trials , but in which the non-chosen value is simply zero , the trajectories for no-choice trial should be similar to the ones on choice trials with small non-chosen value . However , that is not the case . Instead , the trajectories for no-choice trials always reach an end point that is less extended along both the value and direction axis than the respective trajectories for choice trials . Thus , the neural population responds differently on choice and no-choice trials . SEF , supplementary eye field . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 01910 . 7554/eLife . 09418 . 020Figure 7—figure supplement 2 . Fraction of variance in state vector position explained by task-related axes of chosen direction ( pink and purple ) and chosen value ( green ) . The lines underneath show when the fraction of variance explained is different from random ( permutation test , p<=0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09418 . 020 First , we compare only trajectories for saccades in two different directions and three different value levels in a simplified two-dimensional space spanned by the value and the vertical movement direction axis ( Figure 7A ) . The trajectories of upward and downward saccades ( as indicated by the different color ) are clearly well separated along the direction axis . In addition , the trajectories for different chosen value are also separate from each other along the value axis ( as indicated by different line style ) . Thus , the trajectories move in an orderly fashion with respect to the two task-related axis . As a result of the separation across both axis , the trajectories reach six different points in state space , when the respective saccade is initiated . Similarly , in the full three-dimensional ( 3-D ) task space , the trajectories for all four directions and the different chosen values are also well separated from each other and do not converge ( Figure 7B ) . As a result , the trajectories reach different positions in 3-D task space at the moment of saccade initiation , and their distance is significantly correlated with the difference in chosen value for all four different saccade directions ( Figure 7C ) . Our previous analysis suggests that the neurons encoding the relative action value of saccades in different direction compete through mutual inhibition . This mutual inhibition should change the direction and endpoint of the different trajectories , so that they should not only depend on the saccade direction and subjective value of the chosen target , but also on the value of the non-chosen target . To test this , for a fixed direction and chosen value , we computed the distance between the trajectory with the largest non-chosen value and all other trajectories with decreasing non-chosen values . The regression analysis shows that , when saccade direction and chosen value is fixed , the distance between trajectories is significantly modulated by non-chosen value ( Figure 7D ) . The larger the non-chosen value difference is the further apart the trajectories are when the saccade is initiated . This can also be observed by comparing population activity trajectories in the state space ( Figure 7—figure supplement 1 ) . Thus , the trajectories in state space reflect both the chosen and the non-chosen target values . In this context , no-choice trials could be considered as choice trials with zero non-chosen value . In this case , one would expect the trajectory to be similar to the trials with large chosen and small unchosen value . However , this is not the case ( Figure 7—figure supplement 1 ) . Although trajectories for no-choice trials are also modulated by both value and direction , they always reach a point along both the value and direction axis that is less extended than during choice trials . Lastly , we asked whether the sequential value and action selection can be observed in the state space analysis . An indication of this can be seen in Figure 7A . The trajectories first start to separate along the value axis , before they separate along the direction axis . Consistent with this observation and the single neuron analysis , the variance explained by value axis increased earlier than the variance explained by saccade direction axis ( Figure 7—figure supplement 2 ) . Our results indicate that SEF represents the relative action value of all possible saccades , forming a relative action value map . During value-based decisions , the SEF population first encodes the chosen gamble option and only later the chosen direction . Our data suggest that neural activity in SEF reflects the action selection , the second step in the decision process . This selection process occurs likely through a process of competitive inhibition between groups of neurons carrying relative action value signals for different saccades . This inhibition could occur locally between SEF neurons , either through mutual inhibition between different SEF neurons within the relative action value map , or through a global pool of inhibitory neurons that receive input from excitatory neurons in SEF ( Schlag et al . , 1998; Wang , 2008; Nassi et al . , 2015 ) . Alternatively , the neural activity in SEF and the competitive process manifested by it could reflect shared signals within the more distributed action selection network that SEF is part of . Similar action value signals have been reported in lateral prefrontal cortex ( Matsumoto et al . , 2003; Wallis and Miller , 2003 ) , anterior cingulate cortex ( Matsumoto et al . , 2003 ) , and basal ganglia ( Samejima et al . , 2005; Lau and Glimcher , 2008 ) . Of course , it is also possible that the action selection involves both interactions between neurons in a larger network and local inhibitory interactions . Future perturbation experiments will be required to test whether SEF plays a causal role in decision making and also if the relative action value encoding is at least partly the result of local inhibitory mechanism or whether it reflects only input from connected brain regions . Independent of these considerations , our results allow us to draw some conclusions about the basic functional architecture of decision making in the brain . Specifically , they invalidate a number of previously suggested decision models and instead support a new sequential model of decision making . Currently , three major hypotheses about the mechanism underlying value-based decision making have been suggested: the goods-based model ( Figure 1A ) , the action-based model ( Figure 1B ) , and the distributed consensus model ( Figure 1C ) . Our gamble experiment design allows us to test these models by dissociating the value selection process from the action selection process . Due to the uncertainty of reward for each individual gamble , the large number of the gamble option pairs , and the fact that each gamble option pairing could be presented in multiple spatial configurations , the task design prevent the subjects from making direct associations between the visual representation of the gamble options and action choice . Therefore , the task required on each trial a good-based , as well as an action-based selection . Our findings support none of the previously suggested models . First , the pure good-based model of decision making would predict that action-related representations are downstream of the decision stage and should therefore only represent the decision outcome ( i . e . the chosen action; see Figure 1A ) . However , we found evidence for competition between relative action value-encoding neurons in SEF that is spatially organized , that is , in an action-based frame of reference ( Figure 4 ) . This fact , together with the fact that we find activity corresponding to both response options ( Figure 6 ) clearly rules out the pure good-based model ( Figure 1A ) . Second , the pure action-selection model B would predict that the competition would only happen in the action value space . In Figure 1B , this is indicated by the absence of inhibitory connections between the nodes representing the reward options ( or goods ) . This model therefore predicts that information about chosen saccade direction should appear simultaneously with or even slightly earlier than information about chosen reward option , since the selected action value signal contains both direction and value information . However , this prediction is contradicted by our observation that the chosen value information is present earlier than the chosen action information ( Figure 3 , Figure 5 and Figure 7 ) . Thus , there is a moment during the decision-making process ( 100–50 ms before saccade onset ) , when the SEF neurons encode which option is chosen , but not yet , what saccade will be chosen . This could clearly never happen in an action-selection model of decision-making . Lastly , the distributed consensus model ( Figure 1C ) suggests strong recurrent connections from the action selection level back to the option selection level . The reciprocal interaction should lead to a synchronization of the selection process in both stages and the chosen gamble option and chosen action should be selected simultaneously . This prediction is clearly not supported by our findings , given the robust 100 ms time difference in the onset of chosen option and direction information . Instead , our data are most consistent with a model that predicts selection both on the option and the action representation level , but with asymmetric connections between them , so that the option selection level influences the action selection level , but not vice versa . This is the sequential model of decision making ( Figure 1D ) . According to this model , value-based decisions require two different selection processes within two different representational spaces . First , a preferred option has to be chosen within an offer or goods space by comparing their value representations ( Padoa-Schioppa , 2011 ) . Second , within an action space the response has to be chosen that most likely will bring about the preferred option . A similar sequence has been found , when initially only information relevant for the selection of an economic good is provided , and information relevant for the selection of an action is only given after a delay ( Cai and Padoa-Schioppa , 2014 ) . Here , we show that this sequence is obligatory , since it even occurs , when information guiding the selection of goods and actions is given simultaneously . This suggests that these two selection steps are related , but functionally independent from each other and involve different brain circuits . This explains why evidence for decision-related neural activity has been found at both selection stages ( Shadlen et al . , 2008; Cisek and Kalaska , 2010; Wunderlich et al . , 2010; Padoa-Schioppa , 2011; Cisek , 2012 ) . A similar separation between stimulus categorization and action selection has been also found in other decision processes ( Schall , 2013 ) . The competition between subjective value representations takes most likely place in OFC ( Padoa-Schioppa and Assad , 2006 ) and vmPFC ( Wunderlich et al . , 2010; Lim et al . , 2011; Strait et al . , 2014 ) , while the competition between action value representations takes place in SEF and DLPFC ( Wallis and Miller , 2003; Kim et al . , 2008 ) . The selection of action value signals in turn can influence the neural activity in primarily motor-related areas , such as FEF and SC that encode the final commitment to a particular course of action ( Schall et al . , 2002; Brown et al . , 2008; Thura and Cisek , 2014 ) . It has been suggested that different neural and functional architectures underlie different types of value-based decisions making ( Hunt et al . , 2013 ) . In contrast , our sequential decision model predicts that decisions are always made using the same decision architecture: an initial goods-based selection , followed by action-based selection , because both stages are necessary and not functionally interchangeable . Nevertheless , the relative importance of each selection stages likely depends on the behavioral context . To understand strictly economic behavior ( such as savings behavior or consumption of goods ) the goods selection is the more important step . Preferred options can be selected without knowledge about the action necessary to indicate the chosen option ( Gold and Shadlen , 2003; Bennur and Gold , 2011; Grabenhorst et al . , 2012; Cai and Padoa-Schioppa , 2014 ) . In such situations , there is no evidence of ongoing competition between potential actions ( Cai and Padoa-Schioppa , 2012 , 2014 ) . Obtaining a desired good is typically considered to be a trivial act in a well-functioning market ( Padoa-Schioppa , 2011 ) . On the other hand during perceptual or rule-based decision making , the action selection is the most important step in the decision process , because only one type of good can be achieved by engaging in the task . An example are competitive games such as chess , were the goal ( checkmate ) is clear and implicitly chosen when a player starts a game , but were the player still has to find the most appropriate actions to achieve this goal . This implies that within different behavioral contexts , different elements of the decision circuit become critical . Altogether , we think that actual behavior under a wide range of different conditions is best understood by a model that respects that behavioral choices are the result of two independent and functionally different selection mechanisms . In the gambling task , the monkeys had to make saccades to peripheral targets that were associated with different amounts of reward ( Figure 2A ) . The targets were squares of various colors , 2 . 25×2 . 25° in size . They were always presented 10° away from the central fixation point at a 45 , 135 , 225 , or 315° angle . There were seven different gamble targets ( Figure 2B ) , each consisting of two colors corresponding to the two possible reward amounts . The portion the target filled with each color corresponded to the probability of receiving the corresponding reward amount . Four different colors indicated four different reward amounts ( increasing from 1 , 3 , 5 to 9 units of water , where 1 unit equaled 30 µl ) . The minimum reward amount for the gamble option was always 1 unit of water ( indicated by cyan ) , while the maximum reward amount ranged from 3 ( red ) , 5 ( blue ) to 9 units ( green ) , with three different probabilities of receiving the maximum ( 20 , 40 , and 80% ) . This resulted in a set of gambles , whose expected value on the diagonal axis was identical , as shown in the matrix ( Figure 2B ) . The task consisted of two types of trials - choice trials and no-choice trials . All the trials started with the appearance of a fixation point at the center of the screen ( Figure 2B ) , on which the monkeys were required to fixate for 500–1000 ms . In choice trials , two targets appeared on two randomly chosen locations across the four quadrants . Simultaneously , the fixation point disappeared and within 1000 ms the monkeys had to choose between the gambles by making a saccade toward one of the targets . Following the choice , the non-chosen target disappeared from the screen . The monkeys were required to keep fixating on the chosen target for 500–600 ms , after which the target changed color . The two-colored square then changed into a single-colored square associated with the final reward amount . This indicated the result of the gamble to the monkeys . The monkeys were required to continue to fixate on the target for another 300 ms until the reward was delivered . Each gamble option was paired with all other six gamble options . This resulted in 21 different combinations of options that were offered in choice trials . The sequence of events in no-choice trials was the same as in choice trials except that only one target was presented . In those trials , the monkeys were forced to make a saccade to the given target . All seven gamble options were presented during no-choice trials . We presented no-choice and choice trials mixed together in blocks of 28 trials that consisted of 21 choice trials and 7 no-choice trials . Within a block , the order of trials was randomized . The locations of the targets in each trial were also randomized , which prevented the monkeys from preparing a movement toward a certain direction before the target appearance . For reward delivery , we used an in-house built fluid delivery system . The system was based on two syringe pumps connected to a fluid container . A piston in the middle of the two syringes was connected with the plunger of each syringe . The movement of the piston in one direction pressed the plunger of one syringe inward and ejected fluid . At the same time , it pulled the plunger of the other syringe outward and sucked fluid into the syringe from the fluid container . The position of the piston was controlled by a stepper motor . In this way , the size of the piston movement controlled the amount of fluid that was ejected out of one of the syringes . The accuracy of the fluid amount delivery was high across the entire range of fluid amounts used in the experiment , because we used relatively small syringes ( 10 ml ) . Importantly , it was also constant across the duration of the experiment , unlike conventional gravity-based solenoid systems . We used Maximum Likelihood Difference Scaling ( MLDS ) ( Maloney and Yang , 2003; Kingdom and Prins , 2010 ) to estimate the subjective value of different targets . The algorithm is an optimization algorithm which gives the best estimation of the subjective value and internal noise based on the maximum across-trial likelihood , which is defined as: ( 1 ) L ( ψ ( 1 ) , ψ ( 2 ) , …ψ ( N ) , σd\|r ) =∑k=1Tlogep ( rk\|Dk;ψ ( 1 ) , ψ ( 2 ) , …ψ ( N ) , σd ) where ψ ( i ) are the subjective value for all the targets , σd is the internal noise , rk is the response ( chosen:1 or non-chosen:0 ) on the kth trial , and Dk is the estimated subjective value difference between two targets in the kth trial given the set of subjective value and internal noise , r the full set of responses across all trials and T the number of trials . We performed the MLDS using the Matlab based toolbox 'Palamedes' developed by Prins and Kindom ( Kingdom and Prins , 2010 ) . After training , we placed a hexagonal chamber ( 29 mm in diameter ) centered over the midline , 28 mm ( monkey A ) and 27 mm ( monkey I ) anterior of the interaural line . During each recording session , single units were recorded using 1–4 tungsten microelectrodes with an impedance of 2–4 MΩs ( Frederick Haer , Bowdoinham , ME ) . The microelectrodes were advanced , using a self-built microdrive system . Data were collected using the PLEXON system ( Plexon , Inc . , Dallas , TX ) . Up to four template spikes were identified using principal component analysis . The time stamps and local field potential were then collected at a sampling rate of 1 , 000 Hz . Data were subsequently analyzed off-line to ensure only single units were included in consequent analyses . We used custom software written in Matlab ( Mathworks , Natick , MA ) , which are available at the following GitHub respository: https://github . com/XMoChen/Sequential-good-and-action-selection-during-decision-making . To determine the location of the SEF , we obtained magnetic resonance images ( MRI ) for monkey A and monkey I . A 3-D model of the brain was constructed using MIPAV ( BIRSS , NIH ) and custom Matlab codes . As an anatomical landmark , we used the location of the branch of the arcuate sulcus . The locations of neural recording sites are shown in ( Figure 2—figure supplement 1 ) . In both monkeys , we found neurons during the saccade preparation period in the region from 0 to 11 mm anterior to the genu of the arcuate branch and within 5 mm to 2 mm of the longitudinal fissure . We designated these neurons as belonging to the SEF , consistant with previous studies from our lab and existing literature ( Tehovnik et al . , 2000; So and Stuphorn , 2010 ) . We used several criteria to determine whether a neuron was task related . To test whether a neuron was active while the monkey generated saccades to the targets , we analyzed the neural activity in the time period between target onset to saccade initiation . We performed a permutation test on the spike rate in 50 ms intervals throughout the saccade preparation time period ( 150–0 ms before saccade onset or 50–200 ms after target onset ) to compare against the baseline period ( 200–150ms prior to target onset ) . If p value was ≤0 . 05 for any of the intervals , the cell was determined to have activity significantly different form baseline . Out of 516 neurons , 353 were classified as task-related using these criteria . Furthermore , we used a more stringent way to define the task related neuron by fitting a family of regression models to the neural activity and determining the best-fitting model ( So and Stuphorn , 2010 ) . The influence of value ( V ) on neuronal activity was described using a sigmoid function ( 2 ) f ( V ) =b11+e−s ( V−t ) where b1 is the weight coefficient , s ( s∈ ( 0 , 1 ) ) is the steepness , and t ( t∈ ( 0 , 1 ) ) is the threshold value . Often , the influence of expected value on neuronal activity is described using a linear function . However , SEF neurons are better described using a sigmoid function . The reasons for this are twofold: 1 ) Many SEF neurons actually had a ‘curved’ relationship with increasing value ( So and Stuphorn , 2010 ) . 2 ) The more important reason is that a substantial number of value-related SEF neurons showed floor or ceiling effects , that is , they showed no modulation for value increases in a certain range , but started to indicate value above or below a certain threshold . In addition , the sigmoid function is flexible enough to easily approximate a linear value coding . In Equation 2 , by setting t=0 . 5 , b1>1 , the relatively linear part of the sigmoid function can be used for value coding . Thus , the sigmoid function is flexible enough to fit the behavior of a large number of neurons with monotonically increasing or decreasing activity for varying value ( including linearly related ones ) . The influence of saccade direction ( D ) on neuronal activity was described using a circular Gaussian function ( 3 ) g ( D ) =b2×e{w×[cos ( D−p ) ]−1} where b2 is the weight coefficient , w ( w∈ ( 0 , 4π] ) is the turning width , p ( p∈[0 , 2π] ) is the PD of the neuron . The interaction of value and direction was described using the product of f ( V ) and g ( D ) ( 4 ) h ( V , D ) = f ( V ) × g ( D ) =b3 × 11+e-s ( V-t ) × e{w×[cosD-p]-1} where b3 is the weight coefficient . For each neuron , we fitted the average neuronal activity before saccade ( 50ms before saccade onset to 20 ms after saccade onset ) on each no-choice trial with all possible linear combinations of the three terms f ( V ) , g ( D ) , h ( V , D ) as well as with a simple constant model ( b0 ) . We identified the best fitting model for each neuron by finding the model with the minimum Bayesian information criterion ( Burnham and Anderson , 2002; Busemeyer and Diederich , 2010 ) ( 5 ) BIC=n×log ( RSSn ) +k×log ( n ) where n is the number of trials ( a constant in our case ) , and RSS is the residual sum of squares after fitting . We used a loosely defined BIC in order to include more neurons into analysis , where k is the number of independent variables in the equation rather than the number of parameters in the regression model . A lower numerical BIC value indicates that the model fit the data better: with a lower residual sum of squares indicating better predictive power and a larger k penalizes less parsimonious models . All neurons with lower BIC value than the baseline model containing only a constant ( b0 ) were considered task related . Among the 353 task-related neurons , 128 neurons were further classified as directionally tuned and were used in the following analyses . All neurons were tested with all 21 gamble option combinations and at least four diagonal directional combinations in which two targets where 180 degree away . One hundred and six neurons ( 26 from monkey A and 80 from monkey I ) were tested with no less than 8 out of 12 direction combinations ( 4 diagonal and 4 ninety degree away in the same hemi-visual field direction combinations ) , and 86 neurons ( 6 from monkey A and 80 from monkey I ) were tested with all 12 direction-combinations . Averages of neural activity across the entire population of all 128 directionally-tuned SEF neurons were performed after the individual neurons activity was normalized by searching for the minimum and maximum activity across all choice and no-choice trial conditions and setting the minimum activity to 0 and the maximum activity to 1 . The only exception is the construction of the relative action value map , were we used a slightly different definition of the zero reference point . The normalized time-direction maps show the population activity of all directional SEF neurons based on their preferred direction relative to the chosen and non-chosen target ( Cisek and Kalaska , 2005 ) . For each neuron , we generated the mean firing rate separately for all 16 combinations of choices and target configurations ( choice trials: 12; no-choice trials: 4 ) . The neuron’s firing rate was normalized by setting the baseline activity ( mean activity between 50 to 0 before target onset across 16 conditions ) to 0 and the maximum activity across all 16 conditions to 1 . Each cell's preferred direction was defined by the circular Gaussian term in the best fitting model in the BIC analysis . Population data were displayed as a 2D color plot , in which the spike density functions of each neuron were sorted along the vertical axis according to their preferred direction with respect to the location of the selected target . This resulted in a matrix in which the PD distribution within the relative action value map was unevenly sampled . The sorted matrix was therefore smoothed by linear interpolation at an angle of 7 . 2° . The horizontal axis showed the development of the population activity across time aligned to either target or movement onset . For all population maps , the same baseline activity and maximum activity were used for each neuron . Binary linear classification was performed using Matlab toolboxes and custom code . The analysis was performed on neural activity 200 ms before till 20 ms after movement onset at a 1 ms time resolution . For each neuron , we used the neural activity in those choice trials in which the monkey chose a particular value or direction to train the classifier ( around half of the trials for direction , and different numbers of trials for different values depending on the monkeys' choice behavior ) . We then used the classifier to predict either the direction or the value of the chosen target for each trial . When predicting the chosen direction , for example , there are two target locations in a choice trial . We used the neural activity in all choice trials when the monkey chose either one of the target locations to train the classifier , and then used the optimized classifier to predict the chosen saccade direction based on the observed neural activity in a particular trial . The overall classification accuracy was calculated by averaging across all trials for each neuron . We used a permutation test , in which we shuffled the chosen and non-chosen target value and direction , to test whether the classification accuracy was significant ( 1000 shuffle; p≤0 . 05 ) . In order to compare the relative strength of the relationship between neural activity and saccade value and direction , we calculated separately for each neuron the mutual information between neural activity and chosen and non-chosen value or direction , respectively . To capture the dynamics of value and direction encoding , we performed the calculation repeatedly for consecutive time periods during saccade preparation using spike density at a 1 ms time resolution . To reduce the bias in estimating the mutual information and let the estimated information comparable between trial conditions , we discretized the neural activities in the same way . During no-choice trials , we sampled the space of possible values and directions evenly , in contrast to choice trials were the values and direction depended on the monkey’s preferences . We assumed that the neural activity in no-choice trials allowed us to capture how this neuron encoded value and direction information and we used neural activity in no-choice trial to determine the bins for neural activity for all trial conditions . We set the number of bins for neural activity ( NF ) as four . At each particular time window , we collected the mean neural firing rates ( F ) from every no-choice trial and divided them into four bins so that each bin held equal number of no-choice trials . We then got Q1 , Q2 and Q3 as the boundaries for 4 quartiles . For all trial conditions , at the same time window , neural activity below Q1 was classified as F1 , between Q1 and Q2 as F2 , between Q2 and Q3 as F3 , and finally , neural activity above Q3 was classified as F4 . The mutual information between neural activity F and the variable X , which can be either chosen or non-chosen value or chosen or no-chosen direction in our case , was approximated by the following: ( 6 ) I ( F , X ) = ( ∑i=1NF∑j=1NXMijMlogMijMMi·M·j ) −Bias here Mij is the number of trials having both Fi and Xj; is the number of trials having Fi , and M·j is the number of trials having Xj . M is the number of total trials . As mentioned before , we set NF , the number of distinct states of neural activity , to four . In the case of direction , we set Nx , the number of distinct states of the relevant variable , to four , because we tested four different saccade directions . In the case of value , we tested seven different values . However , distinguishing seven different value levels would have resulted in different maximum amounts of mutual information for the two variables ( direction: 2 . 00 bits; value: 2 . 81 bits ) . This would have led to an overestimation of value information relative to directional information . In order to make the value and direction information estimations directly comparable , we set Nx for value to four as well . In grouping the seven different values into four bins , we followed the same binning procedure as we did for the neural activities . The chosen values were divided into four quartiles so that each bin held an equal number of no-choice trials . We computed a first approximation of the bias as follows: ( 7 ) Bias=12Mlog2 ( UFX−UF−UX+1 ) where UFX is the number of nonzero Mij’s for all i and j , UF is the number of nonzero M . i for all i , and UX is the number of nonzero Mi . for all j . This procedure followed the approach described in ( Ito and Doya , 2009 ) . Finally , we performed a Bootstrap procedure to test whether the amount of mutual information was significant , and to further reduce any remaining bias . We generated a random set of Fi and Xj pairs , by permuting both F and X arrays , respectively . We calculated the mutual information between F and X , using the same method described above , and repeated this process for 100 times . The mean of the mutual information obtained from these 100 random processes represented remaining bias and was subtracted from I ( F , X ) . To test whether the final estimated mutual information was significant ( p≤0 . 05 ) , we compared it with the sixth highest information obtained from the 100 random processes . If it was non-significant , we set the mutual information to zero . The bias reductions sometimes lead to negative estimates of mutual information . In that case , we also set the final estimated information to be zero . To determine when the SEF population carried different amounts of information about the chosen and non-chosen direction or value , we compared the information about the chosen and non-chosen option across all neurons in each time bin using a paired t-test . We defined the onset of differences in information as the first time bin in which p-values were less than 0 . 05 for 10 or more consecutive time bins . A linear regression was used to determine the temporally evolving contribution of the chosen and non-chosen target to the neural firing rate in choice trials . First , for each neuron , we calculated the mean firing rate on no-choice trials for each direction ( Sno-choice ( D , t ) or value ( Sno-choice ( V , t ) ) for sequential time steps in the trial , using a sliding time window with 20 ms width and 10 ms step size . Then , in the regression analysis , the contribution of the chosen and non-chosen directions was described as: ( 8 ) Schoice ( t ) =b1Sno−choice¯ ( Dchosen , t ) +b2Sno−chocie¯ ( Dnonchosen , t ) The contribution of the chosen and non-chosen values was described as: ( 9 ) Schoice ( t ) =b1Sno−choice¯ ( Vchosen , t ) +b2Sno−choice¯ ( Vnonchosen , t ) The data were fitted with a linear least-square fitting routine implemented in Matlab ( The Math Works , Natick , MA ) . To determine when the SEF population showed a significant ( p≤0 . 05 ) difference in the influence of the chosen and non-chosen regression coefficients for direction and value , we performed paired t-tests for each time bin . We defined the onset of differences in the strength of coefficients as the first time bin in which p-values were less than 0 . 05 for 3 or more consecutive time bins . Population activity can be represented within the state space framework ( Yu et al . , 2009; Shenoy et al . , 2011 ) . In this framework , the state of activity of all n recorded neurons ( i . e . the activity distribution ) is represented by a vector in an n-dimensional state space . The successive vectors during a trial form a trajectory in state space that describes the development of the neural activity . Our state-space analysis follows generally the one described in Mante et al . ( 2013 ) . The main difference is that we did not perform a principal component analysis to reduce the dimensionality of the state space . To construct population responses , we first computed the average activity of all recorded neurons in both monkeys for each trial condition . Then , we combined the 128 average activity values to a 128-dimensional vector array representing the population activity trajectory in state space for each trial condition . Next , we used linear regression to identify dimensions in state space containing task related variance . For the z-scored responses of neuron i at time t , we have: ( 10 ) ri , t ( k ) =βi , t ( 1 ) chosen_directionleft_right ( k ) +βi , t ( 2 ) chosen_directionup_down ( k ) +βi , t ( 3 ) chosen_value ( k ) +βi , t ( 4 ) +βi , t ( 4 ) nonchosen_directionleft_right ( k ) +βi , t ( 5 ) nonchosen_directionup_down ( k ) +βi , t ( 6 ) nonchosen_value ( k ) +βi , t ( 7 ) where ri , t ( κ ) =βi , t is the z-scored response of neuron i at time t on trial κ , chosen_directionleft_right ( k ) and nonchosen_directionleft_right ( k ) is the monkeys chosen and nonchosen direction on trial k ( +1: right; -1: left ) , chosen_directionup_down ( k ) and nonchosen_directionup_down ( k ) is the monkeys chosen and non-chosen direction on trial k ( +1: right; -1: left ) . There are six independent variables ( var ) that can influence the responses of neuron i in function ( 10 ) . To estimate the respective regression coefficients βi , t ( var ) , for var=1 to 6 , we define , for each unit i , a matrix Fiof size Ncoef×Ntrial , where Ncoef is the number of regression coeffients to be estimated and Ntrial is the number of trial recorded for neuron i . The regression coefficients can be then estimated as: ( 11 ) βi , t= ( FiFiT ) −1Firi , t where βi , t is a vector of length Ncoef with elements βi , t ( var ) , v=1–6 . It corresponds to the regression coefficient for task variable var , time t , and neuron i . For each task variable , we build a set of coefficient vectors βv , t whose entries is βi , t ( var ) . The new vector βvar , t correspond to the directions in state space along which the task variable are represented at the level of the population . For each task variable var , we then determined the time , tvarmax , for which the corresponding set of vectors βvar , t . βvarmax=βvar , tvarmax with tvarmax=argmaxt\|\|βvar , t\|\| . Last , we orthogonalized the axes of direction and value with QR-decomposition . The new axis βvar⊥ span the same ‘regression subspace’ as the original regression vectors; however , it each explains distinct portions of the variance in the responses . Then at a specific time t , the projections of the population response on the time-independent axes are defined by: ( 12 ) pv , arc=βvar⊥TXc where pvar , c is the set of time-series vectors over all task variable and conditions . Xc is the firing rate matrix in different trial conditions with the size of Nunit×T .	Much of our decision making seems to involve selecting the best option from among those currently available , and then working out how to attain that particular outcome . However , while this might sound straightforward in principle , exactly how this process is organized within the brain is not entirely clear . One possibility is that the brain compares all the possible outcomes of a decision with each other before constructing a plan of action to achieve the most desirable of these . This is known as the 'goods-based' model of decision making . However , an alternative possibility is that the brain instead considers all the possible actions that could be performed at any given time . One specific action is then chosen based on a range of factors , including the potential outcomes that might result from each . This is an 'action-based' model of decision making . Chen and Stuphorn have now distinguished between these possibilities by training two monkeys to perform a gambling task . The animals learned to make eye movements to one of two targets on a screen to earn a reward . The identity of the targets varied between trials , with some associated with larger rewards or a higher likelihood of receiving a reward than others . The location of the targets also changed in different trials , which meant that the choice of 'action' ( moving the eyes to the left or right ) could be distinguished from the choice of 'goods' ( the reward ) . By using electrodes to record from a region of the brain called the supplementary eye field , which helps to control eye movements , Chen and Stuphorn showed that the activity of neurons in this region predicted the monkeys’ decision-making behavior . Crucially , it did so in two stages: neurons first encoded the reward chosen by the monkey , before subsequently encoding the action that the monkey selected to obtain that outcome . These data argue against an action-based model of decision making because outcomes are encoded before actions . However , they also argue against a purely goods-based model . This is because all possible actions are encoded by the brain ( including those that are subsequently rejected ) , with the highest levels of activity seen for the action that is ultimately selected . The data instead support a new model of decision making , in which outcomes and actions are selected sequentially via two independent brain circuits .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2015	Sequential selection of economic good and action in medial frontal cortex of macaques during value-based decisions
The heat shock response is a universal homeostatic cell autonomous reaction of organisms to cope with adverse environmental conditions . In mammalian cells , this response is mediated by the heat shock transcription factor Hsf1 , which is monomeric in unstressed cells and upon activation trimerizes , and binds to promoters of heat shock genes . To understand the basic principle of Hsf1 activation we analyzed temperature-induced alterations in the conformational dynamics of Hsf1 by hydrogen exchange mass spectrometry . We found a temperature-dependent unfolding of Hsf1 in the regulatory region happening concomitant to tighter packing in the trimerization region . The transition to the active DNA binding-competent state occurred highly cooperative and was concentration dependent . Surprisingly , Hsp90 , known to inhibit Hsf1 activation , lowered the midpoint temperature of trimerization and reduced cooperativity of the process thus widening the response window . Based on our data we propose a kinetic model of Hsf1 trimerization . To cope with changes in physical and chemical properties of the environment as well as with physiological and pathophysiological conditions which cause protein misfolding , organisms mount a homeostatic transcriptional program , the so-called heat shock response ( Jolly and Morimoto , 2000 ) . In all eukaryotic cells , heat shock transcription factor ( HSF ) 1 is the master regulator of this response and alters transcription of a large number of genes , some of which encode chaperones and proteases ( Anckar and Sistonen , 2011 ) . Although this response is essentially cell autonomous , systemic modulation of this response has been observed in metazoa ( Morimoto , 2008; Prahlad et al . , 2008; Prahlad and Morimoto , 2011 ) . Metazoan Hsf1 consists of a N-terminal winged helix-turn-helix DNA binding domain ( Harrison et al . , 1994; Vuister et al . , 1994 ) , a hydrophobic shorter heptad repeat regions ( HR-A/B ) proposed to function as a leucine zipper coiled-coil trimerization domain ( Clos et al . , 1990; Rabindran et al . , 1993 ) , a regulatory domain , a second heptad repeat ( HR-C ) and a C-terminal transcription activation domain ( Figure 1A ) ( Anckar and Sistonen , 2011; Voellmy , 2004 ) . In unstressed cells metazoan Hsf1 is monomeric and supposed to be in complex with molecular chaperones , including Hsp70 , Hsp90 and TRiC/CCT ( Shi et al . , 1998; Zou et al . , 1998; Neef et al . , 2014 ) . At physiological concentrations monomeric Hsf1 does not bind appreciably to heat shock elements ( nGAAn ) . In the activated state Hsf1 forms trimers or higher order oligomers and binds to its response elements in heat shock gene promoters ( Clos et al . , 1990; Rabindran et al . , 1993 ) . Currently , two models are discussed for the heat-induced activation of Hsf1: ( 1 ) Based on the observation that deletion or mutational alteration of HR-C leads to constitutive trimerization Wu and co-workers proposed that Hsf1 is a thermosensor itself and kept monomeric by intramolecular leucine zipper formation ( Rabindran et al . , 1993 ) . However , activation of human Hsf1 in human and insect cells and Xenopus oocytes occurs at different temperatures , arguing against an Hsf1 intrinsic mechanism of heat shock activation ( Baler et al . , 1993; Clos et al . , 1993 ) . ( 2 ) Owing to the fact that the large variety of Hsf1-inducing signals have in common to cause protein misfolding and in analogy to the regulation of the heat shock response in E . coli ( Guisbert et al . , 2008 ) , chaperones were proposed to prevent Hsf1 activation and to be titrated away from Hsf1 under stress conditions , resulting in heat shock response induction ( Morimoto , 1998 ) . Consistent with this hypothesis is the observation that inhibition of Hsp70 , Hsp90 or TRiC/CCT or knock-down of their expression leads to the induction of the heat shock response ( Powers and Workman , 2007; Powers et al . , 2008; Neef et al . , 2014; Whitesell et al . , 2003; Lee et al . , 2013; Abravaya et al . , 1992; Zou et al . , 1998 ) . 10 . 7554/eLife . 11576 . 003Figure 1 . Recombinant purified human Hsf1 is largely monomeric and trimerizes and acquires DNA binding competence upon heat shock . ( A ) Domain organization of human Hsf1 [modified from Anckar and Sistonen ( 2011 ) ] . ( B ) Size exclusion chromatography separates recombinant human Hsf1 in monomer , dimer and trimer/oligomer as indicated . ( C ) Blue native gel of the three peak indicated in panel B ( monomer and dimer ) , monomeric Hsf1 after 10 min heat shock at 42°C ( monomer HS ) ; and trimeric/oligomeric Hsf1 purified under denaturing conditions and refolded into a DNA binding competent state ( M , monomer; D , dimer; T , trimer; HO , higher order oligomers ) . ( D ) Electrophoretic mobility shift assay ( EMSA ) . Monomeric Hsf1 ( Hsf1m ) , monomeric Hsf1 treated for 10 min at 42°C ( HS ) , or trimeric Hsf1 ( Hsf1t ) were incubated with fluorescent labeled HSE-DNA minus or plus unlabeled HSE-DNA and separated on a native agarose gel . Lane 1 , HSE-DNA in the absence of protein . ( E ) Amide hydrogen exchange of monomeric Hsf1 after 30 s at 20°C in D2O buffer . Exchange was correct for back exchange using a fully deuterated Hsf1 preparation . Error bars are the standard error of mean ( SEM ) of three independent experiments . ( F ) Cartoon representation of the DNA binding ( PDB ID 2LDU ) and trimerization domains of human Hsf1 colored according to deuteron incorporation as indicated . Gray , no sequence coverage . The trimerization domain is a homology model of the HR-A/B region ( residues 130–203 ) of human HSF1 on the structure of Chaetomium thermophilum Skn7 [PDB ID 5D5Z , ( Neudegger et al . , 2016 ) using I-TASSER ( Zhang , 2008; Yang and Zhang , 2015; Yang et al . , 2015; Roy et al . , 2010 ) ] . DOI: http://dx . doi . org/10 . 7554/eLife . 11576 . 003 Further regulation of Hsf1 is provided by posttranslational modifications , including phosphorylation , acetylation , sumoylation and oxidation of cysteines to disulfide bridges ( Hietakangas et al . , 2003; 2006; Sarge et al . , 1993; Westerheide et al . , 2009; Brunet Simioni et al . , 2009; Zhong et al . , 1998; Lu et al . , 2008 ) . The contribution of these modifications to the primary activating mechanism are still unclear ( Budzyński et al . , 2015 ) . To resolve the molecular mechanism of the temperature-induced activation of Hsf1 we analyzed the conformational dynamics of purified monomeric human Hsf1 pretreated at different temperatures using hydrogen-1H/2H-exchange ( HX ) mass spectrometry ( MS ) . We found temperature-dependent unfolding of HR-C and concomitant stabilization of HR-A/B , demonstrating that isolated Hsf1 acts as temperature sensor . At short incubation times the temperature response curve exhibits high cooperativity with a transition midpoint of 36°C . Using fluorescence anisotropy we demonstrate that the acquisition of DNA-binding competency depends on temperature and concentration of Hsf1 . Phosphomimetic Hsf1 variants corresponding to phosphorylation at two serine residues previously shown to negatively affect Hsf1 activation did not have an increased temperature transition midpoint . Hsp90 known to negatively regulate Hsf1-mediated transcription decreased the slope of the temperature response curve , thereby lowering the transition midpoint and widening the response window . Our data suggest a kinetic model of Hsf1 trimerization . To elucidate temperature-induced changes in conformational dynamics , we pre-incubated monomeric Hsf1 at different temperatures for different time intervals followed by incubation at constant temperature in D2O ( Figure 2A ) . As control , we analyzed the pre-treated Hsf1 by blue native polyacrylamide gel electrophoresis ( Wittig et al . , 2006 ) and observed a temperature-dependent increase in trimeric Hsf1 species ( Figure 2B and Figure 2—figure supplement 1 ) . The 10 min-pre-incubation of Hsf1 dramatically changed conformational dynamics of two regions in Hsf1 ( Figure 2 ) : temperature-dependent increase in HX is observed in HR-C , indicating heat-induced unfolding , and a concomitant decrease in HX is observed in HR-A/B , consistent with heat-induced trimerization . Close inspection of the spectra of the peptic fragments exhibiting temperature-induced changes in HX revealed bimodal distributions of the isotope clusters indicative of the coexistence of two populations of molecules with different exchange properties ( Figure 2—figure supplement 2A and C , Figure 2—figure supplement 3 ) . An equation for two Gaussian curves was fitted to the intensity-versus-m/z plots of the data ( Figure 2D and F , and Figure 2—figure supplement 2B and D , Figure 2—figure supplement 3 ) and the equation parameters used to back calculate the contribution of each population to the peak intensities ( see Figure 2—figure supplement 2 ) . For the HR-A/B region the relative frequency of high exchanging population decreases with pre-incubation at increasing temperatures resulting in a sigmoidal temperature response curve ( Figure 2E ) . For the HR-C region the opposite is observed: the frequency of high exchanging species increased with increasing temperatures ( Figure 2G ) . These data clearly demonstrate that Hsf1 has intrinsic thermosensory properties . The midpoint of transition Tm , the temperature at which 50% of the molecules are in the high exchanging conformation after 10 min , was identical for both regions equal to 36 . 15 ± 0 . 14°C . 10 . 7554/eLife . 11576 . 004Figure 2 . Human Hsf1 is a thermosensor . ( A ) Experimental design: monomeric human Hsf1 was pre-incubated at different temperature as indicated for 10 min or 30 min and then either analyzed by blue native polyacrylamide gel electrophoresis ( BN ) or diluted 20-fold into D2O-buffer at 20°C and incubated for 30 s . The reaction was quenched and the samples analyzed by HPLC-MS . ( B ) Analysis of quaternary structure of Hsf1 after pre-incubation at different temperatures for 10 min as indicated . Hsf1 was detected by immunoblotting with an Hsf1 specific antiserum . M , monomer; D , dimer; T , trimer . ( C ) Difference plot of deuteron incorporation into monomeric Hsf1 pre-incubated at the indicated temperature minus deuteron incorporation of Hsf1 pre-incubated at 20°C for peptic peptides as indicated . Error bars are SEM of three independent experiments . Cartoons underneath the X-axis indicate the domains of Hsf1 corresponding to the respective peptic peptides and a homology model of the trimerized human HSF1 ( kindly provided by A . Bracher [Neudegger et al . , 2016] ) , colored according to HX as indicated . First and last amino acid of the model are indicated . ( D , F ) Intensity distributions of the isotope clusters of peptide 1155 . 581+ corresponding to amino acids 159–168 ( D ) and 765 . 301+ corresponding to amino acids 389–395 ( F ) for different pre-incubation temperatures , as indicated . Curves are fits of the sum of two Gaussian peak functions to the data ( see Figure 2—figure supplement 2 ) . Representative plot of three independent experiments . ( E , G ) Fraction of high-exchanging species calculated , as described in Figure 2—figure supplement 2 . Data points for three independent experiments with 10-min ( dark blue ) and 30-min ( light blue ) pre-incubation time at elevated temperatures are shown for peptide 159–168 ( E ) and 389–395 ( G ) . The curve is a fit of a thermal unfolding equilibrium to the data . Data for additional peptides are shown in Figure 2—figure supplement 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 11576 . 00410 . 7554/eLife . 11576 . 005Figure 2—figure supplement 1 . Temperature-induced transition of Hsf1 from the monomeric into the trimeric state as determined by blue native polyacrylamide gel electrophoresis . Quantification of immune blots with Hsf1-specific antisera of blue native gels of two independent experiments , one of which is shown in Figure 2B . Curves are fits of the thermal unfolding equation to the data , restricting the minimal value to 0 and the maximal value to 100 . DOI: http://dx . doi . org/10 . 7554/eLife . 11576 . 00510 . 7554/eLife . 11576 . 006Figure 2—figure supplement 2 . Analysis of the bimodal distributions of the isotope clusters detected by MS . ( A , C ) Original spectra of peptic peptide 1155 . 61+ ( amino acids 159–168; A left ) and 765 . 31+ ( aa 389–395; C , left ) and fractional peak intensities ( A , right , C , right ) for low ( blue ) and high ( red ) exchanges species calculated from the parameters of the fits of the sum of two Gaussian peaks to the data shown in Figure 2 . Unexchanged and 100% control are shown at the bottom and top , respectively . ( B , D ) Individual Gaussian peaks for high ( h ) and low ( l ) exchanging species for the indicated temperatures the sum of which results in the fits shown in Figure 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 11576 . 00610 . 7554/eLife . 11576 . 007Figure 2—figure supplement 3 . Hsf1 is a thermosensor . ( A ) Intensity distributions of the isotope clusters of the peptic peptide 678 . 33+ corresponding to amino acids 378–395 of Hsf1 pre-incubated for 10 min at different temperatures as indicated . One representative plot of three independent experiments is shown . ( B ) Fraction of high-exchanging species calculated , as described in Figure 2—figure supplement 2 . Data points for three independent experiments are shown in different shades of blue for peptide 378–395 . The curve is a fit of a thermal unfolding equilibrium to the data . ( C ) Incubation for 30 min leads to shallower transition curves . Intensity distributions of the isotope clusters of peptic peptide 535 . 82+ corresponding to amino acids 380–388 of Hsf1 pre-incubated for 30 min at different temperatures , as indicated . One representative plot of three independent experiments is shown . ( D ) Fraction of high exchanging species calculated , as described in Figure 2—figure supplement 2 . Data points for three independent experiments are shown in different shades of blue for peptide 380–388 . The curve is a fit of a thermal unfolding equilibrium to the data . DOI: http://dx . doi . org/10 . 7554/eLife . 11576 . 00710 . 7554/eLife . 11576 . 008Figure 2—figure supplement 4 . Three exemplary MS/MS spectra of peptic peptides used in the HX-MS analysis of Hsf1 . DOI: http://dx . doi . org/10 . 7554/eLife . 11576 . 008 We also determined the midpoint of transition for 30 min pre-incubation at different temperatures ( Figure 2 , Figure 2—figure supplement 3 ) . Under these prolonged incubation conditions , the transition curves became more shallow , and the midpoint temperature for HR-C unfolding and HR-A protection ( trimerization ) was not identical anymore but decreased to 32 . 0 ± 0 . 4 and 34 . 7 ± 0 . 2°C , respectively . These data suggest that the temperature-induced conformational changes are not reversible under our conditions , otherwise the steepness of the curves , which is determined by the unfolding enthalpy , should remain the same as for the 10-min-pre-incubation . To investigate this in more detail we heat-shocked Hsf1 for 10 min at 42°C , then incubated the protein for different time intervals at 20°C , and analyzed the conformational state by HX-MS ( Figure 3 ) . As control , we incubated Hsf1 without prior heat shock for 30 min at 20°C before HX-MS analysis . We also tested whether dilution would lead to trimer dissociation in the time scale of our experiments ( Figure 3G ) . Both assays clearly demonstrate that HR-C unfolding , HR-A protection and Hsf1 trimerization are not reversible under these conditions . Therefore , we cannot derive the unfolding enthalpy ∆HU from our temperature response data but only use the fit to determine the temperature at which 50% of the transition occurred . 10 . 7554/eLife . 11576 . 009Figure 3 . Heat-induced trimerization of Hsf1 is not reversible . ( A–F ) Prolonged incubation at 20°C does not revert the heat shock induced changes in Hsf1 conformation . Hsf1 ( 5 µM ) was incubated for 10 min at 42°C and then shifted to 20°C . Aliquots were diluted at different time points ( 0 , 3 , 10 , 30 , 100 min ) for 30 s into D2O buffer and subsequently analyzed by LC-MS . As control , Hsf1 was not heat-shocked and incubated for 30 min at 20°C before dilution into D2O . Shown are the intensity-m/z data for the indicated peptides from HR-A region ( A , aa 159–168 , 578 . 292+; C , aa 169–175 , 430 . 742+ ) and region HR-C ( E , aa 389–395 , 765 . 311+ ) with a global fit of an equation for two Gaussian peaks , as in Figure 2 . The percentage of high exchanging species was calculated as described in Figure 2—figure supplement 2 ( B , D , F ) . Data for two independent experiments are shown . ( G ) Dilution of heat-shocked Hsf1 does not lead to trimer dissociation . Hsf1 ( 5 µM ) was heat-shocked at 42°C for 10 min , subsequently diluted as indicated , incubated at room temperature for 15 min , analyzed by blue-native gel electrophoresis , and detected by immune blotting with an human Hsf1 specific antiserum ( lanes 1–7 ) . Control samples were kept on ice before dilution and incubation at room temperature ( lanes 8–14 ) . M , monomer; T , trimer; HO , higher order oligomers ( not detectable anymore upon dilution ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11576 . 009 To resolve the kinetics of the conformational transitions , we pre-incubated monomeric Hsf1 for 10-1000 s at 35 , 37 , 39 or 42°C before diluting into D2O and incubation for 30 s at 20°C ( Figure 4 ) . For the peptic fragments of HR-C ( amino acids 378–395 and 389–395 ) the low-exchanging population decrease with a rate of 0 . 0028 ± 0 . 0002 , 0 . 0038 ± 0 . 0002 , 0 . 011 ± 0 . 001 , and 0 . 018 ± 0 . 001 s-1 at 35 , 37 , 39 and 42°C , respectively . For the peptic fragments of HR-A/B ( amino acids 159–168 and 169–175 ) the low-exchanging population increased with slightly lower rates of 0 . 0018 ± 0 . 0004 , 0 . 0033 ± 0 . 0002 , 0 . 0094 ± 0 . 0009 and 0 . 016 ± 0 . 001 s-1 ( Figure 4G ) . The Arrhenius plot of the data yielded the activation energy for the temperature transition of 249 ± 47 kJ·mol-1 ( Figure 4H ) . Taken together , our data demonstrate that Hsf1 is a thermosensor that directly senses increasing temperatures with conformational changes in HR-A/B and HR-C . 10 . 7554/eLife . 11576 . 010Figure 4 . Kinetics of heat-induced conformational transitions in human Hsf1 . ( A–D ) Intensity distributions of the isotope clusters for peptide 678 . 323+ ( aa 378–395 ) and 1155 . 581+ ( aa 159–168 ) of Hsf1 incubated at 35°C ( A and B ) or 42°C ( C and D ) for 10 to 1000 s . Curves are fits of the sum of two Gaussian peak functions to the data . ( E and F ) Change in the fraction of low exchanging species for peptides 378–395 ( E ) and 159–168 ( F ) for 35 , 37 , 39 and 42°C , as indicated . Curves are fits of a single exponential equation to the data . ( G ) Transition rates determined by fits as in panels E and F for all four peptides ( 159–168 , 169–175 , 378–395 , 389–395 ) evaluated . ( H ) Arrhenius plot of the data shown in G . Linear regression analysis yielded an activation energy of 258 ± 25 , 273 ± 26 , 239 ± 19 , and 225 ± 22 kJ·mol-1 for peptides 159–168 , 169–175 , 378–395 , and 389–395 , respectively . Error bars are SEM of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 11576 . 010 To uncouple the temperature-induced HR-C unfolding from trimerization , we replaced the hydrophobic heptad repeat residues in HR-A by serine , which should not engage in coiled-coil interactions . HR-A and HR-B are thought to be involved in trimerization , and HR-B had been previously been implicated in negative regulation of trimerization , since deletion of HR-B lead to continuous active trimeric HSF1 ( Zuo et al . , 1994 ) . HX-MS experiments with the mutant protein revealed that HR-A/B and HR-C are constitutively unfolded at all temperatures tested ( Figure 5 ) . These results demonstrate that the ability of HR-A to form a coiled-coil is essential for stabilization of HR-A , HR-B and HR-C , suggesting that HR-C also interacts with HR-A . 10 . 7554/eLife . 11576 . 011Figure 5 . Hydrophobic residues in HR-A are essential for stability of HR-A and HR-C at all temperatures . ( A–D ) HX-MS analysis of Hsf1-I130S , V137S , L140S , V144S , M147S , M154S , L158S , M161S , L168S , V172S , L175S ( Hsf1-HR-A-S11 ) . Mutant protein was incubated at the different temperatures for 10 min and then analyzed by HX-MS . Shown are peptides from HR-A/B ( A , aa 147–169 , 634 . 804+; B , aa 170–189 , 465 . 265+ ) and HR-C ( C , aa 380–388 , 535 . 762+; D , aa 389–395 , 765 . 311+ ) . For all peptides , the 100% deuterated control is shown and for the HR-C peptides a wild-type control , which was incubated for 30 min at 20°C to emphasize the decreased stability of the mutant protein . For the peptides from HR-A , no wild-type control peptides could be shown due to different sequence and cleavage by pepsin . Shown is one of three experiments with identical results . DOI: http://dx . doi . org/10 . 7554/eLife . 11576 . 011 In our in vitro experiments , the midpoint of trimerization of Hsf1 was around 36°C , which seemed rather low given a body core temperature of 37°C , and DNA binding activity of Hsf1 in different human cells is rather low below 40°C and strongly increased above 42°C ( Abravaya et al . , 1991; Mosser et al . , 1990; Baler et al . , 1993 ) . In contrast , in testis the major increase in DNA binding activity of Hsf1 is already observed at 38°C ( Sarge , 1995; Sarge et al . , 1995 ) . Therefore , the above described unfolding of HR-C cannot be the sole determinant for the setpoint of Hsf1 trimerization and DNA binding activity . Trimerization as multi-molecular reaction would be intrinsically concentration dependent . In contrast , inhibition of trimerization by coiled-coil interaction of HR-C with HR-A/B , as originally proposed by Wu and colleagues ( Rabindran et al . , 1993 ) , would be an intramolecular reaction and consequently independent of concentration . Thus , the concentration of Hsf1 could be an important parameter for controlling the setpoint of the temperature response curve . Consistent with this hypothesis we noticed spontaneous trimerization of Hsf1 even at 4°C upon concentrating the protein . To explore this hypothesis more quantitatively , we devised a fluorescence anisotropy assay to determine the fraction of Hsf1 capable of binding DNA after treatment at different temperatures . We treated different concentrations of Hsf1 at 30 , 35 , 37 , 39 , and 42°C for 10 min , serially diluted the samples , incubated them with fluorescent labeled double-stranded DNA containing heat shock elements and measured fluorescence anisotropy ( Figure 6 ) . None of the formed Hsf1 trimers dissociated upon dilution and incubation at room temperature ( Figure 3G ) . If Hsf1 trimerization were not concentration dependent in the concentration range tested , all data points would fall onto the same titration curve . This was obviously not the case . The quadratic solution of the binding equilibrium , modified for fractional active protein was globally fitted to the data , assuming identical KD values for all formed Hsf1 trimers and similar minimum and maximum anisotropy values for no binding and complete binding , respectively . This fit yielded the fraction of DNA binding competent Hsf1 trimers . For 100 nM Hsf1 , no activation was observed up to 42°C ( Figure 6F ) . At 300 nM , less than 20% of the theoretical possible Hsf1 trimers had formed at 42°C within 10 min . In contrast , at 1 and 5 µM concentrations a substantial fraction of Hsf1 trimers had formed already at lower temperatures . At 42°C the apparent fraction of DNA-binding competent Hsf1 species may decrease due to the formation of high order oligomers as observed by blue native gel ( Figure 1C ) . 10 . 7554/eLife . 11576 . 012Figure 6 . Temperature-induced acquisition of DNA-binding competence of Hsf1 is concentration dependent . ( A–E ) DNA-binding competence of monomeric Hsf1 after pre-incubation at the indicated temperature as measured by fluorescence anisotropy of 5’-Alexa 488-labeled HSE-DNA ( 5’ccccTTCccGAAtaTTCcccc3’ ) . Monomeric Hsf1 was pre-incubated at 30-42°C at different concentrations ( 100–5000 nM ) as indicated , then twofold dilution series were prepared and labeled DNA added . Plotted is fluorescence anisotropy ( relative values ) versus theoretical concentration of Hsf1 trimer . Curves represent a global fit of the quadratic solution of the binding equilibrium modified for fractional activity of Hsf1 to all data together resulting in a KD of 1 . 10 ± 0 . 2 nM for fully active trimeric Hsf1 and a fraction of Hsf1 that formed DNA-binding competent trimers as shown in panel ( F ) . Data of one representative experiment is shown . Error bars represent standard error of the mean of four technical replicates . ( F ) Fraction of Hsf1 that formed DNA-binding competent trimers at the given Hsf1 concentration and temperature as calculated from data in panels A to E . Mean and standard error of the mean of three independent sets of experiments are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 11576 . 012 Since these data indicated that the temperature transition of Hsf1 is concentration dependent , we wondered whether only trimerization is affected or HR-C unfolding as well . We therefore repeated the HX-MS experiments at lower concentration ( 2 µM ) . Interestingly , not only trimerization but also unfolding of HR-C was concentration dependent and the differences between the transition temperatures of HR-A/B and HR-C were not statistically significant ( Figure 7B ) . However , the difference in transition temperature between Hsf1 at 5 µM and at 2 µM was highly statistically significant ( p<0 . 0001 ) . We could not test lower concentrations of Hsf1 due to lacking sensitivity in the mass spectrometric detection of the important peptides . 10 . 7554/eLife . 11576 . 013Figure 7 . Temperature response curve of Hsf1 is concentration dependent . ( A ) Fraction of high exchanging species of peptides 159–168 and 389–395 for 5 and 2 µM Hsf1 pre-incubated at the respective temperatures and analyzed by HX-MS as in Figure 2 . Data points and fits of the unfolding equilibrium equation of one representative of three independent experiments are shown . ( B ) Calculated midpoint temperature for wild-type Hsf1 ( 2 and 5 µM ) and two phosphomimetic Hsf1 variants ( 5 µM ) . For 5 µM wild-type Hsf1 ( 10-min and 30-min-incubations ) each data point represents the average of the Tm values for the two peptides observed in the respective region , which were not significantly different from each other ( HR-A: aa 159–168 and 169–175; HR-C: 389–395 and 378–395 or 380–388 ) . For 2 µM wild-type and for 5 µM mutant proteins no statistically significant differences were observed between the Tm values for the different peptides ( 159–168 , 169–175 , 378–395 , 389–395 ) within an experiment , each data point represents the average of the Tm values of all evaluated peptides for an independent experiment . In addition to the data points of three to six independent experiments , the mean and standard error of mean is shown . p<0 . 05; p<0 . 0005; **p<0 . 0001; p-values were determined by Tukey’s multiple comparisons test . DOI: http://dx . doi . org/10 . 7554/eLife . 11576 . 013 Taken together , these data clearly demonstrate that HR-C unfolding , trimerization and the fraction of DNA-competent Hsf1 trimers are a function of temperature and concentration . Human Hsf1 is heavily modified by posttranslational modifications ( 29 phosphorylation sites , 5 acetylation sites and 1 sumoylation site; www . phosphosite . org ) . Some of these modifications have been shown to influence Hsf1 activation ( Holmberg et al . , 2001; Guettouche et al . , 2005; Xia et al . , 1998; Xia and Voellmy , 1997; Chu et al . , 1996; Wang et al . , 2006; Soncin et al . , 2003; Kim et al . , 2005; Hietakangas et al . , 2006; Westerheide et al . , 2009; Raychaudhuri et al . , 2014 ) . Phosphorylation of Ser307 was proposed to negatively regulate the activation of Hsf1 , because Hsf1-S307A was constitutively active in vivo ( Chu et al . , 1996; Xia et al . , 1998 ) . In contrast , phosphorylation of the close-by Ser303 had no influence on Hsf1 activation ( Xia et al . , 1998 ) . Therefore , we constructed phosphomimetic variants of human Hsf1 ( S307D and S303D as control ) and determined their temperature response curves using HX-MS . The transition temperature for the phosphomimetic variants was slightly but statistically significantly lower than for Hsf1wt , indicating that phosphorylation at these sites does not prevent temperature-induced trimerization and might even aid it at physiological concentrations of Hsf1 ( Figure 7B ) . Several lines of evidence suggested that Hsp90 inhibits Hsf1 activation ( Zou et al . , 1998 ) and the current model assumes that Hsp90 binds Hsf1 in the monomeric state in unstressed cells ( Anckar and Sistonen , 2011 ) . We therefore studied the effect of human Hsp90β on the conformational dynamics of Hsf1 , in particular the temperature response curve ( Figure 8A–C ) . Surprisingly , in the presence of Hsp90 the midpoint of transition was lower than in its absence and the response curve was less steep , stretching the transition window from ~10° in the absence of Hsp90 to ~20° . Interestingly , the midpoint temperature of transition in the presence of Hsp90 was slightly lower for the peptide derived from HR-C than for the HR-A/B peptides ( Figure 8G ) . A significant difference in midpoint temperature for the two regions was never observed in other experiments with 10-min-pre-incubation time but with a 30-min-pre-incubation at the different temperatures ( compare Figure 8G with Figure 7B ) . 10 . 7554/eLife . 11576 . 014Figure 8 . Hsp90 modulates midpoint and steepness of the temperature response curves of human Hsf1 . ( A ) Difference plot of deuteron incorporation of human Hsf1 in the presence of Hsp90β minus deuteron incorporation into Hsf1 at the indicated temperatures . ( B and C ) Fraction of high-exchanging species of peptic peptides 169–175 ( B ) and 389–395 ( C ) of Hsf1 ( 5 µM ) pre-incubated in the absence ( blue ) and presence of human Hsp90β ( 15 µM , red ) at the indicated temperature before HX at 20°C for 30 s , quenching with low-pH buffer , peptic digestion , and MS analysis . Data points and fits of the unfolding equilibrium equation for three independent experiments are shown . ( D and E Electrophoretic mobility shift assay ( EMSA ) . HSE-DNA binding of monomeric Hsf1 pre-incubated at the indicated temperature in the absence ( D ) or presence of Hsp90 ( E ) . ( F ) Quantification of data from panels D and E . Fraction of DNA bound by Hsf1 versus temperature is plotted . Data points and fits of the unfolding equilibrium equation to the data of two independent experiments are shown . ( G ) Transition midpoints calculated from the fits of panels B , C , and F . Numbers represent p values as determined by Tukey’s multiple comparisons test; ns , not significant . DOI: http://dx . doi . org/10 . 7554/eLife . 11576 . 014 To verify that the effect of Hsp90 on trimerization is not an artifact of HX-MS methodology we performed electrophoretic mobility shift assays under comparable conditions using a fluorescently labeled DNA probe containing heat shock elements ( Figure 8D–F ) . Within the experimental error , the results of the DNA binding assay were identical to those of the HX-MS experiments ( Figure 8G ) . Taken together , our results demonstrate that in an in vitro assay with purified components , Hsp90 neither inhibits Hsf1 trimerization nor its DNA binding but , on the contrary , lowers transition temperature and widens the activation window . In this study we demonstrate that human Hsf1 is a thermosensor . We are the first to show that the HR-C region of Hsf1 unfolds with temperature-dependent rates , resulting in a release of its repressive effect on Hsf1 trimerization and DNA binding . For relatively short heat shocks unfolding of HR-C and trimerization through intermolecular interactions of HR-A/B exhibit the same temperature response curves and follow identical kinetics , suggesting that these are coupled events . For longer heat shocks the cooperativity of the transition is reduced and HR-C unfolding and HR-A/B trimerization seem to uncouple . Most importantly , HR-C unfolding , temperature-induced trimerization and acquisition of DNA-binding competence depends on the concentration of Hsf1 . Moreover , Hsp90 significantly modulated the temperature response of Hsf1 , reducing midpoint and steepness of the response curve , thus widening the temperature window within which Hsf1 transits from monomer to trimer and low to maximal DNA binding competence . The response curves of Hsf1 in the presence of Hsp90 are similar to the response curves at prolonged incubation time at elevated temperatures . Thus , Hsp90 accelerates the response at intermediate heat shock conditions . Our data with purified human Hsf1 are consistent with previous in vivo and in vitro work ( Rabindran et al . , 1993; Zhong et al . , 1998; Baler et al . , 1993; Zuo et al . , 1994; 1995; Sarge et al . , 1993; Zou et al . , 1998 ) , substantiating the hypothesis that Hsf1 trimerization and DNA binding is controlled by HR-C . Most previous work was performed in cellular systems or complex cell extracts and only few studies used purified components . Purified Drosophila Hsf1 was shown to exist in a trimer-monomer equilibrium that was influenced by temperature and oxidative stress ( Zhong et al . , 1998 ) . However , human HSF1 does not seem to exist in such an equilibrium , as we did not observe dissociation of trimeric human Hsf1 upon dilution and heat-induced trimerization was irreversible in our hands ( Figure 3 ) . These data suggest that there are principle differences between Drosophila and human Hsf1 . For human Hsf1 we observed striking differences between the temperature response curves of 10 and 30-min-incubation at elevated temperatures . How can these differences be explained ? Under our conditions temperature-induced Hsf1 transitions were irreversible . Therefore , at low temperatures even rare unfolding fluctuations of HR-C will eventually lead to trimerization , which will not be observed at short incubation times . Two effects could be responsible for the difference in Tm for HR-A/B and HR-C . At low temperature HR-C dissociation from HR-A/B and re-association might be fast as compared to trimerization . In the free state , HR-C could exist in an unfolding-refolding equilibrium , allowing exchange of protons for deuterons . If refolding is slow in comparison to the intrinsic chemical exchange rate , this would be visible as unfolded species , but HR-C would still refold and reassociate with HR-A/B , repressing trimerization . With increasing temperature unfolding rates would increase , shifting the equilibrium to the completely unfolded state , then allowing HR-A/B trimerization . Alternatively , additional temperature-induced conformational changes in HR-A/B are necessary to allow trimerization and these changes are slow as compared to HR-C unfolding at low temperatures . Phosphorylation of Ser307 was suggested to repress human Hsf1 activation because the Ser307 to Ala replacement caused constitutively active Hsf1 in vivo ( Xia et al . , 1998 ) . The temperature response curve of the phosphomimetic Hsf1-S307D variant measured by HX-MS showed a slightly reduced midpoint of transition as compared to wild-type Hsf1 , suggesting that phosphorylation at this site does not inhibit heat-induced trimerization but might rather favor it and the repressive effect must be at a different level . Our observations are consistent with more recent data on non-phosphorylatable Hsf1 variants which were not constitutively active , casting a doubt on the repressive effect of phosphorylation at this site ( Budzyński et al . , 2015 ) . Most surprising was our finding that Hsp90 does not prevent trimerization and DNA binding of Hsf1 but in the contrary reduces midpoint and steepness of the temperature response curve . This observation seems to be at odds with the known repressive function of Hsp90 on the heat shock response ( Zou et al . , 1998; Whitesell et al . , 2003; 2014; Sittler et al . , 2001; Ali et al . , 1998 ) . This discrepancy may have different reasons . First , Hsp90 also inhibits the heat shock response in yeast , although yeast Hsf1 is constitutively trimeric and bound to DNA , suggesting that Hsp90 could exert its inhibiting function on human Hsf1 after trimerization and DNA binding as well , consistent with observations for human Hsf1 ( Duina et al . , 1998; Sorger et al . , 1987; Guo et al . , 2001 ) . Second , in our experiments we only used one isoform of Hsp90 , Hsp90β , and did not add any of the some 30 co-chaperones known to assist chaperoning by Hsp90 . Further experiments with Hsp90α and different combinations of co-chaperones will be necessary to elucidate whether there exist isoform specificity in Hsf1 regulation or whether Hsp90-co-chaperone complexes have a different effect on Hsf1 trimerization than Hsp90β alone . Third , effects of Hsp90 down-regulation or inhibition could also be indirect , especially because interaction of Hsp90 with Hsf1 seems only to be observed after cross-linking ( Neef et al . , 2014; Zou et al . , 1998 ) , suggesting a very transient interaction . Hsp90 chaperones many kinases , and inhibition of Hsp90 leads to inactivation and degradation of these client proteins . Loss of such kinases could reduce inhibitory effects of phosphorylation or phosphorylation-dependent sumoylation of Hsf1 ( Soncin et al . , 2003; Wang et al . , 2006; Hietakangas et al . , 2003; 2006 ) . Originally , it was proposed that HR-C forms a coiled-coil with HR-A or HR-B to prevent trimerization in unstressed HSF1 ( Rabindran et al . , 1993; Zuo et al . , 1994 ) . Together with the observation that Hsf1 spontaneously trimerizes at high concentrations in the absence of a heat shock ( [Zhong et al . , 1998] and our own observations ) and with our HX-MS data , showing temperature-induced unfolding of HR-C , the model shown in Figure 9A can be derived . Under non-stress conditions , HSF1 is in a conformational equilibrium between a closed conformation with HR-C interacting with HR-A/B and an open conformation , in which the two heptad repeat regions are dissociated . Since association of HR-C with HR-A/B is an intramolecular interaction , limiting the diffusional freedom of the interaction partners , association rates would be very high due to the apparent high local concentration . In addition , the net charge of HR-A/B and HR-C are +6 and -7 , respectively , favoring association of the two regions by electrostatic attraction . Only at high concentrations of HSF1 association of HR-A/B regions of several HSF1 molecules , which are in the open conformation , would be able to compete with the intramolecular reaction , forming the thermodynamically more stable trimer . This would explain the spontaneous Hsf1 trimerization at high concentrations even at low temperatures ( Zhong et al . , 1998 ) . Temperature-induced unfolding of HR-C in the closed conformation favors HR-C dissociation . Alternatively or in addition , HR-C unfolding in the open conformation prevents coiled-coil interaction with HR-A/B and thus reduces the back-reaction to the closed conformation . As a consequence , trimer association is favored even at lower HSF1 concentrations . 10 . 7554/eLife . 11576 . 015Figure 9 . Kinetic models of the thermosensor function of Hsf1 . ( A ) Monomer activation model , based on the originally proposed mechanism modified with our HX-MS data . In unstressed cells monomeric Hsf1 is in equilibrium between a closed , HR-C docked to HR-A/B , and open conformation , with HR-C dissociated from HR-A/B . Owing to high local concentration and electrostatic attraction the intramolecular association rate kon , i of the HR-C–HR-A/B interaction are very high as compared to the dissociation rate koff , i . Since only uncomplexed HR-A/B can trimerize and Hsf1 trimerization therefore depends on the concentration of the open conformation , at low temperatures , trimerization only occurs at high Hsf1 concentrations . Temperature-induced unfolding of HR-C in the docked or undocked state reduces the intramolecular association rates and/or increases the dissociation rate of the intramolecular HR-C–HR-A/B complex , thereby increasing the concentration of Hsf1 in the open conformation and allowing trimerization at low Hsf1 concentrations . ( B ) Dimer activation model . At low temperatures , HR-C is constitutively docked onto HR-A/B and monomeric Hsf1 transiently dimerizes through the free part of HR-A/B . Such transient dimerization may partially destabilize the HR-C–HR-A/B interaction . At high Hsf1 concentrations a third Hsf1 monomer could interact with a transient Hsf1 dimer to form a thermodynamically stable Hsf1 trimer with completely released HR-C even at low temperatures . Increasing temperatures lead to unfolding of HR-C in the dimeric Hsf1 species leading to stabilization of the Hsf1 dimer and increased probability of trimerization . Hsp90 might modulate the temperature response by stabilizing the dimeric Hsf1 species . ( C ) Estimation of the concentration dependence of the transition temperature of Hsf1 . Data points are all the Tm values determined for 10 min incubation at elevated temperatures for Hsf1 wild type in the absence ( black ) or presence ( green ) of Hsp90 by HX-MS and by anisotropy . Black curve is a fit of the quadratic solution of the law of mass action of the monomer-dimer equilibrium , assuming that the fraction of dimer determines the Tm . This fit results in a Tm , M for the monomer of Hsf1 ( extrapolation to 0 nM ) of 53°C , the Tm , D for the dimer of 33°C , and a KD of the monomer-dimer equilibrium of 330 nM . Due to the sensitivity of the fit to data points at low Hsf1 concentrations , these are only rough estimates . The blue and red dotted lines are simulations using a lower value for KD ( 100 nM , blue ) or KD ( 200 nM ) and Tm , D ( 29°C , red ) to simulate the effect of Hsp90 . ( D ) Tentative model of the dimeric Hsf1 based on the recent crystal structure of the trimerization domain of C . thermophilum Skn7 , which formed tetramers in two different crystal forms ( PDB ID 5D5Y and 5D5Z , [Neudegger et al . , 2016] ) . HR-A , HR-B and HR-C were homology modeled on the tetrameric Skn7 using I-TASSER ( Roy et al . , 2010; Zhang , 2008; Yang and Zhang , 2015; Yang et al . , 2015 ) . HR-C was positioned to accommodate interactions with HR-A and HR-B . The homology model is colored according to HX-MS data ( Figure 1E ) . Residues of the heptad repeat involved in the tetramer interface are shown as sticks . DOI: http://dx . doi . org/10 . 7554/eLife . 11576 . 015 For two reasons we do not consider this model as very likely: First , single helices free in solutions are usually not stable but are in a rapid equilibrium with the unfolded state due to the low energy difference between helix-internal hydrogen bonds and hydrogen bonds with water . In such a state , we would not expect to see much protection in HX-MS experiments , which is in contrast to our observation ( Figure 1E ) . The amount of heat necessary to unfold a single helix seems too small to account for the temperature control , the steepness , the kinetics and the substantial activation energy of 249 kJ·mol-1 for the unfolding/trimerization transition , observed in our experiments . Second , HR-C unfolding would be independent of concentration in this model , also inconsistent with our data ( Figure 7 ) . Based on our data we propose a novel ‘dimer activation model’ ( Figure 9B ) . In this model HR-C remains bound to HR-A/B in the unstressed Hsf1 monomer . However , unstressed Hsf1 could transiently dimerize due to the larger size of the HR-A/B region ( 75 amino acids ) as compared to HR-C ( 42 amino acids ) . Such dimers would be destabilized by the interaction of HR-C with HR-A/B , resulting in high dissociation rates and a high fraction of Hsf1 monomers in unstressed cells . However , HR-A/B-HR-A/B interaction could also destabilize the interaction of HR-C with HR-A/B . At high Hsf1 concentrations , association of a third Hsf1 monomer to the transient Hsf1 dimer could occur , displacing the then more-weakly bound HR-C and leading to a stable Hsf1 trimer . Heat-induced unfolding of HR-C and possibly also its binding partner within HR-A/B ( in our experiments obscured by subsequent trimerization ) would lead to HR-C undocking . This would reduce Hsf1 dimer dissociation rates and favor Hsf1 trimerization even at low Hsf1 concentrations . In this model , HR-C unfolding and trimerization are kinetically coupled processes , at least for short heat shocks ( up to 15 min ) , explaining why they occur at identical rate constants ( Figure 4 ) . Furthermore , the average energy to unfold a protein which does not contain a co-factor is 1 . 4 kJ·mol-1 ( Privalov and Gill , 1988; Alexander et al . , 1992 ) . Dividing the activation energy for HR-C unfolding determined by us , 249 kJ·mol-1 , by 1 . 4 results in 175 , suggesting that 175 residues are involved in this unfolding process . This is close to 168 , the number of residues corresponding to a dimer of the coiled-coil between HR-C ( 42 residues ) with a similar sized region in HR-A/B . Our dimer activation model assumes that the proposed Hsf1 dimer has a lower Tm ( Tm , D ) than the Hsf1 monomer and that the measured Tm depends on the fraction of Hsf1 dimer present in the assay . We therefore plotted all of our Tm values derived for wild-type Hsf1 and incubation times of 10 min versus Hsf1 concentration and fitted the quadratic solution of the law of mass action for the monomer-dimer equilibrium to the data ( Figure 9C ) . This fit results in a KD for the monomer-dimer equilibrium , the Tm , M for the monomer and the Tm , D for the dimer . However , the derived values are only very rough estimates due to the sensitivity of the curve to values of very low concentrations of Hsf1 , which , for technical reasons , we could not determine so far . However , the steepness of the curve demonstrates that already small changes in concentration can dramatically change the transition temperature for Hsf1 activation . Although there are no concentration determinations for Hsf1 in different tissues available to our knowledge , using the relative quantification data determined by mass spectrometry for 11 different cancer cell lines ( Geiger et al . , 2012 ) , and assuming that these cells have a total protein concentration of about 150 mg/ml as determined for HEK293 cells ( Gillen and Forbush , 1999 ) , results in Hsf1 concentrations between 10 and 130 nM . The resulting Tm values would be between 47 and 53°C . However , Hsf1 is not equally distributed throughout the cytosol but shuttles in and out of the nucleus with various stress conditions preventing Hsf1 export out of the nucleus ( Vujanac et al . , 2005 ) , resulting in a locally increased Hsf1 concentration between 4- and 17-fold ( Fujioka et al . , 2006 ) and a local concentration between 40 and 2210 nM . This is well within the range that would lead to a functional heat shock response according to our model . Interestingly , the recent crystal structure of the trimerization domain of the Hsf1 homolog Skn7 of Chaetomium thermophilum contains trimers but also tetramers ( PDB ID 5D5Z and 5D5Y , [Neudegger et al . , 2016] ) . This tetramer might be a proxy for the dimer of HR-C-HR-A/B coiled-coils . To visualize how such an Hsf1 dimer might look like , we modeled the structure of HR-A , HR-B and HR-C using I-TASSER ( Roy et al . , 2010; Zhang , 2008; Yang and Zhang , 2015; Yang et al . , 2015 ) and the tetramer structure of the trimerization domain of Skn7 as template ( Figure 9D ) . Hsp90 could modulate the monomer-trimer transition by stabilizing HR-A-HR-A interactions and/or destabilize HR-A/B-HR-C interactions , resulting in a reduced dimer dissociation rate and an increased rate of trimerization at lower temperatures . Stabilization of the HR-A-HR-A dimer and concomitant destabilization of HR-A/B-HR-C interaction would automatically destabilize HR-C , since single helices are not stabile in solution and only stabilized by interaction with other structural elements , leading to a reduced unfolding transition temperature . To distinguish between stabilization of HR-A-HR-A interaction and destabilization of HR-A/B-HR-C interactions we simulated the effect of Hsp90 on the Tm by varying the apparent KD for dimerization and/or the Tm , D of the dimer ( see Figure 9C , dotted lines ) . With the current data available only changing the KD does not reduce the Tm sufficiently to fit the measured values . However , small changes in KD and Tm , D would give satisfying results . This would also explain the observation that Hsp90 reduced the Tm for HR-C unfolding significantly more than the Tm for trimerization . In this respect Hsp90 , curiously , had a similar effect as the prolonged incubation at elevated temperatures ( compare transition curves in Figure 2E and G and Figure 8B , C , and F ) . Thus , Hsp90 accelerates temperature-induced changes in conformation of Hsf1 . It is not surprising that the chaperone Hsp90 destabilizes the HR-C conformation . Chaperones have been shown to locally unfold native proteins ( Rodriguez et al . , 2008; Sharma et al . , 2010; Kirschke et al . , 2014 ) and Hsp90 is believed to destabilize an α-helix in steroid hormone receptors to allow hormone binding . How could Hsf1 be active at non-heat stress conditions , for example during development ( Xiao et al . , 1999 ) , and how could it be activated by salicylate , low pH , Ca2+ ions , hypoxia , or proteotoxic stress other than heat shock as demonstrated previously ( Mosser et al . , 1990; Jurivich et al . , 1992; Huang et al . , 1995; Liu et al . , 1996; Zhong et al . , 1999; Ahn and Thiele , 2003 ) ? According to our model Hsf1 exists in a monomer-dimer equilibrium , and trimerization of Hsf1 with subsequent DNA binding may occur continuously at low levels , promoted by Hsp90 , as shown by us , and inhibited by TriC/CCT ( Neef et al . , 2014 ) . This may ensure basal Hsf1 transcriptional activity under non-stress conditions . At the same time inhibition of Hsf1 activity by Hsp90 , Hsp70-mediated attenuation and continuous Hsf1 monomerization would keep heat shock gene transcription at a low level . Any condition that would favor Hsf1 dimerization and thus trimerization or inhibit chaperone-mediated inhibition or attenuation would , as a consequence , increase heat shock gene transcription . All of the conditions mentioned above , including development , have been associated with an imbalance in proteostasis affecting Hsf1 through titrating away chaperones . However , Hsf1 dimerization could also be affected directly by changing its local concentration , as through transport of Hsf1 into the nucleus ( Dai et al . , 2003 ) or preventing its export ( Vujanac et al . , 2005 ) , and by posttranslational modifications , including glutathionylation of the cysteine in HR-A in response to oxidative stress or alkylating agents ( Liu et al . , 1996 ) , and phosphorylation of Thr142 in HR-A ( Soncin et al . , 2003 ) , both of which would reduce the positive net charge of HR-A and thus the electrostatic repulsion . Our model could explain why the temperature setpoint of activation was lower when human Hsf1 was expressed in Drosophila cells or in Xenopus oocytes ( Baler et al . , 1993; Clos et al . , 1993 ) . In transient or stable transfection experiments , usually a strong promoter is used to express the transfected gene . Thus , the concentration of Hsf1 might have been much higher in the transfected cell than is naturally the case in human cells . Similarly , in Xenopus oocyte-injection experiments the amount of injected mRNA determines the final concentration of HSF1 and might have been so high that the resulting Hsf1 concentration might have allowed activation already at 37°C . Finally , our kinetic Hsf1 activation model would allow each cell to adjust its setpoint of activation by changing the concentration of Hsf1 by producing more Hsf1 or by concentrating it in a smaller compartment , for example by transport from the cytoplasm into the nucleus . This would easily explain the different setpoints in testis ( Sarge et al . , 1995; Sarge , 1995 ) , mouse T-lymphocytes ( Gothard et al . , 2003 ) and mouse motor neurons ( Batulan et al . , 2003 ) . This might be particularly important for cancer cells for which it was shown that Hsf1 is a driver of malignancy ( Dai et al . , 2007 ) . A culture of BL21 Rosetta , freshly transformed with a plasmid encoding the 6xHis-SUMO-Hsf1 wild-type or Hsf1-HR-A-S11 mutant sequence ( Hsf1-I130S , V137S , L140S , V144S , M147S , M154S , L158S , M161S , L168S , V172S , L175S ) , was grown at 37°C to an OD600 of 0 . 6 and then shifted to 20°C . Expression was induced by addition of IPTG to a final concentration of 0 . 1 mM , the culture grown for 2 hr at 20°C and cells were subsequently harvested by centrifugation ( 4500 × g for 15 min ) . All following steps need to be carried out at 4°C . Cell pellets were resuspended in lysis buffer ( 25 mM Hepes pH 7 . 4 , 150 mM NaCl and 10% glycerol , 3 mM β-mercaptoethanol ) containing protease inhibitors ( 10 µg/ml aprotinin , 5 µg/ml leupeptin , 8 µg/ml pepstatin , one cOmplete Protease Inhibitor Cocktail tablet [Roche Diagnostics , Mannheim , Germany] ) . Cells were disrupted by subjecting the suspension two times to a chilled microfluidizer at a pressure of 1000 bar . The resulting lysate was immediately centrifuged ( 16000 × g for 45 min ) to remove cell debris . The supernatant fraction containing 6xHis-tagged HSF1 was incubated for 20 min at 4°C with 1 g of Protino Ni2+-IDA resin ( Macherey-Nagel , Düren , Germany ) in a rotation shaker . The resin was transferred to an empty gravity-flow column and the flow-through was collected . In a first step the resin was washed with 10 column volumes ( CV ) of wash buffer and 10 CV of high salt buffer ( 25 mM Hepes pH 7 . 4 , 1 M NaCl , 10% glycerol , 3 mM β-mercaptoethanol ) . After a final washing step with another 10 CV of wash buffer the protein was eluted by addition of 1 . 5 CV elution buffer ( 25 mM Hepes pH 7 . 4 , 1 M NaCl , 10% glycerol , 3 mM β-mercaptoethanol , 250 mM imidazole ) to the column . The SUMO-Tag was cleaved off by incubation with Ulp1 SUMO-protease for 2 hr at 4°C . The cleaved Hsf1 was further separated by size-exclusion on a S200 HiLoad 16/60 column ( GE Healthcare Europe , Freiburg , Germany ) , equilibrated with Hsf1-buffer ( 25 mM Hepes pH 7 . 4 , 150 mM NaCl , 10% glycerol , 2 mM DTT ) . The fractions containing monomeric Hsf1 were adjusted to a concentration of 10 µM , flash-frozen in liquid nitrogen and stored at -80°C . 300 nM Hsf1 premixed with 200 nM Cy3–labeled HSE-oligonucleotide were incubated for 10 min either on ice or at 42°C . As a positive control purified trimeric Hsf1 was kept on ice for 10 min . After incubation the samples were kept at room temperature for additional 30 min , mixed with glycerol and loaded onto a pre-chilled 1% agarose gel ( TBE ) at 4°C . The agarose gel was run for 30 min at 150 V in the cold room . Labeled HSE-DNA was detected on a FUJI LAS-4000 fluorescence imager ( Fuji Photo Film , Düsseldorf , Germany ) . For Hsp90 experiments , 5 µM Hsf1 was premixed with 2 . 5 µM Cy3–labeled HSE-oligonucleotide and 20 µM Hsp90β in buffer containing 10 mM ATP/20 mM MgCl2 . Samples were incubated for 10 min at different temperatures ( 20°C–42°C ) , diluted 1:6 in Hsf1-buffer and incubated for 30 min at room temperature . An amount of sample containing 850 nM Hsf1 , 3 . 4 µM Hsp90β and 425 nM HSE-DNA was loaded onto a 1% agarose gel and processed as described above . For the determination of the temperature response curve of Hsf1 activation , 5 µM Hsf1 or mutants ( S303D or S307D ) in Hsf1-buffer ( 25 mM Hepes pH 7 . 4 , 150 mM NaCl , 10% glycerol , 2 mM DTT ) were heat shocked for 10 or 30 min at different temperatures ( 20°C–42°C ) . The samples were then diluted 1:20 in D2O buffer and incubated for 30 s at 20°C . Deuterated samples were quenched 1:1 with ice-cold quench buffer ( 400 mM sodium phosphate pH 2 . 2 ) , quickly injected into the injection valve and subjected to LC-MS using an Agilent UPLC and a MaXis mass spectrometer ( Bruker , Bremen , Germany ) . For each experiment at least one unexchanged sample and one fully deuterated control was measured . To determine the Hsf1 activation temperature at lower concentrations , Hsf1 was diluted to a concentration of 2 µM before the experiment . For HX-MS experiments in the presence of Hsp90β , 10 µM Hsf1 were mixed 1:1 with 40 µM human Hsp90β in reaction buffer ( 25 mM Hepes , 150 mM NaCl , 10% Glycerol , 20 mM MgCl2 , 10 mM ATP and 2 mM DTT ) and incubated for 10 min at 20°C . Equilibrated samples were then transferred to a thermomixer for a 10 min-incubation at seven different temperatures ( 20°C–42°C ) . Dilution in D2O buffer and subsequent steps were performed as described above . The unexchanged protein sample was diluted 1:20 in H2O buffer and then mixed 1:1 with quench buffer . The fully deuterated sample ( protein in Hsf1-buffer containing 6 M guanidine hydrochloride , lyophilised and redissolved in pure D2O at least three times ) was treated equally to normal samples . Data analysis was performed manually ( Data Analysis 4 . 1 , Bruker ) . 5 µM HSF1 in Hsf1-buffer were heat-shocked for different amounts of time ( 10 s , 30 s , 60 s , 100 s , 300 s , 600 s , 1000 s ) at four different temperatures ( 35°C , 37°C , 39°C , 42°C ) . The samples were then diluted 1:20 in D2O buffer and incubated for 30 s at 20°C . Deuterated samples were quenched 1:1 with ice-cold quench buffer ( 400 mM sodium phosphate pH 2 . 2 ) and quickly injected into the injection valve and subjected to LC-MS . For each experiment at least one unexchanged sample and one fully deuterated control was measured . Data analysis was performed manually ( Data Analysis 4 . 1 , Bruker ) . Evaluation of bimodal isotope peak distribution: the intensity versus m/z plots of the isotope peaks were fitted with an equation for two Gaussian peaks ( see Figures 2 , 4 or Figure 2—figure supplement 3 ) : I=A1σ·2π·e-12μ-μ1 σ 2 + A2σ·2π·e-12μ-μ2σ2 with A1/2 being the area of the two peaks; μ , the m/z values; μ1/2¯ , the means of the Gaussian peaks , representing the centroid of each of the two subpopulations; and σ , the standard deviation of the Gaussian peaks , representing the width of the isotope peak distribution ( see Figure 2—figure supplement 2B and D for individual Gaussian curves , the sum of which results in the fit curves of Figure 2 ) . For each peptide showing a bimodal distribution all intensity values belonging to one temperature ( Figures 2 , 3 , 5 , 7 , 8 ) or time ( Figure 4 ) series was globally fitted assuming that σ , μ1 and μ2 are constant within this series . Independent experiments were treated independently . Then the parameters of the fit results , A1/2 , μ1/2¯ and σ , were used to calculate for each individual isotope peak which part of the intensity belongs to the low exchanging subpopulation and which part belongs to the high exchanging subpopulation ( see Figure 2—figure supplement 2A and C panels: calculated intensity values for low [blue] and high [red] exchanging subpopulations for each isotope peak stacked on top of each other for comparison with original spectra ) . For all isotope peaks the intensities belonging to one subpopulation ( low or high ) was summed up to calculate the fraction of this subpopulation within the sample . To calculate the temperature midpoint of the transition we used the thermal unfolding equation: F=f0+ ( fmax−f0 ) ∙e ( T−Tm ) ∙ΔHR∙T∙Tm1+e ( T−Tm ) ∙ΔHR∙T∙Tm with f0 and fmax being the fraction of high exchanging subpopulation at low and high temperatures , respectively; T , absolute temperature in K; Tm , temperature at midpoint of activation; R , gas constant; ∆H , unfolding enthalpy . Hsf1 ( 10 µM ) in Hsf1-buffer were incubated for 30 min at 0°C ( control ) or 42°C ( heat shock ) . Natively purified dimer and trimer of Hsf1 were added as additional controls . After incubation 7 µg of Hsf1 were loaded on a 7% native gel or a 4–16% native gradient gel and separated by blue native polyacrylamide gel electrophoresis as described in ( Wittig et al . , 2006 ) except that Coomassie Brillant Blue G250 was only present in the sample buffer ( 0 . 2% ) not in the running buffer . For western blot analysis , an anti-Hsf1 antibody was used ( Santa Cruz Biotech , HSF1 H-311 ) . Aliquots of Hsf1 were thawed and immediately centrifuged ( 4°C , 15 min , 15000 rpm ) . In order to capture any occurring trimeric Hsf1 , the supernatant was incubated for 20 min on ice with DNA containing three HSEs coupled to magnetic beads ( 5’-CCCCTTCCCGAATATTCCCCC-3’ , 0 . 5 mg per aliquot , Dynabeads M-280 by Invitrogen ) . The supernatant concentration was determined by absorbance at 280 nm . Subsequently , four discrete concentrations of Hsf1 ( 5 µM , 1 µM , 300 nM , 100 nM ) were prepared with Hsf1-buffer and heat-shocked for 10 min at five different temperatures ( 30°C , 35°C , 37°C , 39°C and 42°C ) using a temperature-controlled water bath . Additionally , 300 nM Hsf1 was kept on ice for the same period of time as a control . Fluorescence anisotropy measurements were performed with a CLARIOstar microplate reader ( BMG Labtech ) and 384-well black flat-bottom microplates ( Corning ) in a final sample volume of 30 µL . Samples were serial diluted 1:2 until concentrations were below 1 nM . 10 nM of Alexa Fluor 488-labelled DNA containing three HSEs ( 5’-[A488]CCCCTTCCCGAATATTCCCCC-3’ ( Sigma-Aldrich ) was added by the injection system of the plate reader to start the measurement .	Cells cope with excessive heat , toxic compounds and other adverse environmental conditions by triggering an internal repair process called the heat shock response . In mammalian cells , a protein called Hsf1 is activated by stress and regulates the activity of a large set of target genes . These genes code for proteins that help the cell cope with the effects of stress , for example , by repairing or breaking down damaged proteins . Under normal conditions , Hsf1 exists as a single molecule , but when it is activated , three molecules come together to make a complex called a trimer that is able to bind to DNA and activate the target genes . Proteins are made of long chains that then fold into specific three-dimensional shapes . It is not known how Hsf1 is kept in an inactive state in healthy , unstressed cells . One possibility is that the protein folds into a three-dimensional shape that prevents it from being activated . Alternatively , Hsf1 may be bound to other proteins called chaperones that move away when the cell is under stress because they are needed to help the damaged proteins refold into their own three-dimensional shapes . Hentze et al . used a variety of biochemical techniques to study the human Hsf1 protein . The experiments showed that there are two regions of the Hsf1 protein that changed shape dramatically when the temperature increased . A region that regulates the activity of Hsf1 unfolded , while a region involved in making the trimer became more stable . Detailed analysis showed that once the regulatory region unfolded , the protein was able to interact with other Hsf1 units to make the trimer . Therefore , Hsf1 can directly sense and respond to changes in temperature without the aid of any chaperone proteins . Further experiments showed that the formation of Hsf1 trimers and the ability of these trimers to bind to DNA depend upon both the temperature and the amount of Hsf1 present . In addition , a chaperone protein called Hsp90 – which is known to be able to interact with Hsf1 – influenced how Hsf1 responded to changes in temperature . Hentze et al . also present a model for the activation of Hsf1 that allows for flexibility in the response of Hsf1 to changes in temperature . Previous studies have shown that Hsf1 is chemically modified during stress and also while the cell recovers from stressful conditions . Therefore , the next challenge will be to find out how these modifications influence the way in which Hsf1 responds to stress .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2016	Molecular mechanism of thermosensory function of human heat shock transcription factor Hsf1
Acyl-CoA binding domain-containing 7 ( Acbd7 ) is a paralog gene of the diazepam-binding inhibitor/Acyl-CoA binding protein in which single nucleotide polymorphism has recently been associated with obesity in humans . In this report , we provide converging evidence indicating that a splice variant isoform of the Acbd7 mRNA is expressed and translated by some POMC and GABAergic-neurons in the hypothalamic arcuate nucleus ( ARC ) . We have demonstrated that the ARC ACBD7 isoform was produced and processed into a bioactive peptide referred to as nonadecaneuropeptide ( NDN ) in response to catabolic signals . We have characterized NDN as a potent anorexigenic signal acting through an uncharacterized endozepine G protein-coupled receptor and subsequently via the melanocortin system . Our results suggest that ACBD7-producing neurons participate in the hypothalamic leptin signalling pathway . Taken together , these data suggest that ACBD7-producing neurons are involved in the hypothalamic control exerted on food intake and energy expenditure by the leptin-melanocortin pathway . The understanding of the complex brain controls of food intake and energy expenditure is essential to unravel the causes of excess fat deposition and to envision effective strategies to prevent or reverse obesity . The control of food intake and energy expenditure , hence the regulation of energy balance , is assured by interconnected neurons of various brain regions , which include the hypothalamus ( Morton et al . , 2006 , Schwartz et al . , 2000 , Richard , 2015 ) . Among all the hypothalamic nuclei , the arcuate nucleus ( ARC ) , which hosts proopiomelanocortin ( POMC ) and agouti-related peptide ( AgRP ) / neuropeptide ( NPY ) - producing neurons , has emerged as a prominent structure in the control of both food intake and energy expenditure ( Krashes et al . , 2013 , Mayer and Belsham , 2009 , Luquet et al . , 2005 , Gropp et al . , 2005 ) . POMC and NPY/AgRP neurons are major constituents of the melanocortin system , which is recognized to genuinely govern energy balance regulation ( Adan et al . , 2006 , Butler , 2006 , Cone , 2006 , De Jonghe et al . , 2011 , Ellacott and Cone , 2006 , Xu et al . , 2011 ) . POMC neurons release α-melanocyte-stimulating hormone ( α-MSH ) , which induces hypophagic and thermogenic effects , mainly through the stimulation of the melanocortin-4 receptor ( MC4R ) ( Adan et al . , 2006 , Butler , 2006 , Cone , 2006 ) . AgRP has been described as an inverse ( Chai et al . , 2003 , Haskell-Luevano and Monck , 2001 ) or biased ( Buch et al . , 2009 ) agonist of this receptor . In this context , the understanding of the mechanisms whereby the melanocortin regulate energy homeostasis is critical to unravel the “physiopathology” of obesity . Evidence has accumulated in recent years , which suggests that endozepines ( EZs ) could play a role in the regulation of energy balance . EZs are described as glial endogenous peptides ( Tonon et al . , 1990 ) , which are derived from the diazepam-binding inhibitor/Acyl-CoA binding protein ( Dbi/Acbp ) gene ( Mogensen et al . , 1987 , Ferrero et al . , 1984 , Corda et al . , 1984 ) , and which include DBI/ACBP itself , the triakontatetraneuropeptide ( TTN ) and the octadecaneuropeptide ( ODN ) . EZs were initially characterized for their ability to displace benzodiazepines from their binding sites ( i . e . the γ-aminobutyric acid type A receptor ( GABAA-R ) and the translocator protein ( TSPO ) ( Ferrero et al . , 1984 , Slobodyansky et al . , 1989 ) . Recently , ODN emerged as endogenous modulator of the brain melanocortin signalling pathway and as a potential actor in the hypothalamic regulation of energy homeostasis ( Lanfray et al . , 2013 ) , by acting via an uncharacterized G protein-coupled receptor ( GPCR ) ( do Rego et al . , 2007 ) . Further supporting the role of Dbi/Acbp products in energy balance , a study conducted in humans by Comuzzie and collaborators ( Comuzzie et al . , 2012 ) , indicates that DBI/ACBP single nucleotide polymorphism ( SNP ) was associated with obesity , supporting the relevance of the DBI/ACBP production in energy homeostasis . Interestingly , Comuzzie and collaborators ( Comuzzie et al . , 2012 ) also identified a SNP in a well-conserved paralog gene of the DBI/ACBP , referred as Acyl-CoA binding domain containing 7 ( ACBD7 ) , as associated to obesity in humans ( Comuzzie et al . , 2012 ) , suggesting that ACBD7 could be involved in energy homeostasis through encoding products related to EZs . The present study aimed at investigating the role of Acbd7 in energy homeostasis in mice . We hypothesised a role for Acbd7 and its encoded products in the control of energy intake and energy expenditure through modulating the networks genuinely involved in energy homeostasis , including the leptin-melanocortin circuit . In silico analysis indicates that the Acbd7 gene is a well-conserved paralog gene of the Dbi/Acbp ( Figure 1—figure supplement 1a , b ) , that could also lead to the production of bioactive fragments in several vertebrate species , including human and rodent ( Figure 1—figure supplement 1a , b , c ) . We first verified the sequence of Acbd7 mRNA in the murine brain and demonstrated a prevalent sequence ( Figure 1a ) that differs from that already described ( Acbd7-001 ) in the NCBI database ( http://www . ncbi . nlm . nih . gov/nuccore/NM_030063 . 2 ) . The observed splice variant contained 3 additional nucleotides at the splice junction of exons 2 and 3 , ( Figure 1a ) , thereby predicting the insertion of a single glycine between the Ile43 and Ala44 of the originally expected 88-amino acid-encoded ACBD7-001 protein ( herewith referred as ACBD788—Figure 1—figure supplement 1d ) and the production of a 89-amino acid-containing protein isoform ( herewith referred as ACBD789 ) . 10 . 7554/eLife . 11742 . 003Figure 1 . Acbd7 mRNA sequence , distribution and processing in the mouse brain . ( a ) Sequence of the Acbd7 mRNA open reading frame and the expected encoded protein . Additional codon and resulting amino acids are marked in red . ( b , c ) In situ hybridization showing Acbd7 mRNA distribution in mouse hypothalamus . Abbreviations: PVN , paraventricular nucleus; ARC , arcuate nucleus . Dashed lines indicate the boundaries of nuclei ( d ) Western blot performed on mediobasal hypothalamic homogenate using an ACBD7-specific antibody . MRM-MS analysis profile obtained using ( e ) synthetic NDN or ( f ) hypothalamic protein lysate as template . ( g–j ) Mediobasal hypothalamic sections labeled with ACBD7-specific antibody ( g–m; brown labelling ) and GFAP antibody ( g , h; black labelling ) , POMC antibody ( i , j; black labelling ) , NPY antibody ( k , l; black labelling ) or VGAT antibody ( m , n; black labelling ) . ( h , j , l , m ) Higher magnification views of scare panels defined in ( g ) , ( i ) , ( k ) and ( m ) respectively . Apparent co-labelling is indicated by an arrow ( j , n ) . ( o , p ) In situ hybridization analysis of Acbd7 mRNA levels in the ARC ( o ) and PVN ( p ) of ad libitum-fed mice , 18 hr-food deprived mice or 18 hr-food deprived mice having access to food 2 hr before sacrifice . Data were compared to ad libitum-fed mice as control ( n=8 ) . Data are expressed as mean ± SEM . One-way ANOVA , followed by a post-hoc multiple comparison Bonferroni test: *p<0 . 001 . ( q , r ) Western blot analysis of hypothalamic protein lysates from 18 hr-fasted mice or mice having access to food 6 hr before sacrifice , performed using ACBD7 and β-actin antibodies . ( r ) Quantification of the relative ACBD7 protein levels performed using β-actin signal as the loading control . ( s ) MRM-MS analysis of hypothalamic NDN peptide levels performed using exogenous peptide as control ( n=4 ) . Data are expressed as mean ± SEM . Unpaired Student’s t test: p< 0 . 01; **p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 00310 . 7554/eLife . 11742 . 004Figure 1—source data 1 . Synthetic NDN MRM-MS profile . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 00410 . 7554/eLife . 11742 . 005Figure 1—source data 2 . Hypothalamic lysate MRM-MS profile . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 00510 . 7554/eLife . 11742 . 006Figure 1—source data 3 . Impact of body energy status on ARC Acbd7 mRNA levels mRNA levels . In situ hybridization source data . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 00610 . 7554/eLife . 11742 . 007Figure 1—source data 4 . Impact of body energy status on PVN Acbd7 mRNA levels . In situ hybridization source data . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 00710 . 7554/eLife . 11742 . 008Figure 1—source data 5 . Impact of body energy status on hypothalamic ACBD7 levels . Source data . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 00810 . 7554/eLife . 11742 . 009Figure 1—source data 6 . Impact of body energy status on hypothalamic NDN levels . MRM-MS quantification source data . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 00910 . 7554/eLife . 11742 . 010Figure 1—figure supplement 1 . Acbd7 mRNA sequence and conservation . Schematic representation of the organization of the coding DNA sequence of ( a ) the mouse Dbi/Acbp gene and ( b ) the mouse Acbd7 gene , underlining the selective pressure occurring on these two paralog genes . ( c ) Evolutionary conservation of the ACBD7 protein sequence . Amino acids allowing maturation of the whole protein , in expected bioactive fragments , are marked in red . ( d ) Homology between DBI/ACBP and both ACBD7-001 ( i . e . ACBD788 ) and ACBD7-002 ( i . e . ACBD789 ) expected proteins in mice . Conserved amino acids between both proteins are marked in red . Expected protein maturation products resulting from tryptic digestions are surrounded by a rectangle . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 01010 . 7554/eLife . 11742 . 011Figure 1—figure supplement 2 . Relative Acbd7 mRNA levels in mice brain structures . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 01110 . 7554/eLife . 11742 . 012Figure 1—figure supplement 3 . ACBD7 labelling observed without prior colchicine or 3-MA treatment . ( A , B ) Mediobasal hypothalamic sections incoming from colchicine or 3-MA untreated labeled with ACBD7-specific antibody . ( B ) Higher magnification views of scare panels defined in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 012 Using in situ hybridization ( ISH ) , we demonstrated the presence of Acbd7 mRNA , in several brain nuclei , including the paraventricular hypothalamic nucleus ( PVH ) ( Figure 1b ) , ARC ( Figure 1c ) , piriform cortex , nucleus accumbens , median preoptic nucleus , supraoptic nucleus , medial amygdala , and hippocampus ( Figure 1—figure supplement 2 ) . Significant hybridization signal was observed in the ARC and PVH ( Figure 1b , c ) . No labelling was observed when the Acbd7 cDNA specific sense riboprobe was used ( data not shown ) , confirming the specificity of the Acbd7 riboprobe . We also demonstrated that hypothalamic cells could translate the Acbd7 mRNA into ACBD789 and to process it into a 19-amino-acid fragment ( ACBD789 ( 34-52 ) ) that we named nonadecaneuropeptide ( NDN ) . Western blot experiments , using immuno-purified ACBD7 specific antibody , revealed the presence in the mediobasal hypothalamus ( MBH ) of a single band of 10 kDa , ( Figure 1d ) , confirming that ACBD7 was endogenously produced in the MBH . In order to establish the ability of hypothalamic cells to process ACBD789 into NDN , we performed multiple reaction monitoring mass spectrometry ( MRM-MS ) investigations . This protocol performed on mice MBH extracts , using synthetic NDN as control ( Figure 1e ) , indicated that endogenous NDN was found in the hypothalamic lysates ( Figure 1f ) , thus confirming that the ACBD789 isoform is produced but also processed in the mouse brain . Immunohistochemical analysis performed using the ACBD7-specific antibody revealed that ACBD7 labelling was not observed without colchicine or 3-methyladenin ( 3-MA ) pre-treatment ( Figure 1—figure supplement 3 ) . Double immunostaining performed on brain sections of colchicine-treated mice revealed that ACBD7 was not stained in glial cells ( GFAP ) in the MBH ( Figure 1g , h ) . We also observed that ACBD7 was stained in some ARC POMC neurons ( Figure 1i , j ) and some vesicular GABA transporter ( VGAT ) positive neurons ( Figure 1m , n ) in 3-MA-treated mice , indicating that ACBD7 is predominantly produced by neurons in the MBH . However , investigation performed on colchicine-treated mice revealed no apparent ACBD7 staining in NPY neurons ( Figure 1k , l ) . Since ACBD789 appeared produced and processed by neurons of the ARC , whose involvement in energy homeostasis has been acknowledged , we investigated whether the whole body energy variations could affect the Acbd7 mRNA levels as well as the production of ACBD7 and its maturation into NDN . We observed that 18 hr of fasting was sufficient to reduce Acbd7 mRNA levels in the ARC of control mice by about 50% , whereas refeeding ( 2h ) restored Acbd7 expression ( Figure 1o ) . Notably , no effect of fasting or even refeeding challenge was observed on Acbd7 mRNA levels in the paraventricular nucleus ( PVN ) ( Figure 1p ) . Additionally , western blot analysis performed on MBH extract using fasted mice as controls , indicated that refeeding ( 6h ) increased the protein levels of ACBD7 by about 3 times ( Figure 1q , r ) , suggesting that both Acbd7 mRNA and ACBD7 protein levels in the ARC correlated with the body energy status . Thereafter , we assessed by MRM-MS the impact of refeeding challenge on the maturation of ACBD789 into NDN , within the MBH . Consistently with our previous results , the MBH levels of NDN increased 2 fold after 2 hr of refeeding ( Figure 1s ) , suggesting that production of ACBD789 and its processing into NDN was influenced by the energy status . To address the impact of ACBD7 maturation products on feeding behaviour , we examined the effect of intracerebroventricular ( icv ) injections of graded doses of NDN on food intake and compared it with that of the ACBD788-derived peptide ACBD788 ( 34-51 ) . We also tested the effect of the C-terminal octapeptide of NDN , i . e . the NDN ( 12-19 ) ( ACBD789 ( 45-52 ) ) . The injections were done in fasted mice . We observed that all injected compounds led to anorexigenic effects that were seen as early as 30 min after injection . However , dose/response experiments indicate that the optimal concentrations to cause hypophagia differed for each peptide ( Figure 2d ) . Indeed , a potent inhibition of food intake was observed in mice administrated with 10 ng ( 5 . 1 10–12 mol ) of NDN ( Figure 2a ) , while this effect was only achieved when 100 ng ( 5 . 3 10–11 mol ) of the ACBD788 ( 34-51 ) fragment were icv-injected ( Figure 2b ) . Consistently , NDN ( 12-19 ) , the common C-terminal octapeptide of ACBD788 ( 34-51 ) and NDN was efficient for doses as low as 5 ng ( 5 . 6 10–12 mol ) ( Figure 2c ) , suggesting that the biological activity of the ACBD7-derived peptides is ensured by this well-conserved fragment . Furthermore , we observed that the anorexigenic effect induced by NDN ( 12-19 ) was not present 24 hr after the icv injection ( Figure 2d ) . Interestingly , comparison of the efficiency of each peptide , 3 hr after injection , revealed that NDN as well as NDN ( 12-19 ) are both 10 times more potent than ACBD788 ( 34-51 ) , in reducing food intake by 50% ( Figure 2d ) . We also monitored the energy expenditure after acute NDN ( 10 ng ) icv injection in fasted mice . NDN increased the energy expenditure in mice ( Figure 2e , f ) , while increasing the O2 consumption ( Figure 2g ) and CO2 production ( Figure 2h ) , without affecting the respiratory quotient ( Figure 2i ) or the physical activity pattern ( Figure 2j ) , over 6 hr following the icv injection . Consistently with those results , experiments performed on fasted mice also revealed that acute icv injection of NDN ( 10 ng ) led to significant increase in the expression of uncoupling protein 1 ( Ucp1 ) mRNA level in the interscapular brown adipose tissue ( iBAT ) , 4 hr after treatment ( Figure 2k ) , thus suggesting that NDN might stimulate thermogenesis in mice . 10 . 7554/eLife . 11742 . 013Figure 2 . Effects of NDN on food intake and energy expenditure . ( a–c ) Effects of icv injection of graded doses of NDN ( a ) , ACBD788 ( 34-51 ) ( b ) , and NDN ( 12-19 ) C-terminal fragment ( c ) on cumulative food intake in 18 hr-fasted mice . Mice were injected in the right ventricle with the indicated substance , diluted in aCSF as vehicle , and had access to food 20 min later ( n=5 , 6 ) . Data are expressed as mean ± SEM . Two-way ANOVA followed by a post-hoc multiple comparison Bonferroni test: p< 0 . 05; p<0 . 01 , p<0 . 001 . ( d ) Comparison of the efficiency of each compound in terms of inhibiting food intake 3 hr after icv injection . ( e , f ) Effects of icv injection of NDN ( 10 ng ) on energy expenditure in 18 hr-fasted mice . Energy expenditure as represented by area under the curve ( AUC ) between 12 and 18 hr ( e; grey area; n=8 ) . ( g–j ) Effects of icv injection of NDN ( 10 ng ) on the O2 consumption ( g; VO2 ) , the production of CO2 ( h; VCO2 ) , the respiratory quotient ( i; RQ ) and the locomotor activity ( j ) during the first 6 hr following icv injection ( n=8 ) . ( k ) iBAT Ucp1 mRNA levels 4 hr after icv injection of NDN ( 10 ng ) ( n=7 ) . Data are expressed as mean ± SEM . Unpaired Student’s t test: p<0 . 05 , p<0 . 01 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 01310 . 7554/eLife . 11742 . 014Figure 2—source data 1 . Impact of icv injection of NDN on food intake . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 01410 . 7554/eLife . 11742 . 015Figure 2—source data 2 . Impact of icv injection of ACBD788 ( 34-51 ) on food intake . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 01510 . 7554/eLife . 11742 . 016Figure 2—source data 3 . Impact of icv injection of NDN ( 12-19 ) on food intake . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 01610 . 7554/eLife . 11742 . 017Figure 2—source data 4 . Impact of icv injection of NDN on energy expenditure . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 01710 . 7554/eLife . 11742 . 018Figure 2—source data 5 . Impact of icv injection of NDN on iBAT Ucp-1 mRNA levels . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 018 Given the aforementioned effect of central injection of NDN on the food intake behavior , we sought to determine the nature of the receptor relaying these effects . With regard to the sequence proximity between ODN and NDN , we next evaluated whether known EZ receptors could relay the anorexigenic effect of the NDN . Experiments performed in fasted mice have indicated that i . p . injections of flumazenil ( 10 mg/kg—GABAA-R benzodiazepine binding site antagonist ) ( Figure 3a ) and PK11195 ( 10 mg/kg—TSPO antagonist ) ( Figure 3b ) , were unable to blunt the anorexigenic effects of NDN , suggesting that the effect of NDN might not be relayed by classical benzodiazepines/EZ receptors . In contrast , our experiments indicated that icv injection of the cyclo1-8[DLeu5]ODN ( 11-18 ) ( LV-1075; 100 ng ) , an antagonist of the EZ GPCR ( Leprince et al . , 2001 ) , was capable of blocking the anorexigenic effect induced by the central injection of NDN ( 10 ng ) ( Figure 3c ) , suggesting that the anorexigenic effect of NDN was relayed by the uncharacterized EZ GPCR . 10 . 7554/eLife . 11742 . 019Figure 3 . Pharmacological characterization of the receptor relaying the anorexigenic effects of NDN . Mice fasted for 18 hr received a single intraperitoneal injection of flumazenil ( a; 10 mg/kg ) or PK-11195 ( b; 10 mg/kg ) diluted in 0 . 9% NaCl solution , 20 min before icv injection of NDN ( 10 ng ) or vehicle . Mice had access to food 20 min after injection and cumulative food intake was measured during the indicated periods ( n=6 ) . ( c ) Mice fasted for 18 hr were given icv injections of the endozepine metabotropic receptor antagonist cyclo1-8[DLeu5]ODN ( 11-18 ) ( LV-1075; 100 ng ) and NDN ( 10 ng ) ( n=6 ) . Mice had access to food 20 min after icv injection and cumulative food intake was measured at the indicated periods . Data are expressed as mean ± SEM . Two-way ANOVA followed by a post-hoc multiple comparison Bonferroni test:p< 0 . 05; p<0 . 01 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 01910 . 7554/eLife . 11742 . 020Figure 3—source data 1 . Impact of flumazenil treatment on the anorexigenic effect of NDN . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 02010 . 7554/eLife . 11742 . 021Figure 3—source data 2 . Impact of PK-11195 treatment on the anorexigenic effect of NDN . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 02110 . 7554/eLife . 11742 . 022Figure 3—source data 3 . Impact of LV-1075 treatment on the anorexigenic effect of NDN . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 022 We also investigated the nature of the hypothalamic pathway relaying the anorexigenic effect of NDN in mice . Previous investigations performed in rodents , using the non-selective antagonist of the MC3/4 receptors , namely SHU-9119 ( Lanfray et al . , 2013 ) , indicated that the anorexigenic effect of the ODN was relayed by the activation of the melanocortin system ( Lanfray et al . , 2013 ) . In that respect , we next evaluated the impact of NDN on the hypothalamic levels of Pomc , Npy and Agrp mRNAs . We observed that icv injection of NDN in fasted mice induced an increase in Pomc mRNA levels ( Figure 4a , b , c ) but failed to affect the mRNA levels of Agrp ( Figure 4d , e , f ) and Npy ( Figure 4g , h , i ) , suggesting that NDN was able to directly activate POMC but not AgRP/NPY neurons . Converging evidence indicating that the MC4R is the prominent melanocortin receptor in the hypothalamic regulation of whole body energy homeostasis ( Butler and Cone , 2002 , Seeley et al . , 2004 , Butler , 2006 , Ellacott and Cone , 2006 , Richard , 2015 ) , led us to evaluate the potential involvement of the MC4R signalling pathway in the effect of NDN . We observed that the anorexigenic effect induced by the icv injection of NDN ( Figure 4j ) was blocked by the injection of the MC4R-specific antagonist ( HS024—100 ng ) . Consistently , our investigations also revealed that the icv injection of NDN failed to induce anorexigenic effects in MC4R knock-out mice ( Figure 4k ) , further confirming the involvement of the melanocortin system and more specifically the MC4R as a relay in the anorexigenic effect of NDN . Additionally , investigation performed using mice hypothalamic explants revealed that NDN increases the release of the POMC-derived peptide , α-MSH , by 50% times after 45 min of treatment ( Figure 4l ) . We also investigated the impact of intra-MBH injection of NDN on food intake . We observed that NDN injection in the MBH led to a significant reduction in food intake for the first 5 hr following the central administration ( Figure 4m ) , suggesting that the anorexigenic effect of NDN is relayed by the activation of POMC neurons . 10 . 7554/eLife . 11742 . 023Figure 4 . The melanocortin system relayes the anorexigenic effects of NDN . ( a–i ) In situ hybridization analysis of Pomc mRNA levels in the mediobasal hypothalamic sections . Hypothalamic Pomc ( a , b , c ) , AgRP ( d , e , f ) and Npy ( g , h , i ) mRNA levels from 18 hr-food deprived mice icv-injected with aCSF ( a , d , g ) or NDN ( b , e , h; 100 ng ) 2 hr before sacrifice . ( c , f , i ) Relative quantification performed in the ARC . Data were compared to aCSF injected mice as control ( n=6 , 7 ) . Data are expressed as mean ± SEM . Unpaired Student’s t test: p<0 . 05 . ( j ) Mice fasted for 18 hr received icv injection of an MC4R-specific antagonist ( HS024; 100 ng ) alone or with NDN ( 10 ng ) diluted in aCSF ( n=5 , 6 ) . ( k ) Mc4r knock-out or wild type mice fasted for 18 hr were icv injected with NDN ( 10 ng ) . Mice had access to food 20 min after icv injection and cumulative food intake was measured at the indicated period ( n=6 ) . Data are expressed as mean ± SEM . Two-way ANOVA followed by a post-hoc multiple comparison Bonferroni test: p< 0 . 05 , p<0 . 01 , *p<0 . 001 . ( l ) Hypothalamic explants were preincubated in aCSF alone followed by an incubation with or without NDN ( 2 µg/ml ) . α-MSH released during incubation was normalized to the amount released during the preincubation period ( n=5 ) . Data are expressed as mean ± SEM . Unpaired t test: p<0 . 05; NS , not statistically different . ( m ) Mice fasted for 18 hr were bilaterally injected in the MBH with NDN ( 1 ng ) . Mice had access to food 20 min after injection and cumulative food intake was measured at the indicated period ( n=5 ) . Data are expressed as mean ± SEM . Two-way ANOVA followed by a post-hoc multiple comparison Bonferroni test: p< 0 . 05 , p<0 . 01 , p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 02310 . 7554/eLife . 11742 . 024Figure 4—source data 1 . Impact of NDN on the hypothalamic Pomc , AgRP , and Npy mRNA levels . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 02410 . 7554/eLife . 11742 . 025Figure 4—source data 2 . Impact of HS024 treatment on the anorexigenic effect of NDN . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 02510 . 7554/eLife . 11742 . 026Figure 4—source data 3 . Impact of NDN on food intake in MC4R-KO mice . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 02610 . 7554/eLife . 11742 . 027Figure 4—source data 4 . Impact of NDN on α-MSH release by hypothalamic explants . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 02710 . 7554/eLife . 11742 . 028Figure 4—source data 5 . Impact of intra-MBH injection of NDN on food intake . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 028 A recent report indicated that Acbd7 mRNA levels were enriched in hypothalamic cells expressing the leptin receptor ( Allison et al . , 2015 ) . In that regard , we investigated whether NDN could contribute to the hypothalamic leptin signalling . We examined ACBD7 and NDN levels in the hypothalamus of fasted mice icv-injected with leptin ( 2 µg ) using both Western blot and MS-MRM . We observed that the hypothalamic level of ACBD7 was increased more than 3 times ( Figure 5a and b ) while levels of NDN were increased by around 50% ( Figure 5c ) only 2 hr after the icv injection , suggesting that both ACBD7 and NDN levels are dynamically regulated by the leptin in the ARC . We observed that the anorexigenic effect induced by the icv injection of leptin was partially blunted by the co-injection of the LV-1075 ( 1000 ng; Figure 5d ) , while it was totally abolished by the co-injection of HS024 ( 100 ng; Figure 5d ) alone or with LV-1075 ( 1000 ng; Figure 5d ) . Interestingly , co-injection of HS024 and LV-1075 was not associated with any additional effects on leptin induced inhibition of food intake ( Figure 5d ) . Altogether , these results suggest that the endogenous EZ GPCR signalling , specifically its activation by NDN , is a part of the hypothalamic leptin signalling pathway that acts via the melanocortin signalling pathway . Finally , we investigated the impact of chronic injections of either NDN or LV-1075 on the food intake and body weight of mice ( Figure 5e , f ) . We observed that NDN induced a transient inhibition of food intake after two days of treatment ( Figure 5e ) , while a consistent reduction in body weight could be seen until the end of the treatment ( Figure 5f ) . Consistently , we also observed a significant increase in both food intake ( Figure 5e ) and body weight ( Figure 5f ) in the mice injected daily with LV-1075 after two days of treatment . 10 . 7554/eLife . 11742 . 029Figure 5 . Hypothalamic NDN production is potentially involved in the leptin signaling pathway and the control of energy homeostasis . ( a , b ) Western blot analysis of hypothalamic protein lysates from 18 hr-fasted mice icv-injected with aCSF ( control ) or leptin ( Leptin; 2 µg ) 2 hr before sacrifice , performed using ACBD7 and β-actin specific antibodies . Quantification of relative ACBD7 protein levels performed using β-actin signal as loading control ( n=4 ) . Data are expressed as mean ± SEM . Unpaired Student’s t test: p<0 . 001 . ( c ) MRM-MS relative quantification of NDN levels in MBH explants harvested from 18 hr-fasted mice and icv-injected with aCSF , or with leptin ( Leptin , 2 µg ) , 2 hr before sacrifice . Relative quantity was determined using exogenous peptide as an internal control ( n=3 ) . Data are expressed as mean ± SEM . Unpaired Student’s t test: p<0 . 05 . ( d ) Ad libitum-fed mice received icv injection of leptin ( 2 µg ) with or without the endozepine metabotropic receptor antagonist cyclo1-8[DLeu5]ODN ( 11-18 ) ( LV-1075; 1 µg ) and/or the MC4R antagonist HS024 ( 100 ng ) , 2 hr before the beginning of the dark period , and food intake was measured after 4 hr ( n=6 ) . Data are expressed as mean ± SEM . One-way ANOVA followed by a post-hoc multiple comparison Bonferroni test using aCSF injected mice as control **p<0 . 001; p<0 . 05 , or leptin injected mice as control # p<0 . 05 , ### p<0 . 001 . ( e , f ) Effect of two daily injections of NDN ( 100 ng/day ) and LV-0175 ( 1 µg/day ) on food intake ( e ) and body weight ( f ) . Mice were injected each day at ZT2 and ZT10 . Daily food intake ( e ) and body weight ( f ) were measured each day at ZT2 for 5 days ( n=6 ) . Data are expressed as mean ± SEM . Two-way ANOVA followed by a post-hoc multiple comparison Bonferroni test: p< 0 . 05; p<0 . 01 , **p<0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 02910 . 7554/eLife . 11742 . 030Figure 5—source data 1 . Impact of leptin treatment on hypothalamic ACBD7 protein levels . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 03010 . 7554/eLife . 11742 . 031Figure 5—source data 2 . Impact of leptin treatment on hypothalamic NDN levels . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 03110 . 7554/eLife . 11742 . 032Figure 5—source data 3 . Impact of acute pharmacological disruption of the EZ GPCR and the MC4R signaling pathways on the anorexigenic effect of leptin . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 03210 . 7554/eLife . 11742 . 033Figure 5—source data 4 . Impact of chronic treatment ( 2 icv injections / day ) of NDN on food intake . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 03310 . 7554/eLife . 11742 . 034Figure 5—source data 5 . Impact of chronic treatment ( 2 icv injections / day ) of NDN on body weight . DOI: http://dx . doi . org/10 . 7554/eLife . 11742 . 034 Our results demonstrate that Acbd7 is expressed and translated by some POMC neurons , as well as by some GABAergic neurons , but not by NPY/AgRP/GABAergic neurons in the ARC . This study also reveals that a splice variant of the Acbd7 mRNA is expressed in the central nervous system ( CNS ) in response to catabolic signals . This splice variant encoded a protein referred as ACBD789 which is processed into a bioactive 19-amino-acid fragment , that we called NDN ( i . e . ACBD789 ( 34-52 ) ) . Furthermore , our experiments performed in mice indicate that NDN constitutes a anorexigenic peptide , seemingly acting through an uncharacterized EZ GPCR receptor and subsequently through a pathway involving the MC4R signalling . Finally , this work underlines the potential involvement of the ARC ACBD789-producing neurons , in the hypothalamic leptin signalling pathway as well as in the endogenous control of energy homeostasis . At the time this study was initiated , the only information regarding the expression of Acbd7 in the mouse brain consisted in microarray data provides by the Brainstars database ( http://brainstars . org/probeset/1430107_at ? hide=1 ) . Here we confirm that Acbd7 mRNA is found in selective regions in the mouse CNS . Those regions include the accumbens nucleus ( Acb ) , median preoptic nucleus ( MnPO ) , supraoptic nucleus ( SON ) , PVN and ARC , which represent structures known to be involved in various behavioural and physiological regulation processes , including energy homeostasis ( Morton et al . , 2006 , Schwartz et al . , 2000 , Richard , 2015 ) . The distribution of Acbd7 mRNA contrasts with that of Dbi/Acbp mRNA , which is widely distributed in the rodent brain ( Tonon et al . , 1990 , Rouet-Smih et al . , 1992 ) . Our results also reveal that the unexpected splice variant isoform of Acbd7 mRNA ( i . e . Acbd7-002 ) , which produces a 89 amino acid-containing protein ( ACBD789 ) , is expressed in the mouse brain . Interestingly , the additional glycine in the resulting ACBD789 protein sequence , is also present in the rat ACBD7 ( http://www . ncbi . nlm . nih . gov/protein/NP_001119551 . 1 ) , suggesting that this amino acid is relevant for the biological functions of NDN in rodents . Our western blot experiments indicate that ACBD7 is produced in the mouse CNS . Furthermore , in silico analyses suggest that , as for the DBI/ACBP , ACBD789 is likely to be cleaved by tryptic digestion into shorter fragments such as NDN ( Ferrero et al . , 1984 ) . MRM-MS experiments further indicated that endogenous NDN was found in mouse MBH lysates , confirming that hypothalamic cells are capable of processing ACBD789 isoform into shorter fragments . This result further underlines that maturation processes of ACBD789 and DBI/ACBP , which lead to the production of NDN and ODN , respectively , have been well conserved during the evolution , suggesting that both maturation products have essential physiological functions . Nonetheless , our investigation does not allow for excluding the endogenous production of the ACBD788 ( 34-51 ) fragment in the mouse CNS . Finally , given the structure homology between ACBD7 and DBI/ACBP , and specifically the conservation of the Lys18 in the ACBD7 sequence , one can argue that brain cells could have the ability to produce several TTN-like compounds consisting of ACBD788 ( 18-51 ) and ACBD789 ( 18-52 ) . Likewise , considering the marked selective pressure that have occurred on those two paralog genes , future investigations will be important in assessing the ability of ACBD7 to act as an acyl-CoA carrier , as is for DBI/ACBP . Previous investigations have also characterized Dbi/Acbp , as exclusively produced by glial cells in the CNS ( Tong et al . , 1990 , Tonon et al . , 1990 ) . However no information regarding the nature of Acbd7-expressing cells was available prior to our investigation . Using a double immunohistochemistry ( IHC ) approach , we showed that ACBD7 appeared to be exclusively produced by neuronal cells of the MBH . This result indicates that , in addition to glial cells ( Tonon et al . , 1990 ) , neurons are also able to produce ODN-like compounds . Our study , which mainly focused on the ARC , further indicated that ACBD7 is produced by some POMC neurons , as well as by some GABAergic neurons . Our investigation also revealed that ACBD7 is not produced by NPY/AgRP/GABAergic neurons . Altogether these results suggest that ACBD7 is produced by some POMC neurons and GABAergic neurons distinct from the NPY/GABAergic neurons . Interestingly , a previous study conducted on astrocytes reported that the release of the DBI-derived fragment ODN , which does not contain signal peptide , involved an autophagy-based unconventional secretory pathway ( Loomis et al . , 2010 ) . In this study , 3-MA , a potent inhibitor of the autophagosome formation was shown to inhibit degradation of DBI/ACBP leading to its accumulation . In our study , we did not observe immunolabeling of ACBD7 without prior treatment with either colchicine or 3-MA . These results suggest that similar to DBI/ACBP , ACBD7 is likely processed into NDN and secreted in vivo by an autophagosome-based unconventional secretory pathway . The expression of Acbd7 ( and the production of its NDN fragment ) in the ARC provides a neuroanatomical support for a role of this gene in the hypothalamic regulation of energy balance . Further supporting this role are our data , which show that Acbd7 mRNA levels , as well as the production of ACBD7 and NDN , correlate with the energy homeostasis status in the ARC . To further substantiate the role of Acdb7 in energy homeostasis , we assessed the effects NDN on food intake and energy expenditure and demonstrated the anorectic and thermogenic effects of this peptide . Moreover , our investigation reveals that NDN is more potent than ACBD788 ( 34-51 ) in reducing food intake , which suggests that the additional glycine of the ACBD789 sequence contributes to the biological activity of the ACBD789-derived peptides . Moreover , it is noteworthy that the anorexigenic effect of NDN appears to be stronger than that of ODN in mice ( de Mateos-Verchere et al . , 2001 , do Rego et al . , 2007 ) . Interestingly , our investigations also reveal , as it has already been described for ODN ( de Mateos-Verchere et al . , 2001 , do Rego et al . , 2007 , ) , that the C-terminal octapeptide of NDN , also acts as potent anorexigenic peptide . Furthermore , the anorexigenic effect of NDN ( 12-19 ) was not observed 24 hr after icv injection , suggesting that the half-life of NDN-related peptides is short in vivo . In addition , we also observed that dose response investigations performed with either NDN or its related peptides resulted in a U-shaped dose response curve , suggesting that high concentration of NDN can non-specifically stimulate hypothalamic orexigenic pathways . Additionally , NDN induced an increase in energy expenditure in fasted mice , further supporting the role of this peptide in energy homeostasis . Pharmacological experiments performed in mice have demonstrated that the anorexigenic effect of ODN is relayed by the activation of a GPCR , still uncharacterized distinct from the classical benzodiazepine/EZ receptors ( i . e . the GABAA-R and the TSPO ) ( do Rego et al . , 2007 ) . The present results indicate that the anorexigenic effect of NDN is relayed by neither the GABAA-R nor the TSPO , suggesting that even though NDN could bind classical EZ receptor , the anorexigenic effect of this ACBD789-derived peptide seems not to be relayed by those receptors . In contrast , our investigation also revealed that the anorexigenic effect of NDN was blunted by the co-injection of the antagonist of the EZ GPCR , suggesting that similarly to ODN , NDN is also capable to act via the EZ GPCR to induce its effect . Further investigations will be necessary to fully decipher the identity of the uncharacterized EZ GPCR and to assess the ability of ACBD7-derived peptides to bind the GABAA-R and the TSPO . From the potential neuroanatomical association of ACBD7 with POMC neurons in ARC , we were interested in further studying the link between ACBD7 and the melanocortin system . Using in situ hybridization analysis , we demonstrate that the icv injection of NDN increases the ARC levels of Pomc mRNA without affecting Agrp or Npy mRNA levels , which suggests that ARC ACBD789-producing neurons may directly activate POMC neurons . Additionally , using acute pharmacological blockade of the MC4R signalling pathway as well as the MC4R knockout mouse model , we demonstrated that the anorexigenic effects of NDN were relayed via the melanocortin system . Furthermore , experiments performed on mice hypothalamic explants revealed that NDN was able to increase the secretion of α-MSH , suggesting that NDN is able to activate POMC neurons in the ARC . In addition , we demonstrated that intra-MBH injections of NDN induce strong anorexigenic effects . Altogether , these results strongly support that ARC ACBD789-producing neurons are involved in the hypothalamic regulation of food intake and energy expenditure by acting on POMC neurons through releasing NDN . However , the full mechanisms whereby NDN modulates the melanocortin system tone have yet to be fully deciphered . The involvement of the melanocortin system in the action of NDN recapitulates that seen with ODN . Previous investigations performed in rodents have indeed revealed the ability of ODN to increase Pomc mRNA level ( Compere et al . , 2003 ) and the potential involvement of the MC4R receptor in the anorexigenic effect of the ODN ( Lanfray et al . , 2013 ) . Because of the acknowledged role of the leptin - melanocortin association in energy homeostasis ( Goncalves et al . , 2014 , Hill et al . , 2010 ) and the involvement of the MC4R in the effects of NDN , we hypothesized that Acdb7 neurons could be part of the leptin-melanocortin integrated pathway . In that respect , it is noteworthy that recent data indicated that leptin effects on the melanocortin system could be relayed through an uncharacterized class of ARC GABAergic cells diverging from POMC and NPY/AgRP neurons ( Balthasar et al . , 2004 , Hill et al . , 2010 , van de Wall et al . , 2008 , Vong et al . , 2011 ) . Our results indicate that the icv injection of leptin led to an increase in the hypothalamic levels of both ACBD7 and NDN , supporting a connection between leptin and Acdb7 neurons . In that context , we sought to determine the involvement of the endogenous NDN production as part of the hypothalamic leptin signalling pathway . Investigations performed on ad libitum fed mice showed that the anorexigenic effect induced by an icv injection of leptin was partially blunted by the co-injection of the LV-1075 , indicating that the hypothalamic activation of the EZ GPCR is part of the hypothalamic leptin signalling pathway . Additionally , our investigation revealed that the anorexigenic effect of icv injection of leptin is totally abolished by the co-injection of HS024 , confirming that the melanocortin signalling pathway constitutes a major relay in the hypothalamic leptin signalling pathway ( Balthasar et al . , 2004 ) . Interestingly , no additional effects were seen when LV-1075 and HS024 were co-injected , suggesting that NDN relayed the anorexigenic effect of leptin mainly through activating POMC neurons . Although , this result does not establish a direct link between NDN and leptin signalling , it nonetheless provides further evidence for such a link i . e . LV-1075 also blocked NDN anorectic action . It is noteworthy that the hypothalamic Dbi/Acbp mRNA level is unaffected by central injection of leptin ( Compere et al . , 2010 ) , suggesting that glial EZs ( such as ODN ) are not involved in the hypothalamic leptin signalling pathway . Moreover , our observations and speculations are in line with a recent report from the translating ribosome affinity purification analysis of hypothalamic LepR positive neurons , indicating that the Acbd7 mRNA level is enriched by more than four times in hypothalamic neurons expressing the Lepr ( Allison et al . , 2015 ) . From the potential involvement of endogenous NDN production in the hypothalamic leptin signalling pathway , we were interested in further assessing the physiological involvement of endogenous NDN signalling pathway , as well as the impact of long-term exposure to NDN , on the hypothalamic regulation of energy balance . Interestingly , we found that NDN induced a significant reduction in body weight , which could be observed after two days of treatment , while it induced a transient inhibition of food intake that could only be seen on the second day of treatment , suggesting the involvement of compensatory mechanisms as previously shown in rodents for several anorexigenic neuropeptides including the α-MSH ( Lucas et al . , 2015 ) and the ODN ( de Mateos-Verchere et al . , 2001 ) . Additionally , we observed that chronic pharmacological disruption of the EZ GPCR signalling induced a potent increase in both food intake and body weight , suggesting that the endogenous endozepines signalling pathway could be a part of the hypothalamic regulation of energy homeostasis . Although , these results do not establish a direct link between endogenous production of NDN and the hypothalamic regulation of energy homeostasis , they nonetheless provide further evidence for the involvement endozepines in this process . Additionally , it is noteworthy that single nucleotide polymorphism in both the DBI/ACBP and ACBD7 are linked with morbid obesity in humans ( Comuzzie et al . , 2012 ) , suggesting that these two genes are complementary involved in the hypothalamic regulation of energy balance . In conclusion , our study identifies ACBD789 and its maturation product NDN as potential hypothalamic controllers of energy intake and energy expenditure . Our research further indicates that NDN acts as a significant anorexigenic signal through the mediation of an uncharacterized EZs GPCR and via the melanocortin signalling pathway . This work and that from others also provides converging evidence suggesting that MBH ACBD789-producing neurons are part of the hypothalamic leptin-melanocortin system as well as a major actor in the hypothalamic control of energy homeostasis . Adult male C57BL/6 mice weighting 25–30 g , and MC4R knockout mice ( Balthasar et al . , 2005 , Rossi et al . , 2011 ) weighting 45–50 g , were housed under constant temperature ( 22°C ) in a 12/12h light/dark cycle with free access to standard rodent chow ( Teklab lab animal diet , Montreal , Canada ) and drinking water . For icv injection experiments , mice were stereotaxically implanted with a permanent 22-gauge single-guide cannula ( Plastics One , Roanoke , Virginia ) aimed at the lateral ventricle using the following stereotaxic coordinates: 0 . 4 mm posterior to the bregma , 1 mm lateral to the bregma and 2 mm ventral to the skull surface . For intra-MBH injection experiments , mice were stereotaxically implanted with a permanent 22-gauge bilateral-guide cannula ( Plastics One ) targeting the MBH ( 5 . 8 mm depth , 1 . 8 mm caudal to bregma , 0 . 4 mm lateral from the sagittal suture ) . The guide cannulas were secured with screws and cranioplastic cement ( Dentsply Canada , Woodbridge , Canada , ON ) . To prevent clogging and to reduce the potential for brain infection , sterile obturators ( Plastics One ) were inserted into the guide cannulas . For icv injection , cannula placement was functionally verified before experiments using the dipsogenic effects of angiotensin 2 injections ( 5 ng ) as positive control . For intra-MBH injection , cannula placement was confirmed post-mortem . Experiments were conducted according to the Laval University Animal Ethic Committee and the Canadian Guide for the Care and Use of Laboratory Animals . Mouse hypothalamic mRNA was purified using RNeasy Mini Kit ( Qiagen , Toronto , Canada , ON ) , according to the manufacturer's instructions . The RNA concentrations were estimated from absorbance at 260 nm . cDNA synthesis was performed using random hexamer primers and the expand reverse transcriptase ( Roche Diagnostics , Laval , Canada , QC ) . The Acbd7 cDNA was amplified using primers specific to the Acbd7 sequence ( Acbd7-FP: 5’-TCCGTGTCTCATCATTATGTCC-3’; Acbd7-RP: 5’-AGGTAACCATGCTGACAGTCCT-3’ ) designed to cover the entire open reading frame of the Acbd7 mRNA ( 410-bp ) . After purification on agarose gel , PRC product was sequenced by the molecular platform of Laval University ( Quebec City , Canada , QC ) . The mouse Acbd7 cDNA riboprobe was prepared from a 410-bp fragment of the entire open reading frame of the Acbd7 mRNA using the previously described primers ( Acbd7-FP and Acbd7-RP ) . After being subcloned , into a PGEM-T plasmid ( Promega , Madison , Wisconsin ) , and linearized with NcoI and SpeI ( New England Biolabs , Whitby , Canada , ON ) , for antisense and sense probes respectively , radioactive riboprobe was synthesized by incubating the linearized plasmid ( 250 ng ) in 10 mM NaCl , 10 mM 1 , 4-dithiothreitol , 6 mM MgCl2 , 40 mM Tris ( pH 7 . 9 ) , 0 . 2 mM ATP/GTP/CTP , 100 µCi α-35S-UTP ( PerkinElmer , Foster City , California ) , 40 U RNase inhibitor ( Roche Diagnostics ) , and 20 U of RNA polymerase ( SP6 or T7 for antisense and sense probes , respectively ) for 60 min at 37°C . The riboprobe was purified using the RNeasy Mini Kit ( Qiagen ) , eluted in 150 µl of 10 mM Tris/1 mM EDTA buffer , and incorporated in a hybridization solution containing 107 cpm of 35S probe ( per ml ) , 52% formamide , 330 mM NaCl , 10 mM Tris , pH 8 , 1 mM EDTA , pH 8 , Denhardt’s solution 1x , 10% dextran sulfate , 0 . 5 mg/ml tRNA , 10 mM 1 , 4-dithiothreitol , and diethyl pyrocarbonate-treated water . This solution was mixed and heated at 65°C before being spotted on slides . The specificity of the probe was confirmed by the absence of positive signal in sections hybridized with the sense probe . After having been harvested , brains were kept in paraformaldehyde ( 4% ) for 7 days , transferred into a solution containing paraformaldehyde ( 4% ) and sucrose ( 10% ) and thereafter cut using a sliding microtome ( Histoslide 2000 , Heidelberger , Germany ) . Brain sections ( 25 µm ) were taken from the olfactory bulb to the brainstem and stored at -30°C in a cryoprotective solution containing sodium phosphate buffer ( 50 mM ) , ethylene glycol ( 30% ) , and glycerol ( 20% ) . They were mounted onto poly-L-Lysine-coated slides . The Acbd7 mRNA as well as Pomc , Npy and Agrp mRNAs were localized by in situ hybridization as previously described ( Baraboi et al . , 2010 ) . After hybridization the slides were exposed on X-ray film ( Eastman Kodak , Rochester , New York ) for 5 days ( Acbd7 ) or for 1 day ( Pomc , Npy , Agrp ) . Once removed from the autoradiography cassettes , the slides were defatted in toluene and dipped in NTB2 nuclear emulsion ( Kodak , Oakville , Canada , ON ) . The slides were exposed for 5 weeks ( Acbd7 ) or for 1 week ( Pomc , Npy , Agrp ) before being developed in D19 developer ( Kodak ) for 3 . 5 min at 14–15°C and fixed in rapid fixer ( Kodak ) for 5 min . Finally , tissues were rinsed in running distilled water for 2–3 hr , counterstained with thionin ( 0 . 25% ) , dehydrated through graded concentrations of alcohol , cleared in toluene , and coverslipped with dibutylphtalate-xylol ( DPX ) mounting medium . Analysis of amino acid conservation between mouse DBI/ACBP and ACBD7 was performed using the DIALIGN software ( http://bibiserv . techfak . uni-bielefeld . de/dialign/ ) and using the ACBD7 described sequence available from the NCBI ( http://www . ncbi . nlm . nih . gov/protein/NP_084339 . 1 ) and Ensembl website ( http://useast . ensembl . org/Mus_musculus/Gene/Family ? family=ENSFM00670001245867;g=ENSMUSG00000026644;r=2:3336168-3340993;t=ENSMUST00000115089 ) and the ACBD7 splice variant expected sequence as template . Prediction of potential cleavage sites in the ACBD788 and ACBD789 protein was performed by using the Vector NTI advance 11 software ( Invitrogen , Burlington , Canada , ON ) . Antibody directed against ACBD7 was obtained from Novus Biologicals ( Oakville , Canada , ON; NBP1-56527 ) . Rabbit secondary antibody was purchased from Santa Cruz Biotechnology ( Santa Cruz Biotechnology , Santa Cruz , California ) . Tissues ( 10 mg ) were homogenized in a lysis buffer composed of 50 mM HEPES ( pH 7 . 4 ) , 40 mM NaCl , 2 mM EDTA , 10 mM sodium pyrophosphate ( Sigma , Oakville , Canada , ON ) , 10 mM sodium glycerophosphate ( Sigma ) , 50 mM sodium fluoride ( Sigma ) , and 2 mM sodium orthovanadate ( Sigma ) , supplemented with 0 . 1% of sodium dodecyl sulfate , 1% of sodium deoxycholate ( Sigma ) , and 1% of Nonidet-P40 ( Sigma ) . One tablet of protease inhibitor cocktail ( Roche , Indianapolis , United Stated , IN ) and one tablet of phosphatase inhibitor cocktail ( Roche ) were added per 10 ml of lysis buffer . Tissues were rotated at 4°C for 10 min , and then the soluble fractions were isolated by centrifugation at 13 , 000 rpm for 10 min in a refrigerated microcentrifuge . Protein levels were then quantified using a protein assay dye reagent concentrate ( Bio-Rad , Mississauga , Canada , ON ) , and analyzed by western blotting . Briefly , 20 μg of proteins were electrophoresed ( 2 hr , 110 V ) onto NuPAGE Novex 8–16% Bis-Tris gels ( Invitrogen , Burlington , Canada , ON ) using TRIS-Glycine running buffer . Proteins were transferred onto polyvinylidene fluoride ( PVDF ) membrane ( Bio-Rad ) using the Trans-Blot Turbo transfer system ( Bio-Rad ) in transfer buffer for 7 min ( 1 . 3 A; 25V ) . After the transfer , the PVDF membrane was washed in phosphate-buffered saline ( PBS ) -Tween and incubated in blocking buffer ( 5% non-fat milk in PBS-Tween ) for 1 hr at room temperature . The membrane was then washed and incubated overnight at 4°C with antibody directed against ACBD7 ( 1:1 , 000 in 5% non-fat milk in PBS-Tween ) . The membrane was washed and incubated for 1 hr at room temperature with the secondary antibody ( 1:10 , 000 in 5% non-fat milk in PBS-Tween ) . After washing four times , the membrane was incubated with chemiluminescent substrate , ECL ( Amersham Biosciences , Arlington Heights , United Stated , IL ) , for 1 min , and signal was detected by exposing the membranes on a ECL-hyperfilm . Mice were centrally injected with colchicine ( Tocris , Minneapolis , Minnesota - 100 ng ) or 3-methyladenin ( Tocris - 3-MA , 50pmol ) , 24 hr before the sacrifice . Brains were perfused with Bouin fixative solution , harvested and embedded in paraffin . Mice brain coronal sections ( 7µm ) were sequentially incubated at room temperature in methanol containing 3% of H2O2 ( 30 min ) , PBS containing 0 . 1% of triton X-100 ( 10 min ) , HCl ( 1N , 2 min ) and PBS containing 1% of normal goat serum ( NGS ) for 1 hr . They were then incubated at 4°C for 18 hr with rabbit affinity-purified antiserum directed again ACBD7 ( 1:400; Novus biologicals ) . Sections were rinsed with PBS , incubated with SignalStain Boost IHC detection reagent ( Cell Signaling , Whitby , Canada , ON; 30 min , RT ) and labelled using SignalStain DAB substrate kit ( Cell Signaling ) . After inactivation of the initial reaction with methanol containing 3% of H2O2 ( 20 min , RT ) , secondary immunolabelling was processed . Coronal sections were sequentially incubated in PBS containing 1% of NGS and 0 . 1% Triton X-100 for 1 hr , then slices were incubated at 4°C for 18 hr with rabbit purified antiserum directed again the GFAP ( DAKO , 1/400 ) , the POMC ( Phoenix Pharmaceuticals , Burlingame , California , 1/500 ) , the NPY ( Phoenix Pharmaceuticals , 1/500 ) or the VGAT ( Novus biologicals , 1/400 ) . After having been rinsed with PBS , coronal sections were incubated with DAB/Nickel peroxidase substrate ( Vector Laboratories , Burlington , Canada , ON ) . After washes in PBS , slices were mounted in DPX . Ability of hypothalamic cell to process ACBD7 in vivo has been evaluated using multiple reactions monitoring mass spectrometry ( MRM-MS ) analysis of hypothalamic explant ( Proteomic Centre Facility of the Laval University , Québec City , Canada , QC ) . Briefly , fresh MBH hypothalamic explants were solubilized in extraction buffer containing 0 . 5% of sodium deoxycholate , 50 mM of 1 , 4-dithiothreitol and protease inhibitor . After sonication , half of the volume was precipitated by incubating with acetone ( -20°C , 12 hr ) and the remaining half part was purified using Amicon cartridge ( cut-off 10 KDa ) . Both fractions were then combined and purified using HLB Oasis cartridge ( Waters , Mississauga , Canada , ON ) . After adjustment of the sample concentration with formic acid solution ( 0 . 1% ) , 100 fmol/µl of exogenous synthetic peptide was added to each sample in order to normalize each analysis . Total mRNA was isolated from iBAT using QIAzol and the RNeasy Lipid Tissue Kit ( QIAGEN , Mississauga , ON , Canada ) . The RNA concentrations were estimated from absorbance at 260 nm and cDNA synthesis was performed using the expand reverse transcriptase ( Invitrogen , Burlington , ON , Canada ) on 1 µg of total mRNA . mRNA extraction and cDNA synthesis were performed according to the manufacturer’s instructions , and cDNA was diluted in DNAse-free water ( 1:30 ) before quantification by real-time PCR . Ucp1 and acidic ribosomal phosphoprotein P0 ( Arbp ) mRNA levels were measured in duplicate samples using a CFX96 touch real-time PCR ( Bio-Rad Laboratories , Mississauga , Canada , ON ) by using primers specific to the murine Ucp1 mRNA ( Ucp1-FP: 3’-GCAGTGTTCATTGGGCAGCC 5’; Ucp1-RP: 3’- GGACATCGCACAGCTTGGTAC-5’ ) and to the Arbp mRNA ( Arbp-FP: 3’- AGAAACTGCTGCCTCACATC-5’; Arbp-RP: 3’- CATCACTCAGAATTTCAATGG-5’ ) . Chemical detection of the PCR products was achieved with SYBR Green I ( Sigma-Aldrich , Oakville , Canada , ON ) . At the end of each run , melt curve analyses were performed , and representative samples of each experimental group were run on agarose gel to ensure the specificity of the amplification . Fold differences in target mRNA expression were measured using the ΔΔ-cycle threshold method by comparison with the house-keeping gene and expressed as fold change between vehicle versus NDN icv-injected mice . Icv injections were performed in a final volume of 2 µL of artificial cerebrospinal fluid ( aCSF; Harvard Apparatus , Saint-Laurent , Canada , QC ) . Synthetic ACBD788 ( 34-51 ) ( H-Gln-Ser-Val-Ile-Gly-Asp-Ile-Asn-Ile-Ala-Cys-Pro-Ala-Met-Leu-Asp-Leu-Lys-OH; Thermo Scientific Pierce Protein Research , ThermoFisher , Waltham , Massachusetts ) fragment was injected at doses between 10 and 1000 ng . Synthetic NDN ( H-Gln-Ser-Val-Ile-Gly-Asp-Ile-Asn-Ile-Gly-Ala-Cys-Pro-Ala-Met-Leu-Asp-Leu-Lys-OH; ThermoFisher ) corresponding to the ACBD789 ( 34-52 ) fragment was injected at doses between 0 . 1 and 1000 ng for dose-response evaluation and 10 ng for the other experiments . Synthetic NDN ( 12-19 ) ( H-Cys-Pro-Ala-Met-Leu-Asp-Leu-Lys-OH; PRIMACEN , Rouen , France ) corresponding to the ACBD789 ( 45-52 ) was injected at doses between 5 to 500 ng . The doses of icv injected cyclo1-8[DLeu5]ODN ( 11-18 ) ( Arg-Pro-Gly-Leu-DLeu-Asp-Leu-Lys; LV-1075 ) , HS024 and leptin were selected according to previous feeding experiments performed in mice ( do Rego et al . , 2007 ) . Intra-MBH dual injection was performed in a final volume of 0 . 5 µL of aCSF ( Harvard Apparatus ) by cannula . Synthetic NDN ( ThermoFisher ) was injected at the dose of 0 . 5 ng by side . Intraperitoneal injections of PK11195 ( 10mg/Kg of body weight , Tocris ) and Flumazenil ( 10mg/Kg of body weight , Sigma Aldrich ) were performed in saline solution ( NaCl , 0 . 9% ) containing 10% DMSO , 20 min before icv injection . Fasted mice ( ZT9 to ZT27 , total 18 hr ) had access to a weighed food pellet ( 25 g ) 20 min after icv injection . Cumulative food intake was measured by briefly ( less than 20 s ) removing and weighing the pellet at the indicated time points as previously described ( Lanfray et al . , 2013 , do Rego et al . , 2007 ) . Mice hypothalamus were equilibrated in aCSF ( Harvard Apparatus ) at 37°C for 1 hr under constant bubbling of 95% O2 and 5% CO2 . The explants were preincubated with fresh aCSF for 45 min followed by another incubation of 45 min with or without NDN ( 2 µg/mL ) . Measurement of α-MSH-like immunoreactivity in media was performed using ELISA kit ( MyBioSource , San Diego , CA ) . After 48 hr of acclimation in metabolic cages , mice were fasted for 24 hr ( ZT12 to ZT36 ) . They were then icv injected at ZT30 with NDN ( 10 ng ) or vehicle solution ( aCSF ) , and individually monitored by indirect calorimetry to analyse O2 consumption , CO2 production and locomotor activity . Measurements were made continuously over 6 hr ( ZT30 to ZT36 ) in an open circuit system with an oxygen analyzer ( Applied Electrochemistry , Pittsburgh , Pennsylvania , S-3A1 ) and carbon dioxide analyzer ( Applied Electrochemistry , CD-3A ) as previously described ( Sell et al . , 2004 ) . All data are expressed as mean ± SEM . Statistical analysis was performed using Student’s t test or the one way ANOVA , followed by a post hoc multiple comparisons Bonferroni test ( Prism 6 software , GraphPad , La Jolla , California ) . For each test , a value of p<0 . 05 was considered statistically significant .	Obesity is an increasingly common problem worldwide . To treat it effectively , we must understand how the body controls how much food a person consumes and how much energy they expend . The hypothalamus is one region of the brain that plays a critical role in regulating this energy balance . Some of the neurons in the hypothalamus can change their activity when they detect satiety hormones including the leptin , which is produced by fat cells and suppresses appetite . However , it is not clear exactly how the neurons respond to leptin and other energy-related signals . Recent studies have linked the gene that encodes a protein called ACBD7 with obesity , and showed that it is one of the genes that is overexpressed in neurons that are sensitive to leptin . Now , Lanfray et al . have discovered a population of neurons that produce a new variant of the protein in the hypothalamus of mice . When this protein variant matures , it can be cut down to form a small protein-like molecule called NDN . Further experiments showed that leptin stimulates the production of both the new ABCD7 variant and NDN . Lanfray et al . then injected mice that had been denied food for a several hours with NDN . The injected mice ate less than untreated mice , and burn more energy . NDN appears to form part of the signaling pathway through which leptin signals to the hypothalamus to control appetite . In the future , creating mice in which the activity of the gene that encodes ACBD7 can be easily disrupted could help to reveal more about how the hypothalamus helps to control energy balance .	[ "Abstract", "Introduction", "Results", "Discussion", "Material", "and", "methods" ]	[ "cell", "biology", "neuroscience" ]	2016	Involvement of the Acyl-CoA binding domain containing 7 in the control of food intake and energy expenditure in mice
To elucidate new functions of sphingosine ( Sph ) , we demonstrate that the spontaneous elevation of intracellular Sph levels via caged Sph leads to a significant and transient calcium release from acidic stores that is independent of sphingosine 1-phosphate , extracellular and ER calcium levels . This photo-induced Sph-driven calcium release requires the two-pore channel 1 ( TPC1 ) residing on endosomes and lysosomes . Further , uncaging of Sph leads to the translocation of the autophagy-relevant transcription factor EB ( TFEB ) to the nucleus specifically after lysosomal calcium release . We confirm that Sph accumulates in late endosomes and lysosomes of cells derived from Niemann-Pick disease type C ( NPC ) patients and demonstrate a greatly reduced calcium release upon Sph uncaging . We conclude that sphingosine is a positive regulator of calcium release from acidic stores and that understanding the interplay between Sph homeostasis , calcium signaling and autophagy will be crucial in developing new therapies for lipid storage disorders such as NPC . Sphingosine ( Sph ) and other small sphingolipids such as ceramide ( Cer ) and sphingosine 1-phosphate ( S1P ) are bioactive lipids , which play crucial roles in central cellular processes such as apoptosis , cell growth , differentiation and senescence ( Zheng et al . , 2006; Bartke and Hannun , 2009; Hannun and Obeid , 2008 ) . While Cer and S1P have been extensively studied and reviewed ( Futerman and Hannun , 2004; Spiegel and Milstien , 2003; Alvarez et al . , 2007; Pyne and Pyne , 2010; Arana et al . , 2010 ) , only limited information on the signaling mechanisms or cellular targets of Sph is available ( Jefferson and Schulman , 1988; Smith et al . , 1997; Jarvis et al . , 1997; Chang et al . , 2001 ) . In cells , Sph can be generated in lysosomes , the plasma membrane , the endoplasmic reticulum ( ER ) as well as the Golgi complex by hydrolysis of Cer through the action of ceramidases ( Bernardo et al . , 1995; Hwang et al . , 2005; Sun et al . , 2008; Xu , 2006; Mao et al . , 2001 ) . In a physiological context , Sph is reported to mediate cell growth arrest and apoptosis ( Cuvillier , 2002; Suzuki et al . , 2004 ) and is involved in the pathophysiology of the lysosomal storage disorder Niemann-Pick disease type C ( NPC ) . In this progressive neurodegenerative disease , which is caused by mutations in the lysosomal proteins NPC1 or NPC2 , Sph accumulates alongside other lipids such as cholesterol , sphingomyelin , and glycosphingolipids ( te Vruchte et al . , 2004 ) . Specifically , Sph storage is the first detectable biochemical change following inactivation of NPC1 leading to chronically lowered calcium concentration in the acidic compartment , as well as subsequent secondary lipid storage ( Lloyd-Evans et al . , 2008 ) . On the other hand , the calcium release from acidic stores was recently monitored by a genetically encoded calcium indicator directly fused to endolysosomes which showed reduced calcium release from NPC patient cells ( Shen et al . , 2012 ) in agreement with previous studies ( Lloyd-Evans et al . , 2008 ) but found that the calcium content in the lysosomes remains unchanged and that the calcium release is blocked by the accumulated lipids . This is in contrast to previous studies showing that the defect in calcium levels in NPC disease is likely due to a store filling defect and that intraluminal calcium levels were reduced using an intra-lysosomal calcium probe ( Lloyd-Evans et al . , 2008 ) . Further research is necessary to fully elucidate the interplay between calcium uptake and release from the endolysosomes and how this mechanism is deregulated in the NPC disease . Cells use calcium as a versatile tool for modulating intracellular signaling by increasing the intracellular free calcium concentration [Ca2+]i either globally or locally . Second messengers such as myo-inositol 1 , 4 , 5-trisphosphate ( IP3 ) or cyclic ADP-ribose ( cADPR ) open well-characterized calcium channels in the endoplasmic reticulum ( ER ) or the muscle sarcomere , the main intracellular calcium store ( Berridge et al . , 2003; Fliegert et al . , 2007 ) . The intraluminal calcium concentrations in the ER varies from 100–800 µM ( Burdakov et al . , 2005 ) . Since the endosomal/lysosomal system was identified as an important source of intracellular calcium ( Haller et al . , 1996 ) , the luminal calcium concentrations of acidic vesicles was determined to range between 400–600 µM ( Christensen et al . , 2002 ) . The mechanisms responsible for Ca2+filling and release from these acidic stores are not yet fully understood . The main players involved in lysosomal calcium release and signaling are transient receptor potential ( TRP ) channels such as mucolipin 1 ( TRPML1 ) ( Pryor et al . , 2006 ) and two two-pore channels ( TPC1 and TPC2 ) , which are thought to be Ca2+ channels activated by nicotinic acid adenine dinucleotide phosphate ( NAADP ) , Mg2+or phosphatidylinositol 3 , 5-bisphosphate ( PI ( 3 , 5 ) P2 ) ( Brailoiu et al . , 2009; Wang et al . , 2012; Ruas et al . , 2014; Brailoiu et al . , 2010; Rybalchenko et al . , 2012; Pitt et al . , 2010; Pitt et al . , 2014; Jha et al . , 2014; Ruas et al . , 2015; Jentsch et al . , 2015 ) . A very recent study showed that lysosomal calcium signaling regulates autophagy through the actions of the phosphatase calcineurin and the transcription factor EB ( TFEB ) ( Medina et al . , 2015 ) which places the lysosome at the center of this very important signaling hub and underlines the importance of understanding the regulation of lysosomal calcium signaling . In this work , we investigate the role of Sph in intracellular calcium signaling by employing photoactivatable ( “caged” ) Sph . Caged lipids are biologically inactive as long as they are covalently attached to certain photolabile groups ( “cage groups” ) , which prevent recognition and metabolism inside cells . The active species is then released inside cells by short irradiation with a flash of light , which induces a cleavage reaction in the cage group ( Höglinger et al . , 2014 ) . In this manner , it is possible to rapidly elevate the intracellular concentration of the lipid with precise spatiotemporal resolution . It is currently difficult to know the physiological and spatial distribution of sphingosine generated under physiological conditions , particularly inside the lysosome where the majority of sphingosine is generated via the action of acidic ceramidases . This tool therefore provides a very useful system with which to probe sphingosine cell biology in greater detail . We discovered that uncaging Sph in a variety of cell types leads to an immediate and transient increase in cytosolic calcium , which is not released from the ER but from acidic stores . We further pinpoint the endosomal/lysosomal channel TPC1 to be a main contributor to this response and demonstrate that Sph-induced calcium release leads to the nuclear translocation of TFEB . We also show that the calcium release in NPC patient cells is reduced while the Sph levels in the lysosomal compartments are increased . These findings will be helpful in elucidating the pathogenic cascade of NPC and will generally further our understanding of the function of the acidic compartments in calcium homeostasis . We synthesized caged variants of Sph by chemically attaching two different cage groups to the amino functionality of Sph ( Figure 1 ) : R1 is a diethylamino-coumarin group , which allows for very fast , sub-second uncaging kinetics ( Schade et al . , 1999; Hagen et al . , 2003 ) as well as visualization of cages lipids via its fluorescent properties . R2 is a well-established , but non-fluorescent 4 , 5-dimethoxy-2-nitrobenzyl ( NB ) group ( Il'ichev et al . , 2004 ) . In order to exclude potential artifacts caused by irradiation or cleavage of the cage groups , we also synthesized a negative control compound , coumarin-caged dihydrosphingosine ( dhSph ) , a naturally occurring lipid that is structurally very closely related to Sph . 10 . 7554/eLife . 10616 . 003Figure 1 . Structures of coumarin-caged sphingosine ( Sph-Cou ) , nitrobenzyl-caged sphingosine ( Sph-NB ) and the negative control , coumarin-caged dihydrosphingosine ( dhSph-Cou ) , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 00310 . 7554/eLife . 10616 . 004Figure 1—figure supplement 1 . Stability of caged Sph in cells . HeLa cells were labeled with a pulse of 2 µM of Sph-Cou for the indicated times and then washed . Under +UV conditions , the cells were irradiated for 2 min . Cells were lysed , subjected to lipid extraction and separated by thin layer chromatography . The background was subtracted using Fiji software . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 00410 . 7554/eLife . 10616 . 005Figure 1—figure supplement 2 . Comparative lipid analysis by mass spectrometry shows a successful uncaging reaction . Lipid extracts of HeLa cells pulsed with A ) Sph-Cou and B ) dhSph-Cou were AQC derivatized and measured on a TSQ Vantage mass spectrometer . Values are normalized to the C17 internal standards . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 005 Next , we checked whether the caged Sph was taken up by cells and if it was stable in the cellular environment . To this end , HeLa cells were incubated with 2 µM Sph-Cou for different times and subjected to +/- UV uncaging . In +UV cases , the whole dish of labelled cells was exposed to UV light from a mercury arc source for a comparably long time of 2 min to ensure complete uncaging of all cells . Cellular lipids were then extracted and coumarin-containing lipids were visualized by thin layer chromatographic analysis ( TLC , Figure 1—figure supplement 1 ) . We showed that caged Sph was taken up by cells and that its photo-induced cleavage was successful inside the cells . This experiment also demonstrated that caged Sph did not participate in lipid metabolism during up to 60 min of incubation , as no additional fluorescent bands could be detected on the TLC plate . The drawback of this analysis was that only lipids equipped with a coumarin cage could be visualized via TLC . In order to gain a more accurate estimate of the uptake , stability , and successful uncaging of Sph-Cou as well as the negative control compound dhSph , we investigated control HeLa cells , Sph-Cou and dhSph-Cou treated HeLa cells by lipidomic analysis . Figure 1—figure supplement 2 shows the relative amounts of Sph , S1P , dhSph and dhS1P detected in control , non-UV treated ( caged ) and UV-treated ( uncaged ) conditions for uncaging Sph-Cou as well as dhSph-Cou . The stability of the caged Sph ( and caged dhSph ) in the cells was confirmed because comparable levels of sphingoid bases were measured in control and caged conditions , indicating that no cleavage of the carbamate-linked cage group took place . Additionally , incubation with Sph-Cou or dhSph-Cou did not appear to perturb the long chain base homeostasis in the cells as all endogenous sphingoid bases showed similar concentrations compared to control conditions . Under the exaggerated illumination conditions used in the TLC and lipidomics experiments , Sph levels were found to increase 3 . 4 fold relative to control conditions in cells incubated with Sph-Cou , indicative of successful uncaging . This increase was specific for Sph and was not observed for dhSph ( Figure 1—figure supplement 2a ) . The levels of S1P also increased in UV-treated conditions , albeit only at a minute level ( 0 . 2% of Sph ) . We attributed this increase to ongoing metabolism during and after uncaging reaction . It seems that the long time of illumination as well as time for the following steps ( collection of the cells and extraction of the lipids ) was long enough for phosphorylation to occur on a very small fraction of Sph ( Figure 1—figure supplement 2 ) . The negative control dhSph was also uncaged successfully , as indicated by a 13 . 9 fold increase of dhSph in uncaged cells compared to control ( Figure 1—figure supplement 2b ) . This greater increase is explained by the fact that the caged lipids were added at the same concentrations , even though the endogenous levels of dhSph are 5 to 8 fold lower compared to Sph ( according to our mass spectrometric data ) . Finally , it is important to consider that the TLC and mass spectrometric experiments were performed to show that the uncaging reaction is successful in a cellular environment . They should not be interpreted as quantitative measures of the amount of released Sph . Very long illumination conditions were necessary to uncage enough cells so that these analyses were made possible . All following live-cell experiments were performed on a dual scanner confocal microscope with local uncaging of a subcellular area in the cell with simultaneous fluorescence recording . The uncaging area and uncaging times could therefore be kept to a minimum . In this manner , a greatly reduced release of Sph as well as minimal phosphorylation can be expected . In order to observe the immediate effects of photo-induced Sph release on intracellular calcium signaling in living cells , we loaded HeLa cells with the membrane-permeant calcium indicator Fluo-4/AM ( Williams et al . , 1999 ) , which reacts to increases in cytosolic calcium concentrations with corresponding increases in fluorescence intensity . We performed uncaging experiments using a confocal microscope by recording Fluo-4 fluorescence before and after brief irradiation ( 3 s–6 s ) of a local circular area ( ~9 µm2 ) within the cell . Cells reacted with an immediate ( 1–5 s ) , but transient increase in cytosolic calcium after uncaging either variant of caged Sph ( Sph-Cou or Sph-NB , Video 1 and Figure 2a ) . We calculated the mean calcium traces for each lipid and found a pronounced transient increase for both caged Sph variants , whereas the negative control compound dhSph-Cou gave no significant change in calcium levels under exactly the same delivery/uncaging conditions ( Figure 2b ) , indicating that the signal was specifically due to sphingosine . To analyze single cell responses , we calculated the maximum amplitude of each calcium trace and visualized their distribution in a histogram ( Figure 2c ) . We defined individual cells with more than 20% increase over the average baseline fluorescence as responding cells . While all cells responded when Sph-Cou/+UV treated , less than 15% ( 4 out of 29 cells ) responded under control conditions ( dhSph-Cou/+UV ) . Moreover , under control conditions , the few responding cells gave rise to markedly reduced amplitudes . 10 . 7554/eLife . 10616 . 006Video 1 . Local uncaging of Sph leads to calcium transients . Time-lapse movie of HeLa cells loaded with the calcium indicator Fluo-4 and treated with Sph-Cou . Cells were uncaged a t = 7 s for 3 s and the Fluo-4 fluorescence was recorded . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 00610 . 7554/eLife . 10616 . 007Figure 2 . Local uncaging of Sph leads to calcium transients . ( A ) Time-lapse confocal microscopy images of HeLa cells loaded with the calcium indicator Fluo-4 and treated with Sph-Cou . Cells were irradiated within the white circle at t = 7 s for 3 s . ( B ) Mean Fluo-4 fluorescence traces of cells loaded with compounds Sph-Cou ( 17 cells ) , Sph-NB ( 15 cells ) , and dhSph-Cou ( 29 cells ) , respectively . Uncaging was carried out for 3 s for coumarin-caged compounds and for 6 s for nitrobenzyl-caged sphingosine . The duration of uncaging for each lipid is represented by the color-coded bars . Traces represent mean values with the standard error of the mean plotted as error bars . ( C ) Histogram showing the distribution of the maximum observed amplitude compared to baseline of each analyzed cell . The vertical line represents the threshold set at 20% amplitude increase for responding cells . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 00710 . 7554/eLife . 10616 . 008Figure 2—figure supplement 1 . Inhibition of Sph kinases . ( A ) HeLa cells were loaded with 2 µM Sph-Cou in basal conditions or incubated with Sph kinase inhibitor N , N-dimethylsphingosine ( DMS , 1 µM , 20 cells or with SKI-II ( 10 µM , 17 cells ) . Uncaging was performed for 3 s as represented by the bar . Traces represent the mean values with the standard error of the mean plotted as error bars . ( B ) Histogram showing the distribution of the maximum observed amplitude compared to baseline with a threshold of 20% increase over baseline fluorescence . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 00810 . 7554/eLife . 10616 . 009Figure 2—figure supplement 2 . Free Sph does not induce calcium release . ( A ) HeLa cells were loaded with Fluo-4 and different concentrations of free Sph were added to the imaging medium at t=10s . Addition of 5 µM Sph leads to strong artifacts , cell blebbing and death of a fraction of the observed cells . ( B ) HeLa cells were loaded with Fluo-4 and Sph-Cou . Free Sph ( 4 µM ) was added to the imaging medium at t=10s , and uncaging was carried out for 3s as represented by the black bar . Traces represent mean values with the standard error of the mean plotted as error bars . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 009 We then addressed the possibility whether S1P as a potential metabolite of the liberated Sph and known agonist of intracellular calcium release ( Meyer zu Heringdorf et al . , 2003 ) , was essential for the calcium response . We performed uncaging experiments in the presence of two inhibitors of the sphingosine kinases: N , N-dimethylsphingosine ( DMS ) ( Edsall et al . , 1998 ) is a Sph analogue , which acts by competing for the binding of Sph . The second inhibitor , SKI-II acts as a mixed inhibitor of Sph and ATP binding ( Lim et al . , 2012 ) . When performing uncaging experiments in presence of these inhibitors , we found only slight differences of the calcium response compared to basal conditions ( Figure 2—figure supplement 1 ) . These differences might be explained by cell-to-cell variability as well as a general perturbation of the cells when using these inhibitors . We conclude that the observed calcium signals were caused by the rapid , photoinduced Sph release rather than S1P formation . Interestingly , external addition of free Sph to the medium of the cells did not evoke changes in calcium levels ( Figure 2—figure supplement 2a ) . The reasons for this are not fully resolved , but we speculate that Sph is more rapidly metabolized after entering cells than it is delivered . Sufficiently high doses of externally applied Sph ( > 5 µM ) are then cytotoxic . The rapid increase in Sph after uncaging overcomes this problem and hence leads to calcium signaling . We also added free Sph prior to the uncaging experiment and found again no difference in the calcium response . ( Figure 2—figure supplement 2b ) . To determine the source of the released calcium , we inhibited the plasma membrane calcium channels in an unspecific way by adding 5 mM Ni2+ as well as ethylene glycol tetraacetic acid ( EGTA ) to the medium to complex all extracellular calcium ( PM block ) . This inhibition did not block the Sph-induced calcium signal , although the mean amplitudes were reduced and 16 of 41 cells exhibited amplitudes below the 20% response threshold ( Figure 3a and b ) . The initial ( 5 s ) calcium increase was the same as for control conditions , demonstrating that plasma membrane calcium channels were not initially involved in this response . However , blocking PM channels leads to reduced ER calcium and thereby reduces calcium-induced calcium release from the ER . This likely accounts for the reduced amplitudes and response rates in PM block conditions . 10 . 7554/eLife . 10616 . 010Figure 3 . Investigating the source of Sph-induced calcium . ( A ) HeLa cells were loaded with 2 µM Sph-Cou in wild-type conditions ( 41 cells ) or in conditions inhibiting plasma membrane channels by removing extracellular calcium with ethylene glycol tetraacetic acid ( EGTA ) and by inhibition with 5 mM Ni2+ ( PM block , 41 cells ) or by blocking IP3 receptors at the ER using Xestospongin C at 25 µM ( 13 cells ) . Traces represent the mean values with the standard error of the mean plotted as error bars . ( B ) Histograms showing the distribution of the maximum observed amplitude change compared to baseline of all cells , with a threshold at 20% . ( C ) HeLa cells were loaded with 2 µM Sph-Cou . After addition of either 10 µM ATP to stimulate release of ER Ca2+ or addition of 200 µM glycyl-L-phenylalanine-beta-naphthylamide ( GPN ) , which leads to release of Ca2+ from the acidic stores through osmotic rupture , uncaging was performed at t = 100 s after the primary calcium transient had passed , as indicated by the bar . ( D ) Histograms showing the distribution of the maximum observed amplitude of the calcium increase after uncaging . Since these observed effects are second calcium transients , we lowered the threshold for responding cells to 10% amplitude change over baseline . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 010 To investigate the contribution of the ER as the major intracellular calcium store , we inhibited the inositol trisphosphate receptor with Xestospongin C ( Gafni et al . , 1997 ) . The calcium signals still persisted with the same initial calcium increase and only slightly reduced amplitudes , suggesting that IP3 channels were not directly involved ( Figure 3a and b ) . As a complementary approach , we stimulated the release of ER calcium by adding ATP and monitored the increase of cytosolic calcium via G-protein coupled receptor stimulation ( Figure 3c ) . After the ATP-induced calcium transient had passed , we uncaged Sph and observed a second increase in calcium . Since the ER had been previously emptied , we reasoned that this new signal must originate from a different intracellular calcium store . Also , this result further strengthens our hypothesis that liberated Sph conversion to S1P , which is known to induce calcium release from the ER ( Mattie et al . , 1994 ) , is not critical for the effect we observe with caged Sph . To determine whether the acidic compartment calcium stores are involved , we used glycyl-L-phenylalanine-beta-naphthylamide ( GPN ) , an agent which is hydrolyzed by cathepsin C in lysosomes and causes osmotic lysis of acidic vesicles , consequently releasing their content . When GPN was added to the cells , we observed an increase in cytosolic calcium , but uncaging Sph after this transient had passed failed to induce a second calcium release . Taken together , these results point towards the acidic compartment as the primary source of the Sph-induced calcium signal that we observed . To further specify the machinery involved in this Sph-induced calcium efflux from the acidic compartments , we employed mouse embryonic fibroblasts ( MEFs ) derived from mice with a single knock out of either the endosomal/lysosomal calcium channels TPC1 or TPC2 as well as MEFs from a double knock out mouse ( Ruas et al . , 2014 ) . In WT MEFs , uncaging of Sph led to an immediate cytosolic calcium increase in the same way as was observed in HeLa cells , indicating that this effect is not cell-type specific ( Figure 4a ) . Due to the different morphology and size of the fibroblasts , the amplitudes of the calcium transients were less pronounced in MEFs than in HeLa cells , since the area of uncaging was kept identical . This prompted us to set the threshold for responders to 10% increase over baseline . TPC2 knockout MEFs also gave rise to calcium transients with comparable amplitudes . Employing TPC1 knockout fibroblasts , however , resulted in a large reduction of the mean calcium response . Under these conditions , only 6 of 25 cells exhibited amplitudes greater than the response threshold and even those were markedly decreased compared to WT conditions ( Figure 4a and b ) . This striking effect shows the specificity of Sph-induced calcium release via the TPC1 channel . To confirm this , we also employed MEF from the TPC1/TPC2 double knockout mouse . Under these conditions , only 9 of 31 cells gave calcium signals with greater amplitudes than the threshold , much like in the TPC1-KO conditions . We conclude that there is a connection between an increase in cellular Sph and the release of calcium from the acidic compartments through the action of TPC1 , which is known to be localized to early , recycling , and late endosomes as well as lysosomes ( Brailoiu et al . , 2009 ) . 10 . 7554/eLife . 10616 . 011Figure 4 . Knock-out studies of two-pore channels . ( A ) Primary mouse embryonic fibroblasts derived from a wild-type mouse ( 33 cells ) or from two-pore channel 1 ( TPC1 ) knock-out mice ( 25 cells ) , two-pore channel 2 ( TPC2 , 20 cells ) knock-out mice or from a double knock-out mouse ( DKO , 31 cells ) were loaded with 2 µM Sph-Cou and uncaged for 3 s as indicated . Traces represent the mean values with standard errors of the mean plotted as error bars . ( B ) Histograms showing the distribution of the maximum observed amplitude compared to baseline , with the threshold set to 10% increase over baseline . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 01110 . 7554/eLife . 10616 . 012Figure 4—figure supplement 1 . Contribution of the lysosomal calcium channel TRPML1 . ( A ) Human fibroblasts derived from healthy patients ( control , 20 cells ) or patients with mucolipidosis type IV ( MLIV , 35 cells ) were loaded with 2 µM Sph-Cou and uncaged for 3s as indicated . Traces represent the mean values with the standard error of the mean plotted as error bars . ( B ) Histograms showing the distribution of the maximum observed amplitude compared to baseline with the threshold set at 10% increase over baseline . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 01210 . 7554/eLife . 10616 . 013Figure 4—figure supplement 2 . NAADP antagonist Ned-19 . ( A ) HeLa cells were treated with 100 µM Ned-19 for 1 h before loading with Fluo-4 ( 30 min ) and 2 µM Sph-Cou ( 15 min ) . Uncaging was performed for 3s as indicated . Traces represent the mean values with the standard error of the mean plotted as error bars . ( B ) Histograms showing the distribution of the maximum observed amplitude compared to baseline with the threshold set at 20% increase over baseline . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 013 We also investigated another important acidic store calcium channel , TRPML1 , with respect to Sph-induced calcium efflux . We used a fibroblast cell line derived from a patient suffering from mucolipidosis type IV ( MLIV ) ( Sun , 2000 ) . This disease is characterized by a loss-of-function mutation in the acidic store calcium channel TRPML1 ( Bargal et al . , 2000 ) . Again , control human fibroblasts gave an immediate calcium response upon release of Sph ( Figure 4—figure supplement 1 ) . MLIV patient fibroblasts also responded with an immediate and even higher calcium response suggesting that the observed effect is specific to TPC1 and that a loss of TRPML1 function does not interfere with Sph-induced calcium signaling . In fact , the loss of TRPML1 might even increase the importance of calcium efflux through TPCs as the only remaining way of maintaining acidic compartment calcium concentrations . Since two-pore channels are known to be critically involved in NAADP-dependent calcium signaling ( Brailoiu et al . , 2009; Tugba Durlu-Kandilci et al . , 2010 ) , we also tested the effect of a pharmacological antagonist of NAADP , Ned-19 ( Naylor et al . , 2009 ) , on Sph-induced calcium release in HeLa cells ( Figure 4—figure supplement 2 ) . Ned-19 treated HeLa cells did not exhibit a markedly different calcium response after Sph uncaging , which is in line with previous observations that TPC1 is unresponsive to Ned-19 ( Pitt et al . , 2014 ) . Several questions still remain on the interface between Sph and TPC1 . It is unclear if there is direct interaction or if there are additional , yet unidentified mediators in this response . Also , the interaction with other TPC1 ligands such as NAADP needs to be investigated . Niemann-Pick disease type C is marked by deregulated calcium homeostasis in the acidic compartments as well as an accumulation of Sph ( Lloyd-Evans et al . , 2008 ) . We investigated Sph-dependent calcium signaling in this disease by performing uncaging experiments on human fibroblast cell lines derived from a healthy subject ( control ) and from a patient suffering from NPC1 . Control cells reacted with an immediate increase in cytosolic calcium , showing again the broad applicability of these caged compounds across different cell types ( Figure 5a and b ) . In these cells , we also reduced the response threshold to 10% due to the size of the fibroblasts compared to HeLa cells . NPC1 patient cells have been shown to have reduced calcium levels in the acidic stores due to a store filling defect ( Lloyd-Evans et al . , 2008 ) . Using our setup , we could confirm that calcium transients in NPC1 cells after Sph uncaging exhibited significantly lower amplitudes ( Figure 5a and b ) . This observation corroborated our finding that Sph releases calcium from lysosomes and is consistent with other studies ( Lloyd-Evans et al . , 2008 ) showing that in NPC disease these vesicles are filled with less calcium compared to control . It should be mentioned that this is in contrast to other findings in ( Shen et al . , 2012 ) . 10 . 7554/eLife . 10616 . 014Figure 5 . Calcium signaling in NPC disease . ( A ) Human fibroblasts derived from healthy subjects ( control , 25 cells ) or patients with Niemann Pick disease type C ( NPC , 31 cells ) were loaded with 5 µM Sph-Cou ( 1 ) and uncaged for 3s as indicated . Traces represent mean values with the standard error of the mean plotted as error bars . ( B ) Histograms showing the distribution of the maximum observed amplitude compared to baseline of all cells with the response threshold set to 10% increase over baseline . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 014 Next , we investigated whether Sph is localized to the endosomal/lysosomal compartment in control as well as NPC1 fibroblasts by using a newly reported photoactivatable and clickable version of Sph ( pacSph ) ( Haberkant et al . , 2015 ) . This Sph analogue is equipped with two modifications on the hydrophobic tail: a photo-crosslinkable diaziridine moiety and a functionality that allows for click reactions , used for staining Sph in cells with a fluorophore after fixation ( Figure 6a ) . Using pacSph we could , for the first time , visualize the subcellular localization of Sph in individual control and NPC1 fibroblasts . After a short ( 10 min ) pulse of 4 µM pacSph , it was localized in endosomal and lysosomal vesicles in both cell types ( Figure 6b ) as confirmed by co-localization studies with a LAMP1 antibody ( Figure 6—figure supplement 1 ) and quantification by Pearson’s correlation coefficient ( Figure 6c ) . In NPC fibroblasts , these vesicles were bigger and much more numerous than under control conditions . They also exhibited much brighter overall fluorescence , indicative of higher cellular Sph concentrations . It has already been shown that NPC1 disease fibroblasts exhibit both lower acidic compartment calcium levels as well as an increased Sph storage ( Lloyd-Evans et al . , 2008 ) . 10 . 7554/eLife . 10616 . 015Figure 6 . Subcellular localization of Sph . ( A ) Structure of pacSph . ( B ) Sph distribution in control and NPC human fibroblasts . Cells were incubated with 4 µM pacSph for 10 min , washed and either immediately photo-crosslinked and fixed ( 0 min ) or incubated for further 10 min in buffer before crosslinking and fixation . Visualization was achieved by clicking Alexa488-azide to the terminal alkyne bond of pacSph . Confocal analysis shows a striking accumulation of Sph in the late endosomes/lysosomes of the NPC fibroblasts . For co-staining with the lysosomal marker LAMP1 , see Figure 6—figure supplement 1 . ( C ) Quantification of co-localization analysis by calculating Pearson’s correlation coefficient for >6 cells in each condition . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 01510 . 7554/eLife . 10616 . 016Figure 6—figure supplement 1 . Co-localization with lysosomal markers . Co-localization of Sph after photo-crosslinking and labeling via click chemistry with N3-Alexa 488 ( green ) and LAMP1 antibody/anti-rabbit antibody-Alexa555 ( red ) . The respective single cell Pearson’s correlation coefficient r is shown in the top left corner of the merged image . Scale bar represents 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 016 Our findings strengthen the link between Sph and lysosomal calcium homeostasis . Additionally , we followed the metabolism of pacSph in both cell lines by performing pulse chase experiments , fixing and staining cells at different time points . In control fibroblasts , the endosomal/lysosomal staining almost completely disappeared after a chase of only 10 min , as represented by a drop in the Pearson’s coefficient ( Figure 6c ) . This corresponds to the rapid efflux and transport of Sph to other cellular compartments for further metabolism ( Hannun and Obeid , 2008 ) . In contrast , NPC fibroblasts still showed a marked endosomal/lysosomal staining with high fluorescence intensities after 10 min , likely due to failed or diminished efflux from these compartments in the disease . These localization studies further support our hypothesis that elevated endosomal/lysosomal Sph concentrations and reduced calcium levels are closely connected . Lysosomal calcium release was recently shown to lead to activation of calcineurin and subsequent nuclear translocation of transcription factor EB ( TFEB ) , a master regulator of lysosomal biogenesis and autophagy ( Medina et al . , 2015 ) . To test if Sph-induced calcium release also reproduces this effect , we performed uncaging experiments in HeLa cells overexpressing TFEB-GFP . Indeed , uncaging Sph and the subsequent calcium release ( as monitored by the genetically encoded red calcium sensor , R-GECO ( Zhao et al . , 2011 ) ) induced rapid translocation of TFEB-GFP to the nucleus ( Figure 7a ) in 14 out of 19 cells . This was not the case when calcium was released by directly uncaging caged calcium to a similar extent ( Figure 7b ) . Only one in 11 cells showed a slight translocation of TFEB to the nucleus in the 15 min timeframe of the experiment . This further supports the findings that vital cellular signaling pathways are distinctly initiated on the lysosome surface . Medina et al . ( Medina et al . , 2015 ) found the mucolipin channel TRPML1 to be involved in activating calcineurin and translocating TFEB , while our findings suggest a contribution of TPC1 . This could suggest a certain redundancy in lysosomal calcium release and requires further attention . It can be speculated that reduced calcium levels in NPC patients might fail to properly regulatetranscription through TFEB and that the subsequent differences in lysosomal maintenance and autophagy couldcontribute to the NPC phenotype . 10 . 7554/eLife . 10616 . 017Figure 7 . Sph uncaging leads to TFEB translocation to the nucleus . Time-lapse confocal microscopy images of HeLa cells transfected with TFEB-GFP and R-GECO ( a genetically encoded calcium sensor ) and loaded with either Sph-Cou or caged calcium ( o-nitrophenyl EGTA/AM ) . Uncaging was performed in a small area within the cell for 3 s for Sph-Cou ( lower cell ) and 2 s for caged calcium . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 017 In summary , we have developed a method to increase the concentration of Sph in living cells using light in a spatially and temporally precise way . Thus , we interfered with sphingolipid metabolism minimally and were able to investigate resulting effects of elevated sphingosine levels . Dihydrosphingosine , although structurally very similar , served as a good negative control compound and underlined the high structural specificity of the interacting partners of Sph . We applied these tools to a variety of cell types , ranging from cultured HeLa cells to primary mouse embryonic fibroblasts and human patient fibroblasts and showed that in all those cells , an acute increase in Sph was always followed by an immediate release of calcium from GPN-sensitive acidic stores . We also showed that plasma membrane and ER calcium channels were not initially involved in this calcium response . Furthermore , we could pinpoint TPC1 , an endosomal/lysosomal channel of the two-pore channel family to be essential in this newly found signaling pathway . When employing embryonic TPC1 knockout fibroblasts , no calcium increase upon uncaging of Sph was observed . TPC2 knockout fibroblasts , however , reacted similarly to WT fibroblasts , confirming the specific requirement of the TPC1 isoform for Sph induced calcium signaling . The involvement of the other main acidic compartment calcium channel TRPML1 was also tested . Loss-of-function in TRPML1 led to a slight increase in Sph-induced calcium release , which might be explained by some redundancy between TRPML1 and TPCs in the regulation of lysosomal calcium concentrations . Acidic compartment calcium signaling has recently emerged as an important part of intracellular events such as vesicle fusion and secretion ( Patel and Docampo , 2010 ) as well as induction of autophagy and lysosomal biogenesis ( Medina et al . , 2015 ) . Our knowledge of the machinery involved and its regulation , however , remains incomplete . The two-pore channels ( TPCs ) are the subject of current controversies regarding ion selectivity and dependencies on activating ligands . While many reports agree that TPC channels are NAADP-regulated Ca2+ channels ( Brailoiu et al . , 2009; Brailoiu et al . , 2010; Pitt et al . , 2010; Calcraft et al . , 2009 ) , others indicate a broader ion specificity of TPCs ( Wang et al . , 2012; Rybalchenko et al . , 2012; Peiter et al . , 2005 ) or alternative ligands such as PI ( 3 , 5 ) P2 ( Wang et al . , 2012; Cang et al . , 2013 ) . Another study hints towards a convergent regulation of TPC2 by Mg2+ , NAADP , PI ( 3 , 5 ) P2 and two protein kinases ( JNK and p38 ) ( Jha et al . , 2014 ) . A very recent study confirmed that TPCs conduct both Ca2+ and Na+ ions and that they are activated by NAADP as well as PI ( 3 , 5 ) P2 ( Ruas et al . , 2015; Jentsch et al . , 2015 ) . The biological relevance of two-pore channels was recently highlighted by a report showing that TPCs are crucial to successful infection of host cells by the Ebola virus and that disruption of TPC function leads to halted virus trafficking ( Sakurai et al . , 2015 ) . Interestingly , the progression of Ebola infection is also dependent on the NPC1 protein , which was shown to be necessary for viral escape from the lysosomal compartment ( Herbert , et al . , 2015 , Carette , et al . , 2011 ) . These recent findings will certainly encourage further efforts to understand the link between lysosomal calcium signaling and the defects in Niemann-Pick disease . In our study , we could confirm that NPC patient fibroblasts responded with lower amplitudes of the calcium transients after Sph uncaging , which is indicative of a reduced pool of calcium in the acidic compartments of these cells , in accordance with other studies ( Lloyd-Evans et al . , 2008; Shen et al . , 2012 ) . Figure 8 summarizes our results and illustrates that we pinpointed TPC1 as the mediator of a Sph induced rise in cytosolic calcium . We also correlated a reduced calcium release in fibroblasts from NPC patients with a significant storage of Sph in the late endosomes/lysosomes of these patient cells by using a novel bifunctional Sph probe . Short pulse-chase experiments also indicated lack of efflux from lysosomes in NPC fibroblasts compared to cells from healthy controls . 10 . 7554/eLife . 10616 . 018Figure 8 . Schematic summary of findings . Sph causes calcium release from acidic stores through the action of TPC1 . In Niemann-Pick disease type C , lower endolysosomal calcium levels occur together with increased Sph concentration . Stimulating lysosomal calcium release by uncaging Sph in NPC patient cells results in lower calcium amplitudes compared to cells from control subjects . DOI: http://dx . doi . org/10 . 7554/eLife . 10616 . 018 The question whether the NPC1 protein is directly or indirectly involved in the efflux of lipids , specifically Sph , from the acidic compartments has remained a controversial topic . NPC1 has been shown to be involved in the transport of amines ( Kaufmann and Krise , 2008 ) , but conclusive data on the transport of Sph and other potential substrates are absent as there is no direct functional assay for NPC1 . A study using radiolabelled sphingolipids to follow metabolism and transport in NPC disease concluded that Sph efflux from endolysosomes is not restricted in NPC ( Blom et al . , 2012 ) . However , these data were obtained using long , 1–4 h chase times during which the precursor lipid was likely metabolized to a high degree and incorporated into other lipids , leaving only a small fraction of Sph for quantitative analysis . This and other obstacles to the study of lipids affirm the need for new tools , which circumvent the lipid metabolic machinery and give a faithful representation of the subcellular lipid localization . With our tools and shorter chase times in the range of 10–20 min , we observed reduced acidic compartment calcium release as well as increased Sph storage and a failure in Sph efflux in NPC disease cells . Disrupted autophagy is another hallmark of NPC shared by many other neurodegenerative disorders as neurons are particularly vulnerable to defects along the autophagic pathway ( Nixon , 2013 ) . TFEB was shown to up-regulate transcription of genes which support autophagosome formation and lysosomal biogenesis ( Settembre et al . , 2011 ) and could therefore be a considered a novel target to promote lysosomal clearance in lipid storage and neurodegenerative disorders . TFEB overexpression has already been shown to be beneficial in attenuating the storage phenotype in several storage disorders such as Batten disease ( Medina et al . , 2011 ) or Pompe disease ( Spampanato et al . , 2013 ) . Similar overexpression strategies showed promising results even in mouse models of more common neurodegenerative disorders , such as Parkinson disease ( Decressac et al . , 2013 ) , Alzheimer disease ( Polito et al . , 2014 ) or Huntington disease ( Tsunemi et al . , 2012 ) . The regulatory functions of Sph and TFEB are therefore key targets for future investigations of neurodegenerative disorders . Taken together , we describe intracellular Sph as a new small molecule effector of lysosomal calcium homeostasis , the lysosome as triggerable calcium store via the TPC1 channel , and specific transcription factor translocation as a consequence of lysosomal calcium release . Our results further highlight the importance of elucidating the interplay between lysosomal calcium signaling and Niemann-Pick type C disease . This interplay might also be important for the progression of diseases caused by Ebola or Marburg virus infection . Sph should be considered as a new and important player in this field and understanding the regulation of its signaling will likely further our understanding of these named diseases . Common laboratory chemicals were purchased from commercial sources ( Acros , Fluka , Merck , Sigma-Aldrich or VWR ) at highest available grade and were used without further purification . D-erythro sphingosine and D-erythro dihydrosphingosine , the sphingosine kinase inhibitor SKI-II , Xestospongin C and GPN were obtained from Biomol ( Hamburg , DE ) . 7-Diethylamino-4-hydroxymethylcoumarin was a kind gift from Rainer Müller ( EMBL Heidelberg , DE ) . Deuterated solvents for NMR analysis were purchased from Deutero ( Karlsruhe , DE ) . Sphingosine kinase inhibitor N-N-dimethylsphingosine was obtained from Sigma-Aldrich . NAADP-antagonist Ned-19 was purchased from Tocris Biosciences ( Bristol , UK ) . The fluorescent calcium indicator Fluo-4-AM and Alexa488-azide were obtained from Life Technologies ( Thermo Fisher Scientific , Waltham , USA ) . pEGFP-N1-TFEB was a gift from Shawn Ferguson ( Addgene plasmid # 38119 ) ( Roczniak-Ferguson et al . , 2012 ) . All chemical reactions were carried out using dry solvents under inert atmosphere . Thin layer chromatography ( TLC ) was performed on plates of silica gel ( Merck , 60 F254 ) and visualized using UV light ( 254 nm or 366 nm ) or a solution of phosphomolybdic acid in EtOH ( 10% w/v ) . HPLC grade solvents for chromatography were obtained from VMR . Preparative column chromatography was carried out using Merck silica gel 60 ( grain size 0 . 063–0 . 200 nm ) under a pressure of <1 . 5 bar . 1H-NMR spectroscopic measurements were conducted on a 400 MHz Bruker UltraShieldTM spectrometer at 25°C . 13C-NMR measurements were performed on a 500 MHz Bruker UltraShieldTM spectrometer at 25°C and were broadband hydrogen decoupled . Chemical shifts are given in ppm referenced to the residual solvent peak . J values are given in Hz and splitting patterns are designated using s for singlet , d for doublet , t for triplet , q for quartet , m for multiplet and b for broad signal . High-resolution mass spectra were recorded at the Organic Chemistry Institute of the University of Heidelberg A solution of 7-diethylamino-4-hydroxymethylcoumarin ( Schönleber et al . , 2002 ) ( 48 mg , 194 µmol ) in 2 mL dry THF was cooled to 0°C . DIPEA ( 0 . 1 µL , 575 µmol ) and phosgene ( 300 µL , 610 µmol ) were added dropwise and stirred in the dark for 2 h at 0°C . The reaction mixture was extracted with EtOAc/H2O ( 1:1 , 75 mL ) , the layers were separated , the organic layer was washed with brine and dried using Na2SO4 . The solvent was removed under reduced pressure and the product was dried further under high vacuum conditions . 7- ( Diethylamino ) -coumarin-4-yl]-methyl chloroformate was used without further purification . To a solution of D-erythro-sphingosine ( 30 mg , 100 µmol ) in 2 mL dry THF , TEA ( 70 µL , 500 µmol ) and a solution of 7- ( diethylamino ) -coumarin-4-yl ) -methyl chloroformate ( 46 mg , 148 µmol ) in 1 mL dry THF were added . The mixture was stirred at RT for 1 h in the dark . EtOAc ( 50 mL ) was added to stop the reaction and the mixture was washed twice with citric acid ( 5% w/v , 25 mL ) and twice with saturated NaHCO3 . The organic layer was dried with Na2SO4 and the solvent was removed under reduced pressure . The residue was purified by repeated flash chromatography ( first column: eluent: DCM/MeOH13:1 , second column: eluent: cyclohexane/EtOAc 1:5 ( +1% TEA ) ) . The compound 1 was obtained as yellow oil ( 24 mg , 42 µmol , 42% over two steps ) 1H NMR ( 400 MHz , CDCl3 ) δ = 7 . 29 ( d , J=8 . 9 , 1H ) , 6 . 58 ( d , J=8 . 7 , 1H ) , 6 . 50 ( s , 1H ) , 6 . 14 ( s , 1H ) , 5 . 96 ( d , J=8 . 3 , 1H ) , 5 . 80 ( dd , J=14 . 6 , 7 . 2 , 1H ) , 5 . 55 ( dd , J=15 . 2 , 6 . 1 , 1H ) , 5 . 22 ( s , 2H ) , 4 . 39 ( s , 1H ) , 4 . 05–3 . 96 ( m , 1H ) , 3 . 76 ( d , J=11 . 6 , 1H ) , 3 . 68 ( q , J=3 . 8 , 1H ) , 3 . 40 ( q , J=7 . 0 , 3H ) , 3 . 06 ( s , 1H ) , 2 . 05 ( dd , J=13 . 5 , 6 . 6 , 2H ) , 1 . 41–1 . 10 ( m , 28H ) , 0 . 87 ( t , J=6 . 6 , 3H ) . 13C NMR ( 126 MHz , CDCl3 ) δ = 162 . 15 , 156 . 18 , 155 . 60 , 150 . 40 , 134 . 45 , 128 . 70 , 126 . 78 , 124 . 43 , 108 . 98 , 106 . 16 , 98 . 13 , 74 . 75 , 74 . 17 , 66 . 85 , 62 . 21 , 61 . 89 , 55 . 73 , 44 . 96 , 32 . 29 , 31 . 93 , 29 . 70 , 29 . 67 , 29 . 64 , 29 . 51 , 29 . 37 , 29 . 22 , 29 . 10 , 22 . 70 , 14 . 12 , 12 . 40 . HRMS for C33H53N2O6+ calculated: 573 . 39036; found: 573 . 39027 . A solution of D-erythro-sphingosine ( 30 mg , 100 µmol ) in 2 mL dry THF and TEA ( 70 µL , 500 µmol ) was stirred and 4 , 5-Dimethoxy-2-nitrobenzyl-chloroformate ( 41 mg , 150 µmol ) in 2 mL dry THF was added dropwise . The mixture was stirred at RT for 1h in the dark . EtOAc ( 50 mL ) was added and the mixture was washed twice with citric acid ( 5% w/v , 25 mL ) and twice with saturated NaHCO3 . The organic layer was dried with Na2SO4 and the solvent was removed under reduced pressure . The residue was purified by flash chromatography ( eluent: cyclohexane/EtOAc 1:5 ( +1% TEA ) ) . Compound 2 was obtained as colorless oil ( 51 . 4 mg , 95 µmol , 95% ) 1H NMR ( 400 MHz , CDCl3 ) δ = 7 . 70 ( s , 1H ) , 7 . 02 ( s , 1H ) , 5 . 84–5 . 75 ( m , 2H ) , 5 . 55 ( d , J=6 . 0 , 1H ) , 5 . 51 ( s , 2H ) , 4 . 38 ( s , 1H ) , 4 . 01 ( s , 1H ) , 3 . 98 ( s , 3H ) , 3 . 95 ( s , 3H ) , 3 . 73 ( d , J=11 . 5 , 1H ) , 3 . 67 ( s , 1H ) , 2 . 39 ( s , 2H ) , 2 . 12–1 . 96 ( m , 2H ) , 1 . 42–1 . 13 ( m , 22H ) , 0 . 87 ( t , J=6 . 7 , 3H ) 13C NMR ( 101 MHz , CDCl3 ) δ = 155 . 99 , 134 . 47 , 128 . 58 , 110 . 07 , 108 . 15 , 74 . 81 , 63 . 78 , 62 . 24 , 56 . 52 , 56 . 50 , 56 . 40 , 55 . 57 , 32 . 28 , 31 . 94 , 31 . 92 , 29 . 71 , 29 . 68 , 29 . 62 , 29 . 48 , 29 . 36 , 29 . 22 , 29 . 08 , 22 . 71 , 22 . 69 , 14 . 13 . HRMS for C28H47N2O8Na+ calculated: 561 . 31519 , found: 561 . 31557 . To a solution of D-erythro-dihydrosphingosine ( 10 mg , 33 µmol ) in 1 mL dry THF , DIPEA ( 23 µL , 230 µmol ) and a solution of [7- ( diethylamino ) -coumarin-4-yl ) -methyl chloroformate ( 15 mg , 50 µmol ) in 0 , 5 mL dry THF were added . The mixture was stirred at RT for 1 . 5 h in the dark . EtOAc ( 20 mL ) was added to stop the reaction and the mixture was washed twice with citric acid ( 5% w/v , 10 mL ) and twice with saturated NaHCO3 . The organic layer was dried with Na2SO4 and the solvent was removed under reduced pressure . The residue was purified by repeated flash chromatography ( first column: eluent: DCM/MeOH 13:1 , second column: eluent: cyclohexane/EtOAc 1:5 ) . Compound 3 was obtained as yellow oil ( 14 . 5 mg , 25 µmol , 76% over two steps ) 1H NMR ( 400 MHz , CDCl3 ) δ = 7 . 27 ( d , J=8 . 9 Hz , 1H ) , 6 . 56 ( dd , J=9 . 0 , 2 . 3 , 1H ) , 6 . 49 ( d , J=2 . 3 , 1H ) , 6 . 14 ( s , 1H ) , 6 . 08 ( d , J=8 . 4 , 1H ) , 5 . 22 ( s , 2H ) , 4 . 07 ( dd , J=11 . 4 , 2 . 3 , 1H ) , 3 . 82 ( d , J=11 . 7 , 2H ) , 3 . 66 ( s , solvent THF ) , 3 . 62 ( dd , J=8 . 1 , 3 . 3 , 1H ) , 3 . 40 ( q , J=7 . 0 , 4H ) , 2 , 30 ( t , J=7 . 6 , solvent ) , 1 . 67–1 . 45 ( m , 4H ) , 1 . 36–1 . 13 ( m , 30H ) , 0 . 87 ( t , J=6 . 7 , 3H ) 13C NMR ( 126 MHz , CDCl3 ) δ = 162 . 34 , 156 . 17 , 155 . 52 , 150 . 60 , 130 . 02 , 129 . 77 , 124 . 41 , 108 . 83 , 106 . 12 , 105 . 90 , 97 . 90 , 74 . 36 , 62 . 24 , 61 . 83 , 55 . 13 , 51 . 44 , 44 . 82 , 34 . 55 , 31 . 94 , 29 . 71 , 29 . 67 , 29 . 61 , 29 . 59 , 29 . 37 , 27 . 23 , 25 . 99 , 24 . 96 , 22 . 70 , 14 . 12 , 12 . 43 . HRMS for C33H55N2O6+ calculated: 575 . 40601; found: 575 . 40626 . HeLa cells ( LGC-ATCC , No . CCL-2 , authenticated by STR profiling ) were grown in DMEM ( 1 g/L Glutamate , Gibco ) with 10% FBS ( Gibco ) and 1% Primocin ( Invivogen ) , regularly tested for mycoplasma ( using Lookout Mycoplasma PCR detection kit ( Sigma , MP0035 ) ) and only used when negative . Primary MEF from Tpcn1-/- , Tpcn2-/- and double knock out mice were characterized previously ( Ruas et al . , 2015 ) and only used for 3-4 passages . They were not screened for mycoplasma . Human patient fibroblasts from MLIV patients were obtained from the Coriell Institute ( MLIV patient cells: GM02525 , authenticated by Nucleoside Phosphorylase Isoenzyme Electrophoresis , control cells: GM05399 , authenticated by Chromosome Analysis ) and checked for mycoplasma using the Lookout Mycoplasma PCR detection kit ( Sigma , MP0035 ) . In case of contamination , Mycoplasma Removal Agent ( AbD Serotec , BUF035 ) was used . NPC1 cells were provided by Dr . Forbes D . Porter ( NIHCD , USA ) . They are primary cultures from an NPC1 patient enrolled in the NIH program . The cells were not screened for mycoplasma , and used for 3–4 passages only . The culture medium for all mentioned fibroblasts consisted of DMEM ( 4 . 5 g/L glutatmate , Gibco ) with 10% FBS ( Gibco ) , 1% penicillin-streptomycin and 1% L-glutamine ( Gibco ) . Cells were seeded in 8-well Lab-TekTM microscope dishes 48–72 h prior to the experiment . If needed , cells were transfected in OPTIMEM ( Gibco ) with maximal 200 ng DNA per well and Fugene HD ( Promega ) according to the manufacturer’s instructions . For the duration of the experiment , the medium of the cells was changed to imaging buffer ( 20 mM HEPES , 115 mM NaCl , 1 . 2 mM MgCl2 , 1 . 2 mM K2HPO4 , 1 . 8 mM CaCl2 and 10 mM glucose ) . Each Labtek-well was loaded with 100 µl of a 5 µM Fluo-4 AM ( Molecular Probes® ) solution in imaging buffer for 30 min at 37°C . The caged lipids were added to a final concentration of 2 µM 15 min prior to the experiment . Immediately prior to the experiment , the cells were washed and kept in imaging buffer . Cells were grown to ~85% confluency in 6 cm dishes and pulsed with indicated concentrations of Sph-Cou in imaging buffer for the indicated times . Under +UV conditions , the cells were irradiated by a UV mercury arc source ( Newport ) equipped with a 400 nm highpass filter at a distance of 35 cm for 2 min . Cells were washed with PBS and scraped off in 300 µl PBS . The lipids were extracted according to a protocol published by Thiele et al . ( Thiele et al . , 2012 ) . In brief , 600 µl methanol and 150 µl chloroform was added to the cells , vortexed and centrifuged at 14 000 rpm for 3 min . The supernatant was mixed with 300 µl chloroform and 600 µl acetic acid ( 0 . 1% in water ) . The aqueous phase was discarded and the organic phase was dried on a speedvac at 30°C for 15 min . The dried lipids were redissolved in 30 µl chloroform and applied on a 10 x 10 cm HPTLC Silica 60 glass plate ( VWR ) using the automatic Camaq system . TLC plates were developed using first CHCl3/MeOH/H2O/AcOH 65:25:4:1 for 6 cm and then cyclohexane/ethylacetate 1:1 for 9 cm . Fluorescently labeled lipids were visualized using a geldoc system . HeLa cells were grown in 6 cm dishes ( Nunc ) to 85–95% confluency and labelled with 2 µM Sph-Cou or 2 µM dhSph-Cou in imaging buffer or only in imaging buffer for 15 min . After washing with PBS , cells were transferred onto an ice block and UV-irradiated with > 400 nm light on a 450–1 , 000 W high-pressure mercury lamp ( Newport , #66924 , series #1166 ) for 2 min and then immediately pelleted ( by centrifugation with 3000 rpm for 1 min ) and snap-frozen in liquid nitrogen . Long chain base extraction and analysis was performed as follows . Cell pellets were resuspended in extraction solvent EtOH/H2O/diethylether/pyridine ( 15:15:5:1 ) and ammonium hydroxide ( 2 . 1 . 10−3 N ) . Cells were extracted by vortexing at 4°C for 10 min and incubation on ice for 20 min . Cell debris was pelleted by centrifugation at 20000 x g for 2 min at 4°C . The extraction was repeated one more time without the incubation on ice . Supernatants were combined and dried under vacuum in a Centrivap ( Labconco , Kansas City , USA ) . The extract was resuspended in borate buffer ( 200 mM boric acid pH8 . 8 , 10 mM tris ( 2-carboxyethyl ) -phosphine , 10 mM ascorbic acid and 33 . 7 µM 15N13C-valine ) and derivatized by reaction for 15 min at 55°C with 6-aminoquinolyl-N-hydroxysuccinimidyl carbamate ( AQC , 2 . 85 mg/ml in acetonitrile ) . The AQC reagent was synthesized according to Cohen et al ( Cohen and Michaud , 1993 ) . Samples were analyzed after overnight incubation at 24°C using a reverse-phase C18 column ( HPLC EC 100/2 Nucleoshell RP-18 2 . 7 µm ) on an Accela system high-performance liquid chromatography ( ThermoFisher Scientific , Waltham , MA ) , coupled to a TSQ Vantage ( ThermoFisher Scientific , Waltham , USA ) . MRM-MS was used to identify and quantify lipid species . The relative amounts of long chain bases were normalized to a standard for the derivatization process ( 15N13C-valine ) and internal standards added before extraction ( C17-sphingosine , C17-sphinganine , C17-sphingosine-1-phosphate , C17-sphinganine-1-phosphate ) . Fluorescence change of the calcium indicators were captured on a dual scanner confocal laser scanning microscope ( Olympus Fluoview 1200 ) with a 63x oil objective using excitation at 488 nm and emission settings between 500–550 nm at an interval of 1 s per frame . For monitoring the red calcium dye R-GECO , the excitation was at 559 nm and emission between 580–650 nm . A baseline of 10 s was captured before photoactivation . Intracellular photoactivation was performed using the tornado function of the Olympus software with a circular region of interest ( 10 pixel units diameter , 8 . 9 µm2 ) . For uncaging at 405 nm , we used 50% laser intensity for 3 s at 2 µs per pixel . Nitroveratroyl-uncaging at 375 nm was carried out using 100% laser intensity for 5 s at 2 µs per pixel . The acquired time lapse series were analyzed with the Fiji software ( W . Rasband , NIH , USA ) using the FluoQ macro ( Stein et al . , 2013 ) set to the following parameters: Background subtraction method: Mean of an interactively selected ROI Noise reduction / smoothing method: None Threshold method: Interactively with ImageJ’s built-in threshold window ROI segmentation: Semi-automatically with binary mask modification Calculate amplitude changes: Using maximum observed amplitude change The maximum amplitude values were calculated by subjecting the raw traces to a central moving average . The maximal amplitude xresponsemax of these smoothed traces was used to calculate the amplitude change in percent %△xmax according to following formula: %△xmax=xresponsemax−x¯baseline\|x¯baseline\|*100 The resulting intensity series/amplitude values represent mean values of whole cells and were further analyzed using R software ( Development Core Team , R , 2014 ) . Cells were seeded onto 11mm coverslips placed in wells of a 24-well plate and labeled with 4 µM pacSph in imaging buffer for 10 min . Cells were washed , overlaid with 1 mL imaging buffer and UV-irradiated on ice for 2 . 5 min using a 450–1 , 000 W high-pressure mercury lamp ( Newport , #66924 , series #1166 ) equipped with a glass filter to remove wavelengths below 345 nm ( Newport , #20CGA-345 ) , operated at 1 , 000 W . Cells were immediately fixed with MeOH at -20°C for 20 min . Not crosslinked lipids were extracted by washing 3x with 1 mL of CHCl3/MeOH/AcOH 10:55:0 . 75 ( v/v ) at RT . Cells were then incubated with 50 µl of click mixture ( 1 mM ascorbic acid , 100 μM TBTA , 1 mM CuSO4 and 2 μM Alexa 488 azide in PBS ) for 1 h at RT in the dark . Cells were then washed with PBS and incubated with 50 µl of primary antibody ( rabbit α-LAMP1 , Cell Signaling , 1:100 in PBS supplemented with 4% BSA and 0 . 02% Triton ) overnight at 4°C . Coverslips were again washed in PBS and incubated with secondary antibody ( α-rabbit conjugated to AlexaFluor555 , Cell Signaling , 1:800 ) for 1 h , washed and mounted in DAPI-containing mounting medium ( Vectashield , Vector Laboratories , Inc . Burlingame , CA 94010 , #H-1200 ) . Microscopy images were captured at RT using a confocal laser scanning microscope ( Zeiss LSM780 ) with a 63x oil objective . Settings were as follows: DAPI-channel ( 405 nm excitation ( ex ) , 409–475 nm emmission ( em ) ; green channel: 488 nm ex , 489–550 nm em; red channel: 561 nm ex , 569–655 nm em ) . Images were further processed using imageJ ( http://rsb . info . nih . gov/ij/ ) .	Sphingosine is a small fat molecule that has been suggested to act as a signal inside cells . Individuals with a rare neurodegenerative disease called Niemann-Pick disease type C accumulate sphingosine and other fat molecules in cell compartments called lysosomes . Intriguingly , this fat accumulation is accompanied by an altered movement of calcium ions in and out of lysosomes . In healthy cells , an increase in calcium ion levels can trigger a process called autophagy , in which proteins and other cell components are destroyed in a controlled manner . This is thought to be caused by the release of calcium ions from lysosomes , which stimulates a protein called TFEB to move into the nucleus of the cell to activate genes involved in autophagy . Two proteins on the surface of lysosomes called TPC1 and TPC2 are believed to act as channels that can release calcium ions from lysosomes . However , it was not clear how sphingosine could disrupt calcium ion movements in patients with Niemann-Pick disease type C . Here , Hoeglinger et al . have used a new approach to understand how calcium ions and sphingosine are linked in both healthy and diseased cells . The experiments use a form of sphingosine called “caged sphingosine” that is only activated when it is exposed to a flash of light , which makes it possible to increase the levels of this molecule in cells in a precise way . Hoeglinger et al . found that sphingosine triggered the release of calcium ions from lysosomes . This release required the TPC1 protein and resulted in TFEB moving into the cell nucleus . Further experiments confirm that sphingosine accumulates in the lysosomes of cells taken from patients with Niemann-Pick disease type C . In these cells , the activation of caged sphingosine resulted in a much smaller release of calcium ions from lysosomes than that observed in healthy cells . Together , Hoeglinger et al . ’s findings show that sphingosine acts as a signal to trigger the release of calcium ions from lysosomes , which in turn promotes autophagy . The next challenge is to find out exactly how sphingosine opens the calcium ion channels .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2015	Intracellular sphingosine releases calcium from lysosomes
Neoblasts are an abundant , heterogeneous population of adult stem cells ( ASCs ) that facilitate the maintenance of planarian tissues and organs , providing a powerful system to study ASC self-renewal and differentiation dynamics . It is unknown how the collective output of neoblasts transit through differentiation pathways to produce specific cell types . The planarian epidermis is a simple tissue that undergoes rapid turnover . We found that as epidermal progeny differentiate , they progress through multiple spatiotemporal transition states with distinct gene expression profiles . We also identified a conserved early growth response family transcription factor , egr-5 , that is essential for epidermal differentiation . Disruption of epidermal integrity by egr-5 RNAi triggers a global stress response that induces the proliferation of neoblasts and the concomitant expansion of not only epidermal , but also multiple progenitor cell populations . Our results further establish the planarian epidermis as a novel paradigm to uncover the molecular mechanisms regulating ASC specification in vivo . Adult stem cells ( ASCs ) are tissue-specific cells with the capacity to self-renew and differentiate to continually replace cells lost to normal physiological turnover or injury . As a result , ASCs play an essential role in preserving the anatomical form and function of most multicellular organisms . The precise coordination of stem cell proliferation and proper fate specification is of paramount importance to tissue growth and organismal homeostasis . Excessive stem cell divisions can lead to tumorigenesis ( Visvader and Lindeman , 2012 ) , while a loss in proliferation capacity can contribute to premature aging ( Gopinath and Rando , 2008 ) . Understanding the cellular and molecular mechanisms that regulate the balance between stem cell proliferation , differentiation , and cell death will thus provide fundamental insights into tissue maintenance and repair . It will also illuminate the molecular basis of tissue dysfunction , including disease progression and aging . The model planarian Schmidtea mediterranea has emerged as an experimental system that provides a unique window into major aspects of in vivo stem cell biology , including regeneration , fate determination and homeostatic plasticity ( Rink , 2013; Roberts-Galbraith and Newmark , 2015 ) . Neoblasts , the planarian stem cells , are in a state of perpetual action . They are widely distributed throughout the body mesenchyme , driving constitutive renewal of tissues during homeostasis and endowing planarians with the remarkable capacity to regenerate wholly from tiny tissue fragments ( Brøndsted , 1969; Newmark and Sánchez Alvarado , 2000; Wagner et al . , 2011 ) . Neoblasts , the only dividing cells in planarians , are believed to be collectively comprised of both a heterogeneous population of pluripotent cells with broad differentiation potential and also lineage-committed progenitor cells that give rise to specific tissues ( Hayashi et al . , 2010; Scimone et al . , 2014; van Wolfswinkel et al . , 2014; Wagner et al . , 2011 ) . To ensure the integrity of adult tissues during homeostasis and regeneration , neoblasts must perpetuate themselves and generate lineage-committed progenitor cells that give rise to precise numbers of differentiated cell types in a proper spatial and temporal sequence . A general principle used to establish planarian lineages has been to identify tissue-specific transcription factors ( TF ) expressed in subsets of neoblasts ( smedwi-1+ ) ( Reddien et al . , 2005b ) that are also required for the specification of those tissues , including the eye ( Lapan and Reddien , 2011; Lapan and Reddien , 2012 ) , protonephridia ( Scimone et al . , 2011 ) , pharynx ( Adler et al . , 2014; Scimone et al . , 2014 ) , and discrete neuronal sub-types ( Cowles et al . , 2013; Currie and Pearson , 2013; Marz et al . , 2013; Scimone et al . , 2014; Wenemoser et al . , 2012 ) . These TFs have typically been identified through evolutionary conservation , induced expression in neoblasts during regeneration , or through transcriptional profiling of isolated tissues . Additional methods including BrdU incorporation ( Newmark and Sánchez Alvarado , 2000 ) , perdurance of the SMEDWI-1 protein in differentiating progeny cells ( Guo et al . , 2006; Wenemoser and Reddien , 2010; Zhu et al . , 2015 ) , and gamma irradiation ( Eisenhoffer et al . , 2008 ) have also been used to link neoblasts with their progeny . Although RNAi knockdown of lineage-committed TFs blocks the regeneration of their specified tissues , many of these TFs are expressed throughout those tissues , making it difficult to study how different cell types within the same tissue are formed . Therefore , elucidating the cellular mechanisms that bridge the pluripotent and the differentiated state remains a challenge . The planarian epidermis is a simple , monostratified tissue that consists of clear histological organization of multiple differentiated multi-ciliated and non-ciliated cell types ( Rompolas et al . , 2010 ) . Individual epidermal cells must continuously be replaced due to damage from environmental insults and exogenous wounds . To replenish these cells , neoblasts residing deep in the mesenchyme must produce cells that mobilize , undergo multiple determination steps , cross the basement membrane , intercalate and differentiate into the epidermis . Moreover , the epidermis provides the critical first step in regeneration by covering the amputation-induced wound site through cell spreading ( Morita and Best , 1974 ) . Recent work has identified a prominent class of neoblasts referred to as zeta-class possessing a distinct molecular signature , including the novel zinc-finger gene zfp-1 ( van Wolfswinkel et al . , 2014 ) . Zfp-1 ( RNAi ) animals can regenerate tissues including the gut , brain and muscle , but fail to generate cells expressing epidermal markers , indicating that zeta-class neoblasts likely give rise to an epidermal lineage . Zfp-1 , along with the chromatin remodeling factor chd4 , the tumor suppressor gene p53 , and most recently an RNA-binding protein mex3-1 , have all been shown to be required for the maintenance of two related postmitotic , sub-epidermal cell populations expressing the specific marker genes prog-1 and AGAT-1 ( Pearson and Sánchez Alvarado , 2010; Scimone et al . , 2010; Wagner et al . , 2012; Zhu et al . , 2015 ) . These abundant prog-1+ and AGAT-1+ cell populations , originally identified as early and late progeny cells based on their rapid turnover kinetics ( Eisenhoffer et al . , 2008 ) , have been widely used to assess neoblast differentiation . Given that zeta-class neoblasts are required for the generation of prog-1+ , AGAT-1+ cells and other markers of epidermal cell types , prog-1 and AGAT-1 likely mark two major populations of epidermal progeny cells . However , it remains unclear whether the diverse cell types in the planarian epidermis all share common or distinct lineage relationships with each other , and the mechanisms that control the progression of epidermal progenitors along distinct differentiation paths into mature cell types are completely unknown . To understand the molecular mechanisms underlying neoblast differentiation and how they give rise to multiple cell types , we devised a strategy to identify critical factors enriched in the AGAT-1+ cell population required for epidermal lineage progression . We performed RNA-seq analysis of chd4 and p53 RNAi animals and characterized additional markers enriched in AGAT-1+ and related post-mitotic cells . We find that epidermal progeny cells form distinct mesenchymal populations and undergo differentiation in a spatially and temporally graded manner into the mature epidermis . We also describe a conserved transcription factor of the early growth response family , egr-5 , that is expressed in post-mitotic progeny cells and is an essential regulator of post-mitotic epidermal fate specification . Taken together , our results further establish the planarian epidermis as a paradigm to study adult lineage specification in vivo , contributing to our knowledge of mechanisms required for the proper execution of stem cell fate decisions . The prog-1+ ( early progeny ) and AGAT-1+ ( late progeny ) postmitotic cell populations constituting the first neoblast lineage described in planarians ( Eisenhoffer et al . , 2008 ) have been widely used as an assay for neoblast differentiation ( Fraguas et al . , 2011; Pearson and Sánchez Alvarado , 2010; Scimone et al . , 2010; Wagner et al . , 2012 ) . The zeta-class neoblasts are a major subclass of planarian neoblasts identified molecularly by a specific gene signature , mainly two TFs , zfp-1 and soxP-3 ( van Wolfswinkel et al . , 2014 ) . Zeta neoblasts are generated from the collectively pluripotent sigma-class neoblasts , and have recently been shown to generate prog-1+ and AGAT-1+ cells as well as other populations spanning the epidermis ( van Wolfswinkel et al . , 2014 ) , suggesting that they are part of an epidermal lineage ( Figure 1A ) . This lineage is a useful model to dissect the choreography of neoblast progeny differentiation because it is an abundant cell population undergoing rapid turnover , and both prog-1+ and AGAT-1+ cells are molecularly and spatially distinct . Prog-1+ ( early progeny ) cells likely represent a very transient cell population because they are lost about two days after a lethal dose of irradiation , whereas AGAT-1+ ( late progeny ) cells are not completely lost until about seven days post-irradiation ( Eisenhoffer et al . , 2008 ) . Therefore , the precise molecular relationships between these cell populations and the mechanisms that control the progression and maturation of the zeta-class epidermal lineage still must be resolved . 10 . 7554/eLife . 10501 . 003Figure 1 . Identification of a common transcriptional down-regulated gene set in chd4 and p53 RNAi animals . ( A ) Current model of planarian epidermal lineage specification . Sigma-class neoblasts give rise to zeta-class neoblasts , which in turn generate prog-1 ( early progeny ) and AGAT-1 ( late progeny ) expressing cells . The precise molecular relationship between these cell types remains unclear , but collectively give rise to an unknown number of epidermal cell types through an unknown number of transitional states . ( B ) Whole-mount in situ ( WISH ) expression patterns of chd4 and p53 in wild-type planaria . Left panels: chd4 colorimetric WISH and double fluorescent in situ ( FISH ) of chd4 ( red ) , AGAT-1 ( green ) and DAPI ( blue ) . Right panels: p53 colorimetric WISH and double FISH of p53 ( red ) , AGAT-1 ( green ) and DAPI ( blue ) . Magnified regions are single confocal planes from boxed regions . White arrowheads highlight cells with co-localized expression of either chd4 or p53 and AGAT-1 . Yellow arrowheads highlight additional mesenchymal cells that do not express AGAT-1 . Scale bars: 200 μm; 10 μm ( zoomed images ) . ( C ) chd4 ( RNAi ) and p53 ( RNAi ) result in the loss of AGAT-1 expressing cells . Representative colorimetric WISH images shown at 3Fd18 of RNAi treatment . Scale bar: 200 μm . ( D ) Venn diagram of genes down-regulated in chd4 and p53 RNAi data sets . Timeline of RNAi treatment ( Fed d0 , d3 , d6 ) and RNA collected for chd4 ( yellow circles ) and p53 RNAi ( pink circles ) . Arrows highlight time points used to identify down-regulated gene set . Criteria used for genes to make the cut-off are shown . dn , down-regulated genes; p . adj , adjusted p-value; log2 , fold change of RNAi over control ( see Materials and methods ) . For the chd4 and p53 RNAi overlapping data set ( 587 genes ) , a hypergeometric distribution and a universe size of 28 , 668 was used to generate p-value for determining significance of overlap by chance . See also Supplementary file 1 . ( E ) Heat map depicting candidate genes that were selected from the chd4 ( RNAi ) and p53 ( RNAi ) down-regulated data sets for further characterization after in situ hybridization screen . log2 fold changes in RNAi expression relative to each control time point are shown . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 00310 . 7554/eLife . 10501 . 004Figure 1—figure supplement 1 . RNA-seq validation of control genes in chd4 and p53 RNAi animals . ( A ) Colorimetric WISH was performed on control , chd4 and p53 RNAi animals for smedwi-1 ( stem cell marker ) , prog-1 ( early progeny ) and AGAT-1 ( late progeny ) from the same cohort of worms used for RNA-seq analysis . Representative images of 3Fd15 RNAi worms are shown . Scale bar: 200 μm . ( B ) Plot of log2 ratios of control genes ( smedwi-1 , prog-1 , AGAT-1 ) and targeted RNAi genes ( chd4 and p53 ) in chd4 ( RNAi ) ( top panel ) and p53 ( RNAi ) ( bottom panel ) time course . For each time point , reads mapping to gene models were counted in chd4 ( RNAi ) and p53 ( RNAi ) data sets and were then compared to the unc22 ( RNAi ) control , and log2 ratios were generated . ( C-D ) Volcano plots illustrating genes significantly down-regulated ( green dots; adjusted p-value <1e-13 , log2 <-1 ) and up-regulated ( red dots; adjusted p-value <1e-13 , log2 >1 ) at 3Fd15 in chd4 ( RNAi ) data set ( C ) and at d18 in p53 ( RNAi ) data set ( D ) . Control genes are highlighted in yellow . Blue dots represent genes from the 587 gene overlap data set ( Figure 1D ) that are also significantly down-regulated in the plotted time points . See also Supplementary file 1 . See Supplementary file 2 for up-regulated gene list . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 004 To further characterize the molecular transitions implicated in epidermal cell maturation , we compared whole worm expression profiles from two different RNAi knockdown conditions known to drastically reduce the population of AGAT-1+ cells . Previous studies have shown that the chromatin-remodeling factor chd4 and the tumor suppressor gene p53 , both expressed in the neoblast compartment , are required for tissue homeostasis and regeneration ( Pearson and Sánchez Alvarado , 2010; Scimone et al . , 2010 ) . Both chd4 and p53 are also expressed in AGAT-1+ cells . Chd4 is expressed broadly throughout the animal parenchyma as well as in the brain and ventral nerve cords , and p53 is expressed in discrete mesenchymal cells that also include smedwi-1+ and prog-1+ cells ( Figure 1B ) . RNAi knockdown of both chd4 and p53 results in a dramatic loss of AGAT-1+ cells ( Figure 1C ) . However , it is not well understood whether these mechanisms of action are direct or indirect . Based on their close proximity to the basement membrane and on post-irradiation kinetics , AGAT-1+ cells likely represent a relatively stable transition at which multiple fate decisions could be made . Therefore , we reasoned that the union of chd4 and p53 RNAi whole-worm RNA-seq down-regulated datasets would contain genes enriched in AGAT-1+ cells , as well as genes with enriched expression in epidermal cell populations that arise from AGAT-1+ progeny cells , potentially revealing critical factors required for post-mitotic differentiation . We compared whole worm gene expression profiles of chd4 and p53 RNAi animals to those of control animals through multiple time points of RNAi treatment ( Figure 1D ) . To verify the specificity and sensitivity of the data sets , we examined a number of control genes ( smedwi-1 , prog-1 , AGAT-1 and chd4 , p53 ) with known kinetics of disappearance after RNAi knockdown and found strong correlation ( Figure 1—figure supplement 1A , B ) , indicating that known marker genes display predicted behaviors in our RNA-seq data . To identify common down-regulated genes in chd4 and p53 RNAi-treated animals , we adopted criteria where such candidate genes would be required to have significant adjusted p-values ( p . adj <1e-13 ) weighted towards multiple later time points after RNAi treatment ( for chd4: d12 , d15; for p53: d12 , d15 , d18 ) . Using these parameters , a total of 1 , 250 genes were designated to be down-regulated after chd4 RNAi treatment , 1 , 174 genes were down-regulated after p53 RNAi treatment , and a total of 587 common genes were found to be significantly down-regulated in both data sets ( Figure 1D—figure supplement 1C , D ) . Of the 587 common down-regulated genes , 70% ( 411/587 ) are predicted to encode homologs of proteins found in other organisms ( Supplementary file 1 , 2 for up-regulated genes ) . Earlier studies have reported genes involved in creatine metabolism ( AGAT-1 , AGAT-2 , AGAT-3 ) , polyamine biosynthesis ( odc-1 ) , and monooxygenase activities ( cyp1a1 ) , as well as others with unknown function , to be expressed in similar cell populations ( Eisenhoffer et al . , 2008; van Wolfswinkel et al . , 2014; Zhu et al . , 2015 ) . We analyzed the common gene set by assigning gene ontology ( GO ) terms and found that small molecule metabolic processes and transporter activity were the most overrepresented biological processes and molecular functions ( Supplementary file 1 ) . We selected approximately 150 genes from the common down-regulated data set to screen by whole-mount in situ hybridization ( WISH ) ( Supplementary file 1 ) . We also included some genes that were uniquely down-regulated in the p53 RNAi data set ( d18 ) because chd4 ( RNAi ) animals that were to be collected for expression analysis died prematurely ( before d18 ) . A wide net was cast for candidate gene cloning , including genes with homologs predicted to be involved in a diverse array of biological processes and those with no known homologs , together spanning a wide range of expression levels . We performed an in situ expression screen to further narrow down our common gene list to those that are enriched in an AGAT-1-like mesenchymal or a similar epidermal pattern . The majority of genes tested displayed distinctive WISH expression patterns ( Supplementary file 3 ) . We subsequently narrowed our candidate gene list to 29 unique genes based on a combination of their representative expression pattern , relatively strong signal intensity , and predicted gene function , for further analysis ( Figure 1E ) . Based on colorimetric WISH , we classified the gene expression patterns of the 29 candidate genes into four main categories: discrete , AGAT-1-like sub-epidermal mesenchymal expression throughout the animal ( AGAT-1 mesenchymal ) ; mesenchymal cells that appeared more dense than AGAT-1 and/or discrete expression in the epidermis ( AGAT-1 mesenchymal and epidermal ) ; gut-enriched; and expression spanning multiple tissues exhibiting discrete mesenchymal/epidermal cells and in the gut/pharynx ( Figure 2A ) . A wide assortment of biological processes were represented by the 29 candidate genes , including creatine metabolism ( AGAT-1 , gatm ) , monooxygenase activity ( cyp3140A1 , cyp3142A1 , cyp3143A1 , cyp3G1 , cyp3141A1 ) , fatty-acid metabolism ( pla2 , acsl-1 , acsl-2 ) , other metabolic processes ( odc-1 , mpv17 , adss ) , protein-binding ( traf-4 ) , transporter activity ( ttpal , slc25a-19 ) , cytoskeletal ( vim-3 ) , DNA-binding/zinc-finger ( egr-5 , litaf , nkx-2 . 2 , zfp-2 , zfp-3 ) , and multiple novel genes with unknown function . These novel genes with no known homology , predicted to encode small proteins ( ∼100 amino acids ) , contain a conserved signal peptide sequence . Because their expression patterns are down-regulated in zfp-1 RNAi animals ( see below ) , we have named them zeta-class protein of unknown function ( zpuf ) , followed by a unique designation number . 10 . 7554/eLife . 10501 . 005Figure 2 . Expression patterns of candidate genes from chd4 ( RNAi ) and p53 ( RNAi ) data sets . ( A ) WISH expression of representative candidate gene set from Figure 1E and grouped by expression pattern ( see text for details ) . Scale bars: 200 μm . See also Supplementary file 3 . ( B ) Expression pattern of various candidate genes in ( A ) , analyzed by double FISH with AGAT-1 . Images represent single confocal planes from anterior regions . Percentages represent fraction of AGAT-1+ cells that co-express the candidate gene ( ∼200-400 cells were quantified; low but detectable expression was counted as co-localized ) . White arrowheads highlight co-localization; yellow arrowheads highlight additional cells that have no detectable AGAT-1 expression ( AGAT-1neg ) . Scale bar: 10 μm . ( C ) Double FISH of zpuf-6 and AGAT-1 highlighting zpuf-6 expression in discrete cells in the planarian epidermis ( Top panel , white arrowheads ) . Bottom panel , ventral epidermal ( Ep ) view . DapI used for nuclear staining . Scale bar: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 00510 . 7554/eLife . 10501 . 006Figure 2—figure supplement 1 . Validation of select candidate genes in chd4 , p53 and zfp-1 RNAi conditions by WISH . ( A ) Representative WISH of candidate genes encompassing the full range of expression patterns in control , chd4 and p53 RNAi animals fixed at 3Fd21 ( For nkx-2 . 2 , 3Fd18 ) . Genes expressed in mesenchyme , epidermis and gut are reduced in chd4 and p53 RNAi animals . Genes expressed in mesenchyme and epidermis are also reduced in zfp-1 ( RNAi ) animals ( 3Fd21 ) but gut expression is not affected . Scale bar: 200 μm . ( B ) Plot of log2 ratios of candidate genes from ( A ) in chd4 ( RNAi ) ( top panel ) and p53 ( RNAi ) ( bottom panel ) RNA-seq time course . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 00610 . 7554/eLife . 10501 . 007Figure 2—figure supplement 2 . Detailed characterization of zpuf-6 expression pattern by whole-mount FISH . Zpuf-6 is not expressed in muscle cells or in prog-1 cells . Top left panel: Double FISH of zpuf-6 and a muscle marker ( collagen ) . Top right panels: Triple FISH of zpuf-6 , AGAT-1 and prog-1 . Bottom panel: transverse tissue section of triple FISH highlighting the spatial distribution of zpuf-6 , AGAT-1 and prog-1 . zpuf-6 is expressed in both sup-epidermal and epidermal layers . All images shown are single confocal planes . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 00710 . 7554/eLife . 10501 . 008Figure 2—figure supplement 3 . Expression of pla2 and odc-1 in mesenchyme and epidermis . ( A ) Triple FISH of pla2 , zpuf-6 and AGAT-1 in the mesenchyme . Top row: single channel of each individual gene . Middle row: Left panel , double FISH of pla2 and zpuf-6 . Middle panel , double FISH of pla2 and AGAT-1 . Right panel , merge . Yellow arrowheads highlight cells that co-express pla2 and zpuf-6 and are AGAT-1neg . Bottom row: Double FISH of pla2 and zpuf-6 in the epidermis ( Ep ) . White arrowheads highlight co-localization . Scale bars: 50 μm . ( B ) Triple FISH of odc-1 , zpuf-6 and AGAT-1 in the mesenchyme . Top row: single channel of each individual gene . Middle row: Left panel , double FISH of odc-1 and zpuf-6 . Middle panel , double FISH of odc-1 and AGAT-1 . Right panel , merge . Yellow arrowheads highlight cells that co-express odc-1 and zpuf-6 and are AGAT-1neg . Bottom row: Double FISH of odc-1 and zpuf-6 in the epidermis . White arrowheads highlight co-localization . Scale bars: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 008 We were surprised that many candidate genes from our screen were expressed prominently in the gut , as neither chd4 nor p53 have been previously reported to affect intestinal gene expression . For example , nkx-2 . 2 is a homeodomain TF that is enriched in the gut and has previously been shown to be required for proper intestinal regeneration ( Forsthoefel et al . , 2012 ) . To verify the RNA-seq data by WISH , we selected a representative gene from each of the major expression categories and monitored their expression in chd4 , p53 and zfp-1 RNAi backgrounds ( Figure 2—figure supplement 1 ) . The gene expression patterns for all five genes tested ( zpuf-3 , zpuf-6 , vim-3 , nkx-2 . 2 , acsl-2 ) were significantly reduced in chd4 and p53 RNAi animals , whereas discrete mesenchyme and epidermal expression , but not gut expression , were reduced in zfp-1 RNAi animals ( Figure 2—figure supplement 1A ) . We conclude that transcript levels in the RNA-seq data set are highly predictive of gene expression in vivo . We next performed whole-mount fluorescent in situ hybridization ( FISH ) to determine whether the candidate genes exhibiting discrete mesenchymal cell expression patterns overlapped with AGAT-1 expression ( Figure 2B ) . The majority of genes displayed substantial overlap with AGAT-1 , but with varying degrees of signal intensity ( Supplementary file 3 ) . Several genes ( zpuf-6 , cyp3140A1 , ttpal and ascl-2 ) , despite showing substantial overlap with AGAT-1 ( >98% ) , were also expressed in additional mesenchymal cells ( Figure 2B , yellow arrowheads ) . vim-3 was expressed in fewer sub-epidermal cells and showed very little overlap with AGAT-1 . After confirming that candidate genes with discrete mesenchymal cell patterns exhibit overlapping expression with AGAT-1 by whole-mount FISH , we next focused our attention on genes that are expressed in additional AGAT-1-negative ( AGAT-1neg ) mesenchymal cells to determine any potential molecular relationships . We first characterized zpuf-6 ( also known as NB . 36 . 10A ) because it has a robust expression pattern , is expressed in AGAT-1+ mesenchymal cells , and is also expressed in discrete epidermal cells ( Figure 2C ) . We determined that the zpuf-6+ AGAT-1neg mesenchymal cells do not overlap with collagen or prog-1 ( Figure 2—figure supplement 2 ) , suggesting that they are neither muscle cells nor early progeny cells . Rather , these zpuf-6+ AGAT-1neg mesenchymal cells likely mark a unique population linked to epidermal cells . We conducted triple whole-mount FISH for all other candidate genes with zpuf-6 and AGAT-1 to resolve whether they also overlap with zpuf-6+ AGAT-1neg mesenchymal cells or represent some other heterogeneous cell type . Interestingly , all genes tested displayed significant or partial overlap with zpuf-6 in the mesenchyme and epidermis ( Supplementary file 3 ) . For example , pla2 and odc-1 both showed significant overlap with zpuf-6+ AGAT-1neg mesenchymal cells and are very weakly expressed in zpuf-6+ epidermal cells ( Figure 2—figure supplement 3 ) . Furthermore , mpv17 expression overlaps with zpuf-6+ AGAT-1neg mesenchymal cells but is not expressed in the epidermis ( Supplementary file 3 , data not shown ) . Acsl-2 , adss and vim-3 were the only genes found to be expressed in additional epidermal cell types that did not overlap with zpuf-6 . Adss , which encodes the enzyme adenylosuccinate synthase and is involved in purine biosynthesis , was the one exception that appeared to be expressed in additional AGAT-1neg zpuf-6neg mesenchymal cells . Taken together , our in situ expression analysis revealed that the candidate genes identified by RNA-seq are all expressed in similar , overlapping mesenchymal and epidermal cells marked by AGAT-1 and zpuf-6 , and that these cell populations may all be constituents of the same lineage . The spatiotemporal disappearance of various morphological markers in vivo can be used to identify cell populations that belong to the same lineage . Irradiation is a powerful tool that has been demonstrated to rapidly and specifically eliminate neoblasts , and subsequently , their immediate progeny cells ( Eisenhoffer et al . , 2008 ) . The prog-1+ and AGAT-1+ cell populations were identified based on their spatial expression domains and down-regulation kinetics after irradiation . That is , the temporal order of prog-1 ( early ) and AGAT-1 ( late ) down-regulation correlates with the spatial distribution of these markers: the more peripheral the location of the cell , the longer the marker persists after irradiation . We were intrigued by a potential relationship between zpuf-6 and vim-3 , also identified in our screen ( Figure 2 ) , because it is expressed in fewer sub-epidermal mesenchymal cells that do not overlap with AGAT-1 , but more epidermal cells than zpuf-6 . Therefore , both zpuf-6 and vim-3 from our candidate gene set expand the spatial distribution of known progeny markers into the epidermis ( Figure 3A ) . The distinct expression domains of prog-1 , AGAT-1 , zpuf-6 , and vim-3 suggest that these progeny cells undergo outward migration during the course of epidermal differentiation . 10 . 7554/eLife . 10501 . 009Figure 3 . Epidermal progeny markers are expressed in distinct spatiotemporal domains . ( A ) Transverse tissue sections of colorimetric WISH-stained animals expressing putative markers in the epidermal lineage . Sections were counterstained with nuclear fast red . Scale bar: 10 μm . ( B ) Colorimetric WISH of animals after 6000 Rad irradiation exposure for markers of neoblasts ( smedwi-1 ) , early progeny ( prog-1 ) , late progeny ( AGAT-1 ) , and other genes identified in this study ( zpuf-6 and vim-3 ) marking transitions into the epidermis . Rootletin marks differentiated cells of the ciliated epidermis as well as protonephridia . Expression patterns of epidermal progeny markers are lost in a ventral-to-dorsal , anterior-to-posterior manner . A representative image from 4–6 worms for each gene and time point are shown . dpi , days post-irradiation . Dorsal views . See Figure 3—figure supplement 1 for close-up ventral views . Scale bars: 200 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 00910 . 7554/eLife . 10501 . 010Figure 3—figure supplement 1 . Ventral plane sections of epidermal progeny markers after irradiation . Colorimetric WISH of animals after 6 , 000 Rad irradiation exposure for markers of AGAT-1 , zpuf-6 and vim-3 , highlighting loss of ventral expression prior to loss of dorsal expression . A representative image from 4-6 worms for each gene and time point are shown . dpi , days post-irradiation . Scale bars: 200 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 01010 . 7554/eLife . 10501 . 011Figure 3—figure supplement 2 . Correlated spatial expression patterns of prog-1 , AGAT-1 and zpuf-6 . ( A ) Expression of AGAT-1 and zpuf-6 after chd4 and p53 RNAi knockdowns . Colorimetric WISH staining is lost in a ventral-to-dorsal , anterior-to-posterior fashion . Expression of markers in p53 ( RNAi ) animals display a striped pattern of cells ( zoom ) . Ventral views . Scale bars: 200 μm . ( B ) Triple FISH of prog-1 , AGAT-1 and zpuf-6 at 3Fd12 in control , chd4 and p53 RNAi knockdown conditions . The spatial patterns of remaining AGAT-1 and zpuf-6 cells are highly correlated . Images shown are single confocal planes of the ventral anterior region . Scale bar: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 011 We therefore extended this paradigm , that more differentiated cells have slower turnover kinetics , to determine the spatiotemporal irradiation kinetics of these newly characterized markers . After irradiation , both zpuf-6 and vim-3 displayed slower down-regulation kinetics compared to AGAT-1 and exhibited expression loss in a similar ventral-to-dorsal and anterior-to-posterior manner ( Figure 3B , Figure 3—figure supplement 1 ) . We also monitored the expression of rootletin , a gene expressed in the ciliated epidermis and tubule cells of the protonephridia ( Glazer et al . , 2010; Scimone et al . , 2011 ) , as a marker for differentiated cells and confirmed that its expression is not significantly affected after irradiation ( Figure 3B , bottom panel ) . Therefore , the irradiation kinetics data support our interpretation that zpuf-6 and vim-3 mark post-mitotic cells further downstream of the same prog-1 and AGAT-1 epidermal lineage . In other words , AGAT-1+ cells likely give rise to zpuf-6+ and vim-3+ expressing cells . We also looked for a conservation in AGAT-1 and zpuf-6 spatiotemporal expression kinetics in additional contexts . In both chd4 and p53 RNAi backgrounds , AGAT-1+ and zpuf-6+ cells are progressively lost in a ventral-to-dorsal , anterior-to-posterior fashion , with zpuf-6+ cells also undergoing a more delayed loss compared to AGAT-1 ( Figure 3—figure supplement 2A ) . Interestingly , in p53 ( RNAi ) animals , there is a characteristic banding pattern of AGAT-1+ and zpuf-6+ cells on the ventral surface of the animals ( Figure 3—figure supplement 2A bottom panels , 2B ) as cells are progressively lost over the course of RNAi treatment . This banding pattern of cells is also present for vim-3 on the ventral epidermis in p53 ( RNAi ) animals at later time points ( not shown ) , further suggesting that AGAT-1 , zpuf-6 , and vim-3 are markers exhibiting strong spatial correlative patterns . Lineage tracing experiments using the thymidine analog bromodeoxyuridine ( BrdU ) , in combination with whole-mount FISH , have revealed spatial and temporal regulation of neoblasts and the distribution of their differentiating progeny cells ( Eisenhoffer et al . , 2008; Newmark and Sánchez Alvarado , 2000 ) . A single-pulse of BrdU has also revealed the turnover dynamics of prog-1+ and AGAT-1+ and epidermal cells , revealing that prog-1 and AGAT-1 become incorporated markedly earlier than epidermal cells , consistent with the notion that they mark an intermediate stage of epidermal differentiation ( van Wolfswinkel et al . , 2014 ) . We performed BrdU pulse-chase analysis to examine the temporal kinetics of BrdU-labeling from neoblasts to prog-1+ , AGAT-1+ and zpuf-6+ cells during normal tissue turnover . The percentage of prog-1+ BrdU+ cells reached maximum levels around 10 days post-BrdU , followed by AGAT-1+ BrdU+ cells peaking at 14 days post-BrdU , and zpuf-6+ BrdU+ cells reached maximum levels at 22 days post-BrdU ( Figure 4A ) . Confidence tests of the difference between individual data points show that through days 1-–14 , the difference between AGAT-1+ BrdU+ and zpuf-6+ BrdU+ are only marginally significantly different ( most significant data point p = 0 . 04 for d6 ) . However , at 22 days post-BrdU the difference between the points of these same curves is highly significant ( p <0 . 01 ) . These results lend further support to our hypothesis that prog-1 , AGAT-1 , and zpuf-6 are distinct markers representing progressive stages of epidermal progeny differentiation . Because zpuf-6 labels the first transition state into the planarian epidermis , we wondered whether zpuf-6+ epidermal cells embody a specific differentiated epidermal cell type or whether they still possess the potential to differentiate further . To address this question , we built upon the spatial and molecular relationships between AGAT-1 , zpuf-6 and vim-3 defined by our previous irradiation kinetic studies by performing combinatorial whole-mount FISH . Triple whole-mount FISH of AGAT-1 , zpuf-6 and vim-3 revealed that vim-3 overlaps with AGAT-1negzpuf-6+ mesenchymal cells , zpuf-6+ epidermal cells , and there are additional zpuf-6negvim-3+ epidermal cells present on the dorsal side of animals ( Figure 4B ) . vim-3 , predicted to encode an intermediate-filament like protein , also co-localizes with vim-1 ( Figure 4—figure supplement 1A ) , which was shown to be down-regulated in zfp-1 RNAi animals ( van Wolfswinkel et al . , 2014 ) . Intermediate filaments comprise a diverse class of molecules and are expressed in a variety of cell types , including epithelial cells . These filaments generally provide a scaffold to integrate components of the cytoskeleton and organize the internal cell structure ( Snider and Omary , 2014 ) . Thus , we wondered whether the expression overlap of zpuf-6 and vim-3 could reflect cells undergoing a morphological transition within the epidermis . 10 . 7554/eLife . 10501 . 012Figure 4 . Zpuf-6+ epidermal cells are not terminally differentiated . ( A ) Turnover dynamics for prog-1 , AGAT-1 and zpuf-6 expressing cell populations . Animals were soaked with BrdU for 24 hr and chased for the indicated time periods . Quantification of the percentage of prog-1+ , AGAT-1+ or zpuf-6+ cells analyzed that are BrdU+ are plotted . Error bars: SEM ( see Materials and methods ) . Representative images of BrdU+ cells , are maximum intensity projections over 1 cell diameter of a subset of the quantified images , of the minimum and maximum time points for each gene are shown . White arrowheads highlight double-positive cells . Scale bar: 10 μm . ( B ) Lineage relationship between AGAT-1 , zpuf-6 and vim-3 . Top row: Triple FISH showing overlapping expression between AGAT-1 , zpuf-6 and vim-3 in the mesenchyme . Yellow arrowheads highlight cells that are zpuf-6+ vim3+ AGAT-1neg . AGAT-1 and vim-3 exhibit little to no co-expression ( Figure 2B ) . Bottom row: Dorsal epidermal ( Ep ) view of zpuf-6 and vim-3 . White arrowheads highlight cells that co-express zpuf-6 and vim-3 but there are additional cells expressing vim-3 . Images are single confocal planes . Scale bar: 50 μm . Quantifications of the percent colocalizations of combinations of AGAT-1 , zpuf-6 and vim-3 mesenchymal ( magenta ) and zpuf-6 and vim-3 dorsal epidermal cells ( cyan ) are shown ( ∼200-400 cells were quantified over >3 animals; error bars: SD ) . Gene1/gene2 notation signifies percentage of gene1+ cells that are also positive for gene2 expression . Notable percentages: zpuf-6+/AGAT-1+ ( 72% ) , zpuf-6+/vim-3+ ( 26% ) . ( C ) zpuf-6+ epidermal cells express very low levels of rootletin . Top row: double FISH of rootletin and zpuf-6 . 5 . 8%=percentage of strong zpuf-6+ epidermal cells that co-express rootletin ( ~300 cells ) . Scale bars: 50 μm . Bottom row: triple FISH of zpuf-6 , vim-3 and rootletin in dorsal epidermis . Only overlay of vim-3 and rootletin is shown ( far right panel ) . Yellow-dashed shapes outline cells that express very low levels of vim-3 that also express rootletin . 87 . 5%=percentage of weak vim-3+ dorsal epidermal cells that co-express rootletin ( ~200 cells ) . Images are single confocal planes . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 01210 . 7554/eLife . 10501 . 013Figure 4—figure supplement 1 . Vim-3 and vim-1 are co-expressed in the same cell types . ( A ) Images represent single confocal planes from ventral anterior regions of the mesenchyme ( top panels ) and epidermis ( Ep ) ( bottom panels ) . Percentages represent fraction of vim-3+ cells that co-express vim-1 ( ∼200 mesenchymal cells and 400 epidermal cells were quantified ) . Scale bars: 50 μm . ( B ) Vim-1/vim-3 is expressed in NB . 22 . 1E+ and laminB+ cells at the animal body margin ( yellow arrowheads ) . Images represent single confocal planes . Scale bar: 50 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 01310 . 7554/eLife . 10501 . 014Figure 4—figure supplement 2 . zpuf-6+ cells are enriched at regenerating blastemas . Shown are posterior trunk fragments cut from wild-type animals and regenerated for the indicated times: hours or days post-amputation ( hpa and dpa , respectively ) and subjected to double FISH with zpuf-6 and AGAT-1 . Dotted lines demarcate the posterior end of the animal . Maximum intensity projections are shown . Scale bar: 100 μm . Quantification of the fractional area of zpuf-6+ cells in the regenerating blastema compared to non-blastema ( old tissue ) . The overall positive signal of zpuf-6+ cells was measured in a fixed region in the blastema compared to the same area in the non-blastema region . Data represent means from 2 regenerating trunk fragments for each time point; error bars: SD . Student's t test: * , p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 01410 . 7554/eLife . 10501 . 015Figure 4—figure supplement 3 . NB . 22 . 1E+ and laminB+ cells exhibit little overlap with zpuf-6 and display slow cell turnover kinetics after irradiation . ( A ) Whole-mount FISH of zpuf-6 and markers of epidermal edge cells ( NB . 22 . 1E and laminB ) . zpuf-6 exhibits little to no overlap with NB . 22 . 1E in edge cells ( top row ) but overlaps with NB . 22 . 1E+ epidermal cells surrounding the ventral mouth opening ( middle row , white arrowheads ) . zpuf-6 shows very little overlap with laminB+ edge cells ( bottom row , white arrowheads ) . Scale bars: 50 μm . ( B ) Colorimetric WISH of animals after 6 , 000 Rad irradiation exposure for markers NB . 22 . 1E and laminB . By 14dpi ( days post-irradiation ) , cells around the anterior have disappeared , suggesting that these cells do not turnover as quickly as zpuf-6+ epidermal cells . For NB . 22 . 1E marker , sub-epithelial mesenchymal cells and those around the ventral mouth opening disappear by 7dpi , suggesting they turn over more rapidly than the cells at animal margin . Ventral views . Scale bars: 200 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 015 The formation of cilia is a signature of terminal differentiation ( May-Simera and Kelley , 2012 ) . The ventral epidermis is lined with multi-ciliated cells responsible for gliding motility , while the more densely packed dorsal surface contains many non-ciliated and mucous-secreting , rhabdite-containing cells ( Pedersen , 1976 ) . Given that zpuf-6 expression is evenly distributed throughout the dorsal and ventral epidermis , we queried whether zpuf-6+ epidermal cells co-expressed markers for cilia genes . We used the planarian rootletin gene as a marker for ciliated epidermal cells and found that zpuf-6 and rootletin do not significantly overlap in expression ( Figure 4C , top row ) , suggesting that zpuf-6+ epidermal cells are non-ciliated . However , the combination of vim-3 and rootletin in whole-mount FISH reveals that while cells expressing high levels of vim-3 display virtually undetectable levels of rootletin , there are cells with detectable vim-3 expression that do co-express rootletin ( Figure 4C , bottom row ) , suggesting that vim-3 cells may begin to undergo ciliogenesis . Under normal homeostasis conditions , zpuf-6+ epidermal cells appear randomly but broadly distributed on both the dorsal and ventral epidermal surface of intact animals ( Figure 2 ) . We reasoned that if zpuf-6+ epidermal cells still harbor the potential to differentiate further , then the density and distribution of zpuf-6+ cells would increase if epidermal integrity were severely perturbed . Therefore , we amputated wild-type animals and examined zpuf-6 expression in the regenerating blastema over 7 days ( Figure 4—figure supplement 2 ) . Consistent with our hypothesis , the density of zpuf-6+ epidermal cells as measured by fractional area is higher in the new undifferentiated blastema tissue compared to old tissue at 4 and 7 days post-amputation . We also used whole-mount FISH to examine zpuf-6 expression with NB . 22 . 1E and laminB , markers that are expressed in specific domains in the epidermis ( van Wolfswinkel et al . , 2014 ) . NB . 22 . 1E labels marginal adhesive gland cells at the body margin of the animal ( Reddien et al . , 2007; Tazaki et al . , 2002 ) as well as cells lining the ventral mouth opening . Curiously , there was very little overlap between zpuf-6 and NB . 22 . 1E at the edge , but NB . 22 . 1E+ cells around the ventral opening co-expressed zpuf-6 ( Figure 4—figure supplement 3A ) . LaminB is also expressed mainly in marginal adhesive cells and a few subepidermal cells near the edge , and displayed little overlap with zpuf-6 ( Figure 4—figure supplement 3A ) . We wondered whether there was minimal co-expression between zpuf-6 and NB . 22 . 1E and laminB because these marginal adhesive cells do not turnover as quickly as epidermal cells on the dorsal and ventral surface . To test this , we monitored the kinetics of NB . 22 . 1E and laminB expression after a lethal dose of irradiation . After 14 days post-irradiation , NB . 22 . 1E and laminB mesenchyme expression is completely lost , but only cells in the anterior region have disappeared , indicating that the marginal edge cells do not turnover as quickly ( Figure 4—figure supplement 3B ) . Moreover , vim-3 overlaps considerably with both NB . 22 . 1E+ and laminB+ cells at the animal body margin ( Figure 4—figure supplement 1B ) , suggesting that both zpuf-6 and vim-3 are expressed in multiple epidermal cell types . Taken altogether , transcriptional profiling of chd4 and p53 RNAi animals has identified additional epidermal progeny gene markers that are expressed broadly throughout the animal in discrete but overlapping cell populations in the mesenchyme and epidermis . zpuf-6+ epidermal cells can still undergo progressive differentiation into an unknown number of epidermal cell types ( potentially including NB . 22 . 1E+ cells lining the mouth and laminB marginal adhesive cells ) , by expressing markers of cytoskeletal morphogenesis ( vim-1/vim-3 ) followed by cilia gene markers ( rootletin ) . The addition of these new molecular markers in the planarian epidermal lineage will greatly facilitate the study of epidermal progenitor dynamics and differentiation during tissue homeostasis and regeneration . Transcriptional changes are likely drivers of maturation within the epidermal lineage , which occurs in the absence of cell division . We therefore focused on characterizing putative transcription factors from our common down-regulated gene set as likely candidates important for post-mitotic epidermal differentiation . egr-5 is a planarian homolog of the early growth response ( EGR1 ) family of C2H2-type zinc-finger TFs ( Figure 5—figure supplement 1 ) . EGR1 genes are known to be induced by extracellular signals including growth factors , hormones and neurotransmitters , and couple these signals to long-term responses by altering the expression of target genes ( Thiel and Cibelli , 2002 ) . Egr-5 is expressed in discrete mesenchymal cells located throughout the animal in a spatial pattern similar to AGAT-1 cells ( Figure 2B ) . We examined egr-5 expression levels in greater detail by whole-mount FISH in combination with other post-mitotic markers of the epidermal lineage . Robust egr-5 expression appears to correlate with strong AGAT-1 expression , but egr-5 is also lowly expressed in AGAT-1neg zpuf-6+ mesenchymal cells ( Figure 5—figure supplement 2A , top panel ) . By whole-mount FISH metrics , egr-5 expression is also barely detectable in a subset of prog-1+ cells and in the epidermis ( Figure 5—figure supplement 2A , bottom panels ) , demonstrating that egr-5 expression , although varied in signal intensity , spans the domain of all post-mitotic lineage markers . In addition , egr-5 transcripts are predominantly found in the FACS-dissociated Xins population comprised of post-mitotic cells and not in the X1 neoblast dividing fraction ( Figure 5—figure supplement 2B ) , further suggesting that egr-5 mainly functions in post-mitotic cells . To investigate the role of egr-5 during normal tissue turnover , animals were fed egr-5 dsRNA every 3 days and subsequently screened for gross morphological defects . egr-5 ( RNAi ) animals displayed a range of deformities: loss of epidermal integrity , anterior blebbing , and complete loss of the anterior region ( Figure 5A ) . Moreover , the severity of phenotypic progression leading to animal death caused by lysis is inversely correlated with animal size ( Figure 5B ) . Because these gross morphological defects manifested in egr-5 ( RNAi ) animals suggested problems with epidermal morphology or density , we quantified epidermal nuclear density and found that egr-5 ( RNAi ) animals had a significantly reduced number of epidermal cells compared to control animals ( Figure 5C ) . 10 . 7554/eLife . 10501 . 016Figure 5 . Egr-5 is required for the proper differentiation of epidermal progeny cells . ( A ) Intact phenotypes for representative control and egr-5 ( RNAi ) animals . Various stages of phenotypic progression of egr-5 RNAi knockdown ( 4Fd15 ) are shown . Scale bars: 200 μm . ( B ) Survival curves for control and egr-5 ( RNAi ) animals . The efficacy of RNAi and resulting gross phenotype is dependent on size of animals at the start of RNAi feedings . Animals starting around 2-3 mm in size ( red lines ) were fed 5 times ( 5F ) ( n = 99 ) ; animals starting around 5-6 mm in size ( yellow lines ) were fed 6 times ( 6F ) ( n = 65 ) . Death was marked by complete animal lysis . ( C ) Effects of egr-5 ( RNAi ) on epidermal cell density . DapI was used to quantify epidermal ( Ep ) nuclear density in the ventral mid-sections; representative images for 4Fd24 are shown . Scale bar: 50 μm . Quantification of epidermal cell density are plotted for d12 , d18 and d24 for control ( blue ) and egr-5 ( RNAi ) ( red ) animals . Each symbol represents individual animals ( average of 2 regions/animal ) . Black lines and error bars ( colored ) represent mean and SD , respectively . Student's t test: *** , p <0 . 00005 . ( D ) Spatial domain expansion of prog-1 and AGAT-1 cells in egr-5 ( RNAi ) . Tissue transverse sections of control and egr-5 ( RNAi ) animals stained for markers of early epidermal progeny cells ( 8Fd30 ) . To help improve visualization of AGAT-1-expressing cells , an RNA probe mix of other AGAT-1 co-expressed genes were used ( AGAT-1 mix = AGAT-1 , zpuf-1 , zpuf-3 , zpuf-4 ) . Scale bar: 10 μm . ( E ) Misexpression of epidermal progeny markers in egr-5 ( RNAi ) animals . Triple FISH of prog-1 , AGAT-1 and zpuf-6 markers in control and egr-5 ( RNAi ) animals ( 4Fd21 ) . Percentages shown in prog-1 panel represent percentage of prog-1+ cells that also express AGAT-1 ( 400-–700 cells were counted between 3-–4 worms per condition ) . Yellow arrowheads highlight cells that are prog-1+ AGAT-1+ zpuf-6+ ( also shown in inset panels in merge panel ) . Rightmost column: epidermal ( Ep ) view highlighting loss of zpuf-6 expression in epidermis . AGAT-1 mix = AGAT-1 , zpuf-1 , zpuf-3 , zpuf-4; zpuf-6 mix = zpuf-6 , zpuf-5 and zpuf-8 . Scale bar: 50 μm . ( F ) Average fluorescence intensity for AGAT-1 mix and zpuf-6 mix probes in control and egr-5 ( RNAi ) animals from ( E ) . Error bars: SD . p-values are results of unpaired Student's t test . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 01610 . 7554/eLife . 10501 . 017Figure 5—figure supplement 1 . Smed-Egr-5 is a conserved member of the early growth response family of transcription factors . ( A ) Schematic of predicted domain structures of Smed-Egr-5 ( Sm ) with Egr1 from humans ( Hs ) and mouse ( Mus ) , adapted from the Smart Modular Architecture Research Tool ( SMART ) ( http://smart . embl-heidelberg . de ) . ( B ) Alignment of planarian Egr-5 protein sequence ( Smed-Egr-5 ) with Egr1 sequences from human ( Hs ) and Mus musculus ( Mm ) . Predicted Sm ZnF_C2H2 domains are residues 268-346 . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 01710 . 7554/eLife . 10501 . 018Figure 5—figure supplement 2 . Egr-5 is expressed in multiple post-mitotic epidermal progeny cells . ( A ) Whole-mount FISH of egr-5 with various markers in the epidermal lineage . Top panel: Cells expressing the strongest levels of egr-5 exhibit strong AGAT-1 expression and very weak zpuf-6 expression ( red arrowheads ) . Cells expressing lower levels of egr-5 exhibit strong zpuf-6 expression and little to no detectable AGAT-1 expression ( yellow arrowheads ) . Middle panel: Some cells expressing low levels of egr-5 are prog-1+ ( white arrowheads ) . Bottom panel: egr-5 is very weakly expressed in some epidermal cells , which overlap with zpuf-6+ epidermal cells ( not shown ) . Percentages represent fraction of egr-5+ cells ( low but detectable expression was counted ) that co-express the candidate gene ( ∼200-400 cells were quantified ) . Images are single confocal planes . Ep , epidermis . Scale bars: 10 μm . ( B ) egr-5 transcripts are enriched in the post-mitotic , differentiated cell population ( Xins ) . Average RPKM values are plotted for various known neoblast and non-neoblast genes from wild-type planarian cells dissociated by FACS ( data are deposited under GEO accession number: GSE73027 ) . X1 cells are irradiation-sensitive and comprise of genes that are expressed in stem cells ( smedwi-1 ) . Xins cells are irradiation-insensitive and represent a population of various post-mitotic cells . chd4 and p53 are expressed in both stem cells and post-mitotic cells and are therefore present in both X1 and Xins cell population . prog-1 and AGAT-1 are post-mitotic epidermal lineage markers and show enrichment in the Xins population . egr-5 has no detectable expression in stem cells based on FISH ( not shown ) , and consistent with this , few egr-5 transcripts are detected in the X1 population . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 01810 . 7554/eLife . 10501 . 019Figure 5—figure supplement 3 . Molecular and ultrastructural analysis of egr-5 ( RNAi ) epidermis . ( A , B ) Expression patterns of zpuf-6 and rootletin cilia gene marker at ( A ) 4Fd12 and ( B ) 4Fd24 after initial RNAi treatment for control and egr-5 ( RNAi ) animals . Representative images are single confocal planes of ventral epidermis ( Ep ) . Scale bars: 50 μm . ( C ) Scanning electron micrograph of images ( 1500X ) of control and egr-5 ( RNAi ) animals ( 4Fd12 ) of the ventral epidermis ( Ep ) right above the mouth opening . Although cilia are still present in egr-5 ( RNAi ) animals , there is an absence of epidermal pores ( red arrowheads ) that may be a result of epithelial stretching . Scale bar: 10 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 01910 . 7554/eLife . 10501 . 020Figure 5—figure supplement 4 . Reduction of laminB+ cells in egr-5 ( RNAi ) animals . The number of laminB+ cells per mm along the animal margin was quantified in control and egr-5 ( RNAi ) animals . The quantified region was restricted to the anterior because this region exhibits the highest cell turnover of laminB+ cells . Representative images for 4Fd21 are shown . Scale bar: 50 μm . Quantification of laminB+ cell density is plotted from 3 animals for each condition . Data represent means; error bars: SD . Student's t test: * , p < 0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 020 Next , we investigated epidermal morphology by looking at expression of zpuf-6 and rootletin using whole-mount FISH and ultrastructural features by scanning electron microscopy ( SEM ) . We observed a significant reduction in zpuf-6 expression ( see more below ) in the epidermis between day 12 and day 24 of RNAi treatment , whereas rootletin expression did not diminish , although it appeared disorganized compared to controls ( Figure 5—figure supplement 3A–B ) . Analysis of the ventral cilia by SEM qualitatively showed similar levels in abundance . However , the ventral epidermal cells in egr-5 ( RNAi ) animals appeared more smooth and stretched out and lacked epidermal pores/pits as seen in control animals ( Figure 5—figure supplement 3C ) . The smoothness of the ventral epidermis also contributed to difficulty in delineating individual epidermal cells . Our data suggest that the reduced epidermal density in egr-5 ( RNAi ) animals could be caused by the failure of new epidermal progeny cells to properly differentiate and incorporate into the epidermis , resulting in stretching and apparent morphological defects . Consistent with this hypothesis , we also found that there was a significant reduction in the number of laminB+ cells in the anterior region of egr-5 ( RNAi ) animals compared to control animals after 21 days of RNAi treatment ( Figure 5—figure supplement 4 ) . Given that egr-5 knockdown causes a marked reduction in zpuf-6 expression in the epidermis , we looked at our panel of epidermal lineage markers to determine which particular transition may be affected in the course of differentiation . In RNAi conditions such as chd4 , p53 and zfp-1 knockdown where AGAT-1+ cell populations are lost , there is a distinct progressive disappearance in the spatial distribution of cells , but the signal intensity of the gene marker expressed in the remaining cells is not significantly diminished ( see Figure 2—figure supplement 1A; Figure 3—figure supplement 2B ) . This suggests that these factors are required for the generation or maintenance of AGAT-1+ cells , but not necessarily required for the expression of AGAT-1 or other markers in these cells . However , a TF can be responsible for the expression of AGAT-1 or some other marker , and not be required for the maintenance of that cell type , or both . To distinguish between the loss of cell type versus the loss of specific gene expression , we made RNA probe mixes consisting of multiple zpuf genes from our screen expressed in either AGAT-1 or zpuf-6 cells to increase overall signal intensity and detection . We first investigated the spatial distribution of prog-1 and AGAT-1 progeny markers and found that the stereotypical sub-epidermal expression domains of both markers had expanded deeper into the mesenchyme in egr-5 ( RNAi ) animals ( Figure 5D ) , suggesting potential temporal defects in the earlier steps of epidermal lineage progression . We then simultaneously analyzed prog-1 , AGAT-1 and zpuf-6 markers by whole-mount FISH and established that in egr-5 ( RNAi ) animals , epidermal progeny cells in the mesenchyme are not specifically lost , but that AGAT-1 and zpuf-6 expression levels were significantly reduced ( Figure 5E–F ) . Although zpuf-6 mesenchymal expression is very weak after egr-5 knockdown , it is virtually undetectable in the epidermis , potentially highlighting a key transitional defect . Given that egr-5 knockdown reduces both epidermal cell density as well as the expression of AGAT-1 and zpuf-6 markers , we suggest that egr-5 is responsible for both the maturation of epidermal progeny cells and the expression of progeny markers . Interestingly , prog-1 and AGAT-1 normally do not exhibit significant overlapping expression domains , but here they display significant overlapping expression patterns after loss of egr-5 ( Figure 5E , left panels ) . Together , our results demonstrate an essential role for egr-5 in the proper spatial and temporal progression of progeny cells in the epidermal lineage and that misexpression of these marker domains likely causes improper differentiation and failed epidermal maturation . Although egr-5 is expressed in post-mitotic epidermal lineage cells to coordinate their proper temporal transition states , we wondered if loss of egr-5 may have any non-autonomous effects on neoblast and progenitor dynamics . To examine neoblast division kinetics , we quantified histone H3Ser10 phosphorylation ( H3P ) during egr-5 RNAi knockdown and observed a significant increase in neoblast proliferation , followed by an eventual decline ( Figure 6A–B ) . 10 . 7554/eLife . 10501 . 021Figure 6 . Egr-5 knockdown results in the expansion of multiple progenitor populations . ( A , B ) egr-5 knockdown causes an increase in global stem cell proliferation . ( A ) H3P-positive cells over the surface area in control and egr-5 ( RNAi ) intact animals ( 8 feedings ) . Maximum intensity projections of representative H3P patterns for day 12 and day 35 time points are shown . Scale bar: 200 μm . ( B ) Quantification of H3P-positive cells per surface area from ( A ) . Data represent means , error bars: SD . Student's t test: , p <0 . 005 , * , p <0 . 0001 . ( C–G ) Stem cell proliferation results in the expansion of multiple lineage-committed progenitors . For control and egr-5 ( RNAi ) animals ( 3Fd27 ) , animals were fixed and stained for H3P and a marker for the following: ( C ) dividing epithelial progenitors ( zeta class: zfp-1 , egr-1 , fgfr-1 ) , ( D ) dividing protonephridia progenitors ( pou2/3 ) , ( E ) dividing gut progenitors ( hnf4 ) , and ( F ) dividing brain progenitors ( pax6A ) . Quantifications of percentage of lineage-committed progenitors of total H3P+ cells are shown . Error bars: SD . ( G ) Quantification of total mitotic cells ( H3P ) per surface area for control and egr-5 ( RNAi ) animals ( 3Fd27 ) from analysis in ( C–F ) . Data represent means , error bars: SD . Student’s t test: * , p <0 . 0001 . ( H , I ) egr-5 knockdown causes supernumerary protonephridial units ( PU ) . ( H ) Control and egr-5 ( RNAi ) intact animals ( 8 feedings ) were stained for the number of proximal units ( slc6a-13 ) to visualize and quantify total PUs . Maximum intensity projections of representative animals for day 9 and day 35 time points are shown . Scale bar: 200 μm . ( I ) Quantification of PU per surface area . Data represent means , error bars: SD . Student's t test: * , p <0 . 0001 . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 02110 . 7554/eLife . 10501 . 022Figure 6—figure supplement 1 . Analysis of multiple dividing tissue progenitors in egr-5 ( RNAi ) animals . ( A ) Representative images of egr-5 ( RNAi ) animals from Figure 6C–G fixed and stained for H3P ( red ) , neoblast marker ( blue: smedwi-1 ) and progenitor marker ( green ) . Progenitor markers: zeta-class/epidermal: zfp-1 , egr-1 , fgfr-1; protonephridia: pou2/3; gut: hnf4; brain: pax6A . Images represent whole-animal maximum intensity projections and zoomed image highlights an example of a positive dividing progenitor cell ( i . e . H3P+/smedwi-1+/progenitor marker+ ) . Scale bars: 200 μm . ( B-D ) Expansion of multiple differentiated tissue markers in egr-5 ( RNAi ) animals . Representative colorimetric WISH images of control and egr-5 ( RNAi ) intact animals ( 6Fd25 ) monitoring various lineages/tissues: protonephridia ( pou2/3 ) , epidermal progenitors and early/late progeny cells ( p53 ) , and gut morphology/gut progenitors ( hnf4 ) . Scale bars: 200 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 022 Because neoblasts constitute a mixed population of pluripotent stem cells and lineage-committed progenitors , we looked more closely at the composition of dividing neoblasts to determine if egr-5 knockdown expanded all dividing progenitors or only affected a specific subset . We picked a time point where H3P levels were high in egr-5 ( RNAi ) animals ( before the decline ) and then analyzed the proportion of dividing progenitors for major tissues , including the epidermis ( zeta-class ) ( van Wolfswinkel et al . , 2014 ) , protonephridia ( pou2/3 ) ( Scimone et al . , 2011 ) , gut ( hnf4 ) ( Wagner et al . , 2011 ) and brain ( pax6A ) ( Wenemoser et al . , 2012 ) ( Figure 6C–G , Figure 6—figure supplement 1A ) . Despite demonstrating a dramatic increase in overall dividing progenitors , egr-5 ( RNAi ) animals displayed similar proportions of dividing tissue progenitors compared to their control counterparts . These results suggest that egr-5 RNAi knockdown causes an overall expansion of multiple lineage progenitors , including the epidermal progenitors ( zeta-class ) . Therefore , failure to generate mature epidermal cells in egr-5 ( RNAi ) animals is not caused by a failure to generate zeta-class epidermal progenitors . Rather , it is likely caused by a defect at the post-mitotic differentiation level . This decrease over time in the integration of functional , differentiated cells leads to the eventual breakdown of epidermal integrity and results in animal lysis . To assess whether the expanded progenitor population in egr-5 ( RNAi ) animals led to any pronounced defects in differentiation , we examined by colorimetric WISH the protonephrida ( slc6a-13 ) ( Vu et al . , 2015 ) , epidermal progenitors/prog-1/AGAT-1 ( p53 ) , and overall gut morphology ( hnf4 ) . Qualitatively , all markers exhibited prominent increases in number and expression in egr-5 ( RNAi ) animals compared to controls ( Figure 6—figure supplement 1B–D ) . We quantified the number of protonephridial proximal units ( PU ) and found that egr-5 ( RNAi ) animals indeed display a significant increase in protonephridial density ( Figure 6H–I ) . Based on general morphology and the observation that egr-5 RNAi animals did not exhibit defects in osmoregulation by bloating , we conclude that the protonephridia are functional . The strong increase in p53 expression builds upon our previous finding that loss of egr-5 increases the number of zeta-class epidermal progenitors ( Figure 6C , Figure 6—figure supplement 1A ) , as well as prog-1 and AGAT-1 cells ( Figure 5D–E ) . The enriched gut expression marked by hnf4 in egr-5 ( RNAi ) animals also suggests the presence of an increased number of gut cells , without any discernible physiological consequences ( e . g . , animals can still eat ) . Together , our data suggest that the extensive increase in neoblast proliferation caused by egr-5 depletion is remarkably counterbalanced by a corresponding increase in the normal progression of multiple lineage progenitors that differentiate and integrate into functional tissues . The potential non-cell autonomous effect of egr-5 knockdown on neoblast and progenitor dynamics raised the possibility that epidermal defects resulting from abnormal differentiation and defective homeostasis may induce some kind of global stress response . Therefore , we measured apoptosis using a whole-mount TUNEL assay ( Pellettieri et al . , 2010 ) and observed a remarkable increase in cell death throughout the course of egr-5 RNAi treatment compared to control animals ( Figure 7A ) . Notably , the increase in cell death appeared to occur uniformly throughout the animal , with the majority of TUNEL-positive nuclei found in the mesenchymal tissue and not in the epidermis ( not shown ) . A global increase in both cell proliferation and cell death are general features of the initial stages of planarian regeneration ( Pellettieri et al . , 2010; Wenemoser and Reddien , 2010 ) . Simple wounding and amputations causing loss of tissue are both capable of activating the expression of many wound-induced genes within the first 24 hr of insult ( Wenemoser et al . , 2012 ) . We hypothesized that the epidermal defects caused by egr-5 RNAi knockdown may trigger the activation of wound-induced genes expressed in differentiated cells . Delta-1 is a putative Notch signaling pathway ligand whose expression is induced in the epidermis between 6-24 hr after injury , not necessarily localized to the site of wounding ( Wenemoser et al . , 2012 ) . We monitored the expression of delta-1 in intact , uninjured egr-5 ( RNAi ) animals and observed a marked increase in delta-1 in the anterior region of animals compared to their control counterparts through 21 days of RNAi treatment ( Figure 7B , top row ) . We also measured the in vivo expression of other ‘immediate early genes’ that are activated after wounding including fos-1 , jun-1 and egr-3 ( Wenemoser et al . , 2012 ) , and observed notable increases in the anterior region of egr-5 ( RNAi ) animals ( Figure 7B , middle rows ) . Interestingly , the TGF-β inhibitor follistatin , whose expression is specifically induced for multiple days after a loss of animal tissue ( Gavino et al . , 2013 ) , does not appear to be stimulated over the course of egr-5 RNAi treatment ( Figure 7B , bottom row ) . In addition , the immediate early genes fos-1 , jun-1 and egr-3 are enriched in our chd4 and p53 RNAi whole-worm RNA-seq up-regulated datasets ( Supplementary file 2; Figure 7—figure supplement 1 ) . Taken together , these data suggest that both defects in epidermal differentiation ( egr-5 RNAi ) and loss of epidermal progeny cells ( chd4 and p53 RNAi ) likely contribute to a breach in epidermal integrity , which consequently may activate a systemic wound response program . 10 . 7554/eLife . 10501 . 023Figure 7 . A global stress response is induced by loss of egr-5 . ( A ) Whole-mount TUNEL assay measuring apoptosis in control and egr-5 ( RNAi ) animals ( 8 feedings ) . Quantification of TUNEL-positive nuclei per surface area is plotted . Data represent means; error bars: SEM . Student's t test: * , p < . 05 , , p < . 005 , * , p < . 0001 . Representative maximum intensity projection of TUNEL-stained image for day 9 and day 35 are shown . Scale bars: 200 μm . ( B ) Wound-induced genes are up-regulated in egr-5 ( RNAi ) animals . Representative colorimetric WISH images of delta-1 , fos-1 , jun-1 egr-3 and follistatin expression in the anterior regions of intact control and egr-5 ( RNAi ) animals at 4Fd21 of RNAi treatment . Scale bars: 200 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 02310 . 7554/eLife . 10501 . 024Figure 7—figure supplement 1 . Wound-induced genes are upregulated in chd4 and p53 RNAi datasets . Plot of log2 ratios of selected wound-induced genes from Figure 7B in chd4 ( RNAi ) ( top panel ) and p53 ( RNAi ) ( bottom panel ) whole worm RNA-seq time course . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 024 Zeta-class neoblasts , marked by zfp-1 , were recently reported to represent a class of epidermal progenitor cells , giving rise to prog-1+ , AGAT-1+ mesenchymal cells , and other markers of epidermal cells ( van Wolfswinkel et al . , 2014 ) . However , their precise molecular relationships and the mechanisms underlying the differentiation of these progeny cells are poorly understood . Our results extend upon these studies by implementing a whole worm RNA-seq approach to identify genes expressed in AGAT-1+ cells and their progeny and to characterize the post-mitotic transition states downstream of epidermal progenitors . We utilized a combination of assays , including whole-mount FISH , spatiotemporal kinetics after irradiation , and BrdU lineage tracing , to establish the temporal order of epidermal cell differentiation . Combinatorial FISH revealed that zpuf-6 is co-expressed at varying levels in virtually all AGAT-1+ cells , additional AGAT-1neg mesenchymal cells ( AGAT-1negzpuf-6+ ) , and cells evenly distributed throughout the dorsal and ventral epidermis ( Figure 2 ) . We examined the expression patterns of all other genes from our screen that displayed discrete mesenchymal cell patterns in combination with AGAT-1 and zpuf-6 , and found that they all overlapped extensively . Many of these genes , such as pla2 and odc-1 , exhibited moderate expression in AGAT-1neg zpuf-6+ cells , but expression was noticeably reduced in zpuf-6+ epidermal cells ( Figure 2—figure supplement 3 ) , suggesting that these cells are turning off expression of those genes . In addition , AGAT-1 and zpuf-6 showed conserved spatiotemporal down-regulated kinetics after irradiation and in RNAi conditions leading to loss of AGAT-1+ cells ( Figure 3 , Figure 3—figure supplement 1 , 2 ) , and a pulse of BrdU incorporated into dividing neoblasts showed a temporal progression through prog-1+ , AGAT-1+ and zpuf-6+ cells ( Figure 4A ) . Altogether , we interpret these observations to mean that prog-1 , AGAT-1 and zpuf-6 are markers of three major spatially and temporally related cell populations representing distinct transitional stages of epidermal maturation ( Figure 8 ) . 10 . 7554/eLife . 10501 . 025Figure 8 . Model of planarian epidermal lineage progression . Schematic representation of the distribution of neoblasts , epidermal progeny cells in the mesenchyme , and differentiated epidermal cells along the anteroposterior ( top left , dorsal view ) and dorsoventral ( top right , cross section ) axes . Gut branches are shown as reference . Below: sigma-class neoblasts give rise to zeta-class neoblasts , which are epidermal progenitors . zfp-1 is required to generate epidermal progeny cells , which begin to express different markers as cells undergo multiple transitions and intercalate into the mature epidermis . Gradient boxes represent domains and relative expression intensity of specified genes . egr-5 is most strongly expressed in AGAT-1+ cells and is required for proper differentiation of epidermal post-mitotic progeny cells . zpuf-6+ epidermal cells likely represent a branching point in the epidermal cell fate decision and can generate multiple different cell types including marginal adhesive cells ( NB . 22 . 1E and laminB ) , multiciliated cells ( rootletin ) , and other unknown cell types . Bottom panel: schematic of egr-5 ( RNAi ) phenotype . Loss of egr-5 disrupts the proper spatiotemporal transition of post-mitotic epidermal cells , resulting in prog-1 , AGAT-1 and zpuf-6 to be co-expressed in the same cell , and leading to a depletion of mature epidermal cells . The resulting loss of epidermal integrity is sensed by the neoblasts ( induced stress response ) and causes an expansion of neoblasts and multiple progenitors before animals eventually lyse due to irreparable loss of an intact epidermal barrier . DOI: http://dx . doi . org/10 . 7554/eLife . 10501 . 025 Differentiation and migration toward the mature planarian epidermis are temporally correlated , but what are the fates of cells once they pierce through the basement membrane and integrate into the epidermal layer ? We propose that zpuf-6+ epidermal cells may represent a stable transition state with the potential to give rise to an unknown number of differentiated cell types ( Figure 8 ) . Zpuf-6+ epidermal cells are interspersed throughout the epidermis but do not express high levels of a cilia gene , rootletin ( Figure 4C ) , which is a marker for differentiated ciliated epidermal cells . However , zpuf-6+ epidermal cells overlap extensively with a predicted intermediate filament ( IF ) gene , vim-3 , which is expressed more broadly in the dorsal epidermis . Zpuf-6neg epidermal cells express lower levels of vim-3 than zpuf-6+ cells , but these cells begin in turn to express rootletin ( Figure 4C ) . IF proteins are a diverse class of cytoskeletal elements that provide cells with mechanical support , contribute to the structural integrity of cells , and act as markers of morphogenesis and differentiation ( Kim and Coulombe , 2007 ) . In addition , zpuf-6+ cells overlap with two specific subset of cells: one expressing the marker NB . 22 . 1E surrounding the ventral mouth opening and the other expressing the laminB IF gene that marks adhesive cells along the animal body margin ( Figure 4—figure supplement 3 ) . Lastly , zpuf-6+ cells are present at higher cell density in the regenerating blastema than under normal homeostasis conditions ( Figure 4—figure supplement 2 ) , suggesting that they have not completed differentiation . The decisions influencing a particular cell fate path for zpuf-6+ epidermal cells could be determined by the immediate , surrounding environment . Given that dorsal and ventral epidermal cells are morphologically distinct , positional cues and signals are likely to contribute to the cell fate specification process . For example , transplantation experiments involving a plug of tissue from a donor transplanted in the reverse dorsal-ventral ( DV ) orientation with regard to the host can trigger tissue outgrowths on both sides where the grafted tissue maintains its original DV identity ( Kato et al . , 1999 ) . The sub-epidermal layer of body wall muscle cells express most of the polarity cues that organize the planarian body axes , including receptors , ligands and inhibitors of the BMP and Wnt signaling pathways ( Witchley et al . , 2013 ) , and could potentially impact epidermal fates as well . The role of Notch signaling has not been well characterized in planarians but has been implicated in the regulation of cell fate determination in epithelial cells of the mouse intestine ( VanDussen et al . , 2012 ) and the lung ( Rock et al . , 2011; Rock and Hogan , 2011 ) . We cannot exclude the possibility that zpuf-6+ epidermal cells simply reflect an abundant , differentiated cell population that undergoes rapid turnover . Additionally , there could be other progenitors ( zeta-class dependent or independent ) that give rise to specific epidermal cell types that have not been defined molecularly . To build upon our understanding of epidermal cell fate specification , it will be necessary to define all the types and numbers of cells that ultimately compose both the ventral and dorsal planarian epidermis , as well as the factors that contribute to their identity and maintenance . In planarians , TFs play prominent roles in specifying neoblasts to a committed fate ( Reddien , 2013 ) . Key TFs that are important for specifying neoblasts during homeostasis and regeneration often remain expressed in mature cell types , suggesting that these TFs are also required for the maintenance of cellular identity . Studies of planarian eye regeneration currently provide the most detailed characterization of planarian lineage specification . During regeneration , eye progenitors express conserved TFs ovo , six-1/2 and eya at a distance from the eye primordium , forming a trail of cells and undergoing changes in gene expression as they migrate toward the primordium , where differentiation markers begin to be expressed ( Lapan and Reddien , 2011; Lapan and Reddien , 2012 ) . However , ovo , six-1/2 and eya are also expressed in mature eyes and are likely involved in the maintenance of the optic cup cells and photo-sensing neurons of this sensory organ . Our findings also suggest that distinct molecular changes occur both spatially and temporally , and are necessary for the proper differentiation of epidermal progenitors into mature cell types . However , egr-5 , which is required for the stable maturation of epidermal progenitors and their progeny , does not appear to be required for the maintenance of the differentiated state . In fact , the lack of egr-5 expression in the epidermis suggests that shutting off its expression may be an important step in promoting differentiation . Whether this regulation is intrinsic or extrinsic awaits further investigation . Moreover , because loss of egr-5 leads to the misexpression of AGAT-1 and zpuf-6 markers in prog-1+ cells , it could be directly or indirectly involved in turning off prog-1 expression and activating the expression of genes required for the AGAT-1+ transition stage . A key remaining question is how individual epidermal progeny cells undergo distinct differentiation transition stages , including responses to external signals or potential changes in chromatin structure . Genes involved in small molecule metabolic processes ( specifically organonitrogen processes ) were enriched in our common chd4 and p53 RNAi down-regulated gene list . Many of the metabolic genes screened by in situ were expressed in AGAT-1+ and zpuf-6+ cells , including gatm , odc-1 , cytochrome p450s ( cyp ) , and pla2 . Furthermore , genes including acsl-2 , adss , slc25a-19 and zpuf-7 were expressed in both the gut and cells of epidermal lineage , which suggests that these tissues may share common functional roles . There are four arginine:glycine amidinotransferases ( AGAT-1 , AGAT-2 , AGAT-3 , gatm ) identified in planarians . These enzymes catalyze the first step in creatine biosynthesis by transferring the amidino group of arginine to glycine to yield ornithine and guanidinoacetic acid ( Wyss , 2000 ) . Creatine plays an important role in muscle energy homeostasis . In vertebrates , de novo synthesis of creatine mainly takes place in the kidney , pancreas and liver ( Nabuurs et al . , 2013 ) . Thus , it is assumed that AGAT-1+ cells synthesize creatine , which is then released and taken up by neighboring muscle cells and neurons ( Eisenhoffer et al . , 2008 ) , though this has not been formally demonstrated . Polyamines are ubiquitous polycations that are essential for eukaryotic cell growth , and polyamine metabolism is frequently dysregulated in cancer ( Pegg , 2009 ) . Some of the elucidated roles for polyamines in cell growth include maintenance of chromatin conformation , gene regulation , ion channel regulation , and free radical scavenging ( Casero and Marton , 2007 ) . Ornithine decarboxylase ( ODC ) catalyzes the first rate-limiting step in polyamine biosynthesis by converting ornithine to putrescine , which then is converted to spermidine and spermine . In planarians , ODC activity has been reported to be induced early in regeneration near the wound site , and inhibition of ODC activity appears to cause a reduction in cell proliferation and differentiation ( Saló and Baguna , 1989 ) . However , mechanisms of action by odc-1 and the regulation of neoblasts require further characterization . The cytochrome P450s ( CYPs ) are enzymes that use molecular oxygen and complex reaction chemistry to modify their substrates . They are involved in a large number of physiological processes , including steroid hormone synthesis and detoxification of xenobiotics ( Anzenbacher and Anzenbacherova , 2001; Zhang and Yang , 2009 ) . CYPs make up one of the most diverse eukaryotic gene families , and are classified into clans , families and sub-families based on phylogenetics and sequence identity ( Nelson et al . , 2013 ) . We identified a number of cyp genes that were expressed in AGAT-1+ progeny cells as well as in the gut . Steroid metabolism and detoxification are two processes generally performed by the liver , raising the questions of what physiological roles these various planarian cyp genes play and whether AGAT-1+ cells may have an additional endocrine-like function . Although p53 is best known for its central role as a tumor suppressor , it also regulates several aspects of cellular metabolism , including autophagy , central carbon metabolism , and lipid metabolism ( Berkers et al . , 2013 ) . Fatty acids ( FA ) have essential roles in the cell as sources of energy , membrane components , and signaling molecules ( Lopes-Marques et al . , 2013 ) . Long-chain fatty-acid coA-ligases ( ACSL ) are key enzymes involved in the initial steps of FA metabolism . We identified three enzymes involved in fatty acid metabolism , acsl-1 , acsl-2 and pla2 , that are expressed in the gut as well as in epidermal progeny cells and the mature epidermis . Taken together , the presence of multiple metabolic genes co-expressed in AGAT-1+ cells and their progeny suggests that despite these cells being in a temporal stage transitioning to mature epidermal cells , they may also play a dual functional/physiological role in planarian homeostasis . Epithelia are a hallmark of multicellular organisms , forming the interface between the organism and the external environment and lining the cavities and surfaces of organs ( Donati and Watt , 2015 ) . The planarian epidermis provides the first response to amputation-induced injury by covering up the wound site within 30 min ( Reddien and Sánchez Alvarado , 2004 ) . It has been postulated that rhabdites are released at wound sites , where their contents produce a protective mucosal covering , possibly providing immunological functions ( Reisinger and Kelbetz , 1964 ) . Therefore , a breach in epidermal integrity could be the stimulus for initiating regeneration , though how this signal could be transduced to neoblasts is unknown . The egr-5 ( RNAi ) phenotype resulting in the expansion of neoblasts and multiple progenitor populations reveals that complex dynamics are at play between differentiating progeny and how neoblasts respond to loss of tissue integrity . It will be interesting to investigate the neoblast response in other RNAi or pharmacological contexts that directly affect epidermal integrity to determine how these perturbations affect neoblast function . However , it still remains possible that neoblast expansion could be a response to the global increase in apoptosis , although we still do not understand how the processes of cell proliferation and cell death are integrated . Alternatively , egr-5 may be responsible for the expression of a signaling molecule to which neoblasts respond in a feedback control process . Feedback signaling from differentiating progeny to neoblasts may be an important and robust strategy to control stem cell activity during homeostasis and regeneration . Therefore , unveiling the downstream targets of egr-5 will help discern these possibilities , as well as provide more mechanistic insight on the role of egr-5 and the progression of epidermal fate maturation . Several key questions emerge from our findings . Why does epidermal lineage progression require multiple transition states ? Do neoblast progeny contribute to the dynamics of the microenvironment , thereby playing a niche-like role ? How are cell autonomous and non-cell autonomous mechanisms coupled for the maintenance of cellular equilibrium at the organismal level ? How is the overall physiological status of the animal sensed , integrated and converted to the appropriate neoblast output ? With a more nuanced understanding of their adaptive behavior , planarian neoblasts and the epidermal lineage have emerged as an important model system in which to study the dynamics between stem cell self-renewal and the orchestration of progeny differentiation . Schmidtea mediterranea CIW4 asexual strain was maintained in 1X Montjuic salts supplemented with 50 μg/ml Gentamicin and fed homogenized calf liver paste as previously described ( Gurley et al . , 2008; Reddien et al . , 2005a ) . Animals were on average starved between 7–10 days prior to starting experiments and ranged in size from 2 mm to 8 mm . Animals were exposed to 6000 Rads of gamma irradiation using a GammaCell 40 Exactor irradiator . Genes in this study were cloned from a CIW4 cDNA library into pPR-T4P vector ( J . Rink ) as described elsewhere ( Adler et al . , 2014 ) . Primer sequences are provided in Supplementary file 4 . Cloned gene vectors were transformed into bacterial strain HT115 for dsRNA production . RNAi food was prepared by mixing 50 ml of pelleted culture with 250 μl of calf liver paste ( 2X ) or 50 ml of culture with 125 μl of calf liver paste ( 4X ) . Animals were fed every 3 days , with the first day designated as Day 0 of RNAi treatment . The number of RNAi feedings performed for egr-5 knockdown varied depending on starting-size of animals and RNAi food concentration . RNAi feedings and time points analyzed are noted in figure legends as XFdY ( 3Fd18 = 3 RNAi feedings , day18 ) . For all RNAi feeding experiments , unc22 dsRNA was used as the control for the same number of feedings as experimental RNAi animals . Images of live animals were captured using a Leica M205 stereoscope . Whole-mount colorimetric and fluorescent in situ hybridizations were performed using a detailed protocol as previously described ( King and Newmark , 2013; Pearson et al . , 2009 ) . Fluorescence-labeled animals were mounted in ScaleA2 solution ( Hama et al . , 2011 ) . Immunostaining with anti-H3P ( 1:1000 , Millipore , Billerica , MA ) was performed following fluorescent in situ development and was detected using Alexa-conjugated secondary antibody ( 1:1000 , Abcam , Cambridge , MA ) . Animals were fixed and stained for TUNEL using a method previously described ( Pellettieri et al . , 2010 ) with modifications: animals were bleached in 0 . 075% ammonia and 3% hydrogen peroxide and treated with ProteinaseK ( 2 μg/ml ) in PBSTx ( 0 . 3% Triton ) for 10 min followed by 4% formaldehyde incubation for 10 min prior to TdT reaction . Cryosectioned animals were processed after fluorescent in situ development as previously described ( Tu et al . , 2012 ) . For nuclear fast red staining , animals were processed for colorimetric ISH and were subsequently fixed overnight in 4% paraformaldehyde ( PBS ) at 4°C followed by dehydration in 30 , 50 , and 70% ethanol . Fixed specimens were embedded in paraffin and serial sectioned at 10 μm thickness and counter stained with nuclear fast red . For scanning electron microscopy , animals were immersed in a relaxant fixative ( 1% HNO3 , 0 . 85% formaldehyde , 50 mM MgSO4 ) as described elsewhere ( Rompolas et al . , 2013 ) for 5 min and were replaced with fresh fixative and rocked overnight at room temperature . Animals were then transferred to a solution containing 2 . 5% glutaraldehyde , 2% paraformaldehyde , 1% sucrose , 1 mM CaCl2 in 0 . 05 M NaCacodylate buffer pH 7 . 36 and left at 4°C until ready for processing . Animals were rinsed in ultrapure water , and secondary fixation was performed at 4°C overnight in 2% aqueous osmium tetroxide . Samples were dehydrated in a graded series of ethanol and dried in a Tousimis Samdri-795 critical point dryer . Samples were mounted on stubs and sputter coated with gold palladium . Imaging was done with a Hitachi TM-1000 tabletop SEM . Colorimetric WISH images were acquired using a Zeiss Lumar V12 stereomicroscope equipped with an AxioCam HRc . Colorimetric WISH tissue sections were imaged using a Zeiss Axiovert206 . Fluorescent images were acquired with either a Zeiss LSM-510 VIS or a Perkin Elmer Ultraview VOX spinning disk . Stitching and batch processing of images , and H3P quantifications per surface area were performed as previously reported ( Adler et al . , 2014 ) . For H3P/progenitor colocalization thresholding , spots were selected based on signal intensity a multiple above the noise background and filtered based on size . For fluorescence intensity quantification , Z-stack images were acquired and z-projected by average in a fixed neighborhood around the bottom of the animal as determined by DAPI nuclear staining after background subtraction . For epidermal cell quantifications , single focal plane images were acquired and nuclei within a fixed sized rectangle in regions of uniform cell density were quantified using ‘Find Maxima’ with Fiji software . All macros and plugins are available at https://github . com/jouyun . BrdU was administered by soaking animals for 24 hr in 20 mg/ml BrdU ( Sigma , St . Louis , MO ) dissolved in 3% DMSO and 1X Montjuic . Animals were then incubated in 5 g/L Instant Ocean supplemented with 50 μg/ml Gentamicin over the course of BrdU chase period . Animals were fixed and processed using the in situ hybridization protocol except bleached in 6% H2O2 in PBSTx ( 0 . 3% Triton ) for 3–4 hr under direct light . After development , specimens were treated with 2N HCl for 45 min at room temperature . BrdU was detected using a rat anti-BrdU antibody ( 1:1000; Abcam , Cat . No . ab6326 ) . Primary antibody was detected with HRP-conjugated anti-rat antibody ( 1:1000; Jackson ImmunoResearch , West Grove , PA ) . Images were acquired with a Perkin Elmer ( Waltham , MA ) Ultraview VOX spinning disk with a 20x 0 . 8 NA Plan Apochromat objective ( Zeiss , Oberkochen , Germany ) onto an Orca R2 ( Hamamatsu Photonics , Hamamatsu , Japan ) camera . Integration times were adjusted per worm to achieve well-saturated images . Z-stacks were collected at the anterior end of the worm over 250 µm in the Z dimension , with 2 µm steps , which contained nearly the entire depth of the worm . Images were cropped to a 400 pixel2 ( 132 µm2 ) region posterior to the brain . The stack was also cropped in Z to the ventral half of the stack . Target-specific cells were first marked manually in Fiji followed by target and BrdU double-positive cells . At least 6 worms ( up to 10 ) per time point were analyzed and roughly 6 , 500 cells total were quantified . The average percent positive cells were reported ( ( double-positive/total ) *100 ) . Representative images ( Figure 4A ) were first background subtracted in ImageJ with a 25 pixel radius rolling ball and then gaussian-blurred with a 1 pixel radius . For RNA-seq analysis of control , chd4 ( RNAi ) and p53 ( RNAi ) animals , three biological replicates of 10 worms each were collected for RNA isolation . mRNAseq libraries were generated from 500ng of high quality total RNA , as assessed using the LabChip GX ( Perkin Elmer ) . Libraries were made according to the manufacturer’s directions for the TruSeq Stranded mRNA LT– set A and B ( Illumina , San Diego , CA; Cat . No . RS-122-2101 and RS-122-2102 ) . Resulting short fragment libraries were checked for quality and quantity using a LabChip GX and Qubit Fluorometer ( Life Technologies , Carlsbad , CA ) . Libraries were pooled , requantified and sequenced as 50 bp single reads on the Illumina HiSeq 2500 instrument using HiSeq Control Software 2 . 0 . 5 . Following sequencing , Illumina Primary Analysis version RTA 1 . 17 . 20 and Secondary Analysis version CASAVA-1 . 8 . 2 were run to demultiplex reads for all libraries and generate FASTQ files . RNA-seq analysis was carried out by mapping sequence reads to a set of 36 , 035 S . mediterranea transcripts assembled from various sources including a previous transcriptome used in microarray studies ( Adler et al . , 2014 ) , trinity assemblies from lab-generated data involving whole animals , embryos , and sorted X1 cells , a transcriptome from the Bartscherer lab ( Boser et al . , 2013 ) , and the Dresden transcriptome assembly from PlanMine ( http://planmine . mpi-cbg . de ) , reduced as a collection to unique representations of loci via CD-HIT ( Fu et al . , 2012 ) . Sequences can be downloaded from http://smedgd . stowers . org . Reads were aligned using bowtie with the following parameters: --best --strata -v 2 -m 5 , and read counts to genes were tallied from the SAM files with a custom script . Differential gene expression was evaluated using R and the edgeR library ( Robinson et al . , 2010 ) . P-values were adjusted as previously described ( Benjamini and Hochberg , 1995 ) . The RNA-Seq data and the transcriptome against which it was quantified have been archived at GEO and is available under accession number: GSE72389 . Gene Ontology ( GO ) ( Gene Ontology Consortium , 2015 ) terms were assigned to each S . mediterranea gene based on homologous PFAM domains and significant Swissprot hits . GO term enrichment was performed using the R package topGO ( Alexa and Rahnenfuhrer , 2010 ) .	Tissues in adult animals contain cells called adult stem cells , which can divide to generate more adult stem cells ( in a process called self-renewal ) or specialize into other cell types ( via a process called differentiation ) . This means that adult stem cells can replace the specialized cells that are continually lost from animal tissues and organs . This allows the organs to continue to work properly . It is important to understand how adult stem cells decide whether to self-renew or differentiate because if they proliferate too much they may form abnormal growths such as tumors . On the other hand , if adult stem cells do not properly differentiate into specialized cells it can lead to tissue degeneration or even premature aging . Now Tu et al . have used planarian flatworms , which are considered masters of regeneration , as a model to study how adult stem cells differentiate into more specialized cells . In particular , the experiments explored how the flatworm’s adult stem cells ( which are called neoblasts ) develop into the epidermal cells that form the equivalent of the worm’s skin . Tu et al . show that when a neoblast becomes a mature epidermal cell , it has to undergo multiple transition steps . Slightly different genes are expressed during each step , but a gene called egr-5 controls the expression of all of these marker genes . The egr-5 gene is highly expressed when cells start to develop into epidermal cells . Reducing this gene’s activity blocks the cells from differentiating properly , meaning that they do not form mature epidermal cells . The loss of new epidermal cells causes a disruption in the overall integrity of the worm’s outer surface and this triggers a wound response throughout the whole animal . The neoblasts in turn respond by proliferating excessively and generating other differentiated cells such as neurons and gut cells . However , without egr-5 , the flatworms still cannot make new epidermal cells and they ultimately die . The findings highlight that the development of epithelial cells in this relatively simple organism is much more complicated than suspected . In the future , it will be important to understand how the egr-5 gene controls the proper differentiation and maturation of epidermal cells and whether these mechanisms are conserved in other animals .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2015	Egr-5 is a post-mitotic regulator of planarian epidermal differentiation
Seizures are a disruption of normal brain activity present across a vast range of species and conditions . We introduce an organizing principle that leads to the first objective Taxonomy of Seizure Dynamics ( TSD ) based on bifurcation theory . The ‘dynamotype’ of a seizure is the dynamic composition that defines its observable characteristics , including how it starts , evolves and ends . Analyzing over 2000 focal-onset seizures from multiple centers , we find evidence of all 16 dynamotypes predicted in TSD . We demonstrate that patients’ dynamotypes evolve during their lifetime and display complex but systematic variations including hierarchy ( certain types are more common ) , non-bijectivity ( a patient may display multiple types ) and pairing preference ( multiple types may occur during one seizure ) . TSD provides a way to stratify patients in complement to present clinical classifications , a language to describe the most critical features of seizure dynamics , and a framework to guide future research focused on dynamical properties . Epilepsy is one of the most common neurological disorders with an estimated prevalence of 50 million worldwide ( World Health Organization , 2020 ) . It is characterized by spontaneously recurring seizures , which are ‘a transient occurrence of signs and/or symptoms due to abnormal excessive or synchronous neuronal activity in the brain’ ( Fisher et al . , 2005 ) . However , there are a vast array of signs , symptoms , and underlying causes of seizures . Thus , despite high prevalence and considerable morbidity and mortality , it has been challenging to characterize , treat , and understand seizures , which prevents the development of reasoned , mechanistic approaches to therapy and improved patient care . Seizure classifications to date have been purely descriptive of empirical data: clinical manifestations ( e . g . focal vs . generalized ) that are based upon the region of brain affected rather than the seizure itself , and visual descriptions of electroencephalogram ( EEG ) waveforms . These classifications have been subjected to numerous revisions . In its latest position paper , the International League Against Epilepsy states: ‘Because current knowledge is insufficient to form a scientifically based classification , the 2017 Classification is operational ( practical ) ’ ( Fisher et al . , 2017 ) . In effect , that classification is based upon the epilepsy phenotype—the clinical symptoms that arise during a seizure . Over the past several years , a separate approach has arisen: investigating the genotype of specific epilepsies ( McGovern et al . , 2013; Epi4K Consortium et al . , 2013 ) , which may lead to more informed treatment decisions that match deficits with mechanisms . And for decades , intractable epilepsy has been treated with epilepsy surgery , which relies upon the high spatial resolution of imaging and implanted electrodes to find a seizure focus . These approaches are based upon our available tools: clinical expertise , genetics , imaging , pathology , and surgery . However , seizures are by definition dynamic phenomena , and none of these tools characterize the fundamental dynamics of seizures . In this work , we introduce an organizing principle of seizure dynamics based on nonlinear dynamics and bifurcation theory . Bifurcations are sudden qualitative changes in behavior , including onset and offset of oscillations . Here , we introduce the term ‘dynamotype’ to describe a seizure’s composite , observable , dynamic characteristics in electrophysiological recordings comprising seizure onset and offset . Together , dynamotype , phenotype and genotype provide a rich , multifaceted description of the dynamics , clinical manifestation , and underlying pathology of a seizure . The organization of seizures along dynamotypes leads naturally to a Taxonomy of Seizure Dynamics ( TSD ) providing practical , objective metrics for classification . As the periodic table of elements is a tabular display of chemical elements arranged according to proton number and electronic configuration , TSD is a tabular arrangement according to bifurcation type of seizure onset and offset . Furthermore , the organization of the periodic table can be used to derive relationships between the various element properties and predict chemical properties and behaviors of undiscovered or newly synthesized elements . Here , we explore the capacity of TSD to fulfill this second functional part of the analogy also and demonstrate the existence of all dynamotypes in human epilepsy . We then discuss TSD in the context of a canonical model in nonlinear dynamics and identify the relations amongst the seizure dynamotypes . TSD is available for immediate transfer to clinical practice , providing a rational method of characterizing seizures and subsequently a better understanding of the underlying principles governing seizure generation and termination . The basis of TSD is the observation that seizures are characterized by abrupt changes in the EEG waveform at seizure onset and offset , which we interpreted as bifurcations known from dynamic system theory ( Lopes da Silva et al . , 2003a ) . As a seizure evolves , the brain moves from a normal state into a seizure and back again . Recent work has focused on describing these transitions with empirically chosen visual patterns and has found interesting relationships with underlying pathology ( Perucca et al . , 2014 ) , surgical outcome , ( Jiménez-Jiménez et al . , 2015; Lagarde et al . , 2016 ) and sudden unexpected death in epilepsy ( Rajakulendran and Nashef , 2015 ) . However , these transitions can also be described more rigorously and mathematically as bifurcations . Bifurcations represent qualitative changes that both define and constrain the system dynamics ( Strogatz , 2015 ) . The concept has been used to understand neuronal firing: when a neuron goes through a bifurcation , the emergent dynamics often comprise a new set of behaviors such as multiple stable fixed points ( rest activity ) or limit cycles ( oscillatory activity ) ( Izhikevich , 2000 ) . When a neuron oscillates quickly about a limit cycle , it produces fast repetitive activity known generically as bursting , which is a periodic wave form with periods of oscillatory spiking and periods of quiescence . These dynamics can be described by a set of differential equations and accompanying parameters . The concepts about periodic wave forms with alternating periods of oscillations and quiescence have been extended to EEG wave forms in seizures ( Jirsa et al . , 2014 ) , where oscillatory ( ictal ) states and quiescent ( non-ictal ) states alternate . Two variables are the minimum necessary to generate oscillations , and in systems with two variables , there are only six types of bifurcations involved in bursting ( Figure 1 , for further details also see Saggio et al . , 2017 ) . Four can be used to enter the bursting regime , and another four to exit , giving a total of 16 possible dynamotypes ( Izhikevich , 2000 ) . A key benefit of this organizing principle is that it unambiguously identifies the invariant properties of individual events , which may provide mechanistic insight into the underlying causes and response to specific interventions . It also provides a model that not only accounts for the effects of noise on the system ( Suffczynski et al . , 2005 ) and multistability ( Lopes da Silva et al . , 2003b; Milton , 2012 ) , but also generates a time series . Generalizing this to epilepsy , ( Jirsa et al . , 2014 ) proposed the existence of 16 theoretically possible dynamotypes ( i . e . seizure types ) , and found one seizure offset bifurcation that was present across multiple species , brain regions , and pathologies , including a small cohort of humans . Based on that initial work , we now expand and present a taxonomy of seizure dynamotypes . In the following , we begin with the definition of the different types of seizures based on dynamics at their onset and offset . Then , we show that seizures recorded from different centers in the world can be rigorously classified , and how classification can be performed in daily clinical practice . Next , we introduce a canonical model in nonlinear dynamics ( Saggio et al . , 2017 ) with two important properties: 1 ) the model is canonical , which means that under certain mild conditions ( see Materials and methods ) the behaviors of other models of arbitrary physiological detail can be represented and explained by the canonical model; 2 ) the canonical model captures all dynamotypes in a single mathematical representation . Transitions between types can be obtained through an ultra-slow modulation of the model’s parameters providing a map of the parameter space , which systematically predicts relations between dynamotypes , including a hierarchy across dynamotypes . We demonstrate from a large repertoire of empirical data that patients navigate this seizure map to express the different types of seizures . Finally , we discuss how TSD can be used in a wide range of novel applications in clinical care and research . The goal of this work is to characterize seizures by their underlying onset and offset dynamics , which depends upon identifying reliable , canonical dynamic features . While the dynamics of a single neuron have already been described ( Izhikevich , 2000 ) , linking that behavior to a seizure generated by millions of neurons is complex . We chose to analyze the EEG signal from standard intracranial electrodes , as it is the most clinically relevant and widely studied method to measure brain dynamics . These patterns , which are visible within the EEG waveform , identify the bifurcations that define the invariant properties , and thus the first rigorous classification of seizure dynamics . Figure 1 demonstrates these different bifurcations , showing how the signal changes in terms of amplitude and frequency of successive spikes and may contain a shift in the baseline as the seizure starts or stops . Of note , in dynamical terms , a ‘spike’ is defined as any prominent sharp transient associated with the dynamical process . For human EEG , we assume this includes all fast transients < 200 ms with amplitude that is distinguishable from the background . Of note this dynamical definition also includes the fast , low amplitude spiking seen at the beginning of some seizures . In this work , we present algorithms to measure the invariant properties , which can then be used to classify the seizure types . While the theory behind this classification has been proven mathematically ( Kuznetsov , 2004 ) , measurement of these values under real conditions is challenging . This is because 1 ) EEG recordings of the brain are much more complex than single bursting cells , 2 ) EEG is notoriously noisy , and 3 ) there is limited understanding of the underlying physiology that produces the EEG waveforms ( Einevoll et al . , 2013; Reimann et al . , 2013 ) . Despite these limitations , we previously found strong evidence that at least one dynamotype exists across multiple species ( Jirsa et al . , 2014 ) . Herein , we present both an automated algorithm and a visual method to analyze these noisy data . We find that visual analysis is quite reliable and often preferable under clinical conditions , as demonstrated by recent work in other noisy neural signals ( Haddad and Marder , 2018 ) . Therefore , while we do present the algorithm results as validation , the final clinical analysis is based upon the visual classifications . We analyzed seizures from 120 patients recorded on intracranial EEG in seven centers worldwide ( Appendix I . 1 ) ( Ihle et al . , 2012; Cook et al . , 2013; Kanazawa et al . , 2015; Wagenaar et al . , 2015 ) to identify the bifurcations at onset and offset . All patients had focal onset seizures . Results are shown in Figure 2 . Sixteen seizure dynamotypes– The preceding data validate that at least three types of onset and offset are systematically present in human focal epilepsy . As detailed above , real clinical data are challenging: the lack of DC shift makes it difficult to distinguish some bifurcations , and the noisiness of EEG is hard to distinguish from arbitrary dynamics . Nevertheless , these results show robust evidence that human seizures conform to both the onset and offset bifurcations predicted by our framework . These combinations lead to a taxonomy containing 16 dynamotypes of electrographic seizures ( Jirsa et al . , 2014 ) . We identified the dynamotypes in patients with DC recordings . Two patients did not achieve reviewer consensus ( one onset , one offset ) , leaving 49 patients ( Figure 2F ) . We identified 12 different dynamotypes , with the limitation that several of the dynamotypes cannot be fully distinguished in the absence of a DC shift ( e . g . SH ( -DC ) – SNIC offsets ) . The taxonomy was dominated by seizures with either SN or SubH onsets and slowing ( SH-SNIC ) or arbitrary dynamics ( FLC ) at the end . In this cohort of focal onset seizures , the SupH and SNIC onsets were less common , accounting for all four dynamotypes that were absent . We compared all available clinical metadata from patients with their dynamotype and found no correlation between seizure type and patient gender , pathology , or localization . There was a correlation with age , as older patients tended to have more SupH onsets ( Appendix I . 9 ) . We also compared these results with a prior visual classification that identifies seven basic seizure onset patterns ( Perucca et al . , 2014 ) , and found 6/7 patterns without any apparent relationship to clinical data or pathology ( Appendix I . 10 ) . There were no significant similarities between the dynamotype and the visual classification . While analyzing this dataset , we noted that one patient had two consecutive seizures belonging to different types: one supH/supH and one supH/SH , raising the possibility that an individual may express different types of seizures . This finding was surprising , as many clinicians assume that a person’s seizure should be ‘stereotyped’ , that is consistent over time . In fact , multiple medical devices have been designed under the presumption that a patient’s seizures would have similar appearance over time ( RNS System in Epilepsy Study Group and Morrell , 2011; Cook et al . , 2013 ) . To test whether individuals display different types of seizures over time , we used a unique dataset from Melbourne in which patients had intracranial EEG recorded continuously for many months ( Cook et al . , 2013 ) . We analyzed over 2000 seizures from 13 patients . Given the size of the sample , we limited the analysis to the most straightforward metric: the ISI at seizure offset to determine whether there was slowing at seizure termination . This allowed us to differentiate the SH/SNIC from the supH/FLC bifurcations ( i . e . slowing-down or constant ISI ) . There were 658 seizures of sufficient length ( >25 s ) to measure the offset ISI . To be conservative , we only classified seizures as slowing-down or constant if such a determination was unequivocal and labeled the rest as ‘not assessed , ’ meaning that the determination was not readily evident on brief visual inspection . As seen in Figure 3 , all 13 patients expressed at least two offset patterns . Note that this is likely an underestimation of the heterogeneity in seizure types: these recordings did not contain DC coupling so onset bifurcations were not assessed , and we did not distinguish the offsets into all four types . Nevertheless , we can unambiguously conclude that individuals have seizures from different dynamotypes over time . The clinical data above show that seizures can be classified based on their onset/offset bifurcations and that a patient’s seizures may display multiple dynamotypes . To gain a deeper understanding of the relations between dynamotypes , we formalize these findings within a single unifying mathematical framework , which can account for all these behaviors . During seizures , the firing activity of neurons becomes organized , enabling the emergence of oscillatory activity that can be observed in electrographic recordings . This greatly reduces the degrees of freedom necessary to describe the observed activity , that is a small number of differential equations are sufficient to describe the collective behavior ( Figure 4A ) . We here consider a system with the minimum number of variables necessary to produce oscillatory activity , two . Based on the parameter values , two states can be distinguished: resting ( fixed point ) or oscillatory ( limit cycle ) . When these two states coexist for the same range of parameter values ( bistability ) , transitions between them can be promoted by noise if the system is sufficiently close to a bifurcation ( Lopes da Silva et al . , 2003a; Kalitzin et al . , 2010 ) . However , the statistics of ictal durations ( Suffczynski et al . , 2006 ) points to the existence of a deterministic process governing this transition for seizure offset and possibly for the onset . This can be achieved with the addition of a third variable acting on the timescale of ictal duration . We previously validated this approach with the ‘Epileptor , ’ a set of five differential equations able to account for the most dominant dynamotype ( Jirsa et al . , 2014 ) ( SN/SH , also known as ‘square-wave’ bursting Rinzel , 1987 ) . In the Epileptor , the transition from ‘normal’ to seizure state and back again , as well as the seizure dynamics , are controlled by a collective permittivity variable that evolves on a slow time scale . However , the Epileptor accounts in principle for only a single dynamotype and is not sufficient to explain the data presented above ( note that systematic parameter variations also show a range of other bifurcations El Houssaini et al . , 2020 , although these variations are model-specific and not canonical ) . The fact that individuals can express different types of seizures over time leads to two predictions that must be included within a model of human seizure dynamics: different dynamotypes must coexist in the same model , and there must be an endogenous mechanism by which the brain can transition slowly between dynamotypes . Addressing these predictions within the framework of bifurcation analysis provides the entry point to propose a general taxonomy of seizure dynamics and postulate the existence of an ultraslow modulation . Previous mathematical work demonstrated that the procedure to build a minimal model for the SN/SH type ( Golubitsky et al . , 2001 ) provides a two-dimensional map for the parameter space of the fast variables on which all the six bifurcations can be placed ( Dumortier et al . , 1991 ) . In effect , the map is a representation of the range of states in which a brain region can exist , oscillatory ( ictal ) and non-oscillatory ( interictal ) , and the transitions between them . The oscillatory state produces spiking activity that is described by the fast variables ( millisecond scale activity ) . However , on a slower time scale of the order of seizure length , the brain can move toward a transition to an interictal state , as described by a slow variable . Migrations to different locations on the map can occur on a usually even slower timescale ( 10’s-1000’s of seconds ) , which we call here ‘ultraslow’ . Saggio et al . showed that the use of an ultraslow variable allows full exploration of the map ( Saggio et al . , 2017 ) . Applying these general mathematical principles to epilepsy implies that any brain region able to generate SN/SH seizures can potentially generate other types by navigating to different dynamical regimes ( i . e . changing the parameters of differential equations ) . That work also showed that a large number of physiological neuron and neural population models can be mapped upon a canonical dynamic model under certain mild conditions ( existence of a Bogdanov-Takens point ) . All physiological parameters are then absorbed in only three generic parameters , which span a three-dimensional parameter space , in which all bifurcations are represented . Detailed bifurcation analysis reveals that all neighborhood relations between bifurcations can be displayed ( without loss of generality ) as projections onto the spherical surface within the parameter space , yielding a canonical 2D map ( Saggio et al . , 2017 ) , shown in Figure 4B–C . This map displays the basic topology of all possible relations between bifurcation lines , including identity of the bifurcation and the organization of its nearest neighborhood including proximity and intersection of bifurcations . As one ( or multiple ) of the parameters is continuously varied , trajectories are traced out in this map , eventually connecting two bifurcation lines and thus establishing a seizure’s dynamotype . The navigation of the canonical 2D map ( that is a flat projection of a spherical surface ) through continuous parameter changes selectively generates dynamotypes and justifies that we call it a seizure map . The state of a brain region at any moment can be represented as a location on the map , which defines its dynamical properties . Regions in the map that correspond to different regimes , including quiescent and ictal states , are separated by bifurcation curves . Seizures are represented as black arrows , each arrow corresponding to one dynamotype . To produce a seizure , the system , which is initially in the quiescent state within the bistability region , heads toward the onset bifurcation curve . When this curve is reached , the quiescent state disappears and the system is forced to go into the oscillatory seizure regime within the bistability region . This transition in state causes an inversion in the trajectory of brain state , with the system now heading toward the offset bifurcation curve . When the offset is reached , the system goes back to rest and inverts direction again . Movement along the black arrow is produced by slow ( of the order of the ictal length ) mechanisms leading to seizure offset . Note that the movement towards the onset and offset bifurcations at this timescale occurs in both cases from within the bistability region . Ultraslow movements on the seizure map are responsible for changing the location of the brain state while at rest ( as may happen during the night and day cycle ) and enable the expression of different types of seizures as observed clinically . This framework thus provides a potential explanation for the clinical observation of multiple types of seizures in a single patient , and the seizure map provides a hypothesis to describe how a patient’s current state ( i . e . location on the map ) can affect seizure dynamics ( whether a seizure is likely to occur , and what type is most likely ) . Figure 4C depicts paths ( black arrows ) for seven of the 16 dynamotypes placed on a two-parameter map . Adding one additional parameter allows this map to be extended and create seven other types in three dimensions , while the final two types require even higher dimensions to create ( Saggio et al . , 2017 ) . These higher dimensional types require very fine parameter tuning , and thus are less likely to occur ( Golubitsky et al . , 2001; Saggio et al . , 2017 ) . TSD does not predict the likelihood of dynamotypes , but in conjunction with the seizure map and choice of slow dynamics , a hierarchy of seizures can be established , which is supported by our clinical data ( Appendix II . 5 ) , for instance the dynamotypes that occurred the most ( e . g . SN/FLC and SN/SH ) were predicted to be among the most likely to occur . It is important to distinguish the two types of fluctuations within the brain map . The slow permittivity variable affects the general brain state , or position on the map , on the scale of minutes to hours . It represents underlying , and sometimes varying , conditions of the system than determine the position in the map , which has broad physiological implications . There are also fast fluctuations on the scale of ms to s , better described as perturbations of the brain state from its current location on the map . These perturbations are modeled as ‘noise’ within the model , but in reality they also include many physiological phenomena such as afferent signals and neural potentials—in effect anything that perturbs the system ( Jirsa et al . , 2014 ) . Thus , although the model refers to the addition of ‘noise’ to the system , these effects can readily be attributed to physiological neural activity . Both types of fluctuations can lead to seizures by pushing the system across the bifurcation . It is important to note that the topology of the map in Figure 4 was initially proven mathematically to be generic and rigorously valid for bursting ( Dumortier et al . , 1991; Baer et al . , 2006 ) . This invariance establishes the ground truth to define the relationships between the different bifurcations in the proximity of the SN/SH type , which leads to a key prediction: transitions between certain types may be more common due to their proximity on the map . For example , considering the bistability region in the upper part of the map , we note that the offset curves of SH and SupH approach until they meet . When the curves are very close , even small fluctuations in the parameters can cause a transition between types . If fluctuating internal conditions allow individuals to move around these regions of convergence , patients may have seizures belonging to different types over time , as observed in our longitudinal analysis . The model predicts that transitions between specific types are more likely to occur if these types are close in the map , in the sense just shown of bifurcation curves approaching each other within the same bistability region . On the contrary , transitions between types belonging to distant bistability regions require stronger changes in the ultraslow permittivity variable ( s ) and are thus less likely to occur . Two dynamotypes are paired when they share the same seizure onset or the same offset , and a continuous change of parameters in the seizure map can lead from one type to the other , sometimes even during a seizure . The seizure map predicts possible pairings as motions toward different bifurcations while a seizure is ongoing . As proof of concept , we found several examples of such fluctuations in our cohort . In Figure 5A , one patient’s seizure had constant ISI and square-root amplitude scaling for approximately 70 s , properties exhibited when approaching the SupH bifurcation . The seizure appeared to be terminating , but then it abruptly restarted to terminate with slowing-down ISI and constant amplitude ( SH/SNIC bifurcation ) . We found five examples of this behavior in our data ( out of >2000 seizures ) and reproduced it with our model . By definition , the dynamotype includes only the onset and final offset bifurcation , but the behavior during this seizure is intriguing and can be explained by the model . We considered a path for the SN/SupH type with an ultraslow drift of the offset point that changed the path to SN/SH and added noise to all variables to simulate fluctuations . With these settings , we ran 100 simulations , which generated several different dynamotypes , predominantly SN/SH , SupH/SH , and SN/SupH ( Appendix III ) . Several had transitions during the seizure from one bifurcation to another , 10 of which in the same manner as the data in Figure 5A , switching from SupH to SH offset . Thus , the clinical example is one of the most favorable combinations within the model , as the SupH and SH bifurcations are so close that small fluctuations can cause the switch . Status epilepticus: Explorations of the seizure map also demonstrated another effect sometimes seen clinically: status epilepticus . Simulations with the previous settings in some cases produced continuous seizures that did not resolve by the end of the simulation , equivalent to status epilepticus ( El Houssaini et al . , 2015 ) . We analyzed the corresponding trajectories on the map to determine how this had occurred . Prolonged seizures occurred when the brain state crossed the SN onset curve but was unable to return to rest through the offset bifurcation and remained mainly in the violet ‘seizure only’ region in Figure 5G . The slow variable naturally drives the state toward offset , but in these cases was continually overridden by noise , causing the state to ‘escape’ from the bistability region . We then analyzed this effect by simulating various levels of noise and showed that there was a clear correlation between the noise variance and the likelihood of entering status epilepticus ( Appendix IV ) . We compared these results with our clinical data , which had two examples of non-convulsive status epilepticus . In both cases , the seizures began in typical fashion , but instead of terminating began having long periods of constant ISI with varying periods of amplitude fluctuations . There were many abrupt transition periods during which the ISI and amplitudes became arbitrary . After these brief periods of disorganization , the dynamics returned to constant ISI . We compared the dynamics of our model results with these human seizures and found that the transition between different dynamics is quite similar . In Figure 5D , we show a portion of two human seizures and one example of the simulation ( Figure 5E–F ) and movement on the map ( Figure 5H–I ) . Further demonstration of the patients’ status epilepticus is provided in Appendix IV . The patterns of organized- alternating with disorganized- firing , often known clinically as ‘waxing and waning seizures , ’ are entirely consistent with the model: the seizure undergoes periods in which it progresses toward termination , then because of noise it reverts to a point farther away from the offset bifurcation , as described previously ( Kramer et al . , 2012 ) . Accelerating seizure: We then analyzed the seizure offset that increased in frequency described in Appendix V . In this case , we explored conditions on the map that could produce ‘speeding up’ at the end of the seizure . We identified multiple trajectories in the map of brain states capable of producing these unusual seizure dynamics . As in the case of status epilepticus , these unusual patterns are dependent upon the relative position within the brain map , in this case occurring when brain states along a trajectory are affected by multiple bifurcations that are in close proximity ( Appendix 1—figures 17–19 ) . These results demonstrate the explanatory value of the seizure map and show how it provides a rational explanation for a wide range of physiological dynamics . Seizures have been recognized clinically for millennia , but after nearly a century of electrographic recordings we still do not have a translatable method of characterizing their dynamics . We here address this issue and provide the first principled approach toward the organization of seizures in a Taxonomy of Seizure Dynamics ( TSD ) . TSD establishes 16 dynamotypes of seizures , which could be extended to more exotic dynamotypes when considering non-planar bifurcations ( see Appendix II . 6 ) . As TSD provides the classification of seizures , it remains completely unbiased to each seizure type . This invariance is broken by the seizure map , which establishes relations between dynamotypes and introduces a bias in the taxonomy , laying the grounds for a hierarchy of dynamotypes . The hierarchy is based on the mathematical consequences of how bifurcations are related to each other ( Saggio et al . , 2017 ) . The relations can be considered as structural in the sense that they rely on the static properties of location , shape , branching and topology of bifurcation curves in the seizure map . The implications are functional in the sense that they determine the non-static properties of a seizure’s discharge patterns including frequency , acceleration/deceleration , amplitude and amplitude growth . As such TSD and the seizure map provide another example of the ubiquitous link of structure and function in biology . Our classification aims at precisely identifying the seizure type in terms of dynamics , without any dependence upon specific symptoms , pathology , or localization . Thus , it is highly complementary to the classical operational classifications used by clinicians to diagnose and treat patients , which are based upon those factors without addressing dynamics ( Fisher et al . , 2017 ) . The dynamotype describes the behavior of the seizure itself , while the clinical classification describes the patient’s symptoms: together , both classifications are synergistic and can be used to improve patient stratification , providing more insight into diagnosis and treatment . TSD is based on simple , invariant , objective metrics that have compelling scientific rationale . With DC-coupled recordings , it is possible to distinguish the types with high fidelity , even with visual inspection . This method is thus readily available to clinicians , as many standard EEG acquisition devices now have excellent resolution near DC ( <0 . 1 Hz ) . Our interpretation of the results relies on some key assumptions . First , we are assuming that the onset and offset of seizures are brought about by bifurcations . Another common mechanism in the literature is noise-induced transitions ( Lopes da Silva et al . , 2003a ) , which our model can reproduce ( see Appendix II . 3 for a discussion of how this would affect our classification ) . Second , we rely on the assumption of timescale separation between the dynamics of the spikes within seizures and the slower dynamics controlling seizure threshold and termination ( both in theory and in the simulated sample of data ) . If this assumption does not hold , different phenomena could occur and the scaling laws could be impossible to identify . Third , we only considered planar bifurcations for simplification ( Appendix II . 6 ) . These assumptions were the axioms for developing the theory and data analysis . Future work will address the validity and consequences of these simplifying assumptions . Similar to sleep , seizures are universal from insects to humans , leading to the proposal that seizures are an inherent property of a brain , that is they are endogenous to the brain , perhaps as an emergent property of complex neuronal networks ( Jirsa et al . , 2014 ) . This may explain why , despite the vast range of genetic , structural , chemical , and developmental conditions that cause epilepsy , seizures have a remarkably limited set of dynamical behaviors . It is therefore not surprising that elementary mathematical laws can describe their electrophysiological signature . However , clinical interpretation of seizure dynamics has been almost universally based on simple observation , in which clinicians report the frequency and morphology of spikes . This method is helpful to identify primary generalized epilepsies , but within focal seizures has limited clinical use . TSD by itself does not have any bias either regarding the dynamotype , but when linked to the seizure map shows certain dynamotypes to be more prevalent than others . Empirically , low voltage fast activity ( Wetjen et al . , 2009 ) and focal DC shifts Ikeda et al . , 1999 have been found to be highly predictive of the true seizure focus , which are both patterns corresponding to common dynamotypes , suggesting clinical utility . One limitation of previous clinical descriptions of seizures is that it has been unclear which dynamical features are relevant . There is high variability in the frequency and morphology of spikes due to individual fluctuations and noise , as seen in spontaneous seizures recorded in humans and experimental models ( Jirsa et al . , 2014 ) . Our analysis identifies the invariant dynamics that impose important constraints on the system . A crucial aspect of our approach is that it allows us to disentangle characteristics that are necessary to describe the dynamics from other seizure related phenomena that are not fundamental to our simplified model ( e . g . spike and wave complexes , preictal spikes , sentinel spikes ( see Appendix I . 2 ) ) . This is not to suggest that such biomarkers are not relevant to epilepsy: they have well-known correlation with the epileptogenic zone ( Conrad et al . , 2020 ) and can predict the occurrence of the first spontaneous seizure during experimental epileptogenesis ( Chauvière et al . , 2012 ) . The Epileptor model , which comprises the SN-SH dynamotype , generates both spikes and seizures ( Jirsa et al . , 2014 ) . The spikes are important indicators of the organization of the networks , but not part of the generic features of onset and offset dynamics and thus not appear in the generic model ( Saggio et al . , 2017 ) . If additional mechanisms are introduced based on other forms of reasoning , as in Jirsa et al . , 2014 via a second population , spikes can be included in the dynamics . It is important to note that we only analyzed the particular case of drug-resistant epilepsies investigated with invasive intracranial recordings; however , it is unlikely that this theory is specific to such epilepsies as seizures forced in non-epileptic networks follow the same universal rules ( Jirsa et al . , 2014 ) . We note that prior work focusing on individual bifurcations are all also entirely consistent with TSD , which encompasses all these onset and offset possibilities and further shows how they interact ( Appendix VII ) . This work does not include data from generalized onset epilepsies , as these are not typically recorded with intracranial EEG . Our taxonomy is , however , fully consistent with past work on generalized epilepsy dynamics ( Wendling et al . , 2016 ) . Absence seizures , for instance , begin with sudden onset and offset of ~3 Hz large amplitude oscillations without a DC shift ( Slaght et al . , 2004 ) and terminate abruptly without slowing down to zero , which would point to a SubH/FLC type ( which is the most likely dynamotype ( Golubitsky et al . , 2001 ) , see Appendix II . 5 ) . In this work with focal epilepsies , the most common dynamotypes were the SN ( +DC ) /FLC and SN/SH ( which are also likely types ) . When comparing our results with a past visual classification system of spike frequency ( Perucca et al . , 2014 ) , we found no correlation with pathology in our cohort of 120 patients . However , our cohort did not have any patients with tuberous sclerosis , which was the only pathology associated with burst suppression in that prior work . Combining the data from both studies , the different patterns appear to be either evenly distributed or too rare to find robust correlations with pathology . Similarly , dynamotypes are not strongly correlated with pathology . In terms of the seizure map , we hypothesize that what determines the seizure dynamics is not the pathology per se , but the location of the brain on the map . Specific pathologies may predispose to certain regions , but there are many complex dynamics affecting brain state and many conditions that can produce similar dynamics . This coincides with the idea of the seizure map showing the full range of potential seizure onset and offset activity . There is great clinical and research potential in characterizing a seizure’s dynamotype , as it provides a unique perspective on brain networks . The current standards of epilepsy care focus on phenotype , genotype , and the time/location of seizure onset . While those methods have obvious utility , they do not address the underlying dynamics and thus have left several questions unanswered for decades . How do seizures start , stop and spread ? How do we tell the difference between inter- or pre-ictal spiking and seizure initiation ? Is it possible to measure the distance to seizure threshold , that is determine seizure risk at a given moment ? How do we compare two different seizures ? Is it possible to measure if a treatment is working by moving the brain ‘farther away’ from seizure onset , rather than waiting to see if seizures recur ? These questions all require an understanding of the dynamics—an understanding that is not addressed by the current clinical tools . This is where the utility of the dynamotype is manifest . At its most basic , the dynamotype is a quick description of the key dynamics of a particular seizure , a clinical language that focuses on the aspects that are most important . This would supplement current visual descriptions , which typically are limited to amplitude and frequency . But there are many deeper applications of this tool as well . We have previously demonstrated that very different biophysical mechanisms can produce the same dynamotype ( Jirsa et al . , 2014 ) . Here , we show that seizures from 120 patients contain almost the entire taxonomy of dynamotypes , and that a wide array of focal pathologies can be grouped into similar dynamotypes . Our interpretation is that this is because the seizures depend heavily on ‘local’ dynamics , that is the current brain state ( location on the map ) and acute perturbations ( noise in the system ) , more than that a single pathology would necessarily predispose to a specific location on the seizure map . There are many other potential applications for TSD in basic research as well . For instance , our group recently published an analysis quantifying how epileptogenesis progresses in the tetanus toxin model in rats ( Crisp et al . , 2020 ) . That work showed that the dynamotype evolved over time , starting with SN and moving into SNIC ( Appendix 1—figure 4 ) and sometimes SupH onsets over the course of weeks . A clinical trial is currently underway in France using the SN-SH dynamotype to model seizure foci and spread ( HBP , 2018 ) . Future versions of such tools could utilize the whole taxonomy to be much more comprehensive , tailoring models to the key underlying dynamics of specific patients . These models would greatly enhance modern network analytic tools ( Stacey et al . , 2020 ) , which would be greatly enhanced with a rational model to describe the underlying dynamics . One novel aspect of dynamotype is that understanding the underlying dynamics can help in the design of strategies to control seizures , such as with electrical stimulation ( Kalitzin et al . , 2010 ) . Studies on neuronal bursters ( Izhikevich , 2000 ) , which are organized in similar dynamic types , demonstrate that types have different sensitivity to stimulation . For example , SubH onset acts as resonators , which require a resonant frequency in the stimulus to trigger oscillations , while SN onset behaves as an integrator in which the nature of the stimulus ( excitatory or inhibitory ) rather than the frequency plays a key role ( Izhikevich , 2000 ) . There is a long history of using perturbation to probe the proximity of a nearby bifurcation in disciplines such as electrical power ( Chow et al . , 1990 ) and reservoirs ( Heppell et al . , 2000 ) . Past work on stimulation to assess epileptogenicity ( Alarcón , 2005; Kalitzin et al . , 2005; David et al . , 2010 ) is similar to such work and would be greatly enhanced with the insight gained from this model to understand the nearby bifurcations . There is a long history of bifurcation research in other fields that may also be applicable to seizures , such as using perturbations to assess proximity to a SubH or SupH bifurcation ( Bryant and Wiesenfeld , 1986; Vohra et al . , 1994; Yaghoobi et al . , 2001 ) . Further theoretical and clinical work is necessary to assess whether knowledge of the dynamotype could also improve the ability to abort seizures with tailored stimulation . The second important prediction is that the synchronization properties of coupled bursters are bifurcation-dependent ( Wang et al . , 2011; Reimbayev and Belykh , 2014; Belykh et al . , 2015 ) . This is a key issue for seizure propagation , as it predicts that the ability of a seizure to spread is dependent upon its type . Since the spatiotemporal organization of the seizure is part of the data features used to personalize brain network models ( Virtual Epileptic Patient Jirsa et al . , 2017; Proix et al . , 2017 ) and functional connectivity based approaches ( Hutchings et al . , 2015; Sinha et al . , 2017; Taylor et al . , 2017 ) , the choice of the right dynamotype is critical for successful patient modeling and clinical translation . Another significant contribution of this work concerns the dynamics of the slow permittivity variable to explain how slow changes in the behavior/state of a brain region can bring it closer or farther away from different bifurcations , that is seizure threshold . The fact that all 13 patients had seizures belonging to at least two types implies that the permittivity variable moves on a dynamical map in which different types of bifurcations can occur . Each parameter of the map should be considered as a representation of a manifold of physiological variables that cooperate to produce a particular change in the system . Given the slow timescale at which these changes occur , neurochemical substances ( e . g . hormones , neuromodulators etc . ) are the best candidates . Within the permittivity variable we can here distinguish two timescales: a slow timescale of the order of the ictal length , and an ultra-slow timescale of the order of the interictal length ( hours , days , months , years ) . Typical examples include the circadian regulation of seizures ( Karoly et al . , 2017 ) and catamenial epilepsy . Interestingly , both males and females display ultraslow ( weeks ) modulation of seizure probability both in rats ( Baud et al . , 2019 ) and humans ( Karoly et al . , 2017; Baud et al . , 2018 ) , further suggesting that these results are species- and sex-independent . Those results and ours strongly support the proposal that patients move closer and farther away from seizure threshold ( i . e . ‘travel the map’ ) during their lifetime . This interpretation may also be helpful in assessing a brain’s current proximity to seizure bifurcations , that is predict the risk of seizures occurring . Several features are altered when nearing the onset bifurcations , such as preictal spikes ( Jirsa et al . , 2014 ) , variance of the signal ( Meisel et al . , 2015 ) , and reaction to electrical probing of cortical excitability ( Freestone et al . , 2011 ) . Recent work has shown that interictal discharges act like system perturbations that behave like the slow approach to bifurcations , just as predicted by our model ( Chang et al . , 2018 ) . Seizure forecasting , based on electrographic recordings , is already enhanced when circadian rhythms are used to inform the model ( Karoly et al . , 2017 ) . If the ultraslow physiological correlates of the map’s parameter could be identified , measured , and manipulated , this would open new possibilities to assess when the patient is moving toward unsafe regions of the map and to alter their trajectory , that is control seizures before they occur . Our proposed framework provides an organizational principle of seizure dynamics , which , when linked to canonical dynamic systems , identifies a generic seizure map charting out characteristics of dynamotypes including prevalence of a dynamotype and possible pairings of dynamotypes that can occur during a seizure , as well as others that are prohibited . TSD does not describe all possible seizure features , but relies on seizure onset and offset classification , which is why we defend a complementary approach , combining it with traditional operational classifications . However , it provides a unique avenue to classify seizures based upon their key dynamical features , while providing insight into how seizures become more or less likely to occur at a given time . A corollary is that , although TSD has been developed in the context of seizure dynamics , it likely extends to physiological function of the healthy brain ( e . g . alterations between REM and slow wave sleep , and the appearance of gamma frequencies or ripples during slow wave sleep ) and stipulates the existence of at least two time scales in any theory of the brain . Slow time scales are present in many theories of brain function , but typically have been limited to the domain of learning and adaptation , thus functionally separated from fast processes . Here , the functional integration and co-evolution of the fast neuroelectric and slow permittive time scales suggests emergent and inherent properties of brain processes .	Epileptic seizures have been recognized for centuries . But it was only in the 1930s that it was realized that seizures are the result of out-of-control electrical activity in the brain . By placing electrodes on the scalp , doctors can identify when and where in the brain a seizure begins . But they cannot tell much about how the seizure behaves , that is , how it starts , stops or spreads to other areas . This makes it difficult to control and prevent seizures . It also helps explain why almost a third of patients with epilepsy continue to have seizures despite being on medication . Saggio , Crisp et al . have now approached this problem from a new angle using methods adapted from physics and engineering . In these fields , “dynamics research” has been used with great success to predict and control the behavior of complex systems like electrical power grids . Saggio , Crisp et al . reasoned that applying the same approach to the brain would reveal the dynamics of seizures and that such information could then be used to categorize seizures into groups with similar properties . This would in effect create for seizures what the periodic table is for the elements . Applying the dynamics research method to seizure data from more than a hundred patients from across the world revealed 16 types of seizure dynamics . These “dynamotypes” had distinct characteristics . Some were more common than others , and some tended to occur together . Individual patients showed different dynamotypes over time . By constructing a way to classify seizures based on the relationships between the dynamotypes , Saggio , Crisp et al . provide a new tool for clinicians and researchers studying epilepsy . Previous clinical tools have focused on the physical symptoms of a seizure ( referred to as the phenotype ) or its potential genetic causes ( genotype ) . The current approach complements these tools by adding the dynamotype: how seizures start , spread and stop in the brain . This approach has the potential to lead to new branches of research and better understanding and treatment of seizures .	[ "Abstract", "Introduction", "Materials", "and", "methods", "Results", "Discussion" ]	[ "computational", "and", "systems", "biology", "tools", "and", "resources", "neuroscience" ]	2020	A taxonomy of seizure dynamotypes
We demonstrate that human motor memories can be artificially tagged and later retrieved by noninvasive transcranial direct current stimulation ( tDCS ) . Participants learned to adapt reaching movements to two conflicting dynamical environments that were each associated with a different tDCS polarity ( anodal or cathodal tDCS ) on the sensorimotor cortex . That is , we sought to determine whether divergent background activity levels within the sensorimotor cortex ( anodal: higher activity; cathodal: lower activity ) give rise to distinct motor memories . After a training session , application of each tDCS polarity automatically resulted in the retrieval of the motor memory corresponding to that polarity . These results reveal that artificial modulation of neural activity in the sensorimotor cortex through tDCS can act as a context for the formation and recollection of motor memories . Context influences memory encoding and retrieval , including fear conditioning responses in rodents ( Maren et al . , 2013 ) and declarative memory in humans ( Godden and Baddeley , 1975 ) . Although previous studies have examined whether context-dependency also exists in motor memory , context-dependent motor learning based on contextual cues has proven quite difficult . For example , participants have difficulty adapting identical reaching movements to conflicting dynamical environments ( e . g . , velocity-dependent rightward and leftward force fields ) according to a contextual cue ( e . g . , room color , target color , etc [Gandolfo et al . , 1996; Osu et al . , 2004; Nozaki et al . , 2006; Hirashima and Nozaki , 2012; Kadota et al . , 2014] ) . It is only recently that a wide variety of contexts helpful to create distinct motor memories has been discovered ( Osu et al . , 2004; Nozaki et al . , 2006; Hirashima and Nozaki , 2012; Kadota et al . , 2014; Cothros et al . , 2009; Ikegami et al . , 2010; Howard et al . , 2011; Yokoi et al . , 2011; Howard et al . , 2012 , 2010; Sarwary et al . , 2013; Yokoi et al . , 2014; Howard et al . , 2015; Sarwary et al . , 2015 ) . However , the underlying mechanisms regarding context-dependent motor learning and memory are largely unknown . Notably , contexts that are shown to be useful are often associated with different neural activity patterns of the sensorimotor cortex including the primary motor cortex ( M1 ) and the premotor cortex ( PM ) . For example , distinct motor memories for identical reaching movements can be created depending on whether the opposite arm is stationary or moving ( Nozaki et al . , 2006; Kadota et al . , 2014; Nozaki and Scott , 2009 ) . Consistent with this finding , it has been reported that opposite arm movements alter M1 and PM activity during reaching movements ( Donchin et al . , 2001 , 2002; Cisek et al . , 2003; Ganguly et al . , 2009; Rokni et al . , 2003 ) . In agreement with the significant role of these brain areas in motor learning ( Kadota et al . , 2014; Gandolfo et al . , 2000; Li et al . , 2001; Muellbacher et al . , 2002; Arce et al . , 2010; Orban de Xivry et al . , 2011; 2011; 2013 ) , we hypothesized that , when a motor memory is formed under different activity patterns in the sensorimotor cortex associated with different contexts , a motor memory specific to this particular activity pattern is created . As a result , later reinstating this activity pattern in these areas should lead to automatic retrieval of the corresponding memory . Here , we tested this hypothesis directly using transcranial direct current stimulation ( tDCS ) ( Nitsche et al . , 2008; Orban de Xivry and Shadmehr , 2014; Di Lazzaro and Rothwell , 2014 ) . tDCS to the sensorimotor cortex modulates spontaneous M1 activity and excitability according to its polarity . As a result , the size of the motor evoked potential induced by transcranial magnetic stimulation is increased or decreased by , respectively , anodal or cathodal tDCS during ( Nitsche et al . , 2005; 2007 ) and even after the application of stimulation ( Nitsche and Paulus , 2000; 2001; Siebner et al . , 2004 ) . This suggests that different brain activity patterns ( i . e . specific to the polarity of the stimulation ) could be artificially created by tDCS . We predicted that motor learning performed under different background activity patterns of the sensorimotor cortex created by tDCS would yield to the formation of separate motor memories . Participants were trained to perform reaching movements in the presence of two conflicting force fields while receiving tDCS on the sensorimotor cortex . Critically , training for each force field was always associated with a distinct tDCS polarity . We then examined if , after training , applying a particular tDCS polarity reactivated the motor memory associated with this polarity , despite no explicit contextual change . During the test period , the first eight participants received anodal and cathodal tDCS in alternation , starting with anodal stimulation ( Figure 1g , T-TACAC ) . In this group , evolution of the force output during the test period ( Figure 2a , dashed line and open circles ) indicates the presence of a tDCS effect . As predicted , the force output exhibited a clear polarity-dependent change , and the forces during cathodal stimulation ( blue open circles in Figure 2a ) were larger than the forces during anodal stimulation ( red open circles in Figure 2a ) . Another group of participants ( N = 8 , T-TCACA , Figure 1g ) were trained with an identical protocol but first received cathodal stimulation during the test period . This subgroup also exhibited clear polarity-dependent effects on force output during the test period ( Figure 2a , solid line and filled circles ) , which was consistent with our predictions . In addition , these changes were in anti-phase for the two groups . 10 . 7554/eLife . 15378 . 004Figure 2 . Experimental results during the test period . ( a ) Motor memory evaluated as the force output using the error-clamp trials for the T-TACAC ( open circle ) and T-TCACA subgroups ( filled circle ) during the test period . The positive force indicates rightward force output . Circle color represents tDCS polarity ( red and blue for anodal and cathodal tDCS , respectively ) . The solid and dashed lines are the moving average ( 5 data points ) , and the shaded areas indicate standard errors . Gray vertical bars indicate the period during which tDCS polarity changed . Note that the T-TACAC and T-TCACA subgroups received anodal and cathodal tDCS , respectively , in the first block . ( b ) ΔForce was calculated as the difference between the T-TACAC and T-TCACA subgroups in order to reduce the effect of exponential motor memory decay . The bold , solid line and shaded grey area indicates the mean and standard deviation of the bootstrapped samples , respectively ( moving average calculated over 5 data points ) . ( c ) ΔForce averaged over each block . The mean and standard deviation were obtained from bootstrapped samples . A permutation test was used to test the effect of block order ( *p<0 . 0001 as indicated at the right side ) . A permutation test was also used to compare the values between the first and second , second and third , and the third and fourth blocks . #p<0 . 07; p<0 . 01; *p<0 . 005 . ( d–i ) Results for the S-T group ( d–f ) and T-S group ( g–i ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15378 . 00410 . 7554/eLife . 15378 . 005Figure 2—figure supplement 1 . Definition of factors 'period' and 'block order' . DOI: http://dx . doi . org/10 . 7554/eLife . 15378 . 00510 . 7554/eLife . 15378 . 006Figure 2—figure supplement 2 . Trial-dependent changes in the handle’s peak velocity for the T-TACAC ( open circle ) and T-TCACA subgroups ( filled circle ) during the testing period . The format is identical to Figure 2 ( a ) , ( d ) , and ( g ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15378 . 006 Natural exponential decay in force output observed at the beginning of the test period co-occurs with the polarity-dependent effect of tDCS . We took it into account with two different methods . In the first model-free approach , we tried to eliminate this decay effect by contrasting the force output of the two subgroups ( Figure 2 ΔForce obtained by subtracting the data of the ACAC subgroup from the data of the CACA subgroup ) . This method relies on the assumption that exponential decay was similar in both subgroups . In the second model-based approach , a summation of two exponential curves was fitted to the force output data of each participant during the test period and we analyzed the residuals around the curve in function of tDCS polarity ( Figure 3 ) . This method bears the advantage of being applied on each subgroup separately . 10 . 7554/eLife . 15378 . 007Figure 3 . Modulation of the residuals , with block , during the test period . ( a ) The force output residuals from the exponential fitting of 4 blocks for the T-TACAC ( dashed line and open circle ) and T-TCACA ( solid line and filled circle ) groups . The error bars indicate standard errors . Circle color represents tDCS polarity . A 3-way ANOVA indicates a significant interaction between subgroup and block order . p<0 . 005 . ( b ) Results for the S-T group receiving sham tDCS during the training period . ( c ) Data for the T-S group receiving sham tDCS during the test period . Circle color was set to black for data obtained when sham tDCS was used . DOI: http://dx . doi . org/10 . 7554/eLife . 15378 . 00710 . 7554/eLife . 15378 . 008Figure 3—figure supplement 1 . Calculations for residuals . Force data were fitted using a sum of two exponential curves , and fitted curves were subtracted from the data to obtain the residuals . If there were no block-dependent modulation , the residuals should be evenly distributed around the curve ( a ) . In contrast , if there were modulation , the residuals should change , along with the blocks , around the curve ( b ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15378 . 00810 . 7554/eLife . 15378 . 009Figure 3—figure supplement 2 . Polarity-dependent changes in the aftereffect for the Tffrev-T groups . The format is identical to Figures 2 and 3 . p<0 . 05 . DOI: http://dx . doi . org/10 . 7554/eLife . 15378 . 00910 . 7554/eLife . 15378 . 010Figure 3—figure supplement 3 . The residuals obtained using a single exponential curve instead of two exponential curves . The format is identical to Figure 3 . The results were not substantially different from those obtained using two exponential curves . As for the T-T group results ( a ) , there was a significant interaction between subgroup and block order [repeated measures 3-way ANOVA: F ( 1 , 14 ) = 43 . 82 , p = 1 . 15 × 10–5 , ηp2 = 0 . 76] . For both groups , there was a simple main effect of block order on the residuals [T-TACAC: F ( 1 , 14 ) = 28 . 45 , p = 1 . 05 × 10–4 , ηp2 = 0 . 67; T-TCACA: F ( 1 , 14 ) = 16 . 22 , p = 0 . 0012 , ηp2 = 0 . 54] . As for the S-T group ( b ) , there was no significant interaction between subgroup and block order [repeated measures 3-way ANOVA: F ( 1 , 14 ) = 0 . 040 , p = 0 . 844 , ηp2 = 0 . 0028; simple main effect of block order: F ( 1 , 14 ) = 1 . 759 , p = 0 . 206 , ηp2 = 0 . 11] . Furthermore , the modulation of the residuals by stimulation polarity was larger for the T-T group than it was for the S-T group [repeated measures 4-way ANOVA , interaction between group , subgroup , and block order: F ( 1 , 28 ) = 17 . 94 , p = 2 . 22 × 10–4 , ηp2 = 0 . 39] . As for the T-S group ( c ) , there was a significant interaction between subgroup and block order [repeated measures 3-way ANOVA: F ( 1 , 14 ) = 7 . 641 , p = 0 . 015 , ηp2 = 0 . 35] , and there was a simple main effect of block order only for the T-SCACA group [F ( 1 , 14 ) = 16 . 07 , p = 0 . 001 , ηp2 = 0 . 53] , but the modulation due to polarity was opposite to the predicted pattern . Furthermore , the modulation of force output by stimulation polarity was larger for the T-T group than it was for the T-S group [repeated measures 4-way ANOVA , interaction between group , subgroup , and block order: F ( 1 , 28 ) = 43 . 16 , p = 4 . 00 × 10−7 , ηp2 = 0 . 61] . Regarding the Tffrev-T group ( d ) , there was a significant interaction between subgroup and polarity [repeated measures 3-way ANOVA: F ( 1 , 14 ) = 9 . 202 , p = 0 . 008 , ηp2 = 0 . 37] . While in one group ( Tffrev-TACAC ) there was no clear effect of stimulation polarity during the test period [simple main effect of polarity on the residuals: F ( 1 , 16 ) = 2 . 04 , p = 0 . 173 , ηp2 = 0 . 11] , the other group showed a clear effect of stimulation . In this group ( Tffrev-TCACA ) , there was a simple main effect of polarity on the residuals [F ( 1 , 16 ) = 8 . 20 , p = 0 . 011 , ηp2 = 0 . 34] . This modulation was consistent with what was predicted by the association between polarity and force-field during the training period . As for the PPC group ( e ) , there was no significant interaction between subgroup and block order [repeated measures 3-way ANOVA: F ( 1 , 16 ) = 0 . 996 , p = 0 . 333 , ηp2 = 0 . 058] , and modulation was significantly greater for the T-T group than for the PPC group [interaction between group , subgroup , and block order by repeated measures 4-way ANOVA: F ( 1 , 30 ) = 4 . 903 , p = 0 . 035 , ηp2 = 0 . 46] . DOI: http://dx . doi . org/10 . 7554/eLife . 15378 . 010 Figure 2b represents the trial-dependent change in the ΔForce calculated using a bootstrap method . If the stimulation had no effect on motor memories , the evolution of ΔForce over the test period should be flat . However , we observed a clear modulation pattern ( Figure 2b , c ) : the ΔForce was smaller for the first and third block compared to the second and fourth block . This effect of block order ( Figure 2—figure supplement 1 ) on the ΔForce was statistically significant ( Figure 2c; permutation test: p = 0 . 0009 ) . In addition , this effect appeared stable from the first to the second half of the test period ( permutation test: interaction between period and block order: p = 0 . 313 ) . Importantly , these changes were consistent with the pattern predicted by the training . For example , in the first block , the force output of T-TACAC group should be more leftward than that of T-TCACA group and thus the ΔForce was more negative . The degree of ΔForce modulation from the first to second block , from the second to third block and from third to fourth block ( the value was defined positive if the change was congruent with the predicted change ) was 0 . 57 ± 0 . 32 ( permutation test: p = 0 . 0632 ) , 0 . 73 ± 0 . 21 ( p = 0 . 0044 ) , and 0 . 78 ± 0 . 13 N ( p = 0 . 0001 ) , respectively ( mean ± standard deviation of the bootstrapped samples ) . In our experimental setting , the force output of full adaptation varied from -4 N to 4 N ( viscosity was 10 N/ ( m/s ) and the peak velocity of the handle was 0 . 4 m/s ) . Thus , approximately 7–10% force modulation was induced via tDCS for the T-T group . Given the effect of tDCS for muscular force output ( Salimpour and Shadmehr , 2014; Tanaka et al . , 2009 ) , one might expect that the movement kinematics were also modulated by the polarity of the stimulation . Such modulation might confound the actual effect of tDCS on force output because the perturbation was a velocity-dependent force-field . However , we did not observe any polarity-dependent changes in movement peak velocity [Figure 2—figure supplement 2; interaction between group and block order by repeated measures 3-way ANOVA: F ( 1 , 14 ) = 0 . 374 , p = 0 . 551 , ηp2 = 0 . 024; main effect of block order: F ( 1 , 14 ) = 3 . 076 , p = 0 . 101 , ηp2 = 0 . 18] . This demonstrates that polarity-dependent changes in force output were not caused by hand kinematic modulation due to tDCS . We also characterized the effect of tDCS by examining variation in force output around the natural exponential decay of motor memories during the test period ( model-based approach ) . If the stimulation had no effect on motor memories , the decay of the force output should simply follow this exponential function and the residuals should be evenly distributed around the exponential curve in all blocks ( Figure 3—figure supplement 1a ) . In contrast , if the stimulation polarity influences the force output during the test period , the residuals should be more positive ( above the exponential curve ) during cathodal stimulation and more negative ( below the exponential curve ) during anodal stimulation ( Figure 3—figure supplement 1b ) . Given that the two subgroups received different polarity during each of the blocks , the polarity-dependent change in residuals was opposite between the two groups ( Figure 3a ) . That is , the residuals were always more positive for the group receiving cathodal stimulation during one of the blocks and more negative for the group receiving anodal stimulation during the same block ( Figure 3a ) . The modulation of the force output by tDCS was analyzed with a 3-way ANOVA with subgroup as a between-subject factor and period ( first or second half of the test period , 2 blocks each ) and block-order ( first and second block of each half of the test period ) as a within-subject factors ( See Figure 2—figure supplement 1 for the definitions of period and block order ) . Given that the polarity was opposite across groups in function of the block order factor ( anodal in the first and third block for T-TACAC but in the second and fourth block for T-TCACA ) , the modulation of force output by tDCS resulted in a significant interaction between subgroup and block order [repeated measures 3-way ANOVA: F ( 1 , 14 ) = 39 . 24 , p = 2 . 08 × 10–5 , ηp2 = 0 . 74] . For each subgroup separately , there was a polarity-dependent change in the residuals as indicated by the main effect of block order on the residuals [T-TACAC: F ( 1 , 14 ) = 12 . 70 , p = 3 . 10 × 10–3 , ηp2 = 0 . 47; T-TCACA: F ( 1 , 14 ) = 28 . 04 , p = 1 . 13 × 10–4 , ηp2 = 0 . 67] . This effect indicates that , for each subgroup , the force output varied with the polarity of the stimulation . The modulation of force output with tDCS polarity was also examined when the association between tDCS polarity and force direction was reversed ( Tffrev-T group ) , i . e . , when anodal and cathodal stimulations were associated with leftward and rightward force fields , respectively ( Figure 3—figure supplement 2 ) . While the effect appeared clear in one of the subgroups ( Tffrev-TCACA ) , it was absent in the other one ( Tffrev-TACAC ) . This mixed effect yielded a non-significant effect of block order on the ΔForce because the permutation test was based on the two groups , but there was still a significant change in ΔForce from the second to third block ( Figure 3—figure supplement 2b , c; permutation test: p = 0 . 0127 ) . The polarity dependent modulation was more apparent in the model-based approach because it allowed us to study each group separately . In this case , we found a significant interaction between subgroup and block-order [repeated measures 3-way ANOVA: F ( 1 , 16 ) = 5 . 844 , p = 0 . 028 , ηp2 = 0 . 27] . This interaction stems from the polarity-dependent changes in force output observed in the Tffrev-TCACA subgroup [main effect of block order on the residuals: F ( 1 , 16 ) = 7 . 58 , p = 0 . 014 , ηp2 = 0 . 32] . This effect was identical to the one observed for the two T-T subgroups ( Figure 3a ) . In contrast , for the other subgroup ( Tffrev-TACAC ) , there was no clear effect of stimulation polarity during the test period [simple main effect of block-order on the residuals: F ( 1 , 16 ) = 0 . 44 , p = 0 . 515 , ηp2 = 0 . 027] . Overall , we recorded data from four groups in the active tDCS conditions ( i . e . , T-T and Tffrev-T groups ) and our ANOVA results revealed a significant effect in three out of the four groups . Given our statistical power and effect size , this is exactly what would be expected . We compute the probability that there was an effect of stimulation polarity on force during the test period when the effect is observed in 3 out of the 4 groups . To compute this , we used a rationale provided by Ioannidis ( 2005 ) but adapted this justification to our positive results , as was done by Lakens and Evers ( 2014 ) . Given a power of 0 . 8 , the probability of observing three significant and one non-significant finding if there is a true effect is as follows ( Type 2 error ) : 0 . 8 × 0 . 8 × 0 . 8 × 0 . 2 = 0 . 1024 . Any of the four groups could yield a non-significant finding , so the a-priori likelihood of finding three out of four significant effects is 0 . 4096 . We also needed to find the probability of observing these results if there was no effect ( null hypothesis is true ) . Given a Type 1 error of 0 . 05 ( significance threshold ) , this probability is as follows: p = 0 . 05 × 0 . 05 × 0 . 05 × 0 . 95 = 0 . 00011875 . Again , because any group could be non-significant , the probability of finding these results if there is no effect is p = 0 . 000475 . With these data , we can compute the positive predictive value ( PPV ) ( Ioannidis , 2005 ) . This number represents the post-study probability that the effect is true: PPV = 0 . 4096/ ( 0 . 4096 + 0 . 000475 ) = 0 . 998 . That is , the likelihood that there is a true effect despite the fact that one of the four groups yielded a non-significant effect is 99 . 8% . Polarity-dependent modulation in force output could reflect the effect of anodal and cathodal tDCS in facilitating or suppressing force output . However , this is unlikely , as subgroups in which sham tDCS ( tDCS applied only at the beginning part of each block ) was applied during the training period followed by active stimulation during the test period ( S-T group: S-TACAC and S-TCACA subgroups , N = 8 for each subgroup , Figure 1e , h ) showed little influence of tDCS polarity on force output during the test period ( Figure 2d ) . While we detected a main effect of block order on the ΔForce permutation test: p = 0 . 0263 ) , this effect was reduced over time ( permutation test: interaction between period and block order: p = 0 . 025 ) ( Figure 2e , f ) . Indeed , ΔForce did not exhibit significant block-dependent modulation from the second to third block ( permutation test: p = 0 . 905 ) and from the third to fourth block ( p = 0 . 278 ) , although this modulation was significant from the first to second block ( p = 0 . 0132 ) ( Figure 2f ) . We also compared the strength of modulation by subtracting the ΔForce of S-T group from ΔForce of T-T group ( Figure 4a ) . Although we did not find the expected interaction between group and block order ( p = 0 . 205 ) , this appeared to change over time ( permutation test: interaction between block order , period and group: p = 0 . 0595 ) . Indeed , the modulation of ΔForce from the second to third block and from the third to fourth block was significantly larger in the T-T group than in the S-T group ( Figure 4b; permutation test: p = 0 . 0012 and p = 0 . 0146 , respectively ) . Together , this data revealed that the modulation of the force output of tDCS was larger in the T-T group than in the S-T group from the second block onwards . 10 . 7554/eLife . 15378 . 011Figure 4 . Difference in ΔForce between experimental groups . ( a ) ΔForce for the S-T group was subtracted from the T-T group . The solid line and shaded grey area indicates the mean and standard deviation of the bootstrapped samples ( moving average calculated over 5 data points ) . ( b ) The difference in the bootstrap samples of ΔForce was averaged for each block . The error bars indicate the standard deviations of the bootstrapped samples . ( c , d ) Difference in ΔForce between T-T and T-S groups . A permutation test was used to test the effect of block order ( **p<0 . 005 as indicated at the right side of panel ( d ) . A permutation test was also used to compare the values between the first and second , second and third , and the third and fourth blocks . #p<0 . 06; p<0 . 05; *p<0 . 005DOI: http://dx . doi . org/10 . 7554/eLife . 15378 . 011 The significant change in ΔForce from the first to second block in the S-T group ( Figure 2e , f ) and the absence of the differences between T-T and S-T groups for these blocks ( Figure 4a , b; permutation test: p = 0 . 36 ) indicate that the application of tDCS immediately after training can influence the exponential decay of motor memory in a manner opposite to the conventional effect of tDCS . Namely , cathodal stimulation makes motor memories last longer than anodal tDCS . Alternatively , the observed effect could reflect the possibility that the motor memory was tagged with the brain activity without tDCS and the application of different tDCS polarities during the test period might differently recruit the motor memories . Either way , these effects were not strong enough to induce block-dependent modulation in the later blocks ( Figure 2e , f; Figure 4a , b ) . These results were confirmed by the model-based approach . The residuals around the exponential curves in the S-T group did not exhibit any evidence of polarity-dependent modulation of force output [repeated measures 3-way ANOVA: interaction between subgroup and block order: F ( 1 , 14 ) = 0 . 005 , p = 0 . 945 , ηp2 = 0 . 063] ( Figure 3b ) . This discrepancy with the model-free approach is related to the ability of the exponential function to capture the early difference in decay rates . The modulation of the residuals via stimulation polarity was also larger for the T-T group than for the S-T group [interaction between group , subgroup , and block order by repeated measures 4-way ANOVA: F ( 1 , 28 ) = 16 . 80 , p = 2 . 42 × 10–4 , ηp2 = 0 . 37] . These results confirm that the tDCS effect was restricted to the group where active stimulation was received during the training and test periods . In another control experiment , active stimulation was applied during the training period while sham tDCS was applied during the test period ( T-S group ) ( Figure 1f , i ) . In this group , we did not observe any polarity-dependent modulation of force output with tDCS polarity ( Figure 2g ) . Indeed , the model-free approach applied to ΔForce indicated that there was no significant block-dependent modulation ( Figure 2h , i; permutation test: main effect of block order: p = 0 . 163 ) . The modulation of ΔForce was significantly larger in the T-T group than in the T-S group ( Figure 4c , d; permutation test , interaction between group and block order: p = 0 . 0018 ) . In addition , the modulation of ΔForce was significant from the first to second block , from the second to third block , and from the third to fourth block in the T-T group than in the S-T group ( Figure 4d; permutation test: p = 0 . 0573 , 0 . 0213 , and 0 . 0014 , respectively ) . These results were confirmed by the model-based approach . For the T-S group , the residuals around the exponential curve were not modulated by tDCS polarity [Figure 3c , repeated measures 3-way ANOVA , interaction between subgroup and block order: F ( 1 , 14 ) = 2 . 33 , p = 0 . 149 , ηp2 = 0 . 14 ) . In addition , the polarity-dependent modulation of force stimulation observed in the T-T group was significantly larger than the modulation observed in the T-S group [interaction between group , subgroup , and block order by repeated measures 4-way ANOVA: F ( 1 , 28 ) = 29 . 79 , p = 7 . 93 × 10–6 , ηp2 = 0 . 52] . Together , these results indicate that activity patterns in the sensorimotor cortex during motor learning needs to be reinstated in order to retrieve those memories . Finally , to assess the influence of the electrode position on the polarity-dependent modulation of force output , we performed an additional control experiment in which tDCS was applied to the posterior parietal cortex ( PPC: PPC group ) ( Figure 5a ) . Stimulation of the PPC was unable to yield block-dependent modulation of ΔForce as revealed by the model-free approach ( Figure 5b , c; permutation test: p = 0 . 369 ) . In addition , block-dependent changes of ΔForce were significantly smaller in the PPC group than in the T-T group ( Figure 5e , f; permutation test , interaction between group and block order: p = 0 . 0082 ) , although the changes of force output from the first to second ( permutation test: p = 0 . 0538 ) from the second to third ( p = 0 . 113 ) and from the third to fourth block ( p = 0 . 153 ) did not reach the significant level . Similarly , the model-based approach confirmed the absence of effect of tDCS when applied on PPC [Figure 5d , repeated measures 3-way ANOVA on the residuals , interaction between subgroup and block order: F ( 1 , 16 ) = 1 . 98 , p = 0 . 179 , ηp2 = 0 . 11] . In addition , repeated measure 4-way ANOVA indicated that polarity-dependent modulation of force output was significantly greater for the T-T group than for the PPC group [interaction between group , subgroup , and block order: F ( 1 , 30 ) = 5 . 10 , p = 0 . 031 , ηp2 = 0 . 15] , indicating that M1 stimulation was more effective than PPC stimulation in creating and retrieving polarity-dependent motor memories . 10 . 7554/eLife . 15378 . 012Figure 5 . Results during the test period for the PPC group . ( a ) Force output during the test period . ( b ) Trial dependent change in ΔForce obtained with a bootstrap method . The solid line and grey area indicates mean and standard deviation of the bootstrapped samples ( moving average calculated over 5 data points ) . ( c ) ΔForce averaged for each block . The error bars indicate standard deviations of the bootstrapped samples . ( d ) Residuals obtained by subtracting the exponential curves from the original force output . ( e , f ) Trial-dependent ( e ) and block-dependent ( f ) difference in ΔForce between T-T and PPC groups . A permutation test was used to test the effect of block order ( p<0 . 01 as indicated by the right side of panel ( f ) . A permutation test was also used to compare the values between the first and second , second and third , and the third and fourth blocks . #p<0 . 06 . DOI: http://dx . doi . org/10 . 7554/eLife . 15378 . 012 We also examined the potential effect of tDCS polarity alternations on the interference between opposing perturbations during the training period ( Figure 6 ) . To investigate the amount of interference between the two perturbations , we measured the lateral deviations of the first trial during rightward and leftward force-field training . With this measure , higher interference is associated with higher lateral deviation on the first trial . When comparing this measure between the T-T , S-T , and T-S groups , we did not detect any significant differences in lateral deviation during the first trial [repeated measures 3-way ANOVA: F ( 2 , 45 ) = 2 . 369 , p = 0 . 105 , ηp2 = 0 . 088] ( Figure 6b ) . However , when the data from the T-T and T-S groups were grouped together ( they received the same training and stimulation during the training period ) , lateral deviations during the first trial were significantly smaller ( i . e . , less interference ) for groups who received active tDCS whilst training ( T-T and T-S groups ) than for the sham group ( S-T group ) [main effect of group: F ( 1 , 46 ) = 4 . 435 p = 0 . 041 , ηp2 = 0 . 10; interaction between group and force-field direction ( or tDCS polarity ) : F ( 1 , 46 ) = 0 . 148 , p = 0 . 708 , ηp2 = 0 . 003] . This suggests that tagging motor memories with different tDCS might have an effect during the training period . However , this result needs to be confirmed in follow-up studies given the limited effect size and borderline significance here . In contrast , lateral deviation at the end of each block was not different between groups [repeated measures 3-way ANOVA , main effect of group: F ( 1 , 46 ) = 0 . 574 , p = 0 . 453 , ηp2 = 0 . 012; interaction between group and force-field direction: F ( 1 , 46 ) = 0 . 401 , p = 0 . 53 , ηp2 = 0 . 009] ( Figure 6c ) . 10 . 7554/eLife . 15378 . 013Figure 6 . Results during the training period . ( a ) Trial-by-trial changes in lateral deviation evaluated at the peak handle velocity . Only data averaged among participants are displayed; depicting standard errors makes it difficult to observe the data . Red and blue backgrounds indicate the period in which anodal and cathodal stimulation were applied . ( b , c ) Lateral deviation for the first trial ( b ) and last trial ( c ) for each block averaged over participants . The error bar indicates standard error . DOI: http://dx . doi . org/10 . 7554/eLife . 15378 . 013 Recent studies have demonstrated several striking examples of context-dependent motor learning ( Osu et al . , 2004; Nozaki et al . , 2006; Hirashima and Nozaki , 2012; Kadota et al . , 2014; Cothros et al . , 2009; Ikegami et al . , 2010; Howard et al . , 2011; Yokoi et al . , 2011; Howard et al . , 2012; Yokoi et al . , 2014; Howard et al . , 2015 ) . Distinct motor memories for identical movements can be formed and retrieved according to whether the opposite arm is stationary ( i . e . , unimanual ) or moving ( i . e . , bimanual ) ( Nozaki et al . , 2006; Kadota et al . , 2014; Nozaki and Scott , 2009 ) . Motor memories can also be influenced by movement direction of the opposite arm ( Yokoi et al . , 2011; Howard et al . , 2010; Yokoi et al . , 2014 ) , whether the movement is discrete or rhythmic ( Ikegami et al . , 2010; Howard et al . , 2011 ) , or how follow-through movements are performed ( Howard et al . , 2015 ) . Although this type of learning enables us to perform flexible actions for adapting to a wide variety of dynamical environments ( Yokoi et al . , 2011; 2014 ) , it remains unknown how different behaviors result in the formation of distinct motor memories . Intriguingly , previous neurophysiological and/or brain imaging studies have shown that activity in motor-related neural areas also differs according to how an opposite arm is moving ( Donchin et al . , 2001 , 2002; Cisek et al . , 2003; Ganguly et al . , 2009 ) , between discrete and rhythmic movement ( Schaal et al . , 2004 ) , or the direction of the follow-through movement ( Baldauf et al . , 2008 ) . It should be noted that previous studies have also suggested that movement adaptation to a novel dynamical environment is accomplished through neural activity changes in these motor-related areas ( Kadota et al . , 2014; Gandolfo et al . , 2000; Li et al . , 2001; Arce et al . , 2010 ) . Thus , it is natural to speculate that these neurons change their activation patterns during motor adaptation , depending on how they are originally recruited within the behavioral context . For example , distinct motor memories for unimanual and bimanual movements ( Nozaki et al . , 2006 ) can be explained by considering that partially different populations of M1 and PM neurons are involved for motor adaptation ( Nozaki and Scott , 2009 ) . We hypothesized that motor learning performed during distinct background neural activity in the sensorimotor cortex leads to the development of a motor memory that is specifically associated with a particular behavior . We examined this hypothesis by artificially creating two different background activity patterns in the sensorimotor cortex , using anodal and cathodal tDCS , when participants learned to perform reaching movements within 2 conflicting forcefields . Most importantly , each tDCS polarity was always associated with one of two force-fields . Consistent with this hypothesis , our experiments demonstrated that , after training , force output during the test period was modulated by the corresponding polarity ( T-T group; Figure 2a–c , Figure 3a ) . In contrast , force output modulation was not observed when tDCS was applied to the PPC ( PPC group; Figure 5 ) . Furthermore , the ability of tDCS to modulate force output , depending on polarity through mere changes in cortical excitability ( Salimpour and Shadmehr , 2014; Tanaka et al . , 2009 ) , was insufficient to explain our observed force output modulation ( S-T group; Figure 2d–f , Figure 4a , b ) . Participants were also unlikely to modulate the force output according to tDCS-induced skin sensation , as both the PPC and T-T groups experienced similar tDCS sensations throughout the experiments . Taken together , our experiments strongly indicate that differential background activity patterns in the sensorimotor cortex , artificially created when a motor learning task is performed , could lead to the formation of distinct motor memories . These memories could then be artificially retrieved by applying the same-polarity tDCS during a reaching movement . It is important to emphasize that our experimental paradigm fully departs from conventional tDCS applications , which are traditionally restricted to increasing/decreasing motor output and accuracy , motor memory consolidation , rehabilitation outcomes , and memory functioning ( Nitsche et al . , 2008; Orban de Xivry and Shadmehr , 2014 ) . Furthermore , if tDCS can artificially induce motor memory recollection , tDCS may also artificially maintain motor memories when tested in different contexts . Indeed , we have previously shown that reaching movement adaptation to a force-field , while receiving anodal or cathodal tDCS , is more strongly generalized to reaching movements toward another direction performed with different arm postures ( Orban de Xivry et al . , 2011 ) . This improved generalization could be explained by the artificial maintenance of activity patterns in the sensorimotor cortex with tDCS when transitioning from the adaptation phase to the generalization phase . We assumed that tDCS only influenced neural activity in the sensorimotor cortex , particularly in M1 and PM neurons . Recent studies using fMRI ( Lindenberg et al . , 2013 ) and electrical fields models ( Gillick et al . , 2014 ) have indicated that bihemispheric tDCS effects are more localized than when using a conventional stimulation montage ( i . e . , placing one electrode on the forehead ) , but we cannot reject the possibility that broader brain areas are also involved . Given our bilateral stimulation , it is also possible that stimulation to the ipsilateral right sensorimotor cortex ( ipsilateral to the moving arm ) caused the observed effects ( rather than the contralateral left sensorimotor cortex ) . Further studies using a High-Defintion tDCS ( Datta et al . , 2009 ) will be necessary to clarify these issues . Finally , however , given that tDCS had no effect on memory retrieval when applied on the PPC , the involvement of the underlying parietal areas , including sensory areas , is unlikely or relatively small . We also assumed that we could create different background activity patterns via anodal and cathodal tDCS . Indeed , previous studies have demonstrated that M1 excitability ( Nitsche et al . , 2005; Nitsche et al . , 2007 ) or neuronal firing rates ( Creutzfeldt et al . , 1962 ) could be modulated according to tDCS polarities . Although the present study only focused on the immediate modulating effect of tDCS on neural activity , tDCS also has long-lasting or plastic effects on the cortex ( Nitsche et al . , 2008 ) , raising questions as to whether activity patterns in the sensorimotor cortex were similar during the training and testing periods . For example , M1 excitability remains elevated , even after anodal tDCS termination , when stimulus duration is sufficiently long ( Nitsche et al . , 2007; 2008; Nitsche and Paulus , 2000; 2001 ) . However , we considered that any long-lasting effects did not play a dominant role in our results . First , as tDCS polarity switched every 2 min , durations were too short to induce long-lasting effects on cortical excitability . Second , as anodal and cathodal tDCS alternated , any long-lasting effects , including homeostatic plasticity ( Siebner et al . , 2004 ) , were likely cancelled out . Previous studies have suggested that reinstating the same activity pattern as was present during learning might contribute to one’s ability to recall that memory ( Maren et al . , 2013; Fanselow , 2010 ) . For example , invasive electrical stimulation to a particular region of the temporal cortex among patients with epilepsy can induce declarative memory recollection . This is perhaps due to reproducing an activity pattern similar to what was observed during voluntary recollection of that same memory ( Jacobs et al . , 2012 ) . Similarly , we used tDCS to reinstate an activity pattern in the sensorimotor cortex during learning in order to induce implicit retrieval of a particular motor memory . These effects are also similar to optogenetics’ ability to manipulate a fear memory in mice by directly altering hippocampal neural activity ( Liu et al . , 2012; Ramirez et al . , 2013 ) . In summary , our study demonstrates a causal link between background neural activity in the sensorimotor cortex during learning and context-dependent motor memories . Here , we provided initial evidence that human motor memories can be artificially tagged and later retrieved via noninvasive brain stimulation . This manipulation sheds light on the possibility that manipulating the formation and recollection of various memories , including declarative and other types of motor memories ( e . g . , visuomotor rotation task or sequence learning ) , can be achieved by artificially changing activity patterns of the corresponding brain region during the learning period . Thus , our novel tDCS application opens up new avenues for implementing tDCS in human memory research . We recruited 89 healthy participants ( 21–42 years old; 52 men and 37 women ) . Each participant was tested only once . The experiments were terminated for 5 participants ( 2 men and 3 women ) due to impedance failures , strong pain , etc . There were 3 different groups in the main experiments ( T-T , S-T , and T-S groups ) , and each consisted of 2 subgroups [N = 8 ( 5 men and 3 women ) for each subgroup] . Additional control experiments were performed for the Tffrev-T and PPC groups . These groups also consisted of 2 subgroups [N = 9 ( 5 men and 4 women ) for each subgroup] . Participants were not given any information regarding to which group they belonged . According to the Edinburg Handedness Inventory , all participants were right-handed ( Laterality Quotient: 0 . 92 ± 0 . 16 ) , except for 2 women ( -1 . 0 and -0 . 78 ) . There were no significant differences in the quotient between experimental groups . Prior to the experiments , participants provided informed consent and were paid for their participation . Participants , sitting on a chair , grasped a robotic manipulandum handle ( KINARM End-Point Lab , Bkin Technologies , Canada ) with their right hand and moved it horizontally from a starting position toward a target displayed on a mirror placed above the arm . Thus , participants could not directly see their own arm but could see the handle position thorough a white circle ( diameter = 1 . 0 cm ) displayed on the mirror . The upper body was fixed to the chair by straps , and a sling was used to hang the forearm horizontally . The start position was located about 20–30 cm from the middle of the chest , and the target was located 10 cm ahead of the start position . The start and target positions were displayed as a circle ( diameter = 1 . 4 cm ) . After participants maintained their right hand at the starting position for 500–1 , 000 ms , a green target appeared . After an additional waiting time of approximately 800 ms , the color of the target changed to magenta , indicating that participants should reach towards the target . When the handle reached toward the target , the target turned green . Participants were instructed to make their movements with a peak velocity between 0 . 35 m/s and 0 . 45 m/s . The warning 'Slow' or 'Fast' was displayed on the screen if movement speed was outside that range . At the end of the movement , the robot retuned the handle to the starting position . Before the experimental trials , participants performed at least 40 reaching movements ( without a force field or tDCS ) for practice . Velocity-dependent force fields were imposed on the robotic manipulandum handle during the training period ( Shadmehr and Mussa-Ivaldi , 1994 ) . The force , f= ( fx , fy ) ( N ) , imposed on the handle was always set to be perpendicular to the velocity of the handle , v= ( vx′vy ) ( m/s ) , as f=Bv′ , where B = ( 0 10; −10 0 ) and ( 0 −10; 10 0 ) [N/ ( m/s ) ] for the rightward and leftward force field , respectively ( x and y directions indicate right-and-left and anteroposterior directions , respectively ) . To evaluate adaptation to the force fields , error-clamp trials ( Scheidt et al . , 2000 ) were used . During these trials , the handle trajectory was constrained by the robotic manipulandum to a straight line going from the starting position to the target ( i . e . , a force channel ) . A virtual spring [15 , 000 N/m and damper of 100 N/ ( m/s ) ] created the force channel . To evaluate motor memory output , the force exerted by participants against the channel was measured . Rightward forces were defined as positive in the present study . In total , participants performed 373 reaching movements after performing 40 practice trials . Participants performed 20 reaching trials without tDCS or the force field ( baseline trials ) . Twelve trials were error-clamp trials during which the force exerted against the force channel for a baseline was obtained . After the baseline trials , participants performed training trials ( training period ) followed by test trials ( test period ) . The training period consisted of 12 blocks of 22 reaching movement trials ( Figure 1 ) . In each block , the first and last 2 trials were error-clamp trials , and the remaining 18 trials were force field trials . Participants experienced a rightward force field ( T-T , S-T , T-S and PPC groups ) and a leftward force field ( Tffrev-T group ) in the first block . The rightward and leftward force fields alternated in every block . Anodal ( 2 mA ) and cathodal ( -2 mA ) tDCS to the left M1 ( T-T , S-T , T-S and T-T groups ) and to the left PPC ( PPC group ) also alternated in every block . Thus , the rightward force field was learned while receiving anodal tDCS , and leftward force was learned while receiving cathodal tDCS during the training period ( T-T , S-T , T-S , and PPC groups ) . The association between tDCS polarity to M1 and the force-field direction was reversed during an additional control experiment ( Tffrev-T group ) . Participants were not given any explicit contextual cue for the direction of the force-field throughout the experiments . In each block , the transition time from 0 mA to 2 mA ( or −2mA ) , and 2 mA ( −2 mA ) to 0 mA , was 6 s . The reaching target of the first and last trial for each block ( error-clamp trial ) appeared when tDCS intensity reached 1 mA ( or −1 mA ) . As the target appeared every 6 s ( i . e . , inter-trial interval was 6 s ) , tDCS reached 2 mA ( or −2 mA ) at the second and second-to-last trials for each block ( these were also error-clamp trials ) . The test period consisted of 4 blocks of 22 reaching movement trials ( Figure 1g–i ) . Error-clamp trials were used for all 88 reaching movements . The anodal and cathodal tDCS ( ± 2 mA ) were alternated every block as in the training period . In order to maintain participants’ concentration , after 8 blocks of the training period were completed , participants rested for 3 min . After one error-clamp trial without tDCS , the training period was restarted . The T-T group received active tDCS during both the training and test periods ( Figure 1g ) . Training protocols were identical for both subgroups ( T-TACAC and T-TCACA groups ) . In the odd-numbered blocks ( i . e . , first , third , fifth , seventh , ninth , and eleventh ) and even-numbered blocks ( i . e . , second , fourth , sixth , eighth , tenth , and twelfth ) , respectively , participants were trained with the rightward force-field while receiving anodal tDCS and with the leftward force-field while receiving cathodal tDCS ( Figure 1g ) . However , during the test period , the block order during which participants received anodal and cathodal tDCS was reversed: T-TACAC and T-TCACA subgroups received anodal and cathodal tDCS , respectively , in the first block of the test period ( Figure 1g ) . For the S-TACAC and S-TCACA groups , sham tDCS was used during the training period , but active tDCS was used during the test period ( Figure 1e , h ) . Sham tDCS was idential to active tDCS for the initial 9 s of each block ( ramp-up for 6 s and constant for 3 s ) but ramped down to 0 mA for 6 s ( thus , 2 error-clamp trials were performed during these ramp-up and ramp-down phases ) ( Figure 1e ) . For the T-SACAC and T-SCACA groups , sham tDCS was used instead of active tDCS during the test period ( Figure 1f , i ) . In the Tffrev-T group , the order of anodal and cathodal tDCS was identical to that of the T-T group , but the leftward force field was imposed on the first block of the training period . Thus , the leftward and rightward force fields , respectively , were always associated with anodal and cathodal tDCS , respectively . The Tffrev-T group also consisted of Tffrev-TACAC and Tffrev-TCACA subgroups according to the tDCS polarity order received during the test period . The experimental procedure for the PPC group was the same as for the T-T group , except the PPC was stimulated instead of M1 . More specifically , anodal and cathodal tDCS to the left PPC was associated with the rightward and leftward force fields , respectively . The PPC group consisted of PPCACAC and PPCCACA subgroups according to tDCS polarity order received during the test period . An electrical current stimulator applied tDCS ( DPS-133A , Dia-Medical Co Ltd , Japan ) . Two rubber electrodes ( 5 cm × 7 cm ) covered with a sponge soaked in normal saline solution were placed symmetrically at the left and right M1 regions ( around C3 and C4; T-T , S-T , T-S , and Tffrev-T groups ) or at the left and right PPC regions ( around P3 and P4; PPC group ) . We adopted the bihemispheric montage because the stimulated area can be more localized in the sensorimotor cortex , including M1 ( Lindenberg et al . , 2013; Gillick et al . , 2014 ) . Before applying the electrical current , we carefully checked that resistance between the electrodes was below 5 . 0 kΩ , via an LCR meter ( LCR821 , GW Instek , Taiwan ) , in order to reduce pain or burn injury risk . The values before and after data collection were 3 . 86 ± 1 . 09 and 3 . 38 ± 2 . 07 kΩ , respectively . During the experiment , the electrical current was continuously monitored using an ammeter ( Digital Multimeter CD772 , Sanwa , Japan ) . Participants reported the degree of pain using a numerical rating scale ( 1–10 , 10 indicates maximal pain ) after a 3-min break ( i . e . , after 8 blocks of training were completed ) and after the experiment . The reported value was 2 . 19 ± 1 . 06 and 2 . 25 ± 1 . 08 for the T-T group ( mean ± SD for the first 8 blocks and second 8 blocks , respectively ) , 2 . 63 ± 1 . 38 and 3 . 23 ± 1 . 43 for the S-T group , and 2 . 53 ± 1 . 29 and 2 . 07 ± 0 . 81 for the T-S group . It should be noted that participants experienced 4 blocks of sham tDCS in the first 8 blocks for the S-T group and in the second 8 blocks for the T-S group . There was a significant interaction in terms of reported pain between groups and block ( i . e . , first and second block ) [F ( 2 , 44 ) = 6 . 876 , p = 0 . 003] . Thus , although the difference in scale values was less than 1 , participants might feel stronger skin sensations for active tDCS than for sham tDCS . Handle position and exerted force data were sampled at 1000 Hz and then digitally lowpass filtered using a Butterworth filter ( cutoff frequency 10 Hz ) . Handle velocity was obtained by numerical differentiation . In order to quantify adaptation to the force field during the training period and motor output during the test period , the lateral force exerted by participants against the force channel was evaluated at the handle’s peak velocity . Rightward force was defined as positive . We also quantified the handle’s lateral deviation at the peak velocity to evaluate learning during the training period . Before analyzing the lateral force and lateral deviation data , baseline trial values were subtracted . The force output during the test period consisted of natural exponential decay ( Criscimagna-Hemminger and Shadmehr , 2008; Brennan and Smith , 2015 ) which might obscure the polarity-dependent modulation . We adopted two different approaches to eliminate the effect of the decay . In the first , model-free , approach , we calculated the difference in force between the two subgroups ( ΔForce ) obtained by subtracting ACAC subgroup data from CACA subgroup data for each group separately:ΔForce =F¯CACA −F¯ACAC , where F¯ represents the average force across the corresponding subgroup . This approach was based on the assumption that the decaying pattern was almost identical for both subgroups , because they experienced the same training protocol . To obtain the mean and standard deviation of ΔForce for each group and block separately , we carried out a bootstrap analysis ( Hesterberg et al . , 2003 ) by randomly sampling the data sets with replacement ( N = 10 , 000 ) for both the ACAC and CACA subgroups and calculated ΔForce ( Figure 2b , c , e , f , h and i; Figure 4; Figure 5b , c , e , and f ) . To statistically test the effect of polarity on the ΔForce during the test period for each group separately , we used permutation test ( Hesterberg et al . , 2003 ) where the variable of interest was the change in ΔForce as a function of the block order where block order refers to the first or second block of each half of the test period ( first and third vs . second and fourth ) ( Figure 2—figure supplement 1 ) . We expected ΔForce to be more positive during the second and fourth blocks and more negative during the first and third blocks . The modulation of ΔForce as a function of block order was thus computed as follow:BO=mean ( ΔForceb2 ) +mean ( ΔForceb4 ) − ( mean ( ΔForceb1 ) +mean ( ΔForceb3 ) ) , where ΔForcebi represents the ΔForce values of the ith block . The distribution of this variable of interest under the null hypothesis was obtained by computing all the possible values of the polarity contrast under resampling ( N = 10 , 000 ) with random reassignment of the subjects in two subgroups ( without replacement ) . The p-value was defined as the portion of the distribution that was more extreme than the observed polarity contrast ( Hesterberg et al . , 2003 ) . The same technique was used to analyze other variables of interest . The influence of period ( first or second half of the test period: Figure 2—figure supplement 1 ) on the block order contrast was quantified by the difference in block order contrast across the two halves of the test period . This interaction was obtained as following:interaction between period and block order= ( mean ( ΔForceb2 ) −mean ( ΔForceb1 ) ) − ( mean ( ΔForceb4 ) −mean ( ΔForceb3 ) ) , where ΔForcebi represents the ΔForce values of the ith block . Block-dependent changes in ΔForce were analyzed similarly . In this case , the change in Δ between consecutive blocks was computed and submitted to the permutation test . ΔForcei⟶i+1=mean ( ΔForcebi+1 ) −mean ( ΔForcebi ) . Statistical test for force between different groups ( e . g . , T-T vs . S-T groups ) was also performed via a permutation test . The observed value corresponded to the difference between the variable of interest of the two groups . For instance , for the block order contrast , the observed value was BOGR1−BOGR2 where GR1 and GR2 are the two compared groups . In this case , random reassignment of subjects was performed across groups . In addition to the above-mentioned model-free approach , we used a model-based approach where we estimated the exponential decay for each participant individually and analyzed the residuals around the exponential curve . We reasoned that if stimulation polarity had no effect , the force measures should be evenly distributed around the exponential fit ( Figure 3—figure supplement 1a ) . In contrast , if tDCS polarity influences the forces , these residuals should oscillate around that exponential curve across the blocks ( Figure 3—figure supplement 1b ) . For each participant , we fitted 2 exponential functions to the force data , y=A1exp ( B1n ) +A2exp ( B2n ) + C , where n represents trial number , and A , B , and C are free parameters . To this end , the ‘nlinfit’ function in MATLAB was used . We used the 2 exponential functions for data fitting , since a residual calculated using only 1 exponential function could potentially demonstrate artificial polarity-dependent modulation due to data drift . However , our results were not substantially influenced , even when 1 exponential function ( i . e . , y=Aexp ( Bn ) +C ) was used ( Figure 3—figure supplement 3 ) . Residuals between the actual data and exponential functions were averaged for each block ( the first to fourth blocks ) of the test period and tested using ANOVA . The test period consisted of two repetitions of anodal-cathodal tDCS ( ACAC subgroup ) or cathodal-anodal tDCS ( CACA subgroup ) . To test the effect of subgroup and stimulation polarity on the residuals of the force output during the test period , a repeated measures 3-way ANOVA was conducted with within-subject factors period ( the first 2 blocks or last 2 blocks of the test period ) and block order ( the first or second block for each half of the test period , during which the same tDCS polarity was received in both halves of the test period ) and between-subject factor subgroup ( ACAC or CACA ) ( Figure 2—figure supplement 1 ) . In this ANOVA , the factor block order can be assimilated with stimulation-induced polarity-dependent changes in force output . Indeed , tDCS polarity was identical between the first and third blocks of the test period ( where the block order factor has the same value ) and different during the second and fourth blocks of the test period ( where the block order factor had another value ) . Given that the association between block and polarity was opposite between subgroups [the first ( or second ) block for the ACAC subgroup corresponded to anodal ( or cathodal ) stimulation and vice versa for the CACA subgroup] , we expected , if tDCS influenced the force output , a significant interaction between subgroup and block order . In addition , if the force output varies with tDCS polarity in each subgroup , we expect a main effect of block order for each subgroup separately . The same analysis was used for hand velocity during the test period . We used a 4-way ANOVA where the between-subject factor group was added to contrast the effect of active and sham stimulation ( T-T vs S-T or T-T vs T-S ) and to contrast the effect of stimulation site ( T-T vs PPC ) . To test the effect of stimulation type on learning during the training period , the lateral deviation of the first training trial for each block , the averaged value of the last two field trials for each block , and the force exerted during the first error-clamp trial at the end of the block were subjected to an ANOVA with polarity ( anodal and cathodal ) and block ( first to sixth ) as within-subject factors and stimulation type ( T-T and S-T vs . T-S ) as a between-subjects factor . The statistically significant threshold was set at p<0 . 05 both for the ANOVA and permutation test . For the results of ANOVA , we reported effect sizes ( partial eta squared: ηp2 ) as well as F and p-values .	Memory is strongly affected by the context in which a particular memory is formed and remembered . For example , visiting a familiar place can often trigger memories associated or “tagged” with that place . Such tagging also exists for memories related to movement: for instance , distinct motor memories for a limb movement are formed depending on whether the other limb is stationary or moving . However , little is known about how the tagging of such motor memories takes place . Nozaki et al . have now used a technique known as transcranial direct current stimulation to generate artificial “tags” for motor memories . In the experiments , volunteers tried to move a robotic arm towards a goal while the robot pushed their hand off-course . Sometimes the robot pushed the participant’s hand to the left , and sometimes to the right . This makes the task difficult to learn , even when the cue for the direction is provided , as the motor memories that are made to counteract each push overwrite each other . Nozaki et al . used transcranial stimulation to alter the background electrical activity in the sensorimotor regions of the participants’ brains as they performed the robotic arm task . Artificially generating a different pattern of background brain electrical activity for each push direction caused the motor memories associated with leftward and rightward pushes to be tagged differently . Once this association had been learnt , applying the artificial brain stimulation pattern associated with one of the pushes resulted in the participants unconsciously compensating for a push in that direction , even when it was not there . Overall , the results presented by Nozaki et al . suggest that the background electrical activity seen in the brain can influence how a motor memory is created and later recalled . A future challenge is to investigate whether this technique could be used to help athletes improve their performance or to treat people with movement disorders . Further experiments are also needed to test whether the same approach can influence the formation and recollection of other kinds of memories , such as those related to fear .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Tagging motor memories with transcranial direct current stimulation allows later artificially-controlled retrieval
Blood vessels are lined by endothelial cells engaged in distinct organ-specific functions but little is known about their characteristic gene expression profiles . RNA-Sequencing of the brain , lung , and heart endothelial translatome identified specific pathways , transporters and cell-surface markers expressed in the endothelium of each organ , which can be visualized at http://www . rehmanlab . org/ribo . We found that endothelial cells express genes typically found in the surrounding tissues such as synaptic vesicle genes in the brain endothelium and cardiac contractile genes in the heart endothelium . Complementary analysis of endothelial single cell RNA-Seq data identified the molecular signatures shared across the endothelial translatome and single cell transcriptomes . The tissue-specific heterogeneity of the endothelium is maintained during systemic in vivo inflammatory injury as evidenced by the distinct responses to inflammatory stimulation . Our study defines endothelial heterogeneity and plasticity and provides a molecular framework to understand organ-specific vascular disease mechanisms and therapeutic targeting of individual vascular beds . Endothelial cells ( ECs ) line blood vessels in all tissues and organs , and they form a barrier which tightly regulates the trafficking of oxygen , metabolites , small molecules and immune cells into the respective tissue ( Liao , 2013 ) . Previous studies have suggested that the morphology of the microvascular endothelium or the expression of selected genes can vary when comparing the vasculature of multiple tissues , thus allowing ECs to take on tissue-specific EC functions ( Chi et al . , 2003; Potente and Mäkinen , 2017; Aird et al . , 1997 ) . Environmental signals from the tissue microenvironment including mechanical forces , metabolism , cell-matrix , cell-cell interactions , organotypic growth factors likely play an important role in regulating this endothelial heterogeneity ( Potente and Mäkinen , 2017 ) . The tissue-specific interaction between ECs and surrounding cells occurs as early as during development , when , for example , brain ECs instruct neuronal differentiation ( Bussmann et al . , 2011; Matsuoka et al . , 2017 ) . Such tissue-specific endothelial adaptations persist throughout adulthood when brain ECs form a highly selective barrier composed of specialized tight junctions to limit neurotoxicity ( Pozhilenkova et al . , 2017 ) . In the lung , ECs differentiate in parallel with epithelial cells to form gas exchange units which are in contact with the external environment and thus need to ensure a rapid immune response ( Jambusaria et al . , 2018; Rafii et al . , 2016 ) . Heart ECs , on the other hand , are specialized in a manner to ensure ready supply of fatty acids to voracious cardiomyocytes which rely on continuous supply of fatty acids as the primary fuel to generate ATP necessary for cardiac contraction ( Potente and Mäkinen , 2017 ) . Identifying differences in the expression levels of selected genes in endothelial cells from different tissues or organs provides some insights into the molecular underpinnings of endothelial heterogeneity , however unbiased gene expression profiling is likely to yield a more comprehensive evaluation of the genes and regulatory pathways underlying endothelial heterogeneity . Microarray profiling has been used to identify paracrine factors and signaling pathways that characterize endothelial cells in different organs ( Jambusaria et al . , 2018; Nolan et al . , 2013 ) . Single-cell transcriptomic analysis of endothelial cells has also provided a molecular atlas of the brain and lung vasculature at a single cell level ( Vanlandewijck et al . , 2018 ) . The latter work has characterized transcriptomic signatures of distinct endothelial subpopulations . While single cell RNA-sequencing is ideally suited for identifying subpopulations within a single vascular bed , current single cell technologies are limited in their ability to detect the expression of individual genes in a given cell ( Bacher and Kendziorski , 2016; Zhu et al . , 2018; Kharchenko et al . , 2014; Lun et al . , 2016; Vallejos et al . , 2017 ) . The endothelial signatures defined using these transcriptomic approaches are potentially influenced by disassociation and isolation of endothelial cells , a process affecting cellular mRNA levels when cells are removed from their native niche ( Haimon et al . , 2018; Rossner et al . , 2006; Sugino et al . , 2006 ) . Furthermore , conventional global mRNA and single cell mRNA transcriptomic profiling does not discriminate between the total mRNA pool and those mRNAs preferentially translated due to translational regulation ( Zhou et al . , 2016; Piccirillo et al . , 2014 ) . In the present study , to understand further the variegated nature of the endothelium , we used the RiboTag transgenic mouse model , in which LoxP mice express an HA-tag on the ribosomal Rpl22 protein ( Sanz et al . , 2009 ) . These mice enable direct isolation of tissue-specific mRNAs undergoing translation without cell disassociation ( Sanz et al . , 2009 ) . Using an endothelial-specific RiboTag model , we show that organ-specific ECs have distinct translatome patterns of gene clusters during homeostasis . Since the circulating bacterial endotoxin lipopolysaccharide ( LPS ) is a key mediator of tissue inflammation and injury in patients with bacteremia and sepsis ( Cross , 2016 ) ( Charbonney et al . , 2016 ) , we also exposed the RiboTag mice to LPS to induce systemic inflammatory injury and studied the organ-specific EC translatome response . We found that ECs express tissue-specific genes involved in vascular barrier function , metabolism , and substrate-specific transport . In addition , we found that ECs expressed genes thought to be primarily expressed in the surrounding tissue parenchyma , suggesting a previously unrecognized organ-specific endothelial plasticity and adaptation . To allow other researchers to explore the organ-specific EC translatome heterogeneity , we have generated a searchable database ( http://www . rehmanlab . org/ribo ) , in which users can visualize gene expression levels of individual genes . To precisely investigate the in-situ organ-specific EC molecular signature in brain , lung , and heart tissue we crossed the RiboTag mice ( Rpl22HA/+ ) ( Sanz et al . , 2009 ) with the endothelial-specific VE-cadherin-Cre mice ( Jeong et al . , 2017; Sörensen et al . , 2009 ) to generate RiboTagEC ( Cdh5CreERT2/+; Rpl22HA/+ ) mice . At 4 weeks post tamoxifen administration , ribosomes in the endothelial cells of all tissues expressed the HA tag , thus allowing for the specific isolation of mRNA undergoing ribosomal translation from ECs in the brain , heart and lung during homeostatic conditions . We also isolated brain , lung , and heart endothelial mRNA at several time points following systemic inflammatory injury , induced using a sublethal dose of the bacterial endotoxin lipopolysaccharide ( LPS ) , ranging from the acute injury phase at 6 hr post-LPS to the recovery phase at 1 week post-LPS ( Figure 1—figure supplement 1A ) . Log fold change ( logFC ) values were calculated between endothelial mRNA ( immunoprecipitated by an anti-HA antibody ) versus whole tissue mRNA ( immunoprecipitated with control antibody , anti-RPL22 ) using quantitative PCR ( qPCR ) . The analysis of the qPCR data confirmed enrichment of endothelial-specific RNA similar to what has been reported in other studies using the RiboTag model ( Jeong et al . , 2017 ) and also demonstrated minimal expression of RNA from other tissue-resident cell types ( Figure 1—figure supplement 1B–1F ) . After confirming the enrichment of endothelial RNA using qPCR , we performed global transcriptional profiling with RNA-Seq on the RiboTagEC brain , lung , and heart samples . Principal component analysis ( PCA ) of the RNA-Seq data for endothelial mRNA from brain , lung , and heart tissue from all time points showed a clear separation between the replicate brain , lung , and heart translatomes , indicating that ECs from each tissue demonstrated a distinct transcriptional identity at baseline that is maintained even in the setting of profound systemic inflammatory injury ( Figure 1A ) . In order to identify the genes responsible for these distinct tissue-specific EC profiles , we performed a differential expression analysis on the RNA-Seq data . The differential expression analysis was concordant with the PCA and identified 1692 genes which were differentially expressed in brain ECs ( versus ECs from the other two tissues ) , 1052 genes which were differentially expressed in lung ECs , and 570 genes which were differentially expressed in heart ECs ( Figure 1B ) . We next analyzed the baseline heterogeneity of ECs obtained from brain , lung and heart by assessing the gene expression levels of endothelial genes using established databases . We specifically focused our analysis on a pan-endothelial gene set ( Franzén et al . , 2019 ) , glycolysis and fatty acid metabolism gene sets ( Shimoyama et al . , 2015 ) and a solute transport gene set ( Hediger et al . , 2013 ) . Hierarchical clustering of the RNA-Seq profiles on merely 152 pan-endothelial genes from PanglaoDB ( Franzén et al . , 2019 ) separated all replicate baseline samples , indicating that classical endothelial markers are sufficient to differentiate ECs from these three organs ( Figure 1C ) . For example , genes upregulated in brain ECs included T-box transcription factor ( Tbx1 ) and the glucose transporter 1 ( Slc2a1 ) , genes upregulated in the lung endothelium included claudin 5 ( Cldn5 ) and the Hes related family BHLH transcription factor with YRPW Motif 1 ( Hey1 ) , whereas heart ECs demonstrated upregulation of vascular endothelial growth factor receptor 2 ( Kdr ) and the endothelial cell surface expressed chemotaxis and apoptosis regulator ( Ecsr ) . We next focused on the tissue-specific upregulation of metabolic genes . As seen in the glycolysis gene heatmap , we found that most tissue-specific EC genes involved in glycolysis were specifically upregulated in the brain endothelium ( Figure 1D ) , but there were selected glycolytic genes specifically upregulated in other tissues such as 6-phosphofructo-2-kinase/fructose-2 , 6-biphosphatase 3 ( Pfkfb3 ) in lung ECs and alcohol dehydrogenase 1 ( Adh1 ) in heart ECs . In contrast , fatty acid metabolism genes were most upregulated in heart ECs consistent with the heavy reliance of the heart on fatty acids to generate ATP ( Figure 1E ) . Heart ECs exhibited upregulation of 17 fatty acid metabolism genes whereas brain ECs and lung ECs only demonstrated upregulation of 9 and 4 metabolism genes , respectively . Regarding solute transport genes , the brain endothelium showed the most specific upregulation of genes when compared to ECs of the other tissues , both in terms of number of transporters as well as the magnitude of upregulation . We found that 141 transporter genes were upregulated in brain ECs , whereas 43 and 44 genes were upregulated in lung and heart ECs , respectively . As seen in the heatmap ( Figure 1F ) , the expression levels of brain EC-specific transporters were far greater than those of lung and heart ECs , indicative of the central role of solute transport regulation in brain EC function . After confirming the efficiency of the RiboTag immunoprecipitation protocol using qPCR , we next sought to perform an unbiased and systematic analysis of the utility of the RiboTagEC model as a tool to study the organ-specific endothelial translatome heterogeneity . We therefore compared organ-specific RiboTagEC RNA-Seq baseline profiles to healthy whole-tissue RNA-Seq profiles obtained from publicly available whole tissue RNA-Seq datasets ( Athar et al . , 2019 ) . By applying normalization and batch correction techniques , we were able to directly compare the mRNA expression levels of RiboTagEC endothelial samples with those of whole tissue samples . To characterize the whole brain , lung , and heart samples , we identified the genes that were significantly upregulated in each of the tissues and generated a heatmap displaying the 1358 differentially upregulated whole brain-specific genes relative to whole lung and whole heart ( Figure 1—figure supplement 2A ) . By performing a gene set enrichment analysis ( GSEA ) to ascertain the pathways associated with these genes , we confirmed the validity of the samples because the top pathways included ‘neurotransmitter transport’ , ‘synapse organization’ , ‘synaptic vesicle cycle’ ( Figure 1—figure supplement 2B ) . The top 10 most abundant genes in the whole brain RNA-Seq data included myelin basic protein ( Mbp ) , proteolipid protein 1 ( Plp1 ) , calmodulin 1 ( Calm1 ) , synaptosome associate protein 25 ( Snap25 ) , kinesis family member 5A ( Kif5a ) , ATPase Na+/K+ transporting subunit alpha 3 ( Atp1a3 ) , sodium-dependent glutamate/aspartate transporter 2 ( Slc1a2 ) , secreted protein acidic and cysteine rich ( Sparcl1 ) , carboxypeptidase e ( Cpe ) , stearoyl-coA desaturase 2 ( Scd2 ) ( Figure 1—figure supplement 2C ) . Whole lung samples were characterized by 1071 differentially expressed genes ( Figure 1—figure supplement 3A ) on which we performed GSEA ( Figure 1—figure supplement 3B ) . The top 10 most abundant genes in the whole lung were desmoyokin ( Ahnak ) , microtubule-actin crosslinking factor 1 ( Macf1 ) , actin beta ( Actb ) , surfactant protein c ( Sftpc ) , spectrin beta , non-erythrocytic 1 ( Sptbn1 ) , hypoxia inducible factor two alpha ( Hif2a ) , stearoyl-CoA desaturase ( Scd1 ) , filamin a ( Flna ) , adhesion g protein-coupled receptor f5 ( Adgrf5 ) , and ldl receptor related protein 1 ( Lrp1 ) ( Figure 1—figure supplement 3C ) . The signature of the whole heart derived from differential gene expression analysis was composed of 1351 genes ( Figure 1—figure supplement 4A ) . GSEA indicated a preponderance of metabolic and muscle contraction pathways ( Figure 1—figure supplement 4B ) . The top 10 most abundant cardiac genes were myosin heavy chain 6 ( Myh6 ) , ATPase sarcoplasmic/endoplasmic reticulum Ca2+ transporting 2 ( Atp2a2 ) , myoglobin ( Mb ) , actin , alpha , cardiac muscle 1 ( Actc1 ) , phospholamban ( Pln ) , myosin regulatory light chain 2 ( Myl2 ) , titin ( Ttn ) , troponin t2 , cardiac type ( Tnnt2 ) , tropomyosin 1 ( Tpm1 ) , and lipoprotein lipase ( Lpl ) ( Figure 1—figure supplement 4C ) . After establishing and confirming the molecular signatures of the whole brain , whole lung , and whole heart tissue , we next calculated a Kendall’s Tau correlation coefficient to assess the rank correlation between the RiboTagEC samples and the whole tissue samples . We surmised that if the rank of the most abundant whole tissue genes was the same as the rank of these genes in the RiboTagEC samples , then it would indicate possible contamination of the EC samples with whole tissue mRNA; however , if the abundance rank order of whole tissue genes was quite distinct from that in the RiboTagEC samples , then it would indicate tissue specific programming of ECs in situ ( Figure 1—figure supplement 5A ) . We assessed the Kendall’s Tau rank correlation for all three tissues and plotted correlation heatmaps showing the results ( Figure 1—figure supplement 5B–D ) . Our findings indicate that there was no significant correlation between the abundance rank of whole tissue genes and their rank order in the RiboTagEC samples . The rank correlation in the brain samples ranged from −0 . 29 to 0 . 38 ( Figure 1—figure supplement 5B ) . Since the cellular composition of the lung is 40–50% endothelial , we expectantly saw a higher rank correlation between whole lung samples and lung RiboTagEC samples , ranging between 0 . 02 and 0 . 6 ( Figure 1—figure supplement 5C ) . In the heart , we found a range of rank correlations between −0 . 29 to 0 . 24 ( Figure 1—figure supplement 5D ) . These results provide mathematical evidence for the robustness and purity of the RiboTagEC samples . After confirming the robustness and purity of the RiboTagEC samples , we performed differential expression analysis to identify the significantly upregulated genes in the brain endothelial translatome ( Figure 2A , Supplementary file 1 ) . We used these upregulated genes as the input into GSEA to characterize the brain ECs ( Figure 2B ) . Surprisingly , we found that genes involved in processes typically thought of being canonical neuronal functions such as synapse organization , neurotransmitter transport , axon development , and regulation of ion transmembrane transport were significantly enriched in brain ECs ( Figure 2B ) . The top 10 most significantly upregulated genes in the brain ECs included: prostaglandin d synthase ( Ptgds ) , ATPase , Na+/K+ transporting , alpha two polypeptide ( Atp1a2 ) , basigin ( Bsg ) , apolipoprotein e ( Apoe ) , glutamate-ammonia ligase ( Glul ) , apolipoprotein d ( Apod ) , pleiotrophin ( Ptn ) , insulin like growth factor 2 ( Igf2 ) , osteonectin ( Spock2 ) , and glucose transporter 1 ( Slc2a1 ) ( Figure 2C ) . In order to identify brain EC-specific surface markers , which could be of great value for therapeutic targeting of brain ECs , we used the Cell Surface Protein Atlas database ( Bausch-Fluck et al . , 2015 ) and identified the top 10 surface markers for brain ECs ( Figure 2D ) , which included the glutamate/aspartate transporter II ( Slc1a2 ) , thyroxine transporter ( Slco1c1 ) , glial fibrillary acidic protein ( Gfap ) , ATPase Na+/K+ transporting subunit alpha 3 ( Atp1a3 ) , endothelin b receptor-like protein 2 ( Gpr37l1 ) , Delta/Notch like EGF repeat containing transmembrane ( Dner ) , synaptic vesicle glycoprotein 2b ( Sv2b ) , sodium voltage-gated channel beta subunit 2 ( Scn2b ) , glutamate ionotropic receptor NMDA type subunit 2a ( Grin2a ) , and neurofascin ( Nfasc ) . Individual boxplots for the log2 expression levels of each gene show that the expression levels of these cell surface markers are 6–8 log2 units higher in brain ECs than in the lung and heart endothelium . We freshly isolated individual ECs , performed a cytospin and stained for the neurotrophic factor PTN and found that it was expressed on individual brain ECs but at much lower levels in heart or lung ECs ( Figure 2E ) . We next analyzed the lung EC signature using differential expression analysis ( Figure 3A ) . We found that the lung endothelium exhibits significant upregulation of genes involved in biological processes related to immune function such as leukocyte cell-cell adhesion , T cell activation , leukocyte migration , and regulation of immune system processes ( Figure 3B ) . The 10 most significantly upregulated genes in lung ECs included surfactant protein c ( Sftpc ) , advanced glycosylation end-product specific receptor ( Ager ) , norepinephrine transporter ( Slc6a2 ) , chitinase-like protein 3 ( Chil3 ) , WAP four-disulfide cco domain 2 ( Wfdc2 ) , c-type lectin domain containing 7a ( Clec7a ) , mucin 1 ( Muc1 ) , resistin like alpha ( Retnla ) , lysozyme ( Lyz1 ) , homeobox a5 ( Hoxa5 ) ( Figure 3C ) . The top lung endothelial cell surface markers included norepinephrine transporter ( Slc6a1 ) , mucin 1 ( Muc1 ) , tumor necrosis factor c ( Ltb ) , prostaglandin transporter ( Slco2a1 ) , epithelial membrane protein 2 ( Emp2 ) , ATPase sarcoplasmic/endoplasmic reticulum Ca2+ transporting 3 ( Atp2a3 ) , epithelial cell adhesion molecule ( Epcam ) , leukocyte function-associated molecule one alpha chain ( Itgal ) , interleukin three receptor subunit alpha ( Il3ra ) , matriptase ( St14 ) ( Figure 3D ) . We validated our computational analysis by staining freshly isolated ECs for RAGE and found that RAGE was only expressed at significant levels in lung ECs but not heart or brain ECs ( Figure 3E ) . We then studied the differentially expressed genes in the heart endothelium ( Figure 4A , Supplementary file 3 ) . GSEA identified pathways specifically upregulated in heart ECs , as compared to brain and lung ECs ( Figure 4B ) . Strikingly , we found that the genes specifically upregulated in heart ECs were involved in processes such as cardiac muscle tissue development , myofibril assembly and cardiac contraction ( Figure 4B ) . This suggested that the cardiac endothelium expresses genes canonically thought to be cardiomyocyte genes , analogous to the expression of canonical neuronal genes in the brain endothelium . The top expressing heart EC genes included myosin regulatory light chain 2 ( Myl2 ) , myosin regulatory light chain 3 ( Myl3 ) , aquaporin 7 ( Aqp7 ) , ADP-ribosylhydrolase like 1 ( Adprhl1 ) , alpha 2-HS glycoprotein ( Ahsg ) , sodium-coupled nucleoside transporter ( Slc28a2 ) , xin actin binding repeat containing 2 ( Xirp2 ) , myoglobin ( Mb ) , Butyrophilin like 9 ( Btnl9 ) , creatine kinase , mitochondrial 2 ( Ckmt2 ) , leucine rich repeats and transmembrane domains 1 ( Lrtm1 ) , and fatty acid binding protein 4 ( Fabp4 ) ( Figure 4C ) . The top 10 heart EC surface marker genes included alpha 2-HS glycoprotein ( Ahsg ) , sodium-coupled nucleoside transporter ( Slc28a2 ) , titin ( Ttn ) , tumor necrosis factor receptor superfamily member 27 ( Eda2r ) , platelet glycoprotein 4 ( Cd36 ) , laminin subunit alpha 4 ( Lama4 ) , fibulin 2 ( Fbln2 ) , ectonucleotide pyrophosphatase/phosphodiesterase 3 ( Enpp3 ) , t-cadherin ( Cdh13 ) , steroid sensitive gene 1 ( Ccdc80 ) ( Figure 4D ) . We tested the heart EC specificity of AQP7 using confocal analysis on freshly isolated brain , lung , and heart ECs and found that AQP7 was robustly expressed in heart ECs but minimally expressed in brain and lung ECs ( Figure 4E ) . In light of the surprising findings that endothelial cells express genes typically associated with surrounding parenchymal cells such as cardiomyocytes or neuronal cells , we next used single cell RNA-Seq analysis to assess whether the RiboTagEC endothelial signatures are also found in individual endothelial cells by analyzing endothelial single-cell data from the Tabula Muris compendium ( Tabula Muris Consortium et al . , 2018 ) and the single cell RNA-Seq atlas of the brain and lung endothelium ( Vanlandewijck et al . , 2018 ) . Using expression of the endothelial genes Cd31 and Cdh5 as markers of ECs , we analyzed double positive cells for both markers in Tabula Muris brain , lung , and heart tissues and performed PCA to assess the extent of endothelial heterogeneity ( Figure 5A ) . The PCA plot partitioned the cells into groups defined by their tissue of origin , indicating a tissue-specific EC signature even at the single cell level . Similarly , we performed PCA on ECs in Betsholtz dataset ( which relied on Cd31 and Cldn5 as EC markers ) and also found that ECs similarly clustered according to their tissue of origin ( Figure 5B ) . We then used these two scRNA-Seq endothelial datasets for the three organs we had analyzed in our RiboTag experiments and intersected the differentially expressed genes for each organ-specific endothelial population . The intent of this was to ascertain which tissue-specific EC signature genes were present in the single cell datasets as well as our RiboTagEC dataset . We found that the shared brain EC signature across all three datasets ( Tabula MurisEC , BetsholtzEC and RiboTagEC ) for brain ECs was enriched for genes involved in ion transport , acid transport , synapse organization and neurotransmitter transport ( Figure 5C ) . This finding is consistent with the brain EC-specific enrichment of neuronal signaling pathways that had been identified by the RiboTagEC analysis ( Figure 2 ) . We also found that the genes specifically upregulated in the Tabula Muris and Betsholtz lung ECs were involved in T cell activation , TGFβ signaling , and antigen processing and presentation ( Figure 5D ) , again consistent with the ‘immune activation’ signature identified by the RiboTagEC analysis alone ( Figure 3 ) . Similarly , the shared upregulated genes in Tabula Muris single cell heart ECs were involved in processes such as cardiac muscle contraction , myofibril assembly and proliferation ( Figure 5E , Figure 4 ) . We next quantified the intersection of brain , lung and heart endothelial marker genes across the Tabula Muris , brain and lung EC atlas , and RiboTag datasets . For the brain endothelium , 40 of the Tabula Muris top 50 brain EC specific genes were also brain EC specific genes in the RiboTag dataset . In the Betsholtz dataset , 27 of the top 50 brain EC specific genes were present in the RiboTag brain EC specific genes ( Figure 5F ) . We found that 17 of the top lung endothelial specific genes in the Betsholtz data set were also found in the list of lung endothelial-specific genes in the RiboTag model ( Figure 5G ) . Of the 24 top lung endothelial specific genes found in the Tabula Muris data set , the same genes were also found in the list of lung endothelial-specific genes in the RiboTag model ( Figure 5G ) . To address further that the parenchymal signatures ( Supplementary files 4–6 ) identified in the endothelial translatome were simply not driven by low abundance of transcripts , we performed a Spearman correlation analysis to compare organ-matched RiboTag bulk RNA-Seq data with scRNA-Seq data generated by the Betsholtz and the Tabula Muris Compendium ( Figure 6 , Figure 6—figure supplement 1 ) . In each dataset , we first determined the fold change for all genes using a housekeeping gene , Sap30l which we identified as being stably expressed across all datasets , and thus ideally suited to perform relative abundance comparisons ( Supplementary files 7–9 ) . Using the fold change values , we calculated the correlation coefficients between the brain endothelial translatome and single cell brain ECs from the Betsholtz and Tabula Muris datasets . We found that the correlation between RiboTag and Betsholtz was 0 . 53 for all genes detected in the brain endothelium ( Figure 6A ) while the correlation between RiboTag and Tabula Muris was 0 . 47 ( Figure 6—figure supplement 1A ) . We then specifically tested whether the parenchymal signature genes in the brain endothelium were correlated with the Betsholtz and Tabula Muris individual brain ECs . The correlation of the parenchymal gene expression between RiboTag brain EC samples and Betsholtz brain ECs was 0 . 31 ( Figure 6B ) while with Tabula Muris brain ECs the correlation was 0 . 28 ( Figure 6—figure supplement 1B ) . Importantly , the brain EC parenchymal genes including synaptosome associated protein 47 ( Snap47 ) and synaptotagmin 11 ( Syt11 ) were expressed at similar or higher levels in the single cell brain ECs from the Betsholtz and Tabula Muris datasets than in the RiboTag brain EC samples ( Figure 6C ) . We performed identical analysis for the lung and heart endothelium ( Figure 6D–I , Figure 6—figure supplement 1 ) , and found that similar correlation values ranging between 0 . 37 to 0 . 68 . Of note , the heart endothelial gene expression was the most correlated organ across the distinct platforms ( Figure 6G–H ) . In the lung and heart endothelium , we also found that individual genes representing the parenchymal signature were expressed at similar or higher levels in the single cell samples ( Figure 6F , Figure 6G–I ) , such as the cardiac contractile gene Tropomyosin ( Tpm1 ) , which was expressed at higher levels in individual heart ECs from the Tabula Muris dataset . We next analyzed the dynamics of the EC inflammatory response in each tissue , focusing on the early response ( 6 hr post systemic LPS ) and late response ( 24 hr post systemic LPS ) . At these time points , we identified the genes most upregulated by inflammatory injury in each tissue ( Figure 8—figure supplement 1 ) . In the brain endothelium , we identified several differentially expressed acute inflammatory factors including selectins , chemokine receptors , and interleukins which were strongly activated 6 hr post LPS treatment ( Figure 7A–C ) . We analyzed the kinetics during the entire time course for the early inflammatory brain endothelial specific genes such as eosinophil chemotactic protein ( Ccl11 ) ( Figure 7C ) and found that Ccl11 is markedly upregulated at the 6 hr time point and remains significantly higher in the brain endothelium , but by one week post LPS injection the expression level returns to the same level as that seen in lung and heart endothelium . In the lung endothelium , we discovered that the most upregulated inflammatory pathways included chemokines , response to cellular stress , hematopoiesis genes and early immune response mediators ( Figure 7D–F ) . Lymphocyte antigen 96 ( Ly96 ) was strongly upregulated ( Figure 7D ) whereas the apoptosis gene caspase 6 ( Casp6 ) was markedly downregulated 6 hr post LPS treatment and remained lower in lung ECs than in brain or heart ECs throughout the injury period ( Figure 7F ) . In heart ECs , leukocyte migration and neutrophil activation pathways were most upregulated by inflammatory injury ( Figure 7G–I ) . At 24 hr post injury , we found the peak upregulation of inflammatory genes ( Figure 8 ) with a substantial overlap of the inflammatory response pathways , predominantly associated with neutrophil and leukocyte chemotaxis and migration , in the brain ( Figure 8A–C ) , lung ( Figure 8D–F ) , and heart ECs ( Figure 8G–I ) . After establishing the baseline heterogeneity of brain , lung and heart ECs , we next studied the dynamics of the organ-specific baseline endothelial signature during systemic inflammation , we collected translatome data of the brain , lung , and heart endothelium at several time points following LPS treatment . By computationally analyzing RiboTagEC mRNA from brain , lung , and heart at 0 hr , 6 hr , 24 hr , 48 hr , 72 hr , and 168 hr post-LPS administration , we were able to identify tissue-specific molecular mechanisms modulated in endothelial injury , repair , and regeneration . We first investigated the tissue-specific baseline signatures over time in order to address the question of whether the baseline core endothelial functions were disrupted during inflammatory activation . The time-course of the brain endothelium specific endothelial genes were plotted to compare their kinetics to the lung and heart endothelium ( Figure 9A ) . We found that selected genes which constitute the tissue-specific EC signature during homeostasis are modulated during inflammatory injury . For instance , the expression level of von Willebrand factor A domain containing protein 1 ( Vwa1 ) which we found to be a brain endothelial gene during homeostasis decreases during early and late inflammation and then returns to baseline levels one-week post LPS injury , whereas its levels in lung and heart endothelium remain relatively low during the entire time course . On the other hand , there are signature genes such as glucose transporter protein 1 ( Slc2a1 ) which is consistently upregulated in brain ECs throughout the post-injury period . From the analysis of the lung endothelium specific endothelial genes heatmap ( Figure 9B ) , it is apparent that expression of nearly all the canonical endothelial genes drastically decrease during the early and late inflammatory time points . This is an important finding because it suggests that the lung endothelium experiences the most profound dysregulation of core endothelial genes following LPS injury . We also identified lung endothelial specific genes which are solely modulated in the lung endothelium during the inflammatory time course . For instance , the expression levels of forkhead-related transcription factor 1 ( Foxf1 ) and tetraspanin8 ( Tspan8 ) significantly decrease in the lung endothelium at 6 hr and 24 hr post LPS treatment and then gradually recover back to baseline levels , but both genes remain minimally expressed in the brain and heart endothelium . The endothelial genes which were specifically upregulated in the heart endothelium at baseline do not appear to be affected to the extent that the brain and lung endothelium were during LPS stimulation . In the heatmap ( Figure 9C ) , a few genes such as Rho family GTPase 1 ( Rnd1 ) and platelet glycoprotein ( Cd36 ) undergo a robust change in expression during the time course . From our analysis , we found that the endothelial genes specific to the heart endothelium are much more abundant in the heart versus the other tissues . For example , caveolin 1 ( Cav1 ) and vascular endothelial growth factor receptor 2 ( Kdr ) maintained a high expression level in the heart endothelial samples during the entire LPS time course whereas in the brain and lung endothelial samples , we see significantly lower expression . We next focused of the organ-specific endothelial glycolysis signature to investigate the tissue-specific dynamics of glycolytic genes . The brain endothelial basal translatome upregulated the greatest number of glycolytic genes compared to the lung and heart endothelium . Interestingly , when we analyzed the time course of these brain endothelial specific glycolysis genes , we found that they maintain similar levels during the progression and resolution of inflammation ( Figure 9—figure supplement 1A ) . There were only three glycolysis-related genes which were upregulated in the lung endothelium . When we analyzed these three genes over time , we found that two of them remained stable whereas 6-phosphofructo-2-kinase/fructose-2 , 6-biphosphatase 3 ( Pfkfb3 ) was dynamic in all three tissues . Even though this glycolysis regulatory enzyme was specifically upregulated in the lung endothelium at baseline , we found that it was activated in all tissues during late inflammation/early repair and then returned to baseline levels ( Figure 9—figure supplement 1B ) . In the heart endothelium , we found that the upregulated glycolytic genes were not modulated during the LPS injury and recovery ( Figure 9—figure supplement 1C ) . The endothelium which lines the entire vasculature evolves in a tissue-dependent manner during embryonic development to control organ development , homeostasis , and tissue regeneration ( Augustin and Koh , 2017 ) . Under normal physiological conditions , the endothelium maintains a quiescent interface between the blood and tissue . During inflammatory stimulation , the endothelium becomes actively responsible for controlling blood flow , vascular permeability , leukocyte infiltration , and tissue edema ( Pober and Sessa , 2015 ) . Understanding the organotypic endothelial heterogeneity that exists at baseline as well as during the transition from the normal state to the inflammatory state is essential for understanding endothelial plasticity in homeostasis and tissue-specific responses to inflammation ( Chaqour et al . , 2018; Dejana et al . , 2017; Krenning et al . , 2016; Malinovskaya et al . , 2016 ) . The RiboTag strategy was originally applied to expression profiling of neurons and Sertoli cells ( Sanz et al . , 2009 ) . Cell type specificity of the approach depends on the accuracy of the Cre driver that is combined with the Rpl22HA allele . This aspect is highlighted in our study and we revealed the precision of the inducible system for achieving endothelial specificity . Our results demonstrate that the RiboTag approach provides a useful method to identify distinct molecular gene expression signatures of tissue-specific endothelium . Performing high-throughput gene expression analysis on the translatome using the RiboTag approach enabled us to establish tissue-specific molecular signatures underlying in situ endothelial heterogeneity . During homeostasis , we found that the endothelial translatome in each organ is uniquely characterized by a signature adapted to the surrounding parenchymal tissue . The metabolic adaptation of the endothelium is less surprising as the endothelium plays a critical role in supplying nutrients to the host tissue ( Malinovskaya et al . , 2016; Hamuro et al . , 2016 ) . The upregulation of the glucose transporter 1 ( Slc2a1 ) in brain ECs is consistent with the massive glucose consumption of the brain ( Schuenke et al . , 2017 ) , whereas the upregulation of the fatty acid metabolism genes Cd36 and Fabp4 in the heart likely reflects the importance of fatty acids to meet the bioenergetic demands of cardiomyocytes ( Elmasri et al . , 2009; Silverstein and Febbraio , 2009 ) . Similarly , the upregulation of immune and stress response genes in the lung endothelium is expected due to the lung’s continuous exposure to environmental stressors and pathogens contained in the inhaled air ( Al-Soudi et al . , 2017; Kaparakis-Liaskos and Ferrero , 2015 ) . However , the adaptation of the endothelium appears to extend far beyond the supply of metabolites and nutrients to the parenchyma . We surprisingly found that there exists a multidirectional molecular cross-talk of vessel wall cells with the cells of their microenvironment . In the brain endothelium , synapse organization and neurotransmitter transport genes such as Glul were highly enriched , which discloses the molecular mechanisms underlying how excitatory neurotransmitters such as glutamate can be transported among brain endothelial cells , neurons , and astrocytes ( Hawkins , 2009 ) . We also found that lung ECs expressed genes typically found in the lung epithelium such as Surfactant Protein C ( Spc ) and Mucin1 ( Muc1 ) , again indicative of a key interaction of the lung endothelium with the lung parenchymal epithelium . The upregulation of genes involved in cardiomyocyte contraction such as Myl2 and Ckmt2 again points to an unexpected adaptation of the cardiac endothelium to the surrounding cardiomyocytes , possibly suggesting a key role for the endothelium in modulating cardiac contractility ( Cai et al . , 1998; Schnittler et al . , 1990 ) . Studying endothelial heterogeneity in response to the systemic inflammatory stress induced by LPS , we found that the endothelium in each tissue maintains a distinct organ-specific molecular identity . Brain and heart ECs express classical inflammatory adhesion molecules such as E-Selectin and P-Selectin , whereas lung ECs upregulate chemokines such as Cxcl1 and Cxcl9 . The gene expression shifts in the lung may also reflect the severe loss of lung endothelium recently observed during endotoxemia ( Merle et al . , 2019 ) . The marked upregulation P-Selectin in the heart and brain is especially interesting because P-Selectin is a key mediator of thrombosis and platelet aggregation ( Merle et al . , 2019 ) , and both the brain and heart are especially vulnerable to thrombotic events . During the later stage of inflammation at 24 hr , the inflammatory gene expression pathways across all tissues demonstrated significant upregulation of leukocyte migration and chemotaxis genes , suggesting that despite the persistent heterogenous signatures of the ECs in the respective organs , there is a broad shared program of inflammatory signaling pathways in response to systemic endotoxemia . One of the requisites for targeted therapies is the need to deliver such agents to specific organs , thus underscoring the importance of leveraging organ-specific endothelial heterogeneity for such approaches . It has been suggested that vascular endothelial cells in different organs or disease states express specific markers , or ‘zip codes’ ( Folkman , 1999 ) , so that ligands directed against organ-specific vascular endothelial cell surface markers could be used to deliver effector molecules to specific vascular beds . To address this concept , we expanded our analysis by analyzing 1296 cell surface glycoproteins , including 136 G-protein coupled receptors and 75 membrane receptor tyrosine-protein kinases . This allowed us to establish EC surface markers that were specifically upregulated in in each vascular bed . Not only was this integrative analysis valuable for the establishment of EC ‘zip codes’ based on the organs they are derived from , but it may also provide insights about tissue-specific cell-cell contacts of ECs that allow them to interact with niche or parenchymal cells in each tissue ( Maoz et al . , 2018; Zamani et al . , 2018 ) . Among the most intriguing findings of our study was the prominent ‘parenchymal’ signature of endothelial cells in each organ such as contractile genes in the cardiac endothelium and neurotransmitter transport or synaptic vesicle genes in the brain endothelium . A rank-based statistical analysis demonstrated that only selected genes of surrounding parenchymal cells were expressed in the endothelium of each organ . In the setting of a possible contamination , the most abundant genes expressed in the surrounding cells would also be the most abundant genes found in the cell of interest . That the rank order of parenchymal genes abundance in the endothelium differed from that found in the parenchyma suggests tissue-specific programming and adaptation of the endothelium . To further address the concern of possible mRNA contamination by neighboring cells in the RiboTagEC data , we systematically analyzed two independent endothelial single cell RNA-Seq datasets ( Vanlandewijck et al . , 2018; Tabula Muris Consortium et al . , 2018 ) , which can exclude contaminating tissue cells by examining the identity of each sequenced cell . We found that EC signature genes identified by our RiboTagEC approach such as the synaptic vesicle gene Snap47 and cardiac contractile gene Tropomyosin were also expressed in individual brain and heart ECs as identified by scRNA-Seq . Importantly , we found a substantial overlap of individual signature genes across our data and both scRNA-Seq datasets . Even though the approaches to obtain the data were so different , this is a remarkable degree of consilience . We used a genetic VE-cadherin-Cre to label endothelial ribosomes whereas the Tabula Muris scRNA-Seq dataset relied on mRNA markers of endothelial cells and Betsholtz dataset used Claudin5 lineage tracing combined with endothelial gene expression markers to identify individual ECs . Although the bulk of scRNA-Seq tissue-specific genes were found in the Ribotag dataset , the converse was not true . Not all RiboTagEC signature genes were present in the single cell RNA-Seq datasets . We think this likely reflects the greater depth and sensitivity of Ribotag RNA-Seq because current single cell technologies are limited in their ability to detect the expression of individual genes in a given cell ( Bacher and Kendziorski , 2016; Zhu et al . , 2018; Kharchenko et al . , 2014; Lun et al . , 2016; Vallejos et al . , 2017 ) . Not all single ECs expressed parenchymal genes such as Tropomyosin or Snap47 but those expressing them did so at an even higher levels than what we found in the RiboTagEC data . The reason for this might be that RiboTagEC data represent an aggregate of all ECs in a tissue . It is therefore possible that the tissue adaptation of individual ECs may be most prominent in anatomically distinct ECs , for example those in close proximity to parenchymal cells such as neurons and astrocytes . Furthermore , if the expression of parenchymal gene signatures such as synaptic vesicle genes or cardiac contractile genes in the endothelium is dependent on environmental cues from neighboring cells or the extracellular matrix , the disassociation of the cells required for single cell RNA-seq may have further reduced mRNA levels of these genes ( Haimon et al . , 2018; Rossner et al . , 2006; Sugino et al . , 2006 ) . Sequencing a larger number of individual ECs in these tissues may enable identification of additional EC subsets with the most prominent parenchymal signatures , and a single cell sequencing approach that preserves the anatomy of the tissue such as Slide-Seq ( Rodriques et al . , 2019 ) may also be useful to address the in situ transcriptomic signature . Using the RiboTag model , we were able to characterize the endothelial translatome profile from distinct tissues . Our analysis uncovered a previously unrecognized degree of endothelial plasticity and adaptation to the parenchymal tissues , raising intriguing questions about the role that the endothelium plays in modulating parenchymal tissue function that likely go far beyond the classically ascribed roles of supplying oxygen , metabolites and solutes . Further studies such as endothelial-specific deletion of neurotransmitter transport or cardiac contractile genes will be required to establish the functional roles of these tissue-specific genes expressed in the endothelium of each organ . Understanding the biological significance of endothelial plasticity and adaptation to the parenchyma will be important in providing a fuller picture of endothelial function during homeostasis and stress in each tissue . RiboTag ( Rpl22HA/+ ) mice were purchased from Jackson Labs . Endothelial-specific VE-cadherin-Cre mice were provided by Dr . Ralf Adams . We crossed the RiboTag mice ( Rpl22HA/+ ) ( Sanz et al . , 2009 ) with the endothelial-specific VE-cadherin-Cre mice ( Jeong et al . , 2017; Sörensen et al . , 2009 ) to generate RiboTagEC ( Cdh5CreERT2/+; Rpl22HA/+ ) mice . Following tamoxifen-induced recombination at week 4 , HA-tagged Rpl22 was specifically expressed in endothelial cells . To investigate the mechanisms of organ-specific EC injury , repair , and regeneration we performed RNA-Seq analysis of gene expression in ECs isolated at 6 hr , 24 hr , 48 hr , 72 hr , and 1 week post-LPS challenge ( 10 mg/kg LPS i . p . , Sigma-Aldrich Cat#: L2630 ) with PBS-injected mice serving as controls . The C57BL/6J mice were purchased from the Jackson Laboratory . All animal experiments were conducted in accordance with NIH guidelines for the Care and Use of Laboratory Animals and were approved by the IACUC of the University of Illinois ( IACUC Protocol #19–014 , IACUC Protocol #13–175 and IACUC Protocol #16–064 ) . After surgically opening the mouse chest , the brain , lung and heart were harvested after a one-time perfusion of 20 mL PBS through the left and right ventricular chamber . The tissue samples were extracted from RiboTagEC mice , flash-frozen in liquid nitrogen and then stored at −80°C . The samples were then homogenized on ice in ice-cold homogenization buffer ( 50 mM Tris , pH7 . 4 , 100 mM KCl , 12 mM MgCl2 , 1% NP-40 , 1 mM DTT , 1:100 protease inhibitor ( Sigma ) , 200 units/mL RNasin ( Promega ) 1 mg/mL heparin and 0 . 1 mg/mL cycloheximide ( Sigma ) in RNase free DDW ) 10% w/v with a Dounce homogenizer ( Sigma ) until the suspension was homogeneous . To remove cell debris , 1 mL of the homogenate was transferred to an Eppendorf tube and was centrifuged at 10 , 000xg and 4°C for 15 min . Supernatants were subsequently transferred to a fresh Eppendorf tube on ice , then 100 μL was removed for ‘input’ analysis and 3 μL ( =3 µg ) of anti-HA antibody ( ab9110 , Abcam ) or 3 μL ( =1 µg ) of mouse monoclonal IgG1 antibody ( Sigma , Cat# M5284 ) or 6 μL anti-RPL22 ( Invitrogen Cat# PA5-68320 ) was added to the supernatant , followed by 1 hr of incubation with slow rotation in a cold room at 4°C . Meanwhile , Pierce Protein A/G Magnetic Beads ( Thermo Fisher Scientific ) , 100 μL per sample , were equilibrated to homogenization buffer by washing three times . At the end of 1 hr of incubation with antibody , beads were added to each sample , followed by incubation 1 hr in cold room at 4°C . After that , samples were washed three times with high-salt buffer ( 50 mM Tris , 300 mM KCl , 12 mM MgCl2 , 1% NP-40 , 1 mM DTT , 1:200 protease inhibitor , 100 units/mL RNasin and 0 . 1 mg/mL cycloheximide in RNase free DDW ) , 5 min per wash in a cold room on a rotator . At the end of the washes , beads were magnetized , and excess buffer was removed , 350 µL Lysis Buffer was added to the beads and RNA was extracted with RNeasy plus Mini kit ( Qiagen ) . RNA was eluted in 30 μL H2O and taken for RNA-Sequencing . RNA quality and quantity were assessed using an Agilent Bio-analyzer . RNA-Seq libraries were prepared using Illumina mRNA TruSeq kits as protocolled by Illumina . Library quality and quantity were checked using an Agilent Bio-analyzer and the pool of libraries was sequenced using an Illumina HiSeq4000 and Illumina reagents . The sequenced reads from all samples were aligned to the mouse ( mm10 ) reference genome with STAR v . 2 . 4 . 2 ( Dobin et al . , 2013 ) , and the aligned reads were used to quantify mRNA expression by using HTSeq-count v . 0 . 6 . 1 ( Anders et al . , 2015 ) . Gene symbols were mapped to the ENSEMBL features using the biomaRt package v . 2 . 26 . 1 ( Durinck et al . , 2009 ) . Preliminary unsupervised analysis of normalized and processed profiles by principal component analysis ( PCA ) revealed separation into three major clusters . These clusters largely corresponded to the distribution of samples by sequencing batch . Consistent with the PCA plots , the distribution of samples by sequencing batch differed significantly but not by time point after inflammatory treatment or tissue type . To better harmonize profiles prior to analyses reported here , we normalized expression data of all samples using ComBat ( Johnson et al . , 2007 ) . This correction ameliorated the separation by sequencing batch without substantially affecting distributions by time point after inflammatory treatment and tissue type . We calculated the differential expression level of genes using a one versus others approach in order to identify signature genes which were upregulated for each tissue at baseline . For instance , to identify the genes significantly upregulated in brain ECs at baseline , we compared the 0 hr brain EC samples to 0 hr lung ECs and 0 hr heart ECs . We performed these analyses for all three tissues to identify baseline organ-specific EC signatures . We utilized the limma R package and applied the standard limma pipeline ( Ritchie et al . , 2015 ) to RNA-Seq data after voom transformation ( Law et al . , 2014 ) . For each gene , the log fold-change ( logFC ) in expression level is derived from the limma analysis . Genes with FDR < 0 . 05 were identified as being differentially expressed . All upregulated genes for each tissue were plotted using the heatmap . 2 function from the gplots v . 3 . 0 . 1 . 1 ( Warnes , 2011 ) R package . The top 10 significantly differentially expressed genes by logFC were listed . To define the biological function associated with the molecular signature of the tissue-specific ECs , we specifically performed gene set enrichment analysis ( GSEA ) ( Subramanian et al . , 2005 ) on the genes which were significantly upregulated ( logFC >1 ) in the tissue of interest . GSEA was performed on significantly upregulated genes ranked by their p-value using the clusterProfiler package ( Yu et al . , 2012 ) in R with gene ontology ( GO ) gene sets downloaded from the Molecular Signatures Database ( MSigDB ) ( Liberzon et al . , 2015 ) . The top 20 most enriched GO terms were plotted . Tissue-specific cell surface markers were identified by intersecting tissue-specific differentially expressed genes with predicted cell surface markers , as reported in the Cell Surface Protein Atlas ( www . proteinatlas . org ) ( Bausch-Fluck et al . , 2015 ) . The top 10 significantly differentially expressed cell surface proteins by logFC were plotted . The C57BL/6J mice mouse lungs were minced and digested with 3 mL collagenase A at 1 mg/mL in PBS ( Roche , Cat#: 10103586001 ) at 37°C water bath for 1 hr . Mixtures were titrated with #18 needles and then pipetted through a 40 μm disposable cell strainer . After centrifuging 500xg for 5 min and washing with 1x PBS , the isolated cells were treated with red blood cell lysis buffer ( Biolegend , Cat#: 420301 ) for 5 min . After washing with 1x PBS twice , cells were incubated in suspension buffer ( Ca2+ and Mg2+ free PBS , 0 . 5% BSA , 4 . 5 mg/mL D-glucose , and 2 mM EDTA ) with 5 µg anti-CD31 antibody ( BD Pharmingen , Cat#: 553370 ) at 4°C for 60 min with gentle tilting and rotation . After washing , cells were then incubated in suspension buffer with pre-washed Dynabeads ( 20 µL beads in 1 mL buffer , Invitrogen Cat#: 11035 ) at 4°C for 60 min with gentle tilting and rotation . After washing with 1x PBS three times using magnetic separation , lung ECs were dissociated from magnetic beads with trypsin . The forebrains of C57BL/6J mice were micro dissected and minced in collagenase/dispase ( Roche , Cat#: 11097113001 ) and DNAse ( Worthington Biochemical Cat#: LK003170 ) and incubated for 1 hr at 37°C . Myelin Removal Beads ( Miltenyl Biotec , Cat#: 130-096-433 ) and LS columns ( Miltenyl Biotec , Cat#: 130-042-401 ) were used . The resulting pellet after myelin removal contained microglia , astrocytes and endothelial cells . The endothelial cells were further enriched by using CD31 microbeads ( Miltenyl Biotec , Cat#: 130-097-418 ) . Isolated C57BL/6J mice hearts were minced and digested with prewarmed Collagenase/Dispase mix ( 1 mg/mL ) ( Roche ) at 37°C for 30 min . 75 µL DNAse I per 10 mL cell suspension ( 1 mg/mL ) was added and the suspension was incubated at 37°C for 30 min . The digested tissue was filtered using 70 µm cell strainer followed by RBC lysis in RBC lysis buffer ( Biolegend , Cat#: 420301 ) for 7 min at room temperature . The cell suspension was diluted with 10 mL of MACS buffer ( Prepared in phosphate-buffered saline ( PBS ) , pH 7 . 2 , 0 . 5% bovine serum albumin ( BSA ) , and 2 mM EDTA ) by diluting MACS BSA Stock Solution ( Cat#: 130-091-376 ) 1:20 with autoMACS Rinsing Solution ( Cat#: 130-091-222 ) ) and cells were passed through 40 µm cell strainer followed by centrifugation at 500xg for 5 min to pellet the cardiomyocytes . The supernatant containing endothelial cells was centrifuged at 800xg for 5 min to pellet down the ECs . The endothelial cell enriched pellet was resuspended in 500 µL of MACS buffer and the isolated cells were counted . Endothelial cells were further purified by using CD31 microbeads ( Miltenyl Biotec , Cat#: 130-097-418 ) and Miltenyl Biotec MS columns ( Miltenyl Biotec , Cat#: 130-042-201 ) through affinity chromatography according to the manufacturer’s protocol . The Thermo Shandon Cytospin three was used to generate Cytospin slides . Briefly , the Cytoslide with filter card were inserted into a Cytoclip . The Cytoclip was fastened and placed in a recess of the Cytospin rotor after sliding a Cytofunnel into it . The required volume of the cell suspension was pipetted into the Cytofunnel after cell counting and calculation . The Cytospin was centrifuged for 500 rpm for 5 min . The slide was fixed with 4% paraformaldehyde for 10 min and stored in 1x PBS at 4°C . The slides were permeabilized and blocked with 10% donkey serum , 2% BSA , 0 . 05% tween in PBS for 1 hr at room temperature . For lung cells , the slides were incubated with primary antibodies anti-CD31 ( BD Pharmingen , Cat#: 550274 , 1:25 ) and anti-RAGE ( Abcam , Cat#: Ab3611 , 1:3200 ) at 4°C overnight . The brain ECs were incubated with primary antibodies anti-CD31 ( BD Pharmingen , Cat#: 550274 , 1:25 ) and anti-PTN ( Santa Cruz Biotechnology , Cat#: sc-74443 , 1:3200 ) at 4°C overnight . For the heart samples , primary antibodies anti-AQP7 ( Novus Biologicals , Cat#: NBP1-30862 , 1:3200 ) and anti-CD31 ( BD Pharmingen , Cat#: 550274 , 1:25 ) were used and incubated at 4°C overnight . The next day , slides were washed and incubated with the fluorescence-conjugated secondary antibody ( AF488 donkey anti-rat 1:300 , Invitrogen Cat#: A-21208; AF594 donkey anti-rabbit 1:300 , Invitrogen Cat#: A-21207; AF594 goat anti-mouse 1:300 , Invitrogen Cat#: A11032 ) , followed by washing with 1x PBS . Cells were stained with DAPI and mounted on ProLong Gold mounting medium ( Invitrogen , Cat#: P36934 ) . Images were taken with a confocal microscope LSM880 ( Zeiss ) and analyzed by Zen software ( Zeiss ) . Tissue-specific baseline gene expression heatmaps were generated for gene sets related to endothelial function including classical endothelial markers , glycolysis , fatty acid metabolism , and solute transport . The individual genes listed in the heatmaps contain the tissue-specific differentially expressed genes which overlapped with each of the respective gene sets . The classical endothelial markers gene set contains 152 mouse endothelial cell markers downloaded from PanglaoDB ( Franzén et al . , 2019 ) . The mouse glycolysis and fatty acid metabolism gene sets containing 67 and 52 genes respectively were downloaded from the Rat Genome Database ( RGD ) https://rgd . mcw . edu/ ( Shimoyama et al . , 2015 ) . For the transport gene set , the solute carrier family including 423 membrane transport proteins located in the cell membrane were downloaded from the HUGO Gene Nomenclature Committee database ( https://www . genenames . org/ ) ( Hediger et al . , 2013 ) . Due to the endothelial cells being surrounded by other tissue-resident cell types , it is likely that the mRNA isolated from endothelial-specific RiboTagEC samples could contain non-endothelial mRNA . For this reason , we assessed the mRNA purity of RiboTag endothelial samples isolated from whole tissue by comparing the gene expression levels of the endothelial-specific RiboTag samples to the gene expression levels of mRNA from whole tissue . We compared endothelial-specific RiboTagEC mRNA expression levels from brain , lung , and heart tissue to whole brain , lung , and heart mRNA expression levels . We first acquired RNA-Seq data for whole brain , whole lung , and whole heart tissue from Array Express ( Athar et al . , 2019 ) . The three whole brain samples and three whole lung samples were downloaded from accession number E-MTAB-6081 , while the three whole heart samples were downloaded from accession number E-MTAB-6798 . Raw mRNA counts were processed , and batch corrected in a cohort including the 0 hr RiboTag brain , lung , and heart endothelial mRNA counts . The preprocessing and batch correction were performed in the same manner as described above . To identify whether mRNA of tissue-resident cells was isolated during the RiboTag EC mRNA isolation procedure , we calculated a Kendall’s Tau rank coefficient between the most abundant genes in the RiboTag EC mRNA and whole tissue mRNA . The Kendall’s Tau rank coefficient , ranging between −1 and 1 , allowed us to test whether there was contamination of mRNA from the whole tissue in the RiboTag EC samples . As the coefficient approaches −1 , the rank of most abundant genes differs in both sets of samples; while , as the coefficient approaches 1 , the rank of most abundant genes becomes identical . Using this test , we were able to infer that if the rank of the most abundant genes in the RiboTag EC sample and the whole tissue is identical , there is contamination of non-endothelial mRNA in the RiboTag EC mRNA samples . All samples were compared to each other and heatmaps with Kendall’s Tau rank coefficients were generated to visualize the results . To specifically analyze ECs at the single-cell level , we downloaded Tabula Muris data from https://github . com/czbiohub/tabula-muris and Betsholtz Lab data from NCBI Gene Expression Omnibus ( GSE99235 , GSE98816 ) . We filtered out non-ECs from the Tabula Muris brain , lung , and heart data based on Cd31 and Cdh5 expression . We selected ECs from the Betsholtz Lab brain and lung data based on Cd31 and Cldn5 expression . All genes that were not detected in at least 10% of all single cells were discarded . For all further analyses we used 2585 cells expressing 6802 genes from the Tabula Muris dataset and 873 cells expressing 8116 genes from the Betsholtz Lab dataset . Data were log transformed for all downstream analyses . We analyzed the data utilizing the Seurat R package ( https://github . com/satijalab/seurat; http://satijalab . org/seurat/ ) ( Butler et al . , 2018 ) . PCA analysis of organ-specific ECs was performed in each dataset separately using the ‘RunPCA’ function of the Seurat package ( Butler et al . , 2018 ) . Differential expression analysis for organ-specific endothelial cells was performed using a Wilcoxon rank-sum test as implemented in the ‘FindAllMarkers’ function of the Seurat package . GSEA was performed on significantly upregulated genes ranked by their p-value using the clusterProfiler package ( Yu et al . , 2012 ) in R with gene ontology ( GO ) gene sets downloaded from the Molecular Signatures Database ( MSigDB ) ( Liberzon et al . , 2015 ) . Cross-platform comparisons between bulk RNA-Seq data and scRNA-Seq data required computing the fold change of each gene relative to a housekeeping gene . We calculated the relative fold change by dividing the expression value for every gene in every sample by an invariable housekeeping gene . We chose Sap30l as the housekeeping gene because it was invariable in all three datasets . By generating the fold change matrix in all three datasets , we were then able to use these values to compare relative abundances for genes of interest . We next calculated Spearman’s correlation coefficients for all genes shared between the organ-specific endothelial translatome , Tabula Muris scRNA-Seq , and Betsholtz scRNA-Seq datasets , and then separately for all parenchymal ( non-endothelial ) genes . To ascertain the kinetics of the tissue-specific endothelial signatures during inflammation we analyzed the time-series RNA-Seq data with the gene sets referenced above: classical endothelial markers , glycolysis , fatty acid metabolism , and transport . To visualize the tissue-specific dynamics for predominant endothelial functions , we plotted a heatmap which includes the tissue-specific differentially expressed genes for each gene set . To identify the inflammatory genes that were upregulated in the LPS 6 hr samples as compared to the baseline samples , we applied the standard limma pipeline ( Ritchie et al . , 2015 ) for genes in the ‘inflammatory response’ gene ontology term ( GO:0006954 ) . The analysis was carried out on the tissue specific LPS treated samples against the baseline tissue-specific sample . Limma statistically evaluates each inflammatory gene and returns the genes which show statistically significant change between the inflammatory time point and baseline . We applied this approach to the early inflammation time point , 6 hr , and the late inflammatory time point , 24 hr . Heatmaps were generated to visualize the tissue-specific inflammatory genes and their kinetics . The endothelial translatome expression database is hosted on Amazon S3 . The website was constructed using Angular 8 . 0 , JavaScript , HTML5 , and CSS . Barplots and heatmaps were generated for genes of interest using Tableau Public . The visualizations were integrated into the web application using the Tableau JavaScript API . RiboTag log2 normalized baseline and inflammation time-course translatome expression data were uploaded to Tableu . The averages were computed using Tableau calculated fields . Tableau dashboards and workbooks were created to generate bar plots and heatmaps for online publishing .	Blood vessels supply nutrients , oxygen and other key molecules to all of the organs in the body . Cells lining the blood vessels , called endothelial cells , regulate which molecules pass from the blood to the organs they supply . For example , brain endothelial cells prevent toxic molecules from getting into the brain , and lung endothelial cells allow immune cells into the lungs to fight off bacteria or viruses . Determining which genes are switched on in the endothelial cells of major organs might allow scientists to determine what endothelial cells do in the brain , heart , and lung , and how they differ; or help scientists deliver drugs to a particular organ . If endothelial cells from different organs switch on different groups of genes , each of these groups of genes can be thought of as a ‘genetic signature’ that identifies endothelial cells from a specific organ . Now , Jambusaria et al . show that brain , heart , and lung endothelial cells have distinct genetic signatures . The experiments used mice that had been genetically modified to have tags on their endothelial cells . These tags made it possible to isolate RNA – a molecule similar to DNA that contains the information about which genes are active – from endothelial cells without separating the cells from their tissue of origin . Next , RNA from endothelial cells in the heart , brain and lung was sequenced and analyzed . The results show that each endothelial cell type has a distinct genetic signature under normal conditions and infection-like conditions . Unexpectedly , the experiments also showed that genes that were thought to only be switched on in the cells of specific tissues are also on in the endothelial cells lining the blood vessels of the tissue . For example , genes switched on in brain cells are also active in brain endothelial cells , and genes allowing heart muscle cells to pump are also on in the endothelial cells of the heart blood vessels . The endothelial cell genetic signatures identified by Jambusaria et al . can be used as “postal codes” to target drugs to a specific organ via the endothelial cells that feed it . It might also be possible to use these genetic signatures to build organ-specific blood vessels from stem cells in the laboratory . Future work will try to answer why endothelial cells serving the heart and brain use genes from these organs .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "immunology", "and", "inflammation" ]	2020	Endothelial heterogeneity across distinct vascular beds during homeostasis and inflammation
Nucleolar protein interacting with the FHA domain of pKi-67 ( NIFK ) is a Ki-67-interacting protein . However , its precise function in cancer remains largely uninvestigated . Here we show the clinical significance and metastatic mechanism of NIFK in lung cancer . NIFK expression is clinically associated with poor prognosis and metastasis . Furthermore , NIFK enhances Ki-67-dependent proliferation , and promotes migration , invasion in vitro and metastasis in vivo via downregulation of casein kinase 1α ( CK1α ) , a suppressor of pro-metastatic TCF4/β-catenin signaling . Inversely , CK1α is upregulated upon NIFK knockdown . The silencing of CK1α expression in NIFK-silenced cells restores TCF4/β-catenin transcriptional activity , cell migration , and metastasis . Furthermore , RUNX1 is identified as a transcription factor of CSNK1A1 ( CK1α ) that is negatively regulated by NIFK . Our results demonstrate the prognostic value of NIFK , and suggest that NIFK is required for lung cancer progression via the RUNX1-dependent CK1α repression , which activates TCF4/β-catenin signaling in metastasis and the Ki-67-dependent regulation in cell proliferation . The Ki-67 protein plays a specific but enigmatic role in cancer . Since its discovery as an antigen for a monoclonal antibody against a Hodgkin lymphoma cell line more than 30 years ago ( Gerdes et al . , 1983; Gerdes et al . , 1984 ) , Ki-67 serves as the most useful biomarker of cancer in clinical practice ( Brown and Gatter , 2002; Yerushalmi et al . , 2010; Martin et al . , 2004 ) . The finding that Ki-67 is exclusively expressed in the nucleus during all active cell cycle phases ( G1 , S , G2 , and mitosis ) in actively proliferating cells but not in quiescent cells renders it a reliable marker for proliferation fraction assessment ( Gerdes et al . , 1984 ) . In many cancer types including breast ( Luporsi et al . , 2012 ) , lung ( Jakobsen and Sorensen , 2013 ) , brain ( Abry et al . , 2010 ) , and prostate cancer ( Fisher et al . , 2013 ) , the Ki-67 expression level represents a proliferation index , and Ki-67 overexpression predicts poor prognosis . Moreover , Ki-67 has also been used to determine cancer treatment strategies ( Dowsett et al . , 2011 ) . Ki-67 silencing was recently reported to elicit the disappearance of perichromosomal layers in mitosis and the smaller nuclei of daughter cells leading to the problem in mitosis ( Booth et al . , 2014 ) . However , although Ki-67 is known to play an important role in cell proliferation , the mechanism by which Ki-67 regulates cancer progression remains unclear . As a nuclear protein , Ki-67 interacts with other nuclear proteins ( Sueishi et al . , 2000; Takagi et al . , 2001 ) . Increasing research focused on the interaction between Ki-67 and other nuclear proteins has paved the way for the elucidation of the mechanism by which Ki-67 regulates cancer progression . Nucleolar protein interacting with the forkhead-associated ( FHA ) domain of pKi-67 ( NIFK ) is a nucleolar and cytoplasmic protein that interacts with Ki-67 ( Takagi et al . , 2001; Li et al . , 2004; Byeon et al . , 2005 ) . NIFK binds to the FHA domain of Ki-67 by two key regulators of mitosis: cyclin-dependent kinase 1 ( CDK1 ) and glycogen synthase kinase GSK-3 ( Byeon et al . , 2005 ) , and the NIFK sequential phosphorylation at Thr238 and Thr234 is required for the Ki-67 interaction ( Byeon et al . , 2005 ) . During mitosis , both NIFK and Ki-67 are recruited to the chromosome periphery , and the localization of NIFK at the peripheral of mitotic chromosome is disrupted in the absence of Ki-67 ( Booth et al . , 2014; Van Hooser et al . , 2005 ) . This critical role of the NIFK-Ki-67 interaction in regulating mitosis renders NIFK as a promising target of cancer research . However , to date , studies focused on the importance of NIFK in cancer are lacking , and no clear evidence has emerged to elucidate whether and how NIFK regulates cancer progression . The Wnt/β-catenin signaling pathway is critical during tumorigenesis and metastasis in various types of cancer ( Polakis , 2000; Sinnberg et al . , 2010 ) . Stabilization and nuclear import of β-catenin activates downstream transcriptional targets , MMP7 , MYC , TCF4 , CCND1 and CD44 which are related to cell cycle , differentiation and metastasis regulation ( Schwartz et al . , 2003; Dey et al . , 2013; Cho et al . , 2011 ) . β-catenin is constantly synthesized but is normally controlled at restricted low concentration by proteasome-mediated degradation . Degradation of β-catenin is shown to be regulated via sequential phosphorylation by casein kinase 1α ( CK1α ) first , and then by GSK-3 , which facilitates the formation of the destruction complex ( Hernandez et al . , 2012; Li et al . , 2012 ) . CK1 family members including CK1α are constitutively active in cells ( Price MA , 2006 ) . Therefore , CK1α function is determined by its intracellular level . However , the mechanism of CK1α expression regulation in tumors , especially in lung cancer remains obscure . In this study , we aimed to characterize the role of NIFK , an important Ki-67 binding partner , in cancer progression . The significant association between NIFK and Ki-67 expression in approximately 20 cancer types based on samples from over 7000 patients in a public database confirmed the importance of NIFK in cancer . We focused our study on lung cancer due to the strongest prognostic value of NIFK for lung cancer . Surprisingly , our results revealed NIFK significantly promotes cancer migration and invasion in vitro and tumor metastasis in vivo in addition to its ability to regulate cancer proliferation . Furthermore , we demonstrated that NIFK modulates lung cancer metastasis by regulating TCF4/β-catenin signaling via the alternation of Casein kinase 1α ( CK1α ) expression . Our study indicates that NIFK expression promotes cancer metastasis and proliferation leading to poor clinical outcomes; thus , NIFK may represent a prognostic indicator and a promising therapeutic target for lung cancer patients . Due to the well-known characteristics of NIFK as a Ki-67-interacting protein , we first analyzed the expression level of NIFK and Ki-67 based on a public database . Using The UCSC cancer genomics browser web resource , 16 cancer types from the TCGA pan-cancer cohort were analyzed . A significantly positive correlation between MKI67IP ( NIFK ) and MKI67 ( Ki-67 ) was observed in almost all cancer types ( Figure 1A ) . High MKI67IP expression was observed in lung , colorectal , breast , uterine , bladder , head and neck , melanoma , cervical , and ovarian cancer . In these high MKI67IP-expressing cancer types , the most significant positive correlation between MKI67IP and MKI67 expression was detected in lung cancer ( ρ = 0 . 488 , p<0 . 001 ) . Based on the heat map , we also observed that the normal tissue group tended to display low MKI67IP expression . IHC analysis revealed significantly higher expression of NIFK in the samples from our patient cohort than in the paired normal tissue for lung and colorectal cancer but not breast cancer ( Figure 1B ) . To identify the cancer types in which NIFK exerts the most significant impact on cancer progression , we examined the prognostic value of NIFK for various cancer types using the PrognoScan database . High MKI67IP expression was associated with poor survival in several cancer types , including lung , breast , and blood cancer ( Figure 1C ) . By ranking the hazard ratios from the Cox proportional hazards survival model , we determined that high MKI67IP expression corresponded to the highest hazard ratio in lung cancer patients ( hazard ratio = 4 . 71 , Cox p value = 0 . 000308 ) . Based on clinicopathological analysis of lung cancer , the patients displaying high NIFK protein expression exhibited more frequent nodal involvement ( p = 0 . 032 ) and distant metastasis ( p = 0 . 036 ) , as well as a higher pathological stage ( p = 0 . 059 ) ( Figure 1D ) . Similar results were observed in a lung cancer cohort from the TCGA database ( Figure 1—figure supplement 1 ) . According to the above results , NIFK displayed the greatest clinical significance for lung cancer and may be associated with lung cancer progression by regulating tumor metastasis . 10 . 7554/eLife . 11288 . 003Figure 1 . NIFK expression is most concurrently elevated with Ki67 in lung cancer and lung cancer patients displaying high NIFK level exhibit frequent lymph node and distant metastasis . ( A ) In the TCGA pan-cancer cohort , significantly positive correlations between MKI67IP ( NIFK ) and MKI67 ( Ki-67 ) RNA expression were observed in almost all cancer types . Among the cancer types that displayed high MKI67IP expression , lung cancer exhibited the strongest correlation between MKI67IP and MKI67 expression . Red color in heat map represents genes with high expression . Blue color in heat map represents gene with low expression . ( B ) Based on IHC analysis , significant overexpression of NIFK compared with paired normal tissue was observed in lung cancer ( upper ) and colorectal cancer ( lower left ) but not in breast cancer ( lower right ) . ( C ) Forest plot comparison of the hazard ratio of MKI67IP ( NIFK ) overexpression in patients with various cancer types revealed that lung cancer displayed the strongest impact of NIFK RNA expression on survival . ( D ) Lung cancer patients displaying NIFK overexpression exhibited more frequent lymph node and distant metastasis and a higher pathological stage . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 00310 . 7554/eLife . 11288 . 004Figure 1—figure supplement 1 . MKI67IP is overexpressed in tumors and its expression correlates with the pathological TNM stage in lung cancer . The heat map ( A ) illustrates the association of MKI67IP expression with the corresponding ( B ) sample type , ( C ) T stage , ( D ) N stage and ( E ) M stage in the lung cancer cohort . The data for 548 cases were retrieved from the TCGA ( lung adenocarcinoma ( LUAD ) gene expression by RNAseq ( Illumina HiSeq ) ) database and were analyzed using the chi-squared test . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 004 Based on the prognostic significance of NIFK and the strong correlation between NIFK and Ki-67 expression in lung cancer , we further investigated the functional role of NIFK in vitro and in vivo . In 12 lung cancer cell lines , the endogenous NIFK levels were normalized to those of the normal lung cell line Beas2B ( Figure 2A , Top ) , and the relative NIFK levels were statistically analyzed ( Figure 2A , Bottom ) . Then , we stably overexpressed NIFK in A549 and PC13 cells , which exhibit low NIFK expression , via lentiviral infection ( Figure 2B ) . A significant increase in cell migration after NIFK overexpression was observed in the wound-healing assay ( p<0 . 001 for 4 MOI , Figure 2C ) . Increased cell migration and invasion upon NIFK overexpression were confirmed by the transwell assay with short incubation time ( Figure 2—figure supplement 1 ) . In addition , higher endogenous NIFK levels were detected in H661 and H1299 cells , which were derived from metastatic sites in lung cancer patients ( ATCC ) and are considered to be invasive ( Figure 2A ) . Therefore , we established stable NIFK-silenced H661 and H1299 clones using lentivirus-based shRNA-mediated knockdown ( Figure 2D ) . Cell migration was inhibited after NIFK knockdown , especially in clone sh6 ( p<0 . 01 , Figure 2E ) . The in vitro results revealed that NIFK could promote cell migration and invasion . 10 . 7554/eLife . 11288 . 005Figure 2 . NIFK promotes the migration of lung cancer cells in vitro . ( A ) The endogenous expression levels of NIFK in lung adenocarcinoma ( Left ) , squamous and large cell lung cancer cell lines ( Right ) . The relative expression levels were normalized to those of normal Beas2B lung cells , and the average expression levels are presented . ( B ) The relative NIFK levels in A549 and PC13 cells after overexpression of NIFK via lentiviral infection . ( C ) The migratory capacity of NIFK-overexpressing A549 and PC13 cells was assessed using a wound-healing assay . The exposed area was measured after the indicated incubation period and was normalized to that of the 0-h control . ( D ) The NIFK knockdown efficiencies in the lentivirus-based shRNA clones sh5 and sh6 , corresponding to H661 and H1299 cells , respectively . NS , non-silenced control . ( E ) H661 and H1299 cell migration after NIFK knockdown was evaluated at the indicated time points . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 00510 . 7554/eLife . 11288 . 006Figure 2—figure supplement 1 . The overexpression of NIFK promotes cell migration , invasion in vitro and metastasis in vivo . ( A-B ) Cell migration and invasion were evaluated via the transwell assay . A549 cells and PC13 cells were incubated for 4 . 5 hr and 24 hr , respectively , to evaluate cell migration . For the invasion assay , both A549 and PC13 cells were incubated for 42 hr . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 006 Additionally , in vivo animal model experiments were performed to examine whether NIFK promotes tumor metastasis . A549 cells overexpressing NIFK were injected into NSG mice via the tail vein ( Figure 3A ) . Increased lung metastasis was observed in the NIFK-overexpressing group both grossly and microscopically ( p = 0 . 0127 , Figure 3B–C and Figure 3—figure supplement 1 ) . Similarly , significantly reduced numbers of metastatic lung nodules were observed in the NIFK-silenced H661 cell-injected group ( p = 0 . 0054 ) and in the NIFK-silenced H1299 cell-injected group ( p = 0 . 0169 ) compared with the corresponding non-silenced ( NS ) control cell-injected groups ( Figure 3D–E ) . Representative images of HE staining and IHC staining for NIFK are presented in Figure 3F . In our patient cohort , high NIFK IHC expression significantly correlated with poor overall survival in both the Taiwanese ( p = 0 . 018 ) and Korean ( p = 0 . 041 ) lung cancer cohorts ( Figure 3G–H ) . These results including clinical data suggest a regulatory role of NIFK in lung cancer progression , especially in tumor metastasis . 10 . 7554/eLife . 11288 . 007Figure 3 . NIFK facilitates cancer cell metastasis in vivo and is associated with poor survival of lung cancer cohorts . ( A ) The level of NIFK overexpression following infection of A549 cells with 4 MOI virus . ( B ) The indicated cells were injected into NSG mice via the tail vein . Surface lung nodules were statistically quantified . N=5 per group . ( C ) Representative images of surface lung nodules with HE staining and IHC staining for NIFK are presented for the RFP- or NIFK-overexpressing cell-injected groups . ( D ) The NIFK knockdown efficiency in H661 and H1299 cells . ( E ) The cells were injected via the tail vein of the mice . Top , representative images of lung metastasis of the indicated H661 and H1299 cells . Bottom , statistical quantification of lung metastatic nodules in the indicated groups . ( F ) Top , representative images of lung HE staining . Middle , images of HE staining in the indicated areas . Bottom , IHC staining for NIFK in the indicated areas . ( G ) Representative images of IHC staining for NIFK . ( H ) Kaplan-Meier survival analysis revealed that high NIFK IHC expression correlates with poor prognosis in lung cancer . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 00710 . 7554/eLife . 11288 . 008Figure 3—figure supplement 1 . Cells overexpressing NIFK or RFP were injected via the tail vein . After 5 weeks , the mice were sacrificed , and each mouse was assigned a GRCA number . HE and IHC staining for NIFK were performed and are presented . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 008 Due to the previously characterized NIFK as a Ki-67-interacting protein , we therefore studied the regulatory role of NIFK in cell proliferation . NIFK downregulation in H661 and H1299 cells decreased the cancer proliferation at 96 hr in vitro ( Figure 4A ) . Furthermore , H1299 cell with NIFK silencing displayed the reduced tumor weight and volume in vivo p<0 . 01 , Figure 4B–C ) . The NIFK-regulated cell proliferation was Ki-67-dependent that Ki-67 silencing decreased cell growth in both A549 and PC13 cells ( Figure 4D ) . In addition , the effect of NIFK Ki-FHA binding , RNA recognition motif ( RRM ) truncation and T234A/T238A point mutation on cell proliferation was compared with ectopic expression of wild type ( WT ) NIFK ( Figure 4E–F ) . Based on Figure 4F , both RRM and Ki-FHA binding domain were involved in tumor proliferation , but the latter was more significant . In lung cancer cohort , we observed positive correlation of high NIFK/Ki-67 with poor survival at protein level ( Figure 4G , p = 0 . 004 ) and significant expression coefficient ( Figure 4H , p<0 . 001 ) . The results potentially indicated the requirement of Ki-67 in NIFK-increased cell proliferation and the significance of NIFK and Ki-67 in lung cancer progression . 10 . 7554/eLife . 11288 . 009Figure 4 . NIFK promotes cell proliferation via the poor prognosis marker Ki-67 . ( A ) H661 and H1299 cells were seeded at a density 5×104 cells/well . The number of cells was counted at 48 and 96 hr via the Trypan blue exclusion assay . ( B ) Box plot represented tumor weight of H1299 cell NS clone and NIFK-silenced clone at 4 weeks after injection . ( C ) Growth curves of tumor volume in indicated groups . ( D ) Cell numbers of NIFK overexpressing A549 and PC13 cells upon Ki-67 silence . ( E ) Overexpression of GFP-tagged wild type ( WT ) , truncated and point mutated NIFK in PC13 cells . Cells were transiently transfected via liposome and selected . ( F ) Cell numbers of indicated groups were evaluated at 48 hr after cell seeding at density 5×104 cells/well in 6 wells plate . ( G ) Kaplan-Meier survival analysis revealed the correlation of NIFK and Ki-67 IHC expression with poor prognosis in lung cancer . ( H ) The correlation of NIFK and Ki-67 IHC expression in lung cancer . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 009 The importance of NIFK in clinical patients motivated us to explore the molecular mechanism underlying NIFK-mediated lung cancer metastasis . Knocking down Ki-67 expression by shRNA1 in NIFK-overexpressing PC13 cells and H1299 cells did not impact NIFK-induced cell migration or invasion ( Figure 5A–D ) . Therefore , we performed microarray analysis of PC13 cells overexpressing NIFK . The signature of genes displaying a≥1 . 5-fold change in expression upon NIFK overexpression was subjected to Ingenuity Pathway Analysis ( IPA ) and MetaCore database analysis ( Supplementary file 1A ) . Based on the MetaCore enrichment analysis , Wnt signaling , a previously identified pivotal pathway in regulating cancer metastasis was most strongly affected by NIFK overexpression ( Figure 5E–F ) . The levels of CK1α and the adenomatous polyposis coli ( APC ) -AXIN-CK1α-CTNNβ-GSK3β complex were decreased following NIFK overexpression , indicating that CK1α is a key molecule in the NIFK-regulated Wnt signaling pathway ( Figure 5G–H , Figure 5—figure supplement 1 ) . Based on these bioinformatics results , we hypothesized that NIFK overexpression downregulates CK1α , thereby decreasing the levels of the APC-AXIN-CK1α-CTNNβ-GSK3β complex , which in turn increases intracellular β-catenin protein stability . The increased nuclear fraction of β-catenin was observed upon NIFK overexpression ( Figure 5I ) as well as the reduced β-catenin phosphorylation ( Figure 5—figure supplement 2 ) . Furthermore , we determined whether TCF/β-catenin transcriptional activity is affected by the NIFK/CK1α axis . The results of TCF/β-catenin reporter assay revealed the decreased luciferase activity in the NIFK-silenced cells . Additional knockdown of CK1α restored TCF/β-catenin transcriptional activity in both NIFK-silenced H661 and H1299 cells ( p<0 . 01 , Figure 5J ) . In addition , the downstream transcriptional targets of TCF/β-catenin , including TCF4 , CD44 , CCND1 and MMP7 , were regulated by the NIFK-CK1α axis ( Figure 5—figure supplement 3 ) . The upregulated downstream transcriptional targets were further confirmed by another microarray analysis of PC13 cells with lentiviral-based stable NIFK overexpression ( Figure 5—figure supplement 4 and Supplementary file 1B , labeled in red ) . These data indicated that NIFK might regulate TCF/β-catenin-mediated transcriptional events via CK1α . 10 . 7554/eLife . 11288 . 010Figure 5 . Knowledge-based analysis of the microarray data reveals that NIFK regulate CK1α and Wnt signaling . ( A ) The relative knockdown efficiencies of Ki-67 in ( A ) PC13 and ( C ) H1299 cells . Cell migration and invasion were evaluated via the transwell assay using ( B ) PC13 and ( D ) H1299 cells . ( E ) List of the top 10 signaling pathways altered by NIFK overexpression . The microarray data of NIFK overexpression in PC13 cells were analyzed using the MetaCore Maps database . ( F ) List of the top 10 networks affected by NIFK overexpression . The gene signatures were processed via the MetaCore Networks database . ( G ) Representative signature of genes displaying a≥1 . 5-fold change in expression due to NIFK overexpression in PC13 cells . The red and green bars represent upregulation and downregulation , respectively . ( H ) Knowledge-based IPA analysis of the microarray data focusing on CK1α ( CSNK1A1 ) -mediated signaling . The red and green circles represent upregulation and downregulation , respectively . ( I ) The nuclear levels of β-catenin in cells with NIFK overexpression were showed . Nuclear fractions were extracted from A549 and PC13 cells . ( J ) TCF/LEF transcriptional activity alteration in the indicated H661 and H1299 cells . TOPflash: reporter plasmid with TCF binding sites . FOPflash: reporter plasmid with mutated TCF binding sites . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 01010 . 7554/eLife . 11288 . 011Figure 5—figure supplement 1 . NIFK down-regulates CK1α , and up-regulates β-catenin level especially in nucleus . RT-PCR was performed on RNA samples from PC13 cells overexpressing NIFK . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 01110 . 7554/eLife . 11288 . 012Figure 5—figure supplement 2 . Phospho-β-catenin levels are decreased after NIFK overexpression . β-catenin phosphorylation was studied in A549 and PC13 cells . Cells were pre-treated with 20 μM proteasome inhibitor MG132 for 3 hr prior lysate harvest for Western blot . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 01210 . 7554/eLife . 11288 . 013Figure 5—figure supplement 3 . NIFK regulates downstream transcriptional targets of TCF4/LEF via CK1α . RT-PCR was performed on the indicated samples . The expression levels of transcriptional targets of TCF4/LEF are presented . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 01310 . 7554/eLife . 11288 . 014Figure 5—figure supplement 4 . NIFK regulates downstream transcriptional targets of β-catenin . Microarray analysis of PC13 cells stably overexpressing NIFK by lentiviral infection was performed . β-catenin transcriptional targets upregulated by NIFK were examined and identified via IPA analysis . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 014 Knocking down NIFK expression elicited CK1α upregulation , which decreased the downstream inhibitory target β-catenin in the H661 and H1299 cell lines ( Figure 6A , Left ) . In addition , NIFK overexpression using a lentivirus decreased CK1α expression and increased β-catenin level in A549 and PC13 cells ( Figure 6A , Right ) . A similar result was observed following transient overexpression of NIFK ( Figure 6—figure supplement 1 ) . To investigate the potential mechanism underlying the NIFK-mediated regulation of CSNK1A1 ( CK1α ) transcription , we performed a search using the TFSEARCH website to identify putative transcription factors that may bind to the promoter . Several potential transcription factors were identified according to the matched consensus binding sequences of each candidate transcription factor to the CSNK1A1 ( Ensembl:ENSG00000113712 ) promoter region ( Figure 6B ) . RUNX1 and SRY scored 100 with respect to the matched sequences . The promoter region from -969 to -127 bp was found to be important for transcriptional activation , indicating the potential involvement of SRY and RUNX1 ( Figure 6C ) . Thus , we determined the involvement of these transcription factors in regulating the expression level of CK1α . Silencing of RUNX1 and SRY expression in H1299 cells using shRNA decreased the CK1α expression levels ( Figure 6D ) , whereas no change in CK1α expression was detected following knockdown of another putative transcription factor , CdxA ( Figure 6—figure supplement 2 ) . Furthermore , The ChIP assay was performed on NS H1299 cells and the NIFK-silenced H1299 cell clone sh6 to explore whether the binding of the predicted transcription factors to the promoter region is regulated by NIFK . As shown in Figure 6E , increased promoter binding of RUNX1 was observed in the NIFK-silenced H1299 cell clone sh6 compared with the NS control clone ( Top panel ) . In a complementary experiment , the overexpression of NIFK in PC13 cells decreased the promoter binding of RUNX1 ( Figure 6E , Bottom panel ) . However , this regulation was not significantly observed for SRY based on the ChIP experiment ( Figure 6E ) . Thus , we ruled out the involvement of SRY in the NIFK/CK1α axis . To further determine whether RUNX1 affects CK1 promoter activity , we performed the reporter assay . Our results revealed increased CK1 reporter activity after NIFK knockdown in H1299 cells , which was decreased by RUNX1 knockdown ( p<0 . 001 , Left , Figure 6F ) . In addition , the relative luciferase activity was dose-dependently decreased by NIFK overexpression in PC13 cells ( p<0 . 001 , Right , Figure 6F ) . Furthermore , RUNX1 mRNA level was found to be destabilized upon NIFK overexpression ( Figure 6G and Figure 6—figure supplement 3 ) , and the promoter activity was decreased by NIFK ( Figure 6H ) . Inversely , RUNX1 was increased upon NIFK silencing ( Figure 6I ) . In addition , the NIFK-mediated RUNX1 repression might be relevant to its FHA and RRM domain ( Figure 6—figure supplement 4 ) . However , the binding of NIFK with RUNX1 mRNA and with RUNX1 protein were not detected ( RNA binding protein IP and protein IP data not shown ) . Therefore , the mechanism needs further to be investigated . In the lung cancer cohort , the NIFK and RUNX1 or CK1α RNA levels were inversely correlated , whereas the RUNX1 and CK1α RNA expression levels were positively correlated ( p<0 . 001 , Figure 6J ) . These results suggested that RUNX1 is a potential transcriptional factor of CSNK1A1 ( CK1α ) that is negatively regulated by NIFK to decrease CK1α expression . 10 . 7554/eLife . 11288 . 015Figure 6 . NIFK regulates CK1α expression via the destabilization of transcription factor RUNX1 . ( A ) Left , the levels of CK1α and β-catenin in H661 and H1299 cells after NIFK knockdown . Right , the levels of the indicated molecules in A549 and PC13 cells upon NIFK overexpression . ( B ) Identification of cis-regulatory elements within the CSNK1A1 ( Ensembl:ENSG00000113712 ) promoter region . The locations of the consensus binding sites relative to the transcription start site ( TSS ) are presented below the indicated transcription factors . The scores were calculated by the TFSEARCH website according to the matched sequences . BRE , B recognition element . Inr , initiator element . ( C ) The CK1 promoter region from -969 to -127 bp is important for transcriptional activation . The relative luciferase activity was measured 48 hr post-transfection with a reporter plasmid containing the CK1 promoter or the indicated deletion mutant . ( D ) The expression of the indicated molecules after the knockdown of the transcription factors SRY and RUNX1 in the NIFK-silenced H1299 cell clone sh6 . ( E ) ChIP was performed on H1299 ( Top ) and PC13 cells ( Bottom ) using antibodies against RUNX1 and SRY . M , DNA marker . NC , negative beads control . ( F ) A CK1 promoter reporter assay was performed on NIFK-silenced H1299 cells ( Left ) and in NIFK-overexpressing PC13 cells ( Right ) . The cells were lysed 48 hr after reporter plasmid transfection . The relative luciferase activity was normalized to the number of cells and was quantified . ( G ) A549 cells were transiently transfected with RFP control or NIFK by lipofection . After 48 h , cells were treated with actinomycin D ( 5 μg/ml ) for indicated time points . Relative RUNX1 mRNA levels were analyzed by RT-PCR . ( H ) RUNX1 promoter activity was measured after 48 hr of transfection by liposome . ( I ) NIFK , CK1α and RUNX1 protein levels were analyzed in H1299 cells upon NIFK silencing . ( J ) Correlation between the expression levels of the indicated molecules in the lung cancer cohort . The microarray data were retrieved from the TCGA database ( genomic_TCGA_LUAD_exp_HiSeqV2_percentile_clinical ) . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 01510 . 7554/eLife . 11288 . 016Figure 6—figure supplement 1 . A549 and PC13 cells were transfected with Flag-tagged NIFK or a vector control for 48 hr . The expression levels of the indicated molecules are presented . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 01610 . 7554/eLife . 11288 . 017Figure 6—figure supplement 2 . RUNX1 and SRY are potential transcription factors of CK1α . The expression of CK1α after the knockdown of the candidate transcription factors CdxA , SRY , and RUNX1 in NIFK-silenced H1299 cell clone sh6 . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 01710 . 7554/eLife . 11288 . 018Figure 6—figure supplement 3 . NIFK overexpression leads to RUNX1 mRNA instability . A549 cells were transfected with RFP control or NIFK by lipofection . Cells were treated with actinomycin D ( 5 μg/ml ) for indicated time points . Relative RUNX1 mRNA levels were analyzed by Q-PCR . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 01810 . 7554/eLife . 11288 . 019Figure 6—figure supplement 4 . NIFK decreases RUNX1 at RNA level . Overexpression of GFP-tagged wild type ( WT ) , FHA and RRM truncated and T234A/T238A point mutated NIFK in PC13 cells were performed via liposome and selected . RUNX1 mRNA levels were analysed by RT-PCR . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 019 We further investigated whether NIFK-induced tumor metastasis is mediated by the CK1α . CK1α expression was reduced in H661 and H1299 cells via lentivirus-based shRNA-mediated knockdown in conjunction with NIFK silencing using sh6 to determine the functional role of CK1α ( CK1α sh1-3 , Figure 7A ) . Knocking down CK1α expression restored cell migration compared with the NIFK knockdown-only group , although the effects on clones sh1 and sh2 were clearer than those on clone sh3 ( Figure 7B , Figure 7—figure supplement 1 ) . These results suggested that the effect of NIFK on cell migration may be regulated by CK1α . Next , we performed animal experiments to examine whether CK1α is involved in NIFK-regulated tumor metastasis . The formation of surface lung nodules was significantly restored due to CK1α knockdown in the H661 and H1299 sh6 cells ( Figure 7C–D , upper left ) . This restoration of tumor metastasis was further confirmed via HE staining of the lung tissue ( Lower left , Figure 7C–D ) . The quantifications for the number of metastatic lung nodules in each group are presented ( Right , Figure 7C–D ) . These results suggested that NIFK regulates lung cancer metastasis via CK1α . In support of this hypothesis , we observed that high levels of CK1α correlated with a favorable prognosis in lung cancer patients ( p = 0 . 051 ) ( Table 1 and Figure 7E ) . Furthermore , the combination of NIFK ( MKI67IP ) and CK1α ( CSNK1A1 ) may represent a superior prognostic indicator for lung cancer ( p = 0 . 01 ) ( Figure 7E ) . Figure 7F represents the model that NIFK promotes lung cancer progression by ( 1 ) downregulating CK1α expression through the destabilization of its novel transcription factor , RUNX1 , to promote lung cancer metastasis as well as ( 2 ) increasing the Ki-67-dependent cell proliferation . 10 . 7554/eLife . 11288 . 020Figure 7 . NIFK promotes lung cancer metastasis via CK1α and lung cancer patients with high NIFK/low CK1α represent poor survival rate . ( A ) The CK1α knockdown efficiencies of 3 lentivirus-based shRNAs in H661 and H1299 cells . ( B ) Wound-healing assays were performed on the cells in the indicated groups . The quantification of the migration of H661 and H1299 cells is presented . ( C&D ) Animal studies were performed on the indicated groups . Left , representative images of lung metastasis and lung HE staining of NSG mice injected with ( C ) H661 cells or ( D ) H1299 cells . Right , statistical quantification of the number of metastatic nodules in each group is presented . ( E ) Kaplan-Meier plot demonstrating the disease-free survival of 226 lung cancer patients displaying varying ( upper left ) NIFK ( MKI67IP ) and ( upper right ) CK1α ( CSNK1A1 ) expression levels . ( Lower ) Kaplan-Meier plot demonstrating the disease-free survival of cases separated into high NIFK/low CK1α and low NIFK/high CK1α groups . The data were retrieved from the microarray analysis of the GSE31210 dataset . ( F ) The model of NIFK-induced activation of TCF/β-catenin transcriptional activity via the Runx-1-dependent downregulation of CK1α in metastasis . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 02010 . 7554/eLife . 11288 . 021Figure 7—figure supplement 1 . NIFK promotes cancer cell migration via CK1α . Wound-healing assays were performed on H661 and H1299 cells . Representative images of cell migration in the indicated groups are presented . DOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 02110 . 7554/eLife . 11288 . 022Table 1 . Cox univariate and multivariate regression analysis of CSNK1A1 ( CK1α ) expression for relapse-free survival in GSE31210 lung cancer datasetDOI: http://dx . doi . org/10 . 7554/eLife . 11288 . 022UnivariateMultivariateSurvivalVariableHR ( 95% CI ) PHR ( 95% CI ) PRelapse free ( n=246 ) CSNK1A1 ( high/low ) 2 . 261 ( 0 . 975-5 . 245 ) 0 . 0571 . 804 ( 0 . 769-4 . 231 ) 0 . 175Age1 . 658 ( 0 . 984-2 . 796 ) 0 . 0581 . 638 ( 0 . 97-2 . 766 ) 0 . 065Sex ( female/male ) 1 . 271 ( 0 . 778-2 . 075 ) 0 . 3380 . 986 ( 0 . 489-1 . 986 ) 0 . 968Smoking habit ( never/ever ) 1 . 333 ( 0 . 815-2 . 178 ) 0 . 2521 . 19 ( 0 . 589-2 . 402 ) 0 . 628Stage ( I/II ) 3 . 163 ( 1 . 92-5 . 21 ) 0 . 0012 . 912 ( 1 . 745-4 . 862 ) 0 . 001 In this study , we demonstrate that NIFK promotes cancer progression by regulating cancer metastasis and proliferation . Lung cancer patients displaying high NIFK expression exhibited poor prognosis and frequent lymph node and distant metastasis . Although Ki-67 has been a research emphasis of cancer , Ki-67-interacting proteins have not been given the attention that they merit and are generally considered to regulate cancer proliferation via their physiological roles in mitosis and the cell cycle ( Booth et al . , 2014; Byeon et al . , 2005; Florian and Mayer , 2011 ) . Hklp2/KIF15 , another Ki-67 interacting protein , plays a critical role in the maintenance of spindle bipolarity during cell division and was reported to serve as a breast cancer tumor antigen ( Florian and Mayer , 2011; Rath and Kozielski , 2012; Scanlan et al . , 2001 ) . NIFK was previously implicated as a c-Myc-responsive gene in breast cancer ( Musgrove et al . , 2008 ) . The significant cancer metastasis-promoting characteristics of NIFK demonstrated by our results provide novel insight into the role of NIFK in cancer . Furthermore , aside from its role as a proliferation marker , Ki-67 has been demonstrated to be associated with metastasis and lymphovascular invasion in cancer patients ( Pollack et al . , 2004; Inwald et al . , 2013 ) . However , our results imply that Ki-67 may not affect cancer metastasis via its interaction with NIFK . We showed the NIFK-induced cell proliferation is dependent on Ki-FHA binding motif indicating the requirement of NIFK-Ki-67 interaction in lung cancer proliferation . Ki-67 was reported as a key factor organizing chromosomal periphery during cell mitosis , and silence of Ki-67 might elicit cell death ( Booth et al . , 2014 ) . NIFK interacts with Ki-67 in mitotic phase , and whether the interaction is critical in nuclear organization remains to be explored . Our previous study characterized the role of NIFK in U2OS cell proliferation as well as the requirement of its RRM for rRNA maturation ( Pan et al . , 2015 ) . However , Ki-67 interacting motif of NIFK also plays a role in regulating cell proliferation especially after long-term incubation as compared with the immediate response of RRM deletion in U2OS cell ( Pan et al . , 2015 ) . Both results evidently point on NIFK as a gatekeeper in the maintenance of well-regulated cell cycle . The discrepancy of RRM efficacy in proliferation among two studies might potentially due to the difference in cell type-specific mechanism . CK1α , a component of the β-catenin destruction complex , is characterized as a negative regulator that blocks Wnt signaling pathway-mediated metastasis ( Clevers , 2006 ) . In the canonical Wnt signaling pathway , the level of intracellular β-catenin is modulated by proteasomal degradation mediated by destruction complex , which is composed of APC , Axin , CK1α , and GSK3β ( MacDonald et al . , 2009 ) . Given that β-catenin and APC mutation is less commonly observed in lung cancer than in other cancer types such as colon cancer , understanding the mechanism by which these Wnt signaling inhibitors regulate the β-catenin destruction complex is crucial ( Stewart , 2014 ) . Many studies have demonstrated that the downregulation of the members of the β-catenin destruction complex , including CK1α , is commonly observed in lung cancer cell lines and cancer tissue and correlates with poor prognosis or poor clinicopathological characteristics ( Yang et al . , 2013; Srivastava et al . , 2012; Lee et al . , 2013 ) . In our study , we found that NIFK acts as a critical regulator that prevents CK1α-mediated β-catenin degradation , which in turn leads to cancer metastasis . Therefore , these results highlight NIFK as a novel regulator of the Wnt/β-catenin signaling pathway . In our study , we also observed the sole downregulation of CK1α upon NIFK overexpression , whereas the expression levels of other components of the destruction complex were not significantly altered . Recently , a negative feedback loop of Wnt signaling activation was reported via the Huwe-1-dependent ubiquitylation of Dishevelled ( Dvl ) ( de Groot et al . , 2014 ) . Upon Wnt signaling activation , Dvl recruits Axin and GSK3-β away from the destruction complex , thereby inhibiting β-catenin destruction complex formation and , subsequently , β-catenin degradation ( de Groot et al . , 2014; Stambolic et al . , 1996; Behrens et al . , 1998 ) . Correspondingly , we detected significant Huwe-1 upregulation based on the microarray data of NIFK overexpression , suggesting the activation of an alternative negative feedback loop via the decrease in Dvl activity ( Supplementary file 1A ) . Although this negative feedback might serve to regulate lung cancer progression via GSK-3β , CK1α , as the downstream rate-determining enzyme that sequentially phosphorylates β-catenin prior to GSK3β ( Hernandez et al . , 2012 ) , remains to play a significant role in NIFK regulated β-catenin degradation . An additional interesting finding in our data is that higher levels of NIFK were observed in p53 loss-of-function mutation H661 cell lines and in p53-null H1299 cell lines . The repression of CK1α expression significantly restored the tumor metastatic ability inhibited by NIFK downregulation in these p53-deficient cell lines . These results implicated the critical role of the NIFK-CK1α-β-catenin pathway in p53-deficient lung cancer . Previous studies have demonstrated that CK1α plays a role as a tumor suppressor in p53-inactivated cancer cells ( Elyada et al . , 2011; Huart et al . , 2009; Chen et al . , 2005 ) . The loss of heterozygosity of CK1α results in highly invasive carcinoma in a p53-deficient mouse model ( Elyada et al . , 2011 ) . Because p53 inactivation is a major pathogenic event in lung cancer ( Herbst et al . , 2008 ) , further research is necessary to determine the role of NIFK-mediated CK1α expression in p53-deficient lung cancer . Our study revealed that the molecular mechanism underlying NIFK-mediated CK1α downregulation in lung cancer is RUNX1-dependent at the transcriptional level . RUNX1 was originally recognized to display tumor-suppressive ability due to its role in acute myeloid leukemia tumorigenesis ( Miyoshi et al . , 1991; Silva et al . , 2003 ) . The loss of RUNX1 from intestinal epithelial cells significantly induced tumorigenesis in a conditional knockout mouse model ( Fijneman et al . , 2012 ) , and the knockdown of RUNX1 in breast cancer cells resulted in hyperproliferation and abnormal morphogenesis ( Wang et al . , 2011; Janes , 2011 ) . Cancer metastasis due to the inhibitory effect of NIFK on the binding of RUNX1 to the CK1α promoter region , as demonstrated by our study , represents a novel tumor progression mechanism that merits further investigation . In conclusion , NIFK , a Ki-67-interacting protein , is first identified with clinical significance in lung cancer progression . NIFK enhances the metastatic ability of lung cancer cells via the Runx-1-dependent repression of CK1α expression and activates TCF/β–catenin signaling , thereby promoting metastasis in lung cancer . In addition , cell proliferation is positively regulated by NIFK via Ki-67 . High NIFK expression correlates with poor prognosis and tumor metastasis in clinical lung cancer patients , suggesting that NIFK is an independent prognostic indicator and a promising therapeutic target . Tissue samples of non-small cell lung cancer , breast cancer , and colorectal cancer from patients were included to further analyze the clinicopathological role of NIFK . For non-small cell lung cancer , a total of 188 patients from Kaohsiung Medical University Hospital and National Taiwan University Hospital of Taiwan were included . Another tissue microarray using a Korean cohort of non-small cell lung cancer patients was purchased from SuperBioChips ( SuperBioChips Laboratories , Seoul , Korea ) . For breast cancer , we examined samples from 84 patients from Kaohsiung Veterans General Hospital . For colorectal cancer , 62 patients were enrolled from Taipei Medical University Hospital . All cases were staged according to the cancer staging manual of the American Joint Committee on Cancer , and the histological cancer type was classified according to the World Health Organization classification . The tissues used were obtained with approval from the IRBs of Kaohsiung Medical University Hospital ( KMUH-IRB-20110286 ) , National Taiwan University Hospital , Kaohsiung Veterans General Hospital ( VGHKS12-CT9-057 ) , and Taipei Medical University Hospital ( IRB-99049 ) . No informed consent was required because the data were analyzed anonymously PC9 , PC13 and PC14 cells were developed at the National Cancer Center Hospital in Tokyo , Japan ( Lee et al . , 1985 ) . The other nine human lung cancer cell lines were obtained from American Type Culture Collection ( Manassas , VA , USA ) . Cell lines were purchased by Prof . Michael Hsiao at 2012 and 2013 with certificates of analysis including cell authentication by STR analysis and mycoplasma contamination test . The H1355 , PC9 , H358 , H928 , H520 , H157 , H661 , and H460 cells were maintained in RPMI 1640 medium . The PC13 , PC14 , A549 and H1299 cells were maintained in DMEM . Each medium was supplemented with 10% fetal bovine serum , penicillin ( 100 units/ml ) , and streptomycin ( 100 μg/ml ) . The cells were incubated in a humidified atmosphere consisting of 95% air and 5% CO2 at 37°C . The wound healing assay was assessed using culture inserts ( Ibidi , Martinsried , Germany ) . The culture inserts were transferred to plates . The cells were seeded at a density of 2 × 105 cells/well and were allowed to attach . After incubation , the culture inserts were removed using sterile tweezers and washed with PBS . The plates were filled with culture medium supplemented with 2% serum to induce cell migration . The cells were photographed for quantification of closure of the exposed area . The denuded area closure was calculated by ( Denuded distance 0 h − Denuded distance Endpoint ) / Denuded distance 0 h . All animal experiments were conducted in accordance with a protocol approved by the Academia Sinica Institutional Animal Care and Utilization Committee ( IACUC , Protocol# 14-03-665 ) . Age-matched male NSG mice ( 6 to 8 weeks of age ) were used . To evaluate metastasis , 1 × 106 cells were resuspended in 0 . 1 ml of PBS and injected into the lateral tail vein ( n=6 ) . Metastatic lung nodules were counted and were further confirmed via HE staining using a dissecting microscope . The lentiviral shRNA constructs were purchased from Thermo Scientific ( Pittsburgh , PA , USA ) . Lentiviruses were produced via co-transfection of 293T cells with an shRNA-expressing plasmid , an envelope plasmid ( pMD . G ) and a packaging plasmid ( pCMV-dR8 . 91 ) using calcium phosphate ( Invitrogen , Carlsbad , CA , USA ) . The 293T cells were incubated for 18 hr , followed by replacement of the culture medium . The viral supernatants were harvested and titered at 48 and 72 hr post-transfection . The cell monolayers were infected with the indicated lentivirus in the presence of polybrene and were further selected using puromycin . The cells were co-transfected with the TOP-Flash ( for TCF binding sites ) or FOP-Flash ( for mutated TCF binding sites ) reporter plasmid ( Millipore , Bedford , MA , USA ) and pZsGreen ( GFP ) for 48 hr . The cells were harvested and lysed using a Promega luciferase assay kit ( Madison , WI , USA ) according to the manufacturer’s instructions . The luminescence was measured using a luminometer . The TOP/FOP luminescence ratio was used as a measure of TCF/β-catenin transcriptional activity . NIFK was cloned from Beas2B cDNA using TAKARA DNA polymerase ( Mountain View , CA , USA ) according to the manufacture’s instruction . The primer sequences designed were as follows: 5’- ACCCAAGCTGGCTAGCATGGCGACTTTTTCTGGCCCG-3’ ( sense ) and 5’- TCAAGATCTAGAATTCTCACTGATTGCTGCTTCTTCG-3’ ( antisense ) . The PCR products were gel-purified , digested with NheI/EcoRI , and subcloned into lentiviral expression vector pLAS3W ( RNAi Core , Academia Sinica , Taipei , Taiwan ) . The sequences were confirmed via DNA sequencing by Sequencing Core Facility , SIC , Academia Sinica . The cells were lysed at 4°C in RIPA buffer containing 50 mM Tris-HCl ( pH 7 . 4 ) , 150 mM NaCl , 1% Triton X-100 , 0 . 25% sodium deoxycholate , 5 mM EDTA ( pH 8 . 0 ) , and 1 mM EGTA supplemented with protease and phosphatase inhibitors . After 20 min of lysis on ice , the cell debris was removed via microcentrifugation , followed by rapid freezing of the supernatants . The protein concentration was determined using the Bradford method . In our experiments , equivalent loads of 25–50 g of protein were electrophoresed using a SDS-polyacrylamide gel and then electrophoretically transferred from the gel to a PVDF membrane ( Millipore , Bedford , MA , USA ) . After blocking with 5% non-fat milk , the membrane was incubated in specific primary antibodies ( NIFK: Abcam , ab13880 , 1:1000; Ki-67: Dako , Code M7240 , 1:500; CK1α: Abcam , ab88079 , 1:1000; non-phospho-β-catenin: Cell Signaling , #8814 , 1:1000; phospho-β-catenin ( Ser45 ) : Cell Signaling , #9564 , 1:1000; RUNX1: Cell Signaling , #4336 , 1:1000 ) overnight at 4°C and subsequently incubated in a corresponding horseradish peroxidase-conjugated secondary antibody for 1 hr . The membranes were visualized using the ECL-Plus detection kit ( PerkinElmer Life Sciences , Boston , MA , USA ) . The in vitro migration and invasion were assessed using Transwell assay ( Millipore , Bedford , MA , USA ) . For invasion assay , transwell was additional pre-coated with 35 l of 3X diluted matrix matrigel ( Bd Biosciences Pharmingen , San Diego , CA , USA ) for 30 min . Cells of 2 × 105 in serum-free culture medium were added to the upper chamber of the device , and the lower chamber was filled with 10% FBS culture medium . After indicated hours of incubation , upper surface of the filter was carefully removed with a cotton swab . The filter was then fixed , stained and photographed . Cells of migration and invasion were quantified by counting the cells in three random fields per filter . The 1 . 3K bp CSNK1A1 promoter fragment ( Ensembl:ENSG00000113712 ) was cloned from genomic DNA of H1299 cells . The primer sequences designed were as follows: 5’- CCCCGGTACCCTGACTTAAGATGATAGCAT-3’ ( sense ) and 5’-TTTGCTAGCG CTGGGCCACT TGTTTCTCG-3’ ( antisense ) . The 1 . 1K bp RUNX1 promoter was cloned from 293T cells using RUNX1_Promter1_F: 5’-AAAAGGTACCAAGCCAGTGGGGCCGGAAAA-3’ and RUNX1_Promoter1_R: 5’-TGGGGCTAGCGGTTGTTTATGAGGCCCAAA-3’ . PCR was performed using TAKARA DNA polymerase ( Mountain View , CA , USA ) according to the manufacture’s instruction . The PCR products were gel-purified , digested with KpnI/NheI , and subcloned into pGL4 . 20 firefly luciferase vector ( Promega , Madison , WI , USA ) . The sequences of cloned promoter region were confirmed by DNA sequencing . ChIP assay was performed according to the manufacture’s instruction ( Abcam , Cambridge , MA , USA ) . Briefly , Cells were fixed with 1% formaldehyde and quenched by 1 . 25 M glycine . Cells were then lysed and immunoprecipitated overnight with antibodies against RUNX1 ( Cell Signaling Technology , Danvers , MA , USA ) or SRY ( Santa Cruz , CA , USA ) . After immunoprecipitation , protein A beads were added for 1 hr to capture the immune complexes . The beads were washed , and the cross-link was further reversed . DNA was purified and analyzed by RT-PCR analysis . The primer sequences were as follows: RUNX1: 5’- ACTGAGGTTTTCAACAAGACCA -3’ ( sense ) and 5’-ATCCCCCTGCCATCCTATGT-3’ ( antisense ) with product size of 223 bp . SRY: 5’-ACCATGGAGTTTTCTTTCGTGA-3’ ( sense ) and 5’-TGGTCTTGTTGAAAACCTCAGT-3’ ( antisense ) with product size of 214 bp . Total cellular RNA was extracted by TRIzol reagent ( Invitrogen , Carlsbad , CA , USA ) in accordance with the manufacturer's instructions . One microgram of total RNA was reverse-transcribed using Advantage RT for PCR Kit ( Clontech , Mountain View , CA , USA ) at 42°C for 1 hr as described in the manufacturer's protocol . PCR conditions for rat leptin were 94°C for 5 min and 37 cycles at 94°C for 30 s , 56 C for 30 s and 72°C for 60 s , followed by a final extension step at 72°C for 5 min by Bio-Rad icycle ( Bio-Rad ) . Primer sequences were as follows: human CCND1: 5’-GACCTTCGTTGCCCTCTGT-3’ ( sense ) and 5’-TGAGGCGGTAGTAGGACAGG-3’ ( antisense ) with product size of 180 bp; human MMP7: 5’-TGGGAACAGGCTCAGGACTAT-3’ ( sense ) and 5’-CGTCCAGCGTTCATCCTCAT-3’ ( antisense ) with product size of 504 bp; human TCF4: 5’-CCGATGACGAGGGTGATGAG-3’ ( sense ) and 5’-CCGAGGACACCTTCTCTTCC-3’ ( antisense ) with product size of 399 bp; human CD44: 5’-GGATCCACCCCAACTCCATC-3’ ( sense ) and 5’-AGGTCCTGCTTTCCTTCGTG-3’ ( antisense ) with product size of 702 bp; human MKI67IP ( NIFK ) : 5’-AGGTGGCGCAGGTTCGCAAG-3’ ( sense ) and 5’-TGGTGTGGGGCCCTGGCTATC-3’ ( antisense ) with product size of 632 bp; human GAPDH: 5’-GTCCACTGGCGTCTTCACCACC-3’ ( sense ) and 5’-AGGCATTGCTGATGATCTTGAGGC-3’ ( antisense ) with product size of 161 bp . For each combination of primers , the kinetics of PCR amplification was studied . The number of cycles corresponding to plateau was determined and PCR was performed at exponential range . PCR products were then electrophoresed through a 1% agarose gel and visualized by ethidium bromide staining in UV irradiation . The mRNA levels were also determined by real-time PCR with ABI StepOnePlus real-time PCR system according to the manufacturer’s instructions . GAPDH was used as endogenous control . PCR reaction mixture contained the SYBR PCR master mix , 50 ng cDNA , and primers . Relative gene expression level that the amount of target were normalized to endogenous control gene was calculated using the comparative Ct method formula E-△△Ct . The primer sequences were as follows: RUNX1_F:5’-AGCCCCAACTTCCTCTGCTC-3’ RUNX1_R:5’-TCATCATTGCCAGCCATCAC-3’ . Estimates of the survival rates were calculated using the Kaplan-Meier method and were compared using the log-rank test . The association between clinicopathological categorical variables and NIFK expression was analyzed using the chi-squared test . Student’s t-test was used for other statistical analyses . All data are presented as the mean ± S . D . The p values at the following levels were considered to be significant: p<0 . 05 , p<0 . 01 , and **p<0 . 001 . All data was represented after at least three repeated experiments with a similar pattern .	Cancer cells can rapidly divide to form a tumor . Small groups of cells can leave the tumor to migrate to other sites in the body , and it is these “secondary” tumors that are often responsible for the death of cancer patients . Many proteins influence how and when cells divide and migrate . One such protein called Ki67 is only produced when cells are dividing and it is often used in the clinic as a marker to indicate whether cells have become cancerous . However , it is not clear how Ki67 regulates the progression of cancer . Ki67 interacts with another protein called NIFK , and Lin , Su et al . have now investigated the role of NIFK in cancer . First , publicly available data on the levels of proteins in tumor samples from cancer patients were analyzed . This revealed that , in several different types of cancer , tumors that produced more NIFK were more likely to spread to other parts of the body than tumors that produced smaller amounts of NIFK . Next , Lin , Su et al carried out experiments using human lung cancer cells . This revealed that cells that produced larger amounts of NIFK were more likely to migrate , while cells with lower levels of NIFK divided and migrated less often . Further experiments showed that NIFK increases the activity of genes that are involved in cell migration . NIFK achieves this by reducing the production of a protein that inhibits the activity of another protein called β-catenin . Lin , Su et al . ’s findings reveal a new role for NIFK in promoting the development of cancer . A future challenge is to find out whether chemicals that inhibit NIFK could be used in the treatment of lung cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2016	The nucleolar protein NIFK promotes cancer progression via CK1α/β-catenin in metastasis and Ki-67-dependent cell proliferation
Protein interaction is critical molecular regulatory activity underlining cellular functions and precise cell fate choices . Using TWIST1 BioID-proximity-labeling and network propagation analyses , we discovered and characterized a TWIST-chromatin regulatory module ( TWIST1-CRM ) in the neural crest cells ( NCC ) . Combinatorial perturbation of core members of TWIST1-CRM: TWIST1 , CHD7 , CHD8 , and WHSC1 in cell models and mouse embryos revealed that loss of the function of the regulatory module resulted in abnormal differentiation of NCCs and compromised craniofacial tissue patterning . Following NCC delamination , low level of TWIST1-CRM activity is instrumental to stabilize the early NCC signatures and migratory potential by repressing the neural stem cell programs . High level of TWIST1 module activity at later phases commits the cells to the ectomesenchyme . Our study further revealed the functional interdependency of TWIST1 and potential neurocristopathy factors in NCC development . The cranial neural crest cell ( NCC ) lineage originates from the neuroepithelium ( Vokes et al . , 2007; Groves and LaBonne , 2014; Mandalos and Remboutsika , 2017 ) and contributes to the craniofacial tissues in vertebrates ( Sauka-Spengler and Bronner-Fraser , 2008 ) including parts of the craniofacial skeleton , connective tissues , melanocytes , neurons , and glia ( Kang and Svoboda , 2005; Blentic et al . , 2008; Ishii et al . , 2012; Theveneau and Mayor , 2012 ) . The development of these tissues is affected in neurocristopathies , which can be traced to mutations in genetic determinants of NCC specification and differentiation ( Etchevers et al . , 2019 ) . As an example , mutations in transcription factor TWIST1 in human are associated with craniosynostosis ( El Ghouzzi et al . , 2000 ) and cerebral vasculature defects ( Tischfield et al . , 2017 ) . Phenotypic analyses of Twist1 conditional knockout mouse revealed that TWIST1 is required in the NCCs for the formation of the facial skeleton , the anterior skull vault , and the patterning of the cranial nerves ( Soo et al . , 2002; Ota et al . , 2004; Bildsoe et al . , 2009; Bildsoe et al . , 2016 ) . To comprehend the mechanistic complexity of NCC development and its implication in a range of diseases , it is essential to collate the compendium of genetic determinants of the NCC lineage and characterize how they act in concert in time and space . During neuroectoderm development , transcriptional programs are initiated successively in response to morphogen induction to specify neural stem cell ( NSC ) subdomains along the dorsal-ventral axis in the neuroepithelium ( Briscoe et al . , 2000; Vokes et al . , 2007; Kutejova et al . , 2016 ) . NCCs also arise from the neuroepithelium , at the border with the surface ectoderm through the pre-epithelial-mesenchymal transition ( pre-EMT ) which is marked by the activation of Tfap2a , Id1 , Id2 , Zic1 , Msx1 and Msx2 ( Baker et al . , 1997; Mayor et al . , 1997; Saint-Jeannet et al . , 1997; Marchant et al . , 1998; Etchevers et al . , 2019 ) . In the migratory NCCs , gene activity associated with pre-EMT and NCC specification is replaced by that of EMT and NCC identity ( Marchant et al . , 1998 ) . NCC differentiation progresses in a series of cell fate decisions ( Lasrado et al . , 2017; Soldatov et al . , 2019 ) . Genetic activities for mutually exclusive cell fates are co-activated in the progenitor population , which is followed by an enhancement of the transcriptional activities that predilect one lineage over the others ( Lasrado et al . , 2017; Soldatov et al . , 2019 ) . However , more in-depth knowledge of the specific factors triggering this sequence of events and cell fate bias is presently lacking . Twist1 expression is initiated during NCC delamination and its activity is sustained in migratory NCCs to promote ectomesenchymal fate ( Soldatov et al . , 2019 ) . TWIST1 mediates cell fate choices through functional interactions with other basic-helix-loop-helix ( bHLH ) factors ( Spicer et al . , 1996; Firulli et al . , 2005; Connerney et al . , 2006 ) in addition to transcription factors SOX9 , SOX10 , and RUNX2 ( Spicer et al . , 1996; Hamamori et al . , 1997; Bialek et al . , 2004; Laursen et al . , 2007; Gu et al . , 2012; Vincentz et al . , 2013 ) . TWIST1 therefore constitutes a unique assembly point to identify the molecular modules necessary for cranial NCC development and determine how they orchestrate the sequence of events in this process . To decipher the molecular context of TWIST1 activity and identify functional modules , we generated the first TWIST1 protein interactome in the NCCs . Leveraging the proximity-dependent biotin identification ( BioID ) methodology , we captured TWIST1 interactions in the native cellular environment including previously intractable transient and low-frequency events which feature interactions between transcription regulators ( Roux et al . , 2012; Kim and Roux , 2016 ) . Integrating prior knowledge of protein associations and applying network propagation analysis ( Cowen et al . , 2017 ) , we uncovered modules of highly connected interactors as potent NCC regulators . Among the top-ranked candidates were histone modifiers and chromatin remodelers that constitute the functional chromatin regulatory module ( TWIST1-CRM ) in NCC . Genome occupancy , gene expression , and combinatorial perturbation studies of high-ranked members of the TWIST1-CRM during neurogenic differentiation in vitro and in embryos revealed their necessity in stabilizing the identity of early migratory NCC and subsequent acquisition of ectomesenchyme potential . This study also highlighted the concurrent activation and cross-repression of the molecular machinery that governs the choice of cell fates between neural crest and neurogenic cell lineages . The protein interactome of TWIST1 was characterized using the BioID technique which allows for the identification of interactors in their native cellular environment ( Figure 1A ) . We performed the experiment in cranial NCC cell line O9-1 ( Ishii et al . , 2012 ) transfected with TWIST1-BirA* ( TWIST1 fused to the BirA* biotin ligase ) . In the transfected cells , biotinylated proteins were predominantly localized in the nucleus ( Figure 1—figure supplement 1A , B; Singh and Gramolini , 2009 ) . The profile of TWIST1-BirA* biotinylated proteins were different from that of biotinylated proteins captured by GFP-BirA* ( Figure 1—figure supplement 2A ) . Western blot analysis detected TCF4 , a known dimerization partner of TWIST1 , among the TWIST1-BirA* biotinylated proteins but not in the control group ( Figure 1—figure supplement 2A ) . These findings demonstrated the utility and specificity of the BioID technology to identify TWIST1-interacting proteins . We characterized all the proteins biotinylated by TWIST1-BirA* and GFP-BirA* followed by streptavidin purification using liquid chromatography combined with tandem mass spectrometry ( LC-MS/MS ) ( Supplementary file 1 ) . Differential binding analysis of TWIST1 using sum-normalized peptide-spectrum match ( PSM ) values ( Figure 1—figure supplement 2B , C; see Materials and methods ) revealed 140 putative TWIST1 interactors in NCCs ( p<0 . 05; Fold-change >3; PSM#>2; Figure 1B , Supplementary file 1 ) . These candidates included 4 of 56 known TWIST1 interactors , including TCF3 , TCF4 , TCF12 , and GLI3 ( overlap odds ratio = 18 . 05 , Chi-squared test p-value=0 . 0005; Agile Protein Interactomes DataServer [APID] ) ( Alonso-López et al . , 2019; Fan et al . , 2020 ) . Despite that the APID database covers a broad spectrum of protein interaction in different cell line models , it was noted that the TWIST1 partners , TCF3 , 4 , and 12 that were recurrently identified in yeast-two-hybrid , immunoprecipitation and in vitro interaction assays were recovered by BioID ( El Ghouzzi et al . , 2000; Firulli et al . , 2007; Fu et al . , 2011; Teachenor et al . , 2012; Sharma et al . , 2013; Kotlyar et al . , 2015; Li et al . , 2015 ) . This finding has prompted us to explore the rest of the novel partners identified in the BioID analysis . We invoked network propagation analytics to identify functional modules amongst novel TWIST1 BioID-interactors and to prioritize the key NCC regulators ( see Materials and methods ) . Network propagation , which is built on the concept of ‘guilt-by-association’ , is a set of analytics used for gene function prediction and module discovery ( Sharan et al . , 2007; Ideker and Sharan , 2008; Cowen et al . , 2017 ) . By propagating molecular and phenotypic information through connected neighbors , this approach identified and prioritized relevant functional clusters while eliminating irrelevant ones . The TWIST1 functional interaction network was constructed by integrating the association probability matrix of the BioID candidates based on co-expression , protein-interaction , and text mining databases from STING ( Singh and Gramolini , 2009; Szklarczyk et al . , 2015 ) . Markov clustering ( MCL ) was applied to the matrix for the inference of functional clusters ( Figure 1—figure supplement 2D , Supplementary file 2 ) . Additionally , data from a survey of the interaction of 56 transcription factors and 70 unrelated control proteins were used to distinguish the likely specific interactors from the non-specific and the promiscuous TF interactors ( Li et al . , 2015 ) . Specific TF interactors ( red ) and potential new interactors ( blue; Figure 1—figure supplement 2D–i ) clustered separately from the hubs predominated by non-specific interactors ( gray; Figure 1—figure supplement 2D–ii ) . The stringency of the screen was enhanced by increasing the statistical threshold ( adjusted p-value [adjp]<0 . 05 ) and excluding the clusters formed by non-specific interactors such as those containing heat-shock proteins and cytoskeleton components . Gene Ontology analysis revealed major biological activities of proteins in the clusters: chromatin organization , cranial skeletal development , and ribosome biogenesis ( Figure 1C; Supplementary file 2; Chen et al . , 2009 ) . Heat diffusion was applied to prioritize key regulators of NCC development . The stringent TWIST1 interaction network comprises proteins associated with facial malformation phenotypes in human/mouse ( HP:0001999 , MP:0000428 ) , which points to a likely role in NCC development . These factors were used as seeds for a heat diffusion simulation to find near-neighbors of the phenotype hot-spots ( i . e . additional factors that may share the phenotype ) and to determine their hierarchical ranking ( Figure 1C , Supplementary file 2 ) . As expected , that disease causal factors are highly connected and tend to interact with each other ( Jonsson and Bates , 2006 ) , a peak of proteins with high degrees of connectivity emerged among the top diffusion ranked causal factors , most of which are from the chromatin organization module ( Figure 1—figure supplement 2F ) . TWIST1 and these interacting chromatin regulators were referred to hereafter as the TWIST1-chromatin regulatory module ( TWIST1-CRM ) . Among the top 30 diffusion ranked BioID candidates , we prioritized nine for further characterization . These included chromatin regulators that interact with TWIST1 exclusively in NCCs versus 3T3 fibroblasts: the chromodomain helicases CHD7 , CHD8 , the histone methyltransferase WHSC1 and SMARCE1 , a member of the SWI/SNF chromatin remodeling complex ( Figure 1C , candidates name in red; Figure 1—figure supplement 2E , F; Supplementary file 3 ) . We also covered other types of proteins , including transcription factors PRRX1 , PRRX2 , TFE3 and the cytoplasmic phosphoprotein DVL1 ( Dishevelled 1 ) . The genes encoding these proteins were found to be co-expressed with Twist1 in the cranial NCCs of in embryonic day ( E ) 9 . 5 mouse embryos ( Figure 1D , Supplementary file 1; Bildsoe et al . , 2016; Fan et al . , 2016 ) . Co-immunoprecipitation ( co-IP ) assays showed that CHD7 , CHD8 , PRRX1 , PRRX2 , and DVL1 could interact with TWIST1 like known interactors TCF3 and TCF4 , while TFE3 and SMARCE1 did not show any detectable interaction ( Figure 2A ) . Fluorescent immunostaining demonstrated that these proteins co-localized with TWIST1 in the nucleus ( Figure 1—figure supplement 1C ) . The exceptions were DVL1 and TFE3 , which were localized predominantly in the cytoplasm ( Figure 1—figure supplement 1C ) . Among these candidates , CHD7 and CHD8 are known to engage in direct domain-specific protein-protein interactions ( Batsukh et al . , 2010 ) . Three sub-regions of CHD7 and CHD8 were tested for interaction with TWIST1 ( Figure 2B ) . For both proteins , the p1 region , which encompasses helicases and chromodomains , showed no detectable interaction with partial or full-length TWIST1 . In contrast , the p2 and the p3 regions of CHD7 and CHD8 interacted with full-length TWIST1 as well as with its N-terminal region ( Figure 2C ) . Reciprocally , the interaction was tested with different regions of TWIST1 including the bHLH domain , the WR domain , the C-terminal region and the N-terminal region ( Figure 2B ) . CHD7 , CHD8 , and WHSC1 interacted preferentially with the TWIST1 N-terminus whereas the TCF dimerization partners interacted specifically with the bHLH domain ( Figure 2D ) . Consistent with the co-IP result , SMARCE1 and TFE3 did not interact with TWIST1 . Interestingly , the other known factor that binds the TWIST1 N-terminal region is the histone acetyltransferase CBP/P300 which is also involved in chromatin remodeling ( Hamamori et al . , 1999 ) . These findings demonstrated the direct interaction of TWIST1 with a range of epigenetic factors and transcriptional regulators and identified the TWIST1 N-terminal region as the domain of contact . The function of the core components of the TWIST1-CRM was investigated in vivo using mouse embryos derived from ESCs that carried single-gene or compound heterozygous mutations of Twist1 and the chromatin regulators . Mutant ESCs for Twist1 and the three validated NCC-exclusive chromatin regulatory partners Chd7 , Chd8 , and Whsc1 were generated by CRISPR-Cas9 editing ( Figure 3—figure supplement 1A , B; Ran et al . , 2013 ) . ESCs of specific genotype ( non-fluorescent ) were injected into 8 cell host wild-type embryos ( expressing fluorescent DsRed . t3 ) and chimeras were collected at E9 . 5 or E11 . 5 ( Figure 3A; Sibbritt et al . , 2019 ) . Only embryos with predominant contribution of mutant ESCs , indicated by the absence or low level of DsRed . t3 fluorescence , were analyzed . The majority of embryos derived from single-gene heterozygous ESCs ( Twist1+/- , Chd8+/- , and Whsc1+/- ) displayed mild deficiency in the cranial neuroepithelium and focal vascular hemorrhage ( Figure 3B , C ) . Compound heterozygous embryos ( Twist1+/-;Chd7+/- , Twist1+/-;Chd8+/- , and Twist1+/-;Whsc1+/- ) displayed more severe craniofacial abnormalities and exencephaly ( Figure 3B , C ) . Given that CHD8 was not previously known to involve in craniofacial development of the mouse embryo , we focused on elucidating the impact of genetic interaction of Chd8 and Twist1 on NCC development in vivo . While Chd8+/- embryos showed incomplete neural tube closure , compound Twist1+/-;Chd8+/- embryos displayed expanded neuroepithelium , a phenotype not observed in the single-gene mutants ( Figure 3B , E ) . The population of NCCs expressing Tfap2α , a Twist1-independent NCC marker ( Brewer et al . , 2004 ) was reduced in the frontonasal tissue and the trigeminal ganglion ( Figure 3E–i , F ) . In contrast , SOX2 expression was upregulated in the ventricular zone of the neuroepithelium of mutant chimeras ( Figure 3E–ii , iii , G ) . Furthermore , Twist1+/- , Chd8+/- and Twist1+/-;Chd8+/- embryos displayed different degrees of hypoplasia of the NCC-derived cranial nerves ( Figure 3H ) . Cranial nerves III and IV were absent , and nerve bundle in the trigeminal ganglia showed reduced thickness , most evidently in the Twist1+/-;Chd8+/- compound mutant embryos ( Figure 3H , I ) . Altogether , these results suggested that TWIST1 genetically interaction with epigenetic regulators CHD7 , CHD8 , and WHSC1 to guide the formation of the cranial NCC and downstream tissue genesis in vivo . The loss of NCC progenitors and neural tube malformation indicate that the combined activity of TWIST1-chromatin regulators might be required from early in NCC development . To understand the molecular function of TWIST1-chromatin regulators in early NCC differentiation , we performed an integrative ChIP-seq analysis . The ChIP-seq dataset for TWIST1 was generated from the ESC-derived neuroepithelial cells ( NECs ) which are the source of early NCCs ( Figure 4—figure supplement 1 and Materials and methods ) . We retrieved published NEC ChIP-seq datasets for CHD7 and CHD8 and the histone modifications and reanalyzed the data following the ENCODE pipeline ( ENCODE Project Consortium , 2012; Sugathan et al . , 2014; Ziller et al . , 2015; Figure 4—figure supplement 1A ) . Two H3K36me3 ChIP-seq datasets for NECs were included in the analysis ( Du et al . , 2017; Chai et al . , 2018 ) on the basis that WHSC1 trimethyl transferase targets several H3 lysine ( Morishita et al . , 2014 ) and catalyzes H3K36me3 modification in vivo ( Nimura et al . , 2009 ) . Genome-wide co-occupancies of TWIST1 , CHD7 , and CHD8 showed significant overlap ( Fisher's exact test ) and clustered by Jaccard Similarity matrix ( Figure 4A ) . ChIP-seq peaks were correlated with active histone modifications H3K27ac and H3K4me3 but not the inactive H3K27me3 , or the WHSC1-associated H3K36me3 modifications ( Figure 4A ) . TWIST1 , CHD7 , and CHD8 shared a significant number of putative target genes ( Figure 4B ) . TWIST1 shared 63% of target genes with CHD8 ( odds ratio = 16 . 93 , Chi-squared test p-value<2 . 2e-16 ) and 18% with CHD7 ( odds ratio = 8 . 26 , p-value<2 . 2e-16; Figure 4B; Supplementary file 4 ) . Compared with genomic regions occupied by only one factor , greater percentage of regions with peaks for two or all three factors ( TWIST1 , CHD7 , and CHD8 ) showed H3K27ac and H3K4me3 signal ( Figure 4C ) . This trend was not observed for the H3K27me3 modification . Similarly , the co-occupied transcription start sites ( TSS ) showed open chromatin signatures with enrichment of H3K4me3 and H3K27Ac and depletion of H3K27me3 ( Ernst et al . , 2011; Rada-Iglesias et al . , 2011; Figure 4D , E ) . We also did not observe H3K36me3 modifications near the overlapping TSSs , suggesting that WHSC1 may have alternative histone lysine specificity in the NECs . The top Gene Ontology enriched for the co-occupied regulatory regions of two or three core components included neural tube patterning , cell migration and BMP signaling pathway ( Figure 4F ) . Overlapping peaks of the partners were localized within ± 1 kb of the TSS of common target genes ( Figure 4H; Supplementary file 4 ) . This integrative analysis revealed that the TWIST1-chromatin regulators shared genomic targets that are harbored in open chromatin in the NECs . To pinpoint more specifically when TWIST1-chromatin regulators are required and better interpret their transcriptional activities in light of the in vivo dynamics and timing of target gene activity , we examined relevant gene activities in the E8 . 5- E10 . 5 mouse NCCs scRNA-seq datasets of NCCs traced by Wnt1-Cre and Sox10-Cre reporters ( Soldatov et al . , 2019 ) . The first clue came from the expression of Twist1 , Chd7 , Chd8 , and Whsc1 in NCCs clusters that are ordered in developmental pseudotime: neural tube , delaminatory , early migratory , migratory 1 , migratory 2 , sensory , autonomic and mesenchyme ( Figure 4—figure supplement 2A–C ) . Twist1 displayed stage-specific dynamics and initially peaked in the early migratory NCC followed by exponential activation while progressing to the mesenchyme . On the other hand , the three interacting partners expressed rather ubiquitously throughout all NCC populations ( Figure 4—figure supplement 2C ) . We then examined the activities of genes with binding sites for TWIST1-chromatin regulators in their proximal regulatory elements . To narrow down to the most immediate targets , we limited the list to ChIP targets that are also responsive to Twist1 conditional knockdown in the E9 . 5 NCCs ( Bildsoe et al . , 2009 ) . Among all the NCC regulons , the binding sites of the TWIST1-module correlate best with the profile of early migratory NCCs ( Figure 4G ) . We also noted that the marker genes of early migratory and neural tubes are mutually exclusive ( Figure 4—figure supplement 2D ) . A substantial number of early migratory genes are repressed in the neural tube ( 62% ) , while the signature neural tube genes are downregulated at the early migratory stage ( 33% ) . TWIST1 and partners appeared to repress many of the neural tube or neurogenesis genes in the early migratory NCCs , including Sox2 , Foxb1 , Jag1 , En1 , Zic3 , and Dll3 ( Figure 4H ) . In summary , in the in vivo context , the initial manifestation of Twist1-modular activity starts from delamination and correlates best with the early migratory stage . This early function may be important for the newly delaminated NCC progenitors to proceed to migratory stages , through the repression of neural tube signatures and the enhancement of cell migration and early NCC genes . To examine whether TWIST1 is necessary to recruit partner proteins to specific regions of co-regulated genes or vice versa , we examined chromatin binding of the endogenous proteins in NECs by ChIP-qPCR analysis ( Figure 5A ) . As CHD8 correlates best with TWIST1 in their ChIP-seq profile surrounding TSS , we analyzed the pattern of recruitment of TWIST1 and CHD8 at the shared peaks near Sox2 , Epha3 , Pdgfra , and Vegfa . One of the peaks near the Sox2 TSS demonstrated binding by both TWIST1 and CHD8 ( Figure 5A , B ) . In Twist1+/- or Chd8+/- NECs , the binding of TWIST1 or CHD8 at the peak was reduced . Interestingly , Twist1+/- mutation also diminished the binding of CHD8 yet Chd8+/- mutation did not affect TWIST1 binding ( Figure 5A ) . For Epha3 , Vegfa , and Pdgfra , peaks identified by ChIP-seq with H3K4me3 or H3K27ac modifications were tested . Partial loss of Twist1 significantly reduced the recruitment of both TWIST1 and CHD8 but again , the loss of CHD8 only affected its own binding ( Figure 5C–E ) . These findings support that TWIST1 binding is a prerequisite for the recruitment of CHD8 . As the TWIST1 and partners were found to regulate cell migration and BMP signaling pathways through target gene binding , we again took a loss-of-function approach and examined the synergic function of TWIST1-chromatin regulatory factors on cell motility in both NECs and NCCs . The emigration of NECs from the colonies was captured by time-lapse imaging and was quantified ( see Materials and methods ) . While Chd7+/- , Chd8+/- , and Whsc1+/- mutant cells displayed marginally reduced motility , the motility of the Twist1+/- cells was compromised and further reduced in Twist1+/-; Chd7+/- , Twist1+/-; Chd8+/- , and Twist1+/-; Whsc1+/- compound mutant cells ( Figure 6A , B ) . Additionally , to validate the functional interaction of these factors in the later phase of NCC development , and test whether they favor mesenchymal versus the alternative branches of NCC lineages , we performed knockdown analysis in O9-1 neural crest stem cells . O9-1 cells display the transcriptional signature of the mouse ectomesenchymal NCCs and could generate cranial mesenchyme derivatives ( osteoblasts , chondrocytes , smooth muscle cells ) both in vitro and in vivo ( Ishii et al . , 2012 ) . NCCs were treated with siRNA to knockdown Chd7 , Chd8 , and Whsc1 activity individually ( single-gene knockdown ) and in combination with Twist1-siRNA ( compound knockdown; Figure 3—figure supplement 1C , see Materials and methods ) . NCCs treated with Chd8-siRNA or Whsc1-siRNA but not Chd7-siRNA showed impaired motility ( relative to control-siRNA treated cells ) , which was exacerbated by the additional knockdown of Twist1 ( Figure 6C ) . We also performed qPCR on cell type markers highlighted in scRNA-seq analysis of E8 . 5 - E10 . 5 mouse NCCs ( Soldatov et al . , 2019; Figure 6—figure supplement 1 ) . Specifically , we focused on the analysis of genes important for bifurcation events across the ectomesenchyme , autonomic , and sensory branches , as these genes may best report trans-differentiation activity ( Soldatov et al . , 2019 ) . Among these genes , we prioritized those with DNA-binding sites for TWIST1 in their regulatory elements . Impaired motility in Twist1 , Chd8 and Whsc1 knockdowns was accompanied by reduced expression of EMT genes ( Snail2 , Ddr2 , Pcolce , Pdgfrb , Mmp2 ) and ectomesenchyme markers ( Prrx1 , Prrx2 , Foxc1 , Snai1; Figure 6D ) . Combined knockdowns had a stronger impact on the expression of the target genes than individual knockdowns for Twist1 , Chd8 , and Whsc1 ( Figure 6D ) . On the other hand , changes in the expression of non-mesenchymal genes , that is , sensory/ autonomic neuron markers , were less robust ( Figure 6E ) . The significant upregulation of Sox2 , Sox8 , Sox10 , and Tubb3 may indicate a switch to autonomic neuron fate , which is the state immediately adjacent to the mesenchyme ( Soldatov et al . , 2019 ) . Genes more specific for sensory neurons were either below detection or not significantly affected . Prolonged knockdown treatment may be required to detect changes of these cell fate markers . These findings suggested that the persistent activity of TWIST1-CHD8/WHSC1 is required to confer ectomesenchyme propensity ( cell mobility , EMT , and mesenchyme differentiation ) , and repress neurogenic differentiation . The genomic , transcriptomic and embryo phenotypic data collectively pointed to a requirement of TWIST1-chromatin regulators in the newly delaminated NCCs for progression towards early migratory state . To better understand how TWIST1-chromatin regulators coordinate NCC identities at early specification , we studied the role of the module factors during neural differentiation of ESCs in vitro . ESCs were cultured in neurogenic differentiation media , followed by selection and expansion of NECs ( Figure 7A; Bajpai et al . , 2010; Varshney et al . , 2017 ) . All cell lines progressed in the same developmental trajectory ( Figure 7B–i ) and generated Nestin-positive rosettes typical of NECs ( Figure 7D ) . We assessed the lineage propensity of neuroepithelial cells derived from single-gene heterozygous ESCs ( Twist1+/- , Chd7+/- , Chd8+/- , and Whsc1+/- ) and compound heterozygous ESCs ( Twist1+/-;Chd7+/- , Twist1+/-;Chd8+/- , and Twist1+/-;Whsc1+/- ) . Samples were collected at day 0 ( ESCs ) , day 3 and day 12 of differentiation and assessed for the expression of cell markers and ChIP-seq target genes ( Supplementary file 5 ) . Genes were clustered into three groups by patterns of expression: activation , transient activation , and repression ( Figure 7B ii , black , red , gray clusters ) . Notably , Chd7 , Chd8 , and Whsc1 clustered with NCC specifiers that were activated transiently during differentiation ( Figure 7B ii , red ) . NCC and NSC marker genes responded inversely to the combined perturbation of Twist1 and chromatin regulators . Compound loss-of-function ( LOF ) reduced expression of the NCC specifiers and unleashed the expression of NSC TFs in Day-12 NECs ( Figure 7C , second and third row ) . In single-gene heterozygous cells , we observed only modest or no change in the gene expression . Sox2 , a driver of the NSC lineage and a repressor of NCC formation ( Mandalos et al . , 2014 ) , was repressed concurrently with the increased expression of Twist1 and the chromatin regulators during neurogenic differentiation ( Figure 7C ) . However , in the compound heterozygous cells , Sox2 transcript and protein were both upregulated compared to the wild-type cells , together with NSC markers TUBB3 and NES ( Figure 7C–E ) . Finally , the EMT genes were only affected by the compound knockdown at the ectomesenchyme stage ( Figure 7C , bottom row; see also Figure 6D , E ) . We focused on the effect of gene perturbation on the cell fate bias in late NECs by examining the expression of a broader panel of neural tube/NSC signatures ( Briscoe et al . , 2000; Alaynick et al . , 2011; Kutejova et al . , 2016; Figure 4—figure supplement 1D ) . The difference between WT and mutant cells in the dataset is primarily driven by changes in NCC specifiers and NSC TFs . In the compound mutant NECs , in addition to NCC identity ( Tfap2a , Msx1 , Msx2 , Zic1 , Id1 , and Id2 ) , expression of dorsal NSC markers were attenuated ( Gli3 , Rgs2 , Boc , and Ptn; Figure 7F , G ) . Meanwhile , the pan- and ventral-NSC markers Sox2 , Pax6 , Olig2 , Foxa2 , and Cited2 were ectopically induced ( Figure 7F , G: genes in red ) . ChIP-seq data showed that TWIST1 , CHD7 and CHD8 directly bind to the promoters of most of these genes ( Figure 4G , S5A , Supplementary file 4 ) . Collectively , the findings implicated that in the NEC progenitor populations , TWIST1-chromatin regulators may promote the dorsal-most NCC propensity by counteracting SOX2 and other NSC TFs . Loss of function of the module leads to the reversion of the early NCC progenitors to neural-tube-like cells , which may underpin the severe deficiency of NCCs and loss of their derivatives observed in the mutant embryos ( Figure 3 ) . Analysis of protein-protein interaction is a powerful approach to identify the connectivity and the functional hierarchy of different genetic determinants associated with an established phenotype ( Song and Singh , 2009; Mitra et al . , 2013; Sahni et al . , 2015; Cowen et al . , 2017 ) . We used TWIST1 as an anchor point and the BioID methodology to visualize the protein interactome necessary for NCCs development . Network propagation exploiting a similarity network built on prior associations enabled the extraction of clusters critical for neural crest function and pathology . Using this high-throughput analytic pipeline , we were able to identify the core components of the TWIST1-CRM that guides NCC lineage development . Among the interacting factors were members of the chromatin regulation cluster , which show dynamic component switching between cell types , and may confer tissue-specific activities . The architecture of the modular network reflects the biological organization of chromatin remodeling machinery , which comprises multi-functional subunits with conserved and cell-type-specific components ( Meier and Brehm , 2014 ) . Previous network studies reported that disease-causal proteins exist mostly at the center of large clusters and have a high degree of connectivity ( Jonsson and Bates , 2006; Ideker and Sharan , 2008 ) . We did not observe an overall correlation between disease probability and the degree of connectivity or centrality for factors in the TWIST1 interactome ( Figure 1—figure supplement 2F ) . However , the topological characteristics of the chromatin regulatory cluster resembled the features of disease modules and enriched for craniofacial phenotypes . In contrast , the ‘ribosome biogenesis’ module that was also densely inter-connected , was void of relevant phenotypic association ( Figure 1C ) . Network propagation is , therefore , an efficient way to identify and prioritize important clusters while eliminating functionally irrelevant ones . Based on these results , we selected core TWIST1-CRM epigenetic regulators CHD7 , CHD8 , and WHSC1 , and demonstrated their physical and functional interaction with TWIST1 . In the progenitors of the NCCs , these factors displayed overlapping genomic occupancy that correlated with the active chromatin marks in the fate specification genes in the neuroepithelium . Combinatorial perturbation of the disease ‘hot-spots’ in TWIST1-CRM impacted adversely on NCC specification and craniofacial morphogenesis in mouse embryos , which phenocopy a spectrum of human congenital malformations associated with NCC deficiencies ( Johnson et al . , 1998; Chun et al . , 2002; Cai et al . , 2003; Bosman et al . , 2005; Bernier et al . , 2014; Schulz et al . , 2014; Battaglia et al . , 2015; Etchevers et al . , 2019 ) . These observations revealed CHD8 and WHSC1 as putative determinants for NCC development and neurocristopathies . While CHD8 is associated with autism spectrum disorder ( Bernier et al . , 2014; Katayama et al . , 2016 ) , its function for neural crest development has never been reported . Here , we demonstrated that the loss of Chd8 affected NCC migration and trigeminal sensory nerve formation in vivo , in a Twist1-dependent manner . We showed that TWIST1 occupancy is a requisite for CHD8 recruitment to common target genes . CHD8 may initiate chromatin opening and recruit H3-lysine tri-methyltransferases ( Zhao et al . , 2015 ) such as WHSC1 ( Figure 7—figure supplement 1 ) . The TWIST1-CHD8 complex may also repress neurogenic genes by blocking the binding of the competitive TFs . The specificity for NCC differentiation genes might be achieved when TWIST1 , CHD8 and additional factors bind adjacently to each other , either sequentially or simultaneously . ChIP-seq peaks shared between TWIST1 and CHD8 or unique to each of them were enriched in different sets of motifs matched to various transcription factors ( Figure 4—figure supplement 3A ) . Interestingly , only factors binding to TWIST1+CHD8 peaks show enriched expression in the delaminatory and migratory NCC populations ( Figure 4—figure supplement 3B ) . These results suggested that TWIST1-CHD8 module may interact with additional factors , such as DLX1 and SOX10 , to enhance NCC identity . We also showed that WHSC1 is required in combination with TWIST1 to promote NCC fate and tissue patterning . Unlike CHD8 and WHSC1 , CHD7 has been previously implicated in neurocristopathy ( CHARGE syndrome ) and the motility of NCCs ( Schulz et al . , 2014; Okuno et al . , 2017 ) . Our study has corroborated these findings while also showing that CHD7 interacts with TWIST1 to promote NCC differentiation . One limitation of the mutant study is that there was no parallel comparison of the mutant phenotype in chimeras derived from multiple mutant cell lines to rule out the possibility of off-target gene-editing . However , in view of the potential variation of the ESC contribution to the chimera , We took a more productive approach to analyze multiple chimeric embryos from one line to glean a consistent phenotype that may inform the impact of genetic interaction Twist1 and the chromatin regulators on development . In sum , we propose the TWIST1-CRM as a unifying model that connects previously unrelated regulatory factors implicated in different rare diseases and predicts their functional inter-dependency in NCC development ( Figure 7—figure supplement 1 ) . Other epigenetic regulators identified in the TWIST1-interactome , such as PBRM1 , ZFP62 , and MGA , may contribute to the activity of this regulatory module in the NCC lineage . The segregation of neuroepithelial cells to NCC and NSC lineages is the first event of NCC differentiation . Our results show that the lineage allocation may be accomplished by the opposing activity of core members of the TWIST1-CRM and NSC TFs such as SOX2 . Sox2 expression is continuously repressed in the NCC lineage ( Wakamatsu et al . , 2004; Cimadamore et al . , 2011; Soldatov et al . , 2019 ) , likely through direct binding and inhibition by TWIST1-CHD8 at Sox2 promotor . In Twist1+/-;Chd8+/- mutant embryos , the aberrant upregulation of Sox2 correlated with deficiency of NCC derivatives and the expanded neuroepithelium of the embryonic brain . In a similar context , Sox2 overexpression in chicken neuroepithelium blocks the production of TFAP2α-positive NCC ( resulting in the loss of cranial nerve ganglia ) and suppresses the expression of EMT genes and NCC migration ( Wakamatsu et al . , 2004; Remboutsika et al . , 2011 ) . On the contrary , conditional knockout of Sox2 results in ectopic production and abnormal migration of NCCs , and attenuation of the neuroepithelium ( Mandalos et al . , 2014 ) . NCCs are derived from the neuroepithelium in a series of cell fate specification events ( Soldatov et al . , 2019 ) . Classical studies of systemic and conditional Twist1 knockout embryos have shed light on the in vivo implications of our molecular characterizations . Neural tube thickening and distortion were observed in the homozygous null mutant embryo . Cell number in the neural tube was doubled but was not due to altered cell proliferation or reduced cell death ( Vincentz et al . , 2008 ) . Twist1 null mutant has an expanded domain of Wnt1-Cre expression in the neural tube , and the NCC cells are frequently accumulated in the vicinity of the neuroepithelium ( Chen and Behringer , 1995; Soo et al . , 2002; Vincentz et al . , 2008 ) . Based on these phenotypes and our data from the knockout NECs , is possible that in the absence of Twist1 activity , the process of cellular delamination is impaired leading to the retention of the NCC progenitors in the neuroepithelium or at ectopic sites . TWIST1 and the chromatin regulators cooperatively drive the progression along the lineage trajectory at different phases of NCC differentiation . Twist1 is activated from delamination and its expression steadily increases during differentiation . The functional interaction of TWIST1 and the chromatin regulators may commence at the transition from delaminatory to early migratory stage , coinciding with the first peak of Twist1 expression . The expressions of NCC specification ( Msx1/2 , Zic1 ) or migratory genes ( Tfap2a ) were compromised by the loss of the TWIST1-chromatin regulators , albeit they are activated before Twist1 during development ( Soldatov et al . , 2019 ) . The early activity of Twist1 , Chd7 , Chd8 , and Whsc1 module might stabilize the activity of these early NCC genes . Loss of the module in NECs leads to reversion to NSC fate at the expense of the NCC lineage . In the post-migratory NCCs , modular activity of Twist1 and Chd8/ Whsc1 promotes the ectomesenchyme propensity while represses alternative cell fates such as autonomic neurons . Chd7 activity was not connected with EMT in the NCCs , suggesting that its role may be different from CHD8 and WHSC1 . The early versus late TWIST1-module activities are associated with the activation of different groups of target genes , suggesting that phasic deployment of the regulatory module may navigate the cells along the trajectory of NCC development . Overexpression of Twist1 in NCCs disrupts the formation of thoracic sympathetic chain ganglia , of the autonomic nervous system ( Vincentz et al . , 2013 ) . On the other hand , in the Wnt1-Cre conditional Twist1 knockout mutants , differentiation of cardiac neural crest cells into neuronal cells reminiscent of those in the sympathetic ganglia was found ectopically in the cardiac outflow tract ( Vincentz et al . , 2013 ) . These aggregates of cardiac NCCs expressed pan-neuronal Tubb3 and markers specific for the autonomic nerve cells ( Sox10 , Phox2b , and Ascl1 ) , but did not express sensory neuron markers ( TrkA , Pou4f1 , and NeuroD1 ) . Similarly , NCCs losing the function of TWIST1-module in vitro showed sign of mesenchyme to autonomic state trans-differentiation , whereas the sensory markers in these cells remain repressed . In the heterozygote null mutants , we did not observe any ectopic gain of neurons , but loss and disorganization of trigeminal nerves ( composed of sensory and motor neurons ) , similar to the phenotype of null mutants ( Ota et al . , 2004 ) . This may result from the loss of early function of Twist1 in early migratory NCC formation , rather than its late activity on automimic-mesenchymal bifurcation . In conclusion , by implementing an analytic pipeline to decipher the TWIST1 interactome , we have a glimpse of the global molecular hierarchy of NCC development . We have characterized the cooperative function of core components of TWIST1-CRM including the TWIST1 and chromatin regulators CHD7 , CHD8 , and WHSC1 . We demonstrated that this module is a dynamic nexus to drive molecular mechanisms for orchestrating NCC lineage progression and repressing NSC fate , enabling the acquisition of ectomesenchyme propensity . The TWIST1-chromatin regulators and the NSC regulators therefore coordinate the molecular cross-talk between the ectomesenchyme and neurogenic progenitors of the central and peripheral nervous systems , which are often affected concurrently in a range of human congenital diseases . O9-1 cells were purchased from Millipore and 3T3 cells were purchased from ATCC . A2loxCre mouse ESCs were a gift from the Kyba Lab ( Lillehei Heart Institute , Minnesota , USA ) . Derivatives of A2loxCre ESCs were generated in the lab . Cell line identities were authenticated by genotyping , and all cell lines were tested free of mycoplasma . O9-1 cells ( passage 20–22 , Millipore cat . #SCC049 ) were maintained in O9-1 medium: high glucose DMEM ( Gibco ) , 12 . 5% ( v/v ) heat-inactivated FBS ( Fisher Biotec ) , 10 mM β-mercaptoethanol , 1X non-essential amino acids ( 100X , Thermo Fisher Scientific ) , 1% ( v/v ) nucleosides ( 100X , Merck ) and 10 mil U/mL ESGRO mouse leukaemia inhibitory factor ( Merck ) and 25 ng/mL FGF-2 ( Millipore , Cat . #GF003 ) . For each replicate experiment , 1 . 5 × 106 cells per flask were seeded onto 4T75 flasks 24 hr before transfection . The next day PcDNA 3 . 1/ Twist1-BirA-HA plasmid or PcDNA 3 . 1/ GFP-BirA*-HA plasmid was transfected into cells using Lipofectamine 3000 ( Life Technologies ) according to the manufacturer's instructions . Biotin ( Thermo Scientific , cat . #B20656 ) was applied to the medium at 50 nM . Cells were harvested 16 hr post-transfection , followed by snap-freeze liquid nitrogen storage or resuspension in lysis buffer . All steps were carried out at 4°C unless indicated otherwise . Cells were sonicated on the Bioruptor Plus ( Diagenode ) , 30 s on/off for five cycles at high power . An equal volume of cold 50 mM Tris-HCl , pH 7 . 4 , was added to each tube , followed by two 30 s on/off cycles of sonication . Lysates were centrifuged for 15 min at 14 , 000 rpm . Protein concentrations were determined by Direct Detect Infrared Spectrometer ( Merck ) . Cleared lysate with equal protein concentration for each treatment was incubated with pre-blocked streptavidin Dynabeads ( MyOne Streptavidin C1 , Invitrogen , cat . #65002 ) for 4 hr . Beads were collected and washed sequentially in Wash Buffer 1–3 with 8 min rotation each , followed by quick washes with cold 1 mL 50 mM Tris·HCl , pH 7 . 4 , and 500 μL triethylammonium bicarbonate ( 75 mM ) . Beads were then collected by spinning ( 5 min at 2000 × g ) and processed for mass spectrometry analysis . Tryptic digestion of bead-bound protein was performed in 5% w/w trypsin ( Promega , cat . #V5280 ) , 50 mM triethylammonium bicarbonate buffer at 37°C overnight . The supernatant was collected and acidified with trifluoroacetic acid ( TFA , final concentration 0 . 5% v/v ) . Proteolytic peptides were desalted using Oligo R3 reversed phase resin ( Thermo Fisher Scientific ) in stage tips made in-house ( Rappsilber et al . , 2007 ) . Peptides were fractioned by hydrophilic interaction liquid chromatography using an UltiMate 3000 HPLC ( Thermo Fisher Scientific ) and a TSKgel Amide-80 HILIC 1 mm ×250 mm column . Peptides were eluted in a gradient from 100% mobile phase B ( 90% acetonitrile , 0 . 1% TFA , 9 . 9% water ) to 60% mobile phase A ( 0 . 1% TFA , 99 . 9% water ) for 35 min at 50 µL/min and fractions collected in a 96-well plate , followed by vacuum centrifugation to dryness . Dried peptide pools were reconstituted in 0 . 1% formic acid in the water , and 1/10th of samples were analyzed by LC-MS/MS . Mass spectrometry was performed using an LTQ Velos-Orbitrap MS ( Thermo Fisher Scientific ) coupled with an UltiMate RSLCnano-LC system ( Thermo Fisher Scientific ) . A volume of 5 µL was loaded onto a 5 mm C18 trap column ( Acclaim PepMap 100 , 5 µm particles , 300 µm inside diameter , Thermo Fisher Scientific ) at 20 µL/ min for 2 . 5 min in 99% phase A ( 0 . 1% formic acid in water ) and 1% phase B ( 0 . 1% formic acid , 9 . 99% water and 90% acetonitrile ) . The peptides were eluted through a 75 µm inside diameter column with integrated laser-pulled spray tip packed to a length of 20 cm with Reprosil 120 Pur-C18 AQ 3 µm particles ( Dr . Maisch ) . The gradient was from 7% phase B to 30% phase B in 46 . 5 min , to 45% phase B in 5 min , and to 99% phase B in 2 min . The mass spectrometer was used to apply 2 . 3 kV to the spray tip via a pre-column liquid junction . During each cycle of data-dependent MS detection , the ten most intense ions within m/z 300–1500 above 5000 counts in a 120 , 000 resolution orbitrap MS scan were selected for fragmentation and detection in an ion trap MS/MS scan . Other MS settings were: MS target was 1 , 000 , 000 counts for a maximum of 500 ms; MS/MS target was 50 , 000 counts for a maximum of 300 ms; isolation width , 2 . 0 units; normalized collision energy , 35; activation time 10 ms; charge state one was rejected; mono-isotopic precursor selection was enabled; dynamic exclusion was for 10 s . CRISPR-Cas9-edited mESCs were generated as described previously ( Sibbritt et al . , 2019 ) . Briefly , 1–2 gRNAs for target genes were ligated into pSpCas9 ( BB ) −2A-GFP ( PX458 , addgene plasmid #48138 , a gift from Feng Zhang ) . Three µg of pX458 containing the gRNA was electroporated into 1 × 106 A2loxCre ESCs or A2loxCre Twist1+/- cells ( clone T2-3 , generated by the Vector and Genome Engineering Facility at the Children’s Medical Research Institute ) using the Neon Transfection System ( Thermo Fisher Scientific ) . Electroporated cells were plated as single cells onto pre-seeded lawns of mouse embryonic fibroblasts ( MEF ) , and GFP expressing clones grown from single cells were selected under the fluorescent microscope . In total , 30–40 clones were picked for each electroporation . For mutant ESC genotyping , clones were expanded and grown on a gelatin-coated plate for three passages , to remove residue MEFs contamination . For genotyping , genomic lysate of ESCs was used as input for PCR reaction that amplified region surrounding the mutation site ( ± 200–500 bp flanking each side of the mutation ) . The PCR product was gel purified and sub-cloned into the pGEM-T Easy Vector System ( Promega ) as per manufacturer's protocol . At least 10 plasmids from each cell line were sequenced to ascertain monoallelic frameshift mutation and exclude biallelic mutations . ARC/s and DsRed . T3 mice were purchased from the Australian Animal Resources Centre and maintained as homozygous breeding pairs . ESC clones with monoallelic frameshift mutations and the parental A2LoxCre ESC line were used to generate chimeras . Embryo injections were performed as previously described ( Sibbritt et al . , 2019 ) . Briefly , 8–10 ESCs were injected per eight-cell DsRed . T3 embryo ( harvested at 2 . 5 dpc from super-ovulated ARC/s females crossed to DsRed . T3 stud males ) and incubated overnight . Ten to 12 injected blastocysts were transferred to each E2 . 5 pseudo-pregnant ARC/s female recipient . E9 . 5 and E11 . 5 embryos were collected 6 and 8 days after transfer to pseudo-pregnant mice . Embryos showing red fluorescent signal indicating no or low ESC contribution were excluded from the phenotypic analysis . Animal experimentations were performed in compliance with animal ethics and welfare guidelines stipulated by the Children's Medical Research Institute/Children's Hospital at Westmead Animal Ethics Committee , protocol number C230 . Whole-mount fluorescent immunostaining of mouse embryos was performed by following the procedure of Adameyko et al . , 2012 with minor modifications . Embryos were fixed for 6 hr in 4% paraformaldehyde ( PFA ) and dehydrated through a methanol gradient ( 25% , 50% , 75% , 100% ) . After 24 hr of incubation in 100% methanol at 4°C , embryos were transferred into bleaching solution ( 1 part of 30% hydrogen peroxide to 2 parts of 100% methanol ) for another 24 hr ( 4°C ) . Embryos were then washed with 100% methanol ( 10 min x3 at room temperature ) , post-fixed with Dent’s Fixative ( dimethyl sulfoxide: methanol = 1:4 ) overnight at 4°C . Embryos were blocked for 1 hr on ice in blocking solution ( 0 . 2% BSA , 20% DMSO in PBS ) with 0 . 4% Triton . Primary antibodies mouse 2H3 ( for neurofilament 1:1000 ) and rabbit α-TFAP2A ( 1:1000 ) or were diluted in blocking solution and incubated for four days at room temperature , and secondary antibodies ( Goat α-Rabbit Alexa Fluor 633; Goat α-Mouse Alexa Fluor 488 and DAPI , Thermo Fisher Scientific ) were incubated overnight in blocking solution at room temperature . Additional information of the antibodies used are listed in Key Resources Table . Embryos were cleared using BABB ( 1part benzyl alcohol: two parts benzyl benzoate ) , after dehydration in methanol , and imaged using a Carl Zeiss Cell Observer SD spinning disc microscope . Confocal stacks through the embryo were acquired and then collapsed . Confocal stacks were produced containing ~150 optical slices . Bitplane IMARIS software was used for 3D visualization and analysis of confocal stacks . Optical sections of the 3D embryo were recorded using ortho/oblique functions in IMARIS software . The surface rendering wizard tool was used to quantify SOX2 expression in the ventricular zone by measuring the immunofluorescence intensity on three separate z-plane sections per volume of the region of each embryo . The data were presented graphically as the ratio of intensity/ volume . ESC lines generated are listed in Key Resources Table . A2loxCre Mouse ESCs ( Mazzoni et al . , 2011 ) was a gift from Kyba Lab ( Lillehei Heart Institute , Minnesota , USA ) . A2loxCre with Twist1 bi-allelic knockout background was generated by CRISPR-Cas9 , as described below . The inducible Twist1 ESC line was generated using the inducible cassette exchange method described previously ( Iacovino et al . , 2014 ) . The TWIST1 coding sequence was then cloned from the mouse embryo cDNA library into the p2lox plasmid downstream of the Flag tag ( Iacovino et al . , 2014 ) . The plasmid was transfected into A2loxCre ( Twist1 -/- ) treated with 1 μg/mL doxycycline for 24 hr . The selection was performed in 300 μg/mL of G418 ( Gibco ) antibiotic for 1 week . Colonies were then picked and tested for TWIST1 expression following doxycycline treatment . ESC lines generated in this study were differentiated into neural epithelial cells ( NECs ) following established protocols ( Bajpai et al . , 2010; Varshney et al . , 2017 ) with minor modifications . ESCs were expanded in 2i/LIF media ( Ying et al . , 2008 ) for 2–3 passages . Neurogenic differentiation was initiated by plating ESC in AggreWells ( 1 × 106 per well ) using feeder independent mESC . Colonies were then lifted from AggreWells and grown in suspension in Neurogenic Differentiation Media supplemented with 15% FBS with gentle shaking for 3 days . Cell colonies were transferred to gelatin-coated tissue culture plates and cultured for 24 hr at 37°C under 5% CO2 . Cells were selected in insulin-transferrin-selenium ( ITS ) -Fibronectin media for 6–8 days at 37°C and 5% CO2 , with a change of media every other day . Accutase ( Stemcell Technologies ) was used to dissociate cells from the plate , allowing the removal of cell clumps . NECs were collected by centrifugation and plated on Poly-L-ornithine ( 50 μg/mL , Sigma-Aldrich ) and Laminin ( 1 μg/mL , Novus Biological ) coated dishes . For expansion of the cell line , cells were cultured in Neural Expansion Media ( 1 . 5 mg/mL Glucose , 73 μg/mL L-glutamine , 1x N2 media supplement [R and D systems] in Knockout DMEM/F12 [Invitrogen] , 10 ng/mL FGF-2 , and 1 μg/mL Laminin [Novus Biologicals] ) . During this period , cells were lifted using Accutase and cell rosette clusters were let settle and were removed for two passages to enrich for pre-EMT NCC populations . ESC with genotype Twist1-/-; Flag-Twist1 O/E and Twist1-/- were differentiated into NEC for 3 days following established protocol ( Varshney et al . , 2017 ) and were collected in ice-cold DPBS . Following a cell count , approximately 2 × 107 cells were allocated per cell line per ChIP . ChIP-seq assays were performed as previously described ( Bildsoe et al . , 2016 ) . In brief , chromatin was crosslinked and sonicated on the Bioruptor Plus ( Diagenode ) using the following program: 30 s on/off for 40 min on High power . The supernatant was incubated with α-TWIST1 ( Abcam , at . #ab50887 ) antibody conjugated Dynabeads overnight at 4°C . The protein-chromatin crosslinking is reversed by incubation at 65°C for 6 hr . The DNA is purified using RNase A and proteinase K treatments , extracted using phenol-chloroform-isoamyl alcohol ( 25:24:1 , v/v ) and precipitated using glycogen and sodium acetate . The precipitated or input chromatin DNA was purified and converted to barcoded libraries using the TruSeq ChIP Sample Prep Kit ( Illumina ) . Then 101 bp paired-end sequencing was performed on the HiSeq 4000 ( Illumina ) . ChIP-seq quality control results and analysis can be found in Figure 4—figure supplement 1 . Adaptors from raw sequencing data were removed using Trimmomatic ( Bolger et al . , 2014 ) and aligned to the mm10 mouse genome ( GENCODE GRCm38 . p5 ) ; ( Frankish et al . , 2019 ) using BWA aligner ( Li and Durbin , 2009 ) , and duplicates/unpaired sequences were removed using the picardtools ( http://broadinstitute . github . io/picard/ ) . MACS2 package ( Zhang et al . , 2008 ) was used for ChIP-seq peak calling for both Twist1-/-; Flag-Twist1 O/E and Twist1-/- IP samples against genomic input . IDR analysis was performed using the p-value as the ranking measure , with an IDR cut-off of 0 . 05 . Peak coordinates from the two replicates were merged , using the most extreme start and end positions . The raw and processed data were deposited into the NCBI GEO database and can be accessed with the accession number GSE130251 . Public ChIP-seq datasets for CHD7 , CHD8 , and histone modifications in NECs were selected based on the quality analysis from the Cistrome Data Browser ( http://cistrome . org/db/#/ ) and ENCODE guideline ( ENCODE Project Consortium , 2012; Mei et al . , 2017 ) . Datasets imported for analysis are listed in Supplementary file 6 . To facilitate comparison with datasets generated from human samples , TWIST1 ChIP sequences were aligned to the hg38 human genome by BWA . ChIP peak coordinates from this study were statistically compared using fisher’s exact test ( cut-off: p-value<0 . 05 , odds ration >10 ) and visualized using Jaccard similarity score . Analysis were performed with BEDTools ( Quinlan and Hall , 2010 ) . ChIP-seq peaks for TWIST1 , CHD7 and CHD8 were extended to uniform 1 kb regions , and regions bound by single factors or co-occupied by two or three factors were identified . The Genomic Regions Enrichment of Annotations Tool ( GREAT ) was used to assigns biological functions to genomic regions by analyzing the annotations of the nearby genes ( McLean et al . , 2010 ) . Significance by both binomial and hypergeometric test ( p<0 . 05 ) were used as cut-off . Genes with TSS ± 5 kb of the peaks were annotated using ChIPpeakAnno package in R . List of target genes was compared between each CHD7 , CHD8 , and TWIST1 . Bam files for each experiment were converted to bigwig files for ChIP-seq density profile , chromosome footprint , and IGV track visual analysis . scRNA-seq datasets for cranial E8 . 5 , vagal/trunk E9 . 5 , hindlimb/tail E10 . 5 and cardiac E10 . 5 Wnt1-traced , E9 . 5 anterior and E9 . 5 posterior Sox10-traced NCCs were obtained from GEO database ( GSE129114 ) ( Soldatov et al . , 2019 ) . Tables of per-gene read counts in each cell were imported into R . Single-cell datasets were pre-processed using Seurat package ( Stuart et al . , 2019 ) , which includes pre-processing , normalization , and joint analysis of multiple datasets . Only the cells with more than 4000 expressed genes were included in the downstream analysis . Additionally , only genes with more than 10 mapped reads and detected in at least 10 cells were considered in the downstream analysis . For reproducible result , we imported cell clustering , annotation and t-SNE embedding for visualization from the original publication ( http://pklab . med . harvard . edu/ruslan/neural . crest . html ) . Marker genes enriched in each cluster compared to all other clusters were determined using the Wilcoxon Rank Sum test . Enrichment TWIST1-module transcriptional targets in NCC cluster markers genes were analysed using goseq R package ( Young et al . , 2010 ) . To narrow down to the most immediate targets , we limited the list to ChIP targets that are also responsive to Twist1 conditional knockdown in the E9 . 5 NCCs ( Bildsoe et al . , 2009 ) . Random sampling was performed to generate a null distribution for each motif category and calculate its significance for over-representation amongst NCC regulons . Scratch Assays were performed on O9-1 cells following transient siRNA lipofectamine transfections . O9-1 cells were seeded at a density of 0 . 5 × 105 cells per well on Matrigel-coated 24-well plates on the day of transfection . 20 pmol of siRNA for candidate gene ( Chd7 , Chd8 , or Whsc1 ) and 20 pmol siRNA for Twist1 or control was applied per well ( 24-well-plate ) , plus 3 μL lipofectamine RNAiMAX reagent ( Thermo Fisher Scientific , cat . #13778075 ) , following manufacturer protocol . Knockdown efficiency was assessed by qRT-PCR ( Figure 7—figure supplement 1 ) . Forty-eight hr after transfection , a scratch was made in the confluent cell monolayer . Live images were taken with the Cell Observer Widefield microscope ( ZEISS international ) under standard cell culture conditions ( 37°C , 5% CO2 ) . Bright-field images were captured at set tile regions every 15 min over a 10-hr period . The total migration area from the start of imaging to when the first cell line closed the gap was quantified by Fiji software ( Schindelin et al . , 2012 ) . cDNA synthesis , from 1 µg total RNA from each sample , was performed using the RT2 Microfluidics qPCR Reagent System ( Qiagen , Cat . # 330431 ) . cDNAs were pre-amplified using the primer Mix for reporter gene sets ( Supplementary file 5 ) . High-throughput gene expression analysis ( BioMarkTM HD System , Fluidigm ) was then performed using the above primer set . Raw data were extracted using the Fluidigm Real-Time PCR Analysis Software , and subsequent analysis was performed in R-studio . Ct values flagged as undetermined or higher than the threshold ( Ct >24 ) were assigned as missing values . Samples with a measurement for only one housekeeping gene or samples with measurements for <30 genes were excluded from further analysis . Genes missing values for more than 30 samples were also excluded from further analysis . Data were normalized using expressions of the average of three housekeeping genes ( Gapdh , Tbp , Actb ) . Regularized-log transformation of the count matrix was then performed , and the PCA loading gene was generated using functions in the DEseq2 package . Differential gene expression analysis was performed using one-way ANOVA .	Shaping the head and face during development relies on a complex ballet of molecular signals that orchestrates the movement and specialization of various groups of cells . In animals with a backbone for example , neural crest cells ( NCCs for short ) can march long distances from the developing spine to become some of the tissues that form the skull and cartilage but also the pigment cells and nervous system . NCCs mature into specific cell types thanks to a complex array of factors which trigger a precise sequence of binary fate decisions at the right time and place . Amongst these factors , the protein TWIST1 can set up a cascade of genetic events that control how NCCs will ultimately form tissues in the head . To do so , the TWIST1 protein interacts with many other molecular actors , many of which are still unknown . To find some of these partners , Fan et al . studied TWIST1 in the NCCs of mice and cells grown in the lab . The experiments showed that TWIST1 interacted with CHD7 , CHD8 and WHSC1 , three proteins that help to switch genes on and off , and which contribute to NCCs moving across the head during development . Further work by Fan et al . then revealed that together , these molecular actors are critical for NCCs to form cells that will form facial bones and cartilage , as opposed to becoming neurons . This result helps to show that there is a trade-off between NCCs forming the face or being part of the nervous system . One in three babies born with a birth defect shows anomalies of the head and face: understanding the exact mechanisms by which NCCs contribute to these structures may help to better predict risks for parents , or to develop new approaches for treatment .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2021	TWIST1 and chromatin regulatory proteins interact to guide neural crest cell differentiation
Semen is the main vector for HIV transmission and contains amyloid fibrils that enhance viral infection . Available microbicides that target viral components have proven largely ineffective in preventing sexual virus transmission . In this study , we establish that CLR01 , a ‘molecular tweezer’ specific for lysine and arginine residues , inhibits the formation of infectivity-enhancing seminal amyloids and remodels preformed fibrils . Moreover , CLR01 abrogates semen-mediated enhancement of viral infection by preventing the formation of virion–amyloid complexes and by directly disrupting the membrane integrity of HIV and other enveloped viruses . We establish that CLR01 acts by binding to the target lysine and arginine residues rather than by a non-specific , colloidal mechanism . CLR01 counteracts both host factors that may be important for HIV transmission and the pathogen itself . These combined anti-amyloid and antiviral activities make CLR01 a promising topical microbicide for blocking infection by HIV and other sexually transmitted viruses . The majority of new HIV-1 infections are transmitted via sexual intercourse , and semen is the main vector for viral spread . Far from being a passive vehicle , semen potently enhances HIV infectivity ( Münch et al . , 2007; Kim et al . , 2010 ) . This HIV-enhancing activity is attributed to seminal amyloid fibrils ( Münch et al . , 2007; Kim et al . , 2010; Roan et al . , 2011; Usmani et al . , 2014 ) that form by self-assembly of proteolytic fragments of prostatic acid phosphatase ( PAP248-286 and PAP85-120 ) and the homologous proteins semenogelin 1 ( SEM1 ) and semenogelin 2 ( SEM2 ) ( Münch et al . , 2007; Roan et al . , 2011; Arnold et al . , 2012; Castellano and Shorter , 2012; Münch et al . , 2014 ) . Seminal amyloid fibrils are highly cationic , and the positively charged fibrils capture HIV virions , increase viral attachment rates to target cells , and augment fusion ( Roan et al . , 2009; Arnold et al . , 2012; Usmani et al . , 2014 ) . By doing so , fibrils promote HIV infection in vitro by several orders of magnitude , whereas the corresponding monomeric peptides have no effect ( Münch et al . , 2007; Roan et al . , 2011; Arnold et al . , 2012 ) . Importantly , the stimulatory effect of seminal amyloid is the greatest at low virus concentrations ( Münch et al . , 2007 ) , and semen and PAP248-286 fibrils ( termed SEVI for Semen-derived Enhancer of Virus Infection ) may facilitate vaginal virus transmission after exposure to low viral doses ( Miller et al . , 1994; Neildez et al . , 1998; Münch et al . , 2013 ) . HIV transmission rates are relatively low , occurring as infrequently as 1 in 200 to as low as 1 in 10 , 000 coital acts ( Gray et al . , 2001 ) . Thus , counteraction of infectivity promoting amyloids in semen should reduce or even prevent HIV transmission via the sexual route . The lysine- and arginine-specific molecular tweezer , CLR01 ( Figure 1A , B ) ( Fokkens et al . , 2005; Klärner et al . , 2006 , 2010 ) , inhibits amyloid fibrillization by engaging specific lysine , arginine , or both residues within a variety of disease-associated amyloidogenic proteins including amyloid-β protein ( Aβ ) , tau , islet amyloid polypeptide , transthyretin , and α-synuclein ( Sinha et al . , 2011; Attar et al . , 2012; Prabhudesai et al . , 2012; Sinha et al . , 2012; Acharya et al . , 2014; Ferreira et al . , 2014; Lopes et al . , 2015; Zheng et al . , 2015 ) . Furthermore , CLR01 has even been found to slowly remodel preformed Aβ and α-synuclein fibrils over the course of several weeks ( Sinha et al . , 2011; Prabhudesai et al . , 2012 ) . CLR01 binds lysine residues with a Kd of ∼10 µM and also binds arginine residues , albeit with ∼10-fold lower affinity ( Fokkens et al . , 2005; Dutt et al . , 2013 ) . The unprecedented high specificity of CLR01 for basic amino acids relies on a unique binding mode in which the tweezer draws the cationic side chains into its torus-shaped cavity and engages the ammonium cation of lysine or the guanidinium cation of arginine with its anionic phosphate group in a tight ion pair ( Figure 1B ) ( Klärner and Schrader , 2013 ) . No other amino acids fulfill the requirements for this threading mechanism . The structure of the CLR01-lysine complex and the precise mechanism of lysine threading into the CLR01 guest cavity and subsequent ion pairing have been extensively characterized by NMR spectroscopy , crystal structure , molecular dynamics , and quantum mechanics/molecular mechanics ( QM/MM ) calculations ( Bier et al . , 2013; Dutt et al . , 2013; Klärner and Schrader , 2013 ) . Importantly , CLR01 appears only to complex with readily accessible lysine or arginine residues on protein surfaces , as evidenced by crystal structures and NMR experiments ( Bier et al . , 2013 ) . This restriction makes CLR01 more selective for lysine or arginine residues found in intrinsically unfolded proteins or protein sequences . 10 . 7554/eLife . 05397 . 003Figure 1 . CLR01 binds to lysine and arginine residues . ( A ) Chemical structures of CLR01 and CLR03 . ( B ) Stick representation of the structures of CLR01 and CLR03 and their engagement of lysine side chains . ( C–E ) The primary sequences of PAP248-286 ( C ) , PAP85-120 ( D ) , and SEM1 ( 45-107 ) ( E ) are provided . Lysine and arginine residues are in red and hexapeptides predicted to form steric zippers ( Goldschmidt et al . , 2010; Castellano and Shorter , 2012 ) are underlined . ( F ) The average structures of the most populated clusters derived from the REMD simulations of PAP248-286 ( left ) , PAP248-286 with 7 CLR01 molecules ( middle ) , and PAP248-286 with 8 CLR03 molecules ( right ) are shown in the upper row , CLR01 and CLR03 molecules are not shown for clarity . The lower row shows , for each case , a representative structure of the most populated cluster including CLR01 and CLR03 . ( G ) CLR03 establishes only labile interactions with PAP248-286 as shown by the large X-P distances ( Å ) between one P atom of CLR03 and the nitrogen atom of the lysine side chain ( or carbon atom of the guanidinium moiety of arginine ) . Contrarily , the complexes between CLR01 and Lys or Arg were conserved during all the REMD simulations . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 003 Since amyloidogenic seminal peptides are particularly rich in lysine and arginine residues ( Roan et al . , 2009; Arnold et al . , 2012; Castellano and Shorter , 2012 ) ( Figure 1C–E , Lys and Arg residues are highlighted in red ) , we hypothesized that CLR01 might interfere with their HIV-enhancing activity . Here , we establish that CLR01 inhibits amyloidogenesis of PAP and SEM peptides , neutralizes the cationic surface charge of seminal amyloid , and rapidly remodels preformed SEVI and PAP85-120 fibrils . Strikingly , CLR01 also exhibits a direct antiviral effect by selectively disrupting the membrane of enveloped viruses . Thus , CLR01 represents an unprecedented candidate for further development as a microbicide as it not only inactivates HIV and other enveloped viruses but also antagonizes host-encoded seminal amyloids that enhance viral infection . Lysine residues in PAP248-286 , PAP85-120 , SEM1 , and SEM2 peptides are frequently found within or immediately adjacent to hexapeptides predicted to form self-complementary β-strands ( Figure 1C–E , underlined residues ) , termed steric zippers , which often comprise the spine of amyloid fibrils ( Nelson et al . , 2005; Goldschmidt et al . , 2010; Sievers et al . , 2011; Castellano and Shorter , 2012; Frohm et al . , 2015 ) . Moreover , the wealth of basic residues in PAP248-286 , PAP85-120 , and SEM1 ( 45-107 ) ( Figure 1C–E ) led us to hypothesize that the lysine- and arginine-specific tweezer , CLR01 , but not its derivative CLR03 , which lacks hydrophobic sidewalls ( Sinha et al . , 2011 ) ( Figure 1A , B ) , might bind to these residues and interfere with fibril assembly . To test this hypothesis , we first performed replica exchange molecular dynamics simulations using the available structure of PAP248-286 , the best characterized of the amyloid-forming peptides in semen ( Münch et al . , 2007; Castellano and Shorter , 2012; French and Makhatadze , 2012 ) . This analysis revealed that in silico , CLR01 bound at least seven of the eight positively charged residues in PAP248-286 without grossly altering peptide secondary structure ( Figure 1F ) . Indeed , CLR01 engaged Lys251 , Lys253 , Lys281 , and Lys282 ( Figure 1F , G ) , which all reside in predicted steric zippers ( Castellano and Shorter , 2012 ) ( Figure 1C ) . Moreover , CLR01 bound Arg257 , Lys281 , and Lys282 ( Figure 1F , G ) , which form part of the cross-β SEVI fibril core defined by hydrogen–deuterium exchange ( French and Makhatadze , 2012 ) . The CLR01 interaction was very similar among all the different lysine and arginine binding sites as indicated by similar ( X-P ) distances between the lysine ammonium or arginine guanidinium groups ( X ) and the bound phosphate group ( P ) in CLR01 ( Figure 1G ) . By contrast , CLR03 only established more distant and variable interactions with lysine and arginine residues ( Figure 1G ) . Thus , CLR01 but not CLR03 tightly shielded lysine and arginine residues in PAP248-286 suggesting that CLR01 could prevent PAP248-286 assembly into SEVI fibrils . To test this prediction , CLR01 was assessed for its ability to inhibit the spontaneous amyloidogenesis ( i . e . , assembly in the absence of preformed fibril seeds ) of PAP248-286 , PAP85-120 , and SEM1 ( 45-107 ) ( Münch et al . , 2007; Roan et al . , 2011; Arnold et al . , 2012 ) . Using the fluorescence of the diagnostic dye Thioflavin-T ( ThT ) , which increases upon amyloid binding , we found that CLR01 , but not CLR03 , inhibited fibril assembly of all three peptides ( Figure 2A–C ) . The half-maximal inhibitory concentrations ( IC50 ) of CLR01 inhibition of PAP248-286 and SEM1 ( 45-107 ) fibrillization were ∼1 . 19 µM and ∼16 . 2 µM , respectively ( Figure 2D , F ) . The IC50 for inhibition of PAP248-286 fibrillization was significantly lower than the IC50 for inhibition of SEM1 ( 45-107 ) fibrillization ( one-way ANOVA , p < 0 . 0001 ) , suggesting that lysine , arginine , or both residues are more critical for PAP248-286 fibrillization . Moreover , inhibition of PAP248-286 and SEM1 ( 45-107 ) fibrillization exhibited shallow dose–response curves with Hill slopes of ∼0 . 6 for PAP248-286 and ∼0 . 44 for SEM1 ( 45-107 ) ( Figure 2D , F ) , indicating negative co-operativity or CLR01 binding to multiple sites with different affinities . Importantly , these shallow dose responses provide evidence against a non-specific mechanism of inhibition involving colloidal CLR01 aggregates , as aggregating inhibitors tend to exhibit steep dose response curves ( Shoichet , 2006 ) . The IC50 of CLR01 to impede PAP85-120 assembly was significantly higher at ∼228 µM ( one-way ANOVA , p < 0 . 0001; Figure 2E ) . The higher IC50 is likely due to the lower number of lysine or arginine residues in the PAP85-120 sequence located in hexapeptides predicted to have high amyloid propensity ( Figure 1D ) ( Castellano and Shorter , 2012 ) . Indeed , PAP85-120 has 1 lysine or arginine located in predicted steric zippers , whereas PAP248-286 has 4 and SEM1 ( 45-107 ) has 3 ( Figure 1D ) . The Hill slope was ∼1 . 26 , indicating weak positive co-operativity ( Figure 2E ) , but was still below the ∼1 . 5–2 range that might indicate a mechanism of inhibition involving colloidal CLR01 aggregates ( Shoichet , 2006 ) . 10 . 7554/eLife . 05397 . 004Figure 2 . CLR01 inhibits formation of seminal amyloid fibrils . CLR01 inhibits fibril formation by PAP248-286 ( 1 mM ) ( A ) , PAP85-120 ( 1 mM ) ( B ) , and SEM1 ( 45-107 ) ( 500 µM ) ( C ) . Peptides were incubated with equimolar CLR01 , CLR03 , or buffer and agitated at 1400 rpm at 37°C . Aliquots were removed at various time points and fibrillization was assessed using the amyloid-binding dye , Thioflavin-T ( ThT ) . Values represent means ±SEM ( n = 4 for PAP248-286; n = 3 for PAP85-120; n = 9 for SEM1 ( 45-107 ) ) . Transmission electron microscopy ( TEM ) images of PAP248-286 ( A ) , PAP85-120 ( B ) , and SEM1 ( 45-107 ) ( C ) agitated in the presence of CLR01 , CLR03 , or buffer . PAP248-286 and SEM1 ( 45-107 ) samples were visualized after 72 hr of agitation , PAP85-120 after 24 hr . Scale bar: 500 nm . ( D ) Dose-response curve for CLR01 inhibition of PAP248-286 ( 1 mM ) fibrillization after 72 hr of agitation . The IC50 value and Hill slope are indicated . Values represent means ±SEM ( n = 3–12 ) . ( E ) Dose-response curve for CLR01 inhibition of PAP85-120 ( 1 mM ) fibrillization after 24 hr of agitation . The IC50 value and Hill slope are indicated . Values represent means ±SEM ( n = 3–7 ) . ( F ) Dose–response curve for CLR01 inhibition of SEM1 ( 45-107 ) ( 500 µM ) fibrillization after 72 hr of agitation . The IC50 value and Hill slope are indicated . Values represent means ±SEM ( n = 3–12 ) . ( G , H ) CLR01 is unable to block PAP248-286 ( Ala ) fibrillization . Lyophilized PAP248-286 ( Ala ) was dissolved in PBS ( 100 µM ) , incubated with CLR01 ( 100 µM ) , CLR03 ( 100 µM ) , or buffer and agitated at 1400 rpm at 37°C . Aliquots were removed at various time points and fibril assembly was assessed using ThT fluorescence ( G ) or sedimentation analysis ( H ) . Values represent means ±SEM ( n = 7–9 ) ( G ) or means ±SEM ( n = 3 ) ( H ) . ( I ) CLR01 is unable to block PAP85-120 ( Ala ) fibrillization . Lyophilized PAP85-120 ( Ala ) was dissolved in PBS ( 150 µM ) , incubated with CLR01 ( 150 µM ) , CLR03 ( 150 µM ) , or buffer and agitated at 1400 rpm at 37°C . At the indicated times , fibril assembly was assessed using ThT fluorescence . Values represent means ±SEM ( n = 3 ) . ( J ) Inhibition of seeded PAP248-286 fibrillization by CLR01 . Lyophilized PAP248-286 peptide was reconstituted at 1 mM in PBS and incubated with 1 mM CLR01 , 1 mM CLR03 , or buffer . Preformed SEVI fibrils ( 2% wt/wt ) were added to each condition and agitated at 1400 rpm at 37°C . Aliquots were removed at various time points and fibrillization was assessed using ThT fluorescence . Values represent means ±SEM ( n = 6 ) . ( K ) Electron microscopy visualization of CLR01 inhibition of seeded PAP248-286 fibrillization after 48 hr of agitation . Scale bar: 500 nm . ( L ) Dose-response curve for CLR01 inhibition of PAP248-286 fibrillization seeded by preformed SEVI fibrils ( 2% wt/wt ) after 48 hr of agitation . The IC50 value and Hill slope are indicated . Values represent means ±SEM ( n = 3–7 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 00410 . 7554/eLife . 05397 . 005Figure 2—figure supplement 1 . CLR01 does not displace ThT from SEVI , PAP85-120 , or SEM1 ( 45-107 ) fibrils . Preformed SEVI , PAP85-120 , or SEM1 ( 45-107 ) fibrils ( 5 µM monomer ) were preincubated with ThT ( 25 µM ) for 30 min at room temperature . Buffer ( PBS ) , CLR01 ( 250 µM ) , or a known competitor of ThT binding , BTA-1 ( 250 µM ) were then added and incubated for 10 min at room temperature . ThT displacement was then assessed by fluorescence measurements . Values represent means ±SEM ( n = 3 ) . A one-way ANOVA with the post hoc Dunnett's multiple comparisons test was used to compare the buffer control to the CLR01 or BTA-1 condition ( * denotes p < 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 00510 . 7554/eLife . 05397 . 006Figure 2—figure supplement 2 . CLR01 does not impede adsorption of SEVI , PAP85-120 , or SEM1 ( 45-107 ) fibrils to the EM grid . Preformed SEVI , PAP85-120 fibrils , or SEM1 ( 45-107 ) fibrils ( 20 µM ) were treated with a 10-fold excess of CLR01 or buffer for 10 min and processed for TEM . Note the presence of abundant fibrils under each condition indicating that CLR01 does not impede fibril adsorption to the EM grid . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 00610 . 7554/eLife . 05397 . 007Figure 2—figure supplement 3 . Lysine and poly-L-lysine antagonize the ability of CLR01 to inhibit spontaneous formation of seminal amyloid fibrils . ( A ) PAP248-286 ( 1 mM ) was incubated with buffer or CLR01 ( 100 µM ) in the presence or absence of lysine ( 20 mM ) or poly-L-lysine ( 1 mM ) and agitated at 1400 rpm at 37°C for 68 hr . Fibrillization was assessed via ThT fluorescence . Values represent means ±SEM ( n = 3 ) . A one-way ANOVA with the post hoc Dunnett's multiple comparisons test was used to compare the buffer alone control to the other conditions ( * denotes p < 0 . 0001 ) . ( B ) PAP248-286 ( 100 µM ) was incubated with CLR01 ( 100 µM ) or buffer in the presence or absence of lysine ( 20 mM ) or poly-L-lysine ( 1 mM ) and agitated at 1400 rpm at 37°C for 68 hr . Fibrillization was assessed via ThT fluorescence . Values represent means ±SEM ( n = 3 ) . A one-way ANOVA with the post hoc Dunnett's multiple comparisons test was used to compare the buffer alone control to the other conditions ( * denotes p < 0 . 0001 ) . ( C ) SEM1 ( 45-107 ) ( 0 . 5 mM ) was incubated with CLR01 ( 100 µM ) or buffer in the presence or absence of lysine ( 20 mM ) or poly-L-lysine ( 1 mM ) and agitated at 1400 rpm at 37°C for 68 hr . Fibrillization was assessed via ThT fluorescence . Values represent means ±SEM ( n = 3 ) . A one-way ANOVA with the post hoc Dunnett's multiple comparisons test was used to compare the buffer alone control to the other conditions ( * denotes p < 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 00710 . 7554/eLife . 05397 . 008Figure 2—figure supplement 4 . BSA or preclearing CLR01 has no effect on the ability of CLR01 to inhibit formation of seminal amyloid fibrils . PAP248-286 ( 1 mM ) , PAP85-120 ( 1 mM ) , or SEM1 ( 45-107 ) ( 500 µM ) was incubated with CLR01 ( 1 mM ) or buffer in the presence or absence of BSA ( 10 mg/ml ) and agitated at 1400 rpm at 37°C for 68 hr . In some reactions , CLR01 was first cleared by centrifugation at 16 , 100×g for 20 min at 25°C ( precleared ) . Fibrillization was assessed via ThT fluorescence . Values represent means ±SEM ( n = 3 ) . A one-way ANOVA with the post hoc Dunnett's multiple comparisons test was used to compare the buffer alone control to the other conditions ( * denotes p < 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 00810 . 7554/eLife . 05397 . 009Figure 2—figure supplement 5 . Lysine or poly-L-lysine , but not BSA or preclearing CLR01 solutions , antagonize the ability of CLR01 to inhibit seeded assembly of SEVI fibrils . PAP248-286 ( 1 mM ) was incubated with agitation at 37°C for 24 hr with preformed SEVI fibrils ( 2% wt/wt ) plus CLR01 ( 100 µM ) or buffer in the presence or absence of lysine ( 20 mM ) , poly-L-lysine ( 1 mM ) , or BSA ( 10 mg/ml ) . In some reactions , CLR01 was first cleared by centrifugation at 16 , 100×g for 20 min at 25°C ( precleared ) . Fibrillization was assessed via ThT fluorescence . Values represent means ±SEM ( n = 3 ) . A one-way ANOVA with the post hoc Dunnett's multiple comparisons test was used to compare the buffer alone control to the other conditions ( * denotes p < 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 009 We examined the possibility that CLR01 might simply displace ThT from fibrils by employing a ThT displacement assay ( Lockhart et al . , 2005 ) . Thus , preformed SEVI , PAP85-120 , or SEM1 ( 45-107 ) fibrils were preincubated with ThT . Buffer , or an excess of CLR01 or a known competitor of ThT binding , BTA-1 , were then added and ThT displacement was assessed by fluorescence measurements ( Lockhart et al . , 2005 ) . ThT fluorescence decreased drastically in the presence of BTA-1 , confirming its ability to displace ThT from fibrils ( Lockhart et al . , 2005 ) , whereas CLR01 and buffer had no effect ( Figure 2—figure supplement 1 ) . These findings suggest that CLR01 does not simply displace ThT from SEVI , PAP85-120 , or SEM1 ( 45-107 ) fibrils . Thus , any reduction in ThT fluorescence caused by CLR01 can be attributed to an inhibition of fibril assembly . Transmission electron microscopy ( TEM ) analysis confirmed that CLR01 prevented assembly of PAP248-286 and SEM1 ( 45-107 ) into mature fibrils ( Figure 2A , C ) . Only small oligomeric forms ( Figure 2A ) or amorphous aggregates in combination with sparse short fibrils ( Figure 2C ) were detectable in the presence of CLR01 , as opposed to abundant fibrils observed in the presence of buffer or CLR03 ( Figure 2A , C ) . The effect of CLR01 on PAP85-120 assembly was less apparent by TEM ( Figure 2B ) . However , the PAP85-120 assemblies formed in the presence of CLR01 appeared more flexible and curvilinear , and differed from the rigid , straight fibrils formed in the presence of CLR03 or buffer ( Figure 2B ) . Since the structures that formed in the presence of CLR01 were not ThT-reactive ( Figure 2B ) they most likely represent non-amyloid aggregates . Thus , CLR01 appears to impede the transition of PAP85-120 to mature amyloid fibrils and abrogates formation of SEVI and SEM1 ( 45-107 ) fibrils . We excluded the possibility that CLR01 might impede adsorption of SEVI , PAP85-120 , or SEM ( 45-107 ) fibrils to the EM grid . When we mixed CLR01 ( or buffer ) with preformed SEVI , PAP85-120 , or SEM1 ( 45-107 ) fibrils for 10 min ( a time at which no fibril remodeling occurs; Figure 3A–C ) and then adsorbed them to the grid , we observed abundant fibrils in both the CLR01 and buffer control conditions ( Figure 2—figure supplement 2 ) . Thus , any reduction in the presence of fibrils observed by EM can be attributed to inhibition of fibril assembly . 10 . 7554/eLife . 05397 . 010Figure 3 . CLR01 rapidly remodels SEVI and PAP85-120 fibrils . ( A–C ) Preformed SEVI fibrils ( A ) , PAP85-120 fibrils ( B ) , and SEM1 ( 45-107 ) fibrils ( C ) ( 20 µM ) were treated with a 10-fold excess of CLR01 or CLR03 or buffer for 0–6 hr at 37°C . Fibril integrity was assessed using ThT . Values represent means ±SEM ( n = 3 ) . TEM ( middle panel ) and confocal microscopy ( bottom panel ) of SEVI ( A ) , PAP85-120 fibrils ( B ) , and SEM1 ( 45-107 ) fibrils ( C ) obtained after 2 hr treatment with CLR01 , CLR03 , or buffer . Scale bar for TEM images: 500 nm . For confocal microcopy ( CoM ) , samples were stained with Proteostat dye . Scale bar: 20 µm . ( D ) CD spectrum of 50 µM SEVI fibrils incubated with 50 µM CLR01 , 50 µM CLR03 , or buffer . A representative spectrum is shown . ( E ) The mean residue ellipticity ( MRE ) at 218 nm was averaged from three independent experiments . Values represent means ±SEM . ( F ) Seeding with CLR01-remodeled SEVI products . SEVI fibrils ( 20 µM ) were treated with 200 µM CLR01 , 200 µM CLR03 , or buffer for 3 hr and reaction products were used to seed PAP248-286 fibrillization ( 0 . 1% fibril seed , 1 mM peptide ) . Buffer conditions with no fibril seed present were also included . Fibrillization was monitored by ThT fluorescence . Values represent means ±SEM ( n = 8 ) . ( G ) Dose–response curve for CLR01 remodeling of SEVI fibrils ( 20 µM ) after 2 hr of treatment . The EC50 value and Hill slope are indicated . Values represent means ±SEM ( n = 3–15 ) . ( H ) Dose-response curve for CLR01 remodeling of PAP85-120 fibrils ( 20 µM ) after 2 hr of treatment . The EC50 value and Hill slope are indicated . Values represent means ±SEM ( n = 3–14 ) . ( I ) CLR01 is unable to remodel SEVI ( Ala ) fibrils . Preformed SEVI ( Ala ) fibrils ( 20 µM ) were treated with 200 µM CLR01 , 200 µM CLR03 , or buffer for 0–6 hr . Fibril integrity was assessed using ThT fluorescence . Values represent means ±SEM ( n = 3 ) . ( J ) CD spectrum of 50 µM SEVI ( Ala ) fibrils incubated with either 50 µM CLR01 , 50 µM CLR03 , or buffer . One representative example is shown . ( K ) The mean residue ellipticity ( MRE ) at 218 nm was averaged from three independent experiments . Values represent means ±SEM ( n = 3 ) . ( L ) Preformed PAP85-120 ( Ala ) fibrils ( 20 μM ) were treated with CLR01 or CLR03 ( 200 μM ) or buffer for 0–6 hr . Fibril integrity was assessed using ThT fluorescence . Values represent means ±SEM ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 01010 . 7554/eLife . 05397 . 011Figure 3—figure supplement 1 . Lysine and poly-L-lysine antagonize the ability of CLR01 to remodel SEVI and PAP85-120 fibrils . SEVI ( left ) or PAP85-120 fibrils ( right , 20 µM monomer ) were treated with buffer or CLR01 ( 100 µM ) in the presence or absence of lysine ( 20 mM ) or poly-L-lysine ( 1 mM ) for 2 hr at 37°C . Fibril remodeling was then assessed by ThT fluorescence . Values represent means ±SEM ( n = 3–8 ) . A one-way ANOVA with the post hoc Dunnett's multiple comparisons test was used to compare the buffer alone control to the other conditions ( * denotes p < 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 01110 . 7554/eLife . 05397 . 012Figure 3—figure supplement 2 . BSA or preclearing CLR01 has no effect on the ability of CLR01 to remodel SEVI and PAP85-120 fibrils . SEVI ( left ) or PAP85-120 fibrils ( right , 20 µM monomer ) were treated with buffer or CLR01 ( 100 µM ) in the presence or absence of BSA ( 10 mg/ml ) for 2 hr at 37°C . In some reactions , CLR01 was first cleared by centrifugation at 16 , 100×g for 20 min at 25°C ( precleared ) . Fibril remodeling was then assessed by ThT fluorescence . Values represent means ±SEM ( n = 3 ) . A one-way ANOVA with the post hoc Dunnett's multiple comparisons test was used to compare the buffer alone control to the other conditions ( * denotes p < 0 . 0001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 012 To investigate the role of lysine and arginine interactions with CLR01 in the inhibition of fibril assembly , we examined PAP248-286 ( Ala ) or PAP85-120 ( Ala ) , two peptide analogues in which all lysine and arginine residues are replaced by alanine ( Roan et al . , 2009 ) . PAP248-286 ( Ala ) formed amyloid fibrils in the presence of CLR01 , CLR03 , and buffer ( Figure 2G ) . To further analyze the effect of CLR01 on PAP248-286 ( Ala ) assembly , we employed a sedimentation assay . This assay revealed that equal amounts of PAP248-286 ( Ala ) entered the pellet fraction when the peptide was incubated with buffer or CLR01 , indicating that the formation of PAP248-286 ( Ala ) amyloid fibrils is unaffected by CLR01 treatment ( Figure 2H ) . Thus , it is likely that in the presence of CLR01 , PAP248-286 ( Ala ) assembles into a subtly distinct set of cross-beta structures or fibril ‘strains’ that are less ThT-reactive ( Figure 2G , H ) . Amyloidogenesis by PAP85-120 ( Ala ) was also unaffected by CLR01 ( Figure 2I ) . The PAP248-286 ( Ala ) and PAP85-120 ( Ala ) peptides spontaneously assemble into amyloid more rapidly than the wild-type peptides ( Figure 2A , B , G , I ) . However , we have been unable to establish conditions ( e . g . , higher CLR01 concentrations ) where CLR01 prevents assembly of PAP248-286 ( Ala ) and PAP85-120 ( Ala ) into amyloid fibrils ( data not shown ) . These findings suggest that CLR01-lysine contacts , CLR01-arginine contacts , or both , are essential for inhibition of fibrillization . Next , we tested if an excess of free lysine or poly-L-lysine could interfere with the ability of CLR01 to block fibrillization of PAP248-286 , PAP85-120 , and SEM1 ( 45-107 ) ( Figure 2—figure supplement 3 ) . Both 20 mM lysine and 1 mM poly-L-lysine completely abrogated CLR01 inhibition of fibril formation by these three peptides . These results strongly support direct interaction of the tweezer with lysine , arginine , or both residues in PAP248-286 , PAP85-120 , and SEM1 ( 45-107 ) as the underlying mechanism of CLR01 inhibition . We suggest that these specific interactions preclude the conformational rearrangements necessary for amyloidogenesis . Some small molecules must form higher order colloidal aggregates to inhibit amyloid assembly ( Feng et al . , 2008; Young et al . , 2015 ) . BSA inhibits the activity of colloidal small-molecule aggregates via adsorption ( McGovern et al . , 2002; Coan and Shoichet , 2007; Feng et al . , 2008 ) . Additionally , colloidal small-molecule aggregates can be precleared via centrifugation ( McGovern et al . , 2003 ) . Thus , we used these two techniques to test whether inhibition of PAP248-286 , PAP85-120 , or SEM1 ( 45-107 ) aggregation could be mediated via unexpected colloid formation by CLR01 . CLR01 was found to inhibit PAP248-286 , PAP85-120 , and SEM1 ( 45-107 ) fibril assembly in the presence of BSA ( 10 mg/ml ) or when solutions containing CLR01 were centrifuged at 16 , 100×g for 20 min before adding the supernatant solution to the proteins ( Figure 2—figure supplement 4 ) . Thus , CLR01 is not inhibiting PAP248-286 , PAP85-120 , and SEM1 ( 45-107 ) fibril assembly via a mechanism that involves colloidal CLR01 aggregates . The addition of a small amount of preformed SEVI fibrils to soluble PAP248-286 seeds polymerization and eliminates the lag phase for assembly ( Ye et al . , 2009 ) ( compare Figure 2A with Figure 2J ) . Remarkably , in addition to obstructing unseeded PAP248-286 assembly ( Figure 2A ) , CLR01 also completely inhibited seeded fibrillization ( Figure 2J , K ) . Dose–response analysis established an IC50 of ∼15 . 3 µM for CLR01 inhibition of seeded PAP248-286 assembly ( Figure 2L ) compared to ∼1 . 19 µM for spontaneous PAP248-286 assembly ( Figure 2D ) . Thus , higher concentrations of CLR01 are required to inhibit PAP248-286 assembly once SEVI fibrils have formed . Interestingly , the Hill slope for inhibition of seeded PAP248-286 assembly was ∼3 . 1 indicating positive co-operativity and was steeper than the Hill slope of ∼0 . 6 for inhibition of spontaneous PAP248-286 fibrillization ( Figure 2D , L ) . This dissimilarity might indicate a mechanistic difference in how CLR01 inhibits seeded and spontaneous PAP248-286 fibrillization . However , this difference is not likely to be due colloidal CLR01 aggregates , as inhibition of seeded assembly by CLR01 was unaffected by BSA or by preclearing CLR01 solutions via centrifugation ( Figure 2—figure supplement 5 ) . Rather , we suggest that the CLR01-mediated inhibition of fibril elongation in seeded SEVI assembly has requirements distinct from the CLR01-mediated inhibition of initial fibril nucleation in spontaneous , unseeded SEVI assembly . Inhibition of seeded PAP248-286 assembly was alleviated by excess lysine or poly-L-lysine ( Figure 2—figure supplement 5 ) . Thus , inhibition of seeded PAP248-286 assembly by CLR01 likely requires specific interaction with lysine , arginine , or both residues . Importantly , because CLR01 impedes both unseeded and seeded PAP248-286 assembly , it likely acts at multiple stages of the fibrillization process including the initial nucleation and fibril elongation steps . These observations also indicate that lysine , arginine , or both residues play an important role in the primary nucleation and subsequent elongation of PAP248–286 fibrils . Semen-derived fibrils are abundant in liquefied fresh ejaculates ( Usmani et al . , 2014 ) . Thus , agents that not only inhibit fibril formation but also remodel preformed fibrils would be advantageous for microbicide development ( Castellano and Shorter , 2012 ) . To test whether CLR01 could remodel seminal amyloid , SEVI , PAP85-120 and SEM1 ( 45-107 ) fibrils were treated with CLR01 or CLR03 , and ThT fluorescence intensity was monitored . Brief incubations with CLR01 ( 10–20 min ) had no effect , indicating that CLR01 did not simply displace ThT from fibrils ( Figure 2—figure supplement 1; Figure 2—figure supplement 2; Figure 3A–C ) . After 2 hr , however , CLR01 treatment of SEVI ( Figure 3A ) and PAP85-120 ( Figure 3B ) but not SEM1 ( 45-107 ) ( Figure 3C ) fibrils resulted in a reduction in ThT fluorescence intensity by more than 50% . Even after 24 hr , SEM1 ( 45-107 ) fibrils were not remodeled by CLR01 ( data not shown ) . SEVI and PAP85-120 fibril remodeling was confirmed by TEM , which showed few intact fibrils and predominately smaller nonfibrillar species after CLR01 treatment ( Figure 3A , B ) . By contrast , TEM revealed that SEM1 ( 45-107 ) fibrils were still abundant after prolonged CLR01 treatment ( Figure 3C ) . This observation confirms that CLR01 does not simply prevent fibrils from adsorbing to the EM grid ( Figure 2—figure supplement 2 ) . Similar results were obtained by fluorescence microscopy ( CoM ) of samples stained with Proteostat , a red fluorescent aggregate sensing dye ( Figure 3A–C ) ( Usmani et al . , 2014 ) . These effects of CLR01 were remarkably rapid in comparison to the slow disassembly of Aβ or α-synuclein fibrils by CLR01 , which required several weeks ( Sinha et al . , 2011; Prabhudesai et al . , 2012 ) . Thus , CLR01 can remodel SEVI and PAP85-120 fibrils on a time scale that would be useful for prevention of HIV infection , which usually takes place within the first hours after deposition of semen in the anogenital tract ( Shattock and Moore , 2003 ) . We next performed circular dichroism experiments to examine the effect of CLR01 on the amyloid cross-β structure . Untreated SEVI fibrils or SEVI fibrils treated with CLR03 exhibited a pronounced minimum indicative of a characteristic β-sheet rich structure ( Figure 3D , E ) . By contrast , SEVI fibrils incubated with CLR01 showed a loss in the β-sheet minimum ( Figure 3D , E ) , confirming that CLR01 alters the cross-β architecture of SEVI fibrils . Furthermore , SEVI fibrils remodeled by CLR01 were less effective seeds for polymerization of monomeric PAP248-286 ( Figure 3F ) . Thus , CLR01 remodels SEVI fibrils into alternative non-templating conformers . A dose–response curve for CLR01-mediated remodeling of SEVI and PAP85-120 fibrils revealed a half-maximal effective concentration ( EC50 ) value of ∼0 . 89 µM and ∼1 . 41 µM , respectively ( Figure 3G , H ) . Interestingly , in both cases the Hill slope was similar , approximately −0 . 6 for SEVI and approimately −0 . 726 for PAP85-120 , indicating negative co-operativity or CLR01 binding to multiple sites with different affinities . These shallow dose–response curves provide evidence against a colloidal mechanism of CLR01 action ( Shoichet , 2006 ) . Previous studies have reported that a 10-fold molar excess of CLR01 could remodel preformed Aβ40 , Aβ42 , or α-synuclein fibrils slowly over a time period of weeks ( Sinha et al . , 2011; Prabhudesai et al . , 2012 ) . It is therefore remarkable that CLR01 remodels SEVI and PAP85-120 fibrils with such alacrity and at sub-stoichiometric concentrations ( relative to peptide monomers ) . To elucidate the role of lysine- and arginine-tweezer interactions in CLR01-mediated remodeling of SEVI and PAP85-120 fibrils , experiments were performed with fibrils comprised of PAP248-286 ( Ala ) or PAP85-120 ( Ala ) . CLR01 was unable to remodel preformed SEVI ( Ala ) or PAP85-120 ( Ala ) fibrils ( Figure 3I–L ) . Indeed , we have been unable to establish conditions ( e . g . , higher CLR01 concentrations ) where CLR01 remodels SEVI ( Ala ) or PAP85-120 ( Ala ) fibrils ( data not shown ) . Furthermore , addition of excess free lysine or poly-L-lysine to CLR01 prevented its SEVI and PAP85-120 amyloid-remodeling activity ( Figure 3—figure supplement 1 ) . Thus , CLR01-lysine , CLR01-arginine , or both contacts are critical for the remodeling process . By contrast , CLR01 retained SEVI and PAP85-120 amyloid-remodeling activity in the presence of BSA ( 10 mg/ml ) or if CLR01 solutions were precleared via centrifugation ( Figure 3—figure supplement 2 ) . Thus , colloidal CLR01 aggregates do not contribute to the amyloid-remodeling activity of CLR01 . Taken together , our findings suggest that the lysine- and arginine-specific molecular tweezer CLR01 prevents formation of seminal amyloid and remodels mature PAP85-120 and SEVI fibrils via binding to lysine and arginine residues . Since the infectivity-enhancing activity of seminal amyloid fibrils is due to their positive surface charge ( Roan et al . , 2009; Arnold et al . , 2012 ) , we examined next whether CLR01 affected this property . To prevent fibril remodeling from occurring in these experiments , fibrils were mixed with CLR01 , CLR03 , or buffer , and samples were immediately centrifuged to remove unbound CLR01 and CLR03 molecules . Subsequently , the surface charge of resuspended fibrils was determined by zeta potential measurements . We found that CLR01 , but not CLR03 , neutralized the positive surface charge of SEVI , PAP85-120 , and SEM1 ( 49-107 ) fibrils within minutes ( Figure 4A ) . Notably , pre-treatment of CLR01 with lysine or poly-L-lysine abrogated the fibril neutralizing activity of CLR01 ( Figure 4—figure supplement 1 ) , which further supports previous findings concerning the specificity underlying these interactions ( Fokkens et al . , 2005; Klärner et al . , 2006 , 2010 ) . Next , confocal microscopy was used to assess whether this neutralization could abrogate fibril binding to YFP-tagged virions . As previously shown ( Roan et al . , 2011; Arnold et al . , 2012; Yolamanova et al . , 2013 ) , buffer-treated fibrils efficiently sequestered virions ( Figure 4B ) . In contrast , fibrils pretreated with CLR01 , but not CLR03 , were unable to form fibril–virion complexes ( Figure 4B ) . 10 . 7554/eLife . 05397 . 013Figure 4 . CLR01 neutralizes the positive surface charge of seminal amyloids and abrogates their ability to bind virions and enhance HIV infection . ( A ) Surface charge of seminal amyloids determined by zeta potential measurements . Fibrils were mixed with buffer , CLR01 , or CLR03 and centrifuged for 10 min at 20 , 000×g . The pellets were resuspended in 1 mM KCl and zeta potential was measured . Values represent means ±SD ( n = 3 ) . ( B ) Fibrils ( 200 µg/ml ) were pretreated with PBS , CLR01 , or CLR03 in 20-fold excess for 5 min and stained with Proteostat dye . MLV-Gag-YFP particles ( green ) were added 1:2 and allowed to incubate with pretreated fibrils ( red ) for 5 min . Samples were analyzed using confocal microscopy . Scale bar: 5 µm . ( C ) CLR01 displaces virions from fibrils . MLV-Gag-YFP particles ( green ) were incubated with Proteostat-stained seminal amyloids ( red ) for 5 min . PBS , CLR01 , or CLR03 were added at 20-fold excess , and after an additional 5 min of incubation , samples were analyzed by confocal microscopy . Scale bar: 5 µm . ( D ) CLR01 does not quench the fluorescence signal of MLV-Gag-YFP particles . MLV-Gag-YFP particles were treated with 300 µM CLR01 or CLR03 for 30 min at 37°C . Thereafter , virions were recovered in the pellet fraction via high-speed centrifugation and aliquots analyzed by confocal microscopy . Scale bar = 20 µm . ( E ) CLR01 antagonizes the HIV-1 enhancing activity of seminal amyloids . SEVI , PAP85-120 , and SEM1 ( 49-107 ) fibrils were mixed with a 20-fold excess of CLR01 or CLR03 . CCR5-tropic HIV-1 was added and samples were used to inoculate TZM-bl cells . Infection rates were determined 3 days post infection . Values represent mean β-galactosidase activities derived from triplicate infections ±SD ( RLU/s: relative light units per second ) . Numbers above symbols indicate the n-fold enhancement of infection . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 01310 . 7554/eLife . 05397 . 014Figure 4—figure supplement 1 . Lysine and poly-L-lysine dose-dependently antagonize CLR01 binding to SEVI . SEVI fibrils were mixed with CLR01 and increasing amounts of lysine ( 1 , 10 and 100-fold excess over CLR01 ) or poly-L-lysine ( 0 . 02 , 0 . 25 and 2 . 5-fold excess of lysine monomer equivalents over CLR01 ) and centrifuged for 10 min at 20 , 000×g . The pellets were resuspended in 1 mM KCl and zeta potential was measured . Values represent means ±SD ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 01410 . 7554/eLife . 05397 . 015Figure 4—figure supplement 2 . CLR01 does not quench the fluorescence signal of MLV-Gag-YFP particles . MLV-Gag-YFP particles were treated with 300 µM CLR01 or CLR03 for 30 min at 37°C . Thereafter , virions were recovered in the pellet fraction via high-speed centrifugation and aliquots analyzed by confocal microscopy . Z-stack images of the samples revealed a similar amount of fluorescent particles after treatment with CLR01 or CLR03 but altered distribution in the slides . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 01510 . 7554/eLife . 05397 . 016Figure 4—figure supplement 3 . CLR01 is not cytotoxic . TZM-bl cells were incubated for 3 days with the indicated concentrations of ( A ) CLR01 and CLR03 and ( B ) CLR01 , nonoxynol 9 ( N9 ) , sodium dodecyl sulfate ( SDS ) and Triton X-100 ( TX-100 ) . Cell viability was measured in an MTT assay . Values represent means ±SD ( n = 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 016 Since fibril–virion complexes are already present before ejaculation or form rapidly post emission ( Usmani et al . , 2014 ) , we next investigated whether CLR01 could displace virions from preformed fibril–virion complexes . Remarkably , CLR01 but not CLR03 substantially reduced the number of virions covering the surface of SEVI , PAP85-120 , and SEM1 ( 49-107 ) fibrils ( Figure 4C ) . We excluded the possibility that CLR01 quenches the fluorescence of YFP-tagged virions by analyzing virions treated with CLR01 or controls by confocal microscopy . We found that treatment of virions with CLR01 did not affect virion fluorescence ( Figure 4D ) or the number of fluorescent particles ( Figure 4—figure supplement 2 ) . However , CLR01-treated virions were dispersed throughout the chamber indicating that their biophysical properties might be altered ( Figure 4—figure supplement 2 ) . Our results demonstrate that CLR01 not only prevents the interaction of fibrils with virions but also displaces viral particles from preformed fibril–virion complexes . The combined effects of CLR01 on fibril architecture ( Figures 2 , 3 ) and the formation of fibril–virion complexes ( Figure 4A–C ) led us to investigate whether the tweezer might diminish the infection-enhancing property of seminal amyloids . First , we analyzed possible cytotoxic effects of the tweezer in TZM-bl cells , a HIV reporter cell line commonly used to study virus infection ( Münch et al . , 2007; Usmani et al . , 2014 ) . We found that CLR01 and CLR03 did not cause cytotoxic effects at concentrations up to 500 µM , whereas the surfactants Triton X-100 , nonoxynol 9 and sodium dodecyl sulfate ( SDS ) were highly toxic ( Figure 4—figure supplement 3 ) . Thus , CLR01 does not act in a manner similar to non-ionic or anionic surfactants . Next , preformed fibrils were treated for 5 min with CLR01 , CLR03 , or buffer . After addition of a low dose of CCR5-tropic HIV , the resulting mixture was added to TZM-bl cells and infection was measured via β-galactosidase activity 3 days later . As previously reported ( Münch et al . , 2007; Roan et al . , 2011; Arnold et al . , 2012 ) , the three semen-derived amyloids augmented HIV infection in a dose-dependent manner with maximal enhancements between 33- to 64-fold ( Figure 4E ) . Remarkably , pretreatment of SEVI , PAP85-120 , and SEM1 ( 49-107 ) amyloid with CLR01 eliminated the infection-enhancing property of the fibrils , while CLR03 had no effect ( Figure 4E ) . Thus , CLR01 abrogates the infection enhancing activity of seminal amyloids . In addition to engaging amyloid fibrils ( Figures 3 , 4A ) , CLR01 also appeared to have a direct effect on viral particles ( Figure 4—figure supplement 2 ) . To study the consequences of this interaction in more detail , a CXCR4- and CCR5-tropic HIV-1 NL4-3 recombinant ( Papkalla et al . , 2002 ) and two CCR5-tropic transmitted/founder viruses ( Ochsenbauer et al . , 2012 ) were pretreated with CLR01 or controls and then examined for infectivity . Remarkably , CLR01 but not CLR03 abrogated viral infectivity in a dose-dependent manner with IC50 values ranging from ∼13 . 7 to 20 . 1 µM ( Figure 5A ) . This reduction in infectivity was only observed when virions were exposed to CLR01 and not when target cells were pretreated with the tweezer ( Figure 5B ) . Thus , in addition to its anti-amyloid properties , CLR01 also has direct anti-HIV activity that is independent of the viral co-receptor tropism or strain . 10 . 7554/eLife . 05397 . 017Figure 5 . CLR01 has direct anti-HIV-1 activity . ( A ) CLR01 blocks HIV-1 infection by targeting virions . The CXCR4 ( X4 ) -tropic lab-adapted HIV-1 NL4-3 strain , a CCR5 ( R5 ) -tropic V3 loop recombinant thereof , and two R5-tropic transmitted/founder viruses ( THRO and CH058 ) were incubated with CLR01 or CLR03 for 10 min and then used to infect TZM-bl cells . Infection rates were determined 3 days post infection . Values represent mean β-galactosidase activities derived from triplicate infections ±SD ( RLU/s: relative light units per second ) . ( B ) The antiviral activity of CLR01 is not directed against the cell . TZM-bl cells were exposed to CLR01 or CLR03 for 1 hr . Thereafter , cell culture medium was removed and cells were infected with the indicated viruses . Infection rates were determined 3 day post infection . Values represent mean β-galactosidase activities derived from triplicate infections ±SD ( RLU/s: relative light units per second ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 01710 . 7554/eLife . 05397 . 018Figure 5—figure supplement 1 . Poly-L-lysine counteracts the antiviral activity of CLR01 . CLR01 ( 40 µM ) was titrated with poly-L-lysine . After a 10 min incubation with HIV-1 these mixtures were used to infect TZM-bl cells . Infection rates were determined 3 days post infection via a β-galactosidase assay . Values represent normalized mean infection rates ±SEM . Unpaired t-tests were used to compare the buffer control to the CLR01 condition at each poly-L-lysine concentration ( ns denotes not significant; * denotes p < 0 . 05; denotes p < 0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 01810 . 7554/eLife . 05397 . 019Figure 5—figure supplement 2 . Coated BSA does not antagonize the antiviral activity of CLR01 . Microtiter plates were coated over night with the indicated amounts of BSA . After a 20-min incubation at room temperature , serial dilutions of CLR01 were added and incubated with HIV-1 for another 10 min at room temperature . With these mixtures , TZM-bl cells were infected . Infection rates were determined 3 days post infection . ( A ) Values represent mean β-galactosidase activities derived from triplicate infections ±SD ( RLU/s: relative light units per second ) . ( B ) Values represent normalized mean infection rates derived from triplicate infections ±SD . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 01910 . 7554/eLife . 05397 . 020Figure 5—figure supplement 3 . BSA does not antagonize the antiviral activity of CLR01 . CLR01 ( 220 µM ) was titrated with the indicated concentrations of BSA . HIV-1 was incubated with these mixtures for 10 min and then used to infect TZM-bl cells . Infection rates were determined 3 days post infection via a β-galactosidase assay . Values represent normalized mean infection rates derived from triplicate measurements ±SD . Unpaired t-tests were used to compare the buffer control to the CLR01 condition at each BSA concentration ( denotes p < 0 . 01; * denotes p < 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 02010 . 7554/eLife . 05397 . 021Figure 5—figure supplement 4 . Preclearing CLR01 via centrifugation does not affect antiviral activity . CLR01 ( 50 µM ) was centrifuged at 20 , 000×g and 20°C for 20 min . HIV-1 92TH014 was incubated for 10 min with the supernatant ( S ) or the pellet fraction ( P ) . As a control , virus was also incubated either with PBS ( Buffer ) or 50 µM CLR01 that was not centrifuged ( NC ) . After infection of TZM-bl cells with these mixtures infection rates were measured 2 days post infection . Values represent mean β-galactosidase activities derived from triplicate infections ±SD ( RLU/s: relative light units per second ) . Unpaired t-tests were used to compare the buffer control to each CLR01 condition at each poly-L-lysine concentration ( ns denotes not significant; denotes p < 0 . 01; * denotes p < 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 021 Exposure of CLR01 to poly-L-lysine prior to incubation with virions abrogated its anti-HIV-1 activity in a dose-dependent manner ( Figure 5—figure supplement 1 ) . Near-complete inhibition was achieved with 75 nM poly-L-lysine . Thus , the lysine-binding cavity of CLR01 plays an important role in the antiviral ( Figure 5; Figure 5—figure supplement 1 ) and anti-amyloid ( Figure 2; Figure 2—figure supplements 3 , 5; Figure 3; Figure 3—figure supplement 1 ) activities of CLR01 . Incubation of the tweezer in solutions containing up to 5% ( sevenfold molar excess ) coated or free BSA did not reduce the anti-HIV activity of CLR01 ( Figure 5—figure supplements 2 , 3 ) . Similarly , the antiviral activity of CLR01 was not reduced when CLR01 solutions were precleared of any aggregates via centrifugation ( Figure 5—figure supplement 4 ) . Moreover , the pellet fraction did not possess antiviral activity ( Figure 5—figure supplement 4 ) . These data strongly suggest that colloidal CLR01 aggregates do not contribute to the direct antiviral activity of CLR01 . To define the underlying mechanism of this antiviral activity , we tested whether CLR01 disrupts the integrity of the viral membrane , leading to the release of the inner viral p24 capsid protein . HIV virions were exposed to buffer , CLR01 or CLR03 and then separated by centrifugation into a soluble fraction ( containing free p24 ) and a sedimentable fraction ( containing intact viral particles ) . ELISA measurements demonstrated that the amount of p24 was increased in the soluble fraction of CLR01-treated samples as compared to samples treated with CLR03 or buffer ( Figure 6A ) . Time course experiments revealed that a 5-min incubation of virus with 10 µM CLR01 resulted in a 62% decrease in HIV infectivity , and a 10-min incubation achieved almost a 100% reduction ( Figure 6B ) . Atomic force microscopy ( AFM ) of mouse leukemia virus particles ( MLV ) confirmed that treatment with CLR01 destroyed virion architecture ( Figure 6C , D ) . This effect was independent of the presence of viral glycoproteins , since CLR01 also destroyed HIV-1 particles lacking gp120/41 ( ∆env ) ( Figure 6C ) , suggesting that CLR01 disrupts the integrity of the viral membrane . 10 . 7554/eLife . 05397 . 022Figure 6 . CLR01 destroys retroviral particles and selectively disrupts raft-rich membranes . ( A ) CLR01 releases p24 capsid antigen from HIV particles . HIV-1 was incubated with PBS , 100 µM CLR03 , or 100 µM CLR01 and centrifuged at 20 , 000×g and 4°C for 1 hr . The p24 content of the supernatant was determined via p24 ELISA . Values represent means ±SD . Unpaired t-tests were used to compare the buffer control to the CLR03 or CLR01 condition ( ns denotes not significant; denotes p < 0 . 01 ) . ( B ) HIV-1 was incubated at 37°C with 10 µM CLR01 or buffer control . Aliquots were taken after different time points and analyzed regarding their infectivity using TZM-bl reporter cells . Values represent normalized mean infection rates derived from triplicate measurements ±SD compared to the buffer control ( 100% ) . Unpaired t-tests were used to compare the buffer control to the CLR01 condition at each time point ( *** denotes p < 0 . 001 ) . ( C ) CLR01 destroys retroviral particles . Images obtained by atomic force microscopy ( AFM ) show single MLV and glycoprotein-deficient HIV particles before and after treatment with 100 µM CLR01 . Scale bar: 100 nm . ( D ) CLR01 destroys MLV particles . Height distribution of MLV particles after treatment with buffer ( left panel ) or 100 µM CLR01 ( right panel ) . Values were derived from AFM images shown in the insets . Scale bar: 2 µm . ( E ) CLR01 selectively destroys membranes with high lipid raft content . Giant unilamellar vesicles ( GUVs ) consisting of pure DOPC were labeled with N-Rh-DHPE ( red ) . GUVs containing a mixture of DOPC , SM and Chol ( 45/25/30 mol% ) were labeled with N-Rh-DHPE ( red ) and Bodipy-Chol ( green ) . Both types of GUVs were filled with buffer containing the fluorophore ATTO 647 ( blue ) and treated with 150 µM CLR01 for the indicated times before images were taken by confocal microscopy . Note that ATTO 647 remains inside the DOPC GUVs treated with CLR01 , but escapes the DOPC/SM/Chol GUVs treated with CLR01 . Scale bar: 10 μm . ( F ) Upper panel: AFM images ( 10 µm scans ) of a pure DOPC lipid membrane on mica before injection ( 0 min ) and 1 min and 60 min after injection of 800 µl of 150 µM CLR01 in 10 mM NaH2PO4 , pH 7 . 6 into the AFM fluid cell . The whole scan area is shown with a vertical color scale from dark brown to white corresponding to an overall height of 8 nm . The thickness of the hydrated membrane is 3 . 7 nm . Lower panel: AFM image ( 10 µm scan ) of a DOPC/SM/Chol ( 45/25/30 mol% ) lipid membrane on mica before injection ( 0 min ) and 1 min and 60 min after injection of 800 µl of 150 µM CLR01 in 10 mM NaH2PO4 , pH 7 . 6 into the AFM fluid cell . The whole scan area is shown with a vertical color scale from dark brown to white corresponding to an overall height of 8 nm and indicating a homogeneous lipid bilayer with coexisting domains in lo ( liquid-ordered ) and ld ( liquid-disordered domain ) phase . The height difference between domains is 1 nm; the ld phase has a thickness of 4 . 0 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 022 The result that CLR01 destroys retroviral particles with IC50 values between ∼10 and 20 µM by compromising virion integrity was unexpected , particularly because the tweezer does not affect cell viability at these concentrations ( Figure 4—figure supplement 3 ) ( Sinha et al . , 2011; Attar et al . , 2012; Herzog et al . , 2015 ) . Viral membranes differ from cellular membranes in that they are two to threefold enriched in specific lipids , such as sphingomyelin and cholesterol , which can form microdomains termed lipid rafts ( Aloia et al . , 1988; Brügger et al . , 2006; Chan et al . , 2008; Lorizate et al . , 2009 , 2013; Gerl et al . , 2012 ) . We hypothesized that CLR01 might selectively disrupt lipid-raft enriched membranes . To test this hypothesis , dye-loaded giant unilamellar vesicles ( GUV ) were generated that consisted of either 1 , 2-dioleoyl-sn-glycero-3-phosphocholine ( DOPC ) to recapitulate a bilayer devoid of lipid rafts or a mixture of DOPC , sphingomyelin and cholesterol ( DOPC/SM/Chol ) to more closely resemble a lipid raft-rich viral membrane . Strikingly , CLR01 permeabilized the model viral membrane within 5–10 min while exhibiting no effect on the DOPC membrane , even after 60 min of incubation ( Figure 6E ) . These results were confirmed by AFM at higher spatial resolution ( Figure 6F ) . CLR01 treatment of a DOPC membrane that was deposited on mica surface did not affect bilayer stability , and the tweezer ( white dots in Figure 6F ) was homogenously distributed on the scan area . By contrast , the DOPC/SM/Chol raft mixture appeared as coexisting liquid-disordered ( ld ) and liquid-ordered ( lo ) domains , and CLR01 addition induced changes in phase coexistence and decreased height differences between both phases . After 60 min , distinct ld and lo domains were no longer visible , and CLR01 was homogenously distributed in the remaining fluid phase . Collectively , these data suggest that CLR01 selectively disrupts heterogeneous lipid-raft enriched membranes . Inhibitors of amyloid formation may block fibril polymerization by forming colloidal aggregates that sequester peptide via non-specific interactions ( Feng et al . , 2008 ) . Likewise , small molecule aggregates might also interfere non-specifically with viral infection . However , neither inhibition of amyloidogenesis , amyloid remodeling nor inhibition of viral infection by CLR01 were affected by BSA , which would quench small-molecule aggregates ( McGovern et al . , 2002 ) , or by clearance of any potential CLR01 aggregates by centrifugation ( McGovern et al . , 2003 ) ( Figure 2—figure supplement 4; Figure 3—figure supplement 2; Figure 5—figure supplements 2 , 3 , 4 ) . To investigate whether CLR01 even forms colloidal aggregates , we performed various tests for CLR01 aggregation using the same buffers and CLR01 concentrations as employed in our biological experiments described above ( Figures 2A–C , 3A–C , 5A ) . NMR dilution titrations provided experimental evidence for a very weak CLR01 dimer formation ( Ka = ∼60 M−1 ) in PBS buffer ( 10 mM sodium phosphate , 137 mM NaCl , 2 . 7 mM KCl , pH 7 . 4 ) , with ∼5% CLR01 dimers present when the CLR01 concentration was 500 µM ( Figure 7A , Figure 7—figure supplement 1 ) . This observation is consistent with previously published data ( Dutt et al . , 2013 ) . Dimerization was totally absent in HEPES buffer ( 25 mM HEPES , 150 mM KOAc , 10 mM Mg ( OAc ) 2 , pH 7 . 4; Figure 7B , Figure 7—figure supplement 2 ) . Thus , CLR01 is predominantly monomeric . 10 . 7554/eLife . 05397 . 023Figure 7 . CLR01 does not form colloid or micelle aggregates . ( A ) 1H NMR dilution titration of CLR01 from 6 mM to 100 μM in PBS leads to very weak dimerization; ( B ) in HEPES buffer CLR01 remains monomeric over the entire concentration range . ( C ) DOSY spectrum of 6 mM CLR01 in PBS buffer; cross peaks at 4 . 8 ppm correspond to aqueous solvent . ( D ) DOSY spectrum of 2 mM CLR01 in HEPES buffer; cross peaks between 2 and 4 ppm correspond to HEPES buffer , cross peaks at 4 . 8 ppm correspond to the aqueous solvent . ( E ) Attempted cmc determination by analysis of the I392/I373 emission intensity ratio of 10 μM PBS-buffered pyrene solutions containing CLR01 from 30 nM to 500 μM produces a straight line , as opposed to the SDS control ( data not shown ) . ( F ) DLS measurements of CLR01 ( 1 mM ) in PBS or HEPES buffer show formation of minute amounts of particles with RH ∼5–8 nm or ∼20–40 nm in PBS . No particles were detected in HEPES buffer . The vast majority of CLR01 molecules are too small to be detected and are presumably monomeric . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 02310 . 7554/eLife . 05397 . 024Figure 7—figure supplement 1 . Aromatic proton shifts during dilution of PBS-buffered CLR01 . Upfield shifts of the inner aromatic protons ( 2-H , 3-H , 13-H , 14-H ) document their mutual inclusion inside the cavity of the other host molecule , leading to specific anisotropic shielding exclusively in the dimer . Thus , the NMR shift changes prove formation of weak dimers ( Ka ∼60 M−1 ) and exclude unspecific aggregation , which would manifest as smaller chemical shift changes evenly distributed among the CH protons of CLR01 . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 02410 . 7554/eLife . 05397 . 025Figure 7—figure supplement 2 . Aromatic proton shifts during dilution of HEPES-buffered CLR01 . Dilution titration of CLR01 in HEPES buffer did not produce any upfield shift of the aromatic protons; on the contrary , the monomer spectrum appeared at all concentrations , with sharp signals for all three protons . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 02510 . 7554/eLife . 05397 . 026Figure 7—figure supplement 3 . Stejskal and Tanner Plot . Signal intensity in diffusion NMR experiments obeys the following law: ln ( I/I0 ) = −γ2δ2G2 ( ∆ − δ/3 ) D , where I = intensity at a given G , I0 = Intensity at G = 0 , γ = Gyromagnetic ratio ( for 1H: γ = 2 . 675 × 108T−1s−1 ) , δ = Length of diffusion gradient , ∆ = Diffusion shift , G = Gradient field intensity , and D = Diffusion coefficient . Signal intensities are used to generate a Stejskal Tanner-Plot lg ( I/I0 ) vs G2 . Its slope yields the diffusion coefficient . Diffusion coefficient for CLR01 ( 6 . 4 mM in PBS ) : 2 . 433 × 10−10 m2s−1 . By means of the Stokes Einstein equation one calculates the hydrodynamic radius: r ( s ) = ( kT ) / ( 6piηD ) . The hydrodynamic radius for CLR01 ( 6 . 4 mM ) was determined at 1 . 0075 × 10−9 m ∼10 Å in PBS buffer; for CLR01 ( 0 . 2 mM ) the diffusion coefficient dropped to 2 . 702 × 10−10 m2s−1 , corresponding to a hydrodynamic radius of 9 . 072 × 10−9 m ∼9 Å . In HEPES buffer , the diffusion coefficient for CLR01 ( 2 . 0 mM ) was determined at 2 . 449 × 10−10 m2s−1 , corresponding to hydrodynamic radius r ( s ) of 1 . 001 × 10−9 m ∼10 Å; for CLR01 ( 0 . 5 mM ) the diffusion coefficient dropped to 2 . 579 × 10−10 m2s−1 , corresponding to a hydrodynamic radius of 9 . 504 × 10−9 m ∼9 Å . In summary , almost identical r ( s ) values were found for PBS and HEPES buffer . Thus , at various concentrations even above those used in experiments , the tweezers only produced a hydrodynamic radius slightly above the monomeric species . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 026 Diffusion NMR ( DOSY ) experiments in both buffers revealed CLR01 hydrodynamic radii of ∼0 . 9–1 . 0 nm , slightly above the monomeric species ( Figure 7C , D , Figure 7—figure supplement 3 ) . Microcalorimetric dilution titrations revealed only minute endothermic heat changes , which argue strongly against an extensive aggregation process ( data not shown ) . Pyrene fluorescence revealed that no critical micelle concentration ( cmc ) could be determined in a wide concentration range ( 0–0 . 5 mM CLR01 ) encompassing fibril assembly and remodeling conditions ( Figure 7E ) . Dynamic light scattering ( DLS ) experiments showed CLR01 particles with hydrodynamic radius RH = 5–8 nm in PBS at concentration between 200 and 1000 μM , plus a very minor fraction of particles with a RH = 20–40 nm ( Figure 7F; data not shown ) . We did not observe CLR01 particles with a RH of ∼95–400 nm , which is the size range typically associated with colloidal small-molecule aggregates ( McGovern et al . , 2002 ) . No CLR01 particles could be detected at 10 or 50 μM in PBS ( data not shown ) . In HEPES buffer , no particles were detected at any of the concentration tested ( between 10 and 1000 μM ) ( Figure 7F; data not shown ) . The ∼0 . 9–1 nm hydrodynamic radius of CLR01 revealed by DOSY ( Figure 7C , D ) is below the detection limit of our DLS instrument , and hence , we cannot resolve the CLR01 monomer ( Figure 7F ) . Importantly , DLS overemphasizes large particles because the scattered-light intensity is proportional to the square of the particle mass . The scattering intensity in the solutions of CLR01 ( 1 mM ) in PBS was ∼3% that of similar samples containing Mg3 ( PO4 ) colloids ( 10 mM Mg3 ( PO4 ) ) or SDS micelles ( 2% SDS ( wt/vol ) ) , suggesting that the observed species represented a small fraction of the CLR01 molecules ( data not shown ) . Thus , under all assay concentrations the vast majority of the tweezer molecules are monomeric . Minute amounts of dimers or higher order species may be present at high concentrations of CLR01 in some cases in PBS ( fibril assembly buffer ) but are absent from the HEPES buffer ( fibril remodeling buffer ) . Thus , it is highly unlikely that colloidal CLR01 aggregates contribute to the observed activity of the tweezer . CLR01 exhibited a surprising ability to disrupt viral membranes ( Figures 5A , 6A , C–F ) , but did not affect cell viability like various non-ionic and anionic surfactants ( Figure 4—figure supplement 3 ) . We therefore explored whether CLR01 could act as a general inhibitor of enveloped viruses . To test this concept , human cytomegalovirus ( HCMV ) , herpes simplex virus type 2 ( HSV-2 ) , and hepatitis C virus ( HCV ) were treated with CLR01 or CLR03 and then assessed for their ability to infect target cells . Remarkably , CLR01 , but not CLR03 , reduced infection rates of all three analyzed enveloped viruses ( Figure 8A–C ) . By contrast , CLR01 did not inhibit infection by the non-enveloped human adenovirus type 5 ( HAdV5 ) ( Figure 8D ) . Thus , the tweezer is a broad-spectrum inhibitor of enveloped viruses , including viruses that can be sexually transmitted such as HIV-1 , HSV-2 , and HCV . 10 . 7554/eLife . 05397 . 027Figure 8 . CLR01 is a broad-spectrum inhibitor of enveloped viruses . ( A ) Human cytomegalovirus was incubated with PBS , 100 µM CLR03 , or 100 µM CLR01 . Afterwards , HFF cells were infected and immediate early ( IE ) antigen positive cells were counted 1 day post infection as a measure for infectivity . Values are means ±SD ( n = 3 ) . Unpaired t-tests were used to compare the buffer control to the CLR03 or CLR01 condition ( ns denotes not significant; * denotes p < 0 . 001 ) . ( B ) Herpes simplex virus type 2 comprising a GFP reporter gene was treated with PBS , 100 µM CLR03 or 100 µM CLR01 and added to Vero cells . GFP-positive cells were counted using flow cytometry 2 days post infection . Values represent means ±SD ( n = 3 ) . Unpaired t-tests were used to compare the buffer control to the CLR03 or CLR01 condition ( ns denotes not significant; * denotes p < 0 . 001 ) . ( C ) A luciferase encoding hepatitis C virus was treated with 150 µM CLR01 or 150 µM CLR03 and used for infection of Huh-7 . 5 reporter cells . Infection was measured 3 days post infection . Values represent means ±SEM ( n = 3 ) . Unpaired t-tests were used to compare the buffer control to the CLR03 or CLR01 condition ( ns denotes not significant; * denotes p < 0 . 001 ) . ( D ) A GFP-reporter adenovirus type 5 was added to A549 cells after treatment with 158 µM CLR01 or 158 µM CLR03 . GFP positive cells were counted using flow cytometry 1 day post infection . Values represent means ±SD ( n = 3 ) . Unpaired t-tests were used to compare the buffer control to the CLR03 or CLR01 condition ( ns denotes not significant ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 027 CLR01 binds exposed lysine and arginine residues in amyloidogenic peptides ( Sinha et al . , 2011; Attar et al . , 2012; Prabhudesai et al . , 2012; Sinha et al . , 2012; Acharya et al . , 2014; Ferreira et al . , 2014; Zheng et al . , 2015 ) . In vivo , the amyloids we examined are present in the complex environment of human semen . We wondered whether the interactions between CLR01 and seminal peptides might be hindered in conditions resembling those in seminal fluid . To test whether this is the case , lyophilized PAP248-286 was dissolved in an artificial semen simulant ( AS ) ( Owen and Katz , 2005; Olsen et al . , 2012 ) containing 50 mg/ml BSA and agitated until fibril formation was complete . These PAP248-286 ( AS ) fibrils were then diluted in AS and incubated with CLR01 . A reduction in ThT fluorescence intensity to 43% of the initial value was detected ( Figure 9A ) . This finding confirms that CLR01 maintains its amyloid-remodeling activity in a complex solution resembling seminal fluid . 10 . 7554/eLife . 05397 . 028Figure 9 . CLR01 diminishes the infection enhancing property of semen . ( A ) SEVI fibrils were formed in an artificial semen simulant ( AS ) ( Owen and Katz , 2005 ) . The resulting fibrils were then diluted to 20 µM in AS and treated with 200 µM CLR01 or CLR03 , or with buffer for 2 hr . Fibril integrity was assessed using ThT fluorescence . Values represent means ±SEM ( n = 4 ) . A one-way ANOVA with the post hoc Dunnett's multiple comparisons test was used to compare the buffer alone control to the other conditions ( ns denotes not significant; denotes p < 0 . 01 ) . ( B ) CLR01 abrogates semen-mediated enhancement of HIV infection . Seminal plasma ( 10% ) or cell culture medium containing CLR01 or CLR03 was mixed with CCR5-tropic HIV-1 or transmitter/founder HIV-1 CH058 . After 10 min , TZM-bl cells were infected and infectivity was measured 3 day post infection . Shown are the n-fold increased infection rates obtained for semen-treated virus relative to those of medium-treated virus . Values represent means ±SD ( n = 4 ) . Unpaired t-tests were used to compare the buffer control ( 0 µM compound ) to the CLR03 or CLR01 condition at each concentration ( ns denotes not significant; * denotes p < 0 . 05; denotes p < 0 . 01; * denotes p < 0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 028 Finally , we tested whether the molecular tweezer inhibited semen-mediated infection enhancement , as described ( Müller and Münch , 2016 ) . Seminal plasma , which was obtained from pooled semen of 10 donors , was treated with CLR01 or CLR03 . Thereafter , a CCR5-tropic lab-adapted virus or a transmitted/founder virus were incubated with 10% of these solutions followed by infection of TZM-bl cells . Consistent with previous studies ( Münch et al . , 2007; Hauber et al . , 2009; Roan et al . , 2009; Kim et al . , 2010 ) , seminal plasma enhanced infection of both viruses by 10- or 8-fold , respectively ( Figure 9B ) . CLR01 decreased viral infectivity enhancement in a concentration-dependent manner , while CLR03 had no effect on semen-mediated infection enhancement ( Figure 9B ) . At a CLR01 concentration of 31 µM , any stimulatory effect of semen was abolished ( Figure 9B ) . Thus , CLR01 prevents HIV infection in the presence of semen , the main vector for viral transmission in the human population . Despite the development of several different classes of topical microbicides , none have proven safe and effective at HIV prevention . The failure of topical microbicide candidates in previous clinical trials has been attributed to lack of adherence ( Van Damme et al . , 2008; Marrazzo et al . , 2015 ) , adverse effects ( McGowan et al . , 2011 ) , and a greatly diminished antiviral efficacy in the presence of semen ( Neurath et al . , 2006; Zirafi et al . , 2014 ) . This bodily fluid is not only the main vector for HIV transmission but also contains cationic amyloid fibrils that markedly increase viral infectivity ( Castellano and Shorter , 2012; Münch et al . , 2014 ) . Various biological polyanions such as heparin or other glycosaminoglycans can prevent these cationic fibrils from enhancing HIV infection ( Roan et al . , 2009 ) . Unfortunately , however , such anionic polymers have been unsuccessful in past clinical microbicide trials due to their poor bioavailability and induction of inflammatory responses in the genital tract , which actually augment HIV transmission by recruiting HIV-susceptible target cells to the genital mucosa ( Lüscher-Mattli , 2000; van de Wijgert and Shattock , 2007 ) . We have suggested that future microbicide endeavors should focus on agents that simultaneously and safely target HIV and the host factors that are exploited by the virus to facilitate its transmission ( Castellano and Shorter , 2012; Zirafi et al . , 2014; Roan and Münch , 2015 ) . Here , we report that CLR01 , a lysine- and arginine-specific molecular tweezer ( Fokkens et al . , 2005 ) , not only counteracts the infection-enhancing activity of seminal amyloids and semen , but also directly destroys HIV virions ( Figure 10 ) . CLR01 is a highly promising topical microbicide candidate because it possesses potent antiviral and anti-amyloid activity , displays minimal toxicity in vivo ( Prabhudesai et al . , 2012; Attar et al . , 2014; Ferreira et al . , 2014 ) , and is efficacious in human seminal fluid . 10 . 7554/eLife . 05397 . 029Figure 10 . CLR01 acts as a dual-function inhibitor of viral infection . Schematic overview of the anti-amyloid and antiviral effects of CLR01 . DOI: http://dx . doi . org/10 . 7554/eLife . 05397 . 029 It was surprising that CLR01 not only affected the formation and function of seminal amyloids but also displayed a broad and direct antiviral activity against HIV and other enveloped viruses ( Figure 8 ) . These diverse activities made us concerned that CLR01 might disrupt amyloidogenesis and viral infection via a non-specific mechanism involving the formation of colloidal CLR01 aggregates ( McGovern et al . , 2002 , 2003; Shoichet , 2006; Feng et al . , 2008 ) . Nevertheless , multiple lines of evidence argue against a non-specific , colloidal mechanism of CLR01 activity . First , using a variety of biophysical techniques , we were unable to detect significant quantities of colloidal CLR01 aggregates under the conditions employed in our experiments ( Figure 7 ) . Second , the anti-amyloid and antiviral activity of CLR01 was unaffected by high concentrations of BSA ( Figure 2—figure supplement 4; Figure 3—figure supplement 2; Figure 5—figure supplement 2 and 3; Figure 9A ) , which would adsorb and quench the activity of any potential colloidal micelles ( McGovern et al . , 2002; Feng et al . , 2008 ) . Third , attempts to preclear CLR01 solutions by centrifugation to remove any colloidal aggregates ( McGovern et al . , 2003; Feng et al . , 2008 ) had no effect on CLR01 anti-amyloid or antiviral activity ( Figure 2—figure supplement 4; Figure 3—figure supplement 2; Figure 5—figure supplement 4 ) . Fourth , the shallow CLR01 dose–response curves are not typical of colloidal aggregate inhibitors , which typically exhibit steep dose–response curves ( Shoichet , 2006 ) . Finally , a feature of small molecules that form colloidal aggregates is that they are non-specific inhibitors ( McGovern et al . , 2002 , 2003; Feng et al . , 2008 ) . By contrast , CLR01 displays specificity for amyloidogenesis by peptides that harbor lysine and arginine residues . Thus , if the lysine and arginine residues of PAP248-286 or PAP85-120 are replaced with alanine , then CLR01 can neither inhibit fibril assembly nor remodel preformed fibrils ( Figures 2G–I , 3I–L ) . Likewise , the antiviral activity of CLR01 is restricted to enveloped viruses and CLR01 is ineffective against non-enveloped viruses ( Figure 8 ) . This specificity is inconsistent with an indiscriminate , non-specific mechanism of action mediated by colloidal CLR01 aggregates . Importantly , the anti-amyloid and antiviral activity of CLR01 could be eliminated by excess lysine or poly-L-lysine . Thus , the lysine-binding cavity of CLR01 plays an important role in both anti-amyloid and antiviral activity . The ability of CLR01 to antagonize seminal amyloid in the presence of BSA and to prevent HIV infection in the presence of seminal fluid ( Figure 9 ) raises another question . Why is CLR01 activity not quenched by lysine or arginine residues on the surface of BSA or on other proteins abundant in seminal fluid ? Here , it is important to consider the profound difference between freely accessible lysines on intrinsically unstructured domains , which can all be complexed by the tweezers , and lysines in folded proteins whose surface often prohibits the sterically demanding tweezer molecule to approach a lysine residue close enough for the threading mechanism ( Figure 1B ) . The interaction between free lysine and CLR01 has a Kd of ∼10 µM , with fast on and off rates ( Fokkens et al . , 2005; Bier et al . , 2013; Dutt et al . , 2013 ) . On folded proteins , however , only a few lysines are accessible for the relatively large molecular skeleton of the tweezers . For example , only 5 of 17 surface lysines in a 14-3-3 protein are most likely to be complexed by CLR01 ( Bier et al . , 2013 ) . Even in an unstructured small protein , such as Aβ , CLR01 has clear preference for the free Lys16 over Lys28 ( Sinha et al . , 2011 ) because the latter is engaged in stabilizing hydrophobic interactions and salt bridges ( Petkova et al . , 2002; Lazo et al . , 2005; Xiao et al . , 2015 ) . Moreover , because of the high on and off rates , CLR01 is expected to only disrupt relatively weak interactions as well as events that depend on critical and accessible lysine or arginine residues . Therefore , the tweezer has shown selectivity in multiple cases that on first impression may appear counter-intuitive . For example , inhibition of abnormal protein aggregation is typically achieved at ∼1:1–1:5 protein:CLR01 concentration ( Sinha et al . , 2011 , 2012; Acharya et al . , 2014 ) , whereas disruption of controlled protein polymerization , for example , of tubulin , requires ∼55-fold excess CLR01 ( Attar et al . , 2014 ) and enzyme inhibition requires ∼1000-fold excess CLR01 ( Talbiersky et al . , 2008 ) . Moreover , in vivo , CLR01 shows no toxicity at doses 2–3 orders of magnitude above those needed to enable clearance of amyloid ( Attar et al . , 2012; Prabhudesai et al . , 2012; Attar et al . , 2014; Ferreira et al . , 2014 ) . Thus , selectivity for lysine-bearing amyloid conformers in complex biological fluids is achieved by CLR01 . We suggest that the lysine-rich nature of seminal amyloid peptides ( Figure 1C–E ) makes them unusually sensitive to CLR01 activity . Indeed , substoichiometric CLR01 concentrations ( relative to peptide monomers ) suffice to inhibit PAP248-286 , PAP85-120 , and SEM1 ( 45-107 ) fibrillization ( Figure 2D–F ) . Remarkably , substoichiometric CLR01 concentrations ( relative to peptide monomers ) also sufficed to effectively remodel SEVI and PAP85-120 fibrils ( Figure 3G , H ) . Curiously , SEM1 ( 45-107 ) fibrils could not be remodeled by CLR01 . Why SEM1 ( 45-107 ) fibrils were refractory to CLR01 remains unclear and requires further study . CLR01 may require longer time periods to remodel SEM1 ( 45-107 ) fibrils as with remodeling of Aβ fibrils , which required several weeks ( Sinha et al . , 2011 ) . It will also be important to determine whether some lysine and arginine residues in PAP248-286 , PAP85-120 , and SEM1 ( 45-107 ) are more critical than others for CLR01 activity . How does CLR01 directly antagonize viral infection ? Since the CLR03 control , which has a similar negative charge as CLR01 ( Sinha et al . , 2011 ) , was entirely ineffective at blocking viral infection , a purely polyanion-like mechanism that prevents the interaction of virions and cells by increasing charge repulsions ( Neurath et al . , 2002; Weber et al . , 2005 ) could be excluded . Instead , we demonstrate that CLR01 selectively disrupts membranes containing elevated levels of sphingomyelin and cholesterol ( Figure 6 ) . Envelopes of HIV-1 , herpes viruses , and HCV differ significantly from the cellular plasma membrane as they are more highly enriched in these and other lipids ( van Genderen et al . , 1994; Brügger et al . , 2006; Chan et al . , 2008; Lorizate et al . , 2009 , 2013; Merz et al . , 2011 ) . Indeed , the HIV membrane possesses a highly unusual lipid composition , which is distinct even from the detergent-resistant microdomains typically found in the plasma membrane , indicating that HIV budding involves a specific lipid raft clustering process ( Brügger et al . , 2006 ) . For example , although HIV membrane resembles lipid raft microdomains in that it is enriched for saturated lipids , phosphatidylserine , plasmalogen-phosphatidylethanolamine , cholesterol , and sphingolipids , it is also enriched for the unusual sphingolipid dihydrosphingomyelin ( Brügger et al . , 2006; Lorizate et al . , 2009 , 2013 ) . The unique lipid composition of viral membranes may render them uniquely sensitive to CLR01 . Importantly , unlike various non-ionic and anionic surfactants , CLR01 has minimal effects on cell viability ( Figure 4—figure supplement 3 ) . Thus , CLR01 might leverage a therapeutic opportunity provided by the physiological difference between static viral membranes and biogenic cellular membranes capable of self-repair . Static viral membranes are unable to withstand disruption by CLR01 , whereas cells can likely repair lipid-raft-rich regions of the plasma membrane that might be disrupted by CLR01 . In this regard , CLR01 may possess similarities to other broad-spectrum antivirals such as LJ001 , a lipophilic aryl methyldiene rhodanine derivative , which selectively disrupt diverse viral membranes but are ineffective against non-enveloped viruses , and also do not disrupt cell membranes ( Wojcechowskyj and Doms , 2010; Wolf et al . , 2010; Vigant et al . , 2013 ) . We investigated whether the tweezer might recognize the charged lipid head groups but found via NMR chemical-shift experiments that this was not the case ( data not shown ) . However , experiments on various phase boundaries strongly indicated that the tweezer could migrate into lipid bilayers due to its amphiphilic character ( data not shown ) . A pronounced destabilizing effect may be triggered if this happens at the edge of lipid rafts that are enriched in virion membranes . Importantly , HSV-2 , HCMV , and HCV can also be transmitted via sexual intercourse ( Handsfield et al . , 1985; Rapp , 1989; Tedder et al . , 1991 ) . Thus , a CLR01-based microbicide may not only protect from HIV-1 acquisition but also from other major human pathogens . CLR01 binding to lysine residues could potentially disrupt normal protein function leading to toxicity or side effects . However , CLR01 showed no signs of cytotoxicity in this study and has previously been safely employed in multiple cell and animal models ( Sinha et al . , 2011; Attar et al . , 2012; Prabhudesai et al . , 2012; Attar et al . , 2014 ) . For example , transgenic mice that were given 10 mg CLR01 a day per 1 kg body weight intraperitoneally for 30 days showed no apparent toxicity or adverse effects ( Attar et al . , 2014 ) . Moreover , it is conceivable that the application of CLR01 on mucosal surfaces should be even safer compared to its systemic administration . Unlike previous microbicide candidates that block HIV infection by a single mechanism , a CLR01-based formulation would interfere with various essential steps in sexual HIV transmission ( Figure 10 ) . On a minute time frame , the tweezer destroys infectious HIV particles that are present in semen , prevents the formation of amyloid–virus complexes by neutralization of the fibril surface charge , and also displaces pre-bound virions from fibrils ( Figure 10 ) . On a longer time frame ( hours ) , CLR01 not only prevents the formation of infection-enhancing seminal amyloids upon ejaculation but also remodels fibrils that are already abundant in semen after ejaculation or formed during semen liquefaction ( Figure 10 ) . Given the time frame of these effects , we suggest that the primary mechanism of CLR01 action is via its direct effect on the virus , and that the anti-amyloid effects , although potent , play a secondary role . Regardless , these combined antiviral and anti-amyloid activities and the encouraging safety data make CLR01 a promising broad-spectrum , topical microbicide against HIV-1 and other sexually transmitted viruses . We anticipate that large-scale synthesis of CLR01 could be relatively inexpensive ( less than $1 per mg ) , which would enable facile development as a broad-spectrum antiviral microbicide . The interactions of PAP248-286 with the molecular tweezer CLR01 and the spacer CLR03 were investigated using Replica Exchange Molecular Dynamics ( REMD ) simulations ( Sugita and Okamoto , 1999; Okabe et al . , 2001 ) performed with Gromacs 4 . 6 ( Hess et al . , 2008 ) and the CHARMM27 force field ( Mackerell et al . , 2004; Bjelkmar et al . , 2010 ) . The water solvent was treated explicitly using the TIP3P model ( Jorgensen et al . , 1983 ) . The parameters for CLR01 and CLR03 were obtained using the Swissparam server and tested previously ( Zoete et al . , 2011; Bier et al . , 2013 ) . The initial coordinates of PAP248-286 were taken from the Protein Data Bank , code 2L3H ( Nanga et al . , 2009 ) . The temperature range for the REMD simulations was 290–330 K . The temperature distribution was obtained as described by Patriksson and van der Spoel ( 2008 ) . Here , we focused our discussion on the 300 K trajectory . Three systems were investigated: ( 1 ) PAP248-286 , ( 2 ) PAP248-286 with 7 CLR01 molecules ( one for each Lys or Arg residue , with the exception of Lys272 which was not sterically accessible in the initial structure but became available during the simulations ) , and ( 3 ) PAP248-286 with 8 CLR03 molecules interacting via the hydrogen phosphate groups with each Lys or Arg residue of PAP248-286 . For each system , 22 replicas were simulated during 60 ns . The cluster analysis of the REMD simulations was performed with the gromos method using a cut-off value of 0 . 2 nm ( Daura et al . , 1999 ) . All chemicals or biochemicals were from Sigma–Aldrich ( St . Louis , MO ) unless otherwise stated . CLR01 and CLR03 were generated as described previously ( Fokkens et al . , 2005 ) , and 7 . 4–20 mM stock solutions were prepared in water or in PBS . Synthetic peptides PAP248-286 , PAP248-286 ( Ala ) , PAP85-120 , and SEM1 ( 45-107 ) or ( 49-107 ) were purchased from Keck Biotechnology Resource Laboratory ( Yale University , New Haven , CT ) or Celtek peptides ( Franklin , TN ) . PAP85-120 ( Ala ) was purchased from Bachem ( King of Prussia , PA ) . For fibril formation , peptides were reconstituted and assembled into fibrils as previously described ( Münch et al . , 2007; Roan et al . , 2011; Arnold et al . , 2012 ) . L-lysine , poly-L-lysine hydrobromide ( molecular weight 4000–15 , 000 by viscosity ) , and poly-L-lysine hydrobromide ( molecular weight 70 , 000–150 , 000 by viscosity ) were from Sigma–Aldrich . Seminal plasma represents the cell-free supernatant fraction of pooled human semen centrifuged at 20 , 000×g for 30 min at 4°C . CLR01 was dissolved in PBS ( 10 mM sodium phosphate , 137 mM NaCl , 2 . 7 mM KCl , pH 7 . 4 ) or HEPES buffer ( 25 mM HEPES , 150 mM KOAc , 10 mM Mg ( OAc ) 2 , pH 7 . 4 ) and diluted from 6 . 4 mM to 100 μM . In the corresponding 1H NMR spectra , the chemical shift changes of the inner aromatic proton forming a binding isotherm , were analyzed by nonlinear regression and furnished the 1:1 dimerization constant . CLR01 was dissolved in PBS or HEPES buffer ( pH 7 . 4 ) and diluted from 6 . 4 mM to 200 μM . DOSY spectra were measured for all concentrations of CLR01 , and the diffusion coefficient was determined from the corresponding Stejskal and Tanner Plot . The diffusion coefficient was converted into the hydrodynamic radius by way of the Stokes–Einstein equation . A dilution experiment was performed with a HEPES buffered solution of CLR01 ( 6 mM–100 μM ) . Evolved heats were very small ( 0 . 1–0 . 2 μcal per injection ) and endothermic . Pyrene ( 10 μM ) was dissolved in PBS or HEPES buffer containing 5% of DMSO , and the intensity ratio of both fluorescence emission maxima I392/I373 was determined at various CLR01 concentrations ( 10–500 μM ) . As a reference , SDS solutions in the same buffer were analyzed at 0 . 1–50 mM ( cmc at 6–8 mM ) . CLR01 was dissolved in phosphate-buffered saline ( PBS 10 mM sodium phosphate , 137 mM NaCl , 2 . 7 mM KCl , pH 7 . 4 ) or 25 mM HEPES , 150 mM KOAc , 10 mM Mg ( OAc ) 2 , pH 7 . 4 at 10 , 50 , 200 , 500 , or 1000 μM . The solutions were transferred to DLS cuvettes and centrifuged for 30 min at 5000×g to pellet dust particles . Solutions were measured using an in-house-built system with a He-Ne laser , model 127 ( wavelength 633 nm , power 60 mW; Spectra Physics Lasers , Mountain View , CA ) . Light scattered at 90° was collected using image-transfer optics and detected by an avalanche photodiode built into a 256-channel PD2000DLS correlator ( Precision Detectors , Bellingham , MA ) . The size distribution of scattering particles was reconstructed from the scattered light correlation function using PrecisionDeconvolve software ( Precision Detectors ) based on the regularization method by Tikhonov and Arsenin ( Tikhonov and Arsenin , 1977 ) . As a reference , Mg3 ( PO4 ) colloids ( 10 mM Mg3 ( PO4 ) ) or SDS micelles ( 2% SDS ( wt/vol ) ) were analyzed . For assembly experiments , each reconstituted peptide was incubated with CLR01 or CLR03 and agitated at 37°C at 1400 rpm . At various time points , aliquots ( 1 µl ) were removed and added to 25 µM ThT in PBS ( 200 µl ) . Changes in fluorescence ( excitation: 440 nm , emission: 482 nm ) were measured using a Tecan Safire2 microplate reader ( Tecan , Männedorf , Switzerland ) . To assess the extent of fibril assembly using sedimentation analysis , fibril samples ( 20 µM fibrils , 100 µl volume ) were centrifuged for 10 min at 13 , 200 rpm . The supernatant was carefully removed and transferred to a new tube , and the pellet was redissolved in an equal volume ( 100 µl ) of buffer . Then , 50 µl of 3× sample buffer ( 6% SDS , 187 . 5 mM Tris , 30% glycerol , 10% β-mercaptoethanol , 0 . 05% bromophenol blue , pH 6 . 8 ) was added to the supernatant and pellet fractions . Samples were analyzed by SDS-PAGE using 10–20% Tris-Tricine peptide gels and XT Tricine running buffer ( Bio-Rad , Hercules , CA ) and visualized by coomassie staining . A gradient of soluble peptide controls were also run on the gels . Densitometry ( using Image J software ) was used to quantify the percent of protein in the pellet fractions by comparing to a standard curve created from the soluble peptide controls . For amyloid-remodeling experiments , fibrils ( 20 µM , based on peptide monomer concentrations ) were diluted into an assay buffer ( 25 mM HEPES , 150 mM KOAc , 10 mM Mg ( OAc ) 2 , pH 7 . 4 ) in the presence of ATP ( 5 mM ) and incubated with either CLR01 or CLR03 . Aliquots ( 5 µl ) were removed at various time points and added to 25 µM ThT in PBS ( 55 µl ) . ThT fluorescence was measured as above . The artificial semen simulant was prepared as described previously ( Owen and Katz , 2005; Olsen et al . , 2012 ) , and is comprised of: 18 mM citrate; 40 mM chloride; 7 mM calcium; 4 . 5 mM magnesium; 28 mM potassium; 220 mM sodium; 2 mM zinc; 15 mM fructose; 6 mM glucose; 50 . 4 mg/ml BSA; 7 mM lactic acid; 7 . 5 mM urea; all in a 123 mM sodium phosphate base , pH 7 . 7 . In some fibril assembly and remodeling experiments , BSA ( 10 mg/ml ) was also included . To assess any contribution of potential colloidal CLR01 aggregates , CLR01 was first reconstituted at the requisite concentration for fibril assembly or remodeling reactions , and then centrifuged at 16 , 100×g for 20 min at 25°C . The supernatant fraction was then collected and used in fibril assembly or remodeling reactions . For some fibril assembly and remodeling reactions , CLR01 was incubated with a 200-fold molar excess of L-lysine or 10-fold molar excess of poly-L-lysine ( molecular weight 4000–15 , 000 by viscosity ) for 10 min on ice prior to addition to unassembled peptide or preformed fibrils . To determine IC50 or EC50 values and Hill slopes for dose–response relationships , the data were analyzed using GraphPad Prism software . A nonlinear regression analysis ( log ( inhibitor ) vs response–variable slope ) was used and fitted with the least squares ( ordinary ) fit . Preformed SEVI , PAP85-120 , or SEM1 ( 45-107 ) fibrils ( 5 µM monomer ) were preincubated with ThT ( 25 µM ) for 30 min at room temperature . Buffer ( PBS ) , CLR01 ( 250 µM ) , or a known competitor of ThT binding , BTA-1 ( 250 µM ) ( Lockhart et al . , 2005 ) were then added and incubated for 10 min at room temperature . ThT displacement was then assessed by fluorescence measurements . Aliquots were removed from the assembly and remodeling reactions , spotted for 10 min on Formvar carbon-coated grids ( EM Sciences ) , stained for 5 min with 2% uranyl acetate , and washed with distilled water . Samples were visualized using a Jeol-1010 transmission electron microscope . CD spectra were collected on an AVIV Model 410 Circular Dichroism Spectrometer . Mean residue ellipticity ( MRE ) was calculated using the equation MRE = θ/ ( 10lcN ) where θ is the measured ellipticity in millidegrees , l is the pathlength in cm , c is the molar protein concentration , and N is the number of residues . PAP248-286 , PAP85-120 , and SEM1 ( 49-107 ) fibrils were treated with a 10-fold excess of CLR01 or CLR03 . After centrifugation at 20 , 000×g for 10 min , the pellets were resuspended in 1 mM KCl . Zeta potential was measured using the Zeta Nanosizer ( Malvern Instruments , UK ) . To analyze the impact of lysine or poly-L-lysine on the binding of CLR01 to semen amyloids , SEVI fibrils were treated with a twofold excess of CLR01 and increasing concentrations of lysine ( 1 , 10 and 100-fold excess over CLR01 ) or poly-L-lysine ( molecular weight 70 , 000–150 , 000 by viscosity; 0 . 02 , 0 . 25 and 2 . 5-fold excess of lysine monomer equivalents over CLR01 ) . After centrifugation at 20 , 000×g for 10 min , samples were processed and measured as described above . Fibrils ( 200 µg/ml in PBS ) were stained with Proteostat Amyloid Plaque Detection Kit ( Enzo Life Sciences , Plymouth Meeting , PA ) . Fibrils were then treated with 20-fold excess CLR01 or CLR03 and mixed 1:2 with MLV-Gag-YFP virions . Samples were transferred to µ-slides VI0 . 4 ( Ibidi , Munich , Germany ) and imaged with a Zeiss LSM confocal microscope . The reporter cell line TZM-bl was obtained through the NIH ARRRP and cultured in cell culture medium ( DMEM medium supplemented with 120 µg/ml penicillin , 120 µg/ml streptomycin , 350 µg/ml glutamine and 10% inactivated fetal calf serum ( FCS ) , Gibco , Life Technologies , Frederick , MD ) . This cell line is stably transfected with an LTR-lacZ cassette and expresses CD4 , CXCR4 , and CCR5 . Upon infection with HIV-1 , the viral protein Tat is expressed which activates the long terminal repeat ( LTR ) resulting in the generation of β-galactosidase molecules . Virus stocks of X4-tropic HIV-1 NL4-3 , R5-tropic HIV-1 NL4-3 92TH014 , and of the transmitter/founder viruses THRO . c and CH058 . c ( kindly provided by B Hahn ) were generated by transient transfection of 293T cells as described ( Münch et al . , 2007 ) . After transfection and overnight incubation , the transfection mixture was replaced with 2 ml cell culture medium with 2% inactivated FCS . After 40 hr , the culture supernatant was collected and centrifuged for 3 min at 330×g to remove cell debris . Virus stocks were analyzed by p24 antigen ELISA and stored at −80°C . To assess the effect of CLR01 and CLR03 on amyloid-mediated enhancement of HIV-1 infection , 104 TZM-bl cells in 180 µl cell culture medium were seeded in 96-well flat-bottom plates the day before infection . 200 µg/ml fibrils ( 44 µM SEVI , 45 µM PAP85-120 fibrils , 30 µM SEM1 ( 49-107 ) fibrils ) were treated with a 20-fold molar excess of CLR01 or CLR03 for 10 min at room temperature , serially diluted fivefold and then mixed with R5-tropic HIV-1 NL4-3 92TH014 ( 0 . 5 ng/ml p24 antigen ) . After 5 min , 20 µl of these mixtures were added to TZM-bl cells and infection rates were determined 3 days post infection by detecting β-galactosidase activity in cellular lysates using the Tropix Gal-Screen kit ( Applied Biosystems , Life Technologies , Frederick , MD ) and the Orion microplate luminometer ( Berthold , Bad Wildbad , Germany ) . All values represent reporter gene activities ( relative light units per second; RLU/s ) derived from triplicate infections minus background activities derived from uninfected cells . To assess the effect of CLR01 and CLR03 on semen-mediated enhancement of HIV-1 infection , 104 TZM-bl cells were seeded in 280 µl cell culture medium supplemented with 50 µg/ml gentamycin in 96-well flat-bottom plates the day before infection . Seminal plasma ( 20% ) was treated with different concentrations of CLR01 or CLR03 ( highest 925 µM ) for 10 min at room temperature and then mixed with R5-tropic HIV-1 NL4-3 and CH058 ( 0 . 5 ng/ml p24 antigen ) . After 5 min , 20 µl of these mixtures were added to 280 µl TZM-bl cells . To minimize cytotoxic effects mediated by seminal plasma , the inoculums were replaced 2 hr later with fresh cell culture medium . Infection rates were determined as described above . The effect of CLR01 , CLR03 , nonoxynol-9 , sodium dodecyl sulfate and Triton X-100 on the metabolic activity of TZM-bl cells was analyzed using the MTT assay . After 3 days of incubation , 20 µl of 5 mg/ml MTT ( 3-[4 , 5-dimethyl-2-thiazolyl]-2 , 5-diphenyl-2H-tetrazolium bromide ) solution was added to the cells . After 4 hr the cell-free supernatant was discarded and formazan crystals were dissolved in 100 µl DMSO:Ethanol ( 1:2 ) . Absorption was detected at 490 nm and corrected by the background absorption at 650 nm . R5-tropic HIV-1 NL4-3 92TH014 was treated with 40 µM CLR01 in the presence of buffer or increasing concentrations of poly-L-lysine ( molecular weight 70 , 000–150 , 000 by viscosity ) ( 0 . 3–75 nM ) . After a 10-min incubation , TZM-bl cells were infected and β-galactosidase activity was measured 2 days post infection . Microtiter plates ( 96 well , flat bottom , Sarstedt 83 . 3924 ) were coated over night with 100 µl of 0 , 0 . 2 , 1 or 5% BSA dissolved in PBS at 4°C . After washing three times with PBS the coated wells were filled with 70 µl of 900 µM CLR01 and incubated 30 min at 37°C . Then , CLR01 was serially diluted and added to HIV-1 NL4-3 92TH014 . Following a 10-min incubation , TZM-bl cells were infected and β-galactosidase activity was measured 2 days post infection . R5-tropic HIV-1 NL4-3 92TH014 was treated with 220 µM CLR01 in the presence of buffer or increasing concentrations of BSA ( 0 . 01–10% ) . After a 10-min incubation , TZM-bl cells were infected and β-galactosidase activity was measured 2 days post infection . HIV-1 NL4-3 92TH014 was incubated for 10 min at 37°C with PBS , 100 µM CLR03 or 100 µM CLR01 and centrifuged at 20 , 000×g and 4°C for 1 hr . The p24 content of the supernatant and pellet was determined using an in house p24-antigen ELISA . Virus solutions ( 20 µl ) were deposited on aminopropyl-modified glass cover slips ( AP-Glass ) and incubated for 1 hr at room temperature . After removing excess liquid , the deposited virus particles were treated with 40 µl of 100 µM CLR01 or CLR03 and incubated for 10 min at RT . The samples were rinsed with PBS and imaged in PBS on a Nanowizard 3 AFM ( JPK ) in Quantitative Imaging mode using silicon nitride cantilevers with a spring constant of 0 . 03 N/m ( Bruker , Billerica , MA ) . Sodium dihydrogen phosphate , chloroform , and cholesterol ( Chol ) were purchased from Sigma–Aldrich . Lissamine rhodamine B 1 , 2-dihexadecanoyl-sn-glycero-3-phosphoethanolamine ( N-Rh-DHPE ) , ATTO 647 were purchased from Life Technologies and ATTO-TEC , respectively . 1 , 2-dioleoyl-sn-glycero-3-phosphocholine ( DOPC ) , sphingomyelin ( SM ) , and 23- ( dipyrrometheneboron difluoride ) -24-norcholesterol ( Bodipy-Chol ) were purchased from Avanti Polar Lipids ( Alabaster , AL ) . All chemicals used were of the highest analytical grade available and used without further purification . Giant unilamellar vesicles ( GUVs ) were prepared by electroformation on optically transparent and electrically conductive indium tin oxide ( ITO ) -coated glass slides ( SPI Supplies ) in a preparation chamber consisting of a closed bath imaging chamber RC-21B affixed to a P-2 platform ( both Warner Instruments Co . ) topped with a flow-through temperature block . A solution of pure DOPC-containing 0 . 2 mol% N-Rh-DHPE or a lipid mixture of DOPC/SM/Chol ( 45/25/30 mol% ) containing 0 . 2 mol% N-Rh-DHPE and 0 . 1 mol% Bodipy-Chol in chloroform was spread on an ITO-coated cover slip ( 20 μl , 1 mg/ml ) , spin-coated at 800 rpm for 1 min , and subsequently dried under vacuum for at least 2 hr . Afterwards , the lipids were hydrated in 10 mM NaH2PO4 pH 7 . 6 , containing the water-soluble fluorophore ATTO 647 ( 5 μM ) within the preparation chamber . The electroformation of pure DOPC and the DOPC/SM/cholesterol mixture was performed at room temperature and 60°C , respectively , by applying a frequency alternating current field ( 500 Hz , 100 mV for 10 min , 1 V for 20 min and 1 . 6 V for 2 . 5 hr ) to the ITO electrodes by a function generator ( Thurlby Thandar Instruments TG315 ) . Afterwards , the preparation chamber was cooled down to room temperature in the case of the lipid mixture and carefully rinsed with 10 mM NaH2PO4 pH 7 . 6 , to remove the water-soluble ATTO 647 that was not enclosed in the interior of the vesicles . Once a region of interest for imaging of the GUVs was chosen under the microscope , ∼500 µl of CLR01 ( 150 µM ) in 10 mM NaH2PO4 pH 7 . 6 , was added . Images were recorded by a confocal laser-scanning microscope ( Biorad MRC 1024 ) coupled via a side port to an inverted microscope ( Nikon; Eclipse TE-300DV ) enabling fluorescence excitation in the focal plane of an objective lens ( Nikon Plan Apo 60× WI , NA 1 . 2 ) . Fluorescence of Bodipy-Chol , N-Rh-DHPE , and ATTO 647 was sequentially acquired by alternating the excitation with the 488 , 568 , and 647 nm lines of a Kr/Ar laser ( Dynamic Laser , Salt Lake City , UT , USA ) . Signals were detected in three different PMT channels ( emission band pass filters 522 nm/FWHM [full width at half-maximum] 35 nm , 580 nm/FWHM 32 nm , and 680 nm/FWHM 32 nm ) . Image acquisition was controlled by the software LaserSharp2000 ( Biorad ) . Analysis of the data was performed using the software Fiji ( Max Planck Society for the Advancement of Science e . V . , Munich , Germany ) . Images were background and brightness/contrast corrected . The phospholipids 1 , 2-dioleoyl-sn-glycero-3-phosphocholine ( DOPC ) and sphingomyelin ( egg , chicken ) ( SM ) were purchased from Avanti Polar Lipids ( Alabaster , AL , USA ) . Sodium dihydrogen phosphate , chloroform , and cholesterol ( Chol ) were purchased from Sigma–Aldrich ( Steinheim , BW , Germany ) . Stock solutions of 10 mg/ml lipids ( DOPC , SM , Chol ) in chloroform were dissolved to obtain 1 . 95 mg of total lipid with the composition of DOPC ( 100 mol% ) and DOPC/SM/Chol ( 45/25/30 mol% ) for the liquid AFM experiments . The majority of the chloroform was evaporated with a nitrogen stream and the rest of the solvent was removed afterwards by drying under vacuum overnight . The sodium dihydrogen phosphate buffer was filtered through filters of 0 . 02-µm pore size ( Whatman , Dassel , Germany ) before use . The dry lipid films were hydrated with 1 ml of 10 mM NaH2PO4 , pH 7 . 6 . Afterwards , the lipid mixtures were vortexed , kept in a water bath at 65°C for 15 min , and then sonicated for 10 min . After five freeze-thaw-vortex cycles and brief sonication , large multilamellar vesicles were formed and transformed to large unilamellar vesicles of uniform size by use of an extruder ( Avanti Polar Lipids , Alabaster , USA ) with polycarbonate membranes of 100 nm pore size at 65°C ( Weise et al . , 2009 , 2011 ) . The vesicle fusion on mica was carried out by depositing 35 µl of the extruded lipid vesicle solution together with 35 µl of NaH2PO4 buffer on freshly cleaved mica and incubation in a wet chamber at 70°C for 2 hr . After the vesicle fusion , the samples were rinsed carefully with NaH2PO4 buffer to remove unspread vesicles . For the tweezer–membrane interaction studies , 800 µl of 150 µM CLR01 in NaH2PO4 buffer were slowly injected into the AFM fluid cell at room temperature and allowed to incubate for different time periods ( 1 min and 1 hr ) . Afterwards , the fluid cell was rinsed carefully with NaH2PO4 buffer before imaging to remove unbound tweezer . The measurements were performed on a MultiMode scanning probe microscope with a NanoScope IIIa controller ( Bruker , Camarillo , CA , USA ) and usage of a J-Scanner ( scan size 125 µm ) . Images were obtained by applying the tapping mode in liquid with sharp nitride lever ( SNL ) probes mounted in a fluid cell ( Bruker , Camarillo , CA , USA ) . Tips with nominal force constants of 0 . 24 N m−1 were used at driving frequencies around 9 kHz and drive amplitudes between 343 and 570 mV . Slow scan frequencies between 1 . 0 kHz and 1 . 97 kHz were required for high-resolution images . The height and phase images of sample regions were acquired with resolutions of 512 × 512 pixels . All AFM measurements were carried out at room temperature and the partitioning of the tweezer was analyzed by using analysis and NanoScope version 5 processing software ( Weise et al . , 2009 , 2011 ) .	Human Immunodeficiency Virus ( HIV ) is a sexually transmitted virus that can cause a serious disease that weakens the immune system . The virus is most commonly transmitted between individuals in semen , the male reproductive fluid . Semen contains deposits of protein fragments called amyloid fibrils , which can increase the transmission of HIV by trapping viral particles . This helps the virus to attach to the membranes surrounding human cells , which increases the risk of infection . Therefore , therapies that reduce the levels of amyloid fibrils in semen might be able to reduce the transmission of HIV . Drugs that prevent amyloid formation are already being developed because structurally similar fibrils can also form in the brains of individuals with neurodegenerative diseases . One such molecule—called CLR01—works by binding to particular sites on the proteins that form fibrils in the brain . This inhibits fibril formation and slowly disassembles the fibrils that have already formed . CLR01 physically interacts with these residues in a way that resembles a tweezer . The peptides in the amyloid fibrils in semen also have these sites , which suggests that CLR01 might also disrupt amyloid fibrils from forming in semen . Here Lump and Castellano et al . show that CLR01 can both disrupt fibril formation and remodel fibrils that have already formed . In addition , CLR01 prevents HIV particles from interacting with these fibrils and can displace the virus particles that have already bound to the fibrils . In the presence of CLR01 , human cells exposed to semen that contained HIV were less likely to become infected with the virus . Unexpectedly , CLR01 also directly destroys HIV and other enveloped viruses such as HCV or HSV particles by disrupting the membranes that surround the virus . Therefore , Lump and Castellano et al . 's findings reveal that CLR01 has considerable potential to be used as an agent for reducing the transmission of HIV and other sexually transmitted viral diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2015	A molecular tweezer antagonizes seminal amyloids and HIV infection
MicroRNAs ( miRNAs ) direct post-transcriptional regulation of human genes by guiding Argonaute proteins to complementary sites in messenger RNAs ( mRNAs ) targeted for repression . An enigmatic feature of many conserved mammalian miRNA target sites is that an adenosine ( A ) nucleotide opposite miRNA nucleotide-1 confers enhanced target repression independently of base pairing potential to the miRNA . In this study , we show that human Argonaute2 ( Ago2 ) possesses a solvated surface pocket that specifically binds adenine nucleobases in the 1 position ( t1 ) of target RNAs . t1A nucleotides are recognized indirectly through a hydrogen-bonding network of water molecules that preferentially interacts with the N6 amine on adenine . t1A nucleotides are not utilized during the initial binding of Ago2 to its target , but instead function by increasing the dwell time on target RNA . We also show that N6 adenosine methylation blocks t1A recognition , revealing a possible mechanism for modulation of miRNA target site potency . MicroRNAs ( miRNAs ) are an abundant class of regulatory molecules with diverse biological functions in plants and animals ( Lagos-Quintana et al . , 2001; Lau et al . , 2001; Lee and Ambros , 2001; Reinhart et al . , 2002 ) . Over two thousand unique human miRNA sequences have been reported and more than half of all protein-coding human genes are predicted to contain a conserved miRNA recognition site ( Friedman et al . , 2009; Kozomara and Griffiths-Jones , 2011 ) . miRNAs function as guides for Argonaute proteins , which use the sequence information encoded in each miRNA to identify complementary sties in mRNAs targeted for repression ( Liu et al . , 2004; Meister et al . , 2004 ) . Argonautes then recruit additional silencing factors that mediate translational repression and degradation of the targeted mRNAs ( Huntzinger and Izaurralde , 2011 ) . Pioneering studies in the nematode Caenorhabditis elegans showed that perfect complementarity between miRNAs and their targets is not necessary for silencing ( Lee et al . , 1993; Wightman et al . , 1993 ) . Examination of regulatory elements in fly mRNAs then revealed a striking degree of complementarity to the 5′ ends of subset of conserved miRNAs , suggesting that base paring interactions with nucleotides towards the 5′ end of the miRNA are particularly important for target recognition ( Lai , 2002 ) . Indeed , phylogenetic analysis showed that pairing to the miRNA ‘seed region’ ( nt 2–7 or 2–8 , from the miRNA 5′ end ) is the most evolutionarily conserved feature of miRNA targets in animals ( Lewis et al . , 2003; Brennecke et al . , 2005; Krek et al . , 2005; Lewis et al . , 2005 ) , and complementarity to the miRNA seed is generally sufficient to elicit significant levels of target recognition and repression ( Doench and Sharp , 2004; Brennecke et al . , 2005; Lim et al . , 2005 ) . Careful comparison of miRNA seed-matched sites in the 3′ untranslated regions ( UTRs ) of human , mouse , rat , dog , and chicken genomes also revealed that an adenosine ( A ) nucleotide opposite miRNA nucleotide-1 is a conserved feature of many vertebrate miRNA target sites ( Lewis et al . , 2005 ) . These t1A nucleotides confer enhanced repression of miRNA targets beyond pairing to the seed region alone ( Grimson et al . , 2007; Nielsen et al . , 2007; Baek et al . , 2008; Selbach et al . , 2008 ) . Curiously , unlike the seed-matched region , t1A nucleotides function independently of the identity of miRNA nucleotide-1 , indicating that t1A recognition occurs through a mechanism that is distinct from base pairing with the miRNA guide ( Grimson et al . , 2007; Nielsen et al . , 2007; Baek et al . , 2008 ) . We recently reported crystal structures of human Argonaute2 ( Ago2 ) bound to a guide RNA and short , seed-paired target RNAs ( Schirle et al . , 2014 ) . These structures revealed that Ago2 cradles the duplex formed by miRNA nucleotides 2–7 and the complementary target RNA , explaining why seed pairing is critical for recognition of miRNA target sites ( Figure 1A ) . We also noted that t1A nucleotides interact with a surface pocket formed at the interface of the L2 and MID domains of Ago2 , and proposed that this additional interaction may contribute to the affinity of Ago2 for miRNA target sites . However , the t1-nucleotide binding pocket is large enough to accommodate any of the four natural RNA bases , and the contacts between Ago2 and t1A appear to be almost entirely non-specific ( Figure 1B ) . Therefore , it is not clear how A nucleotides in the t1 position of miRNA targets are recognized and confer enhanced repression . In this study , we show that the t1-nucleotide binding pocket in Ago2 preferentially interacts with A nucleotides through water-mediated contacts to adenosine N6 amine . Adding a methyl group to the t1A N6 amine reduces target affinity , raising the possibility that adenosine methylation could , in principle , lead to partial derepression of miRNA targets containing 7mer-A1 or 8mer sites . We also present data indicating that t1A is not used in the initial search for target sites , but instead provides an anchor that helps retain Ago2 on seed-matched sites on target RNAs . 10 . 7554/eLife . 07646 . 003Figure 1 . Structure of the t1 nucleotide binding pocket . ( A ) Linear schematic of the Argonaute2 ( Ago2 ) primary structure . ( B ) Crystal structure of Ago2 bound to a guide RNA ( red ) and target RNA bearing t1A nucleotide ( blue; PDB ID: 4W5O ) . ( C ) Close-up view of the t1-binding pocket . Ordered water molecules shown as pink spheres . Protein shown in stick and surface representations . Target RNA shown as sticks . ( D ) Linear schematic of crystallized guide and target RNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 07646 . 003 We first visualized how Ago2 engages non-A t1 nucleotides by determining crystal structures of Ago2 bound to short target RNAs with cytosine ( C ) , uracil ( U ) , or guanine ( G ) in the t1 position ( Table 1 ) . The overall conformation of Ago2 is nearly identical in the four different t1 structures , indicating that identity of the t1 nucleotide does not affect the structure of the protein . Inspection of target RNA omit maps , in which the target RNAs were excluded from refinement , revealed little or no well-defined electron density for all of the non-A nucleotides , indicating that the non-A t1 nucleotides were mostly disordered ( Figure 2 ) . In contrast , the target omit map from the t1A data set had clearly defined density for the t1A inside the t1-nucleotide binding pocket . Moreover , although residual electron density was observed for t1G and t1C nucleotides , binding experiments using the same target RNAs show that these interactions do not contribute substantially to target affinity ( Figure 2 ) . We conclude that , although the t1-binding site is large enough to accommodate any of the four natural RNA bases , adenine is the only nucleobase that associates stably enough to contribute to overall target affinity . 10 . 7554/eLife . 07646 . 004Table 1 . Crystallographic and refinement statistics for wild-type Ago2-guide-target complexesDOI: http://dx . doi . org/10 . 7554/eLife . 07646 . 004Target RNAt1-Ct1-Gt1-Ut1-DAPt1-InosinePDB code4Z4C4Z4D4Z4E4Z4F4Z4GSpace groupP1211P1211P1211P1211P1211Unit cell dimensions a ( Å ) 55 . 6955 . 7455 . 6455 . 8655 . 66 b ( Å ) 116 . 56117 . 02116 . 84116 . 60117 . 0 c ( Å ) 69 . 6169 . 8769 . 7470 . 3870 . 1 β ( ° ) 92 . 4392 . 4392 . 4392 . 5292 . 40Ago2 per ASU11111Data collection Wavelength ( Å ) 0 . 979450 . 979500 . 979180 . 979500 . 97950 Resolution ( Å ) 38 . 85–2 . 30 ( 2 . 38–2 . 30 ) 39 . 01–1 . 60 ( 1 . 63–1 . 60 ) 55 . 60–1 . 80 ( 1 . 90–1 . 80 ) 38 . 86–2 . 80 ( 2 . 95–2 . 80 ) 39 . 01–2 . 70 ( 2 . 83–2 . 70 ) Total reflections133 , 678532 , 622351 , 63465 , 58775 , 249 Unique reflections38 , 61411377582 , 07121 , 07824 , 061 Completeness ( % ) 98 . 4 ( 96 . 3 ) 96 . 8 ( 93 . 5 ) 99 . 7 ( 99 . 7 ) 95 . 0 ( 92 . 6 ) 97 . 2 ( 91 . 9 ) Redundancy3 . 5 ( 3 . 4 ) 4 . 7 ( 4 . 6 ) 4 . 3 ( 3 . 7 ) 3 . 1 ( 3 . 0 ) 3 . 1 ( 2 . 9 ) I/σI13 . 1 ( 2 . 2 ) 13 . 7 ( 2 . 0 ) 10 . 3 ( 1 . 9 ) 9 . 9 ( 2 . 4 ) 9 . 6 ( 2 . 2 ) Rmerge7 . 7 ( 53 . 0 ) 5 . 5 ( 74 . 9 ) 9 . 8 ( 81 . 9 ) 9 . 8 ( 57 . 7 ) 8 . 4 ( 48 . 2 ) Rpim7 . 3 ( 49 . 2 ) 3 . 1 ( 59 . 0 ) 5 . 3 ( 47 . 4 ) 9 . 7 ( 57 . 1 ) 6 . 4 ( 36 . 3 ) Refinement Resolution ( Å ) 35 . 30–2 . 3039 . 01–1 . 6040 . 27–1 . 8035 . 47–2 . 8035 . 41–2 . 70 R-free/R-factor21 . 86/16 . 9318 . 90/16 . 3118 . 54/15 . 7523 . 30/18 . 3922 . 44/17 . 86 R . M . S . deviation Bond distances ( Å ) 0 . 0140 . 0140 . 0070 . 0060 . 008 Bond angles ( ° ) 1 . 4541 . 4601 . 1310 . 9030 . 989 Number of atoms Non-hydrogen , protein64296469642964216404 Non-hydrogen , RNA568580552571572 Phenol2828282128 Isopropanol420800 Phosphate100500 Mg33333 Water27955454516105 Ramachandran Plot Most favored regions95 . 33%96 . 79%96 . 97%94 . 93%95 . 17% Additionally allowed4 . 54%3 . 21%3 . 03%4 . 94%4 . 83% Generously llowed0 . 13%0 . 00%0 . 00%0 . 13%0 . 00%10 . 7554/eLife . 07646 . 005Figure 2 . The t1 nucleotide binding pocket specifically recognizes adenosine Strong electron density was observed for adenosine in the t1-binding pocket , but not t1C , G , or U . ( A ) T1A target RNA omit map contoured at 3σ ( grey mesh ) . ( B ) T1C target RNA omit map contoured at 3σ ( grey mesh ) . ( C ) T1G target RNA omit map contoured at 3σ ( grey mesh ) . ( D ) T1U target RNA omit map contoured at 3σ ( grey mesh ) . ( E ) Plot of target bound vs Ago2-guide concentration for target RNAs with different t1-nucleotides . DOI: http://dx . doi . org/10 . 7554/eLife . 07646 . 005 The contacts between Ago2 and t1A appear to be largely non-specific ( Figure 1 ) . Therefore , the mechanism by which Ago2 distinguishes the t1A purine ring from t1G is unclear . Two major differences between adenine and guanine are: position-6 , where adenine has an exocyclic amine , and guanine has a carbonyl; and position-2 , where guanine has an exocyclic amine , and adenine has no substituent group ( Figure 3A ) . We dissected the influence of these two elements on t1-binding by examining interactions with target RNAs with either inosine ( I ) or 2 , 6-diaminopurine ( DAP ) in the t1 position . DAP has an N6 amine ( like adenine ) , and an N2 amine ( like guanine ) . Conversely , inosine has an O6 carbonyl ( like guanine ) , but lacks an N2 amine ( like adenine ) . Ago2 bound the t1I target with an affinity similar to t1G ( Table 2 ) and no electron density was observed for the t1I nucleotide target omit maps , indicating that removal of the guanine N2 amine is insufficient to promote interactions with the t1-binding site ( Figure 3B ) . In contrast , we observed unambiguous electron density for the t1DAP nucleotide , which bound Ago2 in the same position and anti conformation as t1A ( Figure 3C ) . Moreover , the affinity of the t1DAP target is ∼1 . 7-fold greater than that of the equivalent t1A target , indicating that Ago2 has a modest , though significant , preference for t1DAP over t1A ( p-value = 0 . 0025; two-tailed , unpaired Student's t-test ) . We conclude that the purine N6 and N2 amines both have positive effects on t1-binding , and that the N6 amine is the major determinant for distinguishing the t1A purine ring from tG . Consistent with this idea , adding a methyl group to the t1A N6 amine reduced target affinity to that of a non-t1A target ( Figure 3D ) . 10 . 7554/eLife . 07646 . 006Figure 3 . A purine N6 amine is required for t1 nucleotide recognition . ( A ) Chemical structures of adenosine , 2 , 6-diaminopurine ( DAP ) , guanosine , and inosine ( I ) nucelobases . Hydrogen bond donors highlighted in blue; hydrogen bond acceptors highlighted in pink . ( B ) t1I target RNA omit map contoured at 3σ ( grey mesh ) . No electron density was observed for t1I . ( C ) t1DAP target RNA omit map contoured at 3σ ( grey mesh ) . Clear electron density was observed for t1DAP . ( D ) Plot of target bound vs Ago2-guide concentration for t1A , t1I , t1DAP , and t1m6A target RNAs . DOI: http://dx . doi . org/10 . 7554/eLife . 07646 . 00610 . 7554/eLife . 07646 . 007Table 2 . Ago2-target affinitiesDOI: http://dx . doi . org/10 . 7554/eLife . 07646 . 007t1 nucleotideKD ( nM ) WT Ago2A481T-Ago2A0 . 75 ± 0 . 041 . 5 ± 0 . 13G1 . 9 ± 0 . 091 . 8 ± 0 . 11U1 . 9 ± 0 . 10–C1 . 8 ± 0 . 12–DAP0 . 45 ± 0 . 03–I1 . 7 ± 0 . 09–m6A1 . 8 ± 0 . 12–Dissociation constants for wild-type ( WT ) and mutant Ago2 binding short target RNAs with different t1 nucleotides . The only direct contact between the t1A N6 amine and Ago2 is a hydrogen bond to the side chain of Ser-561 ( Figure 1 ) . Because the serine alcohol can function as both a hydrogen bond acceptor and donor , it is unlikely that this interaction alone can discriminate a purine N6 amine from an O6 carbonyl . Inspection of the atoms surrounding Ser-561 revealed no set of interactions that could conceivably steer the hydrogen-bonding capacity of the serine alcohol . However , examination of the solvent surrounding the t1A nucleobase identified four ordered water molecules in a hydrogen-bonding network that could selectively interact with purine N6 amines . In this model , both hydrogen atoms in water ‘A’ are engaged in hydrogen bonds with the main chain carbonyls of Lys-440 and Met-437 ( and possibly the main chain carbonyl of Asp-480 ) , leaving only lone pair electrons available for interactions with water ‘B’ ( Figure 4B ) . Water B must therefore donate one hydrogen atom to establish a hydrogen bond with water A . The second hydrogen atom in water B is likely involved in a hydrogen bond with the main chain carbonyl of Ile-477 . Thus , our model suggests that Ago2 uses main chain carbonyls inside the t1-binding pocket to direct the hydrogen atoms on water B away from the purine 6-position , leaving only lone pair electrons available for establishing interactions with t1 nucleotides . In the proposed orientation , the lone pair electrons on water B are positioned to accept a hydrogen bond from purine N6 amines and to repel O6 carbonyls . Waters ‘C’ and ‘D’ likely provide an additional layer of selectivity through interactions with the unprotonated N1 amine on t1A , and may interact with the purine N2 amine , which does not directly contact Ago2 . 10 . 7554/eLife . 07646 . 008Figure 4 . Water-mediated recognition of t1A . ( A ) Water network within the t1-binding pocket . Protein main chain shown as sticks , with side-chains ( except S561 ) hidden for clarity . Water molecules shown as pink spheres . Water omit map shown contoured at 2 . 5σ ( grey mesh ) . Potential hydrogen bonds shown as black dashed lines; t1A shown as blue sticks . ( B ) Flattened cartoon schematic of image shown in ( A ) . Drawing illustrates proposed positions of hydrogen atoms within the t1A recognition network . DOI: http://dx . doi . org/10 . 7554/eLife . 07646 . 008 We tested the role of water molecules in t1A recognition by designing a mutation in Ago2 that disrupts the water network inside the t1-binding site . Ala-481 , which resides in the back of the t1-binding pocket and is within van der Waals contact distance of water B , was mutated to threonine . Our rational was that the A481T mutation would alter the hydrogen-bonding landscape of the t1-binding pocket and thereby perturb placement of the associated water molecules . We first assessed the effects of the A481T mutation on Ago2 by determining the crystal structure of the mutant protein bound to a short target RNA bearing a t1A nucleotide ( Table 3 ) . The overall structure of A481T Ago2 is nearly identical to the wild type protein , and the structure of the t1-binding pocket is almost completely unaffected by the A481T mutation . The Thr-481 side chain is in the same orientation as that of Ala-481 , with the Cγ methyl pointed towards the protein interior and the Oγ alcohol extending into the t1-binding site ( Figure 5A ) . Thus , the A481T mutation adds a hydrogen-bond acceptor/donor to the back of the t1-binding pocket without otherwise perturbing the structure of Ago2 . 10 . 7554/eLife . 07646 . 009Table 3 . Crystallographic and refinement statistics for mutant ( A481T ) Ago2-guide-target complexesDOI: http://dx . doi . org/10 . 7554/eLife . 07646 . 009Target RNAt1-At1-GPDB code4Z4H4Z4ISpace groupP1211P1211Unit cell dimensions a ( Å ) 55 . 6955 . 60 b ( Å ) 116 . 60116 . 60 c ( Å ) 70 . 1069 . 62 β ( ° ) 92 . 2992 . 42Ago2 per ASU11Data collection Wavelength ( Å ) 0 . 999990 . 99999 Resolution ( Å ) 44 . 81–2 . 50 ( 2 . 61–2 . 50 ) 44 . 69–2 . 80 ( 2 . 95–2 . 80 ) Total reflections11333285 , 923 Unique reflections30 , 41121 , 701 Completeness ( % ) 98 . 6 ( 98 . 3 ) 99 . 2 ( 97 . 6 ) Redundancy3 . 7 ( 3 . 8 ) 4 . 0 ( 4 . 0 ) I/σI9 . 2 ( 2 . 3 ) 8 . 7 ( 2 . 4 ) Rmerge12 . 7 ( 69 . 7 ) 13 . 1 ( 60 . 0 ) Rpim12 . 0 ( 65 . 0 ) 12 . 0 ( 55 . 0 ) Refinement Resolution ( Å ) 44 . 81–2 . 5044 . 69–2 . 80 R-free/R-factor21 . 13/17 . 2123 . 32/19 . 18 R . M . S . deviation Bond distances ( Å ) 0 . 0080 . 005 Bond angles ( ° ) 0 . 9400 . 865 Number of atoms Non-hydrogen , protein64326412 Non-hydrogen , RNA568568 Phenol2821 Isopropanol420 Phosphate100 Mg33 Water13523 Ramachandran plot Most favored regions95 . 19%94 . 67% Additionally allowed4 . 68%5 . 33% Generously allowed0 . 13%0 . 00%10 . 7554/eLife . 07646 . 010Figure 5 . Disruption of the t1 pocket water network abolishes t1A recognition . ( A ) Overlay of wild-type and A481T Ago2 structures and close-up view of the region surrounding the A481T mutation ( inset ) . ( B , C ) Crystal structures of A481T-Ago2 bound to t1A ( B ) or t1G ( C ) target RNAs . Target RNA omit maps contoured at 3σ ( grey mesh ) . ( D ) Plot of bound t1A and t1G target RNAs vs A481T-Ago2-guide concentration . ( E ) Overlay of A481T-Ago2 and t1A from wild-type ( WT ) Ago2 ( semi-transparent ) structures . 4 . 2 Å distance between the T481 Oγ hydroxyl and the t1A N6 amine indicated as dashed line . DOI: http://dx . doi . org/10 . 7554/eLife . 07646 . 010 Although the structure of the A481T Ago2 is nearly identical to wild type , t1A recognition is severely impaired . Only residual electron density was observed for the t1A nucleotide in the target RNA omit map ( Figure 5B ) . For comparison , we determined the structure of the A481T mutant bound to a target bearing a t1G nucleotide . Again , the structure of the mutant protein was essentially indistinguishable from the wild type and only residual electron density was observed for the t1G ( Figure 5C ) . Moreover , the A481T Ago2 bound t1A and t1G target RNAs with nearly identical affinities ( Figure 5D ) . Importantly , the Oγ hydroxyl of Thr-481 is 4 . 2 Å from the position of the t1A N6 amine in the wild-type structure , strongly indicating that the mutation affects t1A recognition indirectly ( Figure 5E ) . We conclude that binding specificity in the t1-pocket is achieved by Ago2-directed solvent interactions with the t1 adenosine . Binding t1A nucleotides may increase the affinity of Ago2 for target RNAs by increasing the association rate , decreasing the rate of dissociation , or both . To distinguish between these possibilities , we used a single-molecule technique for directly observing Ago2-target binding events that we recently developed ( Chandradoss et al . , 2015 ) . Briefly , Ago2 is loaded with a Cy3-labeled miRNA and introduced into a microfluidic chamber containing immobilized , Cy5-labled target RNAs . Pairing between the miRNA and the complementary site on the target RNA brings the Cy3 donor and Cy5 acceptor fluorophores into close proximity , leading to high Forster resonance energy transfer ( FRET ) , which is observed at single-molecule resolution by total-internal-reflection microscopy ( Figure 6A–C ) . 10 . 7554/eLife . 07646 . 011Figure 6 . t1A nucleotides increase the dwell time on target sites . ( A ) Cartoon schematic of a single-molecule FRET assay . ( B ) Sequences of miRNA and target RNA with base pairs shown . The donor fluorophore ( Cy3 ) is positioned on the ninth nt of miRNA ( counting from the 5′ end of miRNA ) and the acceptor ( Cy5 ) on target RNA opposite nt 17 of miRNA . ( C ) Representative time trajectory . Δτ , dwell time of interaction; kon ( obs ) , apparent binding rate . The thin grey box indicates the time of a flow that delivers Ago2 and microRNA into the observation chamber . ( D ) Accumulated number of first Ago2-miRNA/target binding events vs time for RNA targets bearing t1U , t1A , or t1DAP . The number of events was normalized by the total number of target RNA strands over an imaging area . ( E ) Binding event dwell times fit to a double exponential decay ( blue ) . The t1A binding events fit two populations ( 49 . 8 ± 6 . 2% that exhibits Δτ1 and 50 . 2 ± 6 . 2% that exhibits Δτ2; R2 = 0 . 998 ) . The t1DAP binding events fit two populations ( 44 . 6 ± 4 . 4% that exhibits Δτ1 and 55 . 4 ± 4 . 4% that exhibits Δτ2; R2 = 0 . 996 ) . Dotted grey lines represent fits to a single exponential decay . The first data columns were excluded to avoid artifacts arising from the time resolution limit . The number of events per bin was normalized by the total amount of binding events per each histogram . DOI: http://dx . doi . org/10 . 7554/eLife . 07646 . 011 We first determined the rates at which Ago2 associates with target RNAs containing six nucleotides of complementarity to the guide seed region ( nt 2–7 ) and either a t1U , t1A , or t1DAP nucleotide . Measuring the time of the first arrival revealed that Ago2 recognizes the three targets at similar rates ( kon ( obs ) = 0 . 040 ± 0 . 009 , 0 . 033 ± 0 . 004 , and 0 . 041 ± 0 . 008 s−1 nM−1 for t1U , t1A , and t1DAP targets , respectively ) , demonstrating that t1-nucleotide identity has little , if any , effect on the association of Ago2 with target RNAs ( Figure 6D ) . In contrast , the average dwell time ( Δτ ) of Ago2 on the t1A and t1DAP targets was more than threefold longer than on the t1U target ( Figure 6E ) . The combined results suggest that t1A nucleotides are not used for initial target site recognition , but instead function primarily by enhancing the stability of the Ago2:target interaction after seed-pairing . We note that , although the Δτ distribution of t1U is well described by a single-exponential decay ( Δτ = 1 . 64 ± 0 . 11 s , R2 = 0 . 994 ) , the Δτ distributions of t1A and t1DAP deviate noticeably from ideal ( R2 = 0 . 972 and 0 . 961 for t1A and t1DAP , respectively; Figure 6E , gray lines ) . This observation indicates that t1A and t1DAP binding events consist of at least two discrete populations . Assuming only two populations exist , we fit each data set to a double-exponential decay and calculated that about half of the binding events ( 49 . 8% ± 6 . 2% and 44 . 6% ± 4 . 4% for t1A and t1DAP , respectively ) had dwell times closely matching that of the t1U target ( Δτ1 ( t1A ) = 1 . 61 ± 0 . 29 s; Δτ1 ( t1DAP ) = 1 . 97 ± 0 . 24 ) . In contrast , the other half of the binding events had an average dwell time more than fivefold longer ( Δτ2 ( t1A ) = 8 . 69 ± 0 . 45 s; Δτ2 ( t1DAP ) = 12 . 87 ± 0 . 47 s ) . The simplest explanation for these observations is that , when using limited seed-pairing as in our experimental setup , Ago2 dissociates from the target RNA without stably engaging t1A about half of the time . However , in cases in which Ago2 does engage t1A , the affinity of the interaction is increased substantially . This effect is exaggerated with t1DAP , which stabilizes the association further still . Notably , the t1A nucleobase inserts into the t1-binding pocket at an angle nearly orthogonal to the trajectory of the paired target RNA as it extends into the central cleft of Ago2 ( Figure 1 ) . Based on this observation , we suggest that t1A binding increases the dwell time on target RNAs by anchoring Ago2 to seed-paired target sites and inhibiting dissociation . In this study we provide evidence that human Ago2 contains a t1-nucleotide binding pocket that specifically recognizes adenine nucleobases through water-mediated contacts . Our model is reminiscent of interactions in the trp operator/repressor complex , which also uses an ordered network of hydrogen-bonded water molecules to achieve nucleobase recognition ( Otwinowski et al . , 1988 ) . We also show that the adenosine N6 amine is a key determinant of specificity and that the addition of a methyl group to the N6 amine blocks t1A recognition . Taken with the finding that m6A modifications are enriched in mammalian mRNA 3′ UTRs ( Meyer et al . , 2012 ) , this result raises the intriguing possibility that , in some cases , miRNA targets may be partially derepressed via adenosine methylation at 7mer-A1 and 8mer sites ( Bartel , 2009 ) . Insights into t1 recognition by Ago2 can also be extended to other eukaryotic members of the Argonaute protein family . The amino acids making up the t1-binding site in Ago2 are conserved in all four human Argonaute proteins , and the structure of the t1-binding pocket in Ago2 is nearly identical to the same region in the structure of human Ago1 ( Faehnle et al . , 2013; Nakanishi et al . , 2013 ) . In fact , two ordered water molecules corresponding to waters A and B were observed in the structure of human Ago1 crystallized in the absence of target RNA ( Nakanishi et al . , 2013 ) . We therefore suggest that all four human Argonautes likely recognize t1A nucleotides using the same mechanism . Additionally , recent studies indicate that some members of the Piwi clade of Argonaute proteins also display t1-nucleotide preferences . Specifically , the Drosophila protein Aubergine , the silkmoth protein Siwi , and the mouse protein Mili preferentially cleave targets bearing a t1A; while mouse protein Miwi2 prefers targets with a t1 purine ( Wang et al . , 2014 ) . We note that most Piwi proteins displaying t1 preferences have conserved residues lining the t1-binding site ( with the notable exception of Mili , which appears to lack several residues homologous to the t1A-binding residues in Ago2 ) . In contrast , Drosophila and silkmoth Ago3 , which do not display t1 preferences ( Wang et al . , 2014 ) , lack homology in key residues making up the t1-binding pocket ( Figure 7 ) . In C . elegans , both t1A and t1U nucleotides are conserved features of many miRNA target sites , indicating that the Argonaute proteins Alg-1 and Alg-2 also have t1 nucleotide preferences ( Jan et al . , 2011 ) . Curiously , there are only two minor differences between residues making up the human Ago2 t1-binding pocket and the equivalent residues in Alg-1 and Alg-2 ( Asn-429 and Arg-479 in Ago2 ) . Conceivably , these substitutions could reorganize the t1-pocket water network to recognize both adenine and uracil nucleobases . Alternatively , we suggest that Alg-1 and Alg-2 may specifically bind t1A , and that conservation of miRNA sites with t1U is connected to recognition by one or more of the other C . elegans Argonaute proteins . 10 . 7554/eLife . 07646 . 012Figure 7 . Conservation in the t1A-binding pocket . ( A ) Structure of the Ago2 t1A-binding pocket with major structural elements indicated . Ago2 shown as sticks with side chains of non-highlighted residues hidden for clarity . Target RNA shown in blue . ( B ) Multiple sequence alignment of Ago2 with other members of the extended Argonaute protein family . Conserved structural elements colored as in panel A . DOI: http://dx . doi . org/10 . 7554/eLife . 07646 . 012 Finally , we suggest that understanding the t1-binding site may provide new inroads for development of novel anti-miR oligonucleotides ( van Rooij and Kauppinen , 2014 ) . We found that the addition of a t1DAP nucleotide increased target affinity fourfold compared to non-t1A targets ( Table 2 ) . More importantly , the t1-binding pocket has a chemically diverse surface and is spacious enough to accommodate nucleotide analogues much more elaborate than DAP . Moreover , displacement of ordered water molecules A and B may lead to entropic gains that translate into higher binding affinity , in a fashion similar to that used in the development of cyclic urea HIV-1 protease inhibitors ( Lam et al . , 1994 ) . We suggest that t1-nucleotide analogs that make favorable contacts inside the t1-binding pocket could lead to anti-miRs that specifically target the Argonaute-miRNA complex . All RNA oligonucleotides were synthesized by Integrated DNA Technologies ( IDT , Coralville , IA ) , with the exception of t1I and t1DAP , which were synthesized by ValueGene , and t1m6A , which was synthesized by GE Healthcare Dharmacon ( Lafayette , CO ) . Prior to binding studies , all target RNAs were 5′ end labeled using γ-32P-ATP and T4 polynucleotide kinase ( New England Biolabs , Ipswich , MA ) and subsequently purified by denaturing 16% polyacrylamide gel electrophoresis ( PAGE ) and ethanol precipitation . RNAs used in crystal structures and equilibrium binding experiments: Guide RNA: 5′ p-UUCACAUUGCCCAAGUCUCUU 3′; Target RNAs ( t1 nucleotides in bold ) :t1A: 5′ CAAUGUGAAAA 3′t1U: 5′ CAAUGUGAUAA 3′t1C: 5′ CAAUGUGACAA 3′t1G: 5′ CAAUGUGAGAA 3′t1I: 5′ CAAUGUGAIAA 3t1DAP: 5′ CAAUGUGA ( DAP ) AA 3′t1m6A: 5′ CAAUGUGA ( m6A ) AA 3′ RNAs used in single molecule experiments Guide RNA ( U with C6 amino modifier for Cy3 attachment underlined ) : 5′ p-UGAGGUAUUUUUUUUUUUUUUU 3′ Target RNAs ( U with C6 amino modifier for Cy5 attachment underlined ) : t1A: 5′ ( U ) 20UUUUUUUUUUUACUACCUCA ( U ) 29-biotin 3′ t1U: 5′ ( U ) 20UUUUUUUUUUUACUACCUCU ( U ) 29-biotin 3′t1DAP: 5′ ( U ) 20UUUUUUUUUUUACUACCUC ( DAP ) ( U ) 29-biotin 3′ Full length wild type and A481T Ago2 proteins were expressed in Sf9 cells using a baculovirus system ( Schirle and MacRae , 2012 ) , and purified as described previously ( Schirle et al . , 2014 ) . Briefly , His6-tagged Ago2 was purified by Ni-chelate chromatography and loaded with single-stranded guide RNAs . Loaded Ago2 proteins where then isolated using the Arpón method for purifying Argonaute complexes loaded with a specified guide RNA ( Flores-Jasso et al . , 2013 ) . Ago2-guide-target complexes were formed by mixing Ago2-guide complexes with target RNAs at a 1:1 . 2 molar ratio and incubating at room temperature for 10 min . Crystals were grown by hanging drop vapor diffusion at 20°C . Drops contained a 1:1 ratio of Ago2-guide-target to reservoir solution ( 16% PEG 3350 , 0 . 1 M Tris , pH 8 . 0 , 0 . 1 M phenol , 12% isopropanol , and 10 mM MgCl2 ) . Crystals typically appeared overnight , were harvested with nylon loops , and flash frozen in liquid N2 . Data were collected under cryogenic conditions remotely at beam line 12-2 at the Stanford Synchrotron Radiation Lightsource ( SSRL ) , and beamline 24-ID-E at the Advanced Photon Source ( APS ) ( McPhillips et al . , 2002; Soltis et al . , 2008 ) . Data were processed using XDS and Scala ( Gonzalez and Tsai , 2010; Kabsch , 2010; Winn et al . , 2011 ) . All structures were refined using the Ago2-guide-target structure ( PDB ID 4W5O ) , with the t1A nucleotide omitted , as a starting model . Models were built using Coot ( Emsley et al . , 2010 ) and were subjected to XYZ coordinate , TLS , and B-factor refinement using PHENIX ( Adams et al . , 2010 ) . Model building and refinement continued iteratively until all interpretable electron density was modeled . Water molecules were identified automatically in Coot ( 2Fobs − Fcalc map , above 1 . 8σ , and greater than 2 . 4 Å and less than 3 . 2 Å from hydrogen bond donors or acceptors ) and by manual inspection of electron density maps . All structures were refined using an Rfree set identical to that used in refinement of the original 4W5O structure . Structure figures were generated with PyMOL ( Schrödinger , LLC , Portland , OR ) . Guide-loaded Ago2 samples ( 0–70 nM ) were incubated with 0 . 1 nM 32P-labeled target RNAs in binding reaction buffer ( 30 mM Tris pH 8 . 0 , 0 . 1 M potassium acetate , 2 mM magnesium acetate , 0 . 5 mM TCEP , 0 . 005% ( vol/vol ) NP-40 , 0 . 01 mg/ml baker's yeast tRNA ( Sigma , St . Louis , MO ) ) , in a total reaction volume of 100 µl , for 45 min at room temperature . Protein–RNA complexes and free RNA were separated using a dot-blot apparatus ( GE Healthcare Life Sciences , Pittsburgh , PA ) , using Protran nitrocellulose membrane ( 0 . 45 µm pore size , Whatman , GE Healthcare Life Sciences ) to bind protein complexes , and Hybond Nylon membrane ( Amersham , GE Healthcare Life Sciences ) to capture free RNA . Samples were applied with vacuum and then washed with 100 µl of ice-cold wash buffer ( 30 mM Tris pH 8 . 0 , 0 . 1 M potassium acetate , 2 mM magnesium acetate , 0 . 5 mM TCEP ) . Membranes were air-dried and 32P signal visualized by phosphorimaging . Quantification was performed using ImageQuant software ( GE Healthcare Life Sciences ) , and dissociation constants calculated using Prism version 5 . 0e ( GraphPad Software , Inc . , La Jolla , CA ) . Ago2 single molecule binding measurements were performed as described elsewhere ( Chandradoss et al . , 2015 ) . Briefly , target RNAs bearing a Cy5 dye ( GE Healthcare ) and a 3′ biotin were immobilized on a polymer ( PEG ) -coated quartz surface in the microfluidic chamber ( Chandradoss et al . , 2014 ) of a prism-type total internal reflection fluorescence microscope . Ago2 was loaded with a guide miRNA containing a Cy3 dye ( GE Healthcare ) . The resulting complex was introduced into the microfluidic chamber and Cy3 molecules were excited with a 532 nm diode laser ( Compass 215M/50 mW , Coherent ) . Fluorescence signals of Cy3 and Cy5 were collected through a 60× water immersion objective ( UplanSApo , Olympus , Center Valley , PA ) with an inverted microscope ( IX73 , Olympus ) . Laser scattering was blocked by a 532 nm long pass filter ( LPD01-532RU-25 , Semrock , Rochester , NY ) . The Cy3 and Cy5 signals were separated with a dichroic mirror ( 635 dcxr , Chroma , Bellows Falls , VT ) and imaged using a EM-CCD camera ( iXon Ultra , DU-897U-CS0-#BV , Andor Technology , United Kingdom ) and described previously ( Selvin and Ha , 2007 ) . CCD images of time resolution 0 . 1 s were recorded , and time traces were extracted from the CCD image series using IDL ( ITT Visual Information Solution , Boulder , CO ) . Colocalization between Cy3 and Cy5 signals was carried out with a custom-made mapping algorithm written in IDL . The extracted time traces were processed using Matlab ( MathWorks , Natick , MA ) and Origin ( Origin Lab , Northampton , MA ) . The binding rate ( kon ) was determined by first measuring the time between when Ago2-miRNA was introduced to a microfluidic chamber and when the first Ago2-miRNA was docked to a target RNA; and then fitting the time distribution with a single-exponential growth curve , A ( 1−e−kont ) . The dissociation rate was estimated by measuring the dwell time of a binding event . A dwell time distribution was fitted by either a single-exponential decay curve ( Ae−t/Δτ ) or a double-exponential decay curve ( A1e−t/Δτ1+A2e−t/Δτ2 ) . In case of a double-exponential decay , the percentages of Δτ1 and Δτ2 populations are determined by A1Δτ1/ ( A1Δτ1 + A2Δτ2 ) and A2Δτ2/ ( A1Δτ1 + A2Δτ2 ) , and the average dwell time is determined by ( A1Δτ12+A2τ22 ) / ( A1Δτ1+A2Δτ2 ) . A microfluidic chamber was incubated with 20 µl Streptavidin ( 0 . 1 mg/ml , Sigma ) for 30 s . Unbound Streptavidin was washed with 100 µl of buffer T50 ( 10 mM Tris–HCl [pH 8 . 0] , 50 mM NaCl buffer ) . The 50 μl of 50 pM acceptor-labelled mRNA construct were introduced into the chamber and incubated for 1 min . Unbound labeled constructs were washed with 100 µl of buffer T50 . The effector complex was formed by incubating 50 nM purified recombinant hAgo2 with 0 . 5 nM of donor-labeled hsa-let-7a miRNA in a buffer containing 50 mM Tris–HCl [pH 8 . 0] ( Ambion , Grand Island , NY ) , 50 mM NaCl ( Ambion ) , and 60 mM KCl ( Ambion ) at 31°C for 20 min . An imaging buffer for single-molecule FRET was added before the mixture was injected to a microfluidic chamber . The final concentration of the imaging buffer consists of the 0 . 8% dextrose ( Sigma ) , 0 . 5 mg/ml glucose oxidase ( Sigma ) , 85 µg/ml Catalase ( Merck ) , and 1 mM Trolox ( ( ± ) -6-hydroxy-2 , 5 , 7 , 8-tetramethylchromane-2-carboxylic acid , 238813 , Sigma ) . The experiments were performed at the room temperature ( 23 ± 2°C ) .	Stretches of DNA known as genes provide the instructions to make the proteins and RNA molecules a cell needs to work . To make a protein , the gene is used as a template to make a type of RNA molecule called messenger RNA ( mRNA ) , which is subsequently ‘translated’ into a protein . Most genes do not need to produce proteins all of the time , and so cells have several ways of stopping proteins from being made . For example , the Argonaute family of proteins prevents mRNA molecules from being translated into proteins . Argonautes are guided to their targets by short RNA molecules called microRNAs . RNA molecules are made up of a sequence of building blocks known as nucleotides , each of which can only bind to one other type of nucleotide . If part of the nucleotide sequence of a microRNA molecule corresponds with part of the nucleotide sequence of the mRNA , the two RNA molecules will bind to each other . This enables the microRNA and the Argonaute protein to prevent the mRNA being translated . If the mRNA has an adenine nucleotide in a particular position ( called ‘t1’ ) near the binding region in the mRNA sequence , Argonaute proteins will prevent translation more effectively . An adenine nucleotide in the t1 position is also known as a t1A nucleotide . In 2014 , researchers revealed the structure of a human Argonaute protein called Argonaute2 when it is bound to a microRNA-mRNA pair . This revealed that t1A nucleotides—but not other nucleotide types in the t1 position—interact with a ‘pocket’ in the Argonaute protein . However , it was not clear how the adenine nucleotide is recognized . Now , Schirle et al . —including several of the researchers involved in the 2014 work—use a technique called X-ray crystallography to examine how the t1A nucleotide interacts with Argonaute2 in more detail . This revealed that the Argonaute2 pocket contains many water molecules that form an organized network . This network interacts with part of the t1A nucleotide and helps to lock Argonaute2 onto its microRNA target sites . The discovery of the pocket and how t1A is recognized may now be used to design more effective ‘anti-miRs’—synthetic microRNA inhibitors that can treat diseases in which microRNAs work incorrectly , a feature common to many forms of cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2015	Water-mediated recognition of t1-adenosine anchors Argonaute2 to microRNA targets
MicroRNA targets are often recognized through pairing between the miRNA seed region and complementary sites within target mRNAs , but not all of these canonical sites are equally effective , and both computational and in vivo UV-crosslinking approaches suggest that many mRNAs are targeted through non-canonical interactions . Here , we show that recently reported non-canonical sites do not mediate repression despite binding the miRNA , which indicates that the vast majority of functional sites are canonical . Accordingly , we developed an improved quantitative model of canonical targeting , using a compendium of experimental datasets that we pre-processed to minimize confounding biases . This model , which considers site type and another 14 features to predict the most effectively targeted mRNAs , performed significantly better than existing models and was as informative as the best high-throughput in vivo crosslinking approaches . It drives the latest version of TargetScan ( v7 . 0; targetscan . org ) , thereby providing a valuable resource for placing miRNAs into gene-regulatory networks . MicroRNAs ( miRNAs ) are ∼22-nt RNAs that mediate post-transcriptional gene repression ( Bartel , 2004 ) . Bound with an Argonaute protein to form a silencing complex , miRNAs function as sequence-specific guides , directing the silencing complex to transcripts , primarily through Watson–Crick pairing between the miRNA seed ( miRNA nucleotides 2–7 ) and complementary sites within the 3′ untranslated regions ( 3′ UTRs ) of target RNAs ( Lewis et al . , 2005; Bartel , 2009 ) . The miRNAs conserved to fish have been grouped into 87 families , each with a unique seed region . On average , each of these families has >400 conserved targeting interactions , and together these interactions involve most mammalian mRNAs ( Friedman et al . , 2009 ) . In addition , many nonconserved interactions also function to reduce mRNA levels and protein output ( Farh et al . , 2005; Krutzfeldt et al . , 2005; Lim et al . , 2005; Baek et al . , 2008; Selbach et al . , 2008 ) . Accordingly , miRNAs have been implicated in a wide range of biological processes in worms , flies , and mammals ( Kloosterman and Plasterk , 2006; Bushati and Cohen , 2007; Stefani and Slack , 2008 ) . Critical for understanding miRNA biology is the accurate prediction of miRNA–target interactions . Although numerous advances have been made , accurate and specific target predictions remain a challenge . Analysis of preferentially conserved miRNA-pairing motifs within 3′ UTRs has led to the identification of several classes of target sites ( Bartel , 2009 ) . The most effective canonical site types , listed in order of decreasing preferential conservation and efficacy , are the 8mer site ( Watson–Crick match to miRNA positions 2–8 with an A opposite position 1 [Lewis et al . , 2005] ) , 7mer-m8 site ( position 2–8 match [Brennecke et al . , 2005; Krek et al . , 2005; Lewis et al . , 2005] ) , and 7mer-A1 site ( position 2–7 match with an A opposite position 1 [Lewis et al . , 2005] ) . Experiments have confirmed that the preference for an adenosine opposite position 1 is independent of the miRNA nucleotide identity ( Grimson et al . , 2007; Nielsen et al . , 2007; Baek et al . , 2008 ) and due to the specific recognition of the target adenosine within a binding pocket of Argonaute ( Schirle et al . , 2014 ) . Two other canonical site types , each associated with weaker preferential conservation and much lower efficacy ( Friedman et al . , 2009 ) , are the 6mer ( position 2–7 match [Lewis et al . , 2005] ) and offset-6mer ( position 3–8 match [Friedman et al . , 2009] ) . Pairing to the 3′ end of the miRNA can supplement canonical sites , although evidence for the use of this 3′-supplementary pairing is observed for no more than 5% of the seed-matched sites ( Brennecke et al . , 2005; Lewis et al . , 2005; Grimson et al . , 2007; Friedman et al . , 2009 ) . Some effective sites lack canonical seed pairing . For example , very extensive pairing to the 3′ region of the miRNA can compensate for a wobble or mismatch to one of the seed positions ( Doench and Sharp , 2004; Brennecke et al . , 2005; Bartel , 2009 ) , as exemplified by the two let-7 sites within the 3′ UTR of Caenorhabditis elegans lin-41 ( Reinhart et al . , 2000 ) . Although these 3′-supplementary sites can be detected above background when searching for preferentially conserved pairing configurations , they are exceedingly rare , with conserved miRNA families in mammals and nematodes each averaging <1 preferentially conserved 3′-supplementary site ( Friedman et al . , 2009 ) . Other relatively rare , yet effective sites include centered sites , which have 11–12 contiguous Watson–Crick pairs to the center of the miRNA ( Shin et al . , 2010 ) , and cleavage sites , which have the very extensive pairing required for Argonaute-catalyzed slicing of the mRNA ( Yekta et al . , 2004; Davis et al . , 2005; Karginov et al . , 2010; Shin et al . , 2010 ) . The existence of additional , still-to-be-characterized types of non-canonical sites is suggested by the large number of mRNA regions that crosslink to the silencing complex in vivo yet lack known site types matching the cognate miRNA ( Chi et al . , 2012; Loeb et al . , 2012; Helwak et al . , 2013; Khorshid et al . , 2013; Grosswendt et al . , 2014 ) . With the prediction of hundreds of conserved targets for most mammalian miRNAs ( and even more nonconserved targets ) , knowing which targets are expected to be most responsive to each miRNA provides important information for both large-scale network analyses and detailed experimental follow-up . As previously mentioned , the type of site ( e . g . , whether the site is an 8mer or a 7mer-A1 ) strongly influences the efficacy of repression . The number of sites also influences efficacy , with each additional site typically acting independently to impart additional repression ( Grimson et al . , 2007; Nielsen et al . , 2007 ) , although sites between 8–40 nt of each other tend to act cooperatively , and those < 8 nt of each other act competitively ( Grimson et al . , 2007 ) . Additional features of site context help explain why a given site ( e . g . , a 7mer-m8 site to miR-1 ) can be more effective in one 3′ UTR than it is in another . These features include the positioning of the site outside of the path of the ribosome ( which includes the first 15 nt of the 3′ UTR [Grimson et al . , 2007] ) and the positioning of the site within 3′-UTR segments that are more accessible to the silencing complex , as measured by either high local AU content ( Grimson et al . , 2007; Nielsen et al . , 2007 ) , high AU content of the entire 3′ UTR ( Robins and Press , 2005; Hausser et al . , 2009 ) , shorter distance from a 3′-UTR terminus ( Gaidatzis et al . , 2007; Grimson et al . , 2007; Majoros and Ohler , 2007 ) , shorter 3′-UTR length ( Hausser et al . , 2009; Betel et al . , 2010; Wen et al . , 2011; Reczko et al . , 2012 ) , or less stable predicted competing secondary structure ( Robins et al . , 2005; Ameres et al . , 2007; Kertesz et al . , 2007; Long et al . , 2007; Tafer et al . , 2008 ) . Conserved sites are also more effective , in part because they tend to reside in more favorable contexts ( Grimson et al . , 2007; Nielsen et al . , 2007 ) . Features of the miRNA can also influence site efficacy , with sites being more effective if the miRNA has lower target-site abundance ( TA ) within the transcriptome ( Arvey et al . , 2010; Garcia et al . , 2011 ) and stronger predicted seed-pairing stability ( SPS ) ( Garcia et al . , 2011 ) . Multiple features can be considered together to build quantitative models of targeting efficacy ( Grimson et al . , 2007; Nielsen et al . , 2007; Wang and El Naqa , 2008; Betel et al . , 2010; Liu et al . , 2010; Garcia et al . , 2011; Wen et al . , 2011; Reczko et al . , 2012; Vejnar and Zdobnov , 2012; Marin et al . , 2013; Gumienny and Zavolan , 2015 ) . Our recent model , called the context-plus ( context+ ) model , considers the features of our original context scores ( i . e . , site type , 3′-supplementary pairing , local AU content , and distance from the closest 3′-UTR end [Grimson et al . , 2007] ) , plus two miRNA features ( TA and SPS [Garcia et al . , 2011] ) . Although the context+ model was trained using multiple regression on 74 high-throughput datasets , the features used to distinguish effective sites ( the three features of the original context scores ) were identified using only 11 datasets , implying that additional features might be identified through analysis of the additional datasets . Here , we examined the function of non-canonical binding sites identified in recent studies and found that mRNAs with these sites are not more repressed than mRNAs without sites , despite compelling evidence that many of these noncanocial sites bind the silencing complex in vivo . This finding justified a focus on the statistical modeling of canonical , seed-matched sites within 3′ UTRs , which mediate the vast majority of repression that can be predicted with current methods . To this end , we pre-processed the 74 datasets to minimize confounding biases and then used stepwise regression to identify the most informative features from a large set of potential targeting features . This approach unbiasedly selected 14 features , which were combined to develop the context++ model of miRNA targeting efficacy . The context++ model was more predictive than any published model and at least as predictive as the most informative in vivo crosslinking approaches . As the engine powering the latest version of TargetScan ( v7 . 0; targetscan . org ) , this model provides a valuable resource for placing the miRNAs of human , mouse , zebrafish , and other vertebrate species into their respective gene-regulatory networks . Several high-throughput crosslinking-immunoprecipitation ( CLIP ) approaches have been applied to identify sites that bind Argonaute in vivo ( Chi et al . , 2009; Hafner et al . , 2010; Helwak et al . , 2013; Grosswendt et al . , 2014 ) . These experiments all observe significant enrichment for cognate seed-matched sites in the vicinity of the crosslinks , which validates their ability to detect authentic sites . Despite this enrichment , some crosslinks do not correspond to canonical sites to the relevant miRNAs , raising the prospect that these results might reveal novel types of non-canonical binding that could mediate repression . Indeed , five studies have reported crosslinking to non-canonical binding sites proposed to mediate repression ( Chi et al . , 2012; Loeb et al . , 2012; Helwak et al . , 2013; Khorshid et al . , 2013; Grosswendt et al . , 2014 ) . In addition , another biochemical study has reported the identification of non-canonical sites without using any crosslinking ( Tan et al . , 2014 ) . Reasoning that these experimental datasets might provide a resource for defining of novel types of sites to be used in target prediction , we re-examined the functionality of these sites in mediating target mRNA repression . We first examined the efficacy of ‘nucleation-bulge’ sites ( Chi et al . , 2012 ) , which were identified from analysis of differential CLIP ( dCLIP ) results reporting the clusters that appear in the presence of miR-124 ( Chi et al . , 2009 ) . Nucleation-bulge sites consist of 8 nt motifs paired to positions 2–8 of their cognate miRNA seed , with the nucleotide opposing position 6 protruding as a bulge but sharing Watson-Crick complementarity to miRNA position 6 . Meta-analyses of miRNA and small-RNA transfection datasets revealed significant repression of mRNAs with the canonical site types but found no evidence for repression of mRNAs that contain nucleation-bulge sites but lack perfectly paired seed-matched sites in their 3′ UTRs ( Figure 1—figure supplement 1A , B ) . Reasoning that the nucleation-bulge site might be only marginally effective , we examined the early zebrafish embryo with and without Dicer , analyzing the targeting by miR-430 , the most highly expressed miRNA of the early embryo . Even in this system , one of the most sensitive systems for detecting the effects of targeting ( where a robust repression is observed for mRNAs with only a single 6mer or offset-6mer sites to miR-430 ) , we observed no evidence for repression of mRNAs with nucleation-bulge sites to miR-430 ( Figure 1A , Figure 1—figure supplement 1C , and Figure 1—figure supplement 4A ) . Because the nucleation-bulge sites were originally identified and characterized as sites to miR-124 , we next tried focusing on only miR-124–mediated repression . However , even in this more limited context , the mRNAs with nucleation-bulge sites were no more repressed than mRNAs without sites ( Figure 1—figure supplement 1D–F ) . 10 . 7554/eLife . 05005 . 003Figure 1 . Inefficacy of recently reported non-canonical sites . ( A ) Response of mRNAs to the loss of miRNAs , comparing mRNAs that contain either a canonical or nucleation-bulge site to miR-430 to those that do not contain a miR-430 site . Plotted are cumulative distributions of mRNA fold changes observed when comparing embryos that lack miRNAs ( MZDicer ) to those that have miRNAs ( WT ) , focusing on mRNAs possessing a single site of the indicated type in their 3′ UTR . Similarity of site-containing distributions to the no-site distribution was tested ( one-sided Kolmogorov–Smirnov [K–S] test , P values ) ; the number of mRNAs analyzed in each category is listed in parentheses . See also Figure 1—figure supplement 1C and Figure 1—figure supplement 4A . ( B and C ) Response of mRNAs to the loss of miR-155 , focusing on mRNAs that contain either a single canonical or ≥1 CLIP-supported non-canonical site to miR-155 . These panels are as in ( A ) , but compare fold changes for mRNAs with the indicated site type following genetic ablation of mir-155 in either T cells ( B ) or Th1 cells ( C ) . See also Figure 1—figure supplement 2 . ( D and E ) Response of mRNAs to the knockdown of miR-92a , focusing on mRNAs that contain either a single canonical or ≥1 CLASH-identified non-canonical site to miR-92a . These panels are as in ( A ) , except CLASH-supported non-canonical sites were the same as those defined previously ( Helwak et al . , 2013 ) and thus were permitted to reside in any region of the mature mRNA , and these panels compare fold changes for mRNAs with the indicated site type following either knockdown of miR-92a ( D ) or combined knockdown of miR-92a and 24 other miRNAs ( E ) in HEK293 cells . See also Figure 1—figure supplement 3A , B . ( F ) As in ( D ) , but focusing on mRNAs that contain ≥1 chimera-identified site . See also Figure 1—figure supplement 3C–E and Figure 1—figure supplement 4B . ( G ) Response of mRNAs to the transfection of 16 miRNAs , focusing on mRNAs that contain either a canonical or MIRZA-predicted non-canonical site . This panel is as in ( A ) , but compares the fold changes for mRNAs with the indicated site type after introducing miRNAs , aggregating results from 16 individual transfection datasets . Fold changes are plotted for the top 100 non-canonical predictions for each of 16 miRNAs compiled either before ( MIRZA , top 100 ) or after ( MIRZA , no 6mers ) removing mRNAs containing 6mer or offset-6mer 3′-UTR sites . ( H ) Response of mRNAs to a transfection of miR-522 , focusing on mRNAs that contain either a single canonical or ≥1 IMPACT-seq–supported non-canonical site to miR-522 . These panels are as in ( A ) , except IMPACT-seq–supported non-canonical sites were the same as those defined previously ( Tan et al . , 2014 ) and thus were permitted in any region of the mature mRNA . ( I ) Response of ribosomes to the loss of miR-155 , focusing on mRNAs that contain either a single canonical or ≥1 CLIP-supported non-canonical site to miR-155 . This panel is as in ( B and C ) but compares the response of mRNAs using ribosome-footprint profiling ( Eichhorn et al . , 2014 ) after genetic ablation of mir-155 in B cells . Ribosome-footprint profiling captures changes in both mRNA stability and translational efficiency through the high-throughput sequencing of ribosome-protected mRNA fragments ( RPFs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 00310 . 7554/eLife . 05005 . 004Figure 1—figure supplement 1 . Inefficacy of nucleation-bulge sites . ( A and B ) These panels are as in Figure 1A but compare the response of cognate site-containing mRNAs in a compendium of either 11 miRNA transfection datasets ( A ) or 74 sRNA transfection datasets ( B ) . The datasets were pre-processed ( Figure 3 ) and are provided in Supplementary file 1 . ( C ) This panel is as in Figure 1A but compares the response of mRNAs in MZDicer embryos in which miR-430 has been injected . ( D–F ) These panels are as in Figure 1A but compare the response of mRNAs with the indicated miR-124 site types after transfecting miR-124 into either HEK293 cells ( D ) , HeLa cells ( E ) , or Huh7 cells ( F ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 00410 . 7554/eLife . 05005 . 005Figure 1—figure supplement 2 . Inefficacy of CLIP-supported non-canonical miR-155 sites . ( A and B ) These panels are as in Figure 1B but compare the response of mRNAs after genetic ablation of miR-155 in Type 2 helper T cells ( Th2 , A ) or B cells ( B ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 00510 . 7554/eLife . 05005 . 006Figure 1—figure supplement 3 . Inefficacy of CLASH- and chimera-supported non-canonical sites . ( A–D ) These panels are as in Figure 1D but compare the response of mRNAs with sites cognate to any one of four miRNA families ( miR-15/16 , miR-19 , miR-17/20/93/106 , or miR-25/92 ) , for either all CLASH-supported targets ( A ) , mRNAs with CLASH-supported 3′-UTR sites ( B ) , all chimera-supported targets ( C ) , or mRNAs with chimera-supported 3′-UTR sites ( D ) . These four miRNA families were chosen because their predicted targets were the most responsive to knockdown of the 25 miRNAs . p values reflect the median p value ( as evaluated by a K–S test ) across 100 trials in which a no-site control cohort with matched 3′-UTR lengths was chosen for each site-containing distribution . Length-matched no-site controls were required for this analysis because longer 3′ UTRs had a greater chance of containing additional sites to at least one of the many miRNAs that were knocked down , and thus had a greater chance of being derepressed as a result of interactions otherwise not considered in the analysis . To populate each control cohort , 500 different no-site mRNAs were chosen , considering the 3′-UTR length of each site-containing mRNA and selecting ( without replacement ) control mRNAs from among the 10 no-site mRNAs with the most similar 3′-UTR lengths . Shown is the response of a control cohort for mRNAs containing non-canonical sites . mRNAs with 3′ UTRs >2000 nt were excluded from the analysis because so many of the 3′ UTRs >2000 nt had a site to at least one of the four miRNA families , making it impossible to select appropriate length-matched controls . ( E ) This panel is as in Figure 1F but compares the response of mRNAs with the indicated miR-302 site types after knocking down miR-302/367 in hESCs . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 00610 . 7554/eLife . 05005 . 007Figure 1—figure supplement 4 . Inefficacy of non-canonical sites in mediating translational repression . ( A ) This panel is as in Figure 1A but compares the response of mRNAs using ribosome footprint profiling ( Bazzini et al . , 2012 ) , which captures changes in both mRNA stability and translational efficiency through the high-throughput sequencing of ribosome-protected mRNA fragments ( RPFs ) . ( B ) This panel is as in ( Figure 1I ) but compares protein fold changes for chimera-supported targets , as evaluated by pulsed SILAC ( Selbach et al . , 2008 ) after transfection of miR-155 in HeLa cells . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 00710 . 7554/eLife . 05005 . 008Figure 1—figure supplement 5 . Re-evaluating conservation of chimera-supported non-canonical sites . ( A ) Conservation of chimera-supported non-canonical sites detected in an analysis modeled after that of Grosswendt et al . ( 2014 ) but modified to control for background conservation . Plotted for the indicated miRNAs is the average conservation of chimera-supported non-canonical sites , as measured by branch-length score ( BLS ) , compared to the average conservation of 100 equally sized cohorts of controls; error bars , standard deviation of cohort averages; ** , p < 0 . 01; * , p < 0 . 05 , one-sided Z test . We considered chimera-supported non-canonical sites that mapped within 3′ UTRs and contained a single mismatch to the 6 nt seed of the miRNA . This set of sites mirrored that analyzed previously ( Grosswendt et al . , 2014 ) , and excluded offset 6mers , which as a class was already known to mediate repression and exhibit preferential conservation ( Friedman et al . , 2009 ) . Cohorts of control sites were generated such that for each chimera-supported site , each control cohort contained a single example of the identical 6 nt motif that was present in the indicated region ( either an AGO cluster or 3′ UTR ) but not supported by chimeric reads . To control for local background conservation and thereby avoid treating sites within slowly evolving 3′ UTRs the same as those within rapidly evolving 3′ UTRs , we used the binning procedure developed for calculating PCT scores ( Friedman et al . , 2009 ) ; 3′ UTRs were partitioned into 10 conservation bins ( based on the median BLS of the nucleotides of the human sequence ) , and control sites were randomly selected ( with replacement ) from 3′ UTRs in the same bin as the actual site . Control AGO clusters were collected as was done previously ( Grosswendt et al . , 2014 ) , using genome-wide data downloaded from clipz . unibas . ch and derived from multiple AGO PAR-CLIP experiments performed in HEK293 cells ( Kishore et al . , 2011 ) . The union of AGO clusters for all experiments was computed and filtered for overlap with Ensembl-annotated 3′ UTRs , using the ‘merge’ and ‘intersectBED’ utilities , respectively , found in BEDTools v2 . 20 . 1 ( parameter ‘-s’ ) ( Quinlan and Hall , 2010 ) . ( B ) Attribution of the conservation signal to the confounding effects of conserved regions . Considered are 1443 non-canonical chimera-supported sites selected as in ( A ) but including sites of all miRNA families . For each chimera-supported site , a z score was generated using the distribution of BLSs for 100 control sites chosen as in panel ( A ) from either AGO clusters or 3′ UTRs , as indicated . Each z score reflected how the conservation of the actual site differed from that of its controls . Compared are cumulative distributions of the z scores for sites of broadly conserved miRNAs and those of less conserved miRNAs , using the previously defined sets of broadly and less conserved miRNAs ( Friedman et al . , 2009 ) . If the chimera-supported non-canonical sites were preferentially conserved because of their function in mediating repression , then sites of broadly conserved miRNAs would be expected to have a right-shifted distribution compared to sites of less conserved miRNAs . However , no significant difference was discerned between each pair of z-score distributions . The remainder of this legend outlines the rationale for the analysis of this panel . One way to reconcile the conservation signal observed in panel ( A ) with our conclusion that a large majority if not all of these sites bind miRNA but do not mediate repression is to consider the potentially confounding biochemical properties of conserved regions , which are illustrated by the observation that artificial siRNAs preferentially target sites that are evolutionarily conserved over those that are not ( Nielsen et al . , 2007 ) . Because these siRNAs are not natural ( and do not share a seed with conserved miRNAs ) the evolutionary conservation of these preferred sites could not have arisen because they function to mediate sRNA-guided repression . Instead , some other function of these 3′-UTR regions , such as greater accessibility to RNA-binding factors , must explain their preferential conservation and also endow them with properties that favor sRNA binding ( Nielsen et al . , 2007 ) . To examine whether confounding properties of conserved 3′-UTR regions might similarly explain the elevated conservation of chimera-supported sites , we compared the z scores for sites bound by broadly conserved miRNAs ( miRNAs in families conserved beyond mammals , as listed in TargetScan7 ) with those bound by less conserved miRNAs . MicroRNAs conserved among mammals but not more broadly were grouped with the less conserved miRNAs because canonical 6mer and 7mer sites to these miRNAs have no conservation signal above background , presumably because these miRNAs have not been present long enough for the number of preferentially conserved 6mer and 7mer sites to rise above the background ( Friedman et al . , 2009 ) ; we reasoned that the same would be true of non-canonical sites , to the extent that any are preferentially conserved . If the conservation signal observed in panel ( A ) were related to miRNA binding , we would have expected a difference between the scores for the sites of broadly and less conserved miRNAs . The lack of a significant difference supports the idea that chimera-supported non-canonical sites tend to be conserved for the same reason that functional sites to artificial siRNAs tend to be conserved . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 008 Another study examined the response of 32 mRNAs that lack canonical miR-155 sites yet crosslink to Argonaute in wild-type T cells but not T cells isolated from miR-155 knockout mice ( Loeb et al . , 2012 ) . As previously observed , we found that the levels of these mRNAs tended to increase in T cells lacking miR-155 ( Figure 1B ) . However , a closer look at the distribution of mRNA fold changes between wild-type and knockout cells revealed a pattern not normally observed for mRNAs with a functional site type . As illustrated for the mRNAs with canonical sites ( including those supported by CLIP ) , when a miRNA is knocked out , the cumulative distribution of fold changes for mRNAs with functional site types diverges most from the no-site distribution at the top of the curve , which represents the most strongly derepressed mRNAs ( Figure 1B ) . However , for the mRNAs harboring non-canonical miR-155 sites , the distribution of fold changes converged with the no-site distribution at the top of the curve ( Figure 1B ) , raising doubt as to whether non-canonical binding of these mRNAs mediates repression . To investigate these mRNAs further , we examined their response to the miR-155 loss in helper T cell subtypes 1 and 2 ( Th1 and Th2 , respectively ) and B cells , which are other lymphocytic cells in which significant derepression of miR-155 targets is observed in cells lacking miR-155 ( Rodriguez et al . , 2007; Eichhorn et al . , 2014 ) . In contrast to mRNAs with canonical sites , the mRNAs with non-canonical sites showed no evidence of derepression in the knockout cells of each of these cell types , which reinforced the conclusion that non-canonical binding of miR-155 does not lead to repression of these mRNAs ( Figure 1C and Figure 1—figure supplement 2 ) . We next probed the functionality of non-canonical interactions identified by CLASH ( crosslinking , ligation , and sequencing of hybrids ) , a high-throughput technique that generates miRNA–mRNA chimeras , which each identify a miRNA and the mRNA region that it binds ( Helwak et al . , 2013 ) . As previously observed , mRNAs with CLASH-identified non-canonical interactions involving miR-92 tended to be slightly up-regulated upon knockdown of miR-92 in HEK293 cells ( Figure 1D ) . However , a closer look at the mRNA fold-change distributions again revealed a pattern not typically observed for mRNAs with a functional site type , with convergence with the no-site distribution in the region expected to be most divergent . Therefore , we examined a second dataset monitoring mRNA changes after knocking down miR-92 and other miRNAs in HEK293 cells ( Hafner et al . , 2010 ) . As reported recently ( Wang , 2014 ) , the slight up-regulation observed for mRNAs with CLASH-identified non-canonical interactions in the original dataset was not reproducible in the second dataset ( Figure 1E ) . Moreover , mRNAs with non-canonical interactions to other miRNAs showed no sign of derepression when the cognate miRNAs were knocked down ( Figure 1—figure supplement 3A ) . To mirror the original analyses of CLASH-identified interactions ( Helwak et al . , 2013 ) , our analyses included sites located in any region of the mature mRNA ( Figure 1D , E and Figure 1—figure supplement 3A ) . No significant difference from the no-site control distribution was observed when restricting our analysis to mRNAs with CLASH-identified non-canonical sites in their 3′ UTRs ( Figure 1—figure supplement 3B ) . Many miRNA–mRNA chimeras can also be found in standard AGO CLIP datasets , presumably generated by an endogenous ligase acting in cell lysates during workup ( Grosswendt et al . , 2014 ) . Global experiments examining function of these interactions group the mRNAs with non-canonical interactions together with those with canonical interactions ( Grosswendt et al . , 2014 ) , and thus the signal for function might arise from only canonical interactions . Indeed , when we re-examined the response of these mRNAs to miRNA knockdown , those with chimera-identified canonical sites tended to be derepressed , whereas those with only chimera-identified non-canonical sites did not ( Figure 1F and Figure 1—figure supplement 3C–E ) . Although at first glance this finding might seem at odds with the elevated evolutionary conservation of chimera-identified non-canonical sites ( Grosswendt et al . , 2014 ) , we found that this conservation signal was not smaller for the sites of less conserved miRNAs and therefore was not indicative of functional miRNA binding ( Figure 1—figure supplement 5 ) . Instead , the reported conservation signal might occur for the same reason that artificial siRNAs tend to target conserved regions of 3′ UTRs ( Nielsen et al . , 2007 ) . Next , we evaluated the response of non-canonical sites modeled by MIRZA , an algorithm that utilizes CLIP data in conjunction with a biophysical model to predict target sites ( Khorshid et al . , 2013 ) . As noted by others ( Majoros et al . , 2013 ) , the definition of non-canonical MIRZA sites was more expansive than that used elsewhere and did not exclude sites with canonical 6mer or offset-6mer seed matches . Indeed , when focusing on only targets without 6mer or offset-6mer seed matches , the top 100 non-canonical MIRZA targets showed no sign of efficacy ( Figure 1G ) . Finally , we examined non-canonical clusters identified by IMPACT-seq ( identification of miRNA-responsive elements by pull-down and alignment of captive transcripts—sequencing ) , a method that sequences mRNA fragments that co-purify with a biotinylated miRNA without crosslinking ( Tan et al . , 2014 ) . Although the mRNAs with an IMPACT-seq–supported canonical site were down-regulated upon the transfection of the cognate miRNA , those with an IMPACT-seq–supported non-canonical site responded no differently than mRNAs lacking a site ( Figure 1H ) . Collectively , the novel non-canonical sites recently identified in high-throughput CLIP and other biochemical studies imparted no detectable repression when monitoring mRNA changes . However , monitoring of only mRNA changes leaves open the possibility that these sites might still mediate translational repression . To address this possibility , we examined ribosome-profiling and proteomic datasets , which capture repression also occurring at the level of translation , and again we found that the CLIP-identified non-canonical sites imparted no detectable repression ( Figure 1I and Figure 1—figure supplement 4 ) . All of our analyses of experimentally identified non-canonical sites examined the ability of the sites to act in mRNAs that had no seed-matched site to the same miRNA in their 3′ UTRs . Any non-canonical site found in a 3′ UTR that also had a seed-matched site to the same miRNA was not considered because any response could be attributed to the canonical site . At first glance , excluding these co-occurring sites might seem to allow for the possibility that the experimentally identified non-canonical sites could contribute to repression when in the same 3′ UTR as a canonical site , even though they are ineffective in 3′ UTRs without canonical sites . However , in mammals , canonical sites to the same miRNA typically act independently ( Grimson et al . , 2007; Nielsen et al . , 2007 ) , and we have no reason to think that non-canonical sites would behave differently . More importantly , although the non-canonical sites examined were in mRNAs that had no seed-matched 3′-UTR site to the same miRNA , most were in mRNAs that had seed-matched 3′-UTR sites to other miRNAs that were highly expressed in the cells . Therefore , even if the non-canonical sites could only function when coupled to a canonical site , we still would have observed a signal for their function in our analyses . The inefficacy of recently reported non-canonical sites was surprising when considering evidence that the dCLIP clusters without cognate seed matches are nonetheless enriched for imperfect pairing to the miRNA , which would not be expected if those clusters were merely non-specific background ( Chi et al . , 2012; Loeb et al . , 2012 ) . Indeed , our analysis of motifs within the dCLIP clusters for miR-124 and miR-155 confirmed that those without a canonical site to the miRNA were enriched for miRNA pairing ( Figure 2A ) . Although one of the motifs identified within CLIP clusters that appeared after transfection of miR-124 into HeLa cells yet lacked a canonical miR-124 site did not match the miRNA ( Figure 2—figure supplement 1C ) , the top motif , as identified by MEME ( Bailey and Elkan , 1994 ) , had striking complementarity to the miR-124 seed region ( Figure 2A ) . This human miR-124 non-canonical motif matched the ‘nucleation-bulge’ motif originally found for miR-124 in the mouse brain ( Chi et al . , 2012 ) . Although the top motif identified within the subset of miR-155 dCLIP clusters that lacked a canonical site to miR-155 was not identified with confidence , it had only a single mismatch to the miR-155 seed , which would not have been expected for a motif identified by chance . 10 . 7554/eLife . 05005 . 009Figure 2 . Confirmation of experimentally identified non-canonical miRNA binding sites . ( A ) Sequence logos corresponding to motifs enriched in dCLIP clusters that either appear following transfection of miR-124 into HeLa cells ( Chi et al . , 2009 ) ( top ) or disappear following knockout of miR-155 in T cells ( Loeb et al . , 2012 ) ( bottom ) . Shown to the right of each logo is its E-value among clusters lacking a seed-matched or offset-6mer canonical site and the fraction of these clusters that matched the logo . Shown below each logo are the complementary regions of the cognate miRNA family , highlighting nucleotides 2–8 in capital letters . ( B ) Position of the top-ranked motif corresponding to non-canonical sites enriched in CLASH ( Helwak et al . , 2013 ) ( left ) or chimera ( Grosswendt et al . , 2014 ) ( right ) data for each human miRNA family supported by at least 50 interactions without a seed-matched or offset-6mer canonical site . For each family the most enriched logo was aligned to the reverse complement of the miRNA . In cases in which a logo mapped to multiple positions along the miRNA , the positions with the best and second best scores are indicated ( red and blue , respectively ) . ( C ) Sequence logos of motifs enriched in chimera interactions that lack canonical sites . As in ( A ) , but displaying sequence logos identified in the chimera data of panel ( B ) for a sample of nine human miRNAs . Logos identified for the other human miRNAs are also provided ( Figure 2—figure supplement 1B ) . A nucleotide that differs between miRNA family members is indicated as a black ‘n’ . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 00910 . 7554/eLife . 05005 . 010Figure 2—figure supplement 1 . Comparison of CLASH and chimera data and identification of motifs enriched in human chimera interactions that lack canonical sites . ( A ) Comparison of CLASH ( left ) and chimera ( right ) reads from human cells , showing the proportion possessing a canonical site ( blue ) and overlapping 3′ UTRs ( red ) . In total , 18 , 514 CLASH and 10 , 567 chimera interactions were analyzed . ( B ) Sequence logos of motifs enriched in chimera interactions that lack canonical sites . This panel is as in Figure 2C but displays the remaining motifs identified from the chimera data analyzed in Figure 2B . In cases of alignment ambiguity , both alignments are shown below the logo . For some miRNA families , multiple motifs were significantly enriched ( E ≤ 0 . 001 ) and are shown separately . Significantly enriched motifs ( or a top-ranked motif matching the miRNA ) were not found for miR-21 , and miR-3168 was excluded from the analysis due to poor support for its authenticity as a miRNA . ( C ) Sequence logos of motifs that do not match the cognate miRNA but are nonetheless enriched in miR-124 dCLIP ( Chi et al . , 2009 ) and miR-522 IMPACT-seq ( Tan et al . , 2014 ) clusters that lack canonical sites to the miRNA . The miR-124 logo was nearly identical to a non-specific motif previously identified as enriched in CLIP data from the mouse brain ( Chi et al . , 2012 ) . The miR-522 logo was found instead of the previously reported miRNA-matching logo ( Tan et al . , 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 01010 . 7554/eLife . 05005 . 011Figure 2—figure supplement 2 . Identification of motifs enriched in mouse and nematode chimera interactions that lack canonical sites . ( A ) Sequence logos of motifs enriched in M . musculus chimera interactions that lack canonical sites; otherwise as in Figure 2C . Significantly enriched motifs ( or a top-ranked motif matching the miRNA ) were not found for let-7 and miR-142-3p . ( B ) Sequence logos of motifs enriched in C . elegans chimera interactions that lack canonical sites; otherwise as in Figure 2C . Significantly enriched motifs ( or a top-ranked motif matching the miRNA ) were not found for miR-1 . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 011 Previous analysis of CLASH-identified interactions shows that the top MEME-identified motifs usually pair to the miRNA , although for many miRNAs this pairing falls outside of the seed region ( Helwak et al . , 2013 ) . Repeating this analysis , but focusing on only interactions without canonical sites , confirmed this result ( Figure 2B ) . Applying this type of analysis to non-canonical interactions identified from miRNA–mRNA chimeras in standard AGO CLIP datasets confirmed that these interactions are also enriched for pairing to the miRNA ( Grosswendt et al . , 2014 ) . As previously shown ( Grosswendt et al . , 2014 ) , these interactions were more specific to the seed region than were the CLASH-identified interactions ( Figure 2B ) . Comparison of all the chimera data with all the CLASH data showed that a higher fraction of the chimeras captured canonical interactions and that a higher fraction captured interactions within 3′ UTRs ( Figure 2—figure supplement 1A ) . These results , implying that the chimera approach is more effective than CLASH at capturing functional sites that mediate repression , motivated a closer look at the chimera-identified interactions that lacked a canonical site , despite our finding that these interactions do not mediate repression . In the human and nematode datasets ( and less so in the mouse dataset ) , these interactions were enriched for motifs that corresponded to non-canonical sites that paired to the miRNA seed region ( Figure 2B–C , Figure 2—figure supplement 1B , and Figure 2—figure supplement 2 ) . Inspection of these motifs revealed that the most enriched nucleotides typically preserved Watson–Crick pairing in a core 4–5 nts within the seed region , with tolerance to mismatches or G:U wobbles observed at varied positions , depending on the miRNA , potentially reflecting seed-specific structural or energetic features , or perhaps context-dependent biases in crosslinking or ligation . Motifs for only a few miRNAs had a bulged nucleotide , and if a bulge was observed it was in the mRNA strand and not in the miRNA strand , as expected if the Argonaute protein imposed geometric constraints in the seed of the miRNA . The miR-124 nucleation-bulge site was enriched in mouse chimera interactions ( Figure 2—figure supplement 2A ) , as it had been in the human and mouse dCLIP clusters ( Figure 2A ) ( Chi et al . , 2012 ) . However , despite identification of this miR-124 interaction in datasets from two methods and two species , this style of bulged pairing was not detected for any other miRNA . Interestingly , for all other cases in which a bulge in the recognition motif was observed ( human miR-33 and miR-374 , and C . elegans miR-50 and miR-58 ) , the bulge was between the nucleotides that paired to miRNA nucleotides 4 and 5 ( Figure 2—figure supplement 1B and Figure 2—figure supplement 2B ) . A bulge is observed between the analogous nucleotides of validated targets of Arabidopsis miR398 ( Jones-Rhoades and Bartel , 2004 ) , whereas single-nucleotide bulges between other seed-pairing positions have not been reported in other validated plant targets . A bulge between these nucleotides is also observed in the first let-7 site in the C . elegans lin-41 3′ UTR , one of the archetypal 3′-compensatory sites ( Reinhart et al . , 2000; Bartel , 2009 ) . Taken together , these observations suggest that the most tolerated bulge in miRNA seed pairing is between the target nucleotides that pair to miRNA nucleotides 4 and 5 . Some motifs , particularly the more degenerate ones , were found in most of the interactions , whereas other motifs were found in only a small minority ( Figure 2C and Figure 2—figure supplement 1B ) . We suspect that many of the interactions lacking the top-scoring motifs also involve non-canonical binding sites , some of which might function through degenerate versions of the motif that happened to have scored highest in the MEME analysis . Nonetheless , some interactions or CLIP clusters lacking the top-scoring motifs might represent background ( Friedersdorf and Keene , 2014 ) , and indeed a few with the motif or even with a canonical site might represent background . In sum , our analyses of the CLIP datasets confirmed that many of the CLIP clusters and CLASH/chimera interactions lacking a seed match nonetheless capture authentic miRNA-binding sites—otherwise the top enriched motifs would not pair so often to the cognate miRNA . Despite this ability to bind the miRNA in vivo and to function in the sense that they contribute to cellular TA ( Denzler et al . , 2014 ) , we classify the CLIP-identified non-canonical sites as non-functional with respect to repression because they showed no sign of mediating repression and no signal for miRNA-dependent conservation ( Figure 1 and Figure 1—figure supplements 1–5 ) . Thus , the only known non-canonical site types that mediate repression are the 3′-supplementary , centered , and cleavage site types , which together comprise <1% of the effective sites that currently can be predicted ( Friedman et al . , 2009; Shin et al . , 2010 ) . Although we cannot exclude the possibility that additional types of functional non-canonical sites might exist but have not yet been characterized to the point that they can be used for miRNA target prediction ( Lal et al . , 2009 ) , our analysis of the CLIP results justified a focus on the abundant site types that are predictive of targeting and are at least marginally functional , that is , the canonical seed-matched sites , including 6mer and offset-6mer sites . To identify features involved in mammalian miRNA targeting , we analyzed the results of microarray datasets reporting the mRNA changes after transfecting either a miRNA or siRNA ( together referred to as small RNAs , abbreviated as sRNAs ) into HeLa cells . From the published datasets , we used the set of 74 experiments that had previously been selected because each ( 1 ) had a clear signal for sRNA-based repression , ( 2 ) was acquired using the same Agilent array platform , and ( 3 ) reported on the effects of a unique seed sequence ( Garcia et al . , 2011 ) . Despite the differences among the 74 transfected sRNAs , mRNA fold changes of some arrays were highly correlated with those of others , which indicated that sRNA-independent effects dominated ( Figure 3A ) . When all 74 datasets were compared against each other , those from either the same group of experiments ( Anderson et al . , 2008 ) or the same transfection protocol ( Jackson et al . , 2006a , 2006b; Grimson et al . , 2007 ) tended to cluster strongly together based on their common transcriptome-wide responses to different transfected sRNAs ( Figure 3B ) , indicating the likely presence of batch effects ( Leek et al . , 2010 ) that could obscure detection of features associated with miRNA targeting . 10 . 7554/eLife . 05005 . 012Figure 3 . Pre-processing the microarray datasets to minimize nonspecific effects and technical biases . ( A ) Example of the correlated response of mRNAs after transfecting two unrelated sRNAs ( sRNA 1 and 2 , respectively ) . Results for mRNAs containing at least one canonical 7–8 nt 3′-UTR site for either sRNA 1 , sRNA 2 , or both sRNAs are highlighted in red , blue , and green , respectively . Values for mRNAs without such sites are in grey . All mRNAs were used to calculate the Spearman correlation ( rs ) . ( B ) Correlated responses observed in a compendium of 74 transfection experiments from six studies ( colored as indicted in the publications list ) . For each pair of experiments , the rs value was calculated as in panel ( A ) , colored as indicated in the key , and used for hierarchical clustering . ( C ) Study-dependent relationships between the responses of mRNAs to the transfected sRNA and either 3′-UTR length or 3′-UTR AU content , focusing on mRNAs without a canonical 7–8 nt 3′-UTR site to the sRNA . Boxplots indicate the median rs ( bar ) , 25th and 75th percentiles ( box ) , and the minimum of either 1 . 5 times the interquartile range or the most extreme data point ( whiskers ) , with the width of the box proportional to the number of datasets used from each study . The studies are colored as in panel ( B ) , abbreviating the first author and year . ( D ) Reduced correlation between the responses of mRNAs to unrelated sRNAs after applying the PLSR technique . This panel is as in ( A ) but plots the normalized mRNA fold changes . ( E ) Reduced correlations in results of the compendium experiments after applying the PLSR technique . This panel is as in ( B ) but plots the correlations after normalizing the mRNA fold changes . ( F ) Reduced study-dependent relationships between mRNA responses and either 3′-UTR length or 3′-UTR AU content . This panel is as in ( C ) but plots the correlations after normalizing the mRNA fold changes . ( G and H ) Cumulative distributions of fold changes for mRNAs containing at least one canonical 7–8 nt 3′-UTR site or no site either before normalization ( raw ) or after normalization ( normalized ) . Panel ( G ) plots the results from experiments shown in ( A ) and ( D ) , and ( H ) plots results from all 74 datasets . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 01210 . 7554/eLife . 05005 . 013Figure 3—figure supplement 1 . Reduced biases from derepression of endogenous miRNA targets . ( A ) Pie chart reflecting the relative proportions of reads for the indicated miRNA families observed when sequencing small RNAs from HeLa cells . Relative miRNA levels were quantified as described previously ( Denzler et al . , 2014 ) . ( B and C ) Cumulative distributions of fold changes for mRNAs with at least one canonical 7–8 nt 3′-UTR site to the indicated miRNA family in the compendium of 74 sRNA transfection datasets , either before ( B ) or after ( C ) normalization . p values were computed using a one-sided Wilcoxon rank-sum test , comparing each of the site-containing distributions to the no-site distribution . This test was a more stringent alternative to the K–S test , which led to highly significant p values for very slight differences , due to the large number of mRNAs in each distribution . To account for multiple hypotheses , an appropriate Bonferroni-corrected significance threshold would be p < 0 . 005 , which was not achieved for most comparisons in panel ( C ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 013 A parameter known to confound the accurate measurement of mRNA responses on microarrays is the relative AU content within 3′ UTRs ( Elkon and Agami , 2008 ) . Indeed , when considering mRNAs without a canonical site to the transfected sRNA , we found that 3′-UTR AU content often correlated with mRNA fold changes . Moreover , the extent and direction of the correlation was similar for different datasets from the same publication but differed when comparing to datasets from other publications ( Figure 3C ) . A second parameter that helped explain the correlated sRNA-independent effects for related datasets was 3′-UTR length ( Saito and Satrom , 2012 ) , which exhibited patterns of correlation similar to those observed for 3′-UTR AU content ( Figure 3C ) . Our observation that AU content and 3′-UTR length correlated so differently with global expression changes when comparing results from different publications helps explain why different 3′-UTR features previously seemed to have such variable predictive power in different experimental contexts ( Hausser et al . , 2009; Wen et al . , 2011; Gumienny and Zavolan , 2015 ) . Another phenomenon known to systematically perturb the levels of mRNAs without sites to the transfected sRNA is the derepression of mRNAs with sites for endogenous miRNAs , presumably through competition between the transfected sRNA and the endogenous miRNAs for limiting components of the silencing pathway ( Khan et al . , 2009; Saito and Satrom , 2012 ) . Statistically significant derepression was indeed observed for mRNAs with sites to eight of the 10 miRNA families most frequently sequenced in HeLa cells ( Figure 3—figure supplement 1A , B ) . To correct for biases that were independent of the sequence of the introduced sRNA , we used partial least-squares regression ( PLSR ) to estimate—for each transfection experiment—the component of the transcriptome response that was similar in other highly correlated experiments , and we then subtracted this estimate from the observed response ( Supplementary file 1 ) . Applying our technique to all the mRNAs in each of the 74 datasets largely eliminated the correlations observed between datasets ( Figure 3D–E ) , as well as the correlations observed between mRNA fold changes and either AU content or 3′-UTR length ( Figure 3F ) , which lowered the risk that these effects that are independent of the sRNA sequence would confound subsequent analyses of sRNA targeting efficacy . Moreover , our technique eliminated the signal for derepression of endogenous miRNA targets ( Figure 3—figure supplement 1C ) , suggesting that it did the same for any other biases unrelated to the sequence of the transfected sRNA that have yet to be identified . Reducing these biases substantially reduced the variance in the response for mRNAs without sites to the sRNA , which substantially enhanced the net signal for sRNA-mediated repression of site-containing mRNAs observed in individual arrays ( Figure 3G ) and all arrays in aggregate ( Figure 3H ) . Previous studies of miRNA targeting have relied on 3′-UTR annotations from databases such as RefSeq , without accounting for abundant alternative 3′-UTR isoforms present in the tissue or cell line of interest ( Tian et al . , 2005 ) . The presence of more than one abundant 3′-UTR isoform for a gene would confound interpretation of 3′-UTR-related features , such as 3′-UTR length , or distance from the closest 3′-UTR end ( Nam et al . , 2014 ) . Moreover , the shorter 3′-UTR isoforms might not include some target sites , which would cause these sites to appear ineffective when in fact they are not present ( Sandberg et al . , 2008; Mayr and Bartel , 2009; Lianoglou et al . , 2013; Nam et al . , 2014 ) . To avoid these complications , we examined 3′-UTR isoform quantifications previously generated for HeLa cells ( Nam et al . , 2014 ) using poly ( A ) -position profiling by sequencing ( 3P-seq ) ( Jan et al . , 2011 ) , and developed our model using the dominant mRNA from the subset of genes for which ≥90% of the 3P-seq tags corresponded to a single 3′-UTR isoform . To isolate the effects of single sites , we also used the subset of these mRNAs for which the 3′ UTR possessed a single seed match to the transfected sRNA ( Supplementary file 1 ) . To improve our model of mammalian target-site efficacy , we considered 26 features as potentially informative of efficacy . These included features of the sRNAs , features of the sites ( including their contexts and positions within the mRNAs ) , and features of the mRNAs , many of which had been used or at least considered in previous efforts ( Table 1 ) . 10 . 7554/eLife . 05005 . 014Table 1 . The 26 features considered in the models , highlighting the 14 robustly selected through stepwise regression ( bold ) DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 014FeatureAbbreviationDescriptionFrequency chosen8mer7mer-m87mer-A16mermiRNA 3′-UTR target-site abundanceTA_3UTRNumber of sites in all annotated 3′ UTRs ( Arvey et al . , 2010; Garcia et al . , 2011 ) 100%100%100%100% ORF target-site abundanceTA_ORFNumber of sites in all annotated ORFs ( Garcia et al . , 2011 ) 9 . 4%0 . 7%68 . 1%93 . 4% Predicted seed-pairing stabilitySPSPredicted thermodynamic stability of seed pairing ( Garcia et al . , 2011 ) 100%100%100%100% sRNA position 1sRNA1Identity of nucleotide at position 1 of the sRNA68%100%99 . 7%97 . 7% sRNA position 8sRNA8Identity of nucleotide at position 8 of the sRNA0%0 . 8%100%100%Site Site position 1site1Identity of nucleotide at position 1 of the siteN/A57 . 1%N/A2% Site position 8site8Identity of nucleotide at position 8 of the site0 . 8%95 . 1%99 . 4%100% Site position 9site9Identity of nucleotide at position 9 of the site ( Lewis et al . , 2005; Nielsen et al . , 2007 ) 15 . 4%7 . 1%0 . 9%93 . 7% Site position 10site10Identity of nucleotide at position 10 of the site ( Nielsen et al . , 2007 ) 0 . 1%100%8 . 5%26 . 3% Local AU contentlocal_AUAU content near the site ( Grimson et al . , 2007; Nielsen et al . , 2007 ) 100%100%100%100% 3′ supplementary pairing3P_scoreSupplementary pairing at the miRNA 3′ end ( Grimson et al . , 2007 ) 42 . 5%100%100%100% Distance from stop codondist_stoplog10 ( Distance of site from stop codon ) 62 . 4%10 . 8%8 . 7%25 . 7% Predicted structural accessibilitySAlog10 ( Probability that a 14 nt segment centered on the match to sRNA positions 7 and 8 is unpaired ) 100%100%100%100% Minimum distancemin_distlog10 ( Minimum distance of site from stop codon or polyadenylation site ) ( Gaidatzis et al . , 2007; Grimson et al . , 2007; Majoros and Ohler , 2007 ) 99 . 9%100%87 . 4%100% Probability of conserved targetingPCTProbability of site conservation , controlling for dinucleotide evolution and site context ( Friedman et al . , 2009 ) 100%100%100%20 . 8%mRNA 5′-UTR lengthlen_5UTRlog10 ( Length of the 5′ UTR ) 98 . 2%8 . 2%4 . 6%17 . 2% ORF lengthlen_ORFlog10 ( Length of the ORF ) 100%100%100%100% 3′-UTR lengthlen_3UTRlog10 ( Length of the 3′ UTR ) ( Hausser et al . , 2009 ) 100%100%100%100% 5′-UTR AU contentAU_5UTRFraction of AU nucleotides in the 5′ UTR13%38 . 9%91 . 1%31 . 3% ORF AU contentAU_ORFFraction of AU nucleotides in the ORF1 . 2%72 . 4%28 . 4%35 . 8% 3′-UTR AU contentAU_3UTRFraction of AU nucleotides in the 3′ UTR ( Robins and Press , 2005; Hausser et al . , 2009 ) 5 . 4%73 . 3%65 . 3%80 . 6% 3′-UTR offset-6mer sitesoff6mNumber of offset-6mer sites in the 3′ UTR ( Friedman et al . , 2009 ) 65 . 9%89 . 6%99 . 8%100% ORF 8mer sitesORF8mNumber of 8mer sites in the ORF ( Lewis et al . , 2005; Reczko et al . , 2012 ) 99 . 5%99 . 1%100%100% ORF 7mer-m8 sitesORF7m8Number of 7mer-m8 sites in the ORF ( Reczko et al . , 2012 ) 4 . 7%4 . 3%85 . 3%100% ORF 7mer-A1 sitesORF7A1Number of 7mer-A1 sites in the ORF ( Reczko et al . , 2012 ) 68 . 4%34 . 2%97 . 8%98 . 4% ORF 6mer sitesORF6mNumber of 6mer sites in the ORF ( Reczko et al . , 2012 ) 91%13 . 3%0 . 7%36 . 7%The feature description does not include the scaling performed ( Table 3 ) to generate more comparable regression coefficients . One of the 26 features was site PCT ( probability of conserved targeting ) , which estimates the probability of the site being preferentially conserved because it is targeted by the cognate miRNA ( Friedman et al . , 2009 ) . Prior to use , our PCT scores were updated to take advantage of improvements in both mouse and human 3′-UTR annotations ( Harrow et al . , 2012; Flicek et al . , 2014 ) , the additional sequenced vertebrate genomes aligned to the mouse and human genomes ( Karolchik et al . , 2014 ) , and our expanded set of miRNA families broadly conserved among vertebrate species , which increased from 87 to 111 families ( with the 111 including 16 isomiR families , that is , cases in which a second or third miRNA was produced from a pri-miRNA hairpin , through either conserved expression of miRNAs from both arms of the hairpin or conserved 5′ heterogeneity ) . Using these updates increased sensitivity , with our estimate for the number of human 3′-UTR sites conserved above background increasing from ∼46 , 400 ( Friedman et al . , 2009 ) to ∼62 , 300 . The PCT score on its own correlates with site efficacy , and when using the same set of 3′ UTRs this correlation increased only modestly for the new scores ( data not shown ) , consistent with the notion that the evolutionary signal was already nearly saturated in the previous analysis of 23 species spanning the vertebrate tetrapods ( Friedman et al . , 2009 ) . Nonetheless , we used our updated PCT score as a feature for sites of broadly conserved miRNAs within our training set . A second feature that we re-evaluated was the predicted structural accessibility of the site . As scored previously , the degree to which the site nucleotides were predicted to be free of pairing to flanking 3′-UTR regions was not informative after controlling for the contribution of local AU content ( Grimson et al . , 2007 ) . However , analysis inspired by work on siRNA site accessibility ( Tafer et al . , 2008 ) suggested an improved scoring scheme for this feature . For this analysis we used RNAplfold ( Bernhart et al . , 2006 ) to predict the unpaired probabilities for variable-sized windows in the proximity of the site and then examined the relationship between these probabilities and the repression associated with sites in our compendium of normalized datasets , while controlling for local AU content and other features of the context+ model ( Figure 4A ) . Based on these results , which resembled those reported previously ( Tafer et al . , 2008 ) , we scored predicted structural accessibility ( SA ) as proportional to the log10 value of the unpaired probability for a 14-nt region centered on the match to miRNA nucleotides 7 and 8 . 10 . 7554/eLife . 05005 . 015Figure 4 . Developing a regression model to predict miRNA targeting efficacy . ( A ) Optimizing the scoring of predicted structural accessibility . Predicted RNA structural accessibility scores were computed for variable-length windows within the region centered on each canonical 7–8 nt 3′-UTR site . The heatmap displays the partial correlations between these values and the repression associated with the corresponding sites , determined while controlling for local AU content and other features of the context+ model ( Garcia et al . , 2011 ) . ( B ) Performance of the models generated using stepwise regression compared to that of either the context-only or context+ models . Shown are boxplots of r2 values for each of the models across all 1000 sampled test sets , for mRNAs possessing a single site of the indicated type . For each site type , all groups significantly differ ( P < 10−15 , paired Wilcoxon sign-rank test ) . Boxplots are as in Figure 3C . ( C ) The contributions of site type and each of the 14 features of the context++ model . For each site type , the coefficients for the multiple linear regression are plotted for each feature . Because features are each scored on a similar scale , the relative contribution of each feature in discriminating between more or less effective sites is roughly proportional to the absolute value of its coefficient . Also plotted are the intercepts , which roughly indicate the discriminatory power of site type . Dashed bars indicate the 95% confidence intervals of each coefficient . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 01510 . 7554/eLife . 05005 . 016Figure 4—source data 1 . Coefficients of the trained context++ model corresponding to each site type . Using these coefficients and corresponding scaling factors ( Table 3 ) , context++ scores can be computed essentially as illustrated in Supplementary Figure 5 of Garcia et al . ( 2011 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 016 Having assembled a set of candidate features , we used the stepAIC function from the ‘MASS’ R package ( Venables and Ripley , 2002 ) to determine which features were most useful for modeling site efficacy . This function uses stepwise regression to build models with increasing numbers of features until it reaches the optimal Akaike Information Criterion ( AIC ) value . The AIC evaluates the tradeoff between the benefit of increasing the likelihood of the regression fit and the cost of increasing the complexity of the model by adding more variables . For each of the four seed-matched site types , models were built for 1000 samples of the dataset . Each sample included 70% of the mRNAs with single sites to the transfected sRNA from each experiment ( randomly selected without replacement ) , reserving the remaining 30% as a test set . Compared to our context-only and context+ models ( Grimson et al . , 2007; Garcia et al . , 2011 ) , the new stepwise regression models were significantly better at predicting site efficacy when evaluated using their corresponding held-out test sets , as illustrated for the each of four site types ( Figure 4B ) . Reasoning that features most predictive would be robustly selected , we focused on 14 features selected in nearly all 1000 bootstrap samples for at least two site types ( Table 1 ) . These included all three features considered in our original context-only model ( minimum distance from 3′-UTR ends , local AU composition and 3′-supplementary pairing ) , the two added in our context+ model ( SPS and TA ) , as well as nine additional features ( 3′-UTR length , ORF length , predicted SA , the number of offset-6mer sites in the 3′ UTR and 8mer sites in the ORF , the nucleotide identity of position 8 of the target , the nucleotide identity of positions 1 and 8 of the sRNA , and site conservation ) . Other features were frequently selected for only one site type ( e . g . , ORF 7mer-A1 sites , ORF 7mer-m8 sites , and 5′-UTR length; Table 1 ) . Presumably these and other features were not robustly selected because either their correlation with targeting efficacy was very weak ( e . g . , the 7 nt ORF sites ) or they were strongly correlated to a more informative feature , such that they provided little additional value beyond that of the more informative feature ( e . g . , 3′-UTR AU content compared to the more informative feature , local AU content ) . Using the 14 robustly selected features , we trained multiple linear regression models on all of the data . The resulting models , one for each of the four site types , were collectively called the context++ model ( Figure 4C and Figure 4—source data 1 ) . For each feature , the sign of the coefficient indicated the nature of the relationship . For example , mRNAs with either longer ORFs or longer 3′ UTRs tended to be more resistant to repression ( indicated by a positive coefficient ) , whereas mRNAs with either structurally accessible target sites or ORF 8mer sites tended to be more prone to repression ( indicated by a negative coefficient ) . Based on the relative magnitudes of the regression coefficients , some newly incorporated features , such as 3′-UTR length , ORF length , and SA , contributed similarly to features previously incorporated in the context+ model , such as SPS , TA , and local AU ( Figure 4C ) . New features with an intermediate level of influence included the number of ORF 8mer sites and site conservation as well as the presence of a 5′ G in the sRNA ( Figure 4C ) , the latter perhaps a consequence of differential sRNA loading efficiency . The weakest features included the sRNA and target position 8 identities as well as the number of offset-6mer sites . The identity of sRNA nucleotide 8 exhibited a complex pattern that was site-type dependent . Relative to a position-8 U in the sRNA , a position-8 C further decreased efficacy of sites with a mismatch at this position ( 6mer or 7mer-A1 sites ) , whereas a position-8 A had the opposite effect ( Figure 4C ) . Similarly , a position-8 C in the site also conferred decreased efficacy of 6mer and 7mer-A1 sites relative to a position-8 U in the site ( Figure 4C ) . Allowing interaction terms when developing the model , including a term that captured the potential interplay between these positions , did not provide sufficient benefit to justify the more complex model . We compared the predictive performance of our context++ model to that of the most recent versions of 17 in silico tools for predicting miRNA targets , including AnTar ( Wen et al . , 2011 ) , DIANA-microT-CDS ( Reczko et al . , 2012 ) , ElMMo ( Gaidatzis et al . , 2007 ) , MBSTAR ( Bandyopadhyay et al . , 2015 ) , miRanda-MicroCosm ( Griffiths-Jones et al . , 2008 ) , miRmap ( Vejnar and Zdobnov , 2012 ) , mirSVR ( Betel et al . , 2010 ) , miRTarget2 ( Wang and El Naqa , 2008 ) , MIRZA-G ( Gumienny and Zavolan , 2015 ) , PACCMIT-CDS ( Marin et al . , 2013 ) , PicTar2 implemented for predictions conserved through mammals , chicken , or fish ( PicTarM , PicTarC , and PicTarF , respectively ) ( Anders et al . , 2012 ) , PITA ( Kertesz et al . , 2007 ) , RNA22 ( Miranda et al . , 2006 ) , SVMicrO ( Liu et al . , 2010 ) , TargetRank ( Nielsen et al . , 2007 ) , and TargetSpy ( Sturm et al . , 2010 ) ; as well as successive versions of TargetScan , which offer context scores ( Grimson et al . , 2007 ) , PCT scores ( Friedman et al . , 2009 ) , or context+ scores ( Garcia et al . , 2011 ) as options for ranking predictions ( TargetScan5 , TargetScan . PCT , or TargetScan6 , respectively ) for either all mRNAs with a canonical 7–8 nt 3′-UTR site ( TargetScan . All ) or those with only broadly conserved sites ( TargetScan . Cons ) . To the best of our knowledge , algorithms excluded from the comparison either were not de novo prediction algorithms ( relying on consensus techniques or experimental data ) , did not provide a pre-computed database of results , or lacked a numerical value ( or ranking ) of either target-prediction confidence or mRNA responsiveness . To test the performance of the included methods , we used the results of seven microarray datasets that each monitor mRNA changes after transfection of a conserved miRNA into HCT116 cells containing a hypomorphic mutant for Dicer ( Linsley et al . , 2007 ) . These datasets differ from those used during development and training of our model with respect to both the cell type and the identities of the sRNAs . To prevent our model from gaining an advantage over methods that used standard 3′-UTR annotations , we used RefSeq-annotated 3′ UTRs ( rather than 3P-seq–supported annotations ) to generate the context++ test-set predictions . For genes with multiple annotated 3′ UTRs we chose the longest isoform because the microarray probes of the test set often matched only this isoform . For each 3′ UTR containing multiple sites to the cognate miRNA , the context++ scores of individual sites were summed to generate the total context++ score to be used to rank that predicted target . The number of potential miRNA–mRNA interactions considered by the different methods varied greatly ( Figure 5A ) , which reflected the varied strategies and priorities of these prediction efforts . Out of a concern for prediction specificity , many efforts only consider interactions involving 7–8 nt seed-matched sites . Accordingly , we first tested how well each of the methods predicted the repression of mRNAs with at least one canonical 7–8 nt 3′-UTR site ( Figure 5B ) . The context++ model performed substantially better than the most predictive published model , which was TargetScan6 . All . Of algorithms derived from other groups , DIANA-microT-CDS , miRTarget2 , miRanda-miRSVR , MIRZA-G ( and its derivatives ) , and TargetRank were the most predictive , with performance within range of TargetScan5 . All ( Figure 5B ) . 10 . 7554/eLife . 05005 . 017Figure 5 . Performance of target prediction algorithms on a test set of seven experiments in which miRNAs were individually transfected into HCT116 cells . ( A ) Average number of targets predicted by the indicated algorithm for each of the seven miRNAs in the test set ( let-7c , miR-16 , miR-103 , miR-106b , miR-200b , miR-200a , and miR-215 ) . The numbers of predictions with at least one canonical 7–8 nt 3′-UTR site to the transfected miRNA ( dark blue ) are distinguished from the remaining predictions ( light blue ) . Names of algorithms are colored according to whether they consider only sequence or thermodynamic features of site pairing ( grey ) , only site conservation ( orange ) , pairing and contextual features of a site ( red ) , or pairing , contextual features , and site conservation ( purple ) . The most recently updated predictions were downloaded , with year that those predictions were released indicated in parentheses . ( B and C ) Extent to which the predictions explain the mRNA fold changes observed in the test set . For predictions tallied in panel ( A ) , the explanatory power , as evaluated by the r2 value for the relationship between the scores of the predictions and the observed mRNA fold changes in the test set , is plotted for either mRNAs with 3′ UTRs containing at least one canonical 7–8 nt 3′-UTR site ( B ) or other mRNAs ( C ) . Algorithms designed to evaluate only targets with seed-matched 7–8 nt 3′-UTR sites are labeled ‘N/A’ in ( C ) . ( D ) Repression of the top predictions of the context++ model and of our previous two models , focusing on an average of 16 top predicted targets per miRNA in the test set . The dotted lines indicate the median fold-change value for each distribution , otherwise as in Figure 1A . ( E and F ) Median mRNA fold changes observed in the test set for top-ranked predicted targets , considering either all predictions ( E ) or only those with 3′ UTRs lacking at least one canonical 7–8 nt site ( F ) . For each algorithm listed in panel ( A ) , all reported predictions for the seven miRNAs were ranked according to their scores , and the indicated sliding threshold of top predictions was implemented . For example , at the threshold of 4 , the 28 predictions with the top scores were identified ( an average of 4 predictions per miRNA , allowing miRNAs with more top scores to contribute more predictions ) , mRNA fold-change values from the cognate transfections were collected , and the median value was plotted . When the threshold exceeded the number of reported predictions , no value was plotted . Also plotted is the median mRNA fold change for all mRNAs with at least one cognate canonical 7–8 nt site in their 3′ UTR ( dashed line; an average of 1366 mRNAs per miRNA ) , the median fold change for all mRNAs with at least one conserved cognate canonical 7–8 nt site in their 3′ UTR ( dotted line; an average of 461 mRNAs per miRNA ) , and the 95% interval for the median fold change of randomly selected mRNAs , determined using 1000 resamplings ( without replacement ) at each cutoff ( shading ) . Conserved sites were defined as in TargetScan6 , with conservation cutoffs for each site type set at different branch-length scores ( cutoffs of 0 . 8 , 1 . 3 , and 1 . 6 for 8mer , 7mer-m8 , and 7mer-A1 sites , respectively ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 01710 . 7554/eLife . 05005 . 018Figure 5—figure supplement 1 . Performance of miRNA prediction algorithms on the test set . ( A ) This panel is as in Figure 5D , but shows the results for all algorithms evaluated in Figure 5A . Algorithm names are listed in the order of the median fold change for their top predictions , with each name colored using the color used for its cumulative distribution . ( B and C ) These panels are as in Figure 5E–F , respectively , but compare mean fold changes instead of median fold changes . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 018 Part of the reason that some algorithms performed more poorly is that they consider relatively few potential miRNA–target interactions ( Figure 5A ) . For example , the drop in performance observed between TargetScan . All and TargetScan . Cons illustrates the effect of limiting analysis to the more highly conserved sites . Nonetheless , the performance of TargetScan . Cons relative to other methods that consider relatively few sites shows that a signal can be observed in this assay even when a very limited number of interactions are scored ( Figure 5A , B ) , presumably because much of the functional targeting is through conserved interactions . Indeed , the performance of ElMMO and TargetScan . PCT illustrate what can be achieved by scoring just the extent of site conservation and no other parameter . In an attempt to maximize prediction sensitivity , some efforts consider many interactions that lack a canonical 7–8 nt 3′-UTR site ( Figure 5A ) . However , all of these algorithms performed poorly in predicting the response of mRNAs lacking such sites ( Figure 5C ) . The two algorithms achieving any semblance of prediction accuracy did so by predicting some of the canonical interactions with known marginal efficacy . These were DIANA-microT-CDS , which captured modest effects of canonical sites in ORFs ( Reczko et al . , 2012; Marin et al . , 2013 ) , and the context++ model , which captured the modest effects of canonical 6mers in 3′ UTRs ( as modified by the 14 features , which included offset 6mers and 8mer ORF sites ) ( Figure 5C ) . The algorithms designed to identify many non-canonical sites performed much more poorly in this test ( r2 < 0 . 004 ) , consistent with the idea that the vast majority of mRNAs without canonical sites either do not change in response to the miRNA or change in an unpredictable fashion as a secondary effect of introducing the miRNA . Another way to evaluate the performance of targeting algorithms is to examine the repression of the top predicted targets . Compared to the r2 test , this approach does not penalize efforts that either impose more stringent cutoffs to achieve higher prediction specificity or implement scoring schemes that are not designed to correlate directly with site efficacy . Perhaps most importantly , this approach aligns with the goals of a biologist considering the top-ranked predictions in an attempt to focus on those most likely to undergo substantial repression . When choosing an average of 16 predicted targets for each of the seven test-set miRNAs , we found that these top 112 predictions of the context++ model were significantly more repressed than the top predictions from earlier versions of TargetScan ( Figure 5D ) and the top predictions of the other algorithms ( Figure 5—figure supplement 1A ) . Despite the success of the context++ model , not all of the fold changes for its top predicted targets were negative; for the test set , the distribution of these fold changes intersected 0 . 0 at a cumulative fraction of 0 . 92 , indicating that mRNAs for 8% of the top predictions increased rather than decreased with transfection of the cognate miRNA ( Figure 5D ) . In principle , these mRNAs could still be authentic targets that are repressed in these cells but nonetheless had increased expression values because either experimental noise or secondary effects of introducing the miRNA overwhelmed the signal for miRNA-mediated repression . Alternatively , some or all of these mRNAs could be false-positive predictions . Because only half of the false-positive predictions would be expected to have positive fold changes in the presence of the miRNA , our best estimate of the upper limit on the false-positive predictions was 2 × 8% , or 16% , at this cutoff ( for which an average of 16 top predictions per miRNA is considered ) . At the same cutoff , the distribution of fold changes for each of the previous algorithms intersected 0 . 0 at a cumulative fractions ranging from 0 . 50–0 . 88 ( Figure 5—figure supplement 1A ) , which implied lower prediction specificity than that observed for the context++ model , with correspondingly higher estimates for the upper limits of false positives among their top predictions , ranging from 24–100% . To evaluate the performance of top-ranked predictions more systematically , we examined median repression of the predicted targets over a broad spectrum of cutoffs , ranging from an average of 4–4096 predictions per miRNA ( Figure 5E ) . Regardless of the cutoff , the top context++ predictions were the most repressed . The top predictions of most other algorithms were repressed significantly more than expected by chance , although the median repression of some ( MBSTAR , RNA22 , PACCMIT-CDS , and AnTarCLIP ) did not exceed the median repression of all mRNAs with a canonical 7–8 nt 3′-UTR site ( Figure 5E ) . Plotting average fold changes rather than median fold changes resulted in very similar relative performances ( Figure 5—figure supplement 1B ) . After eliminating interactions that could involve canonical 7–8 nt 3′-UTR sites , the remaining top predictions were modestly repressed at best ( Figure 5F and Figure 5—figure supplement 1C ) . The most repressed predicted targets without canonical 7–8 nt 3′-UTR sites were those of the context++ model , which scored predictions with canonical 6mer 3′-UTR sites . For algorithms designed to identify many non-canonical sites , the top predictions without 7–8 nt 3′-UTR sites were essentially unresponsive to the transfected miRNA , which indicated that if effective non-canonical sites for these seven miRNAs exist , they are not enriched among the top predictions of these algorithms . We used our context++ model to overhaul the TargetScan predictions ( as described in the next section ) , and as a third way of testing this model , we compared the performance of these TargetScan7 predictions with that of in vivo CLIP experiments . When doing this comparison we took care to evaluate sets of predictions that each were the same size as the cognate set of CLIP-supported targets , whereas some previous analyses compare expansive sets of computational predictions ( e . g . , all mRNAs with a 6mer site ) to relatively small sets of biochemically supported predictions ( Chi et al . , 2009; Lipchina et al . , 2011; Loeb et al . , 2012; Grosswendt et al . , 2014; Tan et al . , 2014 ) . mRNAs with expression signals approaching the array background were not considered . This exclusion was particularly important when comparing to CLIP results; CLIP can only evaluate mRNAs expressed in the cells , which would impart a trivial relative advantage if the computational predictions included targets that appeared unresponsive because they were expressed below the array background . The non-canonical CLIP-supported targets were also not considered , as we had already shown that they do not respond to the miRNA ( Figure 1 and Figure 1—figure supplements 1–4 ) , and we did not want the inclusion of these easily recognized false positives to impart a disadvantage to CLIP . Regardless of the set of canonical CLIP-supported targets examined , we did not find a setting in which they responded significantly better than did the cohort of TargetScan7 predictions , and in some cases , the TargetScan7 predictions performed significantly better ( Figure 6A–J ) . Similar results were observed when comparing the repression of our predictions to that of mRNAs identified biochemically without crosslinking , using either pulldown-seq or IMPACT-seq ( Tan et al . , 2014 ) , again focusing on only mRNAs with canonical sites ( Figure 6K , L ) . Thus , for identifying consequential miRNA–target interactions , the TargetScan7 model is not only more convenient than experimental determination of binding sites , it is also at least as effective . The analogous conclusion was reached from analyses that used the context++ model without using the improved annotation and quantification of 3′-UTR isoforms ( data not shown ) . 10 . 7554/eLife . 05005 . 019Figure 6 . Response of predictions and mRNAs with experimentally supported canonical binding sites . ( A–E ) Comparison of the top TargetScan7 predicted targets to mRNAs with canonical sites identified from dCLIP in either HeLa cells with and without transfected miR-124 ( Chi et al . , 2009 ) or lymphocytes with and without miR-155 ( Loeb et al . , 2012 ) . Plotted are cumulative distributions of mRNA fold changes after transfection of miR-124 in HeLa cells ( A ) , or after genetic ablation of miR-155 in either T cells ( B ) , Th1 cells ( C ) , Th2 cells ( D ) , and B cells ( E ) ( one-sided K–S test , P values ) . For genes with alternative last exons , the analysis considered the score of the most abundant alternative last exon , as assessed by 3P-seq tags ( as is the default for TargetScan7 when ranking predictions ) . Each dCLIP-identified mRNA was required to have a 3′-UTR CLIP cluster with at least one canonical site to the cognate miRNA ( including 6mers but not offset 6mers ) . Each intersection mRNA ( red ) was found in both the dCLIP set and top TargetScan7 set . Similarity between performance of the TargetScan7 and dCLIP sets ( purple and green , respectively ) and TargetScan7 and intersection sets ( blue and red , respectively ) was tested ( two-sided K–S test , P values ) ; the number of mRNAs analyzed in each category is in parentheses . TargetScan7 scores for mouse mRNAs were generated using human parameters for all features . ( F–H ) Comparison of top TargetScan7 predicted targets to mRNAs with canonical binding sites identified using photoactivatable-ribonucleoside-enhanced CLIP ( PAR-CLIP ) ( Hafner et al . , 2010; Lipchina et al . , 2011 ) . Plotted are cumulative distributions of mRNA fold changes after either transfecting miR-7 ( F ) or miR-124 ( G ) into HEK293 cells , or knocking down miR-302/367 in hESCs ( H ) . Otherwise these panels are as in ( A–E ) . ( I ) Comparison of top TargetScan7 predicted targets to mRNAs with canonical sites identified using CLASH ( Helwak et al . , 2013 ) . Plotted are cumulative distributions of mRNA fold changes after knockdown of 25 miRNAs from 14 miRNA families in HEK293 cells . For each of these miRNA families , a cohort of top TargetScan7 predictions was chosen to match the number of mRNAs with CLASH-identified canonical sites , and the union of these TargetScan7 cohorts was analyzed . The total number of TargetScan7 predictions did not match the number of CLASH-identified targets due to slightly different overlap between mRNAs targeted by different miRNAs . Otherwise these panels are as in ( A–E ) . ( J ) Comparison of top TargetScan7 predicted targets to mRNAs with chimera-identified canonical sites ( Grosswendt et al . , 2014 ) . Otherwise this panel is as in ( I ) . ( K ) Comparison of top TargetScan7 predicted targets to mRNAs with canonical binding sites within 3′ UTRs of mRNAs identified using pulldown-seq ( Tan et al . , 2014 ) . Plotted are cumulative distributions of mRNA fold changes after transfecting miR-522 into triple-negative breast cancer ( TNBC ) cells . Otherwise this panel is as in ( A–E ) . ( L ) Comparison of top TargetScan7 predicted targets to mRNAs with canonical sites identified using IMPACT-seq ( Tan et al . , 2014 ) . Otherwise this panel is as in ( K ) . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 019 As mentioned earlier , mRNAs that increase rather than decrease in the presence of the miRNA can indicate the presence of false positives in a set of candidate targets . Examination of the mRNA fold-change distributions from the perspective of false positives revealed no advantage of the experimental approaches over our predictions . When compared to the less informative CLIP datasets , the TargetScan7 predictions included fewer mRNAs that increased , and when compared to the CLIP datasets that performed as well as the predictions , the TargetScan7 predictions included a comparable number of mRNAs that increased , implying that the TargetScan7 predictions had no more false-positive predictions than did the best experimental datasets . Because some sets of canonical biochemically supported targets performed as well as their cohort of top TargetScan7 predictions , we considered the utility of focusing on mRNAs identified by both approaches . In each comparison , the set of mRNAs that were both canonical biochemically supported targets and within the cohort of top TargetScan7 predictions tended to be more responsive . However , these intersecting subsets included much fewer mRNAs than the original sets , and when compared to an equivalent number of top TargetScan7 predictions , each intersecting set performed no better than did its cohort of top TargetScan7 predictions ( Figure 6 ) . Therefore , considering the CLIP results to restrict the top predictions to a higher-confidence set is useful but not more useful than simply implementing a more stringent computational cutoff . Likewise , taking the union of the CLIP-supported targets and the cohort of predictions , rather than the intersection , did not generate a set of targets that was more responsive than an equivalent number of top TargetScan7 predictions ( data not shown ) . As already mentioned , we used the context++ model to rank miRNA target predictions to be presented in version 7 of the TargetScan database ( targetscan . org ) , thereby making our results accessible to others working on miRNAs . For simplicity , we had developed the context++ model using mRNAs without abundant alternative 3′-UTR isoforms , and to make fair comparisons with the output of previous models , we had tested the context++ model using only the longest RefSeq-annotated isoform . Nevertheless , considering the usage of alternative 3′-UTR isoforms , which can influence both the presence and scoring of target sites , significantly improves the performance of miRNA targeting models ( Nam et al . , 2014 ) . Thus , our overhaul of the TargetScan predictions incorporated both the context++ scores and current isoform information when ranking mRNAs with canonical 7–8 nt miRNA sites in their 3′ UTRs . The resulting improvements applied to the predictions centered on human , mouse , and zebrafish 3′ UTRs ( TargetScanHuman , TargetScanMouse , and TargetScanFish , respectively ) ; and by 3′-UTR homology , to the conserved and nonconserved predictions in chimp , rhesus , rat , cow , dog , opossum , chicken , and frog; as well as to the conserved predictions in 74 other sequenced vertebrate species , thereby providing a valuable resource for placing miRNAs into gene-regulatory networks . Because the main gene-annotation databases ( e . g . , RefSeq and Ensembl/Gencode ) are still in the process of incorporating the information available on 3′-UTR isoforms , the first step in the TargetScan overhaul was to compile a set of reference 3′ UTRs that represented the longest 3′-UTR isoforms for representative ORFs of human , mouse , and zebrafish . These representative ORFs were chosen among the set of transcript annotations sharing the same stop codon , with alternative last exons generating multiple representative ORFs per gene . The human and mouse databases started with Gencode annotations ( Harrow et al . , 2012 ) , for which 3′ UTRs were extended , when possible , using RefSeq annotations ( Pruitt et al . , 2012 ) , recently identified long 3′-UTR isoforms ( Miura et al . , 2013 ) , and 3P-seq clusters marking more distal cleavage and polyadenylation sites ( Nam et al . , 2014 ) . Zebrafish reference 3′ UTRs were similarly derived in a recent 3P-seq study ( Ulitsky et al . , 2012 ) . For each of these reference 3′-UTR isoforms , 3P-seq datasets were used to quantify the relative abundance of tandem isoforms , thereby generating the isoform profiles needed to score features that vary with 3′-UTR length ( len_3UTR , min_dist , and off6m ) and assign a weight to the context++ score of each site , which accounted for the fraction of 3′-UTR molecules containing the site ( Nam et al . , 2014 ) . For each representative ORF , our new web interface depicts the 3′-UTR isoform profile and indicates how the isoforms differ from the longest Gencode annotation ( Figure 7 ) . 10 . 7554/eLife . 05005 . 020Figure 7 . Example display of TargetScan7 predictions . The example shows a TargetScanHuman page for the 3′ UTR of the LRRC1 gene . At the top is the 3′-UTR profile , showing the relative expression of tandem 3′-UTR isoforms , as measured using 3P-seq ( Nam et al . , 2014 ) . Shown on this profile is the end of the longest Gencode annotation ( blue vertical line ) and the total number of 3P-seq reads ( 339 ) used to generate the profile ( labeled on the y-axis ) . Below the profile are predicted conserved sites for miRNAs broadly conserved among vertebrates ( colored according to the key ) , with options to display conserved sites for mammalian conserved miRNAs , or poorly conserved sites for any set of miRNAs . Boxed are the predicted miR-124 sites , with the site selected by the user indicated with a darker box . The multiple sequence alignment shows the species in which an orthologous site can be detected ( white highlighting ) among representative vertebrate species , with the option to display site conservation among all 84 vertebrate species . Below the alignment is the predicted consequential pairing between the selected miRNA and its sites , showing also for each site its position , site type , context++ score , context++ score percentile , weighted context++ score , branch-length score , and PCT score . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 02010 . 7554/eLife . 05005 . 021Figure 7—figure supplement 1 . Flowchart of the computational pipeline used to build the TargetScan7 database . DOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 021 3P-seq data were available for seven developmental stages or tissues of zebrafish , enabling isoform profiles to be generated and predictions to be tailored for each of these . For human and mouse , however , 3P-seq data were available for only a small fraction of tissues/cell types that might be most relevant for end users , and thus results from all 3P-seq datasets available for each species were combined to generate a meta 3′-UTR isoform profile for each representative ORF . Although this approach reduces accuracy of predictions involving differentially expressed tandem isoforms , it nonetheless outperforms the previous approach of not considering isoform abundance at all , presumably because isoform profiles for many genes are highly correlated in diverse cell types ( Nam et al . , 2014 ) . For each 6–8mer site , we used the corresponding 3′-UTR profile to compute the context++ score and to weight this score based on the relative abundance of tandem 3′-UTR isoforms that contained the site ( Nam et al . , 2014 ) . Scores for the same miRNA family were also combined to generate cumulative weighted context++ scores for the 3′-UTR profile of each representative ORF , which provided the default approach for ranking targets with at least one 7–8 nt site to that miRNA family . Effective non-canonical site types , that is , 3′-compensatory and centered sites , were also predicted . Using either the human or mouse as a reference , predictions were also made for orthologous 3′ UTRs of other vertebrate species . As an option for tetrapod species , the user can request that predicted targets of broadly conserved miRNAs be ranked based on their aggregate PCT scores ( Friedman et al . , 2009 ) , as updated in this study . The user can also obtain predictions from the perspective of each protein-coding gene , viewed either as a table of miRNAs ( ranked by either cumulative weighted context++ score or aggregate PCT score ) or as the mapping of 7–8 nt sites ( as well as non-canonical sites ) shown beneath the 3′-UTR profile and above the 3′-UTR sequence alignment ( Figure 7 ) . A flowchart summarizing the TargetScan overhaul is provided ( Figure 7—figure supplement 1 ) . Starting with an expanded and improved compendium of sRNA transfection datasets , we identified 14 features that each correlate with target repression and add predictive value when incorporated into a quantitative model of miRNA targeting efficacy . This model performed better than previous models and at least as well as the best high-throughput CLIP approaches . Because our model was trained on data derived from a single cell type , a potential concern was its generalizability to other cell types . Heightening this concern is the recent report of widespread dependency of miRNA-mediated repression on cellular context ( Erhard et al . , 2014 ) . However , other work addressing this question shows that after accounting for the different cellular repertoires of expressed mRNAs , the target response is remarkably consistent between different cell types , with alternative usage of 3′-UTR isoforms being the predominant mechanism shaping cell-type-specific differences in miRNA targeting ( Nam et al . , 2014 ) . Testing the model across diverse cell types confirmed its generalizability; it performed at least as well as the best high-throughput CLIP approaches in each of the contexts examined ( Figure 6 ) . Of course , this testing was restricted to only those predicted targets that were expressed in each cellular context . Likewise , to achieve this highest level of performance , any future use of our model or its predictions would also require filtering of the predictions to focus on only the miRNAs and mRNAs co-expressed in the cells of interest . One of the more interesting features incorporated into the context++ model is SA ( the predicted structural accessibility of the site ) . Freedom from occlusive mRNA structure has long been considered a site-efficacy determinant ( Robins et al . , 2005; Ameres et al . , 2007; Kertesz et al . , 2007; Long et al . , 2007; Tafer et al . , 2008 ) and proposed as the underlying mechanistic explanation for the utility of other features , including global 3′-UTR AU content ( Robins and Press , 2005; Hausser et al . , 2009 ) , local AU content ( Grimson et al . , 2007; Nielsen et al . , 2007 ) , minimum distance of the site ( Grimson et al . , 2007 ) , and 3′-UTR length ( Hausser et al . , 2009; Betel et al . , 2010; Wen et al . , 2011; Reczko et al . , 2012 ) . The challenge has been to predict and score site accessibility in a way that is informative after controlling for local AU content , which is important for speaking to the importance of less occlusive secondary structure as opposed to involvement of some AU-binding activity ( Grimson et al . , 2007 ) . The selection of the SA feature in all 1000 bootstrap samples of all four site types showed that it provided discriminatory power apart from that provided by local AU content and other correlated features , which reinforced the idea that the occlusive RNA structure does indeed limit site efficacy . This being said , local AU content , minimum distance of the site , and 3′-UTR length were each also selected in nearly all 1000 bootstrap samples for most site types ( Table 1 ) , which suggests that either these features were selected for reasons other than their correlation with site accessibility or the definition and scoring of our SA feature has additional room for improvement . Our ability to confidently identify additional features that each contribute to improved prediction of targeting efficacy was enhanced by our pre-processing of the experimental datasets , which minimized variation from biases unrelated to the sRNA sequence . Yet despite applying this same normalization procedure to our test set , the observed r2 value of 0 . 14 implied that our model explained only 14% of the variability observed among mRNAs with canonical 7–8 nt 3′-UTR sites ( Figure 4B ) . The r2 value increased to 0 . 15 when considering the usage of alternative 3′-UTR isoforms , but 85% of the variability remained unexplained . Error in the microarray measurements , different sRNA transfection efficiencies , variable incorporation of sRNAs into the silencing complex , and secondary effects of introducing the sRNA presumably made major contributions to the unexplained variability . Nonetheless , imperfections of the context++ model also contributed , raising the question of how much the model might be improved by identifying additional features or developing better methods for scoring and combining existing features . In analyses not described , we evaluated the utility of other types of regression ( e . g . , linear regression models with interaction terms , lasso/elastic net-regularized regression , multivariate adaptive regression splines , random forest , boosted regression trees , and iterative Bayesian model averaging ) and found their performance to be comparable to that of stepwise regression but their resulting models to be considerably more complex and thus less interpretable . One way to evaluate the extent to which the context++ model might be improved is to consider the degree to which its performance depends on the site-conservation feature . Because sites under selective pressure preferentially possess molecular features required for efficacy , inclusion of the site-conservation feature indirectly recovers some of the information that would otherwise be lost when informative molecular features are missing or imperfectly scored . As more informative molecular features are identified and included in a model , less information remains to be captured , and thus the site-conservation feature cannot contribute as much to the performance of the model . The site-conservation feature ( PCT ) was chosen in all 1000 bootstrap samples of each of the three major site types , which showed that the molecular features of our model still do not fully capture all the determinants under selective pressure . However , PCT was not one of the most informative features ( Figure 4C ) . Moreover , when tested as in Figure 5B , a model trained on only site type and the other 13 molecular features performed nearly as well as the full context++ model ( r2 of 0 . 126 , compared to 0 . 139 for the full model ) . This drop in r2 of only 0 . 013 was substantially less than the 0 . 044 r2 observed for the site-conservation feature on its own ( Figure 5B , TargetScan . PCT ) , which suggested that when predicting the response of the test-set mRNAs with the major canonical site types , the context++ model captured 70% ( calculated as [0 . 044–0 . 013]/0 . 044 ) of the information potentially imparted by molecular features . The relatively minor contribution of site conservation highlights the ability of the context++ model to predict the efficacy of nonconserved sites . Although , everything else being equal , its score for a conserved site is slightly better than that for a nonconserved site , this difference does not prevent inclusion of nonconserved sites from the top predictions . Its general applicability to all canonical sites is useful for evaluating not only nonconserved sites to conserved miRNAs but also all sites for nonconserved miRNAs ( e . g . , Figure 6K , L ) , including viral miRNAs , as well as the off-targets of synthetic siRNAs and shRNAs . Our analyses show that recent computational and experimental approaches , including the different types of CLIP , all fail to identify non-canonical targets that are repressed more than control transcripts ( Figures 1 , 5C , F ) , which reopens the question of whether more than a miniscule fraction of miRNA-mediated repression is mediated through non-canonical sites . Although CLIP approaches can identify non-canonical sites that bind the miRNA with some degree of specificity ( Figure 2 ) , these non-canonical binding sites do not function to mediate detectable repression . Thus far , the only functional non-canonical sites that can be predicted are 3′-compensatory sites , cleavage sites , and centered sites , which together comprise only a very small fraction ( <1% ) of the functional sites that can be predicted with comparable accuracy ( Bartel , 2009; Shin et al . , 2010 ) . The failure of computational methods to find many functional non-canonical sites cannot rule out the possibility that many of these sites might still exist; if such sites are recognized through unimagined determinants , computational efforts might have missed them . CLIP approaches , on the other hand , provide information that is independent of proposed pairing rules or other hypothesized recognition determinants . Therefore , our analyses of the CLIP results , which detected no residual repression after accounting for canonical interactions , provide the most compelling evidence to date on this issue . Unless there is a substantial technical bias in the CLIP approach ( such as a large unanticipated disparity in the propensity of non-canonical interactions to crosslink ) , the inability of current CLIP approaches to identify non-canonical targets that are repressed more than control transcripts argues strongly against the existence of many functional non-canonical targets . Why might the CLIP-identified non-canonical sites fail to mediate repression ( Figure 1 ) despite binding the miRNA in vivo ( Figure 2 ) ? Perhaps these sites are ineffective because perfect seed pairing is required for repression . For example , perfect seed pairing might favor binding of a downstream effector , either directly by contributing to its binding site or indirectly through an Argonaute conformational change that favors its binding . However , this explanation is difficult to reconcile with the activity of 3′-compensatory and centered sites , which can mediate repression despite their lack of perfect seed pairing ( Bartel , 2009; Shin et al . , 2010 ) , and the activity of Argonaute artificially tethered to an mRNA , which can mediate repression without any pairing to the miRNA ( Pillai et al . , 2004; Eulalio et al . , 2008 ) . Therefore , a more plausible explanation is that the CLIP-identified non-canonical sites bind the miRNA too transiently to mediate repression . This explanation for the inefficacy of the recently identified non-canonical sites in the 3′ UTRs resembles that previously proposed for the inefficacy of most canonical sites in ORFs: in both cases the ineffective sites bind to the miRNA very transiently—the canonical sites in ORFs dissociating quickly because of displacement by the ribosome ( Grimson et al . , 2007; Gu et al . , 2009 ) , and the CLIP-identified non-canonical sites in 3′ UTRs dissociating quickly because they lack both seed pairing and the extensive pairing outside the seed characteristic of effective non-canonical sites ( 3′-compensatory and centered sites ) and thus have intrinsically fast dissociation rates . The idea that newly identified non-canonical sites bind the miRNA too transiently to mediate repression raises the question of how CLIP could have identified so many of these sites in the first place; shouldn't crosslinking be a function of site occupancy , and shouldn't occupancy be a function of dissociation rates ? The answers to these questions partially hinge on the realization that the transcriptome has many more non-canonical binding sites than canonical ones . The motifs identified in the non-canonical interactions have information contents as low as 5 . 6 bits , and thus are much more common in 3′ UTRs than canonical 6mer or 7mer sites ( 12 bits and 14 bits , respectively ) . This high abundance of the non-canonical binding sites would help offset the low occupancy of individual non-canonical sites , such that at any moment more than half of the bound miRNA might reside at non-canonical sites , yielding more non-canonical than canonical sites when using experimental approaches with such high specificity that they can identify a site with only a single read ( Figure 2—figure supplement 1A ) . Although the high abundance of non-canonical sites partly explains why CLIP identifies these sites in such high numbers , it cannot provide the complete answer . Some non-canonical sites in the CLASH and chimera datasets are supported by multiple reads , and all the dCLIP-identified non-canonical sites of the miR-155 study ( Loeb et al . , 2012 ) are supported by multiple reads . How could some CLIP clusters with ineffective , non-canonical sites have as much read support as some with effective , canonical sites ? Our answer to this question rests on the recognition that cluster read density does not perfectly correspond to site occupancy ( Friedersdorf and Keene , 2014 ) , with the other key factors being mRNA expression levels and crosslinking efficiency . In principle , normalizing the CLIP tag numbers to the mRNA levels minimizes the first factor , preventing a low-occupancy site in a highly expressed mRNA from appearing as well supported as a high-occupancy site in a lowly expressed mRNA ( Chi et al . , 2009; Jaskiewicz et al . , 2012 ) . Accounting for differential crosslinking efficiencies is a far greater challenge . RNA–protein UV crosslinking is expected to be highly sensitive to the identity , geometry , and environment of the crosslinking constituents , leading to the possibility that the crosslinking efficiency of some sites is orders of magnitude greater than that of others . When considered together with the high abundance of non-canonical sites , variable crosslinking efficiency might explain why so many ineffective non-canonical sites are identified . Overlaying a wide distribution of crosslinking efficiencies onto the many thousands of ineffective , non-canonical sites could yield a substantial number of sites at the high-efficiency tail of the distribution for which the tag support matches that of effective canonical sites . Similar conclusions are drawn for other types of RNA-binding interactions when comparing CLIP results with binding results ( Lambert et al . , 2014 ) . Variable crosslinking efficiency also explains why many top predictions of the context++ model are missed by the CLIP methods , as indicated by the modest overlap in the CLIP identified targets and the top predictions ( Figure 6 ) . The crosslinking results are not only variable from site to site , which generates false negatives for perfectly functional sites , but they are also variable between biological replicates ( Loeb et al . , 2012 ) , which imposes a challenge for assigning dCLIP clusters to a miRNA . Although this challenge is mitigated in the CLASH and chimera approaches , which provide unambiguous assignment of the miRNAs to the sites , the ligation step of these approaches occurs at low frequency and presumably introduces additional biases , as suggested by the different profile of non-canonical sites identified by the two approaches ( Figure 2B and Figure 2—figure supplement 1A ) . For example , CLASH identifies non-canonical pairing to the 3′ region of miR-92 ( Helwak et al . , 2013 ) , whereas the chimera approach identified non-canonical pairing to the 5′ region of this same miRNA ( Figure 2C ) . Because of the false negatives and biases of the CLIP approaches , the context++ model , which has its own flaws , achieves an equal or better performance than the published CLIP studies . Our observation that CLIP-identified non-canonical sites fail to mediate repression reasserts the primacy of canonical seed pairing for miRNA-mediated gene regulation . Compared to canonical sites , effective non-canonical sites ( i . e . , 3′-compensatory sites and centered sites ) are rare because they require many more base pairs to the miRNA ( Bartel , 2009; Shin et al . , 2010 ) and thus together make up <1% of the effective target sites predicted to date . The requirement of so much additional pairing to make up for a single mismatch to the seed is proposed to arise from several sources . The advantage of propagating continuous pairing past miRNA nucleotide 8 ( as occurs for centered sites ) might be largely offset by the cost of an unfavorable conformational change ( Bartel , 2009; Schirle et al . , 2014 ) . Likewise , the advantage of resuming pairing at the miRNA 3′ region ( as occurs for 3′-compensatory sites ) might be partially offset by either the relative disorder of these nucleotides ( Bartel , 2009 ) or their unfavorable arrangement prior to seed pairing ( Schirle et al . , 2014 ) . In contrast , the seed backbone is pre-organized to favor A-form pairing , with bases of nucleotides 2–5 accessible to nucleate pairing ( Nakanishi et al . , 2012; Schirle and MacRae , 2012 ) . Moreover , perfect pairing propagated through miRNA nucleotide 7 creates the opportunity for favorable contacts to the minor groove of the seed:target duplex ( Schirle et al . , 2014 ) . Our overhaul of the TargetScan website integrated the output of the context++ model with the most current 3′-UTR-isoform data to provide any biologist with an interest in either a miRNA or a potential miRNA target convenient access to the predictions , with an option of downloading code or bulk output suitable for more global analyses . In our continuing efforts to improve the website , several additional functionalities will also soon be provided . To facilitate the exploration of co-targeting networks involving multiple miRNAs ( Tsang et al . , 2010; Hausser and Zavolan , 2014 ) , we will provide the option of ranking predictions based on the simultaneous action of several independent miRNA families , to which relative weights ( e . g . , accounting for relative miRNA expression levels or differential miRNA activity in a cell type of interest ) can be optionally assigned . To offer predictions for transcripts not already in the TargetScan database ( e . g . , novel 3′ UTRs or long non-coding RNAs , including circular RNAs ) , we will provide a mechanism to compute context++ scores interactively for a user-specified transcript . Likewise , to offer predictions for a novel sRNA sequence ( e . g . , off-target predictions for an siRNA ) , we will provide a mechanism to retrieve context++ scores interactively for a user-specified sRNA . To visualize the expression signature that results from perturbing a miRNA , we will provide a tool for the user to input mRNA/protein fold changes from high-throughput experiments and obtain a cumulative distribution plot showing the response of predicted targets relative to that of mRNAs without sites . Thus , with the current and future improvements to TargetScan , we hope to enhance the productivity of miRNA research and the understanding of this intriguing class of regulatory RNAs . A list of microarray , RNA-seq , ribosome profiling , and proteomic datasets used for analyses , as well as the corresponding figures in which they were used , is provided ( Table 2 ) . We considered developing the model using RNA-seq data rather than microarray data , but microarray datasets were still much more plentiful and were equally suitable for measuring the effects of sRNAs . Unless pre-processed microarray data were provided by previous studies ( as indicated in Table 2 ) , raw data were processed using Bioconductor release 2 . 14 in the R programming language v3 . 1 . 1 ( Gentleman et al . , 2004; Development Core Team , 2014 ) . Affymetrix data were first background-corrected with the ‘gcrma’ R package ( Wu et al . , 2004 ) , whereas Illumina BeadArray data from the miR-302 knockdown and miR-522 transfection datasets ( Lipchina et al . , 2011; Tan et al . , 2014 ) were processed and background-corrected using the ‘lumiR’ and ‘lumiExpresso’ functions in the ‘lumi’ R package ( Du et al . , 2008 ) . A robust linear regression model was then used to fit to the probe intensities using the ‘lmFit’ function ( parameter ‘method = ‘robust’’ ) in the ‘limma’ R package v3 . 6 . 9 ( Smyth , 2004 , 2005 ) , computing differential expression information with the provided eBayes function . Probe IDs were then converted to RefSeq or Ensembl IDs ( e . g . , using the hgu133plus2ENSEMBL and IlluminaID2nuID/lumiHumanAllENSEMBL functions to convert Affymetrix and BeadArray probe IDs , respectively ) , and the fold change for each mRNA was computed as the median fold change for all probes corresponding to the mRNA . Finally , because about half of the genes in the genome were either not expressed in the cell type examined , or were expressed at a level that was so close to the background that they were prone to have noisy fold-change measurements , the following filters were applied:i . For microarray datasets examining the effect of either knocking down either miR-92 or 25 miRNA families in HEK293 cells ( Hafner et al . , 2010; Helwak et al . , 2013 ) , transfecting miR-7 or miR-124 into HEK293 cells ( Hausser et al . , 2009 ) , knocking out miR-155 in Th1 or Th2 cells ( Rodriguez et al . , 2007 ) , or transfecting each of the 7 miRNAs in HCT116 cells ( Linsley et al . , 2007 ) , we computed the mean signal for each mRNA ( averaging the signal with and without the miRNA ) , and retained mRNAs exceeding the median of this distribution . ii . For microarray datasets examining the effect of injecting miR-430 into MZDicer embryos ( Giraldez et al . , 2006 ) or knocking out miR-155 in T cells ( Loeb et al . , 2012 ) , we required the mean signal intensity of an mRNA to exceed 3 . 0 and 2 . 5 , respectively . iii . For Illumina BeadArray datasets examining the effect of either knocking down miR-302/367 ( Lipchina et al . , 2011 ) or transfecting miR-522 ( Tan et al . , 2014 ) , we required the mean signal intensity to exceed 7 . 5 and 7 . 0 , respectively . iv . For all 74 small-RNA transfections of HeLa cells , we required mRNA expression levels to exceed 10 reads per million ( RPM ) , as quantified by RNA-seq in mock-transfected HeLa cells ( Guo et al . , 2010 ) . v . For analysis of RNA-seq or RPF datasets examining the effect of either losing Dicer in zebrafish embryos ( Bazzini et al . , 2012 ) , transfecting miR-124 into HEK293 , HeLa , or Huh7 cells ( Nam et al . , 2014 ) , or knocking out miR-155 in B cells ( Eichhorn et al . , 2014 ) , we required mRNA expression levels to exceed 10 RPM , as quantified in the condition lacking the perturbed miRNA . vi . For analysis of proteomic results , we used the pre-computed data provided in the table of significantly detectable peptides ( Selbach et al . , 2008 ) . 10 . 7554/eLife . 05005 . 022Table 2 . Summary of datasets analyzed in this study , and corresponding figures using the datasetsDOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 022FigureGene expression omnibus ( GEO ) ID , ArrayExpress ID , or data sourceReferenceFigure 1A , Figure 1—figure supplement 4AGSM854425 , GSM854430 , GSM854431 , GSM854436 , GSM854437 , GSM854442 , GSM854443 ( Bazzini et al . , 2012 ) Figure 1B , Figure 6BGSM1012118 , GSM1012119 , GSM1012120 , GSM1012121 , GSM1012122 , GSM1012123 ( Loeb et al . , 2012 ) Figure 1C , Figure 1—figure supplement 2A , Figure 6C , DE-TABM-232 ( Rodriguez et al . , 2007 ) Figure 1D , FGSM1122217 , GSM1122218 , GSM1122219 , GSM1122220 , GSM1122221 , GSM1122222 , GSM1122223 , GSM1122224 , GSM1122225 , GSM1122226 ( Helwak et al . , 2013 ) Figure 1E , Figure 1—figure supplement 3A–D , Figure 6I , JGSM538818 , GSM538819 , GSM538820 , GSM538821 ( Hafner et al . , 2010 ) Figure 1GGSM156524 , GSM156532 , GSM210897 , GSM210898 , GSM210901 , GSM210903 , GSM210904 , GSM210907 , GSM210909 , GSM210911 , GSM210913 , GSM37599 , http://psilac . mdc-berlin . de/download/ ( let7b_32h , miR-30_32h , miR-155_32h , miR-16_32h ) ( Lim et al . , 2005; Grimson et al . , 2007; Linsley et al . , 2007; Selbach et al . , 2008 ) Figure 1H , Figure 6K , LE-MTAB-2110 ( Tan et al . , 2014 ) Figure 1I , Figure 1—figure supplement 2B , Figure 6EGSM1479572 , GSM1479576 , GSM1479580 , GSM1479584 ( Eichhorn et al . , 2014 ) Figure 1—figure supplement 1AGSM210897 , GSM210898 , GSM210901 , GSM210903 , GSM210904 , GSM210907 , GSM210909 , GSM210911 , GSM210913 , GSM37599 , GSM37601 ( Lim et al . , 2005; Grimson et al . , 2007 ) Figure 1—figure supplement 1B , Figure 3 , Figure 3—figure supplement 1B , C , Figure 474 datasets compiled in Supplementary data 4 of Garcia et al . ( 2011 ) , used as is or after normalization ( Supplementary file 1 ) ; GSM119707 , GSM119708 , GSM119710 , GSM119743 , GSM119745 , GSM119746 , GSM119747 , GSM119749 , GSM119750 , GSM119759 , GSM119761 , GSM119762 , GSM119763 , GSM133685 , GSM133689 , GSM133699 , GSM133700 , GSM134325 , GSM134327 , GSM134466 , GSM134480 , GSM134483 , GSM134485 , GSM134511 , GSM134512 , GSM134551 , GSM210897 , GSM210898 , GSM210901 , GSM210903 , GSM210904 , GSM210907 , GSM210909 , GSM210911 , GSM210913 , GSM37599 , GSM37601; E-MEXP-1402 ( 1595297366 , 1595297383 , 1595297389 , 1595297394 , 1595297399 , 1595297422 , 1595297427 , 1595297432 , 1595297491 , 1595297496 , 1595297501 , 1595297507 , 1595297513 , 1595297518 , 1595297524 , 1595297530 , 1595297535 , 1595297564 , 1595297588 , 1595297595 , 1595297605 , 1595297614 , 1595297621 , 1595297627 , 1595297644 , 1595297650 , 1595297662 ) ; E-MEXP-668 ( 16012097016666 , 16012097016667 , 16012097016668 , 16012097016669 , 16012097017938 , 16012097017939 , 16012097017952 , 16012097017953 , 16012097018568 , 251209725411 ) ( Lim et al . , 2005; Birmingham et al . , 2006; Schwarz et al . , 2006; Jackson et al . , 2006a; Jackson et al . , 2006b; Grimson et al . , 2007; Anderson et al . , 2008 ) Figure 1—figure supplement 1CGSM95614 , GSM95615 , GSM95616 , GSM95617 , GSM95618 , GSM95619 ( Giraldez et al . , 2006 ) Figure 1—figure supplement 1D , FGSM1269344 , GSM1269345 , GSM1269348 , GSM1269349 , GSM1269350 , GSM1269351 , GSM1269354 , GSM1269355 , GSM1269356 , GSM1269357 , GSM1269360 , GSM1269361 , GSM1269362 , GSM1269363 ( Nam et al . , 2014 ) Figure 1—figure supplement 3E , Figure 6Hhttp://icb . med . cornell . edu/faculty/betel/lab/betelab_v1/Data . html ( Lipchina et al . , 2011 ) Figure 1—figure supplement 4Bhttp://psilac . mdc-berlin . de/media/database/release-1 . 0/protein/pSILAC_all_protein_ratios_OE . txt ( miR155 ) ( Selbach et al . , 2008 ) Figure 3—figure supplement 1AGSM416753 ( Mayr and Bartel , 2009 ) Figure 5 , Figure 5—figure supplement 1GSM156522 , GSM156580 , GSM156557 , GSM156548 , GSM156533 , GSM156532 , GSM156524 , processed and normalized ( Supplementary file 2 ) ( Linsley et al . , 2007 ) Figure 6AGSM37601 ( Lim et al . , 2005 ) Figure 6F , GGSM363763 , GSM363766 , GSM363769 , GSM363772 , GSM363775 , GSM363778 ( Hausser et al . , 2009 ) These thresholds were chosen based upon visual inspection of plots evaluating the relationship between mean expression level and fold change ( commonly known as ‘MA plots’ in the context of microarrays ) , attempting to balance the tradeoff between maximal sample size and reduced noise . The overall conclusions were robust to the choice of the threshold . After imposing the threshold , all fold-change values were centered by subtracting the median fold-change value of the ‘no-site’ mRNAs in each sRNA perturbation experiment , except in the case of Figure 5—figure supplement 1B , C , in which data were mean-centered . When available , target genes identified using high-throughput CLIP data were collected from the supplemental materials of the corresponding studies ( Lipchina et al . , 2011; Loeb et al . , 2012; Helwak et al . , 2013; Grosswendt et al . , 2014 ) . For the original PAR-CLIP study ( Hafner et al . , 2010 ) , targets were inferred from an online resource of all endogenous HEK293 clusters ( http://www . mirz . unibas . ch/restricted/clipdata/RESULTS/CLIP_microArray/Antago_mir_vs_ALL_AGO . txt ) as well as clusters observed after transfection of either miR-7 ( http://www . mirz . unibas . ch/restricted/clipdata/RESULTS/miR7_TRANSFECTION/miR7_TRANSFECTION . html ) or miR-124 ( http://www . mirz . unibas . ch/restricted/clipdata/RESULTS/miR124_TRANSFECTION/miR124_TRANSFECTION . html ) . For dCLIP-supported miR-124 sites identified in the original high-throughput CLIP study ( Chi et al . , 2009 ) , we used clusters whose genomic coordinates were provided by SW Chi ( Supplementary file 3 ) , extracting the corresponding sequences using the ‘getfasta’ utility in BEDTools v2 . 20 . 1 ( parameters ‘-s -name -tab ’ ) ( Quinlan and Hall , 2010 ) . When evaluating the function of non-canonical sites supported by CLIP or IMPACT-seq ( Figure 1 and Figure 1—figure supplements 1–4 ) , a cluster ( or CLASH/chimera interaction ) with a 6–8mer site ( but not only an offset-6mer site , unless otherwise indicated in the figure legends ) corresponding to the cognate miRNA was classified as harboring a canonical site . Otherwise , the cluster ( or CLASH/chimera interaction ) was classified as containing a non-canonical site , and the corresponding mRNA was carried forward for functional evaluation as a non-canonical CLIP-supported target if it also had no cognate 6–8mer sites ( but allowing offset-6mer sites ) in its 3′ UTR ( using either RefSeq or Ensembl 3′-UTR annotations as appropriate for the gene IDs published by the CLIP study ) . When comparing the response of canonical CLIP-supported targets to that of TargetScan7 predictions ( Figure 6 ) , the canonical CLIP-supported sites were additionally required to fall within ( and on the same DNA strand as ) annotated 3′ UTRs , as evaluated by the intersectBED utility in BEDTools v2 . 20 . 1 ( parameter ‘-s’ ) ( Quinlan and Hall , 2010 ) . To identify non-canonical modes of binding , all CLASH interactions assigned to a particular miRNA family ( defined as all mature miRNA sequences sharing a common sequence in nucleotide positions 2–8 ) were collected . Interactions containing the cognate canonical site type ( offset 6mer , 6mer , 7mer-m8 , 7mer-A1 , or 8mer ) were removed . For all miRNA families with at least 50 unique CLASH interactions remaining , enriched motifs were evaluated using MEME version 4 . 9 . 0 ( parameters ‘-p 100 -dna -mod zoops -nmotifs 10 -minw 4 -maxw 8 -maxsize 1 , 000 , 000 , 000’ ) ( Bailey and Elkan , 1994 ) . All motifs with an E-value < 10−3 are reported along with their E-values rounded to the nearest log-unit . Instances in which a top-ranked motif exceeded this E-value were also reported if the motif was an approximate complementary match to the miRNA . For each miRNA family , the top motif identified by MEME was aligned to a representative mature miRNA using FIMO ( parameters ‘--norc--motif 1 --thresh 0 . 01’ ) ( Grant et al . , 2011 ) , considering the reverse complement of the mature miRNA with the last nucleotide of this reverse complement changed to an A ( to capture the enrichment of an adenosine across from the 5′ nucleotide of a miRNA , as occurs in 8mer and 7mer-A1 sites ) . Logos were also manually examined to determine if any mapped to the mature miRNA with a bulged nucleotide . The same procedure was performed for chimera interactions , for dCLIP clusters reported for miR-124 and miR-155 , and for IMPACT-seq clusters reported for miR-522 . For each of the 74 transfection experiments of the compendium ( Table 2 ) , data were first partitioned into the mRNA fold changes ( log2 ) measured in the given experiment ( the response variable ) as well as a matrix of the corresponding mRNA fold changes for the remaining 73 datasets ( the predictor variables ) . A PLSR model was then trained to predict the response using information from the predictor variables . When training the model , PLSR took into account the correlated structure of the predictor matrix , decomposing it into a low-dimensional representation that maximally explained the response variable . Stating the procedure more formally , let Z be an n x m matrix consisting of log2 ( mRNA fold change ) measurements of n mRNAs in response to the sRNA transfected in each of m experiments . Let yi represent measurements for all mRNAs in the ith experiment of Z , and Xī represent measurements for all mRNAs from all experiments except for the ith experiment in Z . Finally , let Tī be a matrix with identical dimensions as Xī , with entries tj , k = 1 if the 3′ UTR of mRNA j in Xī contains a canonical 7–8 nt match to the small RNA transfected in experiment k in Xī , and tj , k = 0 otherwise . Missing values in Z represent cases in which the mRNA signal in the microarray was too low to be reliably measured . The following algorithm was used to normalize each yi for i ∈ {1…74}:i . For values in Tī in which tj , k = 1 , the corresponding value xj , k in Xī was removed , which prevented the loss of signal in yi , j due to sRNA-mediated regulation of the mRNA in two independent experiments . ii . mRNAs in yi , Xī , and Tī were removed if the log2 ( mRNA fold change ) was either undefined in yi or undefined in greater than 50% of experiments in Xī . iii . For the remaining missing values in Xī , values were imputed using the k-nearest neighbors algorithm , using k = 20 , as implemented in the impute . knn function in the ‘impute’ R package ( Troyanskaya et al . , 2001 ) . Results were robust to the choice of imputation algorithm ( data not shown ) . iv . To remove biases afflicting yi , yi was predicted from Xī using partial least squares regression , as implemented in the plsr function in the ‘pls’ R package ( Mevik and Wehrens , 2007 ) . Ten-fold cross-validation was used to choose an appropriate number of components in the regression . Values of yi were then adjusted to their residuals as such: yi ← yi − ŷi , where ŷi was the vector of predicted values of yi from the regression ( Supplementary file 1 ) . An analogous normalization procedure was performed for each of the seven transfection experiments of the test set ( Supplementary file 2 ) . 3′ UTRs were folded locally using RNAplfold ( Bernhart et al . , 2006 ) , allowing the maximal span of a base pair to be 40 nucleotides , and averaging pair probabilities over an 80 nt window ( parameters -L 40 -W 80 ) , parameters found to be optimal when evaluating siRNA efficacy ( Tafer et al . , 2008 ) . For each position 15 nt upstream and downstream of a target site , and for 1–15 nt windows beginning at each position , the partial correlation of the log10 ( unpaired probability ) to the log2 ( mRNA fold change ) associated with the site was plotted , controlling for known determinants of targeting used in the context+ model , which include min_dist , local_AU , 3P_score , SPS , and TA ( Garcia et al . , 2011 ) . For the final predicted SA score used as a feature , we computed the log10 of the probability that a 14-nt segment centered on the match to sRNA positions 7 and 8 was unpaired . We updated human PCT scores using the following datasets: ( i ) 3′ UTRs derived from 19 , 800 human protein-coding genes annotated in Gencode version 19 ( Harrow et al . , 2012 ) , and ( ii ) 3′-UTR multiple sequence alignments ( MSAs ) across 84 vertebrate species derived from the 100-way multiz alignments in the UCSC genome browser , which used the human genome release hg19 as a reference species ( Kent et al . , 2002; Karolchik et al . , 2014 ) . We used only 84 of the 100 species because , with the exception of coelacanth ( a lobe-finned fish more related to the tetrapods ) , the fish species were excluded due to their poor quality of alignment within 3′ UTRs . Likewise , we updated the mouse scores using: ( i ) 3′ UTRs derived from 19 , 699 mouse protein-coding genes annotated in Ensembl 77 ( Flicek et al . , 2014 ) , and ( ii ) 3′-UTR MSAs across 52 vertebrate species derived from the 60-way multiz alignments in the UCSC genome browser , which used the mouse genome release mm10 as a reference species ( Kent et al . , 2002; Karolchik et al . , 2014 ) . As before , we partitioned 3′ UTRs into ten conservation bins based upon the median branch-length score ( BLS ) of the reference-species nucleotides ( Friedman et al . , 2009 ) . However , to estimate branch lengths of the phylogenetic trees for each bin , we concatenated alignments within each bin using the ‘msa_view’ utility in the PHAST package v1 . 1 ( parameters ‘--unordered-ss--in-format SS--out-format SS--aggregate $species_list--seqs $species_subset’ , where $species_list contains the entire species tree topology and $species_subset contains the topology of the subtree spanning the placental mammals ) ( Siepel and Haussler , 2004 ) . We then fit trees for each bin using the ‘phyloFit’ utility in the PHAST package v1 . 1 , utilizing the generalized time-reversible substitution model and a fixed-tree topology provided by UCSC ( parameters ‘-i SS--subst-mod REV--tree $tree’ , where $tree is the Newick format tree of the placental mammals ) ( Siepel and Haussler , 2004 ) . PCT parameters and scores were then calculated as described , estimating the signal of conservation for each seed family relative to that of its corresponding 50 control k-mers , matched for k-mer length and rate of dinucleotide conservation at varying branch-length windows ( Friedman et al . , 2009 ) . All phylogenetic trees and PCT parameters are available for download at the TargetScan website ( targetscan . org ) . The mRNAs were selected to avoid those from genes with multiple highly expressed alternative 3′-UTR isoforms , which would have otherwise obscured the accurate measurement of features such as len_3UTR or min_dist , and also created situations in which the response was diminished because some isoforms lacked the target site . HeLa 3P-seq results ( Nam et al . , 2014 ) were used to identify genes in which a dominant 3′-UTR isoform comprised ≥90% of the transcripts ( Supplementary file 1 ) . For each of these genes , the mRNA with the dominant 3′-UTR isoform was carried forward , together with the ORF and 5′-UTR annotations previously chosen from RefSeq ( Garcia et al . , 2011 ) . Sequences of these mRNA models are provided as Supplemental material at http://bartellab . wi . mit . edu/publication . html . To prevent the presence of multiple 3′-UTR sites to the transfected sRNA from confounding attribution of an mRNA change to an individual site , these mRNAs were further filtered within each dataset to consider only mRNAs that contained a single 3′-UTR site ( either an 8mer , 7mer-m8 , 7mer-A1 , or 6mer ) to the cognate sRNA . Features that exhibited skewed distributions , such as len_5UTR , len_ORF , and len_3UTR were log10 transformed ( Table 1 ) , which made their distributions approximately normal . These and other continuous features were then normalized to the ( 0 , 1 ) interval as described ( e . g . , see Supplementary Figure 5 in Garcia et al . , 2011 ) , except a trimmed normalization was implemented to prevent outlier values from distorting the normalized distributions . For each value , the 5th percentile of the feature was subtracted from the value , and the resulting quantity was divided by the difference between the 95th and 5th percentiles of the feature . Percentile values are provided for the subset of continuous features that were scaled ( Table 3 ) . The trimmed normalization facilitated comparison of the contributions of different features to the model , with absolute values of the coefficients serving as a rough indication of their relative importance . 10 . 7554/eLife . 05005 . 023Table 3 . Scaling parameters used to normalize data to the ( 0 , 1 ) intervalDOI: http://dx . doi . org/10 . 7554/eLife . 05005 . 023Feature8mer7mer-m87mer-A16mer5th %95th %5th %95th %5th %95th %5th %95th %3P_score1 . 0003 . 5001 . 0003 . 5001 . 0003 . 5001 . 0003 . 500SPS−11 . 130−5 . 520−11 . 130−5 . 490−8 . 410−3 . 330−8 . 570−3 . 330TA_3UTR3 . 1133 . 8653 . 0673 . 8873 . 1453 . 8873 . 1133 . 887Len_3UTR2 . 3923 . 6372 . 4093 . 6152 . 4133 . 6302 . 4053 . 620Len_ORF2 . 7883 . 7532 . 7733 . 7292 . 7733 . 7302 . 7753 . 731Min_dist1 . 4153 . 1131 . 4913 . 0961 . 4313 . 1171 . 4773 . 106Local_AU0 . 3080 . 8140 . 2770 . 7820 . 3420 . 8010 . 2950 . 772SA−4 . 356−0 . 661−5 . 218−0 . 725−4 . 230−0 . 588−5 . 082−0 . 666PCT0 . 0000 . 8160 . 0000 . 3640 . 0000 . 4490 . 0000 . 193Provided are the 5th and 95th percentile values for continuous features that were scaled , after the values of the feature were appropriately transformed as indicated ( Table 1 ) . We generated 1000 bootstrap samples , each including 70% of the data from each transfection experiment of the compendium of 74 datasets ( Supplementary file 1 ) , with the remaining data reserved as a held-out test set . For each bootstrap sample , stepwise regression , as implemented in the stepAIC function from the ‘MASS’ R package ( Venables and Ripley , 2002 ) , was used to both select the most informative combination of features and train a model . Feature selection maximized the Akaike information criterion ( AIC ) , defined as: −2 ln ( L ) + 2k , where L was the likelihood of the data given the linear regression model and k was the number of features or parameters selected . The 1000 resulting models were each evaluated based on their r2 to the corresponding test set . To illustrate the utility of adding features not included in our previous models , these r2 values were compared to those obtained when re-training the multiple linear regression coefficients on each bootstrap sample using only the features of either the context-only or the context+ model , and computing r2 values on the corresponding test sets . The stepwise regression was implemented independently for each of the site types , and a final set of features was chosen as those that were selected for at least 99% of the bootstrap samples of at least two site types . Using this group of features and the entire compendium of 74 datasets as a training set , we trained a multiple linear regression model for each site type ( Figure 4—source data 1 ) . As done previously for TargetScan6 predictions , scores for 8mer , 7mer-m8 , 7mer-A1 , and 6mer sites were bounded to be no greater than −0 . 03 , −0 . 02 , −0 . 01 , and 0 , respectively , thereby creating a piece-wise linear function for each site type . To compare predictions from different miRNA target prediction tools , we collected the following freely downloadable predictions: AnTar ( predictions from either miRNA-transfection or CLIP-seq models ) ( Wen et al . , 2011 ) , DIANA-microT-CDS ( September 2013 ) ( Reczko et al . , 2012 ) , ElMMo v5 ( January 2011 ) ( Gaidatzis et al . , 2007 ) , MBSTAR ( all predictions ) ( Bandyopadhyay et al . , 2015 ) , miRanda-MicroCosm v5 ( Griffiths-Jones et al . , 2008 ) , miRmap v1 . 1 ( September 2013 ) ( Vejnar and Zdobnov , 2012 ) , mirSVR ( August 2010 ) ( Betel et al . , 2010 ) , miRTarget2 ( from miRDB v4 . 0 , January 2012 ) ( Wang , 2008; Wang and El Naqa , 2008 ) , MIRZA-G ( sets predicted either with or without conservation features and either with or without more stringent seed-match requirements , March 2015 ) ( Gumienny and Zavolan , 2015 ) , PACCMIT-CDS ( sets predicted either with or without conservation features ) ( Marin et al . , 2013 ) , PicTar2 ( from the doRiNA web resource; sets conserved to either fish , chicken , or mammals ) ( Krek et al . , 2005; Anders et al . , 2012 ) , PITA Catalog v6 ( 3/15 flank for either ‘All’ or ‘Top’ predictions , August 2008 ) ( Kertesz et al . , 2007 ) , RNA22 ( May 2011 ) ( Miranda et al . , 2006 ) , SVMicrO ( February 2011 ) ( Liu et al . , 2010 ) , TargetRank ( all scores from web server ) ( Nielsen et al . , 2007 ) , TargetSpy ( all predictions ) ( Sturm et al . , 2010 ) , TargetScan v5 . 2 ( either conserved or all predictions , June 2011 ) ( Grimson et al . , 2007 ) , and TargetScan v6 . 2 ( either conserved predictions ranked by the context+ model or all predictions ranked by either the context+ model or PCT scores , June 2012 ) ( Friedman et al . , 2009; Garcia et al . , 2011 ) . For algorithms providing site-level predictions ( i . e . , ElMMo , MBSTAR , miRSVR , PITA , and RNA22 ) , scores were summed within genes or transcripts ( if available ) to acquire an aggregate score . For algorithms providing multiple transcript-level predictions ( i . e . , miRanda-MicroCosm , PACCMIT-CDS , and TargetSpy ) , the transcript with the best score was selected as the representative transcript isoform . In all cases , predictions with gene symbol or Ensembl ID formats were translated into RefSeq format . When computing r2 to the test sets , mRNAs that were not predicted by the algorithm to be a target were assigned the worst score in the range of all scores generated by the algorithm . To build databases of human and mouse 3′-UTR profiles , we began with the ‘basic’ set of protein-coding gene models deposited in Gencode v19 ( human hg19 assembly ) and Gencode vM3 ( mouse mm10 assembly ) , respectively ( Harrow et al . , 2012 ) . For each unique stop codon in each set of gene models , we selected the transcript with the longest 3′ UTR as its representative transcript . If other datasets indicated that the 3′ UTRs of these representative transcripts have longer tandem isoforms , we extended them accordingly , using additional annotations provided by ( i ) the ‘comprehensive’ set of Gencode gene models ( Harrow et al . , 2012 ) , ( ii ) all mRNAs in the RefSeq database ( Pruitt et al . , 2012 ) , downloaded from the refGene database through the UCSC table browser ( Kent et al . , 2002 ) , and ( iii ) 3′-UTR extensions supported by RNA-seq evidence ( Miura et al . , 2013 ) , after transforming mm9 to mm10 coordinates using liftOver ( Hinrichs et al . , 2006 ) . We then used 3P-seq clusters from human and mouse ( Nam et al . , 2014 ) ( again after transforming coordinates with liftOver ) to further extend 3′ UTRs when possible , searching within a 5400 nt region downstream of the stop codon ( excluding the regions containing annotated introns ) for a cleavage and polyadenylation site supported by at least one 3P-seq cluster , prohibiting the search to extend beyond the start position of any annotated downstream exon . The 5400 nt window was chosen because the 99th percentile of the lengths of previously annotated mouse and human 3′ UTRs was ∼5400 nt . Zebrafish 3′ UTRs for TargetScanFish were identical to those annotated previously ( Ulitsky et al . , 2012 ) . For each representative transcript , 3P-seq clusters mapping within the extended 3′ UTR were used to quantify the relative levels of alternative tandem isoforms , thereby generating a 3′-UTR profile . For human and mouse transcripts , all 3P-seq datasets for cell lines/tissues of each species were combined , after normalizing for the sequencing depth ( i . e . , number of uniquely mapping tags ) of each dataset , to generate meta profiles . To perform this normalization , the number of tags overlapping the 3′ UTR of each annotated transcript was first summed . A matrix of summed tag counts for each cell line/tissue and for each transcript was then compiled , removing transcripts with no tags in any cell type . This matrix was then upper-quartile normalized by calculating the 75th quantile of counts in each cell type , using the calcNormFactors function ( parameter ‘method = ‘upperquartile’’ ) in the ‘edgeR’ R package ( Robinson et al . , 2010 ) . These scaling factors were then applied to all tags , and the normalized tag counts corresponding to each 3P-seq cluster from different cell lines/tissues were summed . A pseudocount of 0 . 1 tag was assigned to the longest tandem 3′-UTR isoform , which accommodated cases in which the longest annotated 3′ UTR did not have tag support . In addition , 5 pseudocounts were assigned to the longest Gencode 3′-UTR isoform , which gave preference to this Gencode annotation if the UTR had poor 3P-seq coverage . The 3′-UTR profiles were then generated and used to compute the affected isoform ratio ( AIR ) and weighted context++ score for each predicted target site as depicted in Figures 2A , 3A , respectively , of Nam et al . ( 2014 ) . For zebrafish transcripts , profiles were generated for each developmental stage with a 3P-seq dataset . All input and output annotation files as well as scripts are available for download at TargetScan ( targetscan . org ) . When partitioning miRNA families according to their conservation level , we began with a high-confidence set of human miRNAs supported by small-RNA sequencing ( T Tuschl , personal communication ) that shared nucleotides 2–8 with a mouse miRNA supported by small-RNA sequencing ( Chiang et al . , 2010 ) . We then extracted 100-way multiz alignments of each mature miRNA from the UCSC Genome Browser and counted the number of species for which nucleotides 2–8 of the miRNA did not change . As an initial pass , those conserved among ≥40 species were classified as mammalian conserved , and those conserved among >60 species were classified as more broadly conserved among vertebrate species . Due to poorer quality alignments for more distantly related species , this procedure misclassified several more broadly conserved miRNAs as mammalian conserved . Therefore , mammalian conserved miRNAs that aligned with >90% homology to a mature miRNA from chicken , frog , or zebrafish , as annotated in miRBase release 21 ( Kozomara and Griffiths-Jones , 2014 ) , were re-classified as more broadly conserved . In addition , miR-489 was included in the broadly conserved set of TargetScanHuman ( but not TargetScanMouse ) despite having a seed substitution in mouse . Some mammalian pri-miRNAs give rise to two or three abundant miRNA isoforms that have different seeds , either because both strands of the miRNA duplex load into Argonaute with near-equal efficiencies or because processing heterogeneity gives rise to alternative 5′ termini ( Azuma-Mukai et al . , 2008; Morin et al . , 2008; Wu et al . , 2009; Chiang et al . , 2010 ) . To annotate these abundant alternative isoforms , we identified all isoforms expressed at ≥33% of the level of the most abundant isoform , as determined from high-throughput sequencing ( allowing for 3′ heterogeneity within each isoform ) . These isoforms were carried forward as mammalian conserved isoforms if they also satisfied this property in the mouse small-RNA sequencing data ( Chiang et al . , 2010 ) , and as broadly conserved isoforms if they satisfied this property in zebrafish small-RNA sequencing data available in miRBase release 21 . Adhering to the miRNA naming convention , if two isoforms mapped to the 5′ and 3′ arms of the hairpin they were named ‘–5p’ and ‘–3p’ , respectively , and if two isoforms were processed from the same arm they were named ‘ . 1’ and ‘ . 2’ in decreasing order of their abundance , as detected in the human . All mature miRNAs were downloaded from miRBase release 21 ( Kozomara and Griffiths-Jones , 2014 ) . Those that matched a conserved miRNA at nucleotides 2–8 were considered part of that miRNA family . All miRNAs and miRNA isoforms annotated in miRBase but not meeting our criteria for conservation in mammals or beyond were also grouped into families based on the identity of nucleotides 2–8 and were classified as poorly conserved miRNAs ( which included many small RNAs misclassified as miRNAs ) . The miRNA seed families and associated conservation classifications are available for download at TargetScan ( targetscan . org ) . TargetScan ( v7 . 0 ) provides the option of ranking predicted targets of mammalian miRNAs according to either cumulative weighted context++ score ( CWCS ) , which ranks based upon the predicted repression , or aggregate PCT score of the longest 3′-UTR isoform , which ranks based upon the confidence that targeting is evolutionarily conserved ( Figure 7—figure supplement 1 ) . For each predicted target , the CWCS estimated the total repression expected from multiple sites to the same miRNA . This score was calculated using the 3′-UTR profiles to weight the marginal effect of each additional site to the miRNA while also taking into account the predicted mRNA depletion resulting from any downstream sites to the same miRNA . This approach was improved over that we used previously to calculate total wContext+ scores ( Nam et al . , 2014 ) , in that it did not over-estimate the aggregate effect of multiple sites in distal isoforms . For each miRNA family , 8mer , 7mer-m8 , 7mer-A1 , and 6mer sites were first filtered to remove overlapping sites , and for each reference 3′ UTR , nonoverlapping sites to the same miRNA were numbered from 1 to n , starting at the distal end of the 3′ UTR . For each site i , from 1 to n , the cumulative predicted repression at that site ( Ci ) was calculated as Ci = C ( i − 1 ) + ( 1 − 2CSi ) ( AIRi − C ( i − 1 ) ) , in which CSi and AIRi were the context++ score and AIR of site i , and the ( 1 − 2CSi ) ( AIRi − C ( i − 1 ) ) term predicted the marginal repression of site i , in which the predicted repression at the site ( 1 − 2CSi ) was modified based on the fraction of mRNAs containing that site ( AIRi ) as reduced by the mRNA depletion predicted to occur from the action of any more distal sites ( C ( i − 1 ) , assigning C0 as 0 ) . The CWCS was then calculated as log2 ( 1 − Cn ) , in which Cn was the Ci at the most proximal site of the reference 3′ UTR . For each reference 3′ UTR , CWCSs were calculated for each member of a miRNA family , and the score from the member with the greatest predicted repression was chosen to represent that family , and the reference 3′ UTR with the most 3P-seq tags was chosen to represent the gene . When scoring features that can vary with 3′-UTR length ( Min_dist , Len_3UTR , and Off6m ) , a weighted score was used that accounted for the abundance of each 3′-UTR tandem isoform in which the site existed , as estimated from a compendium of 3P-seq datasets from the same species ( Nam et al . , 2014 ) . Although 6mer sites are used to calculate cumulative weighted context++ scores , and 6mer sites are tallied in the tables , the locations of these 6mer sites are not displayed , and targets with only 6mer sites are not listed . When calculating PCT scores , the most abundant 3′-UTR isoform as defined by 3P-seq was used to determine the conservation bin to which the 3′ UTR belonged . Sites corresponding to poorly conserved and mammalian-conserved miRNA seed families or sites overlapping annotated ORF regions were assigned PCT scores of zero . For TargetScanFish , genome-wide alignment quality in zebrafish 3′ UTRs was not of sufficient quality to compute PCT scores , so a PCT value of zero was assigned to all sites when computing context++ scores . All PCT parameters and parameters for tree branch lengths and regression models , along with pre-computed context++ scores for human , mouse , zebrafish , and other vertebrate species are available for download ( targetscan . org ) . Perl scripts using these parameters to compute context++ scores , weighted context++ scores , CWCSs , and aggregate PCT scores are also provided ( targetscan . org ) . Predictions are also made for homologous 3′ UTRs of other vertebrate species , using either human-centric or mouse-centric 3′-UTR definitions and corresponding MSAs .	Proteins are built by using the information contained in molecules of messenger RNA ( mRNA ) . Cells have several ways of controlling the amounts of different proteins they make . For example , a so-called ‘microRNA’ molecule can bind to an mRNA molecule to cause it to be more rapidly degraded and less efficiently used , thereby reducing the amount of protein built from that mRNA . Indeed , microRNAs are thought to help control the amount of protein made from most human genes , and biologists are working to predict the amount of control imparted by each microRNA on each of its mRNA targets . All RNA molecules are made up of a sequence of bases , each commonly known by a single letter—‘A’ , ‘U’ , ‘C’ or ‘G’ . These bases can each pair up with one specific other base—‘A’ pairs with ‘U’ , and ‘C’ pairs with ‘G’ . To direct the repression of an mRNA molecule , a region of the microRNA known as a ‘seed’ binds to a complementary sequence in the target mRNA . ‘Canonical sites’ are regions in the mRNA that contain the exact sequence of partner bases for the bases in the microRNA seed . Some canonical sites are more effective at mRNA control than others . ‘Non-canonical sites’ also exist in which the pairing between the microRNA seed and mRNA does not completely match . Previous work has suggested that many non-canonical sites can also control mRNA degradation and usage . Agarwal et al . first used large experimental datasets from many sources to investigate microRNA activity in more detail . As expected , when mRNAs had canonical sites that matched the microRNA , mRNA levels and usage tended to drop . However , no effect was observed when the mRNAs only had recently identified non-canonical sites . This suggests that microRNAs primarily bind to canonical sites to control protein production . Based on these results , Agarwal et al . further developed a statistical model that predicts the effects of microRNAs binding to canonical sites . The updated model considers 14 different features of the microRNA , microRNA site , or mRNA—including the mRNA sequence around the site—to predict which sites within mRNAs are most effectively targeted by microRNAs . Tests showed that Agarwal et al . 's model was as good as experimental approaches at identifying the effective target sites , and was better than existing computational models . The model has been used to power the latest version of a freely available resource called TargetScan , and so could prove a valuable resource for researchers investigating the many important roles of microRNAs in controlling protein production .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "genetics", "and", "genomics" ]	2015	Predicting effective microRNA target sites in mammalian mRNAs
Ferritins are ubiquitous proteins that oxidise and store iron within a protein shell to protect cells from oxidative damage . We have characterized the structure and function of a new member of the ferritin superfamily that is sequestered within an encapsulin capsid . We show that this encapsulated ferritin ( EncFtn ) has two main alpha helices , which assemble in a metal dependent manner to form a ferroxidase center at a dimer interface . EncFtn adopts an open decameric structure that is topologically distinct from other ferritins . While EncFtn acts as a ferroxidase , it cannot mineralize iron . Conversely , the encapsulin shell associates with iron , but is not enzymatically active , and we demonstrate that EncFtn must be housed within the encapsulin for iron storage . This encapsulin nanocompartment is widely distributed in bacteria and archaea and represents a distinct class of iron storage system , where the oxidation and mineralization of iron are distributed between two proteins . Encapsulin nanocompartments are a family of proteinaceous metabolic compartments that are widely distributed in bacteria and archaea ( Sutter et al . , 2008; Akita et al . , 2007; McHugh et al . , 2014; Contreras et al . , 2014 ) . They share a common architecture , comprising an icosahedral shell formed by the oligomeric assembly of a protein , encapsulin , that is structurally related to the HK97 bacteriophage capsid protein gp5 ( Helgstrand et al . , 2003 ) . Gp5 is known to assemble as a 66 nm diameter icosahedral shell of 420 subunits . In contrast , both the Pyrococcus furiosus ( Akita et al . , 2007 ) and Myxococcus xanthus ( McHugh et al . , 2014 ) encapsulin shell-proteins form 32 nm icosahedra with 180 subunits; while the Thermotoga maritima ( Sutter et al . , 2008 ) encapsulin is smaller still with a 25 nm , 60-subunit icosahedron . The high structural similarity of the encapsulin shell-proteins to gp5 suggests a common evolutionary origin for these proteins ( McHugh et al . , 2014 ) . The genes encoding encapsulin proteins are found downstream of genes for dye-dependent peroxidase ( DyP ) family enzymes ( Roberts et al . , 2011 ) , or encapsulin-associated ferritins ( EncFtn ) ( He and Marles-Wright , 2015 ) . Enzymes in the DyP family are active against polyphenolic compounds such as azo dyes and lignin breakdown products; although their physiological function and natural substrates are not known ( Roberts et al . , 2011 ) . Ferritin family proteins are found in all kingdoms and have a wide range of activities , including ribonucleotide reductase ( Aberg et al . , 1993 ) , protecting DNA from oxidative damage ( Grant et al . , 1998 ) , and iron storage ( Bradley et al . , 2014 ) . The classical iron storage ferritin nanocages are found in all kingdoms and are essential in eukaryotes; they play a central role in iron homeostasis , where they protect the cell from toxic free Fe2+ by oxidizing it and storing the resulting Fe3+ as ferrihydrite minerals within their central cavity . The encapsulin-associated enzymes are sequestered within the icosahedral shell through interactions between the shell’s inner surface and a short localization sequence ( Gly-Ser-Leu-Lys ) appended to their C-termini ( Sutter et al . , 2008 ) . This motif is well-conserved , and the addition of this sequence to heterologous proteins is sufficient to direct them to the interior of encapsulins ( Rurup et al . , 2014; 2015; Cassidy-Amstutz et al . , 2016 ) . A recent study of the Myxococcus xanthus encapsulin showed that it sequesters a number of different EncFtn proteins and acts as an ‘iron-megastore’ to protect these bacteria from oxidative stress ( McHugh et al . , 2014 ) . At 32 nm in diameter , it is much larger than other members of the ferritin superfamily , such as the 12 nm 24-subunit classical ferritin nanocage and the 8 nm 12-subunit Dps ( DNA-binding protein from starved cells ) complex ( Grant et al . , 1998; Andrews , 2010 ) ; and is thus capable of sequestering up to ten times more iron than these ferritins ( McHugh et al . , 2014 ) . The primary sequences of EncFtn proteins have Glu-X-X-His metal coordination sites , which are shared features of the ferritin family proteins ( Andrews , 2010 ) . Secondary structure prediction identifies two major α-helical regions in these proteins; this is in contrast to other members of the ferritin superfamily , which have four major α-helices ( Supplementary file 1 ) . The ‘half-ferritin’ primary sequence of the EncFtn family and their association with encapsulin nanocompartments suggests a distinct biochemical and structural organization to other ferritin family proteins . The Rhodospirillum rubrum EncFtn protein ( Rru_A0973 ) shares 33% protein sequence identity with the M . xanthus ( MXAN_4464 ) , 53% with the T . maritima ( Tmari_0787 ) , and 29% with the P . furiosus ( PF1192 ) homologues . The GXXH motifs are strictly conserved in each of these species ( Supplementary file 1 ) . Here we investigate the structure and biochemistry of EncFtn in order to understand iron storage within the encapsulin nanocompartment . We have produced recombinant encapsulin ( Enc ) and EncFtn from the aquatic purple-sulfur bacterium R . rubrum , which serves as a model organism for the study of the control of the bacterial nitrogen fixation machinery ( Pope et al . , 1985 ) , in Escherichia coli . Analysis by transmission electron microscopy ( TEM ) indicates that their co-expression leads to the production of an icosahedral nanocompartment with encapsulated EncFtn . The crystal structure of a truncated hexahistidine-tagged variant of the EncFtn protein ( EncFtnsH ) shows that it forms a decameric structure with an annular ‘ring-doughnut’ topology , which is distinct from the four-helical bundles of the 24meric ferritins ( Lawson et al . , 1991 ) and dodecahedral DPS proteins ( Grant et al . , 1998 ) . We identify a symmetrical iron bound ferroxidase center ( FOC ) formed between subunits in the decamer and additional metal-binding sites close to the center of the ring and on the outer surface . We also demonstrate the metal-dependent assembly of EncFtn decamers using native PAGE , analytical gel-filtration , and native mass spectrometry . Biochemical assays show that EncFtn is active as a ferroxidase enzyme . Through site-directed mutagenesis we show that the conserved glutamic acid and histidine residues in the FOC influence protein assembly and activity . We use our combined structural and biochemical data to propose a model for the EncFtn-catalyzed sequestration of iron within the encapsulin shell . We produced recombinant R . rubrum encapsulin nanocompartments in E . coli by co-expression of the encapsulin ( Rru_A0974 ) and EncFtn ( Rru_A0973 ) proteins , and purified these by sucrose gradient ultra-centrifugation ( Figure 1A ) ( Sutter et al . , 2008 ) . TEM imaging of uranyl acetate-stained samples revealed that , when expressed in isolation , the encapsulin protein forms empty compartments with an average diameter of 24 nm ( Figure 1B and Figure 1—figure supplement 1A/C ) , consistent with the appearance and size of the T . maritima encapsulin ( Sutter et al . , 2008 ) . We were not able to resolve any higher-order structures of EncFtn by TEM . Protein purified from co-expression of the encapsulin and EncFtn resulted in 24 nm compartments with regions in the center that exclude stain , consistent with the presence of the EncFtn within the encapsulin shell ( Figure 1C and Figure 1—figure supplement 1B/C ) . 10 . 7554/eLife . 18972 . 003Figure 1 . Purification of recombinant R . rubrum encapsulin nanocompartments . ( A ) Recombinantly expressed encapsulin ( Enc ) and co-expressed EncFtn-Enc were purified by sucrose gradient ultracentrifugation from E . coli B834 ( DE3 ) grown in SeMet medium . Samples were resolved by 18% acrylamide SDS-PAGE; the position of the proteins found in the complexes as resolved on the gel are shown with arrows . ( B/C ) Negative stain TEM image of recombinant encapsulin and EncFtn-Enc nanocompartments . Samples were imaged at 143 , 000 x magnification , with scale bar shown as 25 nm . Representative encapsulin and EncFtn-Enc complexes are indicated with red arrows . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 00310 . 7554/eLife . 18972 . 004Figure 1—figure supplement 1 . Full-frame transmission electron micrographs of R . rubrum nanocompartments . ( A/B ) Negative stain TEM image of recombinant R . rubrum encapsulin and EncFtn-Enc nanocompartments . All samples were imaged at 143 , 000 x magnification; the scale bar length corresponds to 50 nm . ( C ) Histogram showing the distribution of nanocompartment diameters . A model Gaussian nonlinear least square function was fitted to the data to obtain a mean diameter of 24 . 6 nm with a standard deviation of 2 . 0 nm for encapsulin ( grey ) and a mean value of 23 . 9 nm with a standard deviation of 2 . 2 nm for co-expressed EncFtn and encapsulin ( EncFtn-Enc , black ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 004 We purified recombinant R . rubrum EncFtn as both the full-length sequence ( 140 amino acids ) and a truncated C-terminal hexahistidine-tagged variant ( amino acids 1–96 plus the tag; herein EncFtnsH ) . In both cases the elution profile from size-exclusion chromatography ( SEC ) displayed two peaks ( Figure 2A ) . SDS-PAGE analysis of fractions from these peaks showed that the high molecular weight peak was partially resistant to SDS and heat-induced denaturation; in contrast , the low molecular weight peak was consistent with monomeric mass of 13 kDa ( Figure 2B ) . MALDI peptide mass fingerprinting of these bands confirmed the identity of both as EncFtn . Inductively coupled plasma mass spectrometry ( ICP-MS ) analysis of the SEC fractions showed 100 times more iron in the oligomeric fraction than the monomer ( Figure 2A , blue scatter points; Table 1 ) , suggesting that EncFtn oligomerization is associated with iron binding . In order to determine the iron-loading stoichiometry in the EncFtn complex , further ICP-MS experiments were performed using selenomethionine ( SeMet ) -labelled protein EncFtn ( Table 1 ) . In these experiments , we observed sub-stoichiometric metal binding , which is in contrast to the classical ferritins ( Le Brun et al . , 2010 ) . Size-exclusion chromatography with multi-angle laser light scattering ( SEC-MALLS ) analysis of samples taken from each peak gave calculated molecular weights consistent with a decamer for the high molecular weight peak and a monomer for the low molecular weight peak ( Figure 2C , Table 2 ) . 10 . 7554/eLife . 18972 . 005Figure 2 . Purification of recombinant R . rubrum EncFtnsH . ( A ) Recombinant SeMet-labeled EncFtnsH produced with 1 mM Fe ( NH4 ) 2 ( SO4 ) 2 in the growth medium was purified by nickel affinity chromatography and size-exclusion chromatography using a Superdex 200 16/60 column ( GE Healthcare ) . Chromatogram traces measured at 280 nm and 315 nm are shown with the results from ICP-MS analysis of the iron content of the fractions collected during the experiment . The peak around 73 ml corresponds to a molecular weight of around 130 kDa when compared to calibration standards; this is consistent with a decamer of EncFtnsH . The small peak at 85 ml corresponds to the 13 kDa monomer compared to the standards . Only the decamer peak contains significant amounts of iron as indicated by the ICP-MS analysis . ( B ) Peak fractions from the gel filtration run were resolved by 15% acrylamide SDS-PAGE and stained with Coomassie blue stain . The bands around 13 kDa and 26 kDa correspond to EncFtnsH , as identified by MALDI peptide mass fingerprinting . The band at 13 kDa is consistent with the monomer mass , while the band at 26 kDa is consistent with a dimer of EncFtnsH . The dimer species only appears in the decamer fractions . ( C ) SEC-MALLS analysis of EncFtnsH from decamer fractions and monomer fractions allows assignment of an average mass of 132 kDa to decamer fractions and 13 kDa to monomer fractions , consistent with decamer and monomer species ( Table 2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 00510 . 7554/eLife . 18972 . 006Table 1 . Determination of the Fe/EncFtnsH protein ratio by ICP-MS . EncFtnsH was purified as a SeMet derivative from E . coli B834 ( DE3 ) cells grown in SeMet medium with 1 mM Fe ( NH4 ) 2 ( SO4 ) 2 . Fractions from SEC were collected , acidified and analysed by ICP-MS . EncFtnsH concentration was calculated based on the presence of two SeMet per mature monomer . Samples where the element was undetectable are labelled with n . d . These data were collected from EncFtnsH fractions from a single gel-filtration run . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 006PeakEncFtnsH retention volume ( ml ) Element concentration ( µM ) Derived EncFtnsHconcentration ( µM ) Derived Fe/ EncFtnsH monomerCaFeZnSeDecamer66 . 5n . d . 6 . 7n . d . 24 . 612 . 30 . 568 . 3n . d . 28 . 4n . d124 . 562 . 30 . 570 . 12 . 993 . 72 . 4301 . 7150 . 90 . 671 . 96 . 9120 . 63 . 7379 . 8189 . 90 . 673 . 71 . 964 . 40 . 8240 . 6120 . 30 . 575 . 50 . 921 . 1n . d . 101 . 750 . 80 . 477 . 3n . d . 6 . 2n . d . 42 . 621 . 30 . 379 . 10 . 12 . 4n . d . 26 . 513 . 30 . 280 . 91 . 01 . 5n . d . 22 . 311 . 20 . 182 . 7n . d . 0 . 2n . d . 29 . 214 . 6n . dMonomer84 . 5n . d . 0 . 1n . d . 34 . 917 . 5n . d86 . 3n . d . n . dn . d . 28 . 914 . 4n . d88 . 1n . d . n . d . n . d . 17 . 48 . 7n . d . 89 . 9n . d . n . d . n . d . 5 . 52 . 8n . d . 91 . 7n . d . n . d . n . d . 0 . 10 . 070 . 210 . 7554/eLife . 18972 . 007Table 2 . Estimates of EncFtnsH molecular weight from SEC-MALLS analysis . EncFtnsH was purified from E . coli BL21 ( DE3 ) grown in minimal medium ( MM ) by nickel affinity chromatography and size-exclusion chromatography . Fractions from two peaks ( decamer and monomer ) were pooled separately ( Figure 1C ) and analysed by SEC-MALLS using a Superdex 200 10/300 GL column ( GE Healthcare ) and Viscotek SEC-MALLS instruments ( Malvern Instruments ) ( Figure 2C ) . The decamer and monomer peaks were both symmetric and monodisperse , allowing the estimation of the molecular weight of the species in these fractions ( Folta-Stogniew , 2006 ) . The molecular weights are quoted to the nearest kDa due to the resolution limit of the instrument . The proteins analyzed by SEC-MALLS came from single protein preparation . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 007Molecular Weight ( kDa ) Decamer peakMonomer peakTheoretical13313EncFtnsH-decamer fractions13215EncFtnsH-monomer fractions12613 We purified EncFtnsH from E . coli grown in MM with or without the addition of 1 mM Fe ( NH4 ) 2 ( SO4 ) 2 . The decamer to monomer ratio in the sample purified from cells grown in iron-supplemented media was 4 . 5 , while that from the iron-free media was 0 . 11 , suggesting that iron induces the oligomerization of EncFtnsH in vivo ( Figure 3A , Table 3 ) . To test the metal-dependent oligomerization of EncFtnsH in vitro , we incubated the protein with various metal cations and subjected samples to analytical SEC and non-denaturing PAGE . Of the metals tested , only Fe2+ , Zn2+ and Co2+ induced the formation of significant amounts of the decamer ( Figure 3B , Figure 3—figure supplement 1/2 ) . While Fe2+ induces the multimerization of EncFtnsH , Fe3+ in the form of FeCl3 does not have this effect on the protein , highlighting the apparent preference this protein has for the ferrous form of iron . To determine if the oligomerization of EncFtnsH was concentration dependent we performed analytical SEC at 90 and 700 µM protein concentration ( Figure 3C ) . At the higher concentration , no increase in the decameric form of EncFtn was observed; however , the shift in the major peak from the position of the monomer species indicated a tendency to dimerize at high concentration . 10 . 7554/eLife . 18972 . 008Figure 3 . Effect of Fe2+ and protein concentration on the oligomeric state of EncFtnsH in solution . ( A ) Recombinant EncFtnsH was purified by Gel filtration Superdex 200 chromatography from E . coli BL21 ( DE3 ) grown in MM or in MM supplemented with 1 mM Fe ( NH4 ) 2 ( SO4 ) 2 ( MM+Fe2+ ) . A higher proportion of decamer ( peak between 65 and 75 ml ) is seen in the sample purified from MM+Fe2+ compared to EncFtnsH-MM , indicating that Fe2+ facilitates the multimerization of EncFtnsH in vivo . ( B ) EncFtnsH-monomer was incubated with one molar equivalent of Fe2+ salts for two hours prior to analytical gel-filtration using a Superdex 200 PC 3 . 2/30 column ( GE Healthcare ) . Both Fe2+ salts tested induced the formation of decamer indicated by the peak between 1 . 2 and 1 . 6 ml . Monomeric and decameric samples of EncFtnsH are shown as controls . Peaks around 0 . 8 ml were seen as protein aggregation . ( C ) Analytical gel filtration of EncFtn monomer at different concentrations to illustrate the effect of protein concentration on multimerization . The major peak shows a shift towards a dimer species at high concentration of protein , but the ratio of this peak ( 1 . 5–1 . 8 ml ) to the decamer peak ( 1 . 2–1 . 5 ml ) does not change when compared to the low concentration sample . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 00810 . 7554/eLife . 18972 . 009Figure 3—figure supplement 1 . Effect of metal ions on the oligomeric state of EncFtnsH in solution . ( A/B ) EncFtnsH-monomer was incubated with one mole equivalent of various metal salts for two hours prior to analytical gel-filtration using a Superdex 200 PC 3 . 2/30 column . Co2+ and Zn2+ induced the formation of the decameric form of EncFtnsH; while Mn2+ , Mg2+ and Fe3+ did not significantly alter the oligomeric state of EncFtnsH . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 00910 . 7554/eLife . 18972 . 010Figure 3—figure supplement 2 . PAGE analysis of the effect of metal ions on the oligomeric state of EncFtnsH . 50 µM EncFtnsH monomer or decamer samples were mixed with equal molar metal ions including Fe2+ , Co2+ , Zn2+ , Mn2+ , Ca2+ , Mg2+ and Fe3+ , which were analyzed by Native PAGE alongside SDS-PAGE . ( A ) 10% Native PAGE analysis of EncFtnsH monomer fractions mixed with various metal solutions; ( B ) 10% Native PAGE analysis of EncFtnsH decamer fractions mixed with various metal solutions; ( C ) 15% SDS-PAGE analysis on the mixtures of EncFtnsH monomer fractions and metal solutions; ( D ) 15% SDS-PAGE analysis on the mixtures of EncFtnsH decamer fractions and metal solutions . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 01010 . 7554/eLife . 18972 . 011Table 3 . Gel-filtration peak area ratios for EncFtnsH decamer and monomer on addition of different metal ions . EncFtnsH was produced in E . coli BL21 ( DE3 ) cultured in MM and MM with 1 mM Fe ( NH4 ) 2 ( SO4 ) 2 ( MM+Fe2+ ) and purified by gel-filtration chromatography using an Superdex 200 16/60 column ( GE Healthcare ) . Monomer fractions of EncFtnsH purified from MM were pooled and run in subsequent analytical gel-filtration runs over the course of three days . Samples of EncFtnsH monomer were incubated with one molar equivalent of metal ion salts at room temperature for two hours before analysis by analytical gel filtration chromatography ( AGF ) using a Superdex 200 10/300 GL column . The area for resulting protein peaks were calculated using the Unicorn software ( GE Healthcare ) ; peak ratios were calculated to quantify the propensity of EncFtnsH to multimerize in the presence of the different metal ions . The change in the ratios of monomer to decamer over the three days of experiments may be a consequence of experimental variability , or the propensity of this protein to equilibrate towards decamer over time . The increased decamer: monomer ratio seen in the presence of Fe2+ , Co2+ , and Zn2+ indicates that these metal ions facilitate multimerization of the EncFtnsH protein , while the other metal ions tested do not appear to induce multimerization . The analytical gel filtration experiment was repeated twice using two independent preparations of protein , of which values calculated from one sample are presented here . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 011MethodSampleMonomer areaDecamer areaDecamer/MonomerGel filtration Superdex 200 chromatographyEncFtnsH-MM64 . 3583 . 60 . 1EncFtnsH-MM+Fe2+1938 . 4426 . 44 . 5Analytical Gel filtration Day1EncFtnsH-decamer fractions20 . 21 . 811 . 2EncFtnsH-monomer fractions2 . 921 . 90 . 1Fe ( NH4 ) 2 ( SO4 ) 2/EncFtnsH-monomer11 . 013 . 00 . 8FeSO4-HCl/EncFtnsH-monomer11 . 311 . 41 . 0Analytical Gel filtration Day2EncFtnsH-monomer fractions8 . 322 . 80 . 4CoCl2/EncFtnsH-monomer17 . 714 . 51 . 2MnCl2/EncFtnsH-monomer3 . 130 . 50 . 1ZnSO4/EncFtnsH-monomer20 . 49 . 02 . 3FeCl3/EncFtnsH-monomer3 . 928 . 60 . 1Analytical Gel filtration Day3EncFtnsH-monomer fractions6 . 323 . 40 . 3MgSO4/EncFtnsH-monomer5 . 830 . 20 . 2Ca acetate/EncFtnsH-monomer5 . 625 . 20 . 2 We determined the crystal structure of EncFtnsH by molecular replacement to 2 . 0 Å resolution ( see Table 1 for X-ray data collection and refinement statistics ) . The crystallographic asymmetric unit contained thirty monomers of EncFtn with visible electron density for residues 7 – 96 in each chain . The protein chains were arranged as three identical annular decamers , each with D5 symmetry . The decamer has a diameter of 7 nm and thickness of 4 nm ( Figure 4A ) . The monomer of EncFtn has an N-terminal 310-helix that precedes two 4 nm long antiparallel α-helices arranged with their long axes at 25° to each other; these helices are followed by a shorter 1 . 4 nm helix projecting at 70° from α2 ( Figure 4B ) . The C-terminal region of the crystallized construct extends from the outer circumference of the ring , indicating that the encapsulin localization sequence in the full-length protein is on the exterior of the ring and is thus free to interact with its binding site on the encapsulin shell protein ( Sutter et al . , 2008 ) . 10 . 7554/eLife . 18972 . 012Figure 4 . Crystal structure of EncFtnsH . ( A ) Overall architecture of EncFtnsH . Transparent solvent accessible surface view with α-helices shown as tubes and bound metal ions as spheres . Alternating subunits are colored blue and green for clarity . The doughnut-like decamer is 7 nm in diameter and 4 . 5 nm thick . ( B ) Monomer of EncFtnsH shown as a secondary structure cartoon . ( C/D ) Dimer interfaces formed in the decameric ring of EncFtnsH . Subunits are shown as secondary structure cartoons and colored blue and green for clarity . Bound metal ions are shown as orange spheres for Fe3+ and grey and white spheres for Ca2+ . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 01210 . 7554/eLife . 18972 . 013Figure 4—figure supplement 1 . Electrostatic surface of EncFtnsH . The solvent accessible surface of EncFtnsH is shown , colored by electrostatic potential as calculated using the APBS plugin in PyMOL . Negatively charged regions are colored red and positive regions in blue , neutral regions in grey . ( A ) View of the surface of the EncFtnsH decamer looking down the central axis . ( B ) Orthogonal view of ( A ) . ( C ) Cutaway view of ( B ) showing the charge distribution within the central cavity . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 013 The monomer of EncFtnsH forms two distinct dimer interfaces within the decamer ( Figure 4 C/D ) . The first dimer is formed from two monomers arranged antiparallel to each other , with α1 from each monomer interacting along their lengths and α3 interdigitating with α2 and α3 of the partner chain . This interface buries one third of the surface area from each partner and is stabilized by thirty hydrogen bonds and fourteen salt bridges ( Figure 4C ) . The second dimer interface forms an antiparallel four-helix bundle between helices 1 and 2 from each monomer ( Figure 4D ) . This interface is less extensive than the first and is stabilized by twenty-one hydrogen bonds , six salt bridges , and a number of metal ions . The arrangement of ten monomers in alternating orientation forms the decamer of EncFtn , which assembles as a pentamer of dimers ( Figure 4A ) . Each monomer lies at 45° relative to the vertical central-axis of the ring , with the N-termini of alternating subunits capping the center of the ring at each end , while the C-termini are arranged around the circumference . The central hole in the ring is 2 . 5 nm at its widest in the center of the complex , and 1 . 5 nm at its narrowest point near the outer surface , although it should be noted that a number of residues at the N-terminus are not visible in the crystallographic electron density and these may occupy the central channel . The surface of the decamer has distinct negatively charged patches , both within the central hole and on the outer circumference , which form spokes through the radius of the complex ( Figure 4—figure supplement 1 ) . The electron density maps of the initial EncFtnsH model displayed significant positive peaks in the mFo-DFc map at the center of the 4-helix bundle dimer ( Figure 5—figure supplement 1 ) . Informed by the ICP-MS data indicating the presence of iron in the protein we collected diffraction data at the experimentally determined iron absorption edge ( 1 . 74 Å ) and calculated an anomalous difference Fourier map using this data . Inspection of this map showed two 10-sigma peaks between residues Glu32 , Glu62 and His65 of two adjacent chains , and a statistically smaller 5-sigma peak between residues Glu31 and Glu34 of the two chains . Modeling metal ions into these peaks and refinement of the anomalous scattering parameters allowed us to identify these as two iron ions and a calcium ion respectively ( Figure 5A ) . An additional region of asymmetric electron density near the di-iron binding site in the mFo-DFc map was modeled as glycolic acid , presumably a breakdown product of the PEG 3350 used for crystallization . This di-iron center has an Fe-Fe distance of 3 . 5 Å , Fe-Glu-O distances between 2 . 3 and 2 . 5 Å , and Fe-His-N distances of 2 . 5 Å ( Figure 5B ) . This coordination geometry is consistent with the di-nuclear ferroxidase center ( FOC ) found in ferritin ( Bertini et al . , 2012 ) . It is interesting to note that although we did not add any additional iron to the crystallization trials , the FOC was fully occupied with iron in the final structure , implying that this site has a very high affinity for iron . 10 . 7554/eLife . 18972 . 014Figure 5 . EncFtnsH metal binding sites . ( A ) Wall-eyed stereo view of the metal-binding dimerization interface of EncFtnsH . Protein residues are shown as sticks with blue and green carbons for the different subunits , iron ions are shown as orange spheres and calcium as grey spheres , and the glycolic acid ligand is shown with yellow carbon atoms coordinated above the di-iron center . The 2mFo-DFc electron density map is shown as a blue mesh contoured at 1 . 5 σ and the NCS-averaged anomalous difference map is shown as an orange mesh and contoured at 10 σ . ( B ) Iron coordination within the FOC including residues Glu32 , Glu62 , His65 and Tyr39 from two chains . Protein and metal ions are shown as in A . Coordination between the protein and iron ions is shown as yellow dashed lines with distances indicated . ( C ) Coordination of calcium within the dimer interface by four glutamic acid residues ( E31 and E34 from two chains ) . The calcium ion is shown as a grey sphere and water molecules involved in the coordination of the calcium ion are shown as crosses . ( D ) Metal coordination site on the outer surface of EncFtnsH . The two calcium ions are coordinated by residues His57 , Glu61 and Glu64 from the two chains of the FOC dimer , and are located at the outer surface of the complex , positioned 10 Å away from the FOC iron . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 01410 . 7554/eLife . 18972 . 015Figure 5—figure supplement 1 . Putative ligand-binding site in EncFtnsH . ( A ) Wall-eyed stereo view of the dimer interface of EncFtn . Protein chains are shown as sticks , with 2mFo-DFc electron density shown in blue mesh and contoured at 1 . 5 σ and mFo-DFc shown in green mesh and contoured at 3 σ . ( B ) Wall-eyed stereo view of putative metal binding site at the external surface of EncFtnsH . Protein chains and electron density maps are shown as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 015 The calcium ion coordinated by Glu31 and Glu34 adopts heptacoordinate geometry , with coordination distances of 2 . 5 Å between the metal ion and carboxylate oxygens of Glu31 and Glu34 ( E31/34-site ) . A number of ordered solvent molecules are also coordinated to this metal ion at a distance of 2 . 5 Å . This heptacoordinate geometry is common in crystal structures with calcium ions ( Figure 5C ) ( Katz et al . , 1996 ) . While ICP-MS indicated that there were negligible amounts of calcium in the purified protein , the presence of 140 mM calcium acetate in the crystallization mother liquor favors the coordination of calcium at this site . The fact that the protein does not multimerize in solution in the presence of Fe3+ may indicate that these metal binding sites have a lower affinity for the ferric form of iron , which is the product of the ferroxidase reaction . A number of additional metal-ions were present at the outer circumference of at least one decamer in the asymmetric unit ( Figure 5D ) . These ions are coordinated by His57 , Glu61 and Glu64 from both chains in the FOC dimer and are 4 . 5 Å apart; Fe-Glu-O distances are between 2 . 5 and 3 . 5 Å and the Fe-His-N distances are 4 and 4 . 5 Å . Structural alignment of the di-iron binding site of EncFtnsH to the FOC of Pseudo-nitzschia multiseries ferritin ( PmFtn , PDB ID: 4ITW ) reveals a striking similarity between the metal binding sites of EncFtnsH and the classical ferritins ( Pfaffen et al . , 2013 ) ( Figure 6A ) . The di-iron site of EncFtnsH is by necessity symmetrical , as it is formed through a dimer interface , while the FOC of ferritin does not have these constraints and varies in different species at a position equivalent to His65 of the second EncFtn monomer in the FOC interface ( His65’ ) ( Figure 6A ) . Structural superimposition of the FOCs of ferritin and EncFtn brings the four-helix bundle of the ferritin fold into close alignment with the EncFtn dimer , showing that the two families of proteins have essentially the same architecture around the di-iron center ( Figure 6B ) . The linker connecting helices 2 and 3 of ferritin is congruent with the start of the C-terminal helix of one EncFtn monomer and the N-terminal 310 helix of the second monomer ( Figure 6C ) . 10 . 7554/eLife . 18972 . 016Figure 6 . Comparison of the symmetric metal ion binding site of EncFtnsH and the ferritin FOC . ( A ) Structural alignment of the FOC residues in a dimer of EncFtnsH ( green/blue ) with a monomer of Pseudo-nitzschia multiseries ferritin ( PmFtn ) ( PDBID: 4ITW ) ( orange ) ( Pfaffen et al . , 2013 ) . Iron ions are shown as orange spheres and a single calcium ion as a grey sphere . Residues within the FOC are conserved between EncFtn and ferritin PmFtn , with the exception of residues in the position equivalent to H65’ in the second subunit in the dimer ( blue ) . The site in EncFtn with bound calcium is not present in other family members . ( B ) Secondary structure of aligned dimeric EncFtnsH and monomeric ferritin highlighting the conserved four-helix bundle . EncFtnsH monomers are shown in green and blue and aligned PmFtn monomer in orange as in A . ( C ) Cartoon of secondary structure elements in EncFtn dimer and ferritin . In the dimer of EncFtn that forms the FOC , the C-terminus of the first monomer ( green ) and N-terminus of the second monomer ( blue ) correspond to the position of the long linker between α2 and α3 in ferritin PmFtn . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 01610 . 7554/eLife . 18972 . 017Figure 6—figure supplement 1 . Comparison of quaternary structure of EncFtnsH and ferritin . ( A ) Aligned FOC of EncFtnsH and Pseudo-nitzschia multiseries ferritin ( PmFtn ) ( Pfaffen et al . , 2013 ) . The metal binding site residues from two EncFtnsH chains are shown in green and blue , while the PmFtn is shown in orange . Fe2+ in the FOC is shown as orange spheres and Ca2+ in EncFtnsH is shown as a grey sphere . The two-fold symmetry axis of the EncFtn FOC is shown with a grey arrow ( B ) Cross-section surface view of quaternary structure of EncFtnsH and PmFtn as aligned in ( A ) ( dashed black box ) . The central channel of EncFtnsH is spatially equivalent to the outer surface of ferritin and its outer surface corresponds to the mineralization surface within ferritin . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 017 In order to confirm the assignment of the oligomeric state of EncFtnsH and investigate further the Fe2+-dependent assembly , we used native nano-electrospray ionization ( nESI ) and ion-mobility mass spectrometry ( IM-MS ) . As described above , by recombinant production of EncFtnsH in minimal media we were able to limit the bioavailability of iron . Native MS analysis of EncFtnsH produced in this way displayed a charge state distribution consistent with an EncFtnsH monomer ( blue circles , Figure 7A1 ) with an average neutral mass of 13 , 194 Da , in agreement with the predicted mass of the EncFtnsH protein ( 13 , 194 . 53 Da ) . Under these conditions , no significant higher order assembly was observed and the protein did not have any coordinated metal ions . Titration with Fe2+ directly before native MS analysis resulted in the appearance of a new charge state distribution , consistent with an EncFtnsH decameric assembly ( +22 to +26; 132 . 65 kDa ) ( Figure 7A2/3 ) . After instrument optimization , the mass resolving power achieved was sufficient to assign iron-loading in the complex to between 10 and 15 Fe ions per decamer ( Figure 7B , inset top right ) , consistent with the presence of 10 irons in the FOC and the coordination of iron in the Glu31/34-site occupied by calcium in the crystal structure ( Δmass observed ~0 . 67 kDa ) . MS analysis of EncFtnsH after addition of further Fe2+ did not result in iron loading above this stoichiometry . Therefore , the extent of iron binding seen is limited to the FOC and Glu31/34 secondary metal binding site . These data suggest that the decameric assembly of EncFtnsH does not accrue iron in the same manner as classical ferritin , which is able to sequester around 4500 iron ions within its nanocage ( Mann et al . , 1986 ) . Ion mobility analysis of the EncFtnsH decameric assembly , collected with minimal collisional activation , suggested that it consists of a single conformation with a collision cross section ( CCS ) of 58 . 2 nm2 ( Figure 7B ) . This observation is in agreement with the calculated CCS of 58 . 7 nm2derived from our crystal structure of the EncFtnsH decamer ( Marklund , 2015 ) . By contrast , IM-MS measurements of the monomeric EncFtnsH at pH 8 . 0 under the same instrumental conditions revealed that the metal-free protein monomer exists in a wide range of charge states ( +6 to +16 ) and adopts many conformations in the gas phase with collision cross sections ranging from 12 nm2 to 26 nm2 ( Figure 7—figure supplement 1 ) . These observations are indicative of an unstructured protein with little secondary or tertiary structure ( Beveridge et al . , 2014 ) . Thus , IM-MS studies highlight that higher order structure in EncFtnsH is mediated/stabilized by metal binding , an observation that is in agreement with our solution studies . Taken together , these results suggest that di-iron binding , forming the FOC in EncFtnsH , is required to stabilize the 4-helix bundle dimer interface , essentially reconstructing the classical ferritin-like fold; once stabilized , these dimers readily associate as pentamers , and the overall assembly adopts the decameric ring arrangement observed in the crystal structure . 10 . 7554/eLife . 18972 . 018Figure 7 . Native mass spectrometry and ion mobility analysis of iron loading in EncFtnsH . All spectra were acquired in 100 mM ammonium acetate , pH 8 . 0 with a protein concentration of 5 µM . ( A ) Native nanoelectrospray ionization ( nESI ) mass spectrometry of EncFtnsH at varying iron concentrations . A1 , nESI spectrum of iron-free EncFtnsH displays a charge state distribution consistent with EncFtnsH monomer ( blue circles , 13 , 194 Da ) . Addition of 100 µM ( A2 ) and 300 µM ( A3 ) Fe2+ results in the appearance of a second higher molecular weight charge state distribution consistent with a decameric assembly of EncFtnsH ( green circles , 132 . 6 kDa ) . ( B ) Ion mobility ( IM ) -MS of the iron-bound holo-EncFtnsH decamer . Top , Peaks corresponding to the 22+ to 26+ charge states of a homo-decameric assembly of EncFtnsH are observed ( 132 . 6 kDa ) . Top Insert , Analysis of the 24+ charge state of the assembly at m/z 5528 . 2 Th . The theoretical average m/z of the 24+ charge state with no additional metals bound is marked by a red line ( 5498 . 7 Th ) ; the observed m/z of the 24+ charge state indicates that the EncFtnsH assembly binds between 10 ( green line , 5521 . 1 Th ) and 15 Fe ions ( blue line , 5532 . 4 Th ) per decamer . Bottom , The arrival time distributions ( ion mobility data ) of all ions in the EncFtnsH charge state distribution displayed as a greyscale heat map ( linear intensity scale ) . Bottom right , The arrival time distribution of the 24+ charge state ( dashed blue box ) has been extracted and plotted . The drift time for this ion is shown ( ms ) , along with the calibrated collision cross section ( CCS ) , Ω ( nm2 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 01810 . 7554/eLife . 18972 . 019Figure 7—figure supplement 1 . Native IM-MS analysis of the apo-EncFtnsH monomer . ( A ) Mass spectrum of apo-EncFtnsH acquired from 100 mM ammonium acetate pH 8 . 0 under native MS conditions . The charge state distribution observed is bimodal , with peaks corresponding to the 6+ to 15+ charge states of apo-monomer EncFtnsH ( neutral average mass 13 , 194 . 3 Da ) . ( B ) The arrival time distributions ( ion mobility data ) of all ions in the apo-EncFtnsH charge state distribution displayed as a greyscale heat map ( linear intensity scale ) . ( B ) Right , the arrival time distribution of the 6+ ( orange ) and 7+ ( green ) charge state ( dashed colored‐box ) has been extracted and plotted; The arrival time distributions for these ion is shown ( ms ) , along with the calibrated collision cross section , Ω ( nm2 ) . ( C ) The collision cross section of a single monomer unit from the crystal structure of the Fe-loaded EncFtnsH decamer was calculated to be 15 . 8 nm2 using IMPACT v . 0 . 9 . 1 . The +8 to +15 protein charge states have observed CCS between 20–26 nm2 , which is significantly higher than the calculated CCS for an EncFtnsH monomer taken from the decameric assembly crystal structure ( 15 . 8 nm2 ) . The mobility of the +7 charge state displays broad drift-time distribution with maxima consistent with CCS of 15 . 9 and 17 . 9 nm2 . Finally , the 6+ charge state of EncFtnsH has mobility consistent with a CCS of 12 . 3 nm2 , indicating a more compact/collapsed structure . It is clear from this data that apo-EncFtnsH exists in several gas phase conformations . The range of charge states occupied by the protein ( 6+ to 15+ ) and the range of CCS in which the protein is observed ( 12 . 3 nm2 – 26 nm2 ) are both large . In addition , many of the charge states observed have higher charge than the theoretical maximal charge on spherical globular protein , as determined by the De La Mora relationship ( ZR = 0 . 0778√m; for the EncFtnsH monomer ZR = 8 . 9 ) Fernandez ( Fernandez de la Mora , 2000 ) . As described by Beveridge et al . , all these factors are indicative of a disordered protein ( Beveridge et al . , 2014 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 01910 . 7554/eLife . 18972 . 020Figure 7—figure supplement 2 . Gas-phase disassembly of the holo-EncFtnsH decameric assembly . The entire charge state distribution of the Fe-loaded holo- EncFtnsH assembly ( green circles ) was subject to collisional-induced dissociation ( CID ) by increasing the source cone voltage to 200 V and the trap voltage to 50 V . The resulting CID mass spectrum ( A ) revealed that dissociation of the holo- EncFtnsH decamer primarily occurred via ejection of a highly charged monomer ( blue circles ) , leaving the ‘stripped’ complex ( a 9mer; 118 . 7 kDa; yellow circles ) . The mass of the ejected-monomer is consistent with apo- EncFtnsH ( 13 . 2 kDa ) , suggesting unfolding of the monomer ( and loss of Fe ) occurs during ejection from the complex . This observation of asymmetric charge partitioning of the sub-complexes with respect to the mass of the complex is consistent with the 'typical' pathway of dissociation of protein assemblies by CID , as described by Hall et al . ( 2013 ) . In addition , a third , lower abundance , charge state distribution is observed which overlaps the EncFtn ejected monomer charge state distribution; this region of the spectrum is highlighted in ( B ) . This distribution is consistent with an ejected EncFtnsH dimer ( orange circles ) . Interestingly , closer analysis of the individual charge state of this dimeric CID product shows that this sub-complex exists in three forms – displaying mass consistent with an EncFtnsH dimer binding 0 , 1 , and 2 Fe ions . This is highlighted in ( C ) , where the 15+ charge state of the EncFtnsH dimer is shown; 3 peaks are observed with m/z 1760 . 5 , 1763 . 8 , and 1767 . 0 Th – the lowest peak corresponds to neutral masses of 26392 . 5 Da [predicted EncFtnsH dimer , ( C572H884N172O185S2 ) 2; 26388 . 6 Da] . The two further peaks have a delta-mass of ~+50 Da , consistent with Fe binding . We interpret these observations as partial ‘atypical’ CID fragmentation of the decameric complex – i . e . fragmentation of the initial complex with retention of subunit and ligand interactions . A schematic summary of these results is displayed in ( D ) . We postulate the high stability of this iron-bound dimer sub-complex is due to the metal coordination at the dimer interface , increasing the strength of the dimer interface . Taken together , these observations support our findings that the topology of the decameric EncFtnsH assembly is arranged as a pentamer of dimers , with two Fe ions at each dimer interface . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 020 We subsequently performed gas phase disassembly of the decameric EncFtnsH using collision-induced dissociation ( CID ) tandem mass spectrometry . Under the correct CID conditions , protein assemblies can dissociate with retention of subunit and ligand interactions , and thus provide structurally-informative evidence as to the topology of the original assembly; this has been termed ‘atypical’ dissociation ( Hall et al . , 2013 ) . For EncFtnsH , this atypical dissociation pathway was clearly evident; CID of the EncFtnsH decamer resulted in the appearance of a dimeric EncFtnsH subcomplex containing 0 , 1 , or 2 iron ions ( Figure 7—figure supplement 2 ) . In light of the crystal structure , this observation can be rationalized as dissociation of the EncFtnsH decamer by disruption of the non-FOC interface with at least partial retention of the FOC interface and the FOC-Fe . Thus , this observation supports our crystallographic assignment of the overall topology of the EncFtnsH assembly as a pentameric assembly of dimers with two iron ions located at the FOC dimer interface . In addition , this analysis provides evidence that the overall architecture of the complex is consistent in the crystal , solution and gas phases . In light of the identification of an iron-loaded FOC in the crystal structure of EncFtn and our native mass spectrometry data , we performed ferroxidase and peroxidase assays to demonstrate the catalytic activity of this protein . In addition , we also assayed equine apoferritin , an example of a classical ferritin enzyme , as a positive control . Unlike the Dps family of ferritin-like proteins , EncFtn showed no peroxidase activity when assayed with the substrate ortho-phenylenediamine ( Pesek et al . , 2011 ) . The ferroxidase activity of EncFtnsH was measured by recording the progress curve of Fe2+ oxidation to Fe3+ at 315 nm after addition of 20 and 100 µM Fe2+ ( 2 and 10 times molar ratio Fe2+/FOC ) . In both experiments the rate of oxidation was faster than background oxidation of Fe2+ by molecular oxygen , and was highest for 100 µM Fe2+ ( Figure 8A ) . These data show that recombinant EncFtnsH acts as an active ferroxidase enzyme . When compared to apoferritin , EncFtnsH oxidized Fe2+ at a slower rate and the reaction did not run to completion over the 1800 s of the experiment . Addition of higher quantities of iron resulted in the formation of a yellow/red precipitate at the end of the reaction . We also performed these assays on purified recombinant encapsulin; which , when assayed alone , did not display ferroxidase activity above background Fe2+ oxidation ( Figure 8B ) . In contrast , complexes of the full EncFtn encapsulin nanocompartment ( i . e . the EncFtn-Enc protein complex ) displayed ferroxidase activity comparable to apoferritin without the formation of precipitates ( Figure 8B ) . 10 . 7554/eLife . 18972 . 021Figure 8 . Spectroscopic evidence for the ferroxidase activity and comparison of iron loading capacity of apoferritin , EncFtnsH , encapsulin , and EncFtn-Enc . ( A ) Apoferritin ( 10 μM monomer concentration ) and EncFtnsH decamer fractions ( 20 μM monomer concentration , 10 μM FOC concentration ) were incubated with 20 and 100 μM iron ( 2 and 10 times molar equivalent Fe2+ per FOC ) and progress curves of the oxidation of Fe2+ to Fe3+ at 315 nm were recorded in a spectrophotometer . The background oxidation of iron at 20 and 100 μM in enzyme-free controls are shown for reference . ( B ) Encapsulin and EncFtn-Enc complexes at 10 μM asymmetric unit concentration were incubated with Fe2+ at 20 and 100 μM and progress curves for iron oxidation at A315 were measured in a UV/visible spectrophotometer . Enzyme free controls for background oxidation of Fe2+ are shown for reference . ( C ) Histogram of the iron loading capacity per biological assembly of EncFtnsH , encapsulin , EncFtn-Enc and apoferritin . The results shown are for three technical replicates and represent the optimal iron loading by the complexes after three hours when incubated with Fe2+ . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 02110 . 7554/eLife . 18972 . 022Figure 8—figure supplement 1 . TEM visualization of iron-loaded bacterial nanocompartments and ferritin . Decameric EncFtnsH , encapsulin , EncFtn-Enc and apoferritin , at 8 . 5 µM , were mixed with 147 µM , 1 mM , 1 mM and 215 µM acidic Fe ( NH4 ) 2 ( SO4 ) 2 , respectively . Protein mixtures were incubated at room temperature for 1 hr prior to TEM analysis with or without uranyl acetate stain . ( A–D ) Unstained EncFtnsH , encapsulin , EncFtn-Enc , apoferritin loaded with Fe2+ , respectively , with 35 , 000 x magnification and scale bars indicate 100 nm . ( E ) Protein-free sample as a control . ( F–I ) Stained EncFtnsH , encapsulin , EncFtn-Enc , apoferritin loaded with Fe2+ , respectively , with 140 , 000 x magnification and scale bars indicate 25 nm . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 022 We attributed the precipitates observed in the EncFtnsH ferroxidase assay to the production of insoluble Fe3+ complexes , which led us to propose that EncFtn does not directly store Fe3+ in a mineral form . This observation agrees with native MS results , which indicates a maximum iron loading of 10–15 iron ions per decameric EncFtn; and the structure , which does not possess the enclosed iron-storage cavity characteristic of classical ferritins and Dps family proteins that can directly accrue mineralized Fe3+ within their nanocompartment structures . To analyze the products of these reactions and determine whether the EncFtn and encapsulin were able to store iron in a mineral form , we performed TEM on the reaction mixtures from the ferroxidase assay . The EncFtnsH reaction mixture showed the formation of large , irregular electron-dense precipitates ( Figure 8—figure supplement 1A ) . A similar distribution of particles was observed after addition of Fe2+ to the encapsulin protein ( Figure 8—figure supplement 1B ) . In contrast , addition of Fe2+ to the EncFtn-Enc nanocompartment resulted in small , highly regular , electron dense particles of approximately 5 nm in diameter ( Figure 8—figure supplement 1C ) ; we interpret these observations as controlled mineralization of iron within the nanocompartment . Addition of Fe2+ to apoferritin resulted in a mixture of large particles and small ( ~2 nm ) particles consistent with partial mineralization by the ferritin and some background oxidation of the iron ( Figure 8—figure supplement 1D ) . Negative stain TEM of these samples revealed that upon addition of iron , the EncFtnsH protein showed significant aggregation ( Figure 8—figure supplement 1F ) ; while the encapsulin , EncFtn-Enc system , and apoferritin are present as distinct nanocompartments without significant protein aggregation ( Figure 8—figure supplement 1G–I ) . The results of the ferroxidase assay and micrographs of the reaction products suggest that the oxidation and mineralization function of the classical ferritins are split between the EncFtn and encapsulin proteins , with the EncFtn acting as a ferroxidase and the encapsulin shell providing an environment and template for iron mineralization and storage . To investigate this further , we added Fe2+ at various concentrations to samples of apo-ferritin , EncFtn , isolated encapsulin , and the EncFtn-Enc protein complex , and subjected these samples to a ferrozine assay to quantify the amount of iron associated with the proteins after three hours of incubation . The maximum iron loading capacity of these systems was calculated as the quantity of iron per biological assembly ( Figure 8C ) . In this assay , the EncFtnsH decamer binds a maximum of around 48 iron ions before excess iron induces protein precipitation . The encapsulin shell protein can sequester about 2200 iron ions before significant protein loss occurs , and the reconstituted EncFtn-Enc nanocompartment sequestered about 4150 iron ions . This latter result is significantly more than the apoferritin used in our assay , which sequesters approximately 570 iron ions in this assay ( Figure 8C , Table 5 ) . Consideration of the functional oligomeric states of these proteins , where EncFtn is a decamer and encapsulin forms an icosahedral cage , and estimation of the iron loading capacity of these complexes gives insight into the role of the two proteins in iron storage and mineralization . EncFtn decamers bind up to 48 iron ions ( Figure 8C ) , which is significantly higher than the stoichiometry of fifteen metal ions visible in the FOC and E31/34-site of the crystal structure of the EncFtnsH decamer and our MS analysis . The discrepancy between these solution measurements and our MS analysis may indicate that there are additional metal-binding sites on the interior channel and exterior faces of the protein; this is consistent with our identification of a number of weak metal-binding sites at the surface of the protein in the crystal structure ( Figure 5D ) . These observations are consistent with hydrated Fe2+ ions being channeled to the active site from the E31/34-site and the subsequent exit of Fe3+ products on the outer surface , as is seen in other ferritin family proteins ( Pesek et al . , 2011; Behera and Theil , 2014 ) . While the isolated encapsulin shell does not display any ferroxidase activity , it binds around 2200 iron ions in our assay ( Table 5 ) . This implies that the shell can bind a significant amount of iron on its outer and inner surfaces . While the maximum reported loading capacity of classical ferritins is approximately 4500 iron ions ( Mann et al . , 1986 ) , in our assay system we were only able to load apoferritin with around 570 iron ions . However , the recombinant EncFtn-Enc nanocompartment was able to bind over 4100 iron ions in the same time period , over seven times the amount seen for the apoferritin . We note we do not reach the experimental maximum iron loading for apoferritin and therefore the total iron-loading capacity of our system may be significantly higher than in this experimental system . Taken together , our data show that EncFtn can catalytically oxidize Fe2+ to Fe3+; however , iron binding in EncFtn is limited to the FOC and several surface metal binding sites . In contrast , the encapsulin protein displays no catalytic activity , but has the ability to bind a considerable amount of iron . Finally , the EncFtn-Enc nanocompartment complex retains the catalytic activity of EncFtn , and sequesters iron within the encapsulin shell at a higher level than the isolated components of the system , and at a significantly higher level than the classical ferritins ( Andrews , 2010 ) . Furthermore , our recombinant nanocompartments may not have the physiological subunit stoichiometry , and the iron-loading capacity of native nanocompartments is potentially much higher than the level we have observed . To investigate the structural and biochemical role played by the metal binding residues in the di-iron FOC of EncFtnsH we produced alanine mutations in each of these residues: Glu32 , Glu62 , and His65 . These EncFtnsH mutants were produced in E . coli cells grown in MM , both in the absence and presence of additional iron . The E32A and E62A mutants eluted from SEC at a volume consistent with the decameric form of EncFtnsH , with a small proportion of monomer; the H65A mutant eluted at a volume consistent with the monomeric form of EncFtnsH ( Figure 9 ) . For all of the mutants studied , no change in oligomerization state was apparent upon addition of Fe2+ in vitro . 10 . 7554/eLife . 18972 . 023Figure 9 . Purification of recombinant R . rubrum EncFtnsH FOC mutants . Single mutants E32A , E62A , and H65A of EncFtnsH produced from E . coli BL21 ( DE3 ) cells grown in MM and MM supplemented with iron were subjected to Superdex 200 size-exclusion chromatography . ( A ) Gel-filtration chromatogram of the E32A mutant form of EncFtnsH resulted in an elution profile with a majority of the protein eluting as the decameric form of the protein and a small proportion of monomer . ( B ) Gel-filtration chromatograhy of the E62A mutant form of EncFtnsH resulted in an elution profile with a single major decameric peak . ( C ) Gel-filtration chromatography of the H65A mutant form of EncFtnsH resulted in a single peak corresponding to the protein monomer . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 023 In addition to SEC studies , native mass spectrometry of the apo-EncFtnsH mutants was performed and compared with the wild-type apo-EncFtnsH protein ( Figure 10 ) . As described above , the apo-EncFtnsH has a charge state distribution consistent with an unstructured monomer , and decamer formation is only initiated upon addition of ferrous iron . Both the E32A mutant and E62A mutant displayed charge state distributions consistent with decamers , even in the absence of Fe2+ . This gas-phase observation is consistent with SEC measurements , which indicate both of these variants were also decamers in solution . Thus it seems that these mutations allow the decamer to form in the absence of iron in the FOC . In contrast to the glutamic acid mutants , MS analysis of the H65A mutant is similar to wild-type apo-EncFtnsH and is present as a monomer; interestingly a minor population of dimeric H65A was also observed . 10 . 7554/eLife . 18972 . 024Figure 10 . Native mass spectrometry of EncFtnsH mutants . All spectra were acquired in 100 mM ammonium acetate , pH 8 . 0 with a protein concentration of 5 µM . ( A ) Wild-type EncFtnsH in the absence of iron displays a charge state distribution consistent with a monomer ( see also Figure 8 ) . ( B ) E32A EncFtnsH displays a charge states consistent with a decamer ( green circles ) ; a minor species , consistent with the monomer of E32A mutant is also observed ( blue circles ) . ( C ) E62A EncFtnsH displays charge states consistent with a decamer ( green circles ) . ( D ) H65A EncFtnsH displays charge states consistent with both monomer ( blue circles ) and dimer ( purple circles ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 024 We propose that the observed differences in the oligomerization state of the E32A and E62A mutants compared to wild-type are due to the changes in the electrostatic environment within the FOC . At neutral pH the glutamic acid residues are negatively charged , while the histidine residues are predominantly in their uncharged state . In the wild-type ( WT ) EncFtnsH this leads to electrostatic repulsion between subunits in the absence of iron . Coordination of Fe2+ in this site stabilizes the dimer and reconstitutes the active FOC . The geometric arrangement of Glu32 and Glu62 in the FOC explains their behavior in solution and the gas phase , where they both favor the formation of decamers due to the loss of a repulsive negative charge . The FOC in the H65A mutant is destabilized through the loss of this metal coordinating residue and potential positive charge carrier , thus favoring the monomer in solution and the gas phase . To understand the impact of the mutants on the organization and metal binding of the FOC , we determined the X-ray crystal structures of each of the EncFtnsH mutants ( See Table 4 for data collection and refinement statistics ) . The crystal packing of all of the mutants in this study is essentially isomorphous to the EncFtnsH structure . All of the mutants display the same decameric arrangement in the crystals as the EncFtnsH structure , and the monomers superimpose with an average RMSDCα of less than 0 . 2 Å . 10 . 7554/eLife . 18972 . 025Table 4 . Data collection and refinement statistics . Statistics for the highest-resolution shell are shown in parentheses . Friedel mates were averaged when calculating reflection numbers and statistics . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 025WTE32AE62AH65AData collectionWavelength ( Å ) 1 . 741 . 731 . 731 . 74Resolution range ( Å ) 49 . 63 - 2 . 06 ( 2 . 10 - 2 . 06 ) 48 . 84 - 2 . 59 ( 2 . 683 - 2 . 59 ) 48 . 87 - 2 . 21 ( 2 . 29 - 2 . 21 ) 48 . 86 - 2 . 97 ( 3 . 08 - 2 . 97 ) Space groupP 1 21 1P 1 21 1P 1 21 1P 1 21 1Unit cell ( Å ) a b c β ( ° ) 98 . 18 120 . 53 140 . 30 95 . 3697 . 78 120 . 28 140 . 53 95 . 4198 . 09 120 . 23 140 . 36 95 . 5098 . 03 120 . 29 140 . 43 95 . 39Total reflections1 , 264 , 922 ( 41 , 360 ) 405 , 488 ( 36 , 186 ) 1 , 069 , 345 ( 95 , 716 ) 323 , 853 ( 32 , 120 ) Unique reflections197 , 873 ( 8 , 766 ) 100 , 067 ( 9 , 735 ) 162 , 379 ( 15 , 817 ) 66 , 658 ( 6 , 553 ) Multiplicity6 . 4 ( 4 . 7 ) 4 . 1 ( 3 . 7 ) 6 . 6 ( 6 . 1 ) 4 . 9 ( 4 . 9 ) Anomalous multiplicity3 . 2 ( 2 . 6 ) N/AN/AN/ACompleteness ( % ) 99 . 2 ( 88 . 6 ) 99 . 0 ( 97 . 0 ) 100 ( 97 . 0 ) 100 ( 99 . 0 ) Anomalous completeness ( % ) 96 . 7 ( 77 . 2 ) N/AN/AN/AMean I/sigma ( I ) 10 . 6 ( 1 . 60 ) 8 . 46 ( 1 . 79 ) 13 . 74 ( 1 . 80 ) 8 . 09 ( 1 . 74 ) Wilson B-factor26 . 9840 . 1033 . 9752 . 20Rmerge0 . 123 ( 0 . 790 ) 0 . 171 ( 0 . 792 ) 0 . 0979 ( 1 . 009 ) 0 . 177 ( 0 . 863 ) Rmeas0 . 147 ( 0 . 973 ) 0 . 196 ( 0 . 923 ) 0 . 1064 ( 1 . 107 ) 0 . 199 ( 0 . 966 ) CC1/20 . 995 ( 0 . 469 ) 0 . 985 ( 0 . 557 ) 0 . 998 ( 0 . 642 ) 0 . 989 ( 0 . 627 ) CC*0 . 999 ( 0 . 846 ) 0 . 996 ( 0 . 846 ) 0 . 999 ( 0 . 884 ) 0 . 997 ( 0 . 878 ) Image DOI10 . 7488/ds/134210 . 7488/ds/141910 . 7488/ds/142010 . 7488/ds/1421RefinementRwork0 . 171 ( 0 . 318 ) 0 . 183 ( 0 . 288 ) 0 . 165 ( 0 . 299 ) 0 . 186 ( 0 . 273 ) Rfree0 . 206 ( 0 . 345 ) 0 . 225 ( 0351 ) 0 . 216 ( 0 . 364 ) 0 . 237 ( 0 . 325 ) Number of non-hydrogen atoms23 , 22222 , 36622 , 69122 , 145macromolecules22 , 27622 , 01921 , 96522 , 066ligands13882474water8083397025Protein residues2 , 7032 , 6862 , 6752 , 700RMS ( bonds ) ( Å ) 0 . 0120 . 0050 . 0110 . 002RMS ( angles ) ( ° ) 1 . 260 . 581 . 020 . 40Ramachandran favored ( % ) 1009910099Ramachandran allowed ( % ) 0101Ramachandran outliers ( % ) 0000Clash score1 . 421 . 421 . 790 . 97Average B-factor ( Å2 ) 33 . 9042 . 3141 . 3447 . 68macromolecules33 . 8042 . 3541 . 3147 . 60ligands40 . 4072 . 8065 . 5572 . 34solvent36 . 2038 . 9541 . 4633 . 85PDB ID5DA55L895L8B5L8G10 . 7554/eLife . 18972 . 026Table 5 . Iron loading capacity of EncFtn , encapsulin and ferritin . Protein samples ( at 8 . 5 µM ) including decameric EncFtnsH , encapsulin , EncFtn-Enc and apoferritin were mixed with Fe ( NH4 ) 2 ( SO4 ) ( in 0 . 1% ( v/v ) HCl ) of different concentrations in 50 mM Tris-HCl ( pH 8 . 0 ) , 150 mM NaCl buffer at room temperature for 3 hrs in the air . Protein-Fe mixtures were centrifuged at 13 , 000 x g to remove precipitated material and desalted prior to the Fe and protein content analysis by ferrozine assay and BCA microplate assay , respectively . Fe to protein ratio was calculated to indicate the Fe binding capacity of the protein . Protein stability was compromised at high iron concentrations; therefore , the highest iron loading with the least protein precipitation was used to derive the maximum iron loading capacity per biological assembly ( underlined and highlighted in bold ) . The biological unit assemblies are a decamer for EncFtnsH , a 60mer for encapsulin , a 60mer of encapsulin loaded with 12 copies of decameric EncFtn in the complex , and 24mer for horse spleen apoferritin . Errors are quoted as the standard deviation of three technical repeats in both the ferrozine and BCA microplate assays . The proteins used in Fe loading experiment came from a single preparation . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 026Protein sampleFe ( NH4 ) 2 ( SO4 ) 2 loading ( µM ) Fe detected by ferrozine assay ( µM ) Protein detected by BCA microplate assay ( µM ) Fe / monomeric proteinMaximum Fe loading per biological assembly unit8 . 46 µM EncFtnsH-10mer04 . 73 ± 2 . 325 . 26 ± 0 . 640 . 90 ± 0 . 4439 . 99 . 93 ± 1 . 205 . 36 ± 0 . 691 . 85 ± 0 . 228417 . 99 ± 2 . 014 . 96 ± 0 . 043 . 63 ± 0 . 4114721 . 09 ± 1 . 944 . 44 ± 0 . 214 . 75 ± 0 . 4448 ± 422428 . 68 ± 0 . 303 . 73 ± 0 . 537 . 68 ± 0 . 0830111 . 27 ± 1 . 102 . 50 ± 0 . 054 . 51 ± 0 . 448 . 50 µM Encapsulin0-1 . 02 ± 0 . 548 . 63 ± 0 . 17-0 . 12 ± 0 . 0622462 . 24 ± 2 . 4910 . 01 ± 0 . 586 . 22 ± 0 . 3530167 . 94 ± 3 . 158 . 69 ± 0 . 427 . 81 ± 0 . 36450107 . 96 ± 8 . 888 . 50 ± 0 . 6912 . 71 ± 1 . 0570097 . 51 ± 3 . 197 . 26 ± 0 . 2013 . 44 ± 0 . 441000308 . 63 ± 2 . 068 . 42 ± 0 . 3436 . 66 ± 0 . 242199 ± 15150057 . 09 ± 0 . 901 . 44 ± 0 . 2139 . 77 ± 0 . 6220009 . 2 ± 1 . 160 . 21 ± 0 . 1444 . 73 ± 5 . 638 . 70 µM EncFtn-Enc03 . 31 ± 1 . 576 . 85 ± 0 . 070 . 48 ± 0 . 23224116 . 27 ± 3 . 747 . 63 ± 0 . 1215 . 25 ± 0 . 49301132 . 86 ± 4 . 036 . 66 ± 0 . 3119 . 96 ± 0 . 61450220 . 57 ± 27 . 336 . 12 ± 1 . 0736 . 06 ± 4 . 47700344 . 03 ± 40 . 386 . 94 ± 0 . 1749 . 58 ± 5 . 821000496 . 00 ± 38 . 487 . 19 ± 0 . 0868 . 94 ± 5 . 354137 ± 3211500569 . 98 ± 73 . 635 . 73 ± 0 . 0399 . 44 ± 12 . 842000584 . 30 ± 28 . 334 . 88 ± 0 . 22119 . 62 ± 5 . 808 . 50 µM Apoferritin03 . 95 ± 2 . 269 . 37 ± 0 . 240 . 42 ± 0 . 2542 . 510 . 27 ± 1 . 128 . 27 ± 0 . 301 . 24 ± 0 . 18212 . 544 . 48 ± 2 . 767 . 85 ± 0 . 775 . 67 ± 0 . 83637 . 5160 . 93 ± 4 . 276 . 76 ± 0 . 8123 . 79 ± 3 . 12571 ± 751275114 . 92 ± 3 . 173 . 84 ± 0 . 3029 . 91 ± 2 . 95170091 . 40 ± 3 . 373 . 14 ± 0 . 3529 . 13 ± 3 . 86 Close inspection of the region of the protein around the FOC in each of the mutants highlights their effect on metal binding ( Figure 11 and Figure 11—figure supplement 1–3 ) . In the E32A mutant the position of the side chains of the remaining iron coordinating residues in the FOC is essentially unchanged , but the absence of the axial-metal coordinating ligand provided by the Glu32 side chain abrogates metal binding in this site . The Glu31/34-site also lacks metal , with the side chain of Glu31 rotated by 180° at the Cβ in the absence of metal ( Figure 11—figure supplement 1 ) . The E62A mutant has a similar effect on the FOC to the E32A mutant , however the entry site still has a calcium ion coordinated between residues Glu31 and Glu34 ( Figure 11—figure supplement 2 ) . The H65A mutant diverges significantly from the wild type in the position of the residues Glu32 and Tyr39 in the FOC . E32 appears in either the original orientation as the wild type and coordinates Ca2+ in this position , or it is flipped by 180° at the Cβ , moving away from the coordinated calcium ion in the FOC . Tyr39 moves closer to Ca2+ compared to the wild-type and coordinates the calcium ion ( Figure 11—figure supplement 3 ) . A single calcium ion is present in the entry site of this mutant; however , Glu31 of one chain is rotated away from the metal ion and is not involved in coordination . 10 . 7554/eLife . 18972 . 027Figure 11 . Comparison of the EncFtnsH FOC mutants vs wild type . The structures of the three EncFtnsH mutants were all determined by X-ray crystallography . The E32A , E62A and H65A mutants were crystallized in identical conditions to the wild type . EncFtnsH structure and were essentially isomorphous in terms of their unit cell dimensions . The FOC residues of the mutants and native EncFtnsH structures are shown as sticks with coordinated Fe2+ as orange and Ca2+ as grey spheres and are colored as follows: wild type , grey; E32A , pink; E62A , green; H65A , blue . Of the mutants , only H65A has any coordinated metal ions , which appear to be calcium ions from the crystallization condition . The overall organization of FOC residues is retained in the mutants , with almost no backbone movements . Significant differences center around Tyr39 , which moves to coordinate the bound calcium ions in the H65A mutant; and Glu32 , which moves away from the metal ions in this structure . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 02710 . 7554/eLife . 18972 . 028Figure 11—figure supplement 1 . FOC dimer interface of EncFtnsH-E32A mutant . ( A ) Wall-eyed stereo view of the metal-binding dimerization interface of EncFtnsH-E32A . Protein residues are shown as sticks with blue and green carbons for the different subunits . The 2mFo-DFc electron density map is shown as a blue mesh contoured at 1 . 5 σ . ( B ) Views of the FOC of the EncFtnsH-E32Amutant . Protein atoms shown as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 02810 . 7554/eLife . 18972 . 029Figure 11—figure supplement 2 . FOC dimer interface of EncFtnsH-E62A mutant . ( A ) Wall-eyed stereo view of the metal-binding dimerization interface of EncFtnsH-E62A . Protein residues are shown as sticks with blue and green carbons for the different subunits . The 2mFo-DFc electron density map is shown as a blue mesh contoured at 1 . 5 σ . The single coordinated calcium ion is shown as a grey sphere . ( B ) Views of the FOC of the EncFtnsH-E62A mutant . Protein atoms shown as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 02910 . 7554/eLife . 18972 . 030Figure 11—figure supplement 3 . FOC dimer interface of EncFtnsH-H65A mutant . ( A ) Wall-eyed stereo view of the metal-binding dimerization interface of EncFtnsH-H65A . Protein residues are shown as sticks with blue and green carbons for the different subunits . The 2mFo-DFc electron density map is shown as a blue mesh contoured at 1 . 5 σ . The coordinated calcium ions are shown as a grey spheres with coordination distances in the FOC highlighted with yellow dashed lines . ( B ) Views of the FOC of the EncFtnsH-H65A mutant . Protein atoms and metal ions shown as in ( A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 030 Taken together the results of our data show that these changes to the FOC of EncFtn still permit the formation of the decameric form of the protein . While the proteins all appear decameric in crystals , their solution and gas-phase behavior differs considerably and the mutants no longer show metal-dependent oligomerization . These results highlight the importance of metal coordination in the FOC for the stability and assembly of the EncFtn protein . To address the question of how mutagenesis of the iron coordinating residues affects the enzymatic activity of the EncFtnsH protein we recorded progress curves for the oxidation of Fe2+ to Fe3+ by the different mutants as before . Mutagenesis of E32A and H65A reduces the activity of EncFtnsH by about 40%-55%; the E62A mutant completely abrogates activity , presumably through the loss of the bridging coordination for the formation of the di-nuclear iron center of the FOC ( Figure 12 ) . Collectively , the effect of mutating these residues in the FOC confirms the importance of the iron coordinating residues for the ferroxidase activity of the EncFtnsH protein . 10 . 7554/eLife . 18972 . 031Figure 12 . Relative ferroxidase activity of EncFtnsH mutants . EncFtnsH , and the mutant forms E32A , E62A and H65A , each at 20 µM , were mixed with 100 µM acidic Fe ( NH4 ) 2 ( SO4 ) 2 . Ferroxidase activity of the mutant forms is determined by measuring the absorbance at 315 nm for 1800 s at 25 °C as an indication of Fe3+ formation . The relative ferroxidase activity of mutants is plotted as a proportion of the activity of the wild-type protein using the endpoint measurement of A315 . Three technical repeats were performed and the plotted error bars represent the calculated standard deviations . The FOC mutants showed reduced ferroxidase activity to varied extents , among which E62A significantly abrogated the ferroxidase activity . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 03110 . 7554/eLife . 18972 . 032Figure 12—figure supplement 1 . Progress curves recording ferroxidase activity of EncFtnsH mutants . 20 µM wild-type EncFtnsH , E32A , E62A and H65A mutants were mixed with 20 µM or 100 µM acidic Fe ( NH4 ) 2 ( SO4 ) 2 , respectively . Absorbance at 315 nm was recorded for 1800 s at 25°C as an indication of Fe3+ formation . Protein free samples ( dashed and dotted lines ) were measured for Fe2+ background oxidation as controls . Assays were performed with three technical repeats . Error bars were showed in shadows behind each curves . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 032 The quaternary arrangement of classical ferritins into an octahedral nanocage and Dps into a dodecamer is absolutely required for their function as iron storage compartments ( Chasteen and Harrison , 1999 ) . The oxidation and mineralization of iron must be spatially separated from the host cytosol to prevent the formation of damaging hydroxyl radicals in the Fenton and Haber-Weiss reactions ( Honarmand Ebrahimi et al . , 2012 ) . This is achieved in all ferritins by confining the oxidation of iron to the interior of the protein complex , thus achieving sequestration of the Fe3+ mineralization product . A structural alignment of the FOC of EncFtn with the classical ferritin PmFtn shows that the central ring of EncFtn corresponds to the external surface of ferritin , while the outer circumference of EncFtn is congruent with the inner mineralization surface of ferritin ( Figure 6—figure supplement 1A ) . This overlay highlights the fact that the ferroxidase center of EncFtn faces in the opposite direction relative to the classical ferritins and is essentially inside out regarding iron storage space ( Figure 6—figure supplement 1B , boxed region ) . Analysis of each of the single mutations ( E32A , E62A and H65A ) made in the FOC highlights the importance of the iron-coordinating residues in the catalytic activity of EncFtn . Furthermore , the position of the calcium ion coordinated by Glu31 and Glu34 seen in the EncFtnsH structure suggests an entry site to channel metal ions into the FOC; we propose that this site binds hydrated iron ions in vivo and acts as a selectivity filter and gate for the FOC ( Haldar et al . , 2011 ) . The constellation of charged residues on the outer circumference of EncFtn ( His57 , Glu61 and Glu64 ) could function in the same way as the residues lining the mineralization surface within the classical ferritin nanocage ( Le Brun et al . , 2010 ) , and given their proximity to the FOC these sites may be the exit portal and mineralization site ( Honarmand Ebrahimi et al . , 2012 ) . The absolute requirement for the spatial separation of oxidation and mineralization in ferritins suggests that the EncFtn family proteins are not capable of storing iron minerals due to the absence of an enclosed compartment in their structure ( Figure 6—figure supplement 1B ) . Our biochemical characterization of EncFtn supports this hypothesis , indicating that while this protein is capable of oxidizing iron , it does not accrue mineralized iron in an analogous manner to classical ferritins . While EncFtn does not store iron itself , its association with the encapsulin nanocage suggests that mineralization occurs within the cavity of the encapsulin shell ( McHugh et al . , 2014 ) . Our ferroxidase assay data on the recombinant EncFtn-Enc nanocompartments , which accrue over 4100 iron ions per complex and form regular nanoparticles , are consistent with the encapsulin protein acting as the store for iron oxidized by the EncFtn enzyme . TEM analysis of the reaction products shows the production of homogeneous iron nanoparticles only in the EncFtn-Enc nanocompartment ( Figure 8—figure supplement 1 ) . Docking the decamer structure of EncFtnsH into the pentamer of the T . maritima encapsulin Tmari_0786 ( PDB ID: 3DKT ) ( Sutter et al . , 2008 ) shows that the position of the C-terminal extensions of our EncFtnsH structure are consistent with the localization sequences seen bound to the encapsulin protein ( Figure 14A ) . Thus , it appears that the EncFtn decamer is the physiological state of this protein . This arrangement positions the central ring of EncFtn directly above the pore at the five-fold symmetry axis of the encapsulin shell and highlights a potential route for the entry of iron into the encapsulin and towards the active site of EncFtn . A comparison of the encapsulin nanocompartment and the ferritin nanocage highlights the size differential between the two complexes ( Figure 14B ) that allows the encapsulin to store significantly more iron . The presence of five FOCs per EncFtnsH decamer and the fact that the icosahedral encapsulin nanocage can hold up to twelve of decameric EncFtn between each of the internal five-fold vertices means that they can achieve a high rate of iron mineralization across the entire nanocompartment . This arrangement of multiple reaction centers in a single protein assembly is reminiscent of classical ferritins , which has 24 FOCs distributed around the nanocage . 10 . 7554/eLife . 18972 . 034Figure 14 . Model of iron oxidation in encapsulin nanocompartments . ( A ) Model of EncFtnsH docking to the encapsulin shell . A single pentamer of the icosahedral T . maritima encapsulin structure ( PDBID: 3DKT ) ( Sutter et al . , 2008 ) is shown as a blue surface with the encapsulin localization sequence of EncFtn shown as a purple surface . The C-terminal regions of the EncFtn subunits correspond to the position of the localization sequences seen in 3DKT . Alignment of EncFtnsH with 3DKT positions the central channel directly above the pore in the 3DKT pentamer axis ( shown as a grey pentagon ) . ( B ) Surface view of EncFtn within the encapsulin nanocompartment ( grey and blue respectively ) . The lumen of the encapsulin nanocompartment is considerably larger than the interior of ferritin ( shown in orange behind the encapsulin for reference ) and thus allows the storage of significantly more iron . The proposed pathway for iron movement through the encapsulin shell and EncFtn FOC is shown with arrows . ( C ) Model ofiron oxidation within an encapsulin nanocompartment . As EncFtn is unable to mineralize iron on its surface directly , Fe2+ must pass through the encapsulin shell to access the first metal binding site within the central channel of EncFtnsH ( entry site ) prior to oxidation within the FOC and release as Fe3+ to the outer surface of the protein where it can be mineralized within the lumen of the encapsulin cage . DOI: http://dx . doi . org/10 . 7554/eLife . 18972 . 034 Our structural data , coupled with biochemical and ICP-MS analysis , suggest a model for the activity of the encapsulin iron-megastore ( Figure 14C ) . The crystal structure of the T . maritima encapsulin shell protein has a negatively charged pore positioned to allow the passage of Fe2+ into the encapsulin and directs the metal towards the central , negatively charged hole of the EncFtn ring ( Figure 4—figure supplement 1 ) . The five metal-binding sites on the interior of the ring ( Glu31/34-sites ) may select for the Fe2+ ion and direct it towards their cognate FOCs . We propose that the oxidation of Fe2+ to Fe3+ occurs within the FOC according to the model postulated by ( Honarmand Ebrahimi et al . , 2012 ) in which the FOC acts as a substrate site through which iron passes and is released on to weakly coordinating sites at the outer circumference of the protein ( His57 , Glu61 and Glu64 ) , where it is able to form ferrihydrite minerals which can be safely deposited within the lumen of the encapsulin nanocompartment ( Figure 14 ) . Here we describe for the first time the structure and biochemistry of a new class of encapsulin-associated ferritin-like protein and demonstrate that it has an absolute requirement for compartmentalization within an encapsulin nanocage to act as an iron store . Further work on the EncFtn-Enc nanocompartment will establish the structural basis for the movement of iron through the encapsulin shell , the mechanism of iron oxidation by the EncFtn FOC and its subsequent storage in the lumen of the encapsulin nanocompartment . Genes of interest were amplified by PCR using R . rubrum ATCC 11 , 170 genomic DNA ( DSMZ ) as the template and KOD Hot Start DNA Polymerase ( Novagen ) . Primers used in this study are listed in Supplementary file 2 . PCR products were visualized in 0 . 8% agarose gel stained with SYBR Safe ( Life Technologies , UK ) . Fragments of interest were purified by gel extraction ( Qiagen , UK ) before digestion by endonuclease restriction enzymes ( Thermo Fisher Scientific , UK ) at 37°C for 1 hr , followed by ligation with similarly digested vector pET-28a ( + ) or pACYCDuet-1 at room temperature for 1 hr . Ligation product was transformed into chemically competent Escherichia coli Top10 cells and screened against 50 ng/μl kanamycin for pET-28a ( + ) based constructs or 34 ng/μl chloramphenicol for pACYCDuet-1 based constructs . DNA insertion was confirmed through Sanger sequencing ( Edinburgh Genomics , The University of Edinburgh , UK ) . Sequence verified constructs were transformed into E . coli BL21 ( DE3 ) or Tuner ( DE3 ) for protein production . Alternatively , plasmids transformed into E . coli B834 ( DE3 ) cells were cultured in selenomethionine medium . A single colony of E . coli BL21 ( DE3 ) or Tuner ( DE3 ) cells , transformed with protein expression plasmid , was transferred into 10 ml LB medium , or M9 minimal medium ( MM ) , supplemented with appropriate antibiotic , and incubated overnight at 37 °C with 200 rpm shaking . The overnight pre-culture was then inoculated into 1 liter of LB medium and incubated at 37 °C with 200 rpm shaking . Recombinant protein production was induced at OD600= 0 . 6 by the addition of 1 mM IPTG and the incubation temperature was reduced to 18°C for overnight incubation . Cells were pelleted by centrifugation at 4000 g for 20 min at 4 °C , and resuspended 10-fold ( volume per gram of cell pellet ) in PBS to wash cells before a second centrifugation step . Cells were resuspended in 10-times ( v/w ) of appropriate lysis buffer for the purification method used ( see details of buffers below ) and lysed by sonication on ice , with ten cycles of 30-second burst of sonication at 10 µm amplitude and 30 s of cooling . Cell lysate was clarified by centrifugation at 20 , 000 x g , 30 min , 4 °C; followed by filtration using a 0 . 22 µM syringe filter ( Millipore , UK ) . Selenomethionine labelled protein was produced by growing a single colony of E . coli B834 ( DE3 ) cells transformed with protein expression plasmids in 100 ml LB medium supplemented with appropriate antibiotic overnight at 37 °C with shaking at 200 rpm . The overnight pre-culture was pelleted by centrifugation 3 , 000 x g , 4 °C , 15 min and washed twice with M9 minimal medium . The washed cells were transferred to 1 liter of SeMet medium , which contains M9 minimal medium , 40 mg/L of each L-amino acid ( without methionine ) , 40 mg/L selenomethionine , 2 mM MgSO4 , 0 . 4% ( w/v ) glucose and 1 mM Fe ( NH4 ) 2 ( SO4 ) 2 . Cells were incubated at 37 °C with 200 rpm shaking and recombinant protein production was induced at OD600= 0 . 6 by the addition of 1 mM IPTG and the incubation temperature was reduced to 18 °C for overnight incubation . Cells were harvested and lysed as above . Clarified cell lysate was loaded onto a 5 ml HisTrap column ( GE Healthcare , UK ) pre-equilibrated with HisA buffer ( 50 mM Tris-HCl , 500 mM NaCl and 50 mM imidazole , pH 8 . 0 ) . Unbound proteins were washed from the column with HisA buffer . His-tagged proteins were then eluted by a step gradient of 50% HisA buffer and 50% HisB buffer ( 50 mM Tris-HCl , 500 mM NaCl and 500 mM imidazole , pH 8 . 0 ) . Fractions containing the protein of interest , as determined by 15% ( w/v ) acrylamide SDS-PAGE , were pooled before loading onto a gel-filtration column ( HiLoad 16/600 Superdex 200 , GE Healthcare ) equilibrated with GF buffer ( 50 mM Tris-HCl , pH 8 . 0 , 150 mM NaCl ) . Fractions were subjected to 15% SDS-PAGE and those containing the protein of interest were pooled for further analysis . Co-expressed encapsulin and EncFtn ( EncFtn-Enc ) and encapsulin protein were both purified according to the protocol used by M . Sutter ( Sutter , 2008 ) . Briefly , EncFtn-Enc or encapsulin was expressed based on pACYCDuet-1 vector . The E . coli cells were grown , induced , harvested and sonicated in a similar way as described above . GF buffer used in this purification contains 50 mM Tris-HCl , pH 8 . 0 , and 150 mM NaCl . To remove RNA contamination , the lysate was supplemented with 50 μg/ml RNase A and rotated at 10 rpm and room temperature for 2 hrs , followed by centrifugation at 34 , 000 x g and 4 °C for 20 min and filtering through 0 . 22 µM syringe filter . Proteins were pelleted through 38% ( w/v ) sucrose cushion by ultracentrifugation at 100 , 000 x g and 4 °C for 21 hrs . 10% - 50% ( w/v ) sucrose gradient ultracentrifugation was applied to further separate the proteins at 100 , 000 x g and 4 °C for 17 hrs . Protein was dialyzed against GF buffer to remove sucrose before being used in chemical assays or TEM . TEM imaging was performed on purified encapsulin , EncFtn , and EncFtn-Enc and apoferritin . Purified protein at 0 . 1 mg/ml concentration was spotted on glow-discharged 300 mesh carbon-coated copper grids and excess liquid wicked off with filter paper ( Whatman , UK ) . The grids were washed with distilled water and blotted with filter paper three times before staining with 0 . 2% uranyl acetate , blotting and air-drying . Grids were imaged using a JEM1400 transmission electron microscope and images were collected with a Gatan CCD camera . Images were analyzed using ImageJ ( NIH , Bethesda , MD ) and size-distribution histograms were plotted using Prism 6 ( GraphPad software ) . To observe iron mineral formation by TEM , protein samples at 8 . 5 µM concentration including EncFtnsH , encapsulin , EncFtn-Enc and apoferritin were supplemented with acidic Fe ( NH4 ) 2 ( SO4 ) 2 at their maximum iron loading ratio in room temperature for 1 hr . The mixtures were subjected to TEM analysis with or without uranyl acetate staining . TEM experiments without Fe loading were repeated three times , a representative set of images are presented here . Proteins loaded with Fe and imaged by TEM were from single preparation . EncFtnsH was purified by anion exchange and Superdex 200 size- exclusion chromatography and concentrated to 10 mg/ml ( based on extinction coefficient calculation ) . Crystallization drops were set up using the hanging drop vapor diffusion method at 292 K . Glass coverslips were set up with 1–2 μl protein mixed with 1 μl well solution ( 0 . 14 M calcium acetate and 15% ( w/v ) PEG 3350 ) and sealed over 1 ml of well solution . Crystals appeared after 5 days and were harvested from the well using a LithoLoop ( Molecular Dimensions Limited , UK ) , transferred briefly to a cryoprotection solution containing well solution supplemented with 1 mM FeSO4 ( in 0 . 1% ( v/v ) HCl ) , 20% ( v/v ) PEG 200 , and subsequently flash cooled in liquid nitrogen . Crystals of the EncFtnsHsingle mutations were produced in the same manner as for the EncFtnsH wild-type protein . All crystallographic datasets were collected on the macromolecular crystallography beamlines at Diamond Light Source ( Didcot , UK ) at 100 K using Pilatus 6M detectors . Diffraction data were integrated and scaled using XDS ( Kabsch , 2010 ) and symmetry related reflections were merged with Aimless ( Evans , 2011 ) . Data collection statistics are shown in Table 4 . The resolution cut-off used for structure determination and refinement was determined based on the CC1/2 criterion proposed by Karplus and Diederichs ( 2012 ) . The structure of EncFtnsH was determined by molecular replacement using PDB ID: 3K6C as the search model , modified to match the sequence of the target protein using Chainsaw ( Stein , 2008 ) . A single solution comprising three decamers in the asymmetric unit was found by molecular replacement using Phaser ( McCoy et al . , 2007 ) . The initial model was rebuilt using Phenix . autobuild ( Adams et al . , 2010 ) followed by cycles of refinement with Phenix . refine ( Afonine et al . , 2012 ) , with manual rebuilding and model inspection in Coot ( Emsley et al . , 2010 ) . The final model was refined with isotropic B-factors , torsional NCS restraints , and with anomalous group refinement . The model was validated using MolProbity ( Chen et al . , 2010 ) . Structural superimpositions were calculated using Coot . Crystallographic figures were generated with PyMOL . Multiple sequence alignment of EncFtn and ferritin family proteins was performed using Clustal Omega Sievers and Higgins , 2014 and displayed with Espript 3 . 0 ( Gouet et al . , 2003 ) . Model refinement statistics are shown in Table 4 . The final models and experimental data are deposited in the PDB and diffraction image files are available at the Edinburgh DataShare repository . Horse spleen apoferritin purchased from Sigma Aldrich ( UK ) was dissolved in deaerated MOPS buffer ( 100 mM MOPS , 100 mM NaCl , 3 g/100 ml Na2S2O4 and 0 . 5 M EDTA , pH 6 . 5 ) ( Bauminger et al . , 1991 ) . Protein was dialyzed against 1 liter MOPS buffer in room temperature for two days before buffer exchanging to GF buffer ( 50 mM Tris-HCl , pH 8 . 0 , 150 mM NaCl ) in a vivaspin column with 5 kDa cut-off ( Sartorius , UK ) for several times . Fe content of apoferritin was detected using ferrozine assay ( Riemer et al . , 2004 ) . Protein concentration was determined using Pierce Microplate BCA Protein Assay Kit . Apoferritin containing less than 0 . 5 Fe per 24-mer was used in the ferroxidase assay . Apoferritin used in the Fe loading capacity experiment was prepared in the same way with 5–15 Fe per 24-mer . 1 mM and 200 µM Fe ( NH4 ) 2 ( SO4 ) 2 stock solutions were prepared in 0 . 1% ( v/v ) HCl anaerobically . Protein solutions with 20 µM FOC were diluted from ~10 mg/ml frozen stock in GF buffer ( 50 mM Tris-HCl , pH 8 . 0 and 150 mM NaCl ) anaerobically . Ferroxidase activity was initiated by adding 450 μl protein to 50 μl of acidic Fe ( NH4 ) 2 ( SO4 ) 2 at the final concentration of 100 µM and 20 µM in the air , respectively . The ferroxidase activity was measured by monitoring the Fe3+ formation which gives rise to the change of the absorbance at 315 nm ( Bonomi et al . , 1996 ) . Absorbance at 315 nm was recorded every second over 1800 s using a quartz cuvette in a JASCO V-730 UV/VIS spectrophotometer ( JASCO Inc . , Easton , MD ) . In recombinantly coexpressed nanocompartments the ratio of EncFtn to Enc was assumed as 2 to 1 , assuming each of the twelve pentameric vertices of the icosahedral encapsulin were occupied with decameric EncFtn . The data are presented as the mean of three technical replicates with error bars indicating one standard deviation from the mean . Proteins used here were from a single preparation . In order to determine the maximum iron loading capacity , around 8 . 5 µM proteins including decameric EncFtnsH , Encapsulin , EncFtn-Enc and apoferritin were loaded with various amount of acidic Fe ( NH4 ) 2 ( SO4 ) 2 ranging from 0 to 1700 µM . Protein mixtures were incubated in room temperature for 3 hrs before desalting in Zebra spin desalting columns ( 7 kDa cut-off , Thermo Fisher Scientific , UK ) to remove free iron ions . The protein concentration was determined using PierceMicroplate BCA assay kit ( Thermo Fisher Scientific ) . The protein standard curve was plotted according to the manufacturer . The Fe content in the samples was determined using modified ferrozine assay ( Riemer et al . , 2004 ) . Briefly speaking , 100 μl protein sample was mixed with 100 μl mixture of equal volume of 1 . 4 M HCl and 4 . 5% ( w/v ) KMnO4 and incubated at 60 °C for 2 hrs . 20 μl of the iron-detection reagent ( 6 . 5 mM ferrozine , 6 . 5 mM neocuproine , 2 . 5 M ammonium acetate , and 1 M ascorbic acid dissolved in H2O ) was added to the cooled tubes . 30 min later , 200 μl of the solution was transferred into a well of 96-well plate and the absorbance at 562 nm was measured on the plate reader Spectramax M5 ( Molecular Devices , UK ) . The standard curve was plotted using various concentrations of FeCl3 ( in 10 mM HCl ) diluted in the gel-filtration buffer . Three technical repeats were performed for both the ferrozine and microplate BCA assays . Samples analyzed by ICP-MS were prepared in the same way by mixing protein and ferrous ions and desalting . The proteins used in the Fe loading experiment came from a single preparation . The peroxidase activity of EncFtnsH was determined by measuring the oxidation of ortho-phenylenediamine ( OP ) by H2O2 Pesek et al . ( 2011 ) . EncFtnsH decameric and monomeric fractions purified from MM were both used in the assay . Ortho-phenylenediamine was prepared as a 92 . 5 mM stock solution in 50 mM Tris-HCl ( pH 8 . 0 ) . 80 , 70 , 60 , 50 , 40 , 30 , 20 and 10 mM of OP were prepared by diluting the stock solution in the 50 mM Tris-HCl ( pH 8 . 0 ) . 100 μl of each diluted OP was added to a 96-well plate in 3 repeats . 1 μl of 32 µM protein was supplemented into each well to a final concentration of 160 nM , followed by the addition of 2 μl of 30% H2O2 . After 15 min shaking in the dark , the reaction was stopped by adding 100 μl of 0 . 5 M H2SO4 . The peroxidase activity was measured by monitoring the absorbance at 490 nm in the SpectraMax M5 Microplate Reader ( Molecular Devices ) ( Pesek et al . , 2011 ) . Protein samples were diluted 50-fold into a solution of 2 . 5% HNO3 ( Suprapur , Merck , UK ) containing 20 µg/L Pt as internal standard . Matrix-matched elemental standards ( containing analyte metal concentrations 0 – 1000 µg/L ) were prepared by serial dilution from individual metal standard stocks ( VWR ) with identical solution compositions , including the internal standard . All standards and samples were analyzed by ICP-MS using a Thermo x-series instrument ( Thermo Fisher Scientific ) operating in collision cell mode ( using 3 . 0 ml min-1 flow of 8% H2 in He as the collision gas ) . Isotopes 44Ca , 56Fe , 66Zn , 78Se , and 195Pt were monitored using the peak-jump method ( 100 sweeps , 25–30 ms dwell time on 5 channels per isotope , separated by 0 . 02 atomic mass units ) in triplicate . The protein samples used in ICP-MS came from a single protein preparation . For native MS analysis , all protein samples were buffer exchanged into 100 mM ammonium acetate ( pH 8 . 0; adjusted with dropwise addition of 1% ammonia solution ) using Micro Biospin Chromatography Columns ( Bio-Rad , UK ) prior to analysis and the resulting protein samples were analyzed at a final concentration of ~5 µM ( oligomer concentration ) . In order to obtain Fe-bound EncFtn , 100 µM or 300 µM of freshly prepared FeCl2 was added to apo-EncFtnsH ( monomer peak ) immediately prior to buffer exchange into 100 mM ammonium acetate ( pH 8 . 0 ) . Samples were analyzed on a quadrupole ion-mobility time of flight instrument ( Synapt G2 , Waters Corp . , Manchester , UK ) , equipped with a nanomate nanoelectrospray infusion robot ( Advion Biosciences , Ithaca , NY ) . Instrument parameters were tuned to preserve non-covalent protein complexes . After optimization , typical parameters were: nanoelectrospray voltage 1 . 54 kV; sample cone 50 V; extractor cone 0 V; trap collision voltage 4 V; source temperature 80°C; and source backing pressure 5 . 5 mbar . For improved mass resolution the sample cone was raised to 155 V . Ion mobility mass spectrometry ( IM-MS ) was performed using the travelling-wave mobility cell in the Synapt G2 , employing nitrogen as the drift gas . Typically , the IMS wave velocity was set to 300 m/s; wave height to 15 V; and the IMS pressure was 1 . 8 mbar . All native MS experiments were performed on samples from two independent protein preparations . For collision cross section determination , IM-MS data was calibrated using denatured equine myoglobin and data was analyzed using Driftscope v2 . 5 and MassLynx v4 . 1 ( Waters Corp . , UK ) . Theoretical collision cross sections ( CCS ) were calculated from pdb files using IMPACT software v . 0 . 9 . 1 ( Marklund , 2015 ) . In order to obtain information on the topology of the EncFtnsH assembly , gas-phase dissociation of the Fe-associated EncFtnsH complex was achieved by increasing the sample cone and/or trap collision voltage prior to MS analysis . Size-exclusion chromatography ( ÄKTA-Micro; GE Healthcare ) coupled to UV , static light scattering and refractive index detection ( Viscotec SEC-MALS 20 and Viscotec RI Detector:VE3580; Malvern Instruments , UK ) were used to determine the molecular mass of fractions decamer and monomer of EncFtnsH in solution individually . Protein concentration was determined by measurement of absorbance at 280 nm and calculated using the extinction coefficient ε0 . 1%= 1 . 462 mg−1 ml-1 cm−1 . 100 μl of 1 . 43 mgml-1 fractions of EncFtnsH decamer and 4 . 03 mg ml -1 fractions of EncFtnsH monomer were run individually on a Superdex 200 10/300 GL size-exclusion column pre-equilibrated in 50 mM Tris-HCl ( pH 8 . 0 ) , 150 mM NaCl at 22°C with a flow rate of 0 . 5 ml/min . Light scattering , refractive index ( RI ) and A280nm were analyzed by a homo-polymer model ( OmniSEC software , v 5 . 1; Malvern Instruments ) using the following parameters for fractions of decamer and monomer: the extinction coefficient ( dA/dc ) at 280 nm was 1 . 46 AU mg ml−1 and specific refractive index increment ( dn/dc ) was 0 . 185 ml g−1 . The proteins analyzed by SEC-MALLS came from single protein preparation . Recombinant EncFtnsH fractions at 50 µM concentration were incubated with one molar equivalent of metal ions at room temperature for 2 hrs . Half of each sample was mixed with 5 x native loading buffer ( 65 mM Tris-HCl , pH 8 . 5 , 20% glycerol and 0 . 01% bromophenol blue ) and run on non-denaturing PAGE gels ( 10% acrylamide ) and run in Tris/glycine buffer , 200 V , 4 °C for 50 min . The remaining samples were left for an additional three hours prior to SDS-PAGE ( 15% acrylamide ) analysis . SDS-PAGE gels were run at room temperature at 200 V , room temperature for 50 min . Gels were stained with Coomassie Brilliant Blue R250 and scanned after de-staining in water . The proteins used in this experiment came from single protein preparation . For analysis of the multimeric state of EncFtn proteins by analytical size-exclusion gel-filtration chromatography ( AGF ) 25 μl of 90 µM protein was loaded into Superdex 200 PC 3 . 2/30 column ( GE Healthcare ) at 15 °C with GF buffer running at 0 . 05 ml/min and pressure limit 0 . 45 MPa . In order to use AGF to determine how metal ions influence the assembly of EncFtnsH , 90 µM EncFtnsH monomer fractions were mixed with equal molar concentrations of metal ion solutions including FeSO4 in 0 . 1% ( v/v ) HCl , Fe ( NH4 ) 2 ( SO4 ) 2 , FeCl3 , CoCl2 , calcium acetate ( CaAc ) , ZnSO4 and MnCl2 at room temperature for 2 hrs prior to AGF analysis . Protein samples without metal titration were also analyzed as a control group . Both monomer and decamer fractions of EncFtnsH left at room temperature for 2 hrs , or overnight , were also analysed as controls to show the stability of the protein samples in the absence of additional metal ions . The AGF results have been repeated twice using two independent preparations of protein , of which only one representative trace is presented in the paper . Coordinates and structure factors for the structures presented in this paper have been deposited in the PDB under the following accession codes: EncFtnsH , 5DA5; EncFtnsH-E32A , 5L89; EncFtnsH-E62A , 5L8B; EncFtnsH-H65A , 5L8G ( DOIs for X-ray diffraction image data are shown in Table 4 ) . All MS datasets presented in this paper can be found , in the raw format at http://dx . doi . org/10 . 7488/ds/1449 .	Iron is essential for life as it is a key component of many different enzymes that participate in processes such as energy production and metabolism . However , iron can also be highly toxic to cells because it readily reacts with oxygen . This reaction can damage DNA , proteins and the membranes that surround cells . To balance the cell’s need for iron against its potential damaging effects , organisms have evolved iron storage proteins known as ferritins that form cage-like structures . The ferritins convert iron into a less reactive form that is mineralised and safely stored in the central cavity of the ferritin cage and is available for cells when they need it . Recently , a new family of ferritins known as encapsulated ferritins have been found in some microorganisms . These ferritins are found in bacterial genomes with a gene that codes for a protein cage called an encapsulin . Although the structure of the encapsulin cage is known to look like the shell of a virus , the structure that the encapsulated ferritin itself forms is not known . It is also not clear how encapsulin and the encapsulated ferritin work together to store iron . He et al . have now used the techniques of X-ray crystallography and mass spectrometry to determine the structure of the encapsulated ferritin found in some bacteria . The encapsulated ferritin forms a ring-shaped doughnut in which ten subunits of ferritin are arranged in a ring; this is totally different from the enclosed cages that other ferritins form . Biochemical studies revealed that the encapsulated ferritin is able to convert iron into a less reactive form , but it cannot store iron on its own since it does not form a cage . Thus , the encapsulated ferritin needs to be housed within the encapsulin cage to store iron . Further work is needed to investigate how iron moves into the encapsulin cage to reach the ferritin proteins . Some organisms have both standard ferritin cages and encapsulated ferritins; why this is the case also remains to be discovered .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2016	Structural characterization of encapsulated ferritin provides insight into iron storage in bacterial nanocompartments
The time it takes a sound to travel from source to ear differs between the ears and creates an interaural delay . It varies systematically with spatial direction and is generally modeled as a pure time delay , independent of frequency . In acoustical recordings , we found that interaural delay varies with frequency at a fine scale . In physiological recordings of midbrain neurons sensitive to interaural delay , we found that preferred delay also varies with sound frequency . Similar observations reported earlier were not incorporated in a functional framework . We find that the frequency dependence of acoustical and physiological interaural delays are matched in key respects . This suggests that binaural neurons are tuned to acoustical features of ecological environments , rather than to fixed interaural delays . Using recordings from the nerve and brainstem we show that this tuning may emerge from neurons detecting coincidences between input fibers that are mistuned in frequency . Acoustical waves produced by a sound source reach the two ears at slightly different times depending on its spatial position . Interaural time differences ( ITDs ) are used by many species to localize sounds in the horizontal plane . In mammals , neurons in the medial superior olive ( MSO ) , just three synapses away from the cochlear receptors , are sensitive to both ITD and sound frequency . It is thought that their activity encodes ITD in a frequency band , and is then interpreted in terms of spatial position . They project to neurons in the inferior colliculus ( IC ) , which inherit these properties . The firing rate of these neurons is strongly modulated by the ITD of a tone presented binaurally through earphones . For a 600 Hz tone , the neuron shown in Figure 1A , recorded in the IC of a cat , responds maximally at a ‘best ITD’ of 345 μs , close to the maximum natural ITD reported for cat ( about 350 μs ) ( Roth et al . , 1980 ) . In the Jeffress model , the textbook model of ITD processing ( Jeffress , 1948; Joris and Yin , 2007 ) , ITD tuning arises from the detection of coincidences between spikes relayed from auditory nerve fibers tuned to the same frequency at the two ears , and the neuron responds maximally when the sound's ITD equals the mismatch in axonal conduction delay between inputs from the two cochleae . This model predicts that , for a given neuron , ITD tuning is independent of the sound's frequency . However , at 900 Hz , the neuron of Figure 1 is tuned to an ITD of 158 μs and barely responds to an ITD of 345 μs , while at 400 Hz the neuron responds maximally at 500 μs and is much less responsive at 345 μs ( Figure 1A ) . In fact , the range of best ITDs that this neuron shows at different stimulus frequencies spans several hundred μs ( Figure 1B ) , which is large considering that the maximum natural ITD in cats is ∼350 μs . Thus , the ITD tuning of this neuron varies broadly with sound frequency . This property has been observed in binaural neurons of many species , including cats ( Yin and Kuwada , 1983 ) , guinea pigs ( McAlpine et al . , 1996; Palmer and Kuwada , 2005 ) , rabbits ( Kuwada et al . , 1987 ) , chinchilla ( Bremen and Joris , 2013 ) , gerbils ( Day and Semple , 2011 ) and dogs ( Goldberg and Brown , 1969 ) , but no functional significance has been associated with it . Readers should note that this property is observed within neurons as a function of frequency , and differs from the population property that has also been widely observed , where neurons tuned to low frequencies tend to have larger best ITDs than high-frequency neurons ( McAlpine et al . , 1996; Hancock and Delgutte , 2004; Joris et al . , 2006; Day and Semple , 2011; Bremen and Joris , 2013 ) . 10 . 7554/eLife . 06072 . 003Figure 1 . Frequency-dependence of best delays . ( A ) Firing rate vs ITD for one neuron , to tones between 400 Hz ( blue ) and 1200 Hz ( orange ) . ( B ) Best interaural time difference ( ITD ) ( colored dots , left axis ) , and sync-rate ( SR ) , ( black line , right axis ) vs frequency for the same cell . Data points with SR higher than 80% of the maximum value are used to calculate the range of best ITD ( shaded area above dotted line ) . ( C ) Distribution of the range of best ITDs across all 186 cells . ( D ) Best phase ( BP ) vs frequency and linear regression . The characteristic phase ( CP , here 0 . 27 cycle ) is the intercept; the characteristic delay ( CD , here −0 . 102 ms ) is the slope . ( E ) Distribution of CP across all cells ( N = 186 ) . ( F ) Distribution of CD . ( G ) CD vs CP across all cells . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 00310 . 7554/eLife . 06072 . 004Figure 1—figure supplement 1 . Linearity of BP vs frequency curves . ( A ) Distribution of residual error in linear fits of BP vs frequency in cells ( green ) and in acoustical predictions ( blue ) . ( B ) Statistical significance of linear fits . Cells are selected for further analysis when p < 0 . 05 . Percentage: proportion of nonlinear cells and prediction from the acoustics . ( C ) BP vs frequency and linear regressions for two cells with the same number of frequency points and different residual errors . ( D ) Same as C , for acoustical recordings . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 00410 . 7554/eLife . 06072 . 005Figure 1—figure supplement 2 . Statistical significance of CP-CD correlation . ( A ) Characteristic delay ( CD ) vs characteristic phase ( CP ) for all cells . Spearman's rank correlation ρ is −0 . 357 . ( B ) Illustration of spurious correlations due to noise . Two subsets of BP vs frequency data points ( red and blue dots ) from the same neuron are fitted with lines: intercept ( CP ) and slope ( CD ) are inversely correlated . The solid curve shows the SR ( see ‘Materials and methods’ ) . ( C ) Linear regression performed on bootstrap samples for 4 cells: CD and CP are inversely correlated for each cell , but positively correlated overall . ( D–G ) Statistical test of CP-CD correlation . ( D ) Data points are generated at random under the hypothesis that CP and CD are independent , using the distributions measured in cells . ( E ) Correlated noise is added to each ( CP , CD ) point shown in D , with the correlated noise distributions previously measured as in panel C . Each new point is shown in green , connected by a line to the original point ( blue ) . ( F ) Correlation is measured across all green points of E ( dashed: linear regression ) . The procedure D–F is reiterated many times with new sets of random samples . ( G ) The distribution of Spearman's ρ in the generated data points has a small negative bias , much smaller than in the original data points ( dashed , p < 10−6 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 00510 . 7554/eLife . 06072 . 006Figure 1—figure supplement 3 . Best frequency ( BF ) and characteristic frequency ( CF ) of recorded cells . ( A ) Each cell's BF plotted against the CF . ( B ) Distributions of BF and CF in the population of cells . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 00610 . 7554/eLife . 06072 . 007Figure 1—figure supplement 4 . BP vs tone frequency for 13 sample cells . ( A ) BP vs frequency for cells with low CP values ( below 0 . 2 cycles ) , with regression lines ( dotted ) . ( B ) BP vs frequency for cells with higher CP values . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 007 On the other hand , it is known that in natural environments , the acoustic ITD itself varies not only with spatial position but also with frequency , due to sound diffraction by the head ( Roth et al . , 1980 ) and early reflections from the ground ( Gourévitch and Brette , 2012 ) . Here we show with acoustical recordings and simulations that the variation of ITD with frequency can be substantial at the scale of a single neuron's receptive field . We then show that the detailed statistics of this variation are matched by the tuning of binaural neurons . Finally , we show that slight mismatches in the frequency tuning of auditory nerve fibers projecting to binaural neurons are a plausible mechanism to explain tuning to complex binaural features of ecological environments , and we show the existence of asymmetries in the spectral properties of MSO inputs using intracellular recordings . We examined ITD tuning in 186 IC neurons of cats tuned at a characteristic frequency ( CF , frequency of lowest rate threshold ) between 100 and 3300 Hz ( see Figure 1—figure supplement 3 ) . We found that the best ITDs of a neuron , at the different frequencies to which they were sensitive , spanned on average a range of 128 μs ( ± 240 μs ) ( Figure 1C ) . This extent is large , considering the maximum natural ITD reported for cats ( ∼350 μs ) and their ability to discriminate ITDs differing by only 20 μs ( Wakeford and Robinson , 1974 ) . The dependence of best ITD on sound frequency can be analyzed more precisely ( Yin and Kuwada , 1983 ) . The best ITD can be expressed relative to the period of the tone's frequency f , and is then called the best phase ( BP ) : BP = best ITD × f ( Figure 1D ) . For a neuron with a fixed best ITD that does not depend on frequency , BP is a linear function of frequency with 0 y-intercept . But the neuron shown in Figure 1D does not fit this simple relationship: a better fit is a linear relationship with an offset at 0 Hz , called the characteristic phase ( CP ) , measured between −0 . 5 and 0 . 5 cycle . We computed circular-linear regressions for all 186 neurons , which were highly significant in most cases ( Figure 1—figure supplement 1; see also other examples on Figure 1—figure supplement 4 ) . We found that CP was broadly distributed across an entire cycle ( Figure 1E ) , indicating that the best ITD of many neurons is not fixed but depends on frequency . The slopes of linear regressions are called characteristic delays ( CD; Figure 1D , F ) ( Yin and Kuwada , 1983 ) . If neurons were tuned to fixed ITDs in the contralateral field , we would expect CDs to be distributed between approximately 0 μs and 350 μs . In our neurons , the CDs are mainly positive ( corresponding to contralateral leading sounds ) and mostly within the natural range of 350 μs , but a minority of cells have negative CDs ( 38% ) and a smaller minority have CDs larger than 350 μs ( 19% , grey area in Figure 1F ) . Most intriguingly , CDs are negatively correlated with CPs ( Figure 1G ) . We checked with bootstrap analysis that this negative correlation is not due to measurement artifacts ( Figure 1—figure supplement 2 ) . All these observations are consistent with previous findings in other species ( Yin and Kuwada , 1983; McAlpine et al . , 1996; Palmer and Kuwada , 2005 ) . We looked in the acoustics for a functional rationale for the frequency-dependence of neural tuning to ITD within single neurons . It is known that the acoustic ITD itself varies not only with spatial position but also with frequency , due to sound diffraction by head and body ( Roth et al . , 1980 ) . This variation can be quantified by analyzing head-related transfer functions ( HRTFs ) , which measure the acoustical filtering of the head and body for sources at various positions . Figure 2A shows the variation of phase ITD with frequency for different source directions in an anaesthetized cat ( Tollin and Koka , 2009 ) . The ‘phase’ ITD reported here is the value of the ITD of a pure tone stimulus at a given frequency ( see ‘Materials and methods’ for additional ITD definitions ) . These patterns are consistent with previous acoustical measurements in cats using tones ( Roth et al . , 1980 ) . We also found similar patterns in high-resolution recordings on a taxidermist model of a cat with a natural posture ( Figure 2B ) . We checked that these patterns were not due to possible limitations of acoustical recordings by comparing them with numerical simulations of HRTF obtained on a 3D model of the same cat ( Rébillat et al . , 2014 ) ( Figure 2C , D ) . The global structure of these patterns is consistent with a spherical model of the head ( Kuhn , 1977 ) ( Figure 2E ) . However , their fine structure depends on posture ( Figure 2C , D , cat's head in a different position ) , on the presence of a ground ( Figure 2E , F , spherical head model without and with a ground plane ) , and on whether the source is in the front or in the back ( Figure 2A–D , solid vs dashed curves ) —because of reflections on the body of the cat . We remark that reflections on the ground or on the body of the cat come too early to be separated from the direct signal ( Gourévitch and Brette , 2012 ) and they must be considered as an integral part of the binaural signal received by the animal . Thus in ecological conditions , the ITD generally varies with frequency for a given source position . 10 . 7554/eLife . 06072 . 008Figure 2 . Frequency-dependence of ITD in several acoustical datasets . ( A ) ITD vs frequency for sound directions on the horizontal plane ( azimuth 15–90° , spaced by 15° ) , measured in a live cat and previously reported ( Roth et al . , 1980 ) . ( B ) Acoustical measurements on a taxidermist model of a cat in a large anechoic room ( same azimuths ) . Dashed curves show symmetric positions for sources to the back of the animal . Note that the head is tilted to the right; azimuths are relative to the head ( not the body ) . ( C ) Numerical calculation of ITDs by boundary element method ( BEM ) simulation on a 3D model of the same cat as B ( grey shape ) , obtained from photographs . ( D ) Same as C , but with a straightened head . ( E ) Analytical calculation of ITDs for a spherical rigid head . ( F ) Same as E , but with an additional reflection from the ground . Head and source are placed 1 . 7 meter from each other and 20 cm above the ground . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 00810 . 7554/eLife . 06072 . 009Figure 2—figure supplement 1 . Envelope and fine-structure ITDs . ( A ) Top: linear IPD curve ( constant ITD ) with ITDg = 5 ms ( group ITD , see ‘Materials and methods’ ) . Bottom: If the IPD is nonlinear ( gray curve ) , then it can be locally approximated with an affine function ( black plain and dotted line ) . This introduces a non-zero IDI ( Interaural Diffraction Index , see ‘Materials and methods’ ) . ( B ) an amplitude modulated tone ( top panel , black: envelope , gray: signal ) models the left monaural signal . The right signal is passed through model head-related transfer functions ( HRTFs ) with either ITDg = 5 ms , and IDI = 0 cycles ( middle panel ) or ITDg = 5 ms and IDI = 0 . 5 cycles ( bottom panel ) . The right signal ( blue: signal , green: envelope ) is delayed by the amount of the ITDg , while the fine structure of the signal undergoes an additional phase shift equal to the IDI ( see text ) . ( C ) interaural cross-correlation functions in the two cases IDI = 0 ( top ) and IDI = 0 . 5 ( bottom ) . The envelope peak ( green segment ) is unaffected by the IDI , while the fine structure peak ( blue segment ) is shifted by an amount ( in phase ) equal to the IDI . ( D ) ITDp ( phase ITD , see ‘Materials and methods’ ) and ITDg for one position of the cat HRTFs ( top panel ) , IDI for the same position ( bottom panel ) as a function of frequency . ( E ) Top: Distribution of ITDp ( dashed line ) and ITDg ( solid line ) in the cat over all positions and frequencies . Bottom: Distribution of IDI . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 009 Natural signals do not consist of a single frequency but typically have a certain bandwidth , and individual binaural neurons integrate signal information from a range of frequencies , which matches the bandwidth of filtering in the cochlea ( Mc Laughlin et al . , 2007 ) . In view of the dependence of the acoustical binaural cue ( ITD ) on frequency ( Figure 2 ) , the next question becomes how this compares to the neural dependence of best ITD on frequency . We tested whether the features seen in the electrophysiological data could be explained by the hypothesis that neurons are tuned to frequency-dependent ITDs as found in ecological environments . This is illustrated in Figure 3 which shows frequency-dependent ITDs and zooms in on the frequency band 600–1000 Hz for 3 azimuths . If a neuron were tuned to a fixed ITD ( e . g . , 325 μs , in Figure 3B , top ) , then it would be most responsive to different azimuths at different frequencies . On the other hand , if the neuron were tuned to a fixed spatial azimuth , then its best ITD would vary with frequency to match the variation of ITD with frequency at that position ( Figure 3B , bottom ) . Since the relationship between ITD and frequency is not fixed but depends on variables in listener , source , and environment—as illustrated in Figure 2—we looked for statistical correspondences between acoustical and neural measurements . 10 . 7554/eLife . 06072 . 010Figure 3 . Tuning to frequency-dependent ITDs . ( A ) Left , head-related transfer functions ( HRTFs ) are measured binaurally for different speaker positions . Right , ITD vs frequency in cat at 60 , 70 and 80° on the horizontal plane . ( B ) A neuron for which best ITD is fixed across frequency ( top , black line ) is tuned to different azimuths depending on frequency , while a neuron with fixed azimuth tuning has a frequency-dependent best ITD ( bottom , purple line ) . ( C ) IPD vs frequency at 70° over a 300 Hz window around 800 Hz ( green curve and circles ) . The black segment represents an ITD of 325 μs that is fixed across frequency , equal to the ITD at 800 Hz . The purple segment represents the best linear approximation of IPD around that frequency ( intercept 0 . 12 cycle , slope 167 μs ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 01010 . 7554/eLife . 06072 . 011Figure 3—figure supplement 1 . IPD vs frequency for six different directions , around 650 Hz and 1600 Hz , with circular-linear fits . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 011 We analyzed the frequency-dependence of ITD in the acoustical recordings in the same way as we analyzed the frequency-dependence of best ITD in cells . For each azimuth and center frequency , we extracted a CD and CP from the acoustical data . For example , for a 70° azimuth and center frequency of 800 Hz ( Figure 3B , bottom ) , we approximate the interaural phase difference ( IPD = ITD × f ) for that location by a linear function of frequency ( Figure 3C , purple line ) . The acoustical CD and CP are the slope and intercept of the linear regression . Other examples are shown on Figure 3—figure supplement 1 . Physically , the acoustical CD is the envelope ITD and the acoustical CP is the difference between envelope and fine structure ITD , expressed in cycles ( see Figure 2—figure supplement 1 ) . We then produced statistics by sampling center frequencies according to the CF distribution in the recorded neurons , and azimuths according to a uniform distribution in the contralateral hemifield ( including front and back ) . In agreement with the physiological data , the acoustic measurements show a unimodal and broad CP distribution , with a small but significant positive bias ( Figure 4A ) . Consistent with the measured neurons , the acoustic CDs are mainly positive and mostly within 350 μs , but with a sizeable number of data points with negative or with large CDs ( Figure 4B , and Figure 4—figure supplement 3 for negative CDs ) . While some large neural CDs ( >500 μs ) lie outside the range of ITDs ( Figure 2 ) , all remain inside the range of acoustical CD . Finally , the acoustic data also show an inverse correlation between CD and CP ( Figure 4C ) . Thus , key properties of neural CD and CP , which are the main metrics that have been used in the description of tuning to ITD , are well-matched to binaural acoustics studied with these same metrics . 10 . 7554/eLife . 06072 . 012Figure 4 . Acoustical analysis . ( A ) Distribution of CP in the cells ( green ) and in the acoustics based on acoustical measurements ( blue ) . Error bars represent the mean ± STD/2 , and percentages the proportion of positive/negative values . ( B ) Distribution of CD in cells ( green ) and in acoustics ( blue ) . ( C ) CD vs CP in the acoustics . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 01210 . 7554/eLife . 06072 . 013Figure 4—figure supplement 1 . Acoustical predictions of CD and CP distributions for various prior spatial distributions . First column: distribution of preferred positions . Second and third columns: prediction of CP and CD distributions ( numbers are proportions of positive and negative values ) . Fourth column: joint CP-CD distributions ( 200 sample cells drawn at random ) . ( A ) Uniform distribution of preferred positions in the 0–90° quadrant . ( B ) Distribution of preferred positions inferred from cell recordings ( best fits to HRTFs , see ‘Materials and methods’ ) . ( C ) Bias for positions near 90° . ( D ) Bias for positions near the center . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 01310 . 7554/eLife . 06072 . 014Figure 4—figure supplement 2 . Acoustical predictions of CD and CP distributions in low and high frequency regions . ( A ) Distribution of CP in the cells ( green ) and in the acoustics based on acoustical measurements ( blue ) , for frequency bands below 1 kHz . ( B ) Distribution of CD in cells ( green ) and in acoustics ( blue ) , for frequency bands below 1 kHz . ( C ) CD vs CP in the acoustics , for frequency bands below 1 kHz . ( D–F ) same as A–C for frequency bands above 1 kHz . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 01410 . 7554/eLife . 06072 . 015Figure 4—figure supplement 3 . Negative acoustical CDs . Proportion of negative ( i . e . , ispilateral-leading ) CDs in the cat HRTFs as a function of azimuth . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 015 Quantitatively , the distributions of CD and CP in the acoustic data depend on how space is sampled , which could alternatively be uniform over directions in the front only ( Figure 4—figure supplement 1A ) , or inferred from the electrophysiological recordings ( Figure 4—figure supplement 1B ) , or biased towards the side ( Figure 4—figure supplement 1C ) or the center ( Figure 4—figure supplement 1D ) , as suggested in the barn owl ( Fischer and Peña , 2011 ) . In particular , the proportion of negative CDs varies between 8% and 31% depending on the choice of azimuth distribution , because negative CDs are observed mostly for azimuth near 0° or 180° ( Figure 4—figure supplement 3 ) . Distributions of CD and CP also quantitatively depend on the frequency range of the analysis ( CF < 1 kHz in Figure 4—figure supplement 2A–C; CF > 1 kHz in Figure 4—figure supplement 2D–F ) . However , despite the quantitative differences , the same qualitative features remain . Other factors may contribute quantitative variations , such as posture , reflections off the ground , distance and elevation . Thus , the statistics of frequency-dependent ITDs in acoustical recordings qualitatively match those of frequency-dependent neural tuning to ITD . The inverse correlation between CP and CD can be explained by the variation of ITD with frequency . In a simple spherical head model ( Figure 5A ) , ITDs are larger at low frequency than at high frequency , but these variations are small on a local scale ( Figure 5B , blue [Kuhn , 1977] ) . However , variations appear on a local frequency scale as soon as features of ecological environments are introduced , such as diffraction on the complex shape of head and body of real animals ( Figure 3 ) , or early reflections off the ground ( Figure 5B , green [Gourévitch and Brette , 2012] ) . As a result of these variations , the IPD vs frequency curve is non-linear ( Figure 5C , D ) . When a tangent is moved along this curve ( dashed lines in Figure 5C ) , the slope decreases at the same time as the intercept increases . Because slope and intercept correspond to acoustical CD and CP , this means that for neurons tuned to the same spatial configuration but different frequencies , CD and CP should be inversely correlated ( Figure 5E , F ) . These variations in CD and CP across frequency are small for a simple spherical head ( Figure 5F ) , but they become large as soon as a ground plane is included ( Figure 5E ) . We note that reflections off the ground cannot be temporally separated from the direct signal because delays are very short ( about 150 μs for a source 1 . 5 m away from the cat's head [Gourévitch and Brette , 2012] ) and must thus be considered as part of the signal reaching the two ears . Thus the acoustical space encountered by the animal cannot be adequately described by fixed ITDs . As a result , it cannot be unambiguously represented by neurons tuned to only fixed ITDs ( corresponding to the vertical line CP = 0 in Figure 5E , F ) . 10 . 7554/eLife . 06072 . 016Figure 5 . Theoretical explanation of inverse CP-CD relationship . ( A ) A spherical head model with a ground reflection . ( B ) ITD vs frequency in the spherical model for a source at 70° , with ( green ) and without ( blue ) ground reflection . ( C , D ) IPD vs frequency for the same position as in B ( green and blue ) and for other positions between 0 and 90° ( light gray curves ) . ( E , F ) Predicted CD vs CP for the two cases . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 016 In the textbook model of ITD processing ( Jeffress , 1948; Joris and Yin , 2007 ) , a binaural neuron in the MSO detects coincidences between spikes produced by monaural neurons driven by the left and right ear that are tuned to the same CF . Best ITDs near 0 ms imply that the left and right signals arrive coincidentally at the binaural neuron; ITDs >0 ms imply that the inputs from the contralateral ear reach the binaural neuron with some delay relative to those of the ipsilateral ear . However , this mechanism produces frequency-independent best ITDs , that is , CP = 0 , which is not consistent with most of the physiological data ( Figure 1E ) ( Kuwada et al . , 1997 ) . Frequency-dependent best delays could be produced by small mismatches in the CFs of the monaural inputs to a binaural cell ( Schroeder , 1977; Shamma et al . , 1989; Bonham and Lewis , 1999 ) , and some features of binaural responses are consistent with such mismatches ( Joris et al . , 2006; Day and Semple , 2011 ) . We studied the effects of mismatches in CF with a coincidence analysis of responses of several hundred cat auditory nerve fibers . We model the response of a binaural coincidence detector neuron receiving inputs from two slightly different points on the cochlea , leading to a CF mismatch ( Figure 6A , top panel ) . This is achieved by counting the coincidences between the spike trains of two recorded fibers with slightly different frequency tuning ( Figure 6A , right panel ) , in response to a range of pure tones . By varying the delay between the spike trains , ‘pseudobinaural’ ITD curves at different frequencies are generated . 10 . 7554/eLife . 06072 . 017Figure 6 . Mechanism for frequency-dependent neural tuning . ( A ) Schematics of the coincidence analysis . The left schematic illustrates the concept of cochlear disparities . The trapezoids schematize the cochlear basilar membrane . A left and right fiber originate from a different cochlear place and converge on a binaural neuron . The right schematic illustrates the counting of coincidences between spike trains from two fibers in response to a single tone . Due to the cochlear traveling wave , the spike trains of the more apical ( green ) fiber are expected to be delayed in time and lagged in phase relative to the more basal ( blue ) fiber . ( B ) Threshold tuning curves of the two example fibers . ( C ) Pseudobinaural tuning curves: Coincidence counts as a function of ITD for a pair of fibers for different tone frequencies . Each curve is color coded with the frequency of the stimulus , scale is presented below the plot . ( D ) BP as a function of frequency for the same nerve pair as C . ( E ) CD vs CP over a population of coincidence detectors receiving inputs from cat auditory nerve fibers with mismatched CF ( <0 . 1 octave; CF < 3 . 3 kHz ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 017 Figure 6C shows the results of coincidence analysis on responses to tones with frequencies between 400 and 2200 Hz ( bin width = 50 μs ) , for two fibers with similar but slightly different frequency tuning ( CF = 1092 Hz and 1133 Hz , Figure 6B ) . The maxima of these ITD-curves are the best ITD of a model binaural coincidence detector receiving inputs from those two fibers: they show a frequency dependent BD with CP = 0 . 234 and CD = −0 . 194 ms ( Figure 6D ) . Using auditory nerve data from a single animal , we simulated the CP and CD of a population of binaural coincidence detector cells . We counted coincidences for different delays between spike trains of 73 pairs of fibers with slightly mismatched CF ( ≤0 . 1 octave ) , and processed the resulting coincidence counts with a CP-CD analysis identical to that used on real binaural neurons ( Figure 1 ) . Figure 6E shows that here as well , CP is broadly distributed and inversely correlated with CD . Note that the CP distribution is centered on 0 because we symmetrized the distribution by representing each fiber pair twice to simulate random mismatches between the inputs from the two sides , where sometimes the ipsilateral fiber is higher in CF and sometimes the contralateral fiber ( i . e . , reflecting both positive and negative CF mismatches ) . The distribution of Figure 6E is consistent with the phase characteristics of the cochlear traveling wave , which generates frequency-dependent delays and ultimately drives the hair cells and auditory nerve fibers ( Schroeder , 1977; Shamma et al . , 1989; Bonham and Lewis , 1999; Day and Semple , 2011 ) , and shows that very small CF mismatches are sufficient to produce CPs of the same magnitude as measured in binaural cells . In order to provide direct evidence of CF disparities in mammals , we obtained in vivo patch clamp recordings from 6 gerbil MSO cells , using the method described in a recent study ( Franken et al . , 2015 ) . The rate of excitatory presynaptic events ( EPSPs ) was measured during monaural ipsi- or contralateral presentation of pure tones at different frequencies ( Figure 7 ) . The data show that afferents to MSO cells can differ in their spectral composition , albeit in a complex manner . This confirms an observation in juxtacellular recordings ( see Figure 5 of ( van der Heijden et al . , 2013 ) ) and suggests that CF mismatches in mammals may play a role in shaping tuning of ITD sensitive cells . 10 . 7554/eLife . 06072 . 018Figure 7 . Asymmetries in frequency tuning in the excitatory inputs to the gerbil medial superior olive ( MSO ) . ( A–F ) Rate of excitatory presynaptic events ( EPSPs ) in 6 MSO cells , with different CF , in response to tones as a function of frequency . Stimuli are presented ipsilaterally ( blue function ) or contralaterally ( green function ) . Only EPSPs ≥ the median EPSP amplitude are included . Functions were smoothed using a 3-point running average . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 018 Random cochlear disparities are not a sufficient mechanism to account for the physiological distribution ( Figure 1E , F ) because they fail to account for the positive bias of CP and CD values . Such a bias can be obtained by systematic cochlear disparities ( i . e . , where the contralateral inputs are tuned to lower CFs than the ipsilateral inputs ) ( Joris et al . , 2006 ) . It can also be obtained by considering mismatches in axonal conduction delays , in addition to CF mismatches: axonal delays add to the CD without changing the CP . Such delay mismatches could result from the longer contralateral than ipsilateral path length to reach the off-midline MSO , or from other structural axonal differences between contra- and ipsilateral branches ( Seidl et al . , 2010 ) . The binaural tuning in our sample of IC neurons , quantified with traditional measures of CD and CP , matches that of many previous physiological studies ( Palmer and Kuwada , 2005 ) . However , we go further by making an explicit comparison with distributions of the same metrics applied to acoustical measurements . This leads to two new insights . First , we provide evidence that the distribution of neural CDs and CPs and acoustical CDs and CPs are similarly constrained . Second , large CDs are present acoustically . Taken together our results suggest that the binaural tuning of IC cells is constrained by the range of delays that the animal experiences . Our findings and interpretation relate to two points that have been much discussed in the literature . Non-zero CPs in the neural data have puzzled investigators since the first reports of ITD-sensitivity , both in terms of their physiological origin and their functional significance ( Rose et al . , 1966; Yin and Kuwada , 1983 ) . Perhaps even more puzzling has been the discrepancy observed between the distribution of physiological CDs and acoustical ( phase ) ITDs , for example , as reported in HRTF measurements ( for cat: [Roth et al . , 1980; Tollin and Koka , 2009] , and Figure 2 ) . As was pointed out since the first physiological data became available ( McFadden , 1973 ) , neuronal CDs seem to cover an ‘unnecessarily’ wide range including ITDs that animals will not naturally encounter . Various interpretations have been given to this discrepancy ( McAlpine et al . , 2001 ) . However , the appropriate comparison is not between physiological CDs and acoustical phase ITDs , but with acoustical CDs ( i . e . , envelope ITDs ) . As shown here for acoustical measurements ( Figure 4B ) , due to the frequency dependence of the phase ITD , CDs have a wide distribution and actually exceed the range of physiological CDs . The range of acoustical CDs that drives a given cell can therefore exceed the range of phase ITDs computed from HRTF data . Because the frequency-dependence of ITD reported here reflects the physical interaction of the sound with the ears , head , body , and ground plane , it should also apply to other species . We applied our analysis to HRTFs of humans ( Figure 8A–C ) , for whom ITDs dominate below ∼1 . 5 kHz ( Wightman and Kistler , 1992; Macpherson and Middlebrooks , 2002 ) . The same basic relationship between acoustical CD and CP is present . 10 . 7554/eLife . 06072 . 019Figure 8 . Acoustical analysis of human . ( A–C ) Predictions using human HRTFs ( IRCAM LISTEN HRTF database ) for the distributions of acoustical CP ( A ) , CD ( B ) and 200 sample points from the joint CP-CD distribution ( C ) over a 100–1500 Hz frequency range . DOI: http://dx . doi . org/10 . 7554/eLife . 06072 . 019 Our study emphasizes the notion that the binaural cues that an auditory system has to process in ecological environments are much more complex and rich than in idealized settings . We have focused here on the contribution of sound diffraction by the head and body , a phenomenon that is always present , even in anechoic environments . Because the effect depends on the detailed morphology of the animal , it varies with posture in a systematic way ( see Figure 2C , D and [Rébillat et al . , 2014] ) . Early reflections , in particular from the ground ( again a constant element in ecological environments of terrestrial animals ) , produce interferences with the direct sound that result in large frequency-dependent variations of ITD , especially at low frequencies ( Gourévitch and Brette , 2012 ) . Ground reflections typically arrive very shortly after the direct sound and are therefore an integral part of the binaural signal received by the animal . The interferences they produce are determined primarily by the delay of the reflection ( related to the distance ) and secondarily by the nature of the reflecting object ( grass , sand , snow , etc ) . There are many other sources of complexity in ecological environments . For example , natural sound sources are rarely point sources producing spherical wavefronts , as assumed in most studies . Many are large ( a river ) and directional ( human speech ) , some are partially occluded by objects ( a prey hiding in a bush ) . The sound of one's own footstep travels through air but also through the body , which produces different acoustical cues ( Hood , 1962 ) . We suggest that the binaural system of animals is adapted to these complex natural acoustical cues . This complexity must be addressed by any sound localization theory and imposes particular difficulties ( Brette , 2010; Goodman and Brette , 2010; Goodman et al . , 2013 ) . For example , if ITD depends on the position of the pinnae , or posture , as is the case for the cat ( Figure 2 , and [Tollin and Koka , 2009] ) , then proprioceptive information must be taken into account to interpret binaural cues . Unarguably , this property makes the task more difficult for the animal , but not more difficult than taking into account other known causes of ITD variation—such as source distance and elevation—when mapping ITD to source direction . The fact that ITD depends on factors other than azimuthal position implies that binaural neurons in the MSO are not tuned to source direction per se , but rather to temporal binaural features of acoustical spaces—a notion that generalizes frequency-independent ITD . If binaural tuning develops from exposure to natural sounds ( Seidl and Grothe , 2005 ) , then it would be expected that its properties reflect those of ecological acoustical environments , especially given that very small cochlear disparities give rise to significant frequency-dependence of ITD tuning in binaural neurons ( Figure 6 ) . We can think of two ways of testing the hypothesis that binaural neurons are tuned to properties of ecological acoustical spaces . One is to raise animals in environments with structured but manipulated acoustical cues ( as opposed to unstructured noise as in [Seidl and Grothe , 2005] ) , for example using earplugs , and observe the changes in ITD tuning of binaural neurons . Another one is to measure spatial receptive fields of binaural neurons ( instead of ITD selectivity curves ) and test whether they depend on spectral properties of sounds . In previous work , nonlinear phase-frequency relationships ( i . e . , where the best ITD is not constant ) are typically considered to be ‘biological noise’ . Thus , even though frequency-dependent best ITDs are clearly an important characteristic of the physiological data ( Figure 1 ) , there are to our knowledge no functional interpretations of this characteristic . The observation ( Figure 6 ) that CF mismatches can generate nonlinear phase-frequency relationships ( CP ≠ 0 ) , combined with the presence of such relationships in the binaural and acoustic data , suggest a rather simple new view on neural binaural properties . The view is that limitations in the accuracy of wiring in the binaural system produce a property that benefits binaural hearing . Refinements of this wiring in combination with mismatches in axonal delays produce a range of binaural sensitivities well-matched to the acoustic scenes that the system is faced with . It was previously shown that random CF mismatches in the wiring to the contralateral and ipsilateral pathways to IC could account for the negative correlation between CF and range of best ITDs measured with delayed noise ( Joris et al . , 2006 ) . Here we show that appropriate CF and axonal delay mismatches can account for the dependence of best ITD on tone frequency within single cells , matching acoustical properties . Besides cochlear disparities , a range of mechanisms have been proposed to account for the best delays of ITD-sensitive neurons: axonal length ( Jeffress , 1948 ) , phase-locked inhibition ( Brand et al . , 2002 ) , asymmetric placement of the axon ( Zhou et al . , 2005 ) , asymmetry in synaptic kinetics between contra- and ipsilateral inputs ( Jercog et al . , 2010 ) , differences in axonal conduction time ( Seidl et al . , 2010 ) , and phase delays generated by the interaction of intrinsic properties with input spike patterns ( Franken et al . , 2015 ) . All of these proposals face difficulties to explain all the data available , and it is at present unclear which of these mechanisms , or perhaps mix of mechanisms , is in place , or whether perhaps the main mechanism has not been identified yet . Moreover , discussions of these various mechanisms usually focus on CD , leaving it unclear whether and how they would affect CP . Axonal length and conduction time are expected to generate pure time delays and would therefore not generate CPs different from 0 . The report first proposing phase-locked inhibition as a source of internal delay ( Brand et al . , 2002 ) provided model results illustrating how best ITD at different frequencies was little affected by stimulus frequency , that is , also predicted that inhibition would be equivalent to a pure time delay ( CP = 0 ) . Later , more extensive modeling ( Day and Semple , 2011 ) showed that adjustment of model inhibitory parameters allows creation of a wide range of non-zero CP values , but that CP remained within 0 . 1 cycle when more realistic inhibitory synaptic time constants were used . To our knowledge , asymmetrical axonal placement and asymmetry in synaptic kinetics have not been examined regarding a possible contribution to CP . However , these two proposals , as well as phase-locked inhibition , received little experimental support from a recent in vivo intracellular MSO study ( Franken et al . , 2015 ) . Cochlear disparities have been proposed before as a mechanism for generating internal delays , in place of axonal delay mismatches ( Schroeder , 1977; Shamma et al . , 1989; Bonham and Lewis , 1999 ) . The original form of that hypothesis , where such disparities are the sole source of internal delays , has been rejected in the barn owl: CF mismatches are observed but they are relatively small and do not correlate with ITD tuning ( Pena et al . , 2001; Fischer and Peña , 2009 ) . They are nonetheless significant as they contribute predicted delays of up to 50 μs ( Pena et al . , 2001 ) . Importantly , in contrast with the original cochlear disparity hypothesis , our proposed mechanism combines small CF mismatches ( just 40 Hz in Figure 6 ) and axonal delay mismatches . A mix of cochlear disparities and pure time delays was also proposed to account for non-linear phase-frequency relationships observed in gerbil MSO responses ( Day and Semple , 2011 ) . In cat MSO , ( Yin and Chan , 1990 ) reported that monaural best frequencies differed by 0 . 2 octaves or less for 13 of the 18 cells ( and more for 5 cells ) . As we show in Figure 6 , this order of magnitude is sufficient to produce the observed frequency-dependence of best ITD . Intracellular recordings from MSO neurons allowed us to directly measure the monaural inputs and confirm that they can differ in spectral properties . Additional experiments are needed to compare frequency-dependent properties of ITD tuning with the mismatched frequency tuning of monaural inputs . Our data are from the IC , one synapse removed from the sites of binaural interaction . In our mechanistic explanation , we have assumed that frequency-dependent properties of ITD tuning observed in IC neurons are inherited from the MSO . This is in agreement with electrophysiological data in the MSO of cats ( Yin and Chan , 1990 ) and gerbils ( Day and Semple , 2011; van der Heijden et al . , 2013 ) , which also show frequency-dependent ITD tuning ( broad distributions of CP ) . Alternatively , nonlinearities at the level of the IC could reflect convergence of inputs , for example , from binaural neurons in MSO , LSO , and DNLL ( McAlpine et al . , 1998 ) . For example , LSO neurons have CP close to 0 . 5 ( Joris , 1996; Tollin and Yin , 2005 ) . Neurons with CP close to 0 have been categorized as ‘peakers’ , because peaks of their ITD selectivity curves align across frequency , while neurons with CP close to 0 . 5 have been categorized as ‘troughers’ , because troughs align across frequency ( Batra et al . , 1997 ) . These two categories are traditionally considered as predominantly reflecting MSO or LSO input , respectively . Neurons with intermediate CP are then categorized as ‘tweeners’ , presumably reflecting a combination of MSO and LSO inputs . However , our electrophysiological data showed no clear categories in the distribution of CP ( Figure 1E ) . Instead , the distribution was broad and unimodal around CP = 0 , which argues against the categorization of cells as peaker , trougher , and tweener . In addition , electrophysiological studies in the MSO of cats ( Yin and Chan , 1990 ) and gerbils ( Day and Semple , 2011 ) also show cells with non-zero CP . Our auditory nerve data shows that very small mismatches in CF of inputs to binaural cells are sufficient to produce significantly non-zero CP , without postulating any additional mechanism than coincidence detection ( Figure 6 ) . In summary , while the mechanisms of ITD tuning remain unclear , one attractive feature of cochlear disparities is that they provide a simple mechanism to generate non-zero CPs , which we show here are a likely desirable property of the binaural system as they match the acoustics that animals face , when combined with a physiological source of pure time delays . Our methods for single unit recording have been described before: in the cat IC and auditory nerve ( Joris et al . , 2005 , 2006 ) , and in the gerbil MSO ( Franken et al . , 2015 ) . All procedures were approved by the institutional Animal Care Committee and were in accordance with the NIH Guide for the Care and Use of Laboratory Animals . In cat experiments , anesthesia was induced with acepromazine and ketamine and maintained for surgical preparation and recording with pentobarbital . Induction of anesthesia in gerbils was with ketamine and xylazine; maintenance was with ketamine and diazepam . All animals were placed on a heating pad in a double-walled sound-attenuated chamber . Sound stimuli were delivered dichotically with speakers coupled to earbars that were tightly coupled to the ear canals . The stimuli were generated digitally and were compensated for the acoustic transfer function measured with a probe microphone near the eardrum . In the cat , bullas were vented with tubing . The IC was exposed anterior to the tentorium; the auditory nerve was exposed via a posterior fossa approach . Single IC neurons were recorded with metal electrodes; auditory nerve fibers with high impedance glass micropipettes . The neural signal was amplified , filtered , timed ( 1 μs resolution ) and displayed using standard techniques . The dorsal border of the central nucleus of the IC was defined physiologically by the presence of background discharges phase-locked to binaural beats of low-frequency pure tones , and the IC was histologically processed to confirm the site of recording to the central nucleus . Binaural IC recordings were obtained from 31 animals , monaural auditory nerve recordings from 1 animal . Binaural beat stimuli were long duration ( typically 1 or 5 s ) tones presented over a range of frequencies bracketing the limits of the response area; the step increment was between 25 and 200 Hz to ensure adequate sampling . The tones to the two ears had a small ( 1 or 2 Hz ) difference . Typically the contralateral ear was at the higher frequency ( positive beat ) but the opposite ( negative beat ) was also often tested . The number of repetitions was typically between 1 and 10 , and the SPL was 60 dB . In vivo whole-cell recordings were obtained from MSO neurons in the gerbil ( Franken et al . , 2015 ) . Membrane potential was recorded in current clamp mode during monaural presentations of pure tones at different frequencies ( typically 1–3 repetitions of 50–250 ms long tones in 0 . 3 octave increments , at 60 dB SPL ) . Excitatory post-synaptic events were detected as described in Franken et al . ( 2015 ) . In both experiments , CF to binaural stimulation was determined with a threshold tracking algorithm . HRTFs of an anesthetized cat were obtained from a previous study ( Tollin and Koka , 2009 ) . They consist of 36 measurements in the horizontal plane with evenly spaced azimuth . The analysis was also performed on other HRTF sets ( Figure 2 ) . We measured HRTFs of a taxidermist model of cat from the Paris Museum of Natural History ( Figure 2B ) in a large anechoic chamber at IRCAM ( Paris ) . We used the same experimental setup as the ( IRCAM LISTEN HRTF database , date unknown ) using the sine-sweep method . Miniature microphones were placed at the entry of the meatus , which had been occluded by the taxidermy procedure . As a control , we obtained a 3D model of the same cat from photographs ( Figure 2C , insert ) and numerically calculated HRTF with a boundary element method ( Otani and Ise , 2006; Rébillat et al . , 2014 ) . The calculations were also performed on the 3D model after manually tilting the head 45° on the 3D model so as to align it with the body . HRTFs of a spherical head model were computed based on the analytic solution of the wave equation ( Figure 2E , F , Figure 5 ) , as detailed in ( Duda and Martens , 1998 ) . The head diameter was measured on the 3D , model of the cat ( d = 7 . 3 cm ) . Naturalistic ground reflections were included in the spherical model ( Figure 2F , Figure 5 ) using the method described in ( Gourévitch and Brette , 2012 ) . The head was placed 20 cm above ground , and the sound source was placed at the same distance of the ground , one meter away from the head . The ground was modeled with flow resistivity of 5 . 105 ( kP . s/m2 ) , which is between grass ( 105 ) and sand ( 106 ) . In Figure 8 , we used previously measured HRTFs of one randomly picked subject of the ( IRCAM LISTEN HRTF database ) , measured for 72 positions on the horizontal plane . The BA of a cell is defined as the azimuth that elicits the largest response in that cell . We assume that it occurs when the IPD is closest to the cell's BP across the relevant frequency range . Therefore the cell's BA is the azimuth that minimizes the following quantity:∑fd ( BP ( f ) , IPD ( f ) ) , where d ( . ) is the circular distance . The minimization is performed as previously described . Frequency points where chosen as in the acoustical analysis . Since BPs and IPDs are not necessarily measured at the same frequency points , the closest available frequency was chosen for IPD ( f ) , which was never further than a few Hz away given the high sampling rate of HRTF measurements .	When you hear a sound , such as someone calling your name , it is often possible to make a good estimate of where that sound came from . If the sound came from the left , it would reach your left ear before your right ear , and vice versa if the sound originated from your right . The time that passes between the sound reaching each ear is known as the ‘interaural time difference’ . Previous research has suggested that specific neurons in the brain respond to specific interaural time differences , and the brain then uses this interaural time difference to locate the sound . Sounds come in various frequencies from high-pitched alarms to low bass tones , and how a neuron responds to interaural time differences appears to change according to the frequency of the sound being played . For example , a given neuron may respond to a 200- microsecond interaural time difference when a tone is played at a high frequency , but show no response to this time difference when the tone is played at a low frequency . To date , researchers had been unable to explain why this occurs . Here , Benichoux et al . investigated this topic by playing a variety of sounds to anaesthetized cats . Electrodes were used to record the responses of individual neurons in the cats' brains , and the properties of the sound waves that reached the cats' ears were also recorded . These experiments revealed that the time it took a sound to travel from a location to each of the cats' ears , and consequently the interaural time difference , depended on whether it was a high-pitched or a low-pitched sound . This happened because different properties of the environment , such as the angle of the cat's head , affected specific frequencies in different ways . As expected , the neurons' responses were also affected by sound frequency . Indeed , the neurons' behaviour mirrored that of the sound waves themselves . This shows that neurons do not , as previously thought , simply react to specific interaural differences . Instead , these neurons use both sound frequency and interaural time differences to produce a thorough approximation of the sound's location . The precise mechanisms that generate this brain adaptation to the animal's environment remain to be determined .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2015	Neural tuning matches frequency-dependent time differences between the ears
Behavioral strategies employed for chemotaxis have been described across phyla , but the sensorimotor basis of this phenomenon has seldom been studied in naturalistic contexts . Here , we examine how signals experienced during free olfactory behaviors are processed by first-order olfactory sensory neurons ( OSNs ) of the Drosophila larva . We find that OSNs can act as differentiators that transiently normalize stimulus intensity—a property potentially derived from a combination of integral feedback and feed-forward regulation of olfactory transduction . In olfactory virtual reality experiments , we report that high activity levels of the OSN suppress turning , whereas low activity levels facilitate turning . Using a generalized linear model , we explain how peripheral encoding of olfactory stimuli modulates the probability of switching from a run to a turn . Our work clarifies the link between computations carried out at the sensory periphery and action selection underlying navigation in odor gradients . Chemosensation is an evolutionarily ancient sense found in nearly every living organism . In bacteria , chemotaxis allows individual cells to detect the presence of food and to accumulate in its vicinity . Multicellular organisms have evolved complex sensory systems to track temporal changes in the concentration of volatile odorant molecules relevant to their survival—food odors , pheromones associated with the presence of conspecifics and substances signaling danger . In turn , sensory perception drives behavioral strategies to forage , locate a mating partner and actively avoid danger ( Fraenkel and Gunn , 1961; Schöne , 1984 ) . Bacterial chemotaxis represents the archetype of orientation behavior in unicellular organisms: phases of relatively straight motion—called runs—alternate with changes in orientation—called tumbles—that randomize the direction of the next run ( Berg , 2004 ) . Accumulation near the source of an attractive chemical results from the elongation of runs in the direction of the gradient . In multicellular organisms , olfactory behaviors have been investigated in detail in the nematode Caenorhabditis elegans ( Bargmann , 2006a ) , which uses a combination of undirected turns ( ‘pirouettes’ ) and continuous correction of the orientation of individual runs ( ‘weathervaning’ ) ( Iino and Yoshida , 2009; Lockery , 2011 ) . The neural computations enabling animals with a central nervous system to orient in odor gradients , however , remain poorly understood . The Drosophila larva has the smallest known olfactory system analogous to that of vertebrates ( Cobb , 1999; Bargmann , 2006b; Gerber and Stocker , 2007; Vosshall and Stocker , 2007 ) . The larva achieves robust odor gradient ascents through an alternation of approximately straight runs and turning events ( Gomez-Marin et al . , 2011; Gershow et al . , 2012 ) . The duration of runs is modulated by the sensory input: runs up the gradient are elongated while runs away from it are shortened ( Gomez-Marin et al . , 2011; Gershow et al . , 2012 ) . Although published results hint at how larval chemotaxis may be achieved ( Gomez-Marin and Louis , 2012 , 2014 ) , a quantitative model of the underlying sensorimotor integration is still missing . Here , we focus on the primary task of the orientation algorithm common to bacteria , C . elegans , and Drosophila: the control of run duration ( Bargmann , 2006a; Lockery , 2011; Gomez-Marin and Louis , 2012 ) . It is known that turns are preceded by stereotyped decreases in odor concentration ( Gomez-Marin et al . , 2011; Gomez-Marin and Louis , 2012 ) , but the key question of how concentration differences are computed is unresolved . In both insects and vertebrates , odor concentrations are represented by time-varying patterns of activity distributed across the olfactory sensory neuron ( OSN ) population ( Wilson and Mainen , 2006; Wilson , 2013; Masse et al . , 2009; Mainland et al . , 2014; Uchida et al . , 2014 ) . Nonetheless , animals with an olfactory system genetically reduced to a single functional OSN are still capable of robust chemotaxis ( Fishilevich et al . , 2005; Louis et al . , 2008 ) , implying that the mechanisms of odor concentration detection can be understood at the level of single OSNs . Here , we rely on this simplification to develop a novel larval preparation in which the neural computations underlying odor gradient ascent can be understood in unprecedented detail . We used optogenetics in larvae with a single type of functional OSNs to substitute turbulent olfactory signals with well-controlled light stimulations ( Suh et al . , 2007; Bellmann et al . , 2010; Smear et al . , 2011; Gaudry et al . , 2013 ) . This allowed us to characterize the modulatory effects of OSN firing patterns on the probability of switching from a run to a turn . Toward this goal , we developed a novel tracker to create virtual olfactory realities ( Kocabas et al . , 2012 ) in which optogenetic stimulations of genetically targeted OSNs are defined based on the behavioral history of the larva . We used this technology to derive a phenomenological model of the OSN transfer function . The model was validated on free behavior in sensory landscapes designed to produce predictable sensorimotor responses , and ultimately , it was found to be applicable to real odor gradients . We found that for positive gradients , the OSN operates as a slope detector: its activity increases with the stimulus derivative , which suppresses the probability of turning . For strongly negative gradients , the OSN acts like an OFF detector: the inhibition of the neural activity facilitates turning in a nearly deterministic manner . Altogether , our results advance our understanding of how peripheral odor encoding guides action selection during chemotaxis . Odors are generally attractive to Drosophila larvae ( Cobb , 1999 ) . Exposure to an odor produces gradient ascent even in larvae with a genetically manipulated olfactory system reduced to a single OSN ( Fishilevich et al . , 2005; Louis et al . , 2008 ) . We examined the behavior of larvae with a single functional OSN expressing Or42a , an odorant receptor with a well-characterized tuning profile that includes the odorant isoamyl acetate ( IAA ) ( Fishilevich et al . , 2005; Kreher et al . , 2008; Asahina et al . , 2009 ) . Behavior was studied in a closed environment with a single source of IAA suspended from the ‘ceiling’ of the arena ( Figure 1 and ‘Materials and methods’ ) . For large odor droplets , diffusion from the source creates a radially symmetric gradient that can be approximated by a stationary Gaussian distribution ( Louis et al . , 2008 ) . For smaller odor droplets such as those used in the present study , the temporal evolution of the odor gradient cannot be neglected . We therefore combined infrared spectroscopy ( IR ) and a partial differential equation ( PDE ) model to experimentally reconstruct the two-dimensional geometry of the odor gradient over time ( see Figure 1B and ‘Materials and methods’ ) . The simulated odor gradient served as a template to reconstruct the average stimulus time course experienced by the larva during real trajectories ( Figure 1C ) . 10 . 7554/eLife . 06694 . 003Figure 1 . Sensory experience corresponding to unconstrained chemotactic behavior . ( A ) Illustrative trajectory of a larva freely moving in an attractive odor gradient ( isoamyl acetate , source concentration: 0 . 25 M ) . Position of the midpoint shown in black; position of the head shown in magenta . Two run segments , R1 and R2 , are underlined in green . Turns are depicted as white disks . White arrows indicate the direction of motion . ( B ) Reconstruction of the odor gradient based on numerical simulations of the odor diffusion process modeled by a partial differential equation ( PDE ) system with realistic boundary conditions ( ‘Materials and methods’ ) . The gradient shown in panel A corresponds to a snapshot obtained 60 s after onset of the odor diffusion . ( C ) Time course of the odor concentration experienced at the head ( magenta ) and midpoint ( black ) of the larva during the trajectory shown in panel A . The sensory experiences are reconstructed based on mapping the positions of interest on the dynamic odor gradient computed upon integration of the PDE system for the entire duration of the trajectory . The green box outlines the sensory experience corresponding to run segments R1 and R2 . Small disks on the abscissa indicate the turns comprised in this behavioral sequence . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 003 Figure 1A presents a trajectory consisting of approximately straight segments ( ‘runs’ ) punctuated by large changes in orientation ( ‘turns’ ) . Where to turn to is determined through lateral exploratory head movements , ‘head casts’ , during which the larva scans the local odor gradient ( Gomez-Marin et al . , 2011; Gershow et al . , 2012 ) . On average , larvae terminate their runs when motion is directed down the gradient where the odor concentration is decreasing ( Gomez-Marin et al . , 2011; Gershow et al . , 2012 ) . In contrast , turns are suppressed when the direction of motion is along the gradient and the odor concentration is increasing . We sought to define the neural computations underlying this behavior by characterizing the neural activity of the Or42a-expressing OSN in response to changes in odor concentration experienced during chemotactic behavior ( Figure 1C ) . To probe the input–output transfer function of the Or42a OSN , we devised an extracellular recording technique based on the suction of the antennal nerve into a glass pipette downstream from the dorsal organ ( DO ) ganglion ( Figure 2A and ‘Materials and methods’ ) . With the use of an optogenetic spike-sorting strategy we identified the spikes originating from the Or42a OSN expressing channelrhodopsin ( ChR2 ) ( denoted as ‘Or42a>ChR2 OSN’ ) ( Figure 2B , C and ‘Material and methods’ ) . We devised a customized olfactometer to produce odor stimuli with controlled temporal profiles ( Figure 2A ) with which we examined the response of the Or42a>ChR2 OSN to a concentration replay defined by the stimulus time course associated with the trajectory shown in Figure 1A . Recordings from this 3-min stimulation led to consistent patterns of neural activity in different preparations ( Figure 2D ) . Although the OSN activity appeared to follow the envelope of the stimulus time course , a closer examination revealed greater complexity in the neural response . The OSN firing rate displayed a clear amplification of changes in stimulus intensity , as illustrated by the activity associated with the replay of two consecutive runs , R1 and R2 ( Figure 2F ) . Run R2 brought the larva close to and then beyond the peak of the gradient . When stimulated by the corresponding time course of the odor concentration , OSN activity peaked several seconds before the stimulus intensity ( Figure 2F , arrows ) . Minute fluctuations in odor concentration were strongly amplified in the OSN spiking dynamics ( bursts marked by a sharp # symbol in Figure 2F ) . 10 . 7554/eLife . 06694 . 004Figure 2 . Response of a single larval olfactory sensory neuron ( OSN ) to naturalistic odor stimulation . ( A ) Illustration of preparation for suction electrode recordings of single functional OSNs expressing channelrhodopsin ( ChR2 ) . The preparation is bathed in saline to prevent the dehydration of the dorsal organ ganglion to which the recording electrode is attached . Controlled odor stimulations are achieved in liquid phase with a customized mass flow controller system . ( B ) Recording from the dorsal organ stimulated by a series of 10-ms light pulses . ( i–ii ) The voltage trace shows spikes with different amplitudes . ( iii ) Close-up view of the voltage trace corresponding to three consecutive light pulses . Action potentials with a stereotyped waveform are observed at a short-latency after the onset of the light pulse ( spikes denoted by a blue star * ) . These spikes are associated with the activity of the Or42a OSN expressing ChR2 . Each light pulse yielded an average of 1 . 8 light-evoked spikes . ( C ) Superimposition of light- and odor-evoked spike waveforms observed for the same OSN . The results of two different recordings are shown in ( i ) and ( ii ) . Spike waveforms associated with the light stimulation ( blue spikes ) are superimposed on spike waveforms collected during an episode of odor stimulation ( magenta spikes ) . The high similarity between the light- and odor-evoked spikes serves as a basis to spike sort the recordings arising from the dorsal organ ganglion ( ‘Materials and methods’ ) . ( D ) Dynamic reconstruction of the concentration time course corresponding to the trajectory of the head position depicted in Figure 1 . ( E ) Results of 9 suction electrode recordings for the Or42a>ChR2 OSN stimulated by the concentration shown in panel D ( 5 preparations ) . ( Top ) Raster plot . ( Bottom ) PSTH of the OSN response to the concentration time course shown in panel D with shade representing the standard deviation . ( F ) Close-up view of the sensory experience ( top ) and OSN response ( bottom ) corresponding to the illustrative runs R1 and R2 shown in panel D ( green box ) . Since the maximum firing rate is attained earlier than the stimulus intensity reaches its maximum , the input–output relationship driving the dynamics of the OSN activity is more complex than a proportional detector . Short increases in odor concentration lead to transient bursts in spiking activity ( bursts indicated by sharp # signs ) . Small disks on the abscissa denote turns in the original trajectory presented in Figure 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 004 To tease apart the sensory features encoded by the Or42a>ChR2 OSN , we examined the OSN response induced by a set of controlled odor ramps with a temporal profile analogous to the run sequence described in Figure 2F . As a first approximation , we used linear ramps with symmetrical 8-s rising and falling phases . During the rising phase of the ramp , the neural activity increased in proportion with the derivative of the odor concentration ( Figure 3A ) . During the falling phase of the ramp , the firing rate appeared to be driven by the stimulus intensity rather than the stimulus derivative ( Figure 3B ) , suggesting that the response properties of the OSN differed for positive and negative gradients . 10 . 7554/eLife . 06694 . 005Figure 3 . Characterization of the dynamical features extracted by the Or42a-expressing olfactory sensory neuron . ( A ) Response to three linear ramps with variable slopes during the rising phase and equal slope during the falling phase . The ‘low’ ( ‘high’ ) ramps have a positive slope that is twice slower ( faster ) than the medium ramp . PSTH computed on a pool of minimum 24 recordings obtained from minimum 8 preparations . During the rising phase of the ramp , the activity of the OSN reaches a peak value that scales with the slope of the ramp . ( B ) Response to three linear ramps with variable slopes during the falling phase and equal slope during the rising phase . The ‘slow’ ( ‘fast’ ) ramp has a negative slope that is twice slower ( faster ) than the medium ramp . During the rising phase of the ramp , the plateau reached by the OSN activity grows with the slope of the ramp . During the falling phase of the ramp , the activity of the OSN is more directly driven by the stimulus intensity . For the three ramps , the OSN activity becomes inhibited when the ramp terminates . PSTH computed on a pool of minimum 24 recordings obtained from minimum 8 preparations . ( B ) Response to nonlinear ramps featuring a symmetrical 8 s-rise and 8 s–fall profiles . From left to right , the ramps tested have the following characteristics: linear ( ∝ t ) , exponential ( ∝ e−8 ( et −1 ) ) , and sigmoid ( ∝ t3/ ( t3+43 ) ) with the time given in s . PSTH computed on a pool of minimum 16 recordings obtained from minimum 9 preparations . For panels C , the odor concentration ( magenta ) is computed from the flow ratio measured experimentally based on the flow controller outputs ( ‘Materials and methods’ ) . The time derivative of the concentration time course is represented according to the y-scale on the right of the graph ( gray ) . The derivative was computed after mild smoothening of the stimulus input . Asterisks denote inhibitory phases of the OSN response where the activity decreases below its basal level ( horizontal dashed line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 00510 . 7554/eLife . 06694 . 006Figure 3—figure supplement 1 . Dose-response of the Or42a-expressing olfactory sensory neuron stimulated by prolonged odor pulses . ( A ) Response to odor step stimulations of low and high concentrations: 52 μM and 522 μM . ( Bottom ) Raster plot of the OSN activity in response to low- and high-concentration odor steps ( light and dark gray , respectively ) . 15 recordings conducted on 5 different preparations . ( B ) PSTH of the OSN activity in response to the low- and high-concentration odor steps ( light and dark gray , respectively ) shown in panel A . Shades represent standard deviations . The response dynamics of the Or42a OSN to static odor pulses is similar to that observed in adult-fly OSNs ( de Bruyne et al . , 1999 , Nagel and Wilson , 2011 , Martelli et al . , 2013 ) . ( C ) Dose-response curve observed for the activity of the OSN after it has relaxed to a steady state value ( mean firing rate computed during the time interval 20–24 s ) . Error bars represent standard deviations . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 006 To assess the slope sensitivity of the OSN , we compared the neural activity elicited by nonlinear ramps in which the first derivative of the stimulus changed over time ( Figure 3C ) . In an exponential ramp , the stimulus derivative increased throughout the rising phase of the ramp . This acceleration correlated with a continuous increase in spiking activity . To further test the hypothesis that the OSN encodes features related to the slope of the stimulus , we examined a sigmoid ramp ( Figure 3C , right panel ) for which the first derivative of the stimulus reached its maximum ( gray arrow ) prior to the stimulus intensity ( magenta arrow ) . Consistent with the slope-sensitivity hypothesis , the OSN spiking activity peaked with the first derivative and not the absolute intensity of the stimulus . During the falling phase of the ramps , the OSN firing rate behaved in a way that could not be predicted from the slope sensitivity observed during the rising phase . At the end of the falling phase , OSN activity decreased below baseline ( star signs * in Figure 3C ) , suggesting offset inhibition similar to that observed in OSNs of adult flies ( Hallem et al . , 2004; Nagel and Wilson , 2011 ) . Our findings on the features encoded by the OSN were corroborated by responses elicited by other odor ramps ( Figure 4—figure supplement 1 ) . We attempted to model the input–output relationship of the OSN by following a linear system-identification approach ( Chichilnisky , 2001 ) . To this end , we applied an M-sequence ( pseudorandom binary sequence with nearly flat frequency spectrum ) and reverse-correlation ( Geffen et al . , 2009 ) , but discovered that the resulting linear filter was insufficient to account for the firing patterns elicited by naturalistic stimuli ( ‘Materials and methods’ ) . We thus turned to dynamical systems theory to capture the nonlinear characteristics of the OSN response . We developed a biophysical model that accounts for the slope-sensitivity of the OSN during stimulus upslopes , proportionality response and offset inhibition during downslopes ( Figure 4 ) . 10 . 7554/eLife . 06694 . 007Figure 4 . Quantitative model for signal processing achieved by a single olfactory sensory neuron . ( A ) Hypothetical physiological processes underlying the olfactory transduction pathway and spike generation . The integral feedback ( IFB ) motif is built on the assumption that inhibitory feedback modulates the activity of the odorant receptor , as was proposed in the adult fly ( Nagel and Wilson , 2011 ) . This motif appears to be essential to olfactory transduction in vertebrates ( De Palo et al . , 2013 ) . The incoherent feed-forward ( IFF motif ) relies on the hypothetical existence of a delayed inhibitory effect , as was proposed for the transduction cascade of C . elegans ( Kato et al . , 2014 ) . ( B ) Biophysical model of the olfactory transduction pathway . ( Bi ) Circuit elements combining the IFF and IFB motifs described in panel A . Variable x represents the stimulus intensity , u , the activity or concentration of the intermediate variable and y , the firing rate of the OSN . Pathway ( 3 ) is specific to the IFB motif ( light blue ) . ( Bii ) ODE system providing a phenomenological description of the reaction scheme outlined in panel A for the combination of the IFF and IFB regulatory motifs . The three pathways regulating the activity of u are outlined by numbers ( 1 ) – ( 3 ) . Reaction ( 1 ) corresponds to a ‘production’ of u through the IFF branch; ( 2 ) corresponds to a first-order ‘decay’ of u; ( 3 ) corresponds to a ‘production’ of u through the IFB branch . ( C ) Simulated activity of u ( green , middle ) and firing rate y ( blue , bottom ) in response to an 8-s linear odor ramp ( magenta , top ) . Numerical simulations were achieved by integrating the ODE system described in panel Bii with the parameter values listed in Table 1 . ( D ) Decomposition of the predicted activity of individual pathways contributing to the regulation of u for the linear odor ramp displayed in panel C . Activity computed from the terms ( 1 ) – ( 3 ) outlined in panel B for the feed-forward activation by the stimulus ( IFF , 1 ) , first-order decay ( 2 ) and coupling of the firing rate with the intermediate variable through the negative feedback ( IFB , 3 ) . Notably , the contribution of the reaction specific to the IFB motif ( 3 ) is dominated by the reaction specific to the IFF motif ( 1 ) . ( E ) Fit of the solution of the ODE model for three linear stimulation ramps introduced in Figure 3A , B and produced with odor ( middle ) and light ( bottom ) . ( Top ) Stimulus intensity given as odor concentration ( μM ) . The time derivative of the concentration profile ( gray lines ) is given according to the y-axis shown on the right side of the graph . The derivative was computed after mild smoothening of the stimulus time course . The same ( idealized ) profile was used for the light stimulation with an intensity ranging between 15 W/m2 and 207 W/m2 . ( Middle ) Comparison of the outcome of the model featuring a pure IFF motif ( green ) and a combination of the IFF and IFB motifs ( blue ) . The parameters of both models were obtained independently through a Simplex optimization procedure ( ‘Materials and methods’ ) . For the pure IFF model , parameter α3 was artificially set to 0 . ( Bottom ) Comparison of the outcome of the experimental PSTH and the model's predictions based on a pure IFF motif ( green ) for light stimulation . Parameter optimization shows that the IFB motif does not contribute to the light-evoked OSN dynamics . ( F ) Fit of the solution of the ODE model for three nonlinear stimulation ramps generated with odor and light . ( Middle ) Results of the model compared to the odor-evoked OSN activity . Close-up view of the 8-s linear ramp highlighting the differences between the behavior of the pure IFF ( green ) and combined IFF+IFB ( blue ) circuit motifs for odor stimulation . ( Bottom ) Comparison of the outcome of the experimental PSTH and model based on a pure IFF motif ( green ) for light stimulation . For all conditions shown in the figure , PSTHs were computed on a pool of a minimum of 10 recordings obtained from a minimum of 10 preparations . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 00710 . 7554/eLife . 06694 . 008Figure 4—figure supplement 1 . Response dynamics of the Or42a>ChR2 OSN to linear and nonlinear stimulation ramps induced with odor and light , and fits of the stimulus-to-neural ( ODE ) models . ( A ) Set of 4-s linear ramps with different rising and falling slopes . ( Ai ) Five ramps with a rising phase that lasts 4 s . Stimulus intensity given as odor concentration ( μM , magenta ) . Upon mild smoothening of the stimulus time course trajectories with a Savitzky-Golay filter , the time derivative ( gray ) of the concentration profile is represented according to the y-axis on the right of the graph . The same ( idealized ) profile was used for the light stimulation with an intensity ranging between 15 W/m2 and 207 W/m2 . ( Aii ) Fit of the solution of the ODE model for the linear stimulation ramps generated with odor . Comparison of the experimental PSTH ( black ) with the outcome of the model featuring a pure IFF motif ( green ) , and a combination of the IFF and IFB motifs ( blue ) . ( Aiii ) Modeling of the OSN activity elicited by the linear light ramps . Comparison of the experimental PSTH with the outcome of the model based on a pure IFF motif ( green ) . ( B ) Set of 8-s linear and nonlinear ramps . ( Bi ) Stimulus intensity given as odor concentration ( μM , magenta ) . Same conditions as panel A . ( Bii ) Modeling of the OSN activity elicited by odor ramps . Comparison of the experimental PSTH ( black ) with the outcome of the model featuring a pure IFF motif ( green ) , and a combination of the IFF and IFB motifs ( blue ) . ( Biii ) Same as Bi for the light ramps . Comparison of the experimental PSTH with the outcome of the model based on a pure IFF motif ( green ) . For all conditions shown in this figure , the PSTHs were computed on a pool of a minimum of 10 recordings obtained from a minimum of 10 preparations . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 00810 . 7554/eLife . 06694 . 009Figure 4—figure supplement 2 . Validation of the stimulus-to-neural ( ODE ) model for the olfactory transduction cascade upon training on partial datasets . ( A ) Training of the ODE model described in Figure 4B on a set of 5 linear ramps with rising phase lasting 4 s . ( B ) The ODE model trained on the 4-s linear ramps alone ( panel A ) is validated on 8-s stimulus ramps: a linear ramp ( left ) , an exponential ramp ( middle ) , and a sigmoid ramp ( right ) . The model is trained independently for odor ( middle row ) and light stimulations ( bottom row ) . The predictions of the model are shown as a dashed line . The PSTH experimentally observed is shown in black ( shades denote standard deviation ) . The output of the model trained on the entire set of ramps ( plain line , parameter set listed in Table 1 ) shows nearly no improvement compared to the model trained on the subset of linear ramps . For both the odor-evoked and light-evoked OSN dynamics , the correlation coefficients ( ρ ) and coefficients of variation of the RMSE ( Table 1 ) corresponding to the model trained on the linear ramps and that trained on all ramps differ by less than 1% and 5% , respectively . ( C ) Validation of the ODE model obtained in panel A on naturalistic stimulation with odor ( middle row ) and light ( bottom row ) . Stimulus time course represented in magenta . The experimental PSTH is shown in black ( shades denote standard deviation ) . The predictions of the model trained on the linear ramps alone ( dashed line ) show nearly no difference with the output of the model trained on the entire set of ramps ( plain line ) . The naturalistic odor stimulus corresponds to the trajectory presented in Figure 2D . The naturalistic light stimulus corresponds to the trajectory presented in Figure 6B . For both the light and the odor-evoked spiking activity , the correlation coefficients ( ρ ) and coefficients of variation of the root-mean-square error ( RMSE ) corresponding to the model trained on the linear ramps alone and that trained on all ramps differ by less than 1% and 7% , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 009 Negative feedback is known to play an important regulatory function in sensory transduction . Integral feedback control underlies perfect adaption in bacterial chemotaxis ( Yi et al . , 2000 ) . In vertebrates , the olfactory transduction cascade involves a metabotropic pathway downstream from a G-protein coupled receptor ( GPCR ) that features negative regulatory feedback ( Kaupp , 2010; Pifferi et al . , 2010 ) . As with phototransduction , adaptive features of the olfactory transduction cascade in vertebrate can be accounted for by integral feedback ( De Palo et al . , 2012; De Palo et al . , 2013 ) . Even though invertebrate olfaction does not rely on GPCR signaling ( Kaupp , 2010 ) , the existence of negative feedback on the odorant receptor has been postulated for olfactory transduction in adult-fly OSNs ( Nagel and Wilson , 2011 ) . This conclusion was drawn from a biophysical model that combined a linear filter accounting for the OSN spiking dynamics with a kinetic formalism to describe ligand–receptor interactions . Research in the moth has revealed a different regulatory mechanism that constitutes an ‘incoherent feed-forward’ loop ( Alon , 2007 ) in which the activity of the odorant receptor has a dual effect on the OSN spike rate ( Gu et al . , 2009 ) : ( 1 ) on a short timescale , the inflow of cations increases the firing rate; ( 2 ) on a longer timescale increasing concentration of intracellular calcium ions inhibits the OSN firing rate through a pathway that involves the binding of calcium to calmodulin . Combining the previous ideas , we hypothesized that the OSN spiking activity is regulated by a negative feedback loop ( or integral feedback , IFB ) coupled with an incoherent feed-forward loop ( IFF ) ( Figure 4A and ‘Materials and methods’ ) . In what follows , this composite model will be denoted as IFB+IFF ( Figure 4Bi ) . Using a mass-action-kinetics formalism originally developed for genetic networks ( Ackers et al . , 1982; Bintu et al . , 2005 ) , each of the two regulatory motifs was described by a system of two ordinary differential equations ( ODEs ) with three variables ( Figure 4Bii ) : x , the stimulus strength ( input: odor concentration or light intensity ) , y the instantaneous firing rate of the OSN ( output ) , and u , a phenomenological variable that might represent the intracellular concentration of calcium . The free parameters of the model were determined through a simplex algorithm which optimized the fit between the experimental spiking activity of the Or42a>ChR2 OSN and that produced by the ODE model . The optimization was achieved on a set of 10 linear and nonlinear ramps listed in Figure 4—figure supplement 1 together with the naturalistic stimulus presented in Figure 2D . The parameter set derived from the 10 odor ramps and naturalistic stimulus ( Table 1 ) led to a remarkably good fit between the output of the ODE model and the experimentally measured spiking activity ( experimental peristimulus time histogram PSTH—black line , results of the IFF+IFB model—blue line; Figure 4C , E , F , Figure 4—figure supplement 1 ) . Throughout the study , this parameter set was used to reproduce and predict the OSN spiking activity elicited by olfactory stimuli . To rule out over-fitting , we trained the IFF+IFB model on a partial dataset containing the linear ramps alone and validated its response against other stimuli not present in the training dataset ( Figure 4—figure supplement 2 ) . 10 . 7554/eLife . 06694 . 010Table 1 . Parameters of ODE model derived from the Simplex optimization procedure ( ‘Materials and methods’ ) applied on the OSN spiking activity elicited by light stimulation ( pure IFF model—left column ) and odor stimulation ( pure IFF model—middle column; combined IFF+IFB model—right column ) DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 010Light: IFF motifOdor: pure IFF motifOdor: IFF+IFB motifsα10 . 1 ( W m−2 ) −1 s−10 . 1 μM−1 s−10 . 13 μM−1 s−1α20 . 88 s−10 . 6 s−10 . 26 s−1α310−6 Hz−1 s−10* Hz−1 s−11 . 1 Hz−1 s−1β11731 . 41 Hz s−11002 . 25 μM s−12903 . 36 μM s−1β21 . 27 W m−28 . 63 μM0 . 01 μMβ32 . 48 W m−22 . 39 μM2 . 65 μMβ41214 . 08 Hz s−1624 . 69 μM s−1795 . 62 μM s−1β513 . 03 s−16 . 44 s−123 . 79 s−1θ0 . 3 Hz1 . 01 Hz1 . 88 Hzn222Parameters were obtained upon training of the model on 10 stereotyped stimulus ramps ( see Figure 4—figure supplement 1 ) together with the naturalistic stimulation patterns shown in Figure 2D ( odor ) or Figure 6B ( light ) . For light stimulation , the parameter of the IFB pathway ( α3 ) was negligible and considered equal to 0 in the rest of the study . For odor stimulation , parameter α3 was artificially set to 0 in the case of the pure IFF motif . Note that the units of the intermediate variable u are undefined . We empirically found that the goodness of fit improved when the value of the offset β4 undergoes a small correction over time . In all numerical simulations of this study , we used β4 ( t ) = ( 1 . 023 t4/ ( t4 + 304 ) ) × β4 . The Hill coefficient n was set equal to 2 . In this table , all concentrations are given for odor stimulation in liquid phase . As described in the ‘Materials and methods’ section , the concentration equivalence in gaseous phase can be approximated by multiplying the liquid phase concentration by a factor ρliquid → gas = 26 . 73 . The parameters listed in this table are used in all numerical simulations of the study , except the validation controls described in Figure 4—figure supplement 2 . *parameter set artificially to 0 . Next , we examined the individual contribution of the IFF and IFB pathways to the response dynamics of the OSN . In Figure 4D , the activity of each pathway was separately computed in response to stimulation by a linear odor ramp . The contribution of the IFB motif to the dynamics of variable u is approximately 30% that of the IFF ( cyan vs magenta curves , Figure 4D ) . This led us to conclude that the IFF pathway dominates the control of OSN spiking activity . The IFB pathway has nonetheless a non-negligible impact on the dynamics . Using the parameter optimization procedure , the pure IFF model was trained on the full set of odor ramps ( Table 1 , middle column ) . At a qualitative level , both the pure IFF and the composite IFF+IFB models reproduced the OSN spiking dynamics ( Figure 4 and Figure 4—figure supplement 1 ) , but a quantification of the goodness of fit established the superiority of the IFF+IFB model ( Table 2 and inset of Figure 4F ) . In addition , none of the other standard 3-element circuit motifs we tested produced a reasonable fit of the OSN spiking activity ( data not shown ) , arguing that the composite IFF+IFB model comprises essential regulatory features of the olfactory transduction cascade . 10 . 7554/eLife . 06694 . 011Table 2 . Quantification of the goodness of fit between the stimulus-to-neural ( ODE ) models and the experimental firing dynamics of the OSN stimulated by the linear and nonlinear rampsDOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 011Odor , IFFOdor , IFF+IFBLight , IFFA Correlation coefficient ( ρ ) Linear ( 4 s ) , low0 . 9820 . 9850 . 954 Linear ( 4 s ) , med0 . 9940 . 9940 . 980 Linear ( 4 s ) , high0 . 9860 . 9950 . 983 Linear ( 4 s ) , slow0 . 9930 . 9940 . 980 Linear ( 4 s ) , fast0 . 9830 . 9790 . 970 Linear ( 8 s ) 0 . 9830 . 9940 . 982 Quadratic0 . 9930 . 9930 . 988 Exponential0 . 9900 . 9840 . 965 Sigmoid0 . 9850 . 9940 . 990 Asymptotic0 . 9900 . 9940 . 972B CV ( RMSE ) Linear ( 4 s ) , low0 . 1430 . 1210 . 225 Linear ( 4 s ) , med0 . 1120 . 0970 . 174 Linear ( 4 s ) , high0 . 1590 . 0930 . 184 Linear ( 4 s ) , slow0 . 1110 . 0910 . 167 Linear ( 4 s ) , fast0 . 1730 . 1790 . 229 Linear ( 8 s ) 0 . 1850 . 1060 . 153 Quadratic0 . 1310 . 1110 . 150 Exponential0 . 1550 . 2260 . 237 Sigmoid0 . 1750 . 1020 . 131 Asymptotic0 . 1240 . 0800 . 193 ( A ) Pearson's correlation coefficient ( ρ ) computed for stimulus ramps listed in Figure 4—figure supplement 1 . ( B ) Coefficient of variation ( CV ) of the root-mean-square error ( RMSE ) . The response properties of the OSN were then studied for a family of stimulus ramps induced by light instead of odor . The temporal profiles of the light ramps were identical to the odor ramps; the intensity range was fixed to coarsely match the low firing rate of the OSN activity observed for the odor stimulations . In this regime , the temporal pattern of the OSN activity elicited by the light ramps was comparable to that elicited by the odor ramps ( experimental PSTH—black lines , Figure 4E , F and Figure 4—figure supplement 1 ) . This close similarity in the input–output relationships permitted us to substitute the odor stimulus with light . Using the full set of linear and nonlinear light ramps together with a naturalistic pattern of light stimulation ( Figure 6B ) , optimization of the parameters of the ODE system showed that the IFB pathway does not contribute to the light-evoked response dynamics of the OSN , suggesting that the integral feedback motif is specific to the odor-evoked activity ( Table 1 ) . The results of the pure IFF model are in good agreement with the experimental observations ( Figure 4E , F and Figure 4—figure supplement 1 ) . Notably , the goodness of fit of the pure IFF motif , when applied to both light and odor stimulations , was comparable ( Table 2 ) . In conclusion , the nonlinear-dynamical response properties of the OSN stimulated by odor and light ramps can be well approximated by the IFF motif , even though the IFB motif brings a non-negligible contribution to the modeling of odor-evoked response dynamics . To clarify the sensory computation achieved by the Or42a OSN , we sought to derive an analytical solution of the ODE system . We restricted this analysis to the pure IFF motif , which provides a good approximation of the dynamics of the composite IFF+IFB motif . The general solution of the IFF motif required solving the ODE system shown in Figure 4Bii . Since the OSN spiking activity evolves on a different timescale than the other two variables , the solution of the ODEs could be simplified through a quasi-steady-state approximation ( QSSA , see ‘Materials and methods’ ) . The mathematical expression of the QSSA solution reveals that the OSN spiking activity ( y ) is determined by a hyperbolic ratio function of the stimulus intensity x: ( 1 ) yQSSA=δ1xx+δ2−S ( x , t ) −δ4 , where δ1 , δ2 , and δ4 are constants ( ‘Materials and methods’ ) . The denominator of this hyperbolic relationship contains a scaling term S ( x , t ) that normalizes the spiking activity by the short-term history of changes in the stimulus intensity dx/dt:S ( x , t ) ∝∫0te−α2 ( t−t′ ) dxdt′ dt′ . The integration-differentiation scaling function S ( x , t ) plays a role similar to the ‘input gain control’ resulting from lateral inhibition of the local interneuron on the projection neurons in the adult antennal lobe ( Olsen et al . , 2010 ) with the notable difference that the rescaling takes place within the primary OSN and that it is driven by the temporal integration of changes in the stimulus intensity . In analogy to the divisive normalization reported in the visual system ( Carandini and Heeger , 2012 ) , we termed the rescaling operation described in Equation 1 as ‘transient normalization’ . This operation appears related to the adaptive rescaling of the spike dynamics observed in adult-fly OSNs ( Kim et al . , 2011; Nagel and Wilson , 2011 ) . By examining the analytical solution of the IFF motif under the quasi-steady-state approximation ( Equation 1 ) , we discovered that the most salient features encoded in the activity pattern of the Or42a>ChR2 OSN are: ( 1 ) rapid increases in firing rate triggered by abrupt positive changes in the stimulus intensity ( accelerations ) ; ( 2 ) a relaxation of the firing rate toward stationary activity when the first derivative of the stimulus is null or constant ( no acceleration or deceleration ) ; ( 3 ) decreases in firing rate in response to stimulus decelerations . In addition , we experimentally observed that ( 4 ) the spiking activity of the neuron is strongly inhibited upon abrupt return to the stimulus baseline . We asked whether these features bore any relevance to the control of run-to-turn transitions during odor gradient ascent . We hypothesized that sustained spiking activity of the OSN would suppress turning while inhibition of the OSN would facilitate turning . To test this hypothesis , we built a tracker to monitor the position and behavioral state of a single larva in real-time at a rate of 30 Hz ( Figure 5A and detailed description in ‘Materials and methods’ ) . Equipped with blue LEDs , the tracker was designed to evoke controlled patterns of spiking activity in the Or42a>ChR2 OSN by means of optogenetics . To avoid innate photophobic behavior ( Sawin-McCormack et al . , 1995; Kane et al . , 2013 ) , experiments were conducted on blind larvae ( ‘Materials and methods’ ) . 10 . 7554/eLife . 06694 . 013Figure 5 . Modulation of run-to-turn transitions by light-evoked activity in the Or42a-expressing OSN . ( A ) Illustration of the closed-loop tracker used to synthesize virtual olfactory realities with light stimulation coupled to optogenetics . Close-up view of the larva illuminated by a red light pad fixed to a moving stage below the agarose slab . The camera and LEDs are mounted on a second moving stage whose position is updated synchronously with the bottom stage to remain locked on the position of the larva . The platform on which the larva behaves is fixed . ( B ) Presentation of the run-based light stimulation paradigm where runs are randomly assigned to constant stimulation ( control ) or to a test ramp with an exponential profile similar to that introduced in Figure 3C . ( Right ) Midpoint position of the larva during a trajectory with the light intensity color-coded in accordance with the color bar on the left . Illustrative runs denoted as R1-4 are interspaced by turns T1-3 denoted by arrows . ( C ) Turn probability estimated from a set of runs associated with constant stimulation ( light gray ) or stimulation by an exponential light ramp ( black ) . The turn probabilities are reported as the fraction of turns occurring during a 1-s window centered on the time point of interest ( ‘Materials and methods’ ) . Error bars are estimated from resampled sets of runs ( shaded areas denote standard deviation , see ‘Materials and methods’ ) . The dashed line depicts the mean turning rate computed for constant light stimulation . The turning rate is in first approximation independent of the run duration . Small disks on the x-axis indicate time points after which fewer than 10% of the total number of runs are left for the constant stimulation ( light gray ) and exponential ramp ( dark gray ) . Beyond these time points , the estimates of the turn probability should be considered as unreliable . ( D ) Generalized linear model ( GLM ) for the modulation of turn probability as a function of the sensory experience ( integrated stimulus-to-behavior model ) . The turn probability is predicted from a linear combination of the predicted neural activity ( γ1 y ( t ) ) and a constant term ( γ0 ) . This linear combination is then fed into a logit transformation to convert the domain of definition of the neural activity into a probability . The two parameters of the model , γ0 and γ1 , are determined from a linear regression on the experimental profiles of turn probability . The OSN activity driven by light , y , is predicted from the pure IFF ( ODE ) model described in Figure 4B . As a control , we consider the same model where the input is the stimulus intensity . The parameters of this control model are denoted as γ′0 and γ′1 . Upon training of the test and control models on the full set of linear and nonlinear ramps ( Figure 5—figure supplement 1 ) , we derived the parameter values reported in Table 3 . ( E ) Predictions of the integrated stimulus-to-behavior GLM for linear ramps of different rising and falling slopes . ( Top ) For the individual ramps , the time course of light intensity is shown in magenta . The time derivative of the light ramp ( gray line ) is computed after mild smoothening of the stimulus input . ( Bottom ) Behavioral predictions based on the neural activity predicted by the IFF ( plain blue line ) and the control model that is purely based on the stimulus ( dashed magenta line ) . The integrated stimulus-to-behavior model clearly outperforms the predictions of the control model , which highlights the importance of the signal processing achieved by the OSN . ( F ) Predictions of the integrated stimulus-to-behavior model for 8-s light ramps . For all conditions tested , the experimental turn probability was estimated on a sample of 490–970 runs . Quantification of the goodness of fit is reported in Table 4 for the test and control models . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 01310 . 7554/eLife . 06694 . 014Figure 5—figure supplement 1 . Fit of the integrated stimulus-to-behavior generalized linear model ( GLM ) for the linear and nonlinear light ramps . ( A ) Set of 4-s linear ramps . ( Ai ) Light intensity profile ( W/m2 ) . ( Aii ) Application of pure IFF motif to model the OSN activity elicited by the light ramps ( green line ) . ( Aiii ) Application of the integrated stimulus-to-behavior GLM to predict behavior in response to stereotyped light ramps . The test model ( blue line ) is based on the neural activity modeled by the IFF motif . The turn probability estimated experimentally is shown in black ( shades denote standard deviation as described in ‘Materials and methods’ ) . As indicated in the textbox on the right , the control model is based on the stimulus without any processing of the OSN ( dashed magenta line ) . The black dashed line in the background represents the average turn probability observed upon stimulation at constant light intensity . The integrated stimulus-to-behavior model clearly outperforms the predictions of the control model . ( B ) Set of 8-s linear and nonlinear ramps . ( Bi–Biii ) Same as panels Ai–Aiii for the behavior elicited by 8-s light ramps . For all test conditions , the experimental turn probability was estimated on a sample of 490–970 runs . The outputs of the test and control models are obtained based on the parameter sets listed in Table 1 ( ODE model ) and Table 3 ( GLM ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 01410 . 7554/eLife . 06694 . 015Figure 5—figure supplement 2 . Validation of the integrated stimulus-to-behavior generalized linear model ( GLM ) upon training on a partial set of light ramps . ( A ) Training of the GLM described in Figure 5D restricted to the set of 4-s linear light ramps . ( B ) Validation of the GLM model trained on the set of 4-s linear ramps ( panel A ) on 8-s ramps ( dashed green line ) . Comparison with the output of the model trained on the entire set of ramps ( plain blue line , parameter set presented in Table 3 ) shows nearly no difference with the model trained on a subset of linear ramps . The turn probability estimated experimentally is shown in black ( shades denote standard deviation as described in ‘Materials and methods’ ) . ( Bi ) Validation of the model on the 8-s linear ramp . ( Bii ) Validation of the model on the 8-s exponential ramp . ( Biii ) Validation of model on the 8-s sigmoid ramp . The correlation coefficients and coefficients of variation of the RMSE corresponding to the model trained on the linear ramps and the model trained on all ramps differ by less than 1% and 9% , respectively . ( C ) Training of the GLM on a single light ramp and validation on the 8-s sigmoid ramp: training on the 8-s linear ramp ( short dashed line ) ; training on the 8-s exponential ramp ( long dashed line ) . The output of the model trained on the complete set of ramps is shown as plain line . The correlation coefficients and coefficients of variation of the RMSE corresponding to the model trained on a single ramp and the model trained on all ramps differ by less than 1% and 10% , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 01510 . 7554/eLife . 06694 . 016Figure 5—figure supplement 3 . Comparison of the predictions of the integrated and control generalized linear model ( GLM ) with and without the contribution of the first derivative of the stimulus intensity . ( A ) Predictions for the linear low ( 4-s ) ramp . ( Top ) Time courses of the stimulus intensity ( magenta line ) and its time derivative ( gray line ) . ( Bottom ) The turn probability estimated experimentally is shown in black ( shades denote standard deviation as described in ‘Materials and methods’ ) . The predictions of the integrated stimulus-to-behavior GLM are shown in blue . The performance of the stimulus-to-behavior control model , based on the stimulus intensity alone ( dashed magenta line ) , improves by combining the light intensity with its first derivative . We note that the integrated stimulus-to-behavior GLM outperforms the two control GLMs . The discrepancies between the controls and the test model are particularly pronounced during the beginning of the rising phase of the ramp . The goodness of fit is quantified in Table 4 . ( B ) Same as panel A for the 8-s linear ramp . ( C ) Same as panel A for the 8-s exponential ramp . ( D ) Same as panel A for the 8-s sigmoid ramp . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 016 We took advantage of our ability to use stereotyped light ramps to elicit predictable and reproducible patterns of firing activity in the Or42a>ChR2 OSN ( Figure 4 , light-evoked activity patterns ) . In a series of experiments , we associated individual runs with a predefined light ramp and correlated the simulated OSN firing rate with the onset of run-to-turn transitions . In the example shown in Figure 5B , we began each run with either an exponential ramp or a constant basal light intensity ( internal control ) . When an exponential ramp was played to the larva , the pattern of light stimulation was executed as long as the larva remained in a run state . Upon interruption of the run , the light intensity was reset to baseline . As the motion of the larva had no influence on the stimulation pattern it experienced , this experimental protocol featured a sensorimotor loop that is essentially ‘open’ . When the behavior was modulated by changes in light intensity , the majority of the runs associated with an exponential ramp did not terminate before the falling phase of the ramp . This trend was quantified through the probability of turning ( or turn rate ) defined as the relative number of runs that switched to a turn during a given time window of 1 s ( ‘Materials and methods’ ) . The turn probability was estimated at every time point by using a sliding window . Upon constant light stimulation , we found that the instantaneous turn probability was largely independent of the duration of the ongoing run ( light gray line , Figure 5C ) . In contrast , the turn probability was strongly modulated by the exponential light ramp ( black line , Figure 5C ) . During the rising phase of the ramp ( 0–8 s ) , turning was suppressed below the value corresponding to basal stimulation . Conversely , a sharp increase in turn probability was observed during the falling phase of the ramp . The modulation of the turn probability by the light-evoked spiking activity corroborated the idea that strong activation of the OSN efficiently suppresses turning , whereas inhibition promotes turning . We set out to develop a quantitative model for the control of run-to-turn transitions by the neural activity . As the probability of turning remained approximately constant when the OSN activity was stationary , we hypothesized that the relationship between the OSN spiking activity and the control of run-to-turn transitions could be captured by a simple model where the time-varying probability of turning was described by the combination of a constant term ( λ0 ) and a term proportional to the current OSN firing rate: λ0+ λ1y ( t ) . To map this linear combination ( which can be positive or negative ) onto the definition domain of a probability ( which varies between 0 and 1 ) , we applied a standard logit transformation and described the turn probability as a generalized linear model ( GLM ) ( Myers et al . , 2002 ) :λ ( t ) =11+e− ( γ0+γ1y ( t ) ) . To define the parameters of the GLM ( λ0 and λ1 ) , we transformed the previous relationship as shown in Figure 5D , and we carried out a linear regression on the open-loop behavior elicited by 10 light ramps identical to those used to characterize the OSN response dynamics ( Figure 5—figure supplement 1 ) . The parameter set obtained through this procedure is reported in Table 3 . It was used to reproduce or predict behavioral transitions throughout the study . From here on , the GLM ( Figure 5D ) was fed with the OSN firing rate predicted from the neural model ( Figure 4B ) . This model will be referred to as the integrated stimulus-to-behavior GLM . 10 . 7554/eLife . 06694 . 017Table 3 . Parameters of the stimulus-to-behavior generalized linear model ( GLM ) obtained upon training of model on the stimulation ramps listed in Figure 5—figure supplement 1DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 017Control without first derivative of stimulusControl with first derivative of stimulusTest model with predicted neural activityγ0 ( constant ) −0 . 8156−0 . 8200−0 . 3534γ1 ( input variable ) −0 . 0114 ( W/m2 ) −1−0 . 0013 ( W/m2 ) −1−0 . 1523 Hz−1γ2 ( derivative of input variable ) –−0 . 0214 ( W/m2 ) −1 s–The first two columns of the table report the value of the stimulus-to-behavior control model without ( left ) and with ( center ) the contribution of the first derivative of the stimulus . The last column reports the value of the integrated stimulus-to-behavior model fed with the predicted firing rate of the OSN . The parameters listed in this table are used in all numerical simulations of the study , except for the validation controls described in Figure 5—figure supplement 2 . For the linear and nonlinear ramps , the stimulus-to-behavior GLM accurately reproduces the time courses of the experimental turn probability ( blue lines , Figures 5E , F ) . The performance of the test model was compared to a control GLM in which the turn probability was directly predicted from the stimulus intensity without any sensory processing from the OSN ( dashed magenta lines , Figure 5E , F ) . For this control model , we independently fitted the same GLM with the simulated OSN firing rate replaced by the stimulus intensity ( ‘Materials and methods’ ) . The values of the parameters of the control model are reported in Table 3 . The goodness of fit of the GLM was clearly contingent on the nonlinear transformation achieved by the OSN ( Table 4 ) . To rule out that the test GLM was overfitted , we trained the model on a subset of linear light ramps and validated the model on a set of nonlinear light ramps ( Figure 5—figure supplement 2 ) . 10 . 7554/eLife . 06694 . 018Table 4 . Quantification of the goodness of fit of the control and the test generalized linear model ( GLM ) DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 018Control GLM without derivativeControl GLM with derivativeTest GLMA Correlation coefficient ( ρ ) Linear ( 4 s ) , low0 . 690 . 750 . 89 Linear ( 4 s ) , med0 . 620 . 900 . 90 Linear ( 4 s ) , high0 . 670 . 960 . 95 Linear ( 4 s ) , slow0 . 540 . 920 . 91 Linear ( 4 s ) , fast0 . 650 . 760 . 90 Linear ( 8 s ) −0 . 050 . 660 . 78 Quadratic0 . 110 . 700 . 88 Exponential0 . 290 . 430 . 92 Sigmoid0 . 130 . 610 . 60 Asymptotic−0 . 100 . 250 . 03 All conditions ( all time points included ) 0 . 530 . 740 . 86B CV ( RMSE ) Linear ( 4 s ) , low0 . 590 . 540 . 33 Linear ( 4 s ) , med0 . 600 . 410 . 32 Linear ( 4 s ) , high0 . 750 . 410 . 33 Linear ( 4 s ) , slow0 . 560 . 340 . 27 Linear ( 4 s ) , fast0 . 530 . 450 . 31 Linear ( 8 s ) 1 . 060 . 710 . 59 Quadratic1 . 060 . 730 . 55 Exponential0 . 840 . 850 . 39 Sigmoid0 . 950 . 650 . 58 Asymptotic0 . 850 . 520 . 97 All conditions ( all time points included ) 0 . 650 . 510 . 39Comparison of the performances of the integrated stimulus-to-behavior GLM and the control model bypassing the OSN processing . The outputs of the test and control GLMs are obtained based on the parameter sets listed in Table 3 . ( A ) Application of Pearson's correlation coefficient ( ρ ) on the time course of the experimental turn probability and the simulated turn probability . Restriction of the quantification to the first 12 s of the ramp where the experimental estimate of the turn probability is reliable . ( B ) Same as panel A for the coefficient of variation of the RMSE . The goodness of fit computed for the entire set of ramps is reported at the bottom of the table for both metrics . The successful application of the stimulus-to-behavior GLM led us to conclude that: ( 1 ) stationary levels of OSN firing rate lead to probabilistic transitions from run to turn . When the OSN spiking activity remains constant , the probability of turning at a given time is largely independent of the duration of the run; ( 2 ) excitation of the OSN suppresses turning ( evident during rising phase of all ramps ) ; ( 3 ) inhibition of the OSN facilitates turning ( most evident during falling phase of the exponential ramp ) . Consistent with our finding that the OSN activity is sensitive to the slope of the ramp , we found that the performance of the control GLM was improved by combining the light intensity ( x ) with its first derivative ( dx/dt ) ( Figure 5—figure supplement 3 ) . For the majority of ramps , this improvement did , however , not match the quality of the fit produced by the integrated stimulus-to-behavior GLM ( Table 4 ) . We concluded that the nonlinear response characteristics of the OSN have a noticeable influence on the control of orientation behavior . To test the relevance of the integrated stimulus-to-behavior GLM in conditions in which the sensorimotor loop is closed , we synthesized a controlled light gradient with a shape comparable to that of the odor gradient ( Figure 6A ) . In this stimulation paradigm , the light intensity was continuously updated based on the position of the larva ( ‘Materials and methods’ ) . Figure 6A illustrates the behavior of an Or42a>ChR2 larva in a light gradient . As observed for the odor-evoked behavior ( Gomez-Marin et al . , 2011 ) , the larva ascended the light gradient and remained in the vicinity of its peak by implementing a series of runs and directed turns . In Figure 6B , we examined how the Or42a>ChR2 OSN responds to a replay of the light intensity changes experienced during the trajectory shown in Figure 6A . The spiking activity of the OSN displayed considerable processing of the stimulation pattern . This transformation of the stimulus was well captured by the IFF model . To predict the temporal evolution of the turn probability associated with individual runs , we fed the predicted spiking activity of the OSN into the GLM trained on the open-loop light ramps ( Figure 5 and Table 3 ) . Correlating the predictions of the model with the termination of the actual runs revealed that the initiation of a turn was typically preceded by a steady increase in the predicted probability of turning ( Figure 6B , C and Video 1 ) . To quantify this trend , we analyzed a large set of runs included in 25 trajectories , each trajectory corresponding to a different animal ( representation of a subset of 10 trajectories in Figure 6D ) . Since every run corresponded to a unique sensory experience , the predictions of the stimulus-to-behavior GLM could only pertain to the average behavior observed over multiple runs . We therefore analyzed the averaged trend of the turn probability preceding individual turns . 10 . 7554/eLife . 06694 . 012Figure 6 . Predictions of the integrated stimulus-to-behavior generalized linear model ( GLM ) for run-to-turn transitions observed in a virtual olfactory gradient . ( A ) Synthetic chemotaxis in a virtual odor gradient produced by light stimulation . The larva experiences a light intensity determined by a predefined stimulus landscape . The landscape displayed in the background of panel A is an exponential gradient centered on a point ‘source’ . Larvae responding to this light gradient accumulate at the gradient's peak as observed for odor gradients . Illustrative trajectory of the midpoint ( black ) and head ( magenta ) . Black arrows indicate the direction of motion . ( B ) Sensorimotor analysis of a representative trajectory . ( Top ) Time course of the light intensity associated with the trajectory displayed in panel A . ( Middle ) PSTH of the OSN activity measured experimentally upon a replay of the intensity time course at the electrophysiology rig ( black line ) . Numerical simulations of the neural activity carried out by the IFF motif ( green line ) presented in Figure 4B . ( Bottom ) Turn probability ( blue line ) predicted from the integrated stimulus-to-behavior GLM presented in Figure 5D ( parameter set listed in Table 3 ) . The neural activity simulated in the middle panel is fed into the GLM to predict the turn probability shown in the bottom panel . Behavioral predictions are only shown for the sequences associated with runs . ( C ) Overlay of the trajectory of the midpoint with the predicted turn probability color-coded in accordance with the color bar on the left . We observe that the turn probability tends to increase ( red color range ) when the larva is moving away from the gradient's peak , whereas it decreases ( blue color range ) when the larva is moving toward the peak . ( D ) Overlay of 10 trajectories recorded in the exponential light gradient shown in panel A . For each trajectory , the position of the midpoint is shown in gray . Turns are indicated as small black circles . ( E ) Turn-triggered average of the predicted turn probability for the exponential light gradient . A comparison is made between predictions based on the simulated OSN activity driven by the stimulus intensity ( test model , blue line ) , predictions based on the simulated OSN activity driven by the time-reversed stimulus time course ( uncoupled control , black line ) , and predictions based on the assumption that the neural activity stays constant over the course of each trajectory ( constant control , magenta line ) . The parameters of the stimulus-to-behavior GLM are listed in Table 3 . We observe that the turn probability steeply increases 4 s before the turn , which coincides with the median duration ( 3 . 8 s ) of the entire set of runs . Analysis conduced over 750 runs with a duration of minimum 1 s . Shaded areas represent SEM . ( F ) Log-likelihood of the predictions of the stimulus-to-behavior test GLM compared to the controls . Bootstrap analysis of the difference in log-likelihood ( logL ) between the test model and the controls normalized by the log-likelihood of the test model ( ΔlogL/logLtest ) . Distribution of the relative difference in logL is shown for the test model against the constant neural activity control ( red ) , and the uncoupled stimulus control ( black ) . The median of the distribution is equal to the value obtained from the original full set of runs; the median of the entire distribution is indicated by a dot in the x-axis . As an internal control , the test model was compared to itself ( blue ) . Out of 10 , 000 resampled subsets of runs , none of the controls was found to be more likely than the test model ( p < 0 . 0001 ) . The analysis included all observed runs with a duration of minimum 1 s ( 750 runs originating from 25 trajectories ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 01210 . 7554/eLife . 06694 . 019Video 1 . Illustrative trajectory sequence in an exponential light gradient with predicted turn probability colored in red . The scale bar at the bottom left of the Video represents 1 cm . The green trace displays a 20 s segment of the past positions of the centroid . White spots represent the position of the head every 20 frames ( or 0 . 66 s ) . The behavioral sequence is accelerated by a factor 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 019 As reported in previous work ( Gomez-Marin et al . , 2011 ) , we found that the stimulus intensity decreases steadily for several seconds prior to a turn ( data not shown ) . Accordingly , the stimulus-to-behavior GLM predicted that the stimulus downslope was transformed into a monotonic increase in turn probability ( light blue line , Figure 6E ) . To establish the sensorimotor control underlying this trend , we computed the turn-triggered averages of the turn probability by using two control models ( ‘Materials and methods’ ) : ( 1 ) behavioral predictions based on the assumption that the OSN spiking activity remained constant throughout the trajectory and ( 2 ) behavioral predictions upon uncoupling of the stimulus and the behavior by temporally inverting the reconstructed time course of the stimulus . In contrast to the test GLM , neither control displayed a substantial increase in turn probability prior to turning ( red and back dashed lines , Figure 6E ) . The significance of the improvement in the predictive power of the test model relative to the controls was established by comparing the log-likelihood computed over the entire set of runs ( Figures 6F , bottom panel and ‘Materials and methods’ ) . This allowed us to conclude that the integrated neural-to-behavior model built on controlled conditions of stimulation ( open-loop paradigm ) was sufficient to predict run-to-turn transitions arising from free behavior in a virtual odor gradient ( closed-loop paradigm ) . Next , we tested the idea that inhibition of the OSN activity at the stimulus offset is sufficient to trigger a nearly deterministic release of turning during free behavior . To this end , we designed radially symmetrical light landscapes with geometrical features producing inhibition or maintenance of OSN activity during free motion . As a control , we considered a landscape with an exponential rise interrupted by an exponential fall at a fixed distance of 8 mm from the center ( Figure 7A , top panel , ‘rim’ indicated by a dashed line ) . The shape of this landscape is reminiscent of a ‘volcano’ . The geometry of the landscape was chosen such that a larva moving at a speed of 1 mm/s from the foot of the gradient toward its center would experience a light pattern similar to the 8-s exponential ramp ( Figure 5F and Figure 7Ai ) . In response to an exponential ramp , the spiking activity of the OSN featured a steady increase followed by a rapid decrease . We therefore expected to observe turn suppression during the rising phase of the ramp and turn facilitation during the falling phase of the ramp . The tendency of larvae to initiate turning upon crossing of the volcano's rim was evident from the set of runs that moved from the outer to the inner edge of the volcano ( Figure 7Aii , bottom panel ) . The alternation between turn suppression and turn facilitation resulted in a zigzagging of trajectories across the rim of the volcano . The integrated stimulus-to-behavior GLM predicted a rise in the turn probability prior to the interruption of a run ( Figure 7E and Video 2 ) . 10 . 7554/eLife . 06694 . 020Figure 7 . Predictions of run-to-turn transitions elicited by a family of radially symmetrical light landscapes . ( Ai–Di ) Stereotyped OSN responses to light ramps starting with a common 8-s exponential rise: ramp with a smooth exponential fall ( ‘volcano’ , see panel A ) , ramp with an abrupt fall to basal intensity ( ‘well’ , see panel B ) , ramp with a prolongation of the maximum intensity ( ‘mesa’ , see panel C ) , and ramp with a linear increase ( ‘linear’ , see panel D ) . The spiking activity of the OSN is computed from the pure IFF ( ODE ) model presented in Figure 4B ( parameter set listed in Table 1 ) . Time course of the light intensity shown in magenta; first derivative of the light intensity shown in gray; simulated OSN activity shown in green . ( Ai ) Experimental and predicted OSN activity elicited by an exponential rise followed by an exponential fall . ( Aii ) Symmetrical two-dimensional light landscapes corresponding to the exponential ‘volcano’ profile described in panel Ai . ( Top ) Set of 54 trajectories superimposed onto the stimulus landscape . ( Bottom ) Set of 48 runs starting from the external edge of the ‘volcano’ and heading toward its center ( minimum run duration: 1 s ) . ( Bi ) Same as Ai for the ‘well’ profile . Strong inhibition of the OSN activity follows the abrupt fall in light intensity . ( Bii ) Crossing of the rim leads to an aversive response . As a consequence , larvae avoid the well at the center of the landscape . Set of 42 trajectories represented in the diagram at the top; set of 63 entering runs represented in the diagram at the bottom . ( Ci ) Same as Ai for the ‘mesa’ profile . The transition from an exponential rise in light intensity to constant intensity leads to a transient drop in neural activity before a steady state value is reached . ( Cii ) For the mesa landscape , crossing of the rim does not lead to an aversive response: larvae tend to maintain their ongoing run . The center of the landscape is therefore visited . Set of 36 trajectories represented in the diagram at the top; set of 79 entering runs represented in the diagram at the bottom . ( Di ) Same as Ai for a ‘linear’ hat profile . The deceleration in stimulus from an exponential rise to a linear rise leads to a transient drop in neural activity before a steady state value is reached . The corresponding OSN dynamics is similar to that elicited by the mesa ( panel Ci ) . ( Dii ) Upon crossing of the rim of the linear hat landscape , larvae undergo a deceleration in light intensity that is expected to modulate behavior in a way similar to the mesa landscape . Set of 47 trajectories represented in the diagram at the top; set of 72 entering runs represented in the diagram at the top . ( E ) Turn-triggered averages of the predicted turn probability for the subset of runs entering the landscape's central area ( bottom graphs of Aii–Dii ) . Each run included in the analysis crosses the rim of the landscape ( minimum run duration: 1 s ) . Shaded areas represent SEM . ( F ) Distribution of run durations following the crossing of the landscape's rim . Analysis restricted to the subset of runs described in the diagram at the tops of panels Aii–Dii ( runs entering the central area of the landscape ) . The experience of an abrupt fall in light intensity promotes rapid turning ( ‘well’ condition ) , whereas runs are elongated by constant light stimulation or by a linear rise in light intensity ( ‘mesa’ and ‘linear’ landscapes ) . Differences between the median of the run durations associated with each of the four landscapes is assessed through a Kruskal–Wallis test ( p < 10−10 ) followed by pair-wise Wilcoxon tests with a Bonferroni correction ( for the non-significant difference p > 0 . 05/6; for all other pairwise comparisons p < 0 . 05/6 ) . The behavior predicted from turn probability ( panel E ) is in good agreement with the shortening or elongation of the runs observed for each of the four landscapes . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 02010 . 7554/eLife . 06694 . 021Video 2 . Illustrative trajectory sequence in ‘volcano’ light gradient with predicted turn probability colored in red . The scale bar at the bottom left of the Video represents 1 cm . The green trace displays a 20 s segment of the past positions of the centroid . White spots represent the position of the head every 20 frames ( or 0 . 66 s ) . The behavioral sequence is accelerated by a factor 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 021 We then considered an extreme version of the volcano: a well ( Figure 7B ) . In this landscape , the light intensity experienced by a larva moving toward the center of the well corresponded to an exponential rise followed by a near-instantaneous drop . The IFF model correctly reproduced the strong inhibitory phase experimentally observed in the OSN spiking activity ( Figure 7Bi , bottom panel ) . This inhibition was expected to generate a nearly deterministic release of turns . Consistently , larvae avoided the well region ( Figure 7Bii ) . Such a behavior was correctly described by the GLM , which predicted a dramatic increase in turn probability following the crossing of the rim ( Figure 7E and Video 3 ) . To probe the idea that sustained OSN spiking activity suppresses turning , we synthesized a landscape complementary to the well: a ‘mesa’ in which the light intensity at the rim was extended to the central area of the landscape ( Figure 7C ) . During the transition from an exponential rise in light intensity to a plateau value , the OSN underwent a mild drop in spiking activity before a stationary value was reached—a feature accurately reproduced by the IFF motif ( Figure 7Ci ) . The stimulus-to-behavior GLM predicted a modest increase in turn probability upon crossing of the rim without significant avoidance of the central area of the mesa ( Figure 7E and Video 4 ) . This prediction was corroborated by our experimental results ( Figure 7Cii , bottom panel ) . 10 . 7554/eLife . 06694 . 022Video 3 . Illustrative trajectory sequence in the ‘well’ light gradient with predicted turn probability colored in red . The scale bar at the bottom left of the Video represents 1 cm . The green trace displays a 20 s segment of the past positions of the centroid . White spots represent the position of the head every 20 frames ( or 0 . 66 s ) . The behavioral sequence is accelerated by a factor 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 02210 . 7554/eLife . 06694 . 023Video 4 . Illustrative trajectory sequence in the plateau light gradient with predicted turn probability colored in red . The scale bar at the bottom left of the Video represents 1 cm . The green trace displays a 20 s segment of the past positions of the centroid . White spots represent the position of the head every 20 frames ( or 0 . 66 s ) . The behavioral sequence is accelerated by a factor 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 023 Finally , we considered an intermediate landscape consisting of a linear ‘hat’ ( a cone ) in which runs moving toward the center underwent a deceleration in stimulus intensity during the transition from the exponential rise to the linear rise ( ‘linear’ , Figure 7D ) . Due to the sensitivity of the OSN activity to deceleration in the stimulus intensity , the linear hat landscape led to a modest drop in firing rate similar to that observed for the mesa ( Figure 7Di ) . The IFF model faithfully reproduced this counterintuitive observation . At a behavioral level , the stimulus-to-behavior GLM predicted no significant difference between the behavior evoked by the mesa and the linear hat landscapes ( Figure 7E and Video 5 ) . These predictions were in good agreement with the free behavior of larvae ( Figure 7Dii , bottom panel ) . 10 . 7554/eLife . 06694 . 024Video 5 . Illustrative trajectory sequence in the linear ‘hat’ light gradient with predicted turn probability colored in red . The scale bar at the bottom left of the Video represents 1 cm . The green trace displays a 20 s segment of the past positions of the centroid . White spots represent the position of the head every 20 frames ( or 0 . 66 s ) . The behavioral sequence is accelerated by a factor 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 024 To assess the predictive power of the integrated stimulus-to-behavior GLM , we compared the average turn probability preceding a turn ( Figure 7E ) with the observed latency to turning upon crossing of the rim ( Figure 7F ) . The well was associated with the prediction of the steepest increase in turn probability , leading to the expectation that most runs stopped within 2 s of the rim crossing . The volcano led to a milder increase in the predicted turn probability for about 3 s , while the mesa and linear hat were predicted to generate an even weaker increase . As shown in Figure 7F , we found that the average latency to turn was shortest for the well landscape ( 0 . 93 s ) followed by the volcano ( 3 . 48 s ) . We observed significantly longer turn latencies for the mesa and linear hat with no difference between the two conditions ( 6 . 6 s and 6 . 7 s , respectively ) . In conclusion , the use of synthetic light landscapes permitted us to experimentally demonstrate that sustained OSN spiking activity suppresses turning during free behavior , whereas inhibition of the OSN activity promotes run-to-turn transitions in a nearly deterministic manner . This approach established that the relatively simple linear control underlying the GLM trained on the behavior of larvae experiencing stereotyped open-loop light stimulations is sufficient to account for the control of behavior elicited under conditions of closed-loop light stimulation . Our ultimate goal was to test the ability of the integrated stimulus-to-behavior model to predict the duration of runs in an odor gradient . In a real odor landscape ( Figure 8A ) , larvae accumulated at the peak of the gradient with a dispersal notably larger than that observed in a light gradient ( Figure 6D ) . This apparent decrease in orientation performances can be partly explained by the shallower geometry of the odor gradient ( Figure 8—figure supplement 1 ) . In Figure 8B , the goodness of fit between the spiking activity of the OSN and the output of the IFF+IFB model can be appreciated for the representative trajectory highlighted in Figure 8A ( magenta trace ) . To predict run-to-turn transitions during free motion in a real odor gradient , we replaced the pure IFF model devised for light-evoked spiking activity by the composite IFF+IFB model as input for the stimulus-to-behavior GLM trained on the open-loop light ramps ( Figure 5D and Table 3 ) . The model predicted that runs were on average associated with a monotonic increase in the turn probability during several seconds before a turn ( Figure 8C ) . Based on these predictions , we computed the likelihood of the ensemble of runs observed in the odor gradient . This likelihood was significantly larger than that computed for two control models ( Figure 8D ) . Together , these results establish that the structure and parameters of the integrated stimulus-to-behavior GLM form a solid conceptual basis to describe how the sensory dynamics of single OSNs influence run-to-turn transitions during naturalistic behavior ( Video 6 ) . 10 . 7554/eLife . 06694 . 025Figure 8 . Predictions of the integrated stimulus-to-behavior model for run-to-turn transitions observed in a real odor gradient . ( A ) Superimposition of 10 consecutive trajectories observed in an odor gradient of isoamyl acetate ( same experimental conditions as Figure 1 ) . For every trajectory , the position of the midpoint is shown in gray . Small black circles indicate turns . The head position of the trajectory presented in Figure 1A is highlighted in magenta . Arrows indicate the direction of motion . The odor gradient shown in the background corresponds to the reconstructed snapshot 60 s after the onset of the diffusion process ( ‘Materials and methods’ ) . ( B ) Sensorimotor analysis of a representative trajectory . ( Top ) Time course of the reconstructed odor concentration associated with the trajectory displayed in panel A . ( Center ) PSTH of the OSN measured experimentally in response to a replay of the odor concentration course ( black line ) . Neural activity simulated by the composite IFF+IFB ODE model ( blue line ) presented in Figure 4B ( parameter set listed in Table 1 ) . ( Bottom ) Turn probability ( blue line ) predicted by the stimulus-to-behavior GLM trained on the light-evoked open-loop behavior reported in Figure 5—figure supplement 1 ( parameter set listed in Table 3 ) . The neural activity simulated in the middle panel is fed into the GLM to predict the turn probability shown in the bottom panel . Behavioral predictions are only shown for the sequences associated with runs . The predicted turn probability is only shown for the behavioral sequences associated with runs . ( C ) Turn-triggered average of the predicted probability of turning for behavior observed in the odor gradient . A comparison is made between predictions based on the simulated OSN activity driven by the stimulus intensity ( coupled test , blue ) , predictions based on the simulated OSN activity driven by the time-reversed stimulus time courses ( uncoupled control , black ) , and predictions based on the assumption that the neural activity stays constant over the course of each trajectory ( constant control , red ) . As for the light gradient , we observe that the predicted turn probability increases 5 s before the turn , which coincides with the median duration ( 5 . 4 s ) of the entire set of runs . Since the geometry of the odor gradient is shallower than the light gradient ( Figure 8—figure supplement 1 ) , the increase in turn probability has a reduced amplitude compared to the behavior elicited by the light gradient ( Figure 6E ) . Shaded areas denote SEM . ( D ) Log-likelihood of the predictions of the integrated stimulus-to-behavior GLM compared to the controls . Bootstrap analysis of the difference in log-likelihood ( logL ) of the test model and the controls normalized by the log-likelihood of the test model ( ΔlogL/logLtest ) . Distribution of the relative difference in logL computed for the test model against itself ( blue ) , against the constant neural activity control ( red ) , and against the uncoupled stimulus control ( gray ) . The median of the distribution is equal to the value obtained from the original full set of runs; the median of the entire distribution is indicated by a dot in the x-axis . Based on 10 , 000 resampled subsets of runs , we conclude that the test model is significantly larger than both controls ( p = 0 . 0001 for the constant neural activity control and the uncoupled stimulus control ) . For panels C and D , the analysis includes all runs with a duration of minimum 1 s ( 304 runs originating from 20 trajectories ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 02510 . 7554/eLife . 06694 . 026Figure 8—figure supplement 1 . Comparison of the geometry of the exponential light gradient and the odor gradient . ( A ) Reconstruction of the odor gradient displayed in Figures 1 , 8 with color-coding in accordance with the scale on the right . Snapshot corresponding to the odor landscape 60 s after the initiation of the diffusion process . Cross-section intersecting the gradient's peak: f ( 5 , y ) . ( B ) Same as panel A for the light gradient presented in Figure 6 . Cross-section along gradient: g ( 5 , y ) . ( C ) Comparison of the cross-sections f and g of the odor and light gradients , respectively . Each profile was normalized by its maximum value reached under the gradient's peak . ( D ) Comparison of the derivative of the cross-sections of the odor and light gradients: df/dy and dg/dy where y represents the variable associated with the vertical axis . From this analysis , we observe that the slope of the light gradient is considerably steeper than the slope of the odor gradient , which is expected to facilitate chemotaxis . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 02610 . 7554/eLife . 06694 . 027Video 6 . Illustrative trajectory sequence in odor gradient with turn probability colored in red . Same trajectory as that shown in Figure 8A . Superimposition of the behavior on the dynamical reconstruction of the odor gradient based on the PDE simulations . The scale bar at the bottom left of the Video represents 5 mm . The green trace displays a 20 s segment of the past positions of the centroid . White spots represent the position of the head every 20 frames ( or 0 . 66 s ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 027 Most primary sensory neurons operate differently from proportional counters ( Rieke , 1997; Song et al . , 2012 ) . Individual OSNs of C . elegans and cockroaches function as bipolar detectors that selectively respond to either increases or decreases in stimulus intensity ( Tichy et al . , 2005; Chalasani et al . , 2007 ) . A similar specialization into ON-OFF detection pathways has been observed for thermotaxis in C . elegans ( Suzuki et al . , 2008 ) and motion perception in adult flies ( Joesch et al . , 2010 ) . In contrast with these binary sensory responses , we discovered that a single larval OSN is sensitive to both the stimulus intensity and its first derivative . The enhanced information-processing capacity of primary olfactory neurons in the larva is consistent with the response characteristics of OSNs in adult flies , which encode complex dynamical features of airborne odorant stimuli ( Kim et al . , 2011; Martelli et al . , 2013 ) . To describe the input–output response properties of single larval OSNs , we set out to build a biophysical model of the olfactory transduction pathway . IFB motifs constitute the core mechanism of chemoreception in bacteria , olfactory transduction , and phototransduction ( Yi et al . , 2000; De Palo et al . , 2013 ) . In adult flies , Nagel and Wilson ( 2011 ) investigated how the potential involvement of negative feedback on the olfactory transduction cascade could account for dynamical and adaptive features of OSN response . On the other hand , IFF motifs are implicated in the regulation of numerous cellular and developmental processes ( Goentoro and Kirschner , 2009; Lim et al . , 2013 ) , and their contribution to sensory processing has been documented in recent work ( Kato et al . , 2014; Liu et al . , 2015 ) . These results led us to conjecture that two regulatory motifs might be involved in larval olfactory transduction: an IFB and an IFF featuring direct excitation and indirect inhibition ( Figure 4A , B ) . Using a parameter optimization approach , we found that a pure IFF motif is sufficient to approximate the response properties of the OSN . Combining the IFF and IFB motifs was nonetheless necessary to recapitulate the richness of OSN dynamics elicited by naturalistic olfactory stimuli ( Figure 4 and Figure 4—figure supplement 1 ) . Consistent with the model proposed by Nagel and Wilson ( 2011 ) , our numerical simulations indicate that the integral feedback applies to the signaling pathway specific to the odorant receptor ( OR ) . Nagel and Wilson have suggested that a diffusible effector—potentially intracellular calcium—inhibits the activity of the OR , thereby affecting the onset and offset kinetics of the OSN response . By contrast , the IFF motif would describe a regulatory mechanism acting on components of the transduction pathway downstream from the OR ( Gu et al . , 2009 ) . It is plausible that the IFF regulation is also mediated by intracellular calcium . What features of the olfactory stimulus are encoded in the spiking dynamics of single larval OSNs ? Our biophysical model of the olfactory transduction cascade shows that the spiking activity of the OSN follows a standard hyperbolic dose-response when stimulated by prolonged pulses of odor ( ‘Materials and methods’ ) . In this regime , the maximum OSN firing rate we observed for IAA is modest ( Figure 3—figure supplement 1 ) . Changes in odor concentration occurring on a timescale relevant to the behavior—a second or shorter—can produce significantly higher ( or lower ) firing rates . This sensitivity to positive and negative changes in stimulus intensity can be explained by the mathematical solution we derived for the OSN dynamics ( Equation 1 ) . Upon changes in odor concentration , the dose–response function describing the OSN spiking activity is transiently rescaled ( or ‘normalized’ ) by the short-term history of the stimulus derivative ( ‘memory’ on characteristic time scale of 1 s , see ‘Materials and methods’ ) . As a result , positive derivatives in stimulus intensity excite the OSN . Negative derivatives can inhibit the OSN firing rate in a manner consistent with the stimulus-offset inhibitions observed in adult-fly OSNs ( Hallem et al . , 2004; Nagel and Wilson , 2011 ) . Our model indicates that a single Or42a OSN combines the function of a slope ( ON ) detector in response to positive gradients and an OFF detector in response to negative gradients . When larvae ascend Gaussian odor gradients originating from single odor sources , we thus expect high OSN firing rates . Robust inhibition of OSN spiking activity would result from motion that takes larvae down the odor gradient . How relevant are the features encoded by the spiking activity of the Or42a OSN to the behavioral dynamics directing chemotaxis ? To address this question , we substituted the odor stimulation with optogenetics-based light stimulation and gained unprecedented control over the spiking activity evoked in a genetically targeted OSN . Under the conditions of open-loop light stimulation , we found that OFF responses ( offset inhibition of the OSN firing ) promote turning , whereas ON responses ( sustained high firing ) suppress turning ( Figure 5 ) . We applied a GLM to describe the link between the OSN spiking dynamics and the probability of switching from a run to a turn ( Figure 5D ) . The accuracy of the model's output showed a striking dependence on the nonlinear transformation achieved by the olfactory transduction cascade ( Figure 5—figure supplement 3 ) . Ultimately , we combined the biophysical model for the OSN spiking dynamics with the GLM to make robust predictions about closed-loop behavior in virtual and in real odor gradients ( Figures 6–8 ) . The integrated stimulus-to-behavior GLM clarifies how features encoded in the activity pattern of individual primary olfactory neurons influence behavioral dynamics . The information transmitted by a single larval OSN is sufficient to represent positive and negative odor gradients through the excitation and inhibition of spiking activity . Unlike for chemotaxis and thermotaxis in C . elegans where the ON and OFF pathways are associated with different cellular substrates ( Chalasani et al . , 2007; Suzuki et al . , 2008 ) , the same larval OSN is capable of controlling up-gradient and down-gradient sensorimotor programs . This observation echoes findings recently made for thermotaxis in the Drosophila larva ( Klein et al . , 2015 ) . Furthermore , it corroborates the idea that sensory representations are rapidly transformed into motor representations in the circuit controlling chemotaxis ( Luo et al . , 2014 ) . In the future , it will be important to define whether the sensorimotor principles proposed for the Or42a OSN can be generalized to OSNs expressing other odorant receptors ( Fishilevich et al . , 2005; Kreher et al . , 2008; Mathew et al . , 2013 ) . In addition , the network of interneurons located in the larval antennal lobe ( Das et al . , 2013 ) is expected to participate in the processing of olfactory information arising from the OSNs ( Asahina et al . , 2009; Larkin et al . , 2010 ) . Although our work suggests that the computations achieved by the antennal lobe are not strictly necessary to guide robust chemotaxis ( see also ‘Materials and methods’ ) , the function of the transformation carried out by the synapse between the Or42a OSN and its cognate projection neuron ( PN ) remains to be elucidated in the larva ( Ramaekers et al . , 2005; Asahina et al . , 2009; Masuda-Nakagawa et al . , 2009 ) . As adult-fly PNs encode the second derivative of olfactory stimuli ( Kim et al . , 2015 ) including circuit elements downstream of the OSNs in the present multilevel model are expected to improve the accuracy of the behavioral predictions of the model . The aim of this study was to clarify the relationships between the peripheral encoding of naturalistic olfactory stimuli and gradient ascent toward an odor source . By exploiting the sufficiency of a single OSN to direct larval chemotaxis ( Fishilevich et al . , 2005; Louis et al . , 2008 ) , we developed a mathematical model accounting for the transformation of time-varying stimuli into the firing rate of an OSN and the conversion of dynamical patterns of OSN activity into the selection between two basic types of action—running and turning . It will be interesting to examine the validity of the present model for the sensorimotor control of other aspects of larval chemotaxis such as turn orientation through lateral head casts ( casting-to-turn transitions ) . In adult flies , turn orientation is determined by the crossing of the boundaries of odor plumes: upon encountering of an odor plume , flies veer upwind whereas exiting the plume initiates lateral and vertical casting ( van Breugel and Dickinson , 2014 ) —an orientation strategy related to the surge-and-cast response of moths ( Carde and Willis , 2008 ) . To orient in a rapidly changing olfactory landscape , the OSNs of various flying insects are capable of tracking rapid odor pulses on sub-second timescales and differentiating these signals ( Kim et al . , 2011; Fujiwara et al . , 2014; Szyszka et al . , 2014 ) . Whether the processing of turbulent olfactory inputs involves more temporal integration than that described by the sensorimotor model proposed here remains to be elucidated . Finally , the Drosophila larva offers a unique opportunity to delineate the neural circuit basis of behavior ( Ohyama et al . , 2013 , 2015 ) . Interdisciplinary approaches combining behavioral screens , functional imaging , and circuit reconstruction on the one hand ( Yao et al . , 2012 ) , and computational modeling and robotics on the other hand ( Grasso et al . , 2000; Webb , 2002; Izquierdo and Lockery , 2010; Ando et al . , 2013 ) , should improve our understanding of how brains with reduced numerical complexity exploit streams of sensory information to direct action selection . All behavioral experiments shown in the main figures were achieved with third instar larvae expressing the co-receptor Orco in only one OSN ( Fishilevich et al . , 2005 ) ( Or42a-Gal4>UAS-Orco , UAS-ChR2-H134R;Orco−/− ) in a double blind background ( GMR-hid/+;dTrpA11 ) ( Kwon et al . , 2008; Xiang et al . , 2010 ) . For the control experiments shown Figure 18 , the double blind background was achieved with the null alleles glass60j and dTrpA11 ( Moses et al . , 1989; Busto et al . , 1999 ) . The UAS-ChR2-H134R transgene was donated by Stefan Pulver and Leslie C Griffith ( Pulver et al . , 2009 ) . Flies were raised on standard fly food containing 0 . 5 mM all-trans-retinal in an incubator in complete darkness ( food vials wrapped in aluminum foil ) . Exposure to ambient light was minimized until the experimental test . Approximately 96 hr after egg laying , third instar larvae were taken out of the food and immersed in a 15% ( wt/V ) glucose solution . A controlled odorant environment was created in a 120 × 120 × 12 mm arena consisting of a polystyrene dish ( the lid of a Greiner square dish ref . number: 688102 , Sigma–Aldrich , St . Louis , MO ) standing on a 2% wt/V agarose surface inside the closed-loop tracker . A 3-μl odor droplet of IAA ( 0 . 25 M ) was placed inside a plastic reinforcement ring at the center of the dish ( internal diameter of disk occupied by the odor droplet: 5 mm ) . Inside the arena , an odor gradient emerged as a result of the diffusion from the source for 30 s prior to the introduction of a single larva . This step required a brief opening of the arena . The tracking was carried out for a minimum duration of 3 min . A minority of trajectories associated with no chemotactic response or with larvae idly dwelling under the odor source was excluded from the dataset . For the experimental conditions used in previous work ( Louis et al . , 2008; Asahina et al . , 2009; Gomez-Marin et al . , 2011; Gomez-Marin and Louis , 2014 ) , we obtained evidence that the odor gradients could be approximated as static . Due to the use of an odor source with reduced volume , this approximation did not hold in the present study . To correlate the behavior of the larva with a more accurate reconstruction of the odor gradient , we developed a physical model for the diffusion of the odor inside the behavioral arena ( Figure 9 ) . We used model-based estimation techniques for parameters underlying this physical model . We considered 3D diffusion with separate diffusion constants for air and the droplet . Exposed plastic surfaces of the chamber were treated as adsorptive boundaries . Since the odor gradient was initially established in the arena for 30 s prior to the introduction of a larva , our model also included non-zero initial concentration of the odor in the air , agarose , and plastic chamber . COMSOL Multiphysics v4 . 3 ( COMSOL , Burlington , MA ) was used to solve the diffusion equation with these boundary conditions . Parameter estimation was performed using the MATLAB/Optimization toolbox ( MathWorks , Natick , MA ) by solving a nonlinear least squares problem that matched the simulated odor concentration to measurements at the same time points . 10 . 7554/eLife . 06694 . 030Figure 9 . Physical model of odor diffusion in behavioral arena . ( A ) Configuration of behavioral arena on which the PDE model is based . The arena consists of a square shaped transparent plastic box with a side length of 120 mm and a height of 12 mm . The lid is inverted on a surface of agarose . The odor source consists of a solution of isoamyl acetate mixed with paraffin oil . A droplet of 3 μl of odor is placed inside a transparent reinforcement ring of a radius rring . This volume fills the ring evenly and , upon inversion of the lid on an agarose slab , the droplet remains suspended due to surface tension . The droplet shape is modeled as a spherical cap . The flat face of the droplet is in contact with the top plastic lid . The volume vdrop of the droplet is related to the radius of the flat face rring and the droplet height hdrop according to the formula of a sphere . The agarose layer at the bottom of the chamber is modeled as a two-dimensional sheet with an independent diffusion constant . The top flat face of the droplet that contacts the plastic cap is treated as a no-flux boundary , and flux continuity is imposed on the spherical interface with air . The remaining boundaries , air-agarose and air-plastic , are modeled as Robin boundary conditions to accommodate the possibility of adsorption-desorption at these boundaries . The establishment of the odor gradient in the arena is modeled by two simultaneous diffusion processes , both of which are described by partial differential equation ( PDE ) : ∂x ( r⃑ , t ) /∂t = D ∇2x ( r⃑ , t ) where x ( r⃑ , t ) denotes the odor concentration at position r⃑ and time t . The diffusion constant D depends on whether the medium is air or the odor droplet . We used a flux continuity condition at the droplet-air boundary . For additional details about the model , see ‘Materials and methods’ . ( B ) As described in Louis et al . ( 2008 ) , infrared spectroscopy was used to estimate the absorbance and thereby the average concentration along sections of the arena ( IR beam depicted in panel A ) . The time course of the cumulated concentration was determined for 7 sections at a distance from the source ranging from 0 to 45 mm ( only first 6 are shown in the graph ) . Each concentration profile results from an average over 2 to 4 independent measurements . The absorbance was measured for a source concentration of 1 . 0 M . As discussed in ‘Materials and methods’ , the parameters of the model are estimated by optimizing the fit of the model with the average concentration profiles along the 7 sections of the arena . The parameters of the model are reported in Table 5 . The PDE model leads to a good fit of the temporal profiles of the average concentrations after an initial transient phase of 30 s . ( C ) Assessment of variability in the concentration estimates at a fixed position of the center of the 1 . 0 M odor source ( 7 . 5 mm ) . Mean concentration obtained from four independent infrared measurements . Error bars denote standard deviation . The time course of the simulated concentration is shown as a dashed line . ( D ) Assessment of variability in the concentration estimates at a fixed position of the center of the 1 . 0 M odor source ( 15 mm ) . Mean concentration obtained from three independent infrared measurements . Error bars denote standard deviation . The time course of the simulated concentration is shown as a dashed line . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 030 The geometry of the experimental arena is described in Figure 9 . The radius rring of the odor ring confines the liquid droplet so that the radius of the flat face is equal to rring . The volume Vdrop of the droplet , made up of odor and solvent , is fixed to be 3 μl . According to the formula of a sphere , Vdrop= π6 hdrop ( hdrop2+3rring2 ) where hdrop is the droplet height . The agarose layer at the bottom of the chamber was modeled as a two-dimensional sheet with an independent diffusion constant . Third instar larvae were transferred from the food vial into 15% ( wt/V ) glucose solution . Dissection of tissues was carried out in cold extracellular saline solution ( Singleton and Woodruff , 1994 ) where the head was separated from the rest of the body while the brain was left intact . Using tissue glue ( Histoacryl B , Braun , Germany ) , the dissected head was then glued in the middle of a glass slide at the bottom of a flow chamber . The cuticle covering the mouth hook was removed using a 3 mm Vanna spring scissor ( Fine Science Tools , Germany ) to make the dorsal organ ganglion accessible to the recording electrode . Throughout the experiment , the head was immersed in extracellular saline . The flow chamber was connected to two syringe pumps ( Aladdin2-220 , World Precision Instruments , Sarasota , FL ) to perfuse the preparation with fresh saline and to ensure the continuous evacuation of the odor out of the chamber . The chamber volume was approximately 500 μl . The flow in the chamber was 28 . 4 μl/s , leading to a turnover of the chamber volume in 17 . 6 s . Recording electrodes were pulled ( P97 , Sutter Instruments ) out of borosilicate glass capillaries ( 1 . 5 mm/1 . 12 mm outer/inner diameters ( OD/ID ) , World Precision Instruments , Novato , CA ) with a 10 μm open tip . Electrodes were then back-filled with 3 μl of extracellular saline . A chlorinated silver wire ( 0 . 38 mm in diameter ) was used to connect the electrode to the head stage of a microelectrode amplifier ( Axon MultiClamp 700B , Molecular Devices , Sunnyvale , CA ) . The electrode was mounted on an automated micromanipulator ( ROE-200 & MPC-200 , Sutter Instruments ) . The antennal nerve in close vicinity of the dorsal ganglion was sucked into the recording pipette by applying a negative pressure ( −20 kpa ) created through vacuum . The extracellular signal was amplified 100 times at the microelectrode amplifier; it was visualized on an oscilloscope ( Tektronix , Beaverton , OR ) , and recorded at a sample rate of 20 kHz by a personal computer ( PC ) equipped with the free data acquisition software SpikeHound ( Lott et al . , 2009 ) . The conception of this preparation and recording technique benefited from pioneering recordings from the larval olfactory organ ( Oppliger et al . , 2000; Kreher et al . , 2005; Hoare et al . , 2008 ) . Standard systems identification approaches have shown that the important aspects of the response of invertebrate photoreceptors can be approximated as linear ( Marmarelis and McCann , 1977 ) , even though the modeling of nonlinear features of the response requires a more sophisticated treatment ( French et al . , 1993 ) . More recently , it has been suggested that the response dynamics of OSNs in C . elegans and primary thermosensory neurons in the Drosophila larva are largely linear ( Kato et al . , 2012; Klein et al . , 2015 ) . We therefore examined whether a linear-nonlinear model could be used to describe the OSN response of the larva . Following a reverse-correlation approach , we stimulated the Or42a>ChR2 OSN by a M-sequence induced with light ( Figure 13A ) . Reproducible patterns of neural activity were observed ( Figure 13C ) , from which a biphasic filter was computed ( Figure 13B ) . This filter had a shape similar to those found in retinal ganglion cells ( Chichilnisky , 2001 ) and insect OSNs ( Geffen et al . , 2009; Martelli et al . , 2013 ) . To test the predictive power of this filter , we reconstructed the activity elicited by an exponential and a sigmoid ramp ( Figure 13D , E ) . Whereas the linear filter led to a reasonable reconstruction of the firing pattern elicited by the M-sequence ( Figure 13C ) , it produced unsatisfactory results for the graded ramps with a mismatch so pronounced that it could not be rectified by a nonlinear function . This conclusion is consistent with the nonlinear response dynamics observed in adult-fly OSNs upon stimulation by graded odor ramps ( Kim et al . , 2011 ) . 10 . 7554/eLife . 06694 . 035Figure 13 . A linear filter alone is insufficient to account for the transfer function of the Or42a>ChR2 OSN . ( A ) Stimulation of the Or42a>ChR2 OSN by a maximum-length ( M ) sequence generated with light . The M-sequence was based on a discretization of the light intensity range 15–207 W/m2 into the following 5 values: 15 , 50 , 100 , 150 , and 207 W/m2 . The M-sequence featured all possible 4-element combinations of these 5 intensities . In the experiments , changes in light intensity occurred with time steps of 33 ms . ( Bottom , left ) PSTH of the neural response computed over 10 trials ( 10 preparations ) and for a bin size of 10 ms . The gray boxes outline a 2-s time window over which the predictions of the linear filter are reported in panel C . ( B ) Computation of the linear filter , h , through the operations described in panel B ( Chichilnisky , 2001; Nagel and Wilson , 2011 ) . Function F represents the Fourier transform from the time domain to the frequency domain; F−1 represents the inverse transformation from the frequency domain to the time domain . The bar above the variables in frequency space denotes the complex conjugate transformation . To cancel any DC drifts in the OSN response , the filter was computed on windows of 5-s slid over the entire duration of stimulus ( 20 s ) . An average filter was computed from this series ( dark blue line ) . The linear filter was used to make predictions about particular stimulus time courses . ( C ) Neural activity predicted from the linear filter in response to the M-sequence . The prediction was obtained by convolving the filter with the time course of the stimulus . The resulting activity was normalized to have the same mean as the experimental activity . The result of the prediction is shown for a 2-s window of the complete stimulus ( dashed gray box in panel A ) . The activity predicted from the linear filter ( dark blue line ) is compared to the output of the IFF ( ODE ) model introduced in Figure 4B ( green line ) . Pearson's correlation coefficient ( ρ ) of the predicted and the experimental activities is 0 . 37 for the linear filter reconstruction compared to 0 . 57 for the IFF model . ( D ) Application of the linear filter derived in panel C for the M-sequence to stimulation by an exponential ramp ( red line ) . The linear filter ( dark blue line ) fails to predict the PSTH that was observed experimentally ( black line; gray error bars represent the standard deviation ) ( ρ = 0 . 56 ) . Prediction of the IFF model shown in green ( ρ = 0 . 93 ) . ( E ) Application of the linear filter derived in panel C to a sigmoid ramp ( red line ) . As in panel D , the linear filter ( dark blue line ) fails to reproduce the experimental PSTH ( ρ = 0 . 54 ) while the IFF model leads to a good fit ( green line , ρ = 0 . 98 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 035 The dynamics of the pure IFF motif ( Figure 4Bi ) is described by a 3-variable ODE system ( Figure 4Bii ) . We hypothesized that the firing rate of the OSN ( y ) results from the combined effects of direct excitation and indirect inhibition of the OSN activity . The excitation is mediated by the gating of the OR by the binding of odorant molecules or the absorption of photons by channelrhodopsin-2 ( ChR2 ) . By analogy to the olfactory transduction cascade in the moth ( Gu et al . , 2009 ) , we speculated that the indirect inhibition is mediated by an intermediate variable ( u ) that might represent the concentration of calcium bound to calmodulin . For the pure IFF motif , the dynamics of variable u results from a production term proportional to the stimulus x and a first-order decay term . To model the direct excitation and indirect inhibition of the OSN activity , we used a control function ( d ( x , u ) where d stands for depolarization ) inspired by the cis-regulatory logic of gene transcription ( Goentoro and Kirschner , 2009 ) :d ( x , u ) =β1xβ2+x+β3u . This expression was built from thermodynamic considerations about the state of a promoter occupied by transcription factors ( Ackers et al . , 1982; Bintu et al . , 2005 ) . Here , we hypothesized that a similar function is suitable to describe the depolarizing effects of the opening of the OR ( or ChR2 ) , and the indirect hyperpolarizing effects that calcium bound to calmodulin might have on the OSN membrane . The contribution of each trend is described by x and β3 u , respectively . In addition , we assumed that the intermediate variable ( u ) and the OSN spiking activity ( y ) undergo a first-order decay . For the OSN activity , the introduction of such a decay can be justified by speculating about the existence of ion pumps that restore the membrane to resting potential after an initial increase of cations following the gating of the OR ( or ChR2 ) ( Gu et al . , 2009 ) . By trial and error , we also discovered the necessity of including a constitutive decay ( offset ) term that vanishes at a low firing rate . Although the molecular correlate of this offset remains undefined , it could be explained by the homeostatic function of ion pumps . The combined effects of the two decays are mathematically described as:h ( y ) =−β4y2y2+θ2−β5y , where h stands for hyperpolarization . To keep the model as simple as possible , the membrane potential was not modeled explicitly . Instead , we assumed that depolarizing d ( x , u ) and hyperpolarizing h ( x ) effects on the OSN membrane can be translated into excitatory and inhibitory effects on the OSN firing rate . While our knowledge about the olfactory transduction cascade in Drosophila was insufficient to justify these assumptions , the goodness of fit resulting from the integration of the ODE model demonstrated that the OSN dynamics could be captured by the combination of d ( x , u ) and h ( x ) . By combining the previous relationships , we obtained the following systems of ODEs:dudt=α1x−α2u , ( 2 ) dydt=β1xβ2+x+β3u−β4y2y2+θ2−β5y . The second regulatory motif we considered is a negative IFB . This motif has been implicated in the process of olfactory transduction and adaptation in adult flies ( Nagel and Wilson , 2011 ) . It also forms the regulatory basis of the transduction pathway underlying adaptive chemoreception in bacterial chemotaxis ( Yi et al . , 2000; Tu et al . , 2008 ) . For this motif , we assumed that the activity of the neuron had an excitatory effect on the intermediate variable u , which in turn has an inhibitory effect on the OSN activity . In a first approximation , the negative feedback was assumed to be linear . The difference between the IFF and IFB motifs lies in the production of the intermediate variable ( u ) , which in the case of the IFB is proportional to the firing rate ( y ) and not the stimulus intensity ( x ) . These considerations yielded the following system of ODEs:dudt=α3y−α2u , ( 3 ) dydt=β1xβ2+x+β3u−β4y2y2+θ2−β5y . Using numerical simulations , we found that the IFB motif alone cannot account for the dynamics of the OSN activity . In contrast , we discovered that the combination of the IFF and IFB motifs leads to substantial improvements in the quality of the fit ( Figure 4 ) . For all simulations achieved in this work , the ODE equations were numerically integrated by the solver ode23s built in Matlab . To optimize the parameters of the model to the experimental data , a standard fitting procedure was applied . As outlined in Figure 4Bi , we considered three possible models: the motif IFF , the motif IFB , and a combination of the two motifs . The joint probability of the observations was maximized as a function of the internal parameters for the neural activity patterns elicited by a set of 6 linear ramps , 5 nonlinear ramps , ( Figure 4—figure supplement 1 ) and a naturalistic stimulus ( Figure 2D , E ) . For each stimulation protocol , the confidence interval of the OSN activity ( PSTH ) was used to achieve a robust fit of the free parameters of the model . This procedure was applied to each of the three models independently . In the ODE systems presented in Figure 4Bii and Equations 2 , 3 , we observe that the time derivative of u can be multiplied by an arbitrary scaling factor reabsorbed by the fitting parameter β3 in the time derivative of y . For this reason one is forced to fix one of the eight parameters to a constant value . For numerical convenience we chose to fix α1 = 0 . 1 . To infer the actual value of this parameter , one would need to experimentally access the value of the intermediate variable u , whose molecular identity remains unknown . For the pure IFF model , the number of free parameters is therefore seven . In addition to these parameters , we considered the scaling of the firing rate y elicited by individual stimulation protocols via a multiplicative factor accounting for variability across experimental conditions ( e . g . , minute differences in the positions of the stimulation pipette ) . The maximization of the likelihood function was achieved by means of the Nelder-Mead ( NM ) method ( Nelder and Mead , 1965 ) , which proved to be fast and reliable . The result of the NM optimization was then refined through a gradient search algorithm ( Brun and Rademakers , 1997 ) . For the dataset corresponding to the light stimulation , the fitting procedure led us to rule out the relevance of the IFB model alone with a probability of χ2 very close to zero . In contrast , the IFF was able to reproduce the experimental observation with good accuracy . On the other hand , in the case of the odor stimulation , we obtained a significant improvement of the model fit by adding an IFB component to the IFF motif ( addition of the term α3 y to the dynamics of u ) . The fitted value of the composite IFF+IFB motif indicated that the IFB component was not negligible during the stimulation and accounted for about 30% of the final firing rate ( Figure 4D ) . In contrast to the pure IFF model , variables u and y of the IFF+IFB model were entangled in the structure of the ODE resulting in a coupling that allowed us to fit the value of parameter α1 . We also examined the effect of introducing additional terms in the denominator of the function defining y , such as the product u × y . Besides the test of other circuit motifs , the introduction of additional free parameters represented a qualitative test against the possibility of over-fitting . The improvements in the fitting obtained in these cases were very marginal . With regards to both light and odor stimulation protocols , the data comprised stimulation patterns on diverse timescales and with varying stimulus durations: 10 linear and nonlinear ramps lasting less than 25 s and one ‘naturalistic’ stimulus lasting more than 200 s . We found that the parameter set leading to a good fit during the first 30 s of the light or odor stimulation did not yield an accurate fit for longer durations . By fitting the activity at the beginning and the end of the naturalistic stimulation , we discovered that the discrepancy between both time ranges was mainly due to a change in the threshold θ of the Hill term in the time derivative of y ( Equation 2 ) . We therefore allowed the threshold θ to change smoothly between the two different time ranges with the functional expression: θ′ = θ× ( τ/t ) 2 for t > τ with τ = 30 s . A third set of measurements of the firing rate at steady state ( time interval 20–24 s in Figure 3—figure supplement 1 ) was used as an independent control of the parameter fit obtained from the fitting of the other stimulation protocols . As observed in Figure 4E , F , the pure IFF motif not only accounts for the response of the OSN stimulated by light , but it also represents a good approximation of the OSN dynamics stimulated by an odor ( Figure 4E , F and Table 2 ) . The general solution of Equation 2 is:u ( t ) =α1e−α2t∫oteα2t′x ( t′ ) dt′+Cstee−α2t . For times t larger than α2 , the second term of the solution converges to zero , and we obtain the more compact form: ( 4 ) u ( t ) =α1∫ote−α2 ( t−t′ ) x ( t′ ) dt′ . By sequentially integrating ( 4 ) by parts , we obtain the following identities: ( 5 ) u ( t ) =α1〈x ( t ) 〉α2=α1α2x ( t ) −α1α2〈dxdt ( t ) 〉α2 . where the brackets < > denote the convolution introduced in relationship ( 4 ) . While timescale of the dynamics of variable u is given by α2 , the time scale of the stimulus can be approximated asτx≃xmax−xmin ( dxdt ) max . For the linear and nonlinear ramps tested in Figure 3 , τx is typically 10 s . As the value of α2 is 0 . 88 s–1 ( Table 1 ) , the variable u evolves on a timescale approximately 10 times faster than the stimulus . Using relationship ( 4 ) , Equation 2 can be rewritten as: ( 6 ) dydt=β1xβ2+x+β3α1∫e−α2 ( t−t′ ) x ( t′ ) dt′−β4y2y2+θ2−β5y . The function multiplying parameter β4 is a steep sigmoid ( or Hill function ) whose value is close to 1 when y is reasonably larger than 0 . More formally , this approximation is valid for values of y larger than the threshold y˜: ( 7 ) y~2y~2+θ2= ( 1−ε ) →y~= ( 1−ε ) εθ . Given that the value of θ of 0 . 3 , we see that the Hill term will be larger than 0 . 95 for values of y larger than 1 . 3 Hz . For this range of values , Equation 6 can be rewritten as: ( 8 ) dydt=f ( x ( t ) , t ) −β5y , where f ( t ) is a function independent of y . This function evolves on a timescale slower than α2 . If we assume that the firing rate y ( 0 ) is initially 0 , the solution of ( 8 ) is: ( 9 ) y ( t ) =∫ote−β5 ( t−t′ ) f ( x ( t′ ) , t′ ) dt′ . Equation 9 shows that the dynamics of y ( t ) obeys a characteristic time given by β5 . Since β5 = 13 . 03 s−1 ( Table 1 ) , variable y evolves on a timescale more than 10 times faster than the stimulus . In view of this separation of the timescales , it is justified to assume that y is at quasi-steady-state ( QSSA , dy/dt≃0 ) during the evolution of the stimulus x and variable u . By combining this assumption with ( 6 ) , we find that: ( 10 ) yQSSA ( t ) =β1β5 ( x ( t ) β2+x ( t ) +α1β3∫e−α2 ( t−t′ ) x ( t′ ) dt′ ) −β4β5 , Based on the values of the parameters of the original ODE system ( Table 1 ) , we obtain β1/β5 = 132 . 9 Hz , β2 = 1 . 27 W/m2 , α1β3 = 0 . 25 s−1 , and β4/β5 = 93 . 17 Hz . It is interesting to note that the convolution of x is necessarily smaller when x takes on larger values . Using the identities ( 5 ) , we can rewrite ( 10 ) as: ( 11 ) yQSSA ( t ) =δ1 ( x ( t ) δ2+x ( t ) −δ3∫e−α2 ( t−t′ ) dxdt′ ( t′ ) dt′ ) −δ4 , whereδ1=β1β5α2 ( α2+α1β3 ) =103 . 66 Hz . δ2=β2α2 ( α2+α1β3 ) =0 . 99 W/m2 . δ3=α1β3 ( α2+α1β3 ) =0 . 22 . δ4=β4β5=93 . 17 Hz . As expected , the QSSA solution is in excellent agreement with the results of the integration of the full ODE system ( Figure 14 ) . Given the values of the parameters ( Table 1 ) , the denominator of relationship ( 11 ) is mostly driven by the stimulus intensity for slowly evolving stimuli . The contribution of the convolution over the first derivative is significant for rapid and large changes of the stimulus intensity . In the limit where the stimulus is constant over time ( dx/dt = 0 ) , the scaling term S ( x , t ) is equal to zero and the QSSA predicts the dose–response function displayed in Figure 3—figure supplement 1C:yQSSA∝xx+δ2−C1 dx/dt=0 , where C1 is a constant . For stimulation patterns in which the intensity changes at a constant rate ( linear ramps , Figure 4E and Figure 4—figure supplement 1A ) , the first derivative is constant with a positive or a negative sign . yQSSA∝xx+δ2−C2−C1 constant dx/dt , with C1 and C2 being constants . The sign of C2 is determined by that of dx/dt . For constant stimulus gradients ( linear increases in concentration ) , we predict that the firing rate also followed a hyperbolic function ( dose–response ) of the stimulus intensity ( Figure 3—figure supplement 1 ) . This time , however , the dose–response is expected to saturate at lower values of the stimulus intensity when the gradient is positive ( rising phase of the linear ramps ) and higher values of the stimulus intensity when the gradient is negative ( falling phase of the linear ramps ) . These predictions are consistent with the firing patterns observed in Figure 4 . 10 . 7554/eLife . 06694 . 036Figure 14 . Quasi-steady state approximation of the IFF model describing the spiking dynamics of a single OSN stimulated by light ramps . Comparisons of the analytical solution of the OSN activity upon numerical integration of the full ODE system ( green line ) with the solution obtained under the quasi-steady state approximation ( QSSA , orange line ) and the experimental PSTH ( black line , shades denote standard deviation ) . The goodness of fit of the QSSA is remarkable for the linear and nonlinear ramps . As discussed in the ‘Materials and methods’ , the QSSA holds for values of y such that the Hill term can be linearized ( y >1 . 3 Hz ) . This domain of validity is depicted by the orange background . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 036 The tracker outlined in Figure 5 is presented in more detail in Figure 15 . It was conceived and built at the Instrumentation Design and Fabrication Facility at the Janelia Research Campus . Part of the construction was carried out by KeyTech ( Baltimore , USA ) . Unlike other tracking systems ( Faumont et al . , 2011; Leifer et al . , 2011; Kocabas et al . , 2012 ) , our setup was designed to keep the stage on which the larva evolves fixed by mounting the camera and stimulation LEDs on a moving stage whose position was continuously updated to stay locked with the animal's position ( Figure 15 ) . The upper and lower moving stage were powered by a pair of stepper motors ( T-LSR450B , Zaber Technologies , Canada ) . The blueprint of the tracker and list of parts are available from the following link: https://github . com/LabLouis/eLife_2015/tree/master/Tracker%20Hardware . The light stimulation module consisted of three LEDs , connected in series ( LCS-0470-03-22 , Mightex Systems ) to an LED controller ( SLA-1200-2 , Mightex Systems ) whose output current limit was set to 750 mA . The angle and position of each LED was fixed to cover the camera's field of view with maximum light intensity . The controller's output current scaled proportionally to the analog voltage fed into the controller board . The light intensity reaching the arena was estimated by measuring the current emitted by a photodiode ( SM05PD7A , Thorlabs ) connected to a benchtop amplifier ( PDA200C , Thorlabs ) . The tracker's video camera ( A622f , Basler , Germany ) was placed at the center of the three blue LEDs; it delivered images at a resolution of 800 × 800 pixels at a frame time interval of 23 ms . Combined with the time required to process the image and actuate the position of the stage , the effective frame rate was 30 Hz . 10 . 7554/eLife . 06694 . 037Figure 15 . Technical description of the closed-loop tracker for virtual olfactory realities . ( A ) Schematic drawing of the closed-loop tracker . The blue LEDs and the camera are mounted on a moving stage that follows the larva while it crawls on an agarose slab ( 40 × 40 cm or approximately 120 × 120 body lengths of the larva ) . ( B ) Depiction and description of the moving camera stage equipped with three LEDs . ( C ) Flow chart outlining the interaction of the core modules of the tracking software ( ‘Materials and methods’ ) . ( D ) Illustration of the spatial trajectory generated by an Or42a>ChR2 larva undergoing closed-loop light stimulation in a virtual odor gradient . ( Top-left ) Predefined light landscape with a geometry approximating the odor distribution produced by a point source . During the behavioral tests , the full gradient is not projected onto the arena: the larva is illuminated by the LEDs at an intensity determined by its position in the virtual light gradient . ( Bottom ) The light intensity is updated based on the motion of the larva , which forms the temporal evolution shown in the graph . ( Top-right ) The spatial trajectory described by an Or42a>ChR2 larva in the virtual light gradient . The orientation response faithfully reproduces chemotactic behavior . In closed-loop experiments , the LED intensity was updated according to the position of the head with respect to a predefined spatial landscape . In open-loop experiments , the LED intensity was determined by a predefined temporal profile implemented only when a larva was in a run mode ( Figure 5B ) . During a run , the motion of the larva had no influence on the intensity of the stimulus . As soon as the larva interrupted a run , the light intensity returned to a baseline value ( 15 W/m2 ) . ( E ) Abolishment of photophobic behavior in blind larvae . Stimulation of larvae with light flashes of 0 . 5 s ( intensity: 207 W/m2 ) . The flashes were interspaced by a time interval picked from a Poisson distribution with mean 7 . 7 s . To ensure that a larva had sufficient time to react to individual light pulses , the minimum inter-flash interval was set to 5 s . The behavior resulting from the flash was characterized by computing the flash-triggered averages of the amplitude of the absolute head angle and its time derivative . Wild-type larvae ( black trace ) display an increase in head motion following the flash ( release of head cast ) . Blind GMR-hid;dTrpA11 larvae ( red trace ) were not affected by the light flashes . The graph represents the means of the kinematic variables computed across trajectories; error bars denote SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 06694 . 037 As summarized in Figure 15C , the larval tracker control unit ( LTCU ) formed the main hardware interface controlling the LEDs and acquiring images from the video camera . While the camera was controlled via a transistor–transistor logic pulse signal , a 12-bit digital-to-analog converter output was used to send the control signals ( 0–5 V analog signal ) to the LED controller . An in-circuit debugging ( ICD ) Port was used to connect the LTCU to an ICD3 programmable interface controller via a registered jack ( RJ11 ) connector interface , while an USB Port enabled the LTCU's communication to the PC . A customized program written in C was used to direct the function of the LTCU . This C program enabled the LTCU to respond to the commands issued by the PC . The software interface of the tracking and image analysis software interfacing the LTCU was written in JAVA . Both the C program and JAVA interface are available from the following links: https://github . com/LabLouis/eLife_2015/tree/master/Wormsign ( tracker ) and https://github . com/LabLouis/eLife_2015/tree/master/Venkman ( JAVA controller interface ) . In the behavioral experiments of Figure 5 , the probability of turning ( ‘turn rate’ ) was estimated from the relative number of turns that took place during a 1-s time window centered on the time point of interest . If we denote the number of runs observed at the beginning and the end of the ith time window as Nbi and Nei , respectively , Nb1 represents the total number of runs contained in the dataset . Over time , Nbi decreases monotonically to 0 . The turn probability associated with the ith time point was estimated as the fraction of runs that ended during the corresponding time window: ( Nbi − Nei ) /Nbi . A sliding time window was then applied to estimate the turn probability corresponding to every time point of the experiment . The turn probability was thus defined as the likelihood of implementing a turn within a 1 s time window , resulting in values smaller than 1 ( in the case of 1 all runs entering a given time window switch to a turn ) and larger or equal to 0 ( in the case of 0 no turn takes place during a given time window ) . While a short time window led to noisy turn probability estimates , long time windows led to undesirable averaging effects . We empirically found that a time window of 1 s ( or 30 time points ) offered a good tradeoff . This section summarizes the methods used for the quantification of physiological and behavioral data . For optimal results , the spike-sorting method ( OpSIN , Figure 11B ) developed in this work required that the recordings of the odor-stimulated OSN activity be achieved in the background of a silenced olfactory system ( Orco−/− ) . For this reason , the genotype of the larvae used throughout the study for physiological and behavioral quantification resulted from the cross: w;Or42a-Gal4 , GMR-hid;Orco2 , dTrpA11 × w;UAS-Orco , UAS-ChR2-H134R;Orco2 , dTrpA11 . Given that interneurons of the larval antennal lobe clearly contribute to the neural representation of odors ( Asahina et al . , 2009; Larkin et al . , 2010 ) , it was important to determine whether the sensorimotor principles controlling chemotactic behavior in a silenced and intact olfactory system are the same . In this aim , we examined the behavior of larvae with ChR2 expressed in the Or42a OSN in the background of a fully functional olfactory system ( 21 intact OSNs ) : UAS-ChR2-H134R/Or42a-Gal4;gl60j , dTrpA11 . Larvae bearing the double mutant alleles gl60j and dTrpA11 were insensitive to light ( data not shown ) . In closed-loop and open-loop conditions of light stimulation , we found a high similarity of the chemotactic responses observed in the background of 1 or 21 functional OSNs ( Figure 18 ) . This result corroborates the idea that the basic control principles learned from single functional OSN larvae are relevant to wild-type larvae . All scripts described in the ‘Materials and methods’ as well as the blueprint of the larval tracker are available for download from the following website: https://github . com/LabLouis/eLife_2015 .	Fruit flies are attracted to the smell of rotting fruit , and use it to guide them to nearby food sources . However , this task is made more challenging by the fact that the distribution of scent or odor molecules in the air is constantly changing . Fruit flies therefore need to cope with , and exploit , this variation if they are to use odors as cues . Odor molecules bind to receptors on the surface of nerve cells called olfactory sensory neurons , and trigger nerve impulses that travel along these cells . While many studies have investigated how fruit flies can distinguish between different odors , less is known about how animals can use variation in the strength of an odor to guide them towards its source . Optogenetics is a technique that allows neuroscientists to control the activities of individual nerve cells , simply by shining light on to them . Because fruit fly larvae are almost transparent , optogenetics can be used on freely moving animals . Now , Schulze , Gomez-Marin et al . have used optogenetics in these larvae to trigger patterns of activity in individual olfactory sensory neurons that mimic the activity patterns elicited by real odors . These virtual realities were then used to study , in detail , some of the principles that control the sensory navigation of a larva—as it moves using a series of forward ‘runs’ and direction-changing ‘turns’ . Olfactory sensory neurons responded most strongly whenever light levels changed rapidly in strength ( which simulated a rapid change in odor concentration ) . On the other hand , these neurons showed relatively little response to constant light levels ( i . e . , constant odors ) . This indicates that the activity of olfactory sensory neurons typically represents the rate of change in the concentration of an odor . An independent study by Kim et al . found that olfactory sensory neurons in adult fruit flies also respond in a similar way . Schulze , Gomez-Marin et al . went on to show that the signals processed by a single type of olfactory sensory neuron could be used to predict a larva's behavior . Larvae tended to turn less when their olfactory sensory neurons were highly active . Low levels and inhibition of activity in the olfactory sensory neurons had the opposite effect; this promoted turning . It remains to be determined how this relatively simple control principle is implemented by the neural circuits that connect sensory neurons to the parts of a larva's nervous system that are involved with movement .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "neuroscience" ]	2015	Dynamical feature extraction at the sensory periphery guides chemotaxis
The maternal-to-zygotic transition ( MZT ) is a conserved step in animal development , where control is passed from the maternal to the zygotic genome . Although the MZT is typically considered from its impact on the transcriptome , we previously found that three maternally deposited Drosophila RNA-binding proteins ( ME31B , Trailer Hitch [TRAL] , and Cup ) are also cleared during the MZT by unknown mechanisms . Here , we show that these proteins are degraded by the ubiquitin-proteasome system . Marie Kondo , an E2 conjugating enzyme , and the E3 CTLH ligase are required for the destruction of ME31B , TRAL , and Cup . Structure modeling of the Drosophila CTLH complex suggests that substrate recognition is different than orthologous complexes . Despite occurring hours earlier , egg activation mediates clearance of these proteins through the Pan Gu kinase , which stimulates translation of Kdo mRNA . Clearance of the maternal protein dowry thus appears to be a coordinated , but as-yet underappreciated , aspect of the MZT . Proper embryogenesis is critical for animal development . Many of the earliest events occur prior to the onset of zygotic transcription , and they are instead directed by maternally deposited proteins and messenger RNAs ( mRNAs ) . During the maternal-to-zygotic transition ( MZT ) , genetic control of developmental events changes from these maternally loaded gene products to newly made zygotic ones ( Vastenhouw et al . , 2019 ) . Thus , the MZT requires both the activation of zygotic transcription and clearance of maternal transcripts . Failure to mediate either of these processes is lethal for the embryo ( Benoit et al . , 2009; Liang et al . , 2008 ) . In contrast to our understanding of the transcriptome during the MZT , much less is known about changes in the proteome . Despite the fact that the maternal dowry of proteins plays key roles in embryogenesis , there are only a handful of examples of cleared maternal proteins ( Guven-Ozkan et al . , 2008; Hara et al . , 2017; Pesin and Orr-Weaver , 2007; Sysoev et al . , 2016; Wang et al . , 2017; Yang et al . , 2016 ) . Recently , we found that three RNA-binding proteins ( ME31B , Trailer Hitch [TRAL] , and Cup ) are rapidly degraded during the MZT in Drosophila melanogaster , at a time point coinciding with the major wave of zygotic transcription ( Wang et al . , 2017 ) . ME31B , TRAL , and Cup form a complex that blocks translation initiation ( Kinkelin et al . , 2012; Nakamura et al . , 2004; Nelson et al . , 2004; Wilhelm et al . , 2003 ) . All three proteins are required for oogenesis , and they appear to bind and repress thousands of deposited maternal mRNAs ( Keyes and Spradling , 1997; Nakamura et al . , 2001; Tritschler et al . , 2008; Wang et al . , 2017; Wilhelm et al . , 2003 ) . The degradation of ME31B , TRAL , and Cup coincides with many of the hallmarks of the MZT , but explorations into this issue have been hindered by a lack of understanding of how their destruction is controlled . We previously made an intriguing observation that genetically linked the clearance of ME31B , TRAL , and Cup , to the Pan Gu ( PNG ) kinase ( Wang et al . , 2017 ) . Composed of three subunits ( PNG , Giant Nuclei [GNU] , and Plutonium [PLU] ) , the PNG kinase is central to the oocyte-to-egg transition and mediates key aspects of embryogenesis , including resumption of the cell cycle , zygotic transcription , and maternal mRNA clearance ( Elfring et al . , 1997; Tadros et al . , 2007; Vardy and Orr-Weaver , 2007 ) . Unlike many animals , the oocyte-to-egg transition in Drosophila does not require fertilization but is instead triggered by egg activation ( Doane , 1960; Heifetz et al . , 2001; Horner and Wolfner , 2008a ) . Here , the PNG kinase is activated by mechanical stress as the oocyte passes through the oviduct , and then phosphorylation and degradation of the GNU subunit quickly inactivates the kinase , restricting its activity to the first half hour after egg activation ( Hara et al . , 2017 ) . One way that PNG mediates the oocyte-to-embryo transition is by rewiring post-transcriptional gene regulation ( Eichhorn et al . , 2016; Kronja et al . , 2014 ) . Possibly by phosphorylating key RNA-binding proteins such as Pumilio , PNG activity leads to changes in the poly ( A ) -tail length and translation of thousands of transcripts during egg activation ( Hara et al . , 2018 ) . Importantly , two targets induced by PNG activity are the pioneer transcription factor Zelda , which is responsible for initial zygotic transcription , and the RNA-binding protein Smaug , which is responsible for clearance of many maternal transcripts ( Benoit et al . , 2009; Eichhorn et al . , 2016; Liang et al . , 2008; Tadros et al . , 2007; Vardy and Orr-Weaver , 2007 ) . The PNG kinase also phosphorylates ME31B , Cup , and TRAL ( Hara et al . , 2018 ) , but it is unclear what effect phosphorylation has on these proteins . One possibility has been that PNG phosphorylation could lead to the degradation of ME31B , TRAL , and Cup , but this model has been thus far unexplored . The ubiquitin-proteasome system is a major protein degradation pathway . Here , a series of ubiquitin activating enzymes , conjugating enzymes , and ligases ( E1 , E2 , and E3 , respectively ) lead to the post-translational addition of a polyubiquitin chain on a target protein , which then serves as a molecular beacon for degradation by the proteasome . E3 ligases are typically thought to recognize target proteins , while E2 conjugating enzymes provide the activated ubiquitin and in turn recognize the E3 ligase ( Komander and Rape , 2012 ) . There are hundreds of different E3 ligases and 29 annotated E2 conjugating enzymes in Drosophila ( Du et al . , 2011 ) , but most of the client substrates are unknown , and few have been implicated in the MZT . Given the key roles of ME31B , Cup , and TRAL in oogenesis and embryogenesis , we wanted to understand the mechanisms controlling their degradation . In particular , we sought to answer how PNG activity at egg activation leads to the degradation of these three RNA-binding proteins several hours later , and how their degradation is coordinated with other elements of the MZT , including zygotic transcription and maternal mRNA clearance . To answer these questions , we performed a selective RNAi screen in Drosophila , and identified the E2 conjugating enzyme as UBC-E2H/Marie Kondo and the E3 ligase as the CTLH complex . Interestingly , structural models based on the S . cerevisiae complex ( Qiao et al . , 2020 ) suggest that the Drosophila version is organized differently than its orthologous complexes . The CTLH complex recognized and bound ME31B and Cup even in the absence of PNG activity , strongly suggesting that phosphorylation is not required for the destruction of these proteins . In contrast , Kdo mRNA is translationally upregulated by more than 20-fold upon egg activation in a PNG-dependent manner . Thus , egg activation through PNG mediates translation upregulation of Kdo and so leads to ME31B , Cup , and TRAL destruction . We previously demonstrated by western blotting that ME31B , TRAL , and Cup were degraded 2–3 hr after egg laying ( Wang et al . , 2017 ) . To understand the mechanisms underlying degradation of these RNA-binding proteins , we decided to establish a fluorescence-based assay so that we could follow ME31B degradation in living embryos . To do so , we took advantage of an ME31B-GFP trap line where the fusion protein is expressed from the endogenous locus ( Buszczak et al . , 2007 ) ; we have previously shown that ME31B-GFP recapitulates the dynamics of the wild-type protein ( Wang et al . , 2017 ) . Consistent with western blotting , the GFP signal in control ( png50/FM7 ) embryos robustly decreased from 2 to 3 hr after egg laying ( Figure 1A , B ) . In contrast , the GFP signal remained constant through this time period in png50/png50 embryos ( hereafter referred to as png50 ) , consistent with our previous study ( Wang et al . , 2017 ) . Note that less heterogeneity in fluorescence was also observed in the mutant embryos; this observation likely stems from the fact that in wild-type embryos some ME31B-GFP degradation occurs during embryo collection and staging , but almost none occurs in png50 embryos . Together , these results confirm that the differences in ME31B-GFP dynamics are observable by microscopy and that the degradation of ME31B-GFP requires PNG . To test the importance of other genes for ME31B degradation , we combined the ME31B-GFP allele with a GAL4-UAS system , where GAL4 was under the control of the matα-tubulin promoter and so is specifically expressed during oogenesis; this system enabled us to induce the expression of dsRNA during oogenesis and monitor the requirement of various genes for ME31B-GFP degradation . Given that png50 embryos did not degrade ME31B-GFP , we first investigated whether the other two components of the PNG kinase , GNU and PLU , were required . When either GNU or PLU were knocked down ( Figure 1C , D ) , ME31B-GFP was again stabilized , thus confirming that its destruction requires the full PNG kinase . We next asked whether the degradation of ME31B-GFP required fertilization by following ME31B-GFP levels with fluorescence microscopy in unfertilized eggs ( Figure 1E ) . In contrast to our results in png50 embryos , ME31B-GFP was still unstable in the activated , but unfertilized , eggs . This result is consistent with numerous studies demonstrating that the major events pre-MZT in Drosophila require egg activation ( primarily through the PNG kinase ) , but not fertilization ( Fenger et al . , 2000; Horner and Wolfner , 2008b; Tadros et al . , 2007; Tadros et al . , 2003; Vardy and Orr-Weaver , 2007 ) . Taken together , we conclude that degradation of ME31B is triggered by egg activation through PNG activity . We hypothesized that ME31B degradation involved ubiquitination . To test this model , we immunoprecipitated ME31B-GFP from 1 to 2 hr embryo lysates under stringent conditions that disrupted protein-protein interactions , such as that with eIF4E . An ubiquitin smear was detected in immunoprecipitants by western blotting at a size consistent with polyubiquitinated ME31B-GFP ( Figure 2A ) . We also detected ubiquitin by mass spectrometry of ME31B-GFP pull-downs ( see below ) . This ubiquitination was not detectable in png50 mutant embryos and thus depended upon the PNG kinase activity ( Figure 2A ) . We next asked whether ME31B-GFP is degraded by autophagy or by the proteasome . To do so , we depleted components of either system using the GAL4-UAS system described above . Knockdown of five autophagy components , such as Atg4a or Atg8a , gave viable embryos . However , ME31B-GFP was not stabilized in any of these knockdown embryos; indeed , in some cases , it appeared to be degraded more quickly than in the control embryos ( Figure 2B , Figure 2—source data 1 ) . Analysis of embryos depleted of core barrel proteasome proteins proved more challenging because knockdown of most components , such as Prosα5 and Prosα7 , resulted in females that did not lay eggs ( Figure 2—source data 1 ) . We were able to obtain embryos from Rpn10 and Rpn11 knockdowns , two components of the regulatory particle of the proteasome , perhaps because these embryos only had partial inhibition of proteasome function or there is functional redundancy . Importantly , we observed partial stabilization of ME31B-GFP in both knockdown embryos ( Figure 2C–E ) . The role of the proteasome in degrading ME31B is consistent with results from a complementary study where injection of MG132 into embryos stabilized endogenous ME31B , TRAL , and Cup ( Cao et al . , 2019 ) . Thus , taken together , these data suggest that the ubiquitin-proteasome system degrades ME31B . We thus set out to identify E2 conjugating enzymes and E3 ligases responsible for the degradation of ME31B by carrying out a medium-scale RNAi screen . As before , we monitored ME31B-GFP decay by GFP fluorescence , taking images every 30 min after egg laying . We focused on those proteins that: ( 1 ) had evidence of expression , based on RNA-seq or mass spectrometry data , and ( 2 ) had available RNAi lines . We screened 137 RNAi lines , targeting E3 ligases as well as related factors ( Figure 2—source data 1 ) . Note that RNAi lines knocking down many cullins and proteasomal components did not lay eggs , presumably because of critical functions during oogenesis . Because we did not measure mRNA or protein levels in the screen or in subsequent experiments ( due to COVID-19 restrictions ) , we do not know the efficiency of RNAi knock-down . Although our initial E3 screen failed to reveal any strong candidates , knockdown of UBC-E2H , an E2 ligase conserved from yeast to humans ( Kaiser et al . , 1995; Kaiser et al . , 1994; Lampert et al . , 2018 ) , blocked degradation of ME31B-GFP and nearly phenocopied the dynamics seen in png50 mutants ( Figure 3A ) . To test this result , we raised an antibody against UBC-E2H and confirmed that the protein was depleted in the knockdown embryos ( Figure 3B ) . We next isolated lysates from staged embryos and performed western blotting , probing for ME31B . Because the maternal line contains both the wild-type and trap ME31B alleles , this experiment revealed that both wild-type and GFP fusion proteins were stabilized when UBC-E2H was depleted ( Figure 3C ) , albeit more so for the fusion protein than wild-type one . Importantly , as determined by western blotting , endogenous , untagged Cup and TRAL were also stabilized in the UBC-E2H knockdown embryos ( Figure 3C ) . Due to its role in removing proteins given in the maternal dowry , we renamed UBC-E2H as ‘Marie Kondo’ ( shortened to ‘Kdo’ ) . Finally , through immunoprecipitation experiments , we determined that ubiquitination of ME31B-GFP was undetectable in Kdo knockdown embryos ( Figure 3D ) . Thus , we conclude that Kdo is required for the destruction of ME31B , TRAL , and Cup during the MZT . Kdo is conserved from yeast to humans and is known to work through the CTLH E3 ligase , a multicomponent complex ( Lampert et al . , 2018; Santt et al . , 2008 ) . ( Note that the S . cerevisiae complex is called the Gid complex . ) Using BLAST for the human CTLH components , we were easily able to identify putative D . melanogaster homologs: RanBPM ( homologous to Hs RanBP9 ) , Muskelin , CG3295 ( homologous to Hs RMND5A/GID2 ) , CG7611 ( homologous to Hs WDR26 ) , CG6617 ( homologous to Hs TWA1/GID8 ) , and CG31357 ( homologous to Hs MAEA ) ( Figure 4A ) . We were unable to find putative homologs for Hs GID5/ARMC8 or Hs GID4 ( see below ) . Notably , none of these genes were annotated as putative E3 components in FlyBase , and thus none were included in our original screen . To ask if the CTLH complex might be involved in the degradation of ME31B , we immunoprecipitated ME31B-GFP and Cup-GFP from pre-MZT embryos ( in conditions that maintain complex formation ) and determined the proteins bound by mass spectrometry ( Figure 4B and C; Figure 4—source datas 1 and 2 ) . Consistent with our previous work ( Wang et al . , 2017 ) , both immunoprecipitations readily pulled-down other members of the Cup–TRAL–ME31B complex . We were able to identify Muskelin , RanBPM , and CG6617 in all four samples . We also detected CG3295 in both ME31B-GFP pull-downs and one Cup-GFP pull-down , and CG31357 in both ME31B-GFP pull-downs . Similar results were seen in previous studies of ME31B complexes in embryonic lysates ( Götze et al . , 2017 ) and in a complimentary study ( Cao et al . , 2019 ) . We were unable to detect CG7611 in any of our samples . We next asked whether destruction of ME31B requires the CTLH E3 ligase . As before , we knocked down various components using available RNAi lines ( CG3295 , CG7611 , Muskelin , and RanBPM ) , and monitored levels of ME31B-GFP by fluorescence microscopy ( Figure 4D ) . In contrast to what we had observed with other E3 ligases ( Figure 2—source data 1 ) , depletion of CG3295 , Muskelin , or RanBPM almost completely stabilized ME31B-GFP . RNAi directed against CG7611 , the one component that we failed to detect by mass spectrometry , did not affect the destruction of ME31B , although we cannot exclude the possibility that CG7611 protein levels were not sufficiently affected . Consistent with these results , when we used western blotting to look at levels of ME31B , Cup and TRAL , we found that all were stabilized when the CTLH complex was depleted ( Figure 4E ) . One trivial explanation for these results is that depletion of the CTLH complex inadvertently reduced levels of Kdo , which is required for the destruction of ME31B ( Figure 3 ) . However , as determined by western blotting , Kdo levels were unaffected in these knockdown embryos ( Figure 4—figure supplement 1A ) . Because antibodies were only available for RanBPM ( Dansereau and Lasko , 2008 ) , we were unable to generally determine how depletion of individual components affected levels of the other components . Nonetheless , in analyzing RanBPM levels , we found that the RanBPM RNAi line was depleted for the protein ( Figure 4—figure supplement 1B ) . We next asked whether we could detect an interaction between ME31B and the CTLH complex using immunoprecipitation followed by western blotting . Consistent with our mass spectrometry analysis ( Figure 4B ) , we detected RanBPM in ME31B-GFP immunoprecipitations ( Figure 4F ) . Because we lack antibodies for other components , we were unable to probe interactions between ME31B and Muskelin , CG3295 , and CG6617 by western blotting . As expected , when we performed control experiments in wild-type embryos , RanBPM was not immunoprecipitated ( Figure 4F ) . Consistent with this result , we were able to immunoprecipitate RanBPM using antibodies recognizing endogenous Cup protein ( see below ) . Taken together , we conclude that the CTLH E3 ligase is required for the destruction of ME31B , Cup , and TRAL , and , in the early Drosophila embryo , is at least composed of RanBPM , Muskelin , CG6617 , CG3295 , and CG31357 , although the role of CG7611 remains unknown . Because of their roles in clearing proteins , we also now refer to CG6617 as Houki ( Hou , Japanese for ‘broom’ ) , CG3295 as Souji ( Sou , Japanese for ‘cleaning’ ) , and CG31357 as Katazuke ( Kaz , Japanese for ‘tidying up’ ) . We next wanted to understand the organization of the Drosophila CTLH complex . Serendipitously , a cryoEM structure of the yeast Gid complex was recently published ( Qiao et al . , 2020 ) . The Gid complex is composed of three sections: a catalytic module made by Gid2 and Gid9; a scaffold of Gid8 , Gid1 , and Gid5; and a substrate adaptor module formed by Gid4 ( Figure 5A ) . By analogy , we were able to assign roles to the known Drosophila components: Kaz and Sou likely form the catalytic module , while RanBPM and Hou are part of the scaffold domain ( Figure 5A ) . Consistent with such organization , when Sou was depleted , the interaction between RanBPM and ME31B-GFP was unaffected ( Figure 5B ) . The interaction between RanBPM and ME31B-GFP did , however , depend upon Muskelin ( Figure 5B ) , but with current structures , it is unclear how Muskelin interacts with the other components of the Drosophila CTLH complex . In considering the organization of the Drosophila CTLH complex , we were surprised by the lack of any putative Gid5 or Gid4 , which are important for the scaffold and substrate recognition , respectively . In yeast , Gid4 is exchangeable with other substrate adaptors , such as Gid10 . Because the substrate adaptor module interacts with the core of the Gid complex predominantly through Gid5 , we focused on identifying a Drosophila Gid5 ortholog . However , initial BLAST searches with both S . cerevisiae Gid5 and H . sapiens ARMC8 failed to identify a putative ortholog , and so we turned Phyre2 to conduct a ‘BackPhyre’ structure-homology-based search for a Drosophila Gid5 with the search model provided by the recent structure of the S . cerevisiae Gid complex ( Kelley et al . , 2015; Qiao et al . , 2020 ) . Although Drosophila proteins were identified that contained the armadillo domains ( which is the major fold in Gid5 ) , none of these were convincing hits to Gid5 overall , predicted to interact with Gid8 , or identified in our mass spectrometry data ( Figure 4—source datas 1 and 2; Figure 5—source data 1 ) . In contrast , the same search performed against the human genome easily identified ARMC8 , which was predicted to interact with Gid8 ( Figure 5—source data 1 ) . Prompted by our continued inability to identify a Drosophila Gid5 ortholog , we threaded RanBPM and Hou into the S . cerevisiae structure so that we could examine the predicted Gid5 interface . These proteins broadly shared predicted structures with their yeast counterparts ( Figure 5C ) . Despite overall predicted similarities with the yeast structure , we found two differences at the predicted interface between RanBPM–Hou and Gid5–Gid4 ( Figure 5D , E ) . First , in S . cerevisiae , Gid4 makes the majority of its contacts only with Gid5 . Nonetheless , one loop in Gid1 extends out to interact with Gid4 ( Qiao et al . , 2020 ) . In contrast , no such loop is predicted in Drosophila RanBPM , although the flanking β sheet strands appear to exist ( Figure 5D , Figure 5—figure supplement 1A ) . Second , in S . cerevisiae , the scaffold module is composed of Gid1 , Gid5 , and Gid8 . Here , most of the Gid8 C-terminus wraps around Gid5 and makes nearly all of the interactions with Gid5 . This domain appears to be also absent from Drosophila Hou , despite structural similarity in the rest of the protein ( Figure 5E; Figure 5—figure supplement 1B ) . The consequence of Drosophila Hou lacking this domain is that there is little predicted interaction between Hou and Gid5 . Taken together , these analyses suggest that substrate recognition is likely different for the Drosophila CTLH complex than in other organisms and leaves open the question of how ME31B , Cup , and TRAL bind the E3 ligase . Having identified the E3 ligase and E2 conjugating enzyme , we next turned to understanding how the destruction of ME31B was triggered by PNG activity . Because recent work has demonstrated that the PNG kinase phosphorylates ME31B , TRAL , and Cup ( Hara et al . , 2018 ) , we explored the idea that this phosphorylation might stimulate an association between ME31B and the E3 ligase . Consistent with a role for the E3 ligase binding a target protein , interaction between RanBPM and ME31B-GFP was unaffected by Kdo depletion ( Figure 6A ) . However , the interaction between RanBPM and ME31B-GFP remained robust in png50 embryos ( Figure 6B ) . Similarly , when we immunoprecipitated endogenous Cup , we were able to detect an interaction with RanBPM in both wild-type and png50 embryos ( Figure 6C ) . Thus , we conclude that PNG activity is not required for the CTLH complex to recognize and interact with ME31B and Cup . Given that PNG phosphorylation of ME31B could not explain how egg activation stimulated its association with the CTLH complex , we searched for alternative explanations , focusing on recent observations that PNG also mediates the translational upregulation of thousands of mRNAs at the oocyte-to-embryo transition ( Eichhorn et al . , 2016 ) . We analyzed published ribosome profiling datasets ( Eichhorn et al . , 2016 ) for evidence of translational upregulation of CTLH components and Kdo mRNAs upon the oocyte-to-embryo transition . Known CTLH component mRNAs were either not affected or downregulated during egg activation ( Figure 7—figure supplement 1 ) , although we cannot exclude the hypothesis that unidentified components may be regulated by PNG . In contrast , the most striking change occurred for translation of Kdo mRNA: although translation of Kdo mRNA was repressed through oogenesis , its translation increased 25-fold during the oocyte-to-embryo transition ( Figure 7A ) , placing it in the top 10% of genes upregulated at this developmental transition . However , Kdo was not translationally upregulated in png50-activated embryos ( Figure 7B ) , and its translation differed by more than 200-fold between wild-type and mutant-activated eggs . In fact , Kdo was the seventh-most affected transcript ( Figure 7C ) , showing a similar dependence on PNG as Smaug , which encodes a well-known and important downstream target of PNG activity . Consistent with this analysis , when we probed Kdo protein levels , we were able to detect a dramatic increase in protein levels over the first 3 hr of embryogenesis in wild-type embryos , but we were unable to detect expression at any time point in png50 mutant embryos ( Figure 7D ) . Pointing to a role of post-transcriptional gene regulation , the poly ( A ) tail length on Kdo mRNA also doubled upon egg activation in a manner dependent on PNG . This result suggests that the translational increase is partially mediated by an increase in poly ( A ) -tail length ( Figure 7E , F ) , which directly impacts translational changes during egg activation ( Eichhorn et al . , 2016 ) . We noted , however , that although a change in poly ( A ) -tail length partially explained the changes in translational efficiency ( Figure 7G ) , it did not fully account for all of the translational changes at egg activation . Because we had previously found that ME31B is broadly associated with translational repression in the early embryo ( Wang et al . , 2017 ) , we wondered if ME31B binding might also change during egg activation and if such changes could impact translation . To test this possibility , we immunoprecipitated ME31B-GFP complexes from stage 14 oocytes and sequenced bound transcripts , as we have done previously ( Wang et al . , 2017 ) . We normalized bound RNA to total abundance to generate an overall occupancy . Although ME31B-GFP binding in stage 14 oocytes was highly correlated with that measured previously in 0–1 hr wild-type embryos ( Spearman r [rs]=0 . 66 , p<10–15 ) , its binding was even more similar in 0–1 hr png50 embryos ( rs = 0 . 84 , p<10–15; Fisher’s r-to-z transformation: p<10–15; Figure 7H , I ) . These data indicate that although ME31B binds broadly similar transcripts in the oocyte and embryo , its binding does change at egg activation in a manner dependent on the PNG kinase . Moreover , we found that the change in ME31B-GFP binding at egg activation was strongly correlated with the change in translational efficiency ( rs = –0 . 48 , p<10–15 ) such that those mRNAs with diminished ME31B-GFP binding were translationally activated and those with increased binding were translationally repressed ( Figure 7J ) . Importantly , this relationship held even after we controlled for changes in poly ( A ) -tail length ( rs = –0 . 40 , p<10–15 ) . Thus , these analyses indicate that PNG activity at egg activation triggers two independent , albeit related , mechanisms for altering translation: changing poly ( A ) -tail length and altering ME31B binding . Importantly , as with poly ( A ) -tail length , ME31B-GFP binding to Kdo mRNA also changed upon egg activation and was substantially reduced in a PNG-dependent manner ( Figure 7H , I ) . Taken together , these data suggest that , during oogenesis , ME31B , presumably via Cup and TRAL , acts to repress translation of Kdo mRNA and thus suppress production of its own E2 . Upon egg activation , PNG activity not only leads to extension of the Kdo mRNA poly ( A ) tail , but also stimulates the dissociation of ME31B from the transcript . Together , these two activities likely promote translation of Kdo , setting the stage for clearance of deposited ME31B , TRAL , and Cup . ME31B , Cup , and TRAL are RNA-binding proteins that are degraded during the MZT . Despite occurring several hours after egg laying , degradation of these proteins is triggered by egg activation through the activity of the PNG kinase and appears to be mediated by the ubiquitin-proteasome system ( Figures 1 and 2 ) . Through a medium-scale RNAi screen , we identified that the E2 conjugating enzyme Kdo is required for the clearance of ME31B , TRAL , and Cup ( Figure 3 ) . Kdo is conserved from yeast to humans and , as in those systems ( Kaiser et al . , 1994; Lampert et al . , 2018 ) , appears to work with the CTLH complex , which acts as the E3 ligase . Components of the CTLH complex physically interact with ME31B and Cup , and the CTLH complex is also required for the degradation of ME31B , TRAL , and Cup during early embryogenesis ( Figure 4 ) . Structure-based homology suggests that , despite its conservation from yeast to humans , the Drosophila CTLH complex has an unusual architecture , and it remains unclear how it recognizes its substrate ( Figure 5 ) . The association of CTLH with ME31B occurs in the absence of PNG activity , suggesting that , although ME31B ( as well as TRAL and Cup ) are phosphorylated by the kinase , phosphorylation may not be required for their destruction ( Figure 6 ) . Instead , translation of Kdo appears to be suppressed during oogenesis by its short poly ( A ) tail length and binding of ME31B . Its translation is dramatically upregulated at the oocyte-to-embryo transition , in a process that depends on PNG activity ( Figure 7 ) . Together , these data suggest a model that egg activation via the PNG kinase leads to translational activation and production of Kdo , which then allows the CTLH complex to ubiquitinate ME31B , TRAL , and Cup and ultimately leads to their destruction ( Figure 7K ) . Interestingly , based on RNA-seq data from FlyBase ( FlyBase Consortium et al . , 2019 ) , Muskelin shows exquisite tissue-specificity and is only strongly expressed in the ovaries . This observation , together with the translational control of Kdo , may partly explain how ME31B , a ubiquitous protein , is specifically destabilized in the early embryo . Although the CTLH complex is conserved , it has not yet been studied in Drosophila . Our data point to this complex being composed of multiple components ( Muskelin , RanBPM , Houki , Souji , and Katazuke ) , as in other organisms . However , due to a lack of available reagents , we do not know about the stoichiometry of these components , and it remains possible that there are additional , Drosophila-specific components . Nonetheless , so far , the CTLH complex in Drosophila appears different than the human and yeast complexes . Although Gid7 and WDR26 are important in the yeast and human versions , respectively , and we identified a Drosophila ortholog ( CG7611 ) , we found no evidence of its association with ME31B or requirement for ME31B degradation; the role of CG7611 in the Drosophila CTLH complex warrants further investigation . We were also unable to identify orthologs of Gid4 and Gid5 , which are critical for substrate recognition in S . cerevisiae ( Lampert et al . , 2018; Maitland et al . , 2019; Santt et al . , 2008 ) . Intriguingly , the residues and domains important for the Gid1–Gid4 and Gid8–Gid5 interactions in budding yeast appear absent to be in RanBPM and Hou , raising the fundamental question of how the Drosophila CTLH complex recognizes and positions its substrate proteins . Answering this question will require future investigation and may shed light on other proteins targeted by the Drosophila CTLH complex and the extent to which ME31B , a ubiquitously expressed protein , is targeted outside of the MZT . One unexpected result is the role of PNG in mediating the destruction of ME31B , TRAL , and Cup . PNG phosphorylates all three proteins ( Hara et al . , 2018 ) , and so our initial hypothesis was that this modification also stimulated their destruction . However , contrary to our expectations , ME31B and Cup interacted with the CTLH complex even in png50 embryos , demonstrating that phosphorylation by PNG was not required for binding of ME31B and Cup by the E3 ligase . An unresolved question , then , is how PNG phosphorylation affects the activities of ME31B , TRAL , and Cup . Intriguing observations from the Orr-Weaver lab suggest that the modification can impact the ability of these proteins to repress gene expression ( Hara et al . , 2018 ) . It is tempting to speculate that phosphorylation may then contribute to the MZT by modulating the activities of ME31B , TRAL , and Cup , rather than their stability . The link between PNG and the destruction of ME31B , TRAL , and Cup instead appears to be mediated through the translational upregulation of Kdo . Although PNG may contribute through other , as-yet undiscovered , mechanisms as well ( such as phosphorylating unknown CTLH adaptor proteins ) , this link is sufficient to explain the PNG requirement for ME31B degradation: in the absence of Kdo , ME31B is stable during the MZT , and in the absence of PNG , Kdo is not detectably expressed . An important question for the future will be to understand what elements in the Kdo mRNA are responsible for its translational repression during oogenesis . One hint may be that the 3'UTR of Kdo contains several putative Pumilio-binding sites , and translation of Kdo is upregulated in ovaries where Pumilio has been knocked down ( Flora et al . , 2018 ) . Pumilio is also a target of PNG ( Hara et al . , 2018 ) , and so a possible model is that translational repressors , such as Pumilio , are phosphorylated and inactivated at egg activation , leading to the production of Kdo . PNG also mediates the translational upregulation of key MZT effectors: Zelda , the pioneer transcription factor , and Smaug , an RNA-binding protein that targets nearly two-thirds of the maternal transcriptome for degradation ( Chen et al . , 2014; Eichhorn et al . , 2016; Tadros et al . , 2007 ) . Together with our results , a picture is emerging that egg activation stimulates the production of multiple key factors that are important for clearing the maternal RNA and protein dowry and for producing zygotic gene products . Although the MZT has typically been considered from the perspective of RNA , a role for maternal protein clearance is becoming clearer . Over the past few years , the list of proteins degraded during the Drosophila MZT has grown and now includes GNU , Matrimony , Cort , Smaug , ME31B , TRAL , and Cup ( Benoit et al . , 2009; Hara et al . , 2017; Pesin and Orr-Weaver , 2007; Wang et al . , 2017; Whitfield et al . , 2013 ) . Unbiased mass spectrometry experiments also suggest that Wispy and Dhd are also robustly degraded ( Sysoev et al . , 2016 ) . As this list of proteins in Drosophila and other developmental systems increases , a new question is emerging: how many maternally deposited proteins are degraded during the MZT ? Understanding the mechanisms controlling protein degradation during the MZT as well as the impact of removing the maternal protein dowry will be key issues to explore in the future . Fly stocks were maintained in a 25°C incubator with 65% humidity . Male flies from the TRiP stocks were crossed with the ME31B-GFP driver line . Female flies from this cross were collected and crossed with w1118 males . For egg collection , flies were transferred in the morning to egg-collection chambers on apple juice/agar plates . Flies were allowed to lay eggs for 1 hr . Eggs were collected into cell strainers , washed with 1X PBS , dechorionated with 25% bleach , and washed with 1X PBS . Dechorionated eggs were transferred onto a glass slide and covered with halocarbon oil 700 ( Sigma ) . Images were taken on a ZEISS SteREO Discovery . V8 microscope with X-Cite 120Q fluorescence illumination system ( Exelitas Technologies ) . For the RNAi screen , phenotypes were scored qualitatively ( Figure 2—source data 1 ) . Quantitation of images was performed using ImageJ , and the data were subsequently processed in R using in-house scripts . Stage 14 oocytes were isolated from a large-scale Drosophila culture established with w1118 flies and were homogenized in lysis buffer B ( with protease inhibitor cocktail ( BioShop ) and additional freshly added protease inhibitors [100 µM Leupeptin , 100 µM Chymostatin , 4 mM Benzamidine HCl , 3 µM Pepstatin; Sigma] and SUPERase-In RNase inhibitors ) . The homogenized lysates were clarified at 15 , 000 rpm , 4°C for 15 min , and the supernatant was stored at –80°C . Embryos were collected at various time points post-egg laying , dechorionated with bleach , and washed with 0 . 1% Triton X-100 . Embryos were then homogenized in lysis buffer B ( 150 mM KCl , 20 mM HEPES-KOH pH 7 . 4 , 1 mM MgCl2 , 1 mM DTT , complete mini EDTA-free protease inhibitors ) , and were clarified at 15 , 000 rpm , 4°C for 15 min . The supernatant was stored at –80°C . The rabbit anti-Kdo antibody was generated by Pacific Immunology . For western blotting , it was used at 1:10 , 000 . Rat anti-Cup ( a gift of C . Smibert ) was used at 1:5000 . Mouse anti-ME31B antibody ( a gift of K . Nakamura ) was used at 1:5000 . Rabbit anti-TRAL ( a gift of K . Nakamura ) was used at 1:5000 . Rabbit anti-eIF4E ( a gift of E . Izaurralde ) was used at 1:10 , 000 . Rabbit anti-PABP ( a gift of E . Izaurralde ) was used at 1:10 , 000 . Rabbit anti-RanBPM ( a gift of P . Lasko ) was used at 1:10 , 000 . Mouse anti-GFP ( Roche ) was used at 1:1000 . Mouse anti-ubiquitin ( ThermoFisher Scientific ) was used at 1:5000 . For immunoprecipitations to probe for interactions with Cup or ME31B-GFP , pre-made lysates ( described above ) were diluted to 1 . 0 mg/ml , and then incubated with anti-GFP ( Roche ) or rabbit IgG ( Abcam ) for 1 hr , rotating at 4°C . EZView protein G affinity beads ( Sigma ) were washed 3X with lysis buffer , and 25 μl of slurry was added to the lysate-antibody mixture and incubated for 1 hr , rotating at 4°C . Beads were washed three times with lysis buffer and transferred to a new tube . For western blot analysis , the beads were boiled in loading sample buffer and reducing agent , and immunoprecipitates were loaded onto an SDS-PAGE gel . For ME31B and RanBPM , 2 . 4% input and 17% IP were loaded . For immunoprecipitations to test for ubiquitination , pre-made lysates ( from above ) were incubated at 4°C with anti-GFP or rabbit IgG for 1 hr , rotating . EZView protein G affinity beads were instead washed 2X with RIPA buffer , and then 25 μl of the slurry was added to the lysate-antibody mixture . The lysate-antibody mixture with the beads was diluted with RIPA buffer supplemented with 50 μM PR-619 and protease inhibitors , then incubated for another hour at 4°C , rotating . After incubation , the beads were washed three times with supplemented RIPA buffer and transferred to a new tube . Beads were boiled in loading sample buffer and reducing agent , and immunoprecipitates were loaded on an SDS-PAGE gel . To probe for ubiquitin , 3% input and 20% IP were loaded; to probe for GFP , 1% input and 10% IP were loaded; to probe for eIF4E , 3% input and 10% IP were loaded . For RNA immunoprecipitation , the beads were blocked overnight with salmon ssDNA and incubated with anti-GFP antibodies . Stage 14 oocyte lysates were diluted to 1 . 1 mg/ml , and then incubated with conjugated beads for 2–3 hr , rotating at 4°C . Beads were washed six times with the lysis buffer . Immunoprecipitates were sent to SPARC BioCentre ( SickKids ) for LC/MS/MS analysis as described previously ( Wang et al . , 2017 ) . Briefly , immunoprecipitates were reduced with 10 mM DTT and then treated with 10 mM iodoacetemaide . Samples were digested in solution with trypsin . Mass spectrometry was performed on Q Exactive with dynamic exclusion . Peptides were searched with Sequest against the Drosophila uniprot database , and spectral counts were reported . Ubiquitin peptides were manually identified by mapping to ubiquitin-regions of fusion genes . RNA was extracted from immunoprecipitates and input lysate using TRI-reagent . To assess the enrichment of the RNA-immunoprecipitation , a fraction of the RNA was treated with DNase and used for RT-qPCR to ensure enrichment of Act5C transcripts . After verifying the quality of the RNA-immunoprecipitation , the RNA was subjected to Ribo-Zero Gold rRNA depletion according to the manufacturer’s protocol . Libraries were then generated using Illumina’s TruSeq stranded mRNA library preparation kit according to the manufacturer’s protocol and sequenced at The Center for Applied Genetics ( SickKids ) . Libraries were pooled and sequenced on an Illumina HiSeq 2500 by The Centre for Applied Genomics at The Hospital for Sick Children . 50 base-pair single-end reads were demultiplexed and converted to FASTQ format using bcl2fastq2 v2 . 17 ( Illumina ) . Library quality was inspected using FastQC v0 . 11 . 5 ( http://www . bioinformatics . babraham . ac . uk/projects/fastqc/ ) . Reads were trimmed for quality and clipped for Illumina adaptors using TrimmomaticSE version 0 . 36 ( Bolger et al . , 2014 ) . Surviving reads were mapped by STAR 2 . 5 . 2a ( Dobin et al . , 2013 ) to the D . melanogaster genome obtained from UCSC on 7 August 2016 . Genes were quantified using Cufflinks 2 . 2 . 1 ( Trapnell et al . , 2010 ) . Downstream analyses were then performed with R version 3 . 1 . 2 , using in-house scripts . Occupancies were calculated for each gene by dividing the IP FPKM by the input FPKM . When calculating occupancy , all the genes were filtered such that only genes with greater than 0 . 5 FPKM were included in the analysis . High-throughput sequencing data described in this paper are available from the GEO: GSE83616 ( Eichhorn et al . , 2016 ) , GSE98106 ( Wang et al . , 2017 ) , and , for the data prepared in this paper , GSE140436 . RanBPM and Hou were visualized using standard Phyre2 parameters ( Kelley et al . , 2015 ) . Outputs were visualized using Pymol and compared with the Gid complex ( Qiao et al . , 2020 ) .	Bestselling author and organizing consultant Marie Kondo has helped people around the world declutter their homes by getting rid of physical items that do not bring them joy . Keeping the crowded environment inside a living cell organized also requires work and involves removing molecules that are no longer needed . A fertilized egg cell , for example , contains molecules from the mother that regulate the initial stages as it develops into an embryo . Later on , the embryo takes control of its own development by destroying these inherited molecules and switches to making its own instead . This process is called the maternal-to-zygotic transition . The molecules passed from the mother to the egg cell include proteins and messenger RNAs ( molecules that include the coded instructions to make new proteins ) . Previous research has begun to reveal how the embryo destroys the mRNAs it inherits from its mother and how it starts to make its own . Yet almost nothing is known about how an embryo gets rid of its mother’s proteins . To address this question , Zavortink , Rutt , Dzitoyeva et al . used an approach known as an RNA interference screen to identify factors required to destroy three maternal proteins in fruit fly embryos . The experiments helped identify one enzyme that worked together with another larger enzyme complex to destroy the maternal proteins . This enzyme belongs to a class of enzymes known as ubiquitin-conjugating enzymes ( or E2 enzymes ) and it was given the name “Kdo” , short for “Marie Kondo” . Further experiments showed that the mRNAs that code for the Kdo enzyme were present in unfertilized eggs , but in a repressed state that prevented the eggs from making the enzyme . Once an egg started to develop into an embryo , these mRNAs became active and the embryo started to make Kdo enzymes . This led to the three maternal proteins being destroyed during the maternal-to-zygotic transition . These findings reveal a new pathway that regulates the destruction of maternal proteins as the embryo develops . The next challenge will be identifying other maternal proteins that do not “spark joy” and understanding the role their destruction plays in the earliest events of embryonic development .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "developmental", "biology" ]	2020	The E2 Marie Kondo and the CTLH E3 ligase clear deposited RNA binding proteins during the maternal-to-zygotic transition
The gene encoding the Insulin-like Growth Factor 2 mRNA binding protein 2/IMP2 is amplified and overexpressed in many human cancers , accompanied by a poorer prognosis . Mice lacking IMP2 exhibit a longer lifespan and a reduced tumor burden at old age . Herein we show in a diverse array of human cancer cells that IMP2 overexpression stimulates and IMP2 elimination diminishes proliferation by 50–80% . In addition to its known ability to promote the abundance of Insulin-like Growth Factor 2/IGF2 , we find that IMP2 strongly promotes IGF action , by binding and stabilizing the mRNA encoding the DNA binding protein HMGA1 , a known oncogene . HMGA1 suppresses the abundance of IGF binding protein 2/IGFBP2 and Grb14 , inhibitors of IGF action . IMP2 stabilization of HMGA1 mRNA plus IMP2 stimulated IGF2 production synergistically drive cancer cell proliferation and account for IMP2’s tumor promoting action . IMP2’s ability to promote proliferation and IGF action requires IMP2 phosphorylation by mTOR . The insulin-like growth factor 2 ( IGF2 ) mRNA-binding proteins ( IGF2BP1-3 or IMP1-3 ) are a family of RNA binding proteins that participate in post-transcriptional gene regulation ( Nielsen et al . , 1999; Yisraeli , 2005 ) ; notably , SNPs in the second intron of the human IMP2 gene are associated with increased risk for Type 2 Diabetes ( Saxena et al . , 2007 ) . IMPs enable translation of RNAs containing the IGF2 leader 3 5’UTR by internal ribosomal entry ( Dai et al . , 2011 ) and also bind to the IGF2 mRNA 3’UTR; IMP1 but not IMP2 participates in IGF2 RNA splicing ( Dai et al . , 2013 ) . The IMPs contain six RNA binding motifs , two RRM domains followed by four KH domains and each are substrates for mTOR , which phosphorylates one ( IMP1/3 ) ( Dai et al . , 2013 ) or two ( IMP2 ) ( Dai et al . , 2011 ) serine residues in the segment that links the second RRM domain with the first KH domain . In RD rhabdomyosarcoma cells , the concurrent dual phosphorylation of IMP2 Ser162 and Ser164 is inhibited by rapamycin coincident with inhibition of IMP2 binding to the IGF2 leader 3 5’UTR ( Dai et al . , 2011 ) and IGF2 leader 3 mRNA translation by internal ribosomal entry . In the mouse embryo , all three IMPs are expressed coordinately starting ~e10 . 5 coincident with the onset of IGF action; the expression of Igf2 , Imp1 and Imp3 is largely extinguished before birth ( Nielsen et al . , 1999 ) , whereas Imp2 is widely expressed postnatally ( Dai et al . , 2011 ) . Despite their architectural and sequence similarity , functional differences between the IMPs exist , as displayed most emphatically by the phenotypes of Imp-deficient mice . Imp1 null mice are ~40% smaller than wildtype with aberrant intestinal development and ~50% mortality at p3 ( Hansen et al . , 2004 ) . Imp1 null mouse embryo fibroblasts ( MEFs ) exhibit deficient Igf2 RNA splicing and translation and greatly slowed proliferation; the latter is rescued entirely by exogenous IGF2 . In contrast , Imp2 null mice are nearly normal in size through weaning , lean and slightly small as adults , highly resistant to diet-induced obesity and long lived ( Dai et al . , 2015 ) . Investigating the prolonged lifespan of Imp2 deficient mice , necropsy of an apparently healthy cohort at ~845–850 d age revealed the presence of malignant tumors in 4/6 Imp2+/+ mice but in 0/6 Imp2−/− mice ( Dai et al . , 2015 ) , raising the possibility that IMP2 contributes to tumorigenesis . Herein we demonstrate that although the oncofetal IMPs are commonly reexpressed in human cancers , IMP2 is usually much more abundant in most human cancers than its paralogs IMP1or IMP3 ( Bell et al . , 2013 , Lederer et al . , 2014 ) ; moreover , the IMP2 gene is amplified at a high frequency in several common solid tumors , a phenomenon rarely seen with the IMP1 or IMP3 genes . We show that IMP2 overexpression promotes , and IMP2 deficiency strongly inhibits the proliferation of both MEFs and an array of human tumor-derived cell lines . Beyond its known ability to promote IGF2 translation , IMP2 controls the abundance of the oncogenic transcriptional regulator HMGA1 ( Fedele and Fusco , 2010; Ozturk et al . , 2014; Sumter et al . , 2016 ) by binding and stabilizing HMGA1 mRNA . In turn , HMGA1 , another oncofetal protein , suppresses the transcription of Igfbp2 , a high affinity extracellular IGF binding protein ( Diehl et al . , 2009; Hoeflich et al . , 1999 ) , and also reduces Grb14 , an inhibitor of IGF1R and Insulin Receptor signaling ( Desbuquois et al . , 2013 ) . IMP2’s stabilization of HMGA1 mRNA together with its stimulation of IGF2 mRNA translation act synergistically to promote cell proliferation through mitogenic signaling by the IGF1R and the type A Insulin Receptor . Data generated by the TCGA research network ( http://cancergenome . nih . gov/ ) indicates that amplification of the IMP2 gene is a relatively common event in comparison to amplification of IMP1 and IMP3 ( Figure 1A ) , occurring in ~35–50% of squamous lung cancers , ~15–27% of ovarian cancers and in 15–20% of head and neck , esophageal , cervical and uterine cancers . Moreover , the absolute abundance of IMP2 mRNA in all but a few cancers far exceeds that of the IMP1 and IMP3 paralogues ( Figure 1B ) , even in those cancers wherein the fold amplification of IMP1/IMP3 RNA over their level in the normal tissue is much greater than that of IMP2 . Thus , IMP2 is nearly always the most abundant IMP paralogue in human cancers and its overexpression occurs at a high frequency . IMP2 polypeptide was overexpressed in several cancer-derived cell lines and in wildtype mouse embryo fibroblasts ( MEFs ) . In each instance , IMP2 overexpression increased proliferation in a dose-dependent manner ( Figure 1C ) . Reciprocally , we used the CRISPR-Cas9 to inactivate the IMP2 gene in a diverse cohort of human cancer cell lines; parental controls used a GFP-directed guide . Each of the lines lacking IMP2 expression showed a substantial ( 52–78% ) reduction in the rate of proliferation ( Figure 1D ) . Similarly , Imp2–/– MEFs uniformly displayed less rapid growth , proliferating at ~19–25% the rate of Imp2+/+ cells ( Figure 1E ) . Seeking the mechanism ( s ) by which loss of IMP2 slows the proliferation we chose to focus on MEFs , which are less genetically heterogeneous than the cancer cells , but comparably responsive to IMP2 overexpression and elimination . IGF2 production by Imp2–/– MEFs is reduced by about 19% compared with Imp2 +/+ MEFs controls ( Figure 1F ) , however saturating amounts of exogenous IGF2 do not rescue the proliferation of Imp2−/− MEFs ( Figure 1G ) . This contrasts sharply with the response of Imp1−/− MEFs to IGF2 , which restores their proliferative rate to 100% the level of Imp1+/+ MEFs ( Dai et al . , 2013 ) . Thus , the modest decrease of IGF2 production caused by Imp2 inactivation contributes minimally to the slowed proliferation of Imp2−/− MEFs . We applied deep sequencing to RNA extracted from lysates and from anti-IMP2 immunoprecipitates from Imp2+/+ and Imp2−/− MEFs ( Supplementary file 1 ) . We also subjected triplicate aliquots of Imp2−/− and Imp2+/+ MEFs to whole cell mass spectroscopic analysis using Tandem Mass Tag ( TMT ) Technology which yielded the quantitative estimation of the abundance of 7964 polypeptides ( Supplementary file 2 ) . The bioinformatic analysis of the RNAs enriched in the IMP2 immunoprecipitates ( Huang et al . , 2009 ) , as well as the comparisons of the RNAome and proteome of Imp2+/+ and Imp2−/− MEFs by GSEA ( Subramanian et al . , 2005 ) did not point to dominant mechanisms or specific elements that might underlie the slowed proliferation of the Imp2−/− MEFs . Seeking polypeptides involved in growth factor signaling and cell cycle progression whose altered abundance ( Supplementary file 2 ) might underlie the slower proliferative rate of the Imp2−/− MEFs , we observed that the abundance of IGFBP2 ( Diehl et al . , 2009; Hoeflich et al . , 1999 ) and Grb14 ( Desbuquois et al . , 2013 ) polypeptides , potential inhibitors of IGF1R and insulin receptor-A signaling , are increased 24 . 7- and 8 . 6-fold respectively in the Imp2−/− MEF proteome; Among cell cycle components , the largest difference is a 4 . 5 fold greater abundance of the cdk inhibitor Cdkn1a/p21Cip1 ( Warfel and El-Deiry , 2013 ) in the Imp2−/− MEFs . The differential abundance of IGFBP2 , Grb14 and p21Cip1 was confirmed by immunoblot , whereas the abundance of IMP1 and IMP3 proteins is not altered by deletion of Imp2 ( Figure 2A ) . Lentiviral encoded , doxycycline-inducible shRNAs against Grb14 , Igfbp2 and p21Cip1 were each stably expressed in both Imp2+/+ and Imp2−/− MEFs . Induction of shRNA expression reduced the abundance of the target polypeptides progressively over the initial 4 days ( Figure 2B ) , with levels in the Imp2−/− MEFs becoming almost as low as in the Imp2+/+ MEFs by day 3 ( Figure 2C ) . As regards proliferation , Imp2+/+ MEFs containing shRNA against Grb14 proliferated at the same rate whether treated with DMSO or doxycycline , increasing in number ~10 fold over seven days; Imp2−/− MEFs containing shRNA against Grb14 treated with DMSO increased in number by ~3 . 5 fold , however treatment with doxycycline increased Imp2−/− MEF numbers ~8 fold over the same interval , that is , they proliferated at ~80% the rate of the Imp2+/+ MEFs ( Figure 2D , bottom ) . A similar response is observed with depletion of IGFBP2; doxycycline treatment of Imp2+/+ MEFs containing shRNA against Igfbp2 increased proliferation slightly ( ~20% ) over treatment with DMSO whereas doxycycline treatment of Imp2−/− MEFs containing shRNA against Igfbp2 increased cell number from ~30% to ~68% that of similarly-treated Imp2+/+ MEFs . ( Figure 2D , second from bottom , 2E ) . Concurrent depletion of Grb14 and IGFBP2 from Imp2−/− MEFs restored their proliferation to ~75% that of similarly treated Imp2+/+ MEFs ( Figure 2D , second from top , 2E ) . In contrast , depletion of p21Cip1 from Imp2+/+ and Imp2−/− MEFs increased their proliferation by 1 . 33 and 1 . 36 fold respectively ( Figure 2D , top , 2E ) . Thus , reversal of the 4 . 5 fold increased abundance of p21Cip1does not enable any rescue of Imp2−/− MEF proliferation . We hypothesize that overexpression of Grb14 and IGFBP2 slow proliferation of Imp2−/− MEFs through inhibition of IGF1R and Type A insulin receptor-initiated signal transduction . To evaluate the signaling capacity of the IGF1R , Imp2+/+ and Imp2−/− MEFs were deprived of serum overnight , rinsed and incubated in fresh serum-free medium and stimulated briefly with IGF1 ( 30 nM ) ; although the abundance of the IGF1R and InsR beta subunits does not differ , the IGF1-induced increase in IGF1R beta subunit tyrosine phosphorylation in Imp2−/− MEFs is substantially reduced as compared with Imp2+/+ MEFs , as is the overall tyrosine phosphorylation of IRS1 +IRS2 and activation of Akt assessed by phosphorylation at Ser473 ( Figure 2F ) . Thus IGF1R signaling is impeded in the Imp2 null MEFs . Surprisingly , the Igfbp2 and Grb14 mRNAs are not among the 636 mRNAs 1 . 5-fold or more enriched in the IMP2 immunoprecipitates ( Supplementary file 1 ) from Imp2+/+ MEFs , i . e . , Igfbp2- and Grb14-mRNAs are not IMP2 clients . The abundance of Igfbp2 mRNA in Imp2−/− MEFs is increased 18 . 6 ( RNAseq; Supplementary file 1 ) to 21 . 7 ( by QPCR; Figure 3A , top ) fold and its half life is prolonged ~1 . 7 fold ( 10 . 1 hr vs 5 . 9 hr ) over Imp2+/+ MEFs ( Figure 3B , top ) ; although there is also some increase in polysomal abundance ( Figure 3C , top ) , the markedly increased abundance of the IGFBP2 polypeptide in Imp2−/− MEFs is attributable primarily to increased Igfbp2 gene transcription . In contrast , Grb14 mRNA is approximately 70% lower ( RNAseq: Supplementary file 1; QPCR: Figure 3A , middle ) and its half-life is ~40% shorter ( 3 . 89 hr vs 6 . 03 hr; Figure 3B , middle ) in Imp2−/− versus Imp2+/+ MEFs . Although there is a ~ 2 . 5 fold increase in Grb14 mRNA polysomal abundance in the Imp2−/− MEFs ( Figure 3C , middle ) , the primary mechanism for ~9 fold increased polypeptide abundance is post-translational , a marked prolongation of Grb14 polypeptide half-life from ~2 hr to ~31 hr ( Figure 3D , middle ) ; p21Cip1 follows a pattern generally similar to Grb14 ( Figure 3A–E , bottom ) . As in MEFs , deletion of IMP2 from the cancer cell lines is accompanied by a marked increase in the abundance of Grb14 and IGFBP2 ( Figure 4A ) , and overexpression of IMP2 reduces the abundance of both the IGFBP2 and Grb14 polypeptides ( Figure 4B ) . Notably , IMP2 deletion from the human cancer cells causes a much greater decrease in IGF2 production than in MEFs , ranging from ~35–50% of parental levels ( Figure 4C ) . Consequently , we attempted to overcome both the relative deficiency of IGF2 as well as the ability of IGFBP2 overabundance to sequester IGF2 by the addition of an excess of IGF2 . The effect of adding exogenous IGF2 and shRNA-induced Grb14 depletion , singly and in combination , on the proliferation of the parental and IMP2 deficient cancer cell lines shown in Figure 4D . Depletion of Grb14 from parental cancer cells increased proliferation by only 1 . 04–1 . 12 fold whereas Grb14 depletion from the IMP2-deficient variants increased their proliferation by 1 . 5–2 . 09 fold . Similarly , addition of saturating amounts of IGF2 to the parental and IMP2 deficient cancer cells causes a substantially greater fractional increase in cell number in each of the IMP2 deficient variants than in the parental cells ( except IMP2 deficient SNU-423 , bottom right ) . Excess IGF2 alone however , like Grb14 depletion , restores total cell number of the IMP2 deficient variants only partially toward the parental levels; in the IMP2 deficient HeLa ( Figure 4D , middle left ) , MB-231 ( Figure 4D , middle right ) and Hep3b cells ( Figure 4D , top left ) , combining depletion of Grb14 with the addition of IGF2 restores cell numbers of the IMP2 deficient variants to 72% , 79% and 96% of the parental levels . Thus in these three lines the large majority of the IMP2-stimulated proliferation is mediated by enhancing the production and action of IGF2 . The IMP2 deficient variants of RD ( Figure 4D , bottom left ) and HCC1359 cells ( Figure 4D , top right ) exhibit a more modest additivity suggesting the contribution of other IMP2-regulated proliferative regulators . Only in the SNU-423 cells ( Figure 4D , bottom right ) did the parental cells show a more robust proliferative response to IGF2 than did the IMP2 deficient variant ( 2 . 44 fold vs . 1 . 63 fold ) . Nevertheless , IMP2 deficient SNU-423 cells responded to depletion of Grb14 with a greater increase in proliferation than the parental SNU-423 cells ( 1 . 5 fold vs 1 . 12 fold ) . We conclude that the pro-proliferative effect of IMP2 in most cancer derived cell lines is mediated predominantly by upregulation of IGF2 production and suppression of Grb14 and IGFBP2 polypeptide abundance which together facilitate IGF2 signaling . Because the increase in Igfbp2 mRNA induced by IMP2 deficiency is due primarily to increased gene transcription , we inquired whether elimination of IMP2 significantly alters the abundance of any of the IMP2 client RNAs encoding transcription factors ( TFs ) . The RNAseq from Imp2+/+ and Imp2−/− MEFs ( Supplementary file 1 ) detected 810 of 1457 mouse TFs at AnimalTFDB ( http://www . bioguo . org/AnimalTFDB/index . php ) , many more than did the proteomic analysis ( 275/1457 ) ; among the 636 RNAs enriched ≥1 . 5X in the IMP2 IP are 64 that encode TFs , of which 14 show a three-fold or greater decreased abundance in Imp2−/− MEFs as compared with Imp2+/+ MEFs ( none show a comparably increased abundance; Supplementary file 1 ) . Only one of polypeptides encoded by these 14 mRNAs , HMGA1 was detected in the proteomic analysis ( Supplementary file 2 ) , and the abundance of HMGA1 polypeptide was approximately 60% lower in the Imp2−/− than in the Imp2+/+ MEFs . We confirmed by PCR that HMGA1 mRNA is specifically enriched in the IMP2 IP , ~2 fold from MEFs and ~5 fold from RD cells ( Figure 5A ) ; RNA seq indicates that IMP2 binds primarily to the 3’UTR of the Hmga1 RNA ( Figure 5—figure supplement 1 ) . Hmga1 mRNA in Imp2−/− MEFs is 90% lower than Imp2+/+ MEFs ( Figure 5B; Supplementary file 1 ) primarily because of faster mRNA turnover; the half life of Hmga1 mRNA , ~3 . 5 hr in Imp2+/+ MEFs is reduced to 0 . 4 hr in Imp2−/− MEFs ( Figure 5C ) . The reduction in the HMGA1 polypeptide ( Figure 5F ) is somewhat ameliorated by a 1 . 8 fold increase in Hmga1 mRNA polysomal abundance ( Figure 5D ) . Depletion of HMGA1 from Imp2+/+ MEFs or RD cells increased the abundance of IGFBP2 mRNA 12 . 1 fold and 20 . 8 fold respectively ( Figure 5G , left ) ; reciprocally , doxycycline-induced expression of HMGA1 reduced IGFBP2 mRNA levels in RD cells and Imp2+/+ MEFs and by 79% and 63% respectively ( Figure 5G , right ) . To determine if HMGA1 directly binds to the mouse Igfbp2 gene promoter , chromatin immunoprecipitation was performed using MEFs . The Igfbp2 gene promoter was enriched 3 . 7 fold in the anti-IMP2 IP as compared with control IgG , an enrichment comparable to that observed of the promoters of the genes encoding Igf1r , Igfbp1 and Igfbp3 ( Figure 5H ) , previously identified as HMGA1 transcriptional targets ( Aiello et al . , 2010; Liritano et al . , 2012 ) ; notably the promoter of Grb14 was not enriched in the HMGA1 IP; similar results were obtained with HMGA1 ChIP using RD cells ( not shown ) . Thus HMGA1 binds and regulates the Igfbp2 gene , suppressing its transcription , so that HMGA1 depletion increases , and overexpression decreases IGFBP2 polypeptide abundance in MEFs; surprisingly , HMGA1 affects the abundance of the Grb14 polypeptide in a virtually identical manner ( Figure 5I ) , an effect not mediated through IGFBP2 ( Figure 5—figure supplement 2 ) . As with loss of IMP2 , HMGA1 depletion greatly increases the abundance of Grb14 polypeptide by reducing Grb14 polypeptide degradation ( Figure 5-Supplementaary Figure 3 ) , acting indirectly through an as yet unidentified transcriptional target . The dependence of HMGA1 abundance on IMP2 demonstrated for MEFs ( Figure 5I ) is observed in all of the other cancer derived cell lines examined; elimination of IMP2 is accompanied by a major reduction in HMGA1 polypeptide ( Figure 4A ) whereas forced overexpression of IMP2 in HCC1419 and H2029 cells causes a dose-dependent increase in HMGA1 ( Figure 4B ) . Doxycycline-induced expression of HMGA1 increases the proliferative rate of Imp2+/+ MEFs by 1 . 3 fold but increases that of Imp2−/− MEFs by 3 . 1 fold , restoring their proliferative rate from 30 . 6% of Imp2+/+ MEFs to 73% ( Figure 5J ) , essentially the same increase as incurred by dual depletion of IGFBP2 and Grb14 ( Figure 2D , E ) . Induced expression of HMGA1 in RD cells stimulates proliferation by 1 . 2 fold , whereas HMGA1 increases the proliferative rate of IMP2-deficient RD cells by 2 . 6 fold , restoring the rate to 83% that of the parental RD cells overexpressing HMGA1 ( Figure 5K ) ; this is a much greater stimulation than induced by addition of excess IGF2 plus depletion of Grb14 ( from 26% to 52% of the parental rate , Figure 4D bottom left ) , suggesting that in RD cells HMGA1 engenders proproliferative outputs in addition its ability to suppress IGFBP2 and Grb14 . We showed previously that transfection of Flag-IMP2 ( Ser162Asp/Ser164Asp ) stimulated the translation of an IGF2-leader 3-luciferase reporter in RD rhabdomyosarcoma cells to an extent similar to Flag-IMP2 wildtype , whereas Flag-IMP2 ( Ser162Ala/Ser164Ala ) was no more effective than empty vector ( Dai et al . , 2011 ) ; this translational response was paralleled by the ability of the IMP2 variants to bind the IMP2 leader 3 5’UTR RNA . To evaluate the functional significance of each phosphorylation site on the ability of IMP2 to promote proliferation , we generated IMP2 variants with either an Ala or an Asp at both Ser162 and Ser164 , which we refer to as IMP2-AA , -AD , -DA and -DD and stably expressed doxycycline regulated Flag-tagged versions of each of the four mutant IMP2 , a Flag-tagged IMP2 wildtype ( IMP2-SS ) and an empty Flag vector in the parental Imp2−/− MEFs . Polyclonal populations were selected and a doxycycline dose was determined that induced polypeptide expression of each Flag-tagged IMP2 variant to a level similar to endogenous IMP2 in Imp2+/+ MEFs ( Figure 6A , immunoblot; Figure 6B , TMT peptides , immunoblot below ) ; 104 cells of the six MEF types were plated in replicate and proliferation was monitored over six days ( Figure 6A ) . Imp2−/− MEFs expressing IMP2-AD and IMP2-DD proliferated at a rate indistinguishable from Imp2−/− MEFs expressing IMP2-SS , whereas Imp2−/− MEFs expressing IMP2-AA IMP2-DA proliferated at ~50% of that rate as did Imp2−/− MEFs containing empty Flag vector . To evaluate how these IMP2 variants altered the MEF proteome , whole cell mass spectroscopic proteomic analysis was performed as before ( Supplementary file 3 ) . If the TMT-determined abundance of each polypeptide in the MEFs expressing Flag-IMP2-AA and Flag-IMP2-DA and empty Flag vector ( slower proliferating MEFs ) were summed , and divided by the sum of the corresponding polypeptide in Imp2−/− MEFs containing Flag-IMP2-SS , Flag-IMP2-DD and Flag-IMP2-AD ( faster proliferating MEFs ) , the two endogenous polypeptides with the highest ratio in the entire proteome are IGFBP2 ( 5 . 7 ) and Grb14 ( 3 . 7 ) ( Figure 6B , right column ) . Immunoblot confirmed the ability of IMP2-AD and DD to sustain HMGA1 polypeptide and suppress IGFBP2 and Grb14 polypeptide abundance in Imp2−/− MEFs comparably to wildtype IMP2 and more effectively than IMP2-AA and IMP2-DA ( Figure 6B , right ) . Thus the ability of IMP2 to maintain the expression of Hmga1 and suppress the abundance of IGFBP2 and Grb14 , which underlies in large part IMP2’s ability to promote MEF proliferation , depends strongly on the status of the IMP2 mTOR phosphorylation site at Ser164 . The regulation of IMP2 phosphorylation was examined in 293E cells . Overnight withdrawal of serum , brief incubation in amino acid and serum-free medium or torin1 ( Figure 6C , column 1 vs columns 3 , 4 , 5 respectively ) fully abolish the concomitant dual phosphorylation at IMP2 Ser162 and Ser164; the sensitivity of this dual phosphorylation to amino acid withdrawal and rapamycin ( Dai et al . , 2011 ) as well as its rapid stimulation by insulin ( Figure 6C , column 2 vs column 3 ) point strongly to the mediation of mTORC1 . Nevertheless , serum withdrawal or torin1 cause little dephosphorylation of IMP2 ( Ser162P ) and only partial dephosphorylation of IMP2 ( Ser164P ) , a pattern previously observed with shRNA-induced depletion of mTOR in RD cells ( Dai et al . , 2011 ) . The modest dephosphorylation of IMP2 ( Ser164 ) caused by torin1 or amino acid withdrawal contrasts with the ability of these treatments to cause the near-total dephosphorylation of sites phosphorylated exclusively by mTORC1 , such as S6K1 ( Ser389 ) or 4E-BP ( Thr37/46 ) . The marked resistance of IMP2 ( Ser162P ) to dephosphorylation by torin1 is similar to that seen previously with the co-translational phosphorylation of IMP1 ( Ser181 ) ( Dai et al . , 2013 ) . We therefore examined whether IMP2 is phosphorylated co-translationally by incubating 293 cells with puromycin , which is covalently incorporated into nascent polypeptides because of its resemblance to the 3’ end of aminoacylated tRNA , thereby terminating elongation . Puromycin immunoprecipitates contain IMP2 polypeptides that are phosphorylated at both Ser162 and Ser164 as well as at both sites concomitantly , which can have only occurred co-translationally; pretreatment of the cells with torin1 20 min . prior to addition of puromycin strongly inhibits these phosphorylations of nascent IMP2 ( Figure 6D ) . Thus IMP2 undergoes mTOR dependent cotranslational phosphorylation at both Ser162 and Ser164 , in addition to an insulin- , serum- and rapamycin-sensitive post-translational phosphorylation primarily ( if not exclusively ) at IMP2 ( Ser164 ) , which results in concomitant dual phosphorylation . How the mature IMP2 polypeptide attains its steady state level of phosphorylation is not entirely understood; it is clear however , that the phosphorylation at either IMP2 site does not affect phosphorylation at the other site; substitution of Ser164 with either Ala or Asp has little effect on the extent of phosphorylation at Ser162 and the same is true of such substitutions at Ser162 on the phosphorylation at Ser164 ( Figure 6E ) . IMP2 is amplified and overexpressed in many cancers and recent reports ( Li et al . , 2015; Liu et al . , 2015; Mu et al . , 2015 ) have pointed to a tumorogenic role in individual cancer types , often with a poor prognosis ( Barghash et al . , 2015; Bigagli et al . , 2016; Davidson et al . , 2014; Kessler et al . , 2013 ) . Consistent with the view that IMP2 contributes broadly to tumor progression , we show herein that IMP2 drives the proliferation of both MEFs and a wide and unselected cohort of cancer cell lines . IMP2 is best considered a tumor promoter rather than an oncogene; although mice with transgenic overexpression of IMP2 in liver exhibit a higher tumor burden after diethylnitrosamine-induced hepatotoxicity ( Kessler et al . , 2015 ) , IMP2 has not as yet been shown capable of initiating tumorigenesis . The conclusion that IMP2 is a tumor promoter raises for consideration the biology of IMP2 in normal , noncancerous circumstances and how this might underlie tumor promotion . IMP1 and IMP3 are so-called ‘oncofetal’ genes because their high expression in embryonic development is strongly downregulated before birth , but frequently reappears in cancers . IMP2 diverges from this model in that it continues to be expressed widely in adult life , however IMP2 appears to be preferentially expressed in tissue resident stem cells , such as satellite cells ( Boudoukha et al . , 2010; Li et al . , 2012 ) , neural stem cells ( Fujii et al . , 2013 ) and white adipocyte precursors ( Dai et al . , 2015 ) , where it is critical for their ability to proliferate , differentiate and gain motility . Thus IMP2 gene amplification and/or upregulated expression may be important to the genesis of cancer stem cells and thereby central to IMP2’s tumor promoting capability , as proposed by ( Degrauwe et al . , 2016a ) and demonstrated in glioblastoma stem cells ( Degrauwe et al . , 2016b ) . The present work also elucidates a major molecular pathway that underlies IMP2’s tumor promoting activity; IMP2 occupies a central place in a network of gene products that drive proliferation through the IGF1R and INSR-A ( Figure 7 ) . The expression of the IMP2 gene is strongly upregulated by the oncofetal HMGA2 oncogene ( Brants et al . , 2004; Cleynen et al . , 2007; Li et al . , 2012 ) and the mRNA encoding the oncogenic transcriptional modifier HMGA1 ( Fedele and Fusco , 2010; Ozturk et al . , 2014 , Sumter et al . , 2016 ) , previously detected as an IMP2 client ( Janiszewska et al . , 2012 ) , is identified here as an important mediator of IMP2’s tumor promoting activity . Restoring HMGA1 abundance largely rescues the proliferative defect engendered by IMP2 deficiency in MEFs and RD cells ( Figure 5J , K ) . IMP2 binds the HMGA1 mRNA 3’UTR , inhibiting its degradation; elimination of IMP2 causes a major decrease in HMGA1 mRNA and protein abundance . Regarding mechanism by which IMP2 protects HMGA1 mRNA , Degrauwe et al . ( 2016b ) demonstrated in Glioblastoma stem cells that IMP2 binding to the HMGA1 mRNA 3’UTR ( Figure 5—figure supplement 1 ) overlaps the binding site of the pro-differentiation tumor suppressor miR , Let-7 and antagonizes Let-7’s ability to promote HMGA1 mRNA degradation . HMGA1 expression is high during embryonic development , much lower in adult tissues , but greatly upregulated in a broad range of benign and malignant tumors . The HMGA1 protein binds to the AT-rich regions in the DNA minor groove; although lacking direct transcriptional regulatory activity , it modulates gene expression by changing chromatin structure and promoting the assembly of other TFs . HMGA1’s output is thus cell context dependent; in addition to oncogenic transformation , HMGA1 can also promote expression of a stem cell-like program and/or an epithelial-to-mesenchymal transition . Numerous transcriptional targets relevant to tumorigenesis have been described , including p53 and Rb ( Sumter et al . , 2016 ) or the oncogenic miR222 ( Wong et al . , 2010; Zhang et al . , 2011; Panneerselvam et al . , 2016 ) ; here we identify the ability of HMGA1 to suppress the abundance of two previously unknown targets , IGFBP2 and Grb14 , inhibitors of IGF1R and Insulin receptor signaling . HMGA1 binds directly within the Igfbp2 gene promoter to suppress transcription . In contrast , HMGA1 does not bind to the Grb14 gene , but acts through an unidentified transcriptional target to ultimately promote Grb14 polypeptide degradation . Envisioning an anti-proliferative role for Grb14 is relatively straightforward , although contrary examples exist ( Balogh et al . , 2012 ) . The role of IGFBP2 in cancer however is controversial ( Pickard and McCance , 2015; Russo et al . , 2015 ) . IGFBP2 has IGF-dependent and independent , proproliferative actions in cell culture; IGFBP2 contains an RGD domain and can interact with the integrins αVβ3 and α5β1 to affect cell survival or motility ( see Russo et al . , 2015 ) , and functions coordinately with IGF1 to promote endothelial proliferation through polymerization and inactivation of the receptor tyrosine phosphatase RTPTβ , causing inhibition of PTEN ( Shen et al . , 2015 ) . Nevertheless , the ability of transgenic overexpression of IGFBP2 to inhibit postnatal growth ( Hoeflich et al . , 1999 ) and colonic tumorigenesis ( Diehl et al . , 2009 ) attests to its ability to sequester IGF in vivo . A central role for IGFs in tumorigenesis had been predicted by the finding that mouse embryo fibroblasts devoid of an IGF1R are resistant to transformation by many cellular and viral oncogenes ( Sell et al . , 1993 ) . Although studies in mouse cancer models were often supportive , interventions directed at inhibition of the IGF1R in man as a single therapy and in combination with chemotherapy have not shown efficacy ( Baserga , 2013; Pollak , 2012 ) . The reason ( s ) for this disappointing outcome are not known , however one likely mechanism is the frequent overexpression of IGF2 , due in part to loss of imprinting at the IGF2 locus and/or loss of the IGF2R ( Brouwer-Visser and Huang , 2015; Livingstone , 2013 ) . IGF2 has high affinity for the fetal-predominant A form of the insulin receptor ( INSR-A ) , also frequently reexpressed in tumors . The present findings identify IMP2 as a vital component of this network; its stabilization of HMGA1 mRNA is necessary for the oncogene-driven upregulation of HMGA1 transcription to generate HMGA1 polypeptide overabundance . In turn , HMGA1 suppression of Grb14 and IGFBP2 synergize with IMP2 stimulation of IGF2 L3 mRNA translation to promote IGF2 mitogenic signaling though the IGF1R and Insulin Receptor-A ( Morrione et al . , 1997; Ulanet et al . , 2010 ) . It is worth noting that the human genes encoding major elements of the tumor promoting pathway described herein , that is , HMGA2 , IMP2 , Grb14 , IGF2 , have each been found by GWAS to encode SNPs that are associated with excess risk for type 2 diabetes ( Flannick and Florez , 2016 ) , a condition accompanied by increased risk for many cancers ( Giovannucci et al . , 2010; García-Jiménez et al . , 2016 ) . Most of these SNPs are noncoding so that the contribution of the encoded polypeptide in conferring risk is unknown . As regards HMGA1 , Foti et al . ( 2005 ) reported 4 individuals with reduced HMGA1 abundance due to either hemizygous gene loss or point mutation , who exhibited severe insulin resistance accompanied by diminished insulin receptor abundance . This group also described variants in the HMGA1 gene that associate in excess with type 2 diabetes ( Chiefari et al . , 2011 ) however this association has been effectively refuted ( Marquez et al . , 2012; Froguel et al . , 2012 ) , so that a role for HMGA1 in human type 2 diabetes remains unsubstantiated . The mouse HMGA1 KO also exhibits reduced insulin receptor abundance but as part of a complex metabolic phenotype with mild glucose intolerance , hypoinsulinemia and enhanced insulin sensitivity ( Foti et al . , 2005 ) . Thus , whereas the ability of the pathway centered around IMP2 to initiate and/or promote tumor progression is well established , the role of these elements , if any , in the development of type 2 diabetes and the increased susceptibility of that condition to various cancers remains to be elucidated . MEFs were isolated from E12 . 5–13 . 5 embryos of Imp2+/– intercrosses ( Dai et al . , 2015 ) . The proliferation , QPCR and polypeptide abundance measurements derived from mouse embryo fibroblasts ( Figure 1E , F , G Figure 2D , E , Figure 3A–E , Figure 5A–H , J , Figure 6A ) represent data from 3 or more separate experiments; each experiment used MEFs derived from different Imp2+/+ and Imp2−/− embryos . Human RD ( RRID:CVCL_1649 ) , Hela ( RRID:CVCL_0058 ) , MB-231 ( RRID:CVCL_0062 ) , Hep3b ( RRID:CVCL_0326 ) and 293E ( RRID:CVCL_6974 ) cell lines were maintained in DMEM supplemented with 10% FBS . HCC-1359 ( RRID:CVCL_5128 ) , NCI-H2029 ( RRID:CVCL_1516 ) , SNU-423 ( RRID:CVCL_0366 ) , HCC-1419 ( RRID:CVCL_1251 ) cells were cultured in RPMI-1640 supplemented with 10% FBS . The corresponding data derived from human cancer cell lines ( Figure 1C , D , Figure 4C , D , Figure 5A , G , K ) represented three separate experiments . The error bars on these figures represent ± one S . D . DMSO , doxycycline , cycloheximide , rapamycin , insulin , human IGF1 , human IGF2 , RPL26 and actin antibodies were from Sigma-Aldrich ( St . Louis , Missouri , USA ) ; Torin 1 from TOCRIS . Antibodies: P-AKT ( Ser473 ) , AKT , pS6K ( Thr389 ) , S6K , P-4E-BP1 ( Thr37/460 ) , 4E-BP1 , P-AMPK , AMPK , IGF1R , IRS1 , IRS2 , P21 , Phos-Tyrosine from Cell Signaling Technology; Grb14 and IGFBP2 antibodies from abcam . IMP antibodies were described previously ( Dai et al . , 2011; Dai et al . , 2013 ) . The inducible Lentiviral shRNA for Grb14 , IGFBP2 , p21 and HMGA1 were from Dharmacon and the stably expressed Flag-IMP2 variants and Flag-HMGA1were generated using pcDNA6/TR and pcDNA5/TO vectors; these reagents were used according to manufacturer’s instructions . The P-site mutants of IMP2 were generated using the QuikChange site-directed mutagenesis kit ( Stratagene ) . CRISPR-Cas reagents were obtained from Addgene ( U6-Chimeric_BB-CBh-hSpCas9 ) . Small guide RNAs ( sgRNAs ) , selected for minimal predicted off-target mutagenesis , were designed using CRISPR Design software ( http://crispr . mit . edu ) ; ( hIMP2_CRISPR 1 , caccgACAAGAACAATTCCTGAGCT; hIMP2_CRISPR 2 , aaacAGCTCAGGAATTGTTCTTGTC ) . Plasmids co-expressing the sgRNA specific to the IMP2 and the Streptococcus pyogenes Cas9 nuclease were transfected into cells by Nucleofector ( Lonza ) according to manufacturer’s instruction . Immunoblot analyses as described ( Dai et al . , 2015 ) ; each immunoblot was repeated at least one time . The levels of IGF2 were determined using Human IGF2 and Mouse IGF2 ELISA kit from Cusabio life science and R&D systems respectively . Cycloheximide was used at a final concentration of 250 μM to inhibit mRNA translation; protein half-life was estimated from immunoblot of extracts prepared before and at various times after cycloheximide addition . The data used to construct the comparison of IGF2BP1/2/3 gene copy number ( Figure 1A ) and mRNA abundance ( Figure 1B ) in human cancers was obtained from TCGA published ( PMID:18772890 , 21720365 , 20579941 , 20601955 , 27158780 , 28112728 , 28052061 , 26061751 ) and provisional datasets . cBioPortal ( http://www . cbioportal . org/index . do ) was used to organize and process the TCGA data . Samples were processed and analyzed through the Thermo Fisher Scientific Center for Multiplexed Proteomics at Harvard Medical School ( Weekes et al . , 2014 ) . Peptides derived from digestion using LysC and trypsin were labeled with Tandem Mass Tag 8-plex reagents and fractionated . Multiplexed quantitative mass spectrometry data were collected on an Orbitrap Fusion mass spectrometer operating in a MS3 mode using synchronous precursor selection for the MS2 to MS3 fragmentation ( McAlister et al . , 2014 ) . MS/MS data were searched against a Uniprot mouse database with both the forward and reverse sequences using the SEQUEST algorithm . Further data processing steps included controlling peptide and protein level false discovery rates , assembling proteins from peptides , and protein quantification from peptides . RNA was isolated using QIAGEN RNase kit . RNA samples were examined for the integrity by Agilent bioanalyzer prior to further processing . Sequencing was carried out on Illumina HiSeq 2500 , resulting in approximately 21 million of 50 bp paired-end reads per sample . STAR aligner ( Pubmed ID: 23104886 ) was used to map sequencing reads to the Mus musculus genome version 10 ( mm 10 ) reference genomes . Read counts for individual transcripts were produced with HTSeq-count ( Pubmed ID:25260700 ) , followed by the estimation of expression values and detection of differentially expressed transcripts using EdgeR ( Pubmed ID:19910308 ) . The RNA sequence data have been deposited in NCBI’s Gene Expression Omnibus and are accessible through GEO series accession number GSE101311 . Cells were treated with 1 μM Actinomycin D ( Sigma ) for 12 hr to block transcription . After the indicated times , the total RNAs were extracted , followed by DNase digestion for eliminating DNA contamination and cDNA syntheses . The concentration of mRNAs were quantified by Real-Time RT-PCR using the SYBR Green . The primers for the real-time PCRs were as follow: murine Grb14 sense ( actatgtggacgacaacagc ) , antisense ( cataattctttctaagatacag ) ; murine Igfbp2 sense ( ccagcaggagttggaccaggt ) , antisense ( cttaaggttgtaccggccatgc ) ; murine P21sense ( cgagaacggtggaactttgac ) , antisense ( cagggctcaggtagaccttg ) . Protein was cross-linked to DNA by treatment of PBS washed cells with formaldehyde . Cells were pelleted and resuspended in cell lysis buffer containing a protease inhibitor cocktail ( Roche ) . After 10 min on ice , the nuclei were collected and chromatin was sheared by sonication . ChIP was performed using the Simple ChIP Enzymatic Chromatin IP kit ( Cell Signaling Technology ) according to the manufacturers’ instructions . The sheared DNA-protein complexes were immune-precipitated using antibodies to HMGA1 . A nonimmune IgG was the negative control .	Some types of cancers develop when genes known as oncogenes or tumor promoters become faulty , and are present at abnormally high levels or inappropriately turned on . For example , cancer cells often have extra copies of the gene IMP2 and therefore produce too much the IMP2 protein . Previous research has shown that mice that lack the IMP2 protein develop fewer cancers and live longer , while patients whose cancers make too much IMP2 have a poorer prognosis . In healthy cells , the IMP2 protein normally helps to make new gene products by stabilising certain newly produced RNA molecules – the precursors of proteins , and in some cases by promoting the translation of these RNAs into proteins . For example , IMP2 binds to the mRNA that encodes the protein IGF2 , which is a protein that helps cells to grow and is commonly produced in large quantities by cancer cells . However , until now it was not clear whether IMP2 only acts by increasing the production of IGF2 or also contributes to cancer growth in other ways . Using a range of human cancer cell lines , and healthy mouse cells , Dai et al . first confirmed that without IMP2 , cancer cells made less IGF2 and grew less quickly . When IGF2 was added to the cells lacking IMP2 , it only partially restored their ability to grow . Further experiments revealed that cells without IMP2 had increased levels of proteins that counteract the effects of IGF2 . Usually , IMP2 binds and stabilizes the mRNA that encodes the oncogenic protein HMGA1 , which is known to regulate the number of ‘anti-IGF2 proteins’ . However , without IMP2 , the HMGA1 levels drop , which causes an increase of the anti-IGF2 proteins . This indicates that IMP2 promotes cancer cell growth both by enabling cells to produce more IGF2 and by suppressing inhibitors of IGF2 action . This suggests that cancer patients whose tumors have abnormally high levels of IMP2 may be especially sensitive to drugs that target and inhibit IGF2 .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cancer", "biology" ]	2017	IGF2 mRNA binding protein-2 is a tumor promoter that drives cancer proliferation through its client mRNAs IGF2 and HMGA1
Coordination of diverse individuals often requires sophisticated communications and high-order computational abilities . Microbial populations can exhibit diverse individualistic behaviors , and yet can engage in collective migratory patterns with a spatially sorted arrangement of phenotypes . However , it is unclear how such spatially sorted patterns emerge from diverse individuals without complex computational abilities . Here , by investigating the single-cell trajectories during group migration , we discovered that , despite the constant migrating speed of a group , the drift velocities of individual bacteria decrease from the back to the front . With a Langevin-type modeling framework , we showed that this decreasing profile of drift velocities implies the spatial modulation of individual run-and-tumble random motions , and enables the bacterial population to migrate as a pushed wave front . Theoretical analysis and stochastic simulations further predicted that the pushed wave front can help a diverse population to stay in a tight group , while diverse individuals perform the same type of mean reverting processes around centers orderly aligned by their chemotactic abilities . This mechanism about the emergence of orderly collective migration from diverse individuals is experimentally demonstrated by titration of bacterial chemoreceptor abundance . These results reveal a simple computational principle for emergent ordered behaviors from heterogeneous individuals . Collective group migration , as an important class of coordinated behaviors , is ubiquitous in living systems , such as navigation , foraging , and range expansion ( Krause et al . , 2002; Partridge , 1982; Sumpter , 2010 ) . In the presence of individual heterogeneity , the migrating group often exhibits spatially ordered arrangements of phenotypes ( Krause et al . , 2002; Parrish and Edelstein-Keshet , 1999; Partridge , 1982; Sumpter , 2010 ) . In animal group migration , individual behavioral abilities ( e . g . , directional sensitivity ) would result in social hierarchy , which further drives the spatial arrangement in a coordinated group ( Couzin et al . , 2005 ) . At the same time , spatial arrangements can lead to different costs and benefits for the individuals participating in the group migration ( Krause , 1994; Parrish and Edelstein-Keshet , 1999; Partridge , 1982 ) . Participating individuals must follow disciplinary rules to organize themselves into coordinated patterns while on the move , which requires complex computational abilities to interact with the group and the environment ( Couzin and Krause , 2003; Couzin et al . , 2002; Vicsek and Zafeiris , 2012 ) . Therefore , understanding how individuals of different phenotypes determine their location in the group is an essential prerequisite to uncover the organization principles of collective populations . The chemotactic microbe , Escherichia coli , provides a simple model to address the emergence of collective decision-making among diverse population , as it can exhibit both individualistic behaviors ( Dufour et al . , 2016; Frankel et al . , 2014; Kussell and Leibler , 2005; Waite et al . , 2016; Waite et al . , 2018 ) and collective migratory patterns ( Adler , 1966a; Fu et al . , 2018; Keller and Segel , 1971a , Wolfe and Berg , 1989 ) . Individual cells perform run-and-tumble random motions by spontaneously switching the rotating direction of flagella ( Berg , 2004; Berg and Brown , 1972 ) . These cells can facilitate the chemotaxis pathway to control the switching frequency to bias their motions toward their favorable direction along the chemoattractant gradient , where the efficiency to climb the gradient is defined as the chemotactic ability ( χ ) ( Celani and Vergassola , 2010; Dufour et al . , 2014; de Gennes , 2004; Si et al . , 2012 ) . In addition , the chemotactic abilities of individual cells exhibit substantial phenotypic heterogeneity even for the clonal bacterial population , which diversifies the chemotactic response to identical signals ( Spudich and Koshland , 1976; Waite et al . , 2016; Waite et al . , 2018 ) . Despite the stochastic solitary behavior and variations in phenotypic ability , the E . coli population can migrate as a coherent group by following a self-generated attractant gradient , via consumption of the whole population ( Adler , 1966a; Saragosti et al . , 2011; Wolfe and Berg , 1989 ) . The migratory group form a stable pattern of phenotypes sorted by their chemotactic abilities ( Figure 1A ) , which is believed to maintain phenotypic diversity in the group ( Fu et al . , 2018; Waite et al . , 2018 ) . Although a previous study showed that behavior modulation helps migrating bacteria to maintain a consistent group ( Saragosti et al . , 2011 ) , how individuals with phenotypic variations manage to determine their relative positions in the group remains to be determined . Here , we analyzed single-cell trajectories of bacterial run-and-tumble motions in the chemotactic migration group ( see Materials and methods ) . We found that the expected drift velocity of individual cells decreased from the back to the front . Such a spatial profile modulates cells to behave as mean reverting processes relative to the entire group , that is , cells effectively tend to revert their direction of runs toward the mean position of the group . Using an Ornstein-Uhlenbeck ( OU ) -type model , we demonstrated that the mean reversion behavior is a result of a pushed wave , where the driving force decreases from the back to the front of the group . Cells of different phenotypes are imposed to the same type of driving force , of which the strength is coupled with their chemotactic abilities . As a result , the pushed wave front , driven by the spatially structured force , can maintain more diverse individuals in the migratory group . By theoretical analysis and stochastic simulations , we also discovered that the balanced locations of diverse phenotypes are spatially ordered by their chemotaxis abilities . Further simulations and experiments with cells of titrated chemoreceptor abundance demonstrated that this spatial modulation of individual behavior enables the ordering of bacteria with diverse chemotaxis abilities . Therefore , although the high-order computational abilities are not available to simple organisms , the spatial modulation of stochastic behaviors at the individual level reveals novel decision-making capabilities at the population level . To directly investigate the ordering mechanism in a coherent migration group , we quantified the stochastic behaviors of bacterial run-and-tumble motions relative to the stable migrating group . To achieve this , we employed a microfluidic device that generated a stable propagating band of bacteria , as previously reported ( Fu et al . , 2018; Saragosti et al . , 2011 ) . Using aspartate ( Asp ) as the only chemoattractant to drive the migration of E . coli , we tracked a small fraction of fluorescent bacteria ( JCY1 ) as representatives of the non-fluorescent wild-type cells ( RP437 ) ( see Materials and methods , Appendix 1—figure 1 , Video 1 ) . Because the group velocity VG=〈Δxi ( t ) /Δt〉 was constant over time , VG∼3 . 0μm/s ( Appendix 1—figure 2 ) , we were able to map the tracks to a moving coordinate z=x − VGt . With the shifted tracks , we calculated the key statistics of the single-cell behaviors . We first checked that the average instantaneous velocity , VI ( z ) =⟨Δxi ( z ) /Δt⟩ , was constant along the density profile ρ ( z ) ( Figure 1B ) . This result confirmed that the group migrates coherently . Then , we identified all the run states and tumble states of individual trajectories using a previously described computer assistant program ( Dufour et al . , 2016; Waite et al . , 2016 ) , to ensure that the tracks are separated into successive tumble-and-run events ( see Materials and methods ) . Comparing the sample events initiated from the back ( b ) , middle ( m ) , and front ( f ) of the migration group , we observed that the runs in the front were longer but distributed more uniformly in terms of the directionality , whereas the runs in the back were shorter but more oriented toward the group migration direction ( Figure 1C , Appendix 1—figure 3A , B ) . Quantitatively , the statistics of run length ( and duration ) displayed exponential distributions with the means for the direction of group migration longer than those for the opposite direction ( Figure 1D , Appendix 1—figure 3C ) . The angular distribution of the run length confirmed the difference in directionality between cells in different spatial locations ( Figure 1E , Appendix 1—figure 3D ) . We also confirmed that cells in the back showed greater directional persistence toward the migration direction ( Appendix 1—figure 4A–C ) , as ( Saragosti et al . , 2011 ) reported previously . All these results suggested that the bacteria in the back run more effectively toward the direction of group migration than those in the front . To quantify the spatial extent of the drift efficiency , we defined the expected drift velocity VD≡⟨lR⋅cos⁡θR ⟩⟨τR+τT⟩ , by the projection of the average run length along the migration direction over the average duration of runs and tumbles ( Dufour et al . , 2014 ) . This quantity reflects the effective run speed of run-and-tumble events that start running on a given location relative to the group . The drift velocity was found to decrease from the back to the front , crossing the group velocity VG in the middle of the group ( Figure 1F ) . This particular trend of VDz suggests a mean reverting behavior of bacteria in the group: the cells at the back drift faster than the group ( VD>VG ) , enabling the cells to catch up with the group; at the same time , the cells in the front drift slower than the group ( VD<VG ) , causing them to slow down and fall back ( Figure 1G ) . Such mean reverting process also results in sub-diffusion of individuals relative to the group ( Appendix 1—figure 3F ) , for which the mean square displacement ( MSD ) is constrained over time ( Appendix 1—figure 3G ) . The slope of the spatial extent of VD ( z ) , −r =−0 . 05min−1 , quantifies the speed at which individuals revert to its center . We noted that the spatial profile of the expected drift velocity VD ( z ) was different from the instantaneous velocity VIz . This is because the instantaneous velocity VIz defines the average speed of cells in a given time interval dt , which reflects the positional shift of a group of cells at a given position . However , the expected drift velocity VD ( z ) defines the average speed of run-and-tumble events that start tumbling at a given position . Since the run duration is explicitly modulated by the gradient of chemoattractant gz and is dependent on the chemotactic ability χ , VD ( z ) =χgz represents the drift velocity driven by the external stimuli ( Celani and Vergassola , 2010; de Gennes , 2004; Dufour et al . , 2014; Si et al . , 2012 ) . To understand how the spatially structured profile of the drift velocity VD impacts on the group migration , we first adopted a Langevin-type model that describes bacterial motions as an active Brownian particle driven by the expected drift velocity VD and a random force: dx=VDdt+ϵdW ( Berg , 2004 ) . In this model , the run-and-tumble random motions are considered as a Gaussian random force ϵdW , while the cell motions are imposed to a deterministic force VD ( Rosen , 1973; Rosen , 1974 ) . The strength of Gaussian noise can be estimated by the effective diffusion coefficient ϵ=2D , while the drift velocity is determined by the product of the perceived gradient gz and the chemotactic ability χ , VD=χgz . Such a stochastic description of bacterial motions has been proven equivalent to the classic Keller-Segel ( KS ) model that described the population dynamics of bacterial chemotactic group migration ( Keller and Segel , 1971a , Rosen , 1973 ) . In the moving coordinate , z=x − VGt , this Langevin-type model specifies that dz=VD ( z ) dt-VGdt+ϵdW . Thereby , the cell motions relative to the migrating group are modulated by two effective forces: one generated by VD ( z ) , which pushes the cells to catch up with the wave; and another generated by -VG , which drags the cells to fall behind the wave . These two ‘forces’ constrain the random motions of individuals in an effective potential well Uz ( Figure 2A ) . To estimate the spatial profile of the effective driven force VD ( z ) , we analyzed the KS model with moving ansatz ( supplementary text ) . Assuming the density profile ρz directly measured from experiments ( Figure 1B ) , we can deduce the chemoattractant concentration profile S ( z ) ( Appendix 1—figure 4D ) , as well as the perceived gradient gz ( Figure 2B ) . Since the perceived gradient gz is almost linear in the main part of the wave profile , we approximated it by gz≈g0+g1z ( Figure 2B , dashed line ) , which allows us to transform the stochastic model to an OU-type equation: ( 1 ) dz=χg1zdt+χg0-VGdt+ϵdW By this simplified model , we obtained a clear picture how individual behaviors are regulated relative to the group: cells are imposed to a driving force linearly dependent on their relative positions in the group , F ( z ) =χg1z+χg0-VG . This suggests that the random motions of bacteria are constrained in a parabolic moving potential well , Uz=12χg1z2+χg0-VGz+U0 ( Figure 2A ) , where U0 is set by U∞=0 . The position that minimizes Uz is also the balanced position , z0=-g0g1-VGχg1 , where the driving force is null F ( z0 ) =0 . Behind the balanced position , cells experience a pushed force to catch up the group . Cells would start to fall back once they exceed the balanced position , where the driving force becomes negative . Therefore , the decreasing profile of the driving force enable cells to perform mean reverting behaviors around the balanced position . In addition , the expected rate that cells tend to revert back to the balanced position is defined by the slope of the spatially dependent driving force , r=χg1 ( Figure 2B ) . Given the knowledge of individual behaviors , we studied the spatial distribution of population on the group migration . The OU model ( Equation 1 ) which describes the single-cell stochastic motions has an equivalent form , known as the Fokker-Planck equation ( Equation S19 ) which describes the spatial-temporal dynamics of cell density distribution . This population model provides a traveling wave solution with the mean position around z0 and standard deviation σ=ϵ2 χ g1 . From the solution , we noted that the effective driving force , as well as the drift velocity , has a negative slope ( g1<0 ) . The decreasing profile of the drift velocity suggests that the leading front of the group migrates as a pushed wave front ( Gandhi et al . , 2016; van Saarloos , 2003 ) . As a key feature of the pushed wave , the leading front of the group drops parabolically , which is much faster than that of diffusion front . This further leads to a tight density profile of group migration for a single phenotype population . To address whether this pushed wave front still holds in presence of phenotypic diversity , we further examined the above analysis with cells of diverse chemotactic abilities χi imposed to the same moving perceived gradient gz ( Figure 2B ) . Given a monotonically decreasing profile of perceived gradient gz , the driving force that each phenotypic individual experiences exhibits the same spatial dependency with the slopes depend on the intrinsic phenotypic properties of each phenotype ri=χig1 . This monotonic dependency means that the balanced positions z0 of the diverse phenotypes are orderly arranged . By the stochastic Langevin-type model with phenotypic diversity in chemotactic ability , we first confirmed that each phenotypic population migrates at a constant speed VG , following the moving gradient gz ( see supplementary text ) . The density profiles of cells with different χi follow the same shape but are spatially orderly aligned ( Figure 2C ) . Under the same moving gradient g ( z ) , the driving force χig ( z ) is phenotype-dependent , so that the bottom position of the potential well , z0 , i=-g0g1-VGχig1 , is also spatially arranged according to χi ( Figure 2D ) . As predicted by the OU model , the balanced positions z0 , i of different phenotypes increase with their chemotactic abilities χi ( Figure 2E , black line ) , while the standard deviation ( σi ) of the density profiles decreases with χi ( Figure 2E , blue line ) . We also confirmed the ordered structure of phenotypes by a particle-based model of the Langevin-type dynamics coupled with chemoattractant consumption ( Appendix 1—figure 5 ) . Therefore , under the spatially decreasing driving force , cells with phenotypic diversity perform the same type of mean reverting processes with spatially ordered mean positions . The spatial order of phenotypes does not directly promise a compact group migration with phenotypic diversity . By close examination of the density profile , we found that each phenotypic subpopulation propagates as a pushed wave front . We further calculated the total density profile of the entire migratory group with Gaussian distribution of chemotactic abilities under a decreasing linear gradient . Simulation reveals a pushed wave front for the combination of these subpopulations with different chemotactic abilities ( Figure 2F , insert ) . In addition , we checked that the width of the entire group maintains in a converged width over long time , suggesting that the pushed wave profile enables the migratory population with diversity to keep in a tight shape ( Figure 2F ) . We examined the migration profiles under other forms of perceived gradient gz : a constant perceived gradient and a spatially increasing perceived gradient . In the first case , individual bacteria of identical phenotype follow the diffusion process relative to the group , where the standard deviation of the population increasing with time by σ=2Dt ( Figure 2G , dashed line ) . Each phenotype subpopulation is expected to have a constant drift velocity over space . However , as the drift velocity would vary by the chemotactic ability , each subpopulation migrates in different group speeds , suggesting a compact group migration of diverse population cannot be maintained in a constant perceived gradient ( Figure 2G ) . In the latter case , when imposed to a spatially increasing driving force ( a pulled wave , by definition ) , individual cells display super-diffusion processes . The density profile easily diverges in this case ( Figure 2H ) . Therefore , we concluded that the pushed wave front , driven by a decreasing shape of driving force , enables a spatially ordered and compact pattern of phenotypes while on the move . Although the simplified OU-type model ( Equation1 ) represents a key aspect of the ordering mechanism of phenotypes , it does not detail the signaling processes of bacterial chemotaxis , such as receptor amplification , adaptation , and motor responses ( Sumpter , 2010 ) , and it cannot predict bacterial run-and-tumble behavior . To consolidate the proposed mechanism underlying the emergence of spatial orders from the individual random motions , we further performed agent-based simulations integrated with the chemotactic pathway , multi-flagella competition , and boundary effect in three dimensions ( 3D ) ( Dufour et al . , 2014; Jiang et al . , 2010; Sneddon et al . , 2012 ) . In the agent-based model , the attractant dynamics governed by diffusion and bacterial consumption is described by a reaction-diffusion equation ( Equation S2 ) as previous multiscale models ( Erban and Othmer , 2005; Xue and Othmer , 2009 ) . We constructed a population with multiple phenotypes defined by different chemotactic abilities χi , where χi was varied by changing the receptor gain N ( for details , see supplementary text ) . Since the receptor gain N only affects the amplification factor by which a cell responds to the gradient , the variation in bacterial motility ϵ is unchanged . As a result , a dense band of migrating cells that follow a self-generated moving chemoattractant gradient via consumption were recaptured as experiments ( Appendix 1—figure 6A ) . The phenotypes were spatially ordered as χ varies ( Appendix 1—figure 6B ) , and the velocity profile of each phenotype decreases from the back to front ( Appendix 1—figure 6C ) as predicted by the OU-type model . To better analyze the simulations , we simplified the simulation with a non-consumable attractant profile Sz moving with constant speed VG ( Appendix 1—figure 4D ) . Using this simplified model , we first checked that the mean positions of the density profiles of cells with different receptor gains N , as well as their peaks , were orderly aligned with respect to chemotactic ability χi . As an important advantage of the agent-based simulations , the model allowed us to analyze single-cell behavior during the ordered group migrations . For each phenotype i , the expected drift velocity VD , iz decreased along the density profile ( Figure 3A ) . Consistent with the ordered structure of the density profiles , the intersection between VD , iz and VG exhibited the same order of chemotactic ability χi ( Appendix 1—figure 7 ) . As the reversion rate ri=dVD , izdz showed a positive correlation to χi , cells with lower receptor gain N ( resulting in a smaller χ ) experienced a weaker reverting force toward the center ( Figure 3A insert ) . Thus , the effective moving potential , Uiz , which constrains the cells around the mean positions sorted by their chemotactic abilities , becomes flat for cells with lower chemotaxis ability χ ( Figure 3B; Long , 2019 ) . As a result , cells of each phenotype perform sub-diffusion , whereby the MSD along the migration coordinate relative to the group was bound at a level negatively correlated to χ ( Figure 3C ) . The width of the density distribution , as an effect of the reversion force , decreased with the reversion rate , as an approximate linear function . Using this agent-based model , we further obtained similar results for populations of different χi through adaptation time τ , or basal CheY protein level Yp0 , which determined the basic tumble bias TB0 ( Dufour et al . , 2014; Jiang et al . , 2010; Sneddon et al . , 2012; Appendix 1—figure 8 ) . To verify the model predictions on the individual behaviors of different phenotypes , we experimentally measured the trajectories of cells with different chemotactic abilities during the group migration . Specifically , we altered the chemotaxis abilities of cells by titrating the expression level of Tar , which is under the control of a small molecule inducer aTc ( Sourjik and Berg , 2004; Zheng et al . , 2016 ) ( see Materials and methods and Figure 4A ) . The variations in the expression of Tar would lead to different receptor gains in response to the Asp gradient ( Adler , 1966b; Adler , 1969; Sumpter , 2010 ) , but the tumble bias and growth rate would not change . Using the migration speed of the bacterial range expansion to quantify the chemotaxis ability of the titrated strains , we found that the chemotaxis ability increases with aTc concentration ( see Materials and methods and Appendix 1—figure 9 ) . The Tar-titrated cells labeled with yellow fluorescent protein ( YFP; strain JCY20 ) were added to the wild-type population at a ratio of 1 in 400 . Within the wild-type population , 1 in 50 cells was labeled with red fluorescent protein ( RFP ) ( strain JCY2 ) . As the Tar-titrated strain constituted a small portion of the pre-mixed population , we considered that the density profile of the population would be invariant to different levels of induction of Tar . The premixed population could generate collective group migration , similar to the wild-type population ( Appendix 1—figure 9 ) . The trajectories of the YFP-labeled cells were tracked to represent the behavior of cells with different chemotactic abilities , while the profile of wild-type cells with RFP was also measured to characterize the density distribution of the entire migratory population . By comparing the statistics of cells with different Tar expression levels , we found that the expected drift velocity VD , i ( z ) followed the same decreasing pattern from back to front ( Figure 4B ) . More importantly , as the Tar expression level ( chemotactic ability ) increased , the slope of the decreasing pattern increased , which was consistent with the model prediction shown in Figure 3A . The intersections between VD , i ( z ) and VG , as well as the peak positions and mean positions of each Tar-titrated density profile ( Figure 4C ) , shifted toward the front as the chemotactic ability increased ( as measured by the migration rate on agar plates; Cremer et al . , 2019; Liu et al . , 2019 ) . The VD cross point was always behind the peak position and the mean position ( Figure 4C ) , suggesting that cells were leaking behind . Moreover , the width of each Tar-titrated density profile ( defined by 2σi ) decreased as the reversion rate ri increased ( Figure 4D ) , consistent with the model results in Figure 3C . Thus , as the OU-type model predicts , the width of the density profile is controlled by the reversion rate determined by the chemotactic ability χi . In summary , coordinated behaviors with ordered spatial arrangements of phenotypes are abundant in a wide range of biological and human-engineered systems , and are believed to involve elaborate control mechanisms . For animal migrations , it is challenging to characterize simultaneously the computational strategy and behavior at the individual level so as to avoid averaging out phenotypic diversity , and the emergent behavior at the population level ( Couzin et al . , 2005; Couzin et al . , 2002; Vicsek and Zafeiris , 2012 ) . For bacterial chemotactic migration , cells with different phenotypes are spatially aligned based on their chemotactic abilities . This observation was explained as a self-consistent result with the decreasing profile of attractant predicted by KS model ( Fu et al . , 2018 ) . In this study , we demonstrated that the collective consumption of attractant by bacterial group generates a spatial structure of individual drift velocity along the migrating group profile . Such a spatial profile of drift velocity results a pushed wave front on population level , and maintains diverse phenotypes in a compact migration group . Moreover , this pushed wave front enables spatial modulation of individuals to perform mean reverting random motions around centers sequentially aligned by their chemotactic abilities , thereby giving rise to a spatially ordered pattern . Therefore , we demonstrated that the population order could emerge among diverse individuals that following the same rule of behavioral modulation . This strategy of self-organization does not require sophisticated communications ( Curatolo et al . , 2020; Karig et al . , 2018; Liu et al . , 2011; Payne et al . , 2013 ) nor other hydrodynamic interactions ( Chen et al . , 2017; Drescher et al . , 2011; Zhang et al . , 2010 ) among individuals . Our observation of the decreasing drift velocity can imply the effective perceived gradient that cells experience . We believe that this spatial profile is mainly contributed by the consumption of the chemoattractant . By using a Tar-only strain ( UU1624 ) ( Gosink et al . , 2006 ) , we demonstrated that the mutant could also generate a stable group migration in our experimental condition similar to the wild-type strain ( Appendix 1—figure 9G ) . This further suggests that the secretion of self-attractants is unlikely a necessary condition of collective group migration ( Cremer et al . , 2019; Fu et al . , 2018; Keller and Segel , 1971a ) , although there are doubts about the existence of self-attractants in high density ( Budrene and Berg , 1995 ) . The spatially dependent drift velocity provides a structured driving force of a migration group , resulting a pushed wave front . Pushed and pulled waves are determined by the spatial distribution of the spreading velocity of a propagation front ( van Saarloos , 2003; Figure 2F–H ) . The properties of pushed and pulled waves have been discussed in growth-driven range expansion ( Birzu et al . , 2018; Erm and Phillips , 2020; Gandhi et al . , 2016 ) , where the wave type is determined by density-dependent growth rates . A prominent example of pulled wave is known as the Fisher wave ( Fisher , 1937 ) , where the population expansion is driven by constant diffusion and logistic growth of individuals . However , in such biological systems , the spatial dependence of front speeds is hardly quantified in experiments . Here , by direct measurement of drift velocity on single-cell level , we identified the chemotactic migration group of bacteria as a pushed wave . This chemotaxis system would provide a unique multi-scale model to study further details of pushed wave . The spatially decreasing profile of driving force does not only cause an ordered pattern of phenotypes , but also results a pushed wave front that enables a negative feedback control on the propagating speed . This further allows a compact density profile for heterogenous population to migrate at the collective level . The advantageous to keep diversity in the pushed wave front was also reported in the growth-driven range expansion system ( Birzu et al . , 2018 ) . Our study revealed that , other than spatial regulation of fitness , the direct modulation of individual drift velocity in space could also maintain diversity in range expansion . Detailed analysis of the spatial structured driving force could also provide the limits of phenotype that is allowed in the group . In the migratory group , the same rule of behavioral modulation applied to cells with different phenotypes , such that the random motions of cells were bound by moving potential wells whose basins were sequentially aligned . However , it is noteworthy that cells could skip the potential wells from the back because the ‘driving force’ decreased again at the far back of the group ( Long , 2019 ) . This results in leakage of cells in the migratory group ( Holz and Chen , 1978; Novickcohen and Segel , 1984; Scribner et al . , 1974 ) . Phenotypes with weaker chemotactic abilities were located at the back of the group , where the effective potential well was shallower ( Figure 3C ) , allowing for more chances to skip . Thus , such collective migration selects bacteria with higher chemotactic abilities ( Liu et al . , 2021; Liu et al . , 2019 ) . The simple computational principle of behavioral modulation to allocate different phenotypes in the collective group is likely not limited to sensing the self-generated signal by consumption of attractant . A prominent example of trail-following migration ( Couzin and Krause , 2003; Helbing et al . , 1997 ) and a typical class of collective behavior is represented by a modified Langevin-type model , where individuals tracing the accumulated signal secreted by all participants ( Equation S20 ) can reproduce similar spatial-temporal dynamics of behavioral modulation , as well as ordered arrangements of phenotypes in the migratory group ( Appendix 1—figure 10 ) . Thus , this mechanism of matching individual abilities by the signal strength might provide an explanation of how other higher organisms organize ordered structures during group migration . The wild-type strain E . coli ( RP437 ) and its mutants were used in this study , where all plasmids were kindly provided by Dr Chenli Liu . Specifically , the Tar-titratable strain was constructed by recombineering according to previous research ( Zheng et al . , 2016 ) . Specifically , the DNA cassette of the Ptet-tetR-tar feedback loop was amplified and inserted into the chromosomal attB site by recombineering with the aid of plasmid pSim5 . The tar gene at the native locus was seamlessly replaced with the aph gene by using the same recombineering protocol . To color-code the strains , we use plasmids with chloramphenicol-resistant gene carrying yfp under constitutive promoter ( for JCY1 strain ) and pLambda-driven mRFP1 plasmids maintained by kanamycin ( for JCY2 ) . To color-code Tar-titratable strain ( JCY20 ) , a plasmid carrying yfp chloramphenicol-resistant gene was transformed into constructed Tar-titratable strain . The tar-only strain ( UU1624 ) was modified from RP734 and was kindly provided by Prof . Johan Sandy Parkinson . For bacterial culture , the M9 supplemented medium was used . The preparation of the M9 supplemented medium follows the recipe in previous study ( Fu et al . , 2018 ) : 1×M9 salts , supplemented with 0 . 4% ( v/v ) glycerol , 0 . 1% ( w/v ) casamino acids , 1 . 0 mM magnesium sulfate , and 0 . 05% ( w/v ) polyvinylpyrrolidone-40 . 1×M9 salts were prepared to be 5×M9 salts stock solution: 33 . 9 g/L Na2HPO4 , 15 g/L KH2PO4 , 2 . 5 g/L NaCl , 5 . 0 g/L NH4Cl . For migration experiments in the micro-channel , the M9 motility buffer was used . The recipe was:1×M9 salts , supplemented with 0 . 4% ( v/v ) glycerol , 1 . 0 mM magnesium sulfate , and 0 . 05% ( w/v ) polyvinylpyrrolidone-40 , 0 . 1 mM EDTA , 0 . 01 mM methionine , and supplemented with 200 µM aspartic acid . For the migration rate measurements , the M9 amino acid medium with 0 . 2% ( w/v ) agar was used to prepare swim plate ( Liu et al . , 2019 ) . The recipe was: 1× M9 salts , supplemented with 0 . 4% ( v/v ) glycerol , 1× amino acid , 200 µM aspartic acid , 1 . 0 mM magnesium sulfate , and 0 . 05% ( w/v ) polyvinylpyrrolidone-40 . 1× amino acid were prepared to be 5× amino acid stock solution: 4 mM alanine , 26 mM arginine ( HCl ) , 0 . 5 mM cysteine ( HCl·H2O ) , 3 . 3 mM glutamic acid ( K salt ) , 3 mM glutamine , 4 mM glycine , 1 mM histidine ( HCl·H2O ) , 2 mM isoleucine , 4 mM leucine , lysine , 1 mM methionine , 2 mM phenylalanine , 2 mM proline , threonine , 0 . 5 mM tryptophane , 1 mM tyrosine , 3 mM valine . All experiments were carried out at 30°C . Plasmids were maintained by 50 µg/mL kanamycin or 25 µg/mL chloramphenicol . The bacteria from frozen stock were streaked onto the standard Luria-Bertani agar plate with 2% ( w/v ) agar and cultured at 37°C overnight . Three to five separate colonies were picked and inoculated in 2 mL M9 supplemented medium for overnight culture with corresponding antibiotics to maintain plasmids . The overnight culture was diluted by 1:100 into 2 mL M9 supplemented medium the next morning . For Tar titration strains , related aTc were added in this step . When the culture OD600 reaches 0 . 2–0 . 25 , it was then diluted into pre-warmed 15 mL M9 supplemented medium so that the final OD600 was about 0 . 05 ( Liu et al . , 2019; Zheng et al . , 2020; Zheng et al . , 2016 ) . Bacteria were washed with the M9 motility buffer and were re-suspended in fresh M9 motility buffer to concentrate cell density at OD600 about 1 . 0 . Then , the wild-type strain and fluorescent strain were mixed with ratio of 400:1 before loaded in the microfluidic chamber ( Fu et al . , 2018; Saragosti et al . , 2011 ) . For Tar titration experiments , the wild-type strain ( RP437 ) was mixed with two fluorescent strains ( JCY2 and JCY20 ) by 400:8:1 . The microfluidic devices were fabricated with the same protocol and the same design as previous research ( Bai et al . , 2018; Fu et al . , 2018 ) , except that the capillary channel was designed longer than that of previous ones . The size of the main channel was 20 mm×0 . 6 mm× 0 . 02 mm and only one gate at the end of the channel was kept ( Appendix 1—figure 1A ) . Fluorescent cells were mixed with non-fluorescent cells by 1:400 for cell tracking in the dense band . Sample of mixed cells with density OD600 ≈1 . 0 was gently loaded into the microfluidic device and then the device was spun for 15 min at 3000 rpm in a 30°C environmental room so that almost 100 , 000-150 , 000 cells were placed to the end of the channel . After spinning , the microfluidic device was placed on an inverted microscope ( Nikon Ti-E ) equipped with a custom environmental chamber set to 50% humidity and 30°C . The microscope and its automated stage were controlled by a custom MATLAB script via the μManager interface ( Edelstein et al . , 2014; Fu et al . , 2018 ) . About 30 min after the sample loading , a 4× objective ( Nikon CFI Plan Fluor DL4X F , NA 0 . 13 , WD 16 . 4 mm , PhL ) was placed in the wave front for imaging . The fluorescent bacteria , seen as randomly picked samples of the migrating group , were captured continuously with frame rate of 9 fps in 10 min until they leave the view . Typical tracks are longer than 300 s . Time-lapsed images were acquired by a ZYLA 4 . 2MP Plus CL10 camera ( 2048 × 2048 arrays of 6 . 5 µm×6 . 5 µm pixels ) at 9 frames/s ( fps ) . An LED illuminator ( 0034R-X-Cite 110LED ) and an EYFP block ( Chroma 49003; Ex: ET500/X 20 , Em: ET535/30 m ) compose the lightening system . For the Tar titration experiments , the channel was first scanned with 10× objective ( CFI Plan Fluor DL 10× A , NA 0 . 30 , WD 15 . 2 mm , PH-1 ) enlighten by an LED illuminator ( 0034R-X-Cite 110LED ) through the RFP block ( Chroma 49005 , Ex: ET545/X 30 , Em: ET620/60 m ) and EYFP block channels for seven neighbored views around the migration group . These images were further combined to two large pictures of the RFP strains and YFP strains . The channel was scanned twice , respectively before and after the 10 min tracking of fluorescent Tar-titrated cells . The acquired movie was first analyzed with the U-track software package to identify bacteria and to get their trajectories ( Jaqaman et al . , 2008 ) . Then the tracks were labeled by run state and tumble state by a custom MATLAB package ( Waite et al . , 2016 ) using a previously described clustering algorithm ( Dufour et al . , 2016 ) . The group velocity VG was calculated by averaging the frame-to-frame velocity ( dt≈0 . 11s ) over all tracks and all time . The cell number for the first frame over a spatial bin of Δx=60μm and a channel section a=12 , 000 µm2 were calculated to get the density profile ρx , t=0=∑ix , ta⋅dx . The peak position of the first frame ( xpeakt=0 ) was then determined by the maximum of ρ ( x , t=0 ) . The position of each bacterium ( xi ( t ) ) was transformed to moving coordinate position zi by the group velocity VG and origin of the axis on the density peak by zi=xit-VGt-xpeakt=0 . Given the relative position of each cell , we recalculated the density profile in moving coordinate ρz=∑iza⋅dx . The width of the density profile was defined by two times the standard deviation of relative positions 2σ=21n-1∑i=1nzi-〈z〉2 . The spatial distribution of the instantaneous velocities 〈VIz〉 was calculated by averaging the velocity in spatial bin of Δz=240μm . A tumble-run event is the minimal element of bacterial behavior . The typical spatial scale of a tumble-run event is about 20 μm , which is much smaller than the spatial bin size chosen in this study ( 240 µm ) . The spatial distributions of run time 〈τRz〉 , tumble time 〈τTz〉 , and run length 〈lRz〉 were calculated by averaging the related values of all the events with tumbling position ( zT ) located in each spatial bin ( z ) . As the displacement of tumble is small , the tumbling position is approximately the starting position of runs . For each tumble-run event , we have the vector linking starting position and end position of the run . The running angle θR is then defined by the angle between run direction and the group migration direction . One can easily deduce all the other quantities with the formulations in Table 1 . Growth rates of Tar-titrated strains were calculated from exponential fitting ( R2>0 . 99 ) over measured curves of cell density ( OD600 ) with respect to time . A 250 mL flask with 20 mL M9 supplement medium were used . All measurements were performed in a vibrator of rotation rate of 150 rpm at 30 °C . OD600 was measured by a spectrophotometer reader every 25 min . Each strain has been measured for at least three times . The semi-solid agar plate was illuminated from bottom by a circular white LED array with a light box as described previously ( Liu et al . , 2011; Liu et al . , 2019; Wolfe and Berg , 1989 ) and was imaged at each 2 hr by a camera located on the top . As bacteria swimming in the plate forms ‘Adler ring’ , we used the first maximal cell density from the edge to define the moving edge of bacterial chemotaxis . The migration rate was then calculated from a linear fit over the data of edge positions in respect to time ( R2>0 . 99 ) . Details of the theoretical models and numerical simulations were presented in the appendix notes . In which , the Langevin equation was deduced and solved numerically with a particle-based simulation; the approximated OU-type equation and its traveling wave solution were deduced; an agent-based simulation of bacterial with chemotaxis pathway was performed following previous works ( Dufour et al . , 2014; Jiang et al . , 2010; Sneddon et al . , 2012 ) .	Organisms living in large groups often have to move together in order to navigate , forage for food , and increase their roaming range . Such groups are often made up of distinct individuals that must integrate their different behaviors in order to migrate in the same direction at a similar pace . For instance , for the bacteria Escherichia coli to travel as a condensed group , they must coordinate their response to a set of chemical signals called chemoattractants that tell them where to go . The chemoattractants surrounding the bacteria are unequally distributed so that there is more of them at the front than the back of the group . During migration , each bacterium moves towards this concentration gradient in a distinct way , spontaneously rotating its direction in a ‘run-and-tumble’ motion that guides it towards areas where there are high levels of these chemical signals . In addition to this variability , how well individual bacteria are able to swim up the gradient also differs within the population . Bacteria that are better at sensing the chemoattractant gradient are placed at the front of the group , while those that are worst are shifted towards the back . This spatial arrangement is thought to help the bacteria migrate together . But how E . coli organize themselves in to this pattern is unclear , especially as they cannot communicate directly with one another and display such diverse , randomized behaviors . To help answer this question , Bai , He et al . discovered a general principle that describes how single bacterial cells move within a group . The results showed that E . coli alter their run-and-tumble motion depending on where they reside within the population: individuals at the rear drift faster so they can catch up with the group , while those leading the group drift slower to draw themselves back . This ‘reversion behavior’ allows the migrating bacteria to travel at a constant speed around a mean position relative to the group . A cell’s drifting speed is determined by how well it moves towards the chemoattractant and its response to the concentration gradient . As a result , the mean position around which the bacterium accelerates or deaccelerates will vary depending on how sensitive it is to the chemoattractant gradient . The E . coli therefore spatially arrange themselves so that the more sensitive bacteria are located at the front of the group where the gradient is shallower; and cells that are less sensitive are located towards the back where the gradient is steeper . These findings suggest a general principle for how bacteria form ordered patterns whilst migrating as a collective group . This behavior could also apply to other populations of distinct individuals , such as ants following a trail or flocks of birds migrating in between seasons .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "physics", "of", "living", "systems", "microbiology", "and", "infectious", "disease" ]	2021	Spatial modulation of individual behaviors enables an ordered structure of diverse phenotypes during bacterial group migration
Mutations in the human BEST1 gene lead to retinal degenerative diseases displaying progressive vision loss and even blindness . BESTROPHIN1 , encoded by BEST1 , is predominantly expressed in retinal pigment epithelium ( RPE ) , but its physiological role has been a mystery for the last two decades . Using a patient-specific iPSC-based disease model and interdisciplinary approaches , we comprehensively analyzed two distinct BEST1 patient mutations , and discovered mechanistic correlations between patient clinical phenotypes , electrophysiology in their RPEs , and the structure and function of BESTROPHIN1 mutant channels . Our results revealed that the disease-causing mechanism of BEST1 mutations is centered on the indispensable role of BESTROPHIN1 in mediating the long speculated Ca2+-dependent Cl- current in RPE , and demonstrate that the pathological potential of BEST1 mutations can be evaluated and predicted with our iPSC-based ‘disease-in-a-dish’ approach . Moreover , we demonstrated that patient RPE is rescuable with viral gene supplementation , providing a proof-of-concept for curing BEST1-associated diseases . The human BEST1 gene encodes a protein ( BESTROPHIN1 , or BEST1 ) that is predominantly expressed in retinal pigment epithelium ( RPE ) ( Marmorstein et al . , 2000; Marquardt et al . , 1998; Petrukhin et al . , 1998 ) . To date , over 200 distinct mutations in BEST1 have been identified to associate with bestrophinopathies ( http://www-huge . uni-regensburg . de/BEST1_database/home . php ? select_db=BEST1 ) , which consist of at least five distinct retinal degeneration disorders , namely: Best vitelliform macular dystrophy ( BVMD ) ( Marquardt et al . , 1998; Petrukhin et al . , 1998 ) , autosomal recessive bestrophinopathy ( ARB ) ( Burgess et al . , 2008 ) , adult-onset vitelliform dystrophy ( AVMD ) ( Allikmets et al . , 1999; Krämer et al . , 2000 ) , autosomal dominant vitreoretinochoroidopathy ( ADVIRC ) ( Yardley et al . , 2004 ) , and retinitis pigmentosa ( RP ) ( Davidson et al . , 2009 ) . Patients with bestrophinopathies are susceptible to progressive vision loss for which there is currently no treatment available . Therefore , understanding how disease-causing mutations affect the biological function of BEST1 in the retina is critical for elucidating the pathology of bestrophinopathies and developing rational therapeutic interventions . A clinical feature of bestrophinopathies associated with BEST1 mutations is abnormal electrooculogram ( EOG ) light peak ( LP ) , measured by the maximum transepithelial potential produced by RPE upon light exposure ( Boon et al . , 2009; Marmorstein et al . , 2009 ) . LP is believed to represent a depolarization of the basolateral membrane of RPE due to activation of a Cl- conductance triggered by changes in intracellular Ca2+ concentration ( [Ca2+]i ) ( Fujii et al . , 1992; Gallemore and Steinberg , 1989; Gallemore and Steinberg , 1993 ) . The simplest hypothesis about the origin of this ion conductance is that it is generated by Ca2+-activated Cl- channels ( CaCCs ) . However , the existence of Ca2+-dependent Cl- current on the plasma membrane of RPE has not yet been directly demonstrated , let alone the identity of the participating channel ( s ) . BEST1 localizes to the basolateral membrane of RPE ( Marmorstein et al . , 2000 ) , and has been functionally identified as a CaCC in heterologous expression studies ( Hartzell et al . , 2008; Kane Dickson et al . , 2014; Sun et al . , 2002; Tsunenari et al . , 2003; Xiao et al . , 2008; Yang et al . , 2014b ) . Consequently , whether or not BEST1 conducts Ca2+-dependent Cl- currents responsible for LP in RPE has been a long-standing question in the field ( Hartzell et al . , 2008; Johnson et al . , 2017 ) . Best1 knock-out mice do not have any retinal phenotype or Cl- current abnormality ( Marmorstein et al . , 2006; Milenkovic et al . , 2015 ) , suggesting that either BEST1 is not the Cl- conducting channel in RPE , or that there are fundamental differences between humans and mice regarding the genetic bases for this electrophysiological response . So far only two studies investigated the Cl- channel function of endogenous BEST1 in human RPE . Although both studies demonstrated that Cl- secretions were partially impaired in iPSC-RPEs ( RPE cells differentiated from induced pluripotent stem cells ) derived from BEST1 patients compared to those from healthy donors ( Milenkovic et al . , 2015; Moshfegh et al . , 2016 ) , whether or not the CaCC function of BEST1 is involved remains unknown . The first study measured volume-regulated Cl- current without testing the involvement of Ca2+ , and used only one WT iPSC-RPE as the control which may not be representative ( Johnson et al . , 2017; Milenkovic et al . , 2015 ) . The second study , by our group , utilized anion sensitve fluorescent dyes to detect changes in Ca2+-stimulated Cl- secretion , which is not a direct measurement of CaCC activity ( Moshfegh et al . , 2016 ) . As BEST1 has also been suggested to regulate intracellular Ca2+ homeostasis by controlling intracellular Ca2+ stores on the endoplasmic reticulum ( ER ) membrane and/or modulating Ca2+ entry through L-type Ca2+ channels ( Barro-Soria et al . , 2010; Constable , 2014; Gómez et al . , 2013; Neussert et al . , 2010; Singh et al . , 2013; Strauß et al . , 2014 ) , our observations could potentially reflect BEST1’s role as a regulator of Ca2+ signaling rather than as a CaCC . Moreover , two recent reports argued that other CaCCs rather than BEST1 are responsible for Ca2+-stimulated Cl- current based on results from porcine and mouse RPEs , and the human RPE-derived ARPE-19 cell line ( Keckeis et al . , 2017; Schreiber and Kunzelmann , 2016 ) . Overall , the physiological role of BEST1 in human RPE and the pathological mechanisms of BEST1 disease-causing mutations are still poorly understood . Here for the first time , we directly measured Ca2+-dependent Cl- currents on the plasma membrane of human RPEs by whole-cell patch clamp , evaluated the physiological influence of two distinct ARB patient-derived BEST1 mutations in this context , and demonstrated rescue of mutation-caused loss of function by complementation . We further investigated the impacts of the two disease-causing mutations on the function and structure of BEST1 by electrophysiological and crystallographic approaches , respectively , and discovered mechanistic bases correlated with patient clinical phenotypes . Our results provide definitive evidence that the CaCC activity of BEST1 is indispensable for Ca2+-dependent Cl- currents in human RPE , reveal the molecular mechanisms of two BEST1 patient mutations , and offer a proof-of-concept for curing BEST1-associated retinal degenerative diseases . Reduced LP is a pathognomonic phenotype associated with BEST1 mutations in bestrophinopathy patients ( Boon et al . , 2009; Marmorstein et al . , 2009 ) . Although LP is believed to be mediated by surface Ca2+-dependent Cl- current in RPE , the existence of the current on the plasma membrane of RPE cells has not been directly demonstrated , let alone the putative physiological role of BEST1 as a contributor to the current . To address these deficits , we generated iPSC-RPEs from the skin fibroblasts of two BEST1 WT donors ( Figure 1—figure supplement 1A , B ) . We first examined the subcellular localization of BEST1 by fluorescent co-immunostaining of the channel together with a plasma membrane marker ( zonula occludens-1 , ZO-1 ) and a nucleus marker ( Hoechst ) followed by confocal microscopy . We found that BEST1 localized on the plasma membrane of iPSC-RPE ( Figure 1A , and Figure 1—figure supplement 1C ) . We examined the Ca2+-dependent Cl- current amplitudes on the plasma membrane of RPE using whole-cell patch clamp across a range of free [Ca2+]i ( Figure 1B–D , and Figure 1—figure supplement 1D ) . Currents were tiny ( < 5 pA/pF ) when [Ca2+]i was 0 ( Figure 1B , C ) , and increased in amplitude as [Ca2+]i was raised from 100 nM to 4 . 2 μM , peaking at 358 ± 15 pA/pF at 1 . 2 and 4 . 2 μM [Ca2+]i ( Figure 1B–D , Figure 1—figure supplement 1D , and Figure 1—source data 1 ) . The measured currents were inhibited by the Cl- channel blocker niflumic acid ( NFA ) ( Figure 1—figure supplement 1D ) , demonstrating that these were indeed Ca2+-dependent Cl- currents . A plot of peak current ( evoked with a +100 mV step pulse ) as a function of [Ca2+]i displayed robust Ca2+-dependent activation with the half maximal effective concentration ( EC50 ) of Ca2+ at 455 nM . Similar Ca2+-dependent Cl- current profiles were recorded in iPSC-RPEs derived from two independent BEST1 WT donors , and in iPSC-RPEs from two distinct clonal iPSCs of the same donor ( Figure 1—figure supplement 1 , and Figure 1—source data 1 ) . These results provide the first direct measurement of Ca2+-dependent Cl- currents on the plasma membrane of RPE . To test if the status of BEST1 and the properties of surface Ca2+-dependent Cl- current in iPSC-RPE represent those in real RPE , we conducted the same set of experiments in fetal human RPE ( fhRPE ) . Consistent with the results from iPSC-RPEs , BEST1 was plasma membrane enriched ( Figure 2A ) , and a similar pattern of Ca2+-dependent Cl- currents was recorded in fhRPEs from two independent fetuses ( Figure 2B–E ) . Interestingly , despite their comparable initial and peak amplitudes , the Ca2+-dependent Cl- current in fhRPEs displayed a lower Ca2+ sensitivity compared to that in iPSC-RPEs ( EC50 1 . 7 μM vs . 455 nM , Figure 2D ) , which may reflect the different requirement of LP generation in RPE during different developmental stages . Overall , the subcellular localization of BEST1 and the properties of Ca2+-dependent Cl- current in iPSC-RPE resemble those in fhRPE , validating iPSC-RPE as a powerful platform to study the influence of BEST1 mutations on RPE surface Ca2+-dependent Cl- currents . It is worth to note that during patch clamp recording with fhRPE , when the pipet solution contained high ( 18 μM ) [Ca2+]i , the currents ran up after patch break with a half-time of ~2 . 5 min and reached a plateau that was on average 7 . 8-fold greater than the initial current ( Figure 2—figure supplement 1A–C ) . In contrast , when the pipet solution contained low ( 0 . 6 μM ) [Ca2+]i , the currents remained stable after patch break ( Figure 2—figure supplement 1C ) . Unlike the other bestrophinopathies caused by autosomal dominant mutations in BEST1 , ARB is associated with recessive mutations . Patients with ARB are characterized by progressive generalized rod-cone degenerations , typically with a visual acuity reading around 20/40 in the first decade of life , and their vision progressively worsens over time ( Burgess et al . , 2008; Johnson et al . , 2017 ) . In this study , we focused on two diagnosed ARB patients from independent families . Both patients exhibit typical ARB phenotypes in fundus autofluorescence imaging , spectral domain optical coherence tomography ( SDOCT ) and full-field electroretinography ( ERG ) ( Figure 3A–C ) . Unlike EOG which mainly represents the electrical responses of RPE ( Figure 3—figure supplement 1 ) , ERG measures the overall activity of various cell types in the retina . Patient 1 , a 12-year-old otherwise healthy boy , who has a previously described homozygous c . 821C > G; p . P274R mutation in BEST1 ( Fung et al . , 2015; Kinnick et al . , 2011 ) , showed reduced visual acuities at 20/60 and 20/70 in the right and left eye , respectively . Color fundus showed bilateral , confluent curvilinear subretinal yellowish vitelliform deposits to the optic disks , which over 3 years of follow-up became more multifocal and dispersed to involve the nasal retinae ( Figure 3A , left ) . SDOCT discovered bilateral , multifocal serous retinal detachments involving the maculae and cystoid changes in the macula ( Figure 3B , left ) . Maximum response of ERG b-wave ( amplitudes between a- and b-wave ) were 132 . 6 μV and 194 . 4 μV in the right and left eye , respectively , contrasting 355 μV ( median value ) in healthy teenagers tested in the same device ( Figure 3C , left ) . Patient 2 , a 72-year-old otherwise healthy man , who has a homozygous c . 602T > C; p . I201T mutation in BEST1 , showed a dropped vision acuity at 20/40 in the right eye , and 20/200 in the left eye mainly due to aging-caused retinal atrophy . His color fundus presented less vitelliform deposits compared with patient 1 , and aging-caused dispersed punctate fleck lesions in the left eye ( Figure 3A , right ) . SDOCT showed milder cystic degeneration compared to that in Patient 1 ( Figure 3B , right ) . Maximum responses of ERG b-waves were 103 . 2 μV and 79 . 6 μV in the right and left eye , respectively , contrasting 287 μV ( median value ) in age-matched healthy people ( Figure 3C , right ) . In summary , even though ARB has progressed for 60 years longer , patient 2 has better vision acuity ( in his more relevant right eye ) , less vitelliform deposit , milder cystic degeneration , and better responses to visual stimuli , suggesting that the I201T mutation is less severe than the P274R mutation . If the recorded Ca2+-dependent Cl- current is responsible for LP , it is logically speculated to be impaired in BEST1 patient iPSC-RPEs , because reduced LP is a clinical feature in BEST1 patients . To directly examine the physiological impact of BEST1 mutations on Ca2+-dependent Cl- current in RPE , iPSCs were derived from the patients’ skin cells and then differentiated to iPSC-RPEs . RPE-specific marker proteins RPE65 ( retinal pigment epithelium-specific 65 kDa protein ) and CRALBP ( cellular retinaldehyde-binding protein ) displayed similar expression levels in the BEST1 WT and two patient-derived iPSC-RPEs by western blot ( Figure 4A ) , confirming the mature status of iPSC-RPEs . Patient iPSC-RPE carrying the BEST1 P274R mutation showed a similar overall BEST1 expression level compared to that in WT iPSC-RPE ( Figure 4A ) in western blot , but exhibited diminished BEST1 antibody staining on the plasma membrane ( Figure 4B , top ) , indicating that the subcellular localization of the channel was severely impaired by the P274R mutation . Strikingly , tiny currents ( < 6 pA/pF ) were detected in P274R patient iPSC-RPE at all tested [Ca2+]i by whole-cell patch clamp ( Figure 4C-E p , and Figure 1—source data 1 ) , indicating that the P274R mutation abolishes Ca2+-dependent Cl- current in RPE . Furthermore , both the membrane localization of BEST1 and the Ca2+-dependent Cl- current were rescued in P274R patient iPSC-RPE by complementation with WT BEST1-GFP expressed from a BacMam baculoviral vector ( Figure 4E , and Figure 4—figure supplement 1A , B , C ) . These results demonstrated that functional BEST1 is necessary for generating Ca2+-dependent Cl- current in human RPE . On the other hand , patient iPSC-RPE carrying the BEST1 I201T mutation showed a similar overall BEST1 level compared to that in WT iPSC-RPE ( Figure 4A ) , and normal BEST1 antibody staining on the plasma membrane ( Figure 4B , bottom ) . However , I201T patient iPSC-RPE displayed robust but significantly decreased Ca2+-dependent Cl- currents compared to those in WT iPSC-RPE ( Figure 4C , D , F , and Figure 1—source data 1 ) . Notably , when current amplitudes were normalized , the pattern of Ca2+ response was similar in WT and I201T iPSC-RPEs ( EC50 455 nM vs . 526 nM , Figure 4F , and Figure 4—figure supplement 1D ) , indicating that the Ca2+ sensitivity of surface Cl- current in RPE is not altered by the I201T mutation . Taken together , our results showed that the P274R mutation leads to a ‘null’ phenotype of Ca2+-dependent Cl- current in RPE associated with a loss of BEST1 plasma membrane enrichment , while the seemingly milder I201T mutation causes reduced Cl- current in RPE without altering Ca2+ sensitivity of the current or subcellular localization of BEST1 . Importantly , the P274R patient exhibits a more severe retinal phenotype compared to the I201T patient , suggesting a strong correlation between the status of BEST1 , the functionality of RPE surface Ca2+-dependent Cl- current , and retinal physiology . As BEST1 is a CaCC located on the plasma membrane of RPE , the next important question is whether the defective Ca2+-dependent Cl- current in BEST1 patient iPSC-RPEs truly reflects deficiency of the BEST1 channel activity . To directly examine the influence of the disease-causing mutations on BEST1 , WT and mutant BEST1 channels were individually introduced into HEK293 cells , which do not have any endogenous CaCC on the plasma membrane ( Figure 5—figure supplement 1A , B ) . Western blot confirmed that both WT and the mutant channels were expressed at similar levels after transient transfection ( Figure 5—figure supplement 1C ) . As previously reported , HEK293 cells expressing WT BEST1 displayed robust Ca2+-dependent currents markedly inhibited by NFA ( Figure 5—figure supplement 1B ) , indicating that they were Ca2+-dependent Cl- currents ( Hartzell et al . , 2008 ) . Consistent with the results in iPSC-RPE , HEK293 cells expressing the P274R mutant yielded no current , while cells expressing the I201T mutant displayed significantly smaller current amplitude compared to that of WT at 1 . 2 μM [Ca2+]i , where HEK293 cells expressing WT BEST1 conduct peak current amplitude ( Figure 5A , B ) ( Hartzell et al . , 2008 ) . As HEK293 cells represent a ‘blank’ background , the recorded Ca2+-dependent Cl- currents are genuinely generated from transiently transfected BEST1 channels . Therefore , the two disease-causing mutations lead to distinct defects of the BEST1 channel activity that match the defects of Ca2+-dependent Cl- current in iPSC-RPEs , strongly suggesting that BEST1 is the bona fide CaCC on the plasma membrane of RPE mediating Ca2+-dependent Cl- current for LP . As an ion channel , how could BEST1 go wrong with the disease-causing mutations ? Multiple mechanisms may exist , including massive disruption of the channel structure , alterations in single channel activity , and dysregulation of the channel ( e . g . expression ) . We sought to find critical clues from the channel structure to answer this question . Since the structure of BEST1 has not been solved , we generated a three-dimensional human homology model based on our previously solved Klebsiella pneumoniae bestrophin ( KpBest ) structure and a chicken bestrophin1 ( cBest1 ) structure ( Kane Dickson et al . , 2014; Moshfegh et al . , 2016; Yang et al . , 2014b ) ( Figure 6A , Figure 6—figure supplement 1A , B , and Figure 6—figure supplement 2 ) . In this BEST1 model , P274 locates at the N-terminal of helix S4a ( Figure 6A , B , Figure 6—figure supplement 1A , B , and Figure 6—figure supplement 2 ) . The presence of Pro in alpha helices normally promotes thermostability of the membrane protein ( Reiersen and Rees , 2001 ) . The restricted torsion angle for the N–Cα bond of Pro allows only a limited number of conformations and imposes stress on secondary structures in proteins . Substitution of Pro with Arg will release the restrictions and induce instability of local structure , predicting a dramatic disruption of the channel . It should be noted that a Pro to Arg mutation based on the structure model would result in a steric clash between this amino acid and helix S3b , thereby highlighting the major contribution of Pro in the structure ( Figure 6—figure supplement 1D ) . On the other hand , I201 resides in a loop between S2h and S3a ( Figure 6A , B , Figure 6—figure supplement 1A , B , and Figure 6—figure supplement 2 ) , surrounded by hydrophobic residues V114 , A195 , L207 , and L210 ( Figure 6C ) , which are conserved among species and thus probably important for the channel function ( Figure 6—figure supplement 1C ) . As the Ile to Thr substitution changes a hydrophobic residue to a polar residue , which weakens the hydrophobic interactions , this mutation may change the channel property by altering the local interplays between spatially adjacent subunits , but will unlikely disrupt the channel structure as its localization on a loop renders flexibility . Importantly , the potential influence of the I201T mutation on the channel function is underlined by its proximity to I205 ( Figure 6A , C ) , a putative activation/permeation gate and the narrowest exit along the ion conducting pathway ( Figure 6A , B ) ( Yang et al . , 2014b ) . Sequence alignment reveals that BEST1 P274 is identical while I201 has a highly conservative substitution in KpBest ( P239 and L177 , respectively , Figure 6—figure supplement 1C , and Figure 6—figure supplement 2 ) , prompting us to test the predictions from the BEST1 homology model with the corresponding KpBest mutants ( P239R and L177T , respectively ) expressed from E . coli . During protein purification , we noticed that the yield of pentameric KpBest P239R was significantly less compared to that of KpBest WT or L177T ( Figure 7A ) , consistent with the prediction that P274R causes massive disruption , and thus instability , to the channel structure . Purified KpBest mutant proteins were set for crystal growing . While no crystal was grown with KpBest P239R , well-diffracted KpBest L177T crystals were obtained under the same condition as KpBest WT ( Yang et al . , 2014b ) , and the structure was solved to 3 . 1 Å resolution ( Figure 8—source data 1 ) . The KpBest L177T structure mirrors that of KpBest WT , with all-atom alignment RMSD ( root-mean-square deviation ) in a protomer 0 . 4 Å ( Coot LSQ superpose ) . However , superposition of KpBest WT with the L177T mutant based on the alignment of single chain residues 174–180 showed an obvious shift of the TM region ( Figure 8 , and Figure 8—figure supplement 1 ) . These results strongly support our structural predictions on the BEST1 P274R and I201T mutations . We next assessed the influence of the disease-causing mutations on BEST1 single channel activity . To circumvent the unavailability of purified human BEST1 , we utilized the corresponding KpBest P239R and L177T mutants . As previously described ( Yang et al . , 2014b ) , purified KpBest channels were fused into planar lipid bilayer with 150 mM NaCl in both the trans ( internal ) and cis ( external ) solutions , and single channel currents were recorded with KpBest WT at 80 mV with mean amplitude of 5 . 5 pA ( Figure 7B ) . By contrast , no currents were obtained with KpBest P239R , while currents with reduced unitary conductance ( mean amplitude 1 . 5 pA ) were recorded with KpBest L177T ( Figure 7B , C ) , suggesting that the BEST1 P274R and I201T mutations result in a complete and partial loss of single channel activity , respectively . Taken together , we concluded that P274R is a null mutation that abolishes both plasma membrane localization and channel activity of BEST1 due to structural disruption , whereas I201T is a partial loss-of-function mutation that retains plasma membrane localization and Ca2+ sensitivity of BEST1 caused by minor structural alterations . Here , we first proved the existence of Ca2+-dependent Cl- currents on the plasma membrane of human RPE by whole-cell patch clamp . Then we comprehensively examined two BEST1 disease-causing mutations ( P274R and I201T ) derived from ARB patients in an interdisciplinary platform , including whole-cell patch clamp with patient-derived iPSC-RPEs and HEK293 cells , immunodetection of endogenous BEST1 in iPSC-RPEs , lipid bilayer with purified bacterial bestrophin proteins , and structural analyses with human models and bacterial homolog crystal structures ( Table 1 ) . Collectively , our results illustrated the physiological influence of these two mutations on RPE surface Ca2+-dependent Cl- current and the BEST1 channel function , and provided structural insights into their disease-causing mechanisms: the P274R mutation abolishes Ca2+-dependent Cl- current in vivo , likely due to disruption of the BEST1 channel structure; while the I201T mutation partially impairs Ca2+-dependent Cl- current in vivo , likely due to non-disruptive structural alteration ( Table 1 ) . The structure of BEST1 has not been solved , and only two bestrophin homolog structures- KpBest and cBest1 , were reported in previous studies ( Kane Dickson et al . , 2014; Yang et al . , 2014b ) . We used both KpBest crystal structures and human homology models mainly based on cBest1 to analyze the possible structural alterations in BEST1 caused by the patient-specific mutations . Results from the two methods are consistent with each other and with functional data . Moreover , it has been proposed that disease mutations may result in wrongly numbered oligomers rather than the correct pentamer formed by WT BEST1 ( Johnson et al . , 2017 ) . The structure of KpBest I177T suggests that the BEST1 I201T mutation does not alter the pentameric conformation of the channel . Although decreased LP in BEST1 patients has been attributed to aberrant RPE surface Ca2+-dependent Cl- current , how BEST1 disease-causing mutations physiologically influence Ca2+-dependent Cl- current in RPE has not been directly examined . Most previous studies investigated the anion channel function of BEST1 in transiently transfected cell lines ( Hartzell et al . , 2008; Johnson et al . , 2017 ) , while the only two studies done in human RPE by other groups did not directly examine Ca2+-dependent Cl- current: one measured transepithelial potential in fhRPE expressing exogenous BEST1 on virus ( Marmorstein et al . , 2015 ) , and the other investigated volume-dependent current ( Milenkovic et al . , 2015 ) . We recently used anion sensitive fluorescent dyes to compare Ca2+-stimulated Cl- secretion in BEST1 WT and mutant donor iPSC-RPEs , but neither the surface Cl- current nor Ca2+ sensitivity was directly measured ( Moshfegh et al . , 2016 ) . Here we clearly demonstrated with whole-cell patch clamp that the surface Ca2+-dependent Cl- current in patient-derived iPSC-RPEs is completely abolished and significantly reduced by the P274R and I201T mutations , respectively , providing the first direct evidence that BEST1 disease-causing mutations impair Ca2+-dependent Cl- current in human RPE . Our results strongly argue that BEST1 is the CaCC mediating Ca2+-dependent Cl- current in human RPE , because: 1 ) the surface Ca2+-dependent Cl- current is completely defective in iPSC-RPE with the P274R mutation , which generates an essentially ‘null’ BEST1 channel with loss of plasma membrane enrichment in RPE and no ion conductivity in HEK293 cells and bilayer ( KpBest P239R ) , suggesting that BEST1 is indispensable for Ca2+-dependent Cl- current in RPE; 2 ) the I201T mutation results in significantly reduced conductivity of the channel in both HEK293 cells and bilayer ( KpBest L177T ) , and concomitantly leads to much smaller Ca2+-dependent Cl- currents in the patient iPSC-RPE , in which the mutant BEST1 channels are still expressed and enriched on the plasma membrane , suggesting that the CaCC function of membrane located BEST1 orchestrates Ca2+-dependent Cl- current in RPE; 3 ) the I201T mutation does not affect the Ca2+ sensitivity of Cl- current in RPE , consistent with the non-involvement of I201 in Ca2+ binding according to the cBest1 crystal structure model ( Kane Dickson et al . , 2014 ) . The simplest and most logical conclusion based on our results is that BEST1 functions as the surface CaCC to generate Ca2+-dependent Cl- current in human RPE . A recent report with primary mouse RPE and the human RPE-derived ARPE-19 cell line suggested that TMEM16B is the CaCC responsible for Ca2+-stimulated Cl- current in those cells ( Keckeis et al . , 2017 ) , while a role of TMEM16A was proposed by another study using Cl- channel blockers in porcine RPE ( Schreiber and Kunzelmann , 2016 ) . It should be noted that Best1 knockout mice do not have any retinal abnormality or aberrant Cl- current ( Marmorstein et al . , 2006; Milenkovic et al . , 2015 ) , unlike the phenotypes seen with human BEST1 mutations , suggesting different genetic requirements for retinal physiology among species . Moreover , the expression of BEST1 in the ARPE-19 cell line may be different from that in iPSC-RPE and fhRPE ( Marmorstein et al . , 2000 ) . In any case , our results do not completely exclude the role of other CaCCs in contributing to Ca2+-dependent Cl- current in human RPE . Besides its CaCC function on the basolateral plasma membrane of RPE , other roles of BEST1 have also been suggested including HCO3- channel , volume-regulated Cl- channel , regulator of Ca2+ channels , and Ca2+ sensor on the endoplasmic reticulum membrane ( Barro-Soria et al . , 2010; Burgess et al . , 2008; Fischmeister and Hartzell , 2005; Gómez et al . , 2013; Qu and Hartzell , 2008; Rosenthal et al . , 2006; Yu et al . , 2008 ) . Here we focused on the CaCC function of BEST1 in conducting surface Ca2+-dependent Cl- current , which directly gives rise to LP , but did not exclude any indirect contribution of BEST1 to LP through its non-CaCC function ( s ) . For instance , BEST1 may affect a downstream CaCC ( e . g . TMEM16A or TMEM16B ) through regulating intracellular Ca2+ . Interestingly , we found remarkable differences in the Ca2+ sensitivity of Cl- current in different cell types . The lower Ca2+ sensitivities in fhRPE ( EC50 1 . 7 μM ) compared to that in BEST1 WT iPSC-RPE ( EC50 455 nM ) may result from the cells’ different developmental stages , considering that fetuses do not have a fully functional visual system and therefore probably only need less sensitive CaCCs in their RPEs . The higher Ca2+ sensitivity of heterologously expressed BEST1 in HEK293 cells ( EC50 ~150 nM ) has been reported in previous studies ( Lee et al . , 2010; Xiao et al . , 2008 ) , while purified cBest1 displays an even smaller EC50 of 17 nM in bilayer ( Vaisey et al . , 2016 ) . Considering the role of BEST1 as the CaCC in RPE , the significant difference of Ca2+ sensitivities may reflect intrinsic differences between RPE where BEST1 is endogenously expressed and other experimental systems with overexpressed or purified proteins . It is also possible that in native RPE , BEST1 senses Ca2+ not only through direct interaction as suggested by the cBest1 model ( Kane Dickson et al . , 2014 ) , but also indirectly via a third-party Ca2+-sensor protein or by posttranslational modification mechanisms ( e . g . phosphorylation ) ( Hartzell et al . , 2008 ) , to function properly under the sophisticated physiological environment . Notably , the Ca2+ sensitivity observed in BEST1 WT iPSC-RPE ( EC50 455 nM ) is at levels more comparable to physiological conditions than that detected in HEK293 cells over-expressing BEST1 ( EC50 ~150 nM ) , let alone cBest1 in bilayer ( EC50 17 nM ) , because basal [Ca2+]i in the human body is typically around 100 nM , meaning that CaCCs with a EC50 near or lower than 100 nM would be readily activated even in resting cells . In regard to the clinical treatment of bestrophinopathies , our study provided an important proof-of-concept for treating ARB caused by BEST1 recessive mutations , as the loss of Ca2+-dependent Cl- current in the ‘null’ BEST1 P274R iPSC-RPE was rescued by viral expression of WT BEST1 . It will be very intriguing to see if ARB patients can be treated by gene therapy delivering functional WT BEST1 to their RPEs . Notably , most of the BEST1 patient-specific mutations are dominant , so that the mutant BEST1 alleles in these cases may be more functionally defective and/or structurally disruptive compared to recessive mutant alleles in ARB patients . Although it is formally possible that overexpression of WT BEST1 can also rescue , in a dominant-negative matter , aberrant Ca2+-dependent Cl- current in RPE caused by BEST1 dominant mutations , further studies will be needed to test this premise . On the other hand , recessive BEST1 mutations from ARB patients provide a unique opportunity to analyze and connect the structure , function and physiological role of BEST1 in a ‘clean’ manner , as only the mutant BEST1 proteins are present in patients and all our experimental systems . By contrast , the co-existence of both WT and mutant BEST1 proteins in the cases of dominant mutations complicates the functional-structural analyses for several reasons: ( 1 ) as the pentameric bestrophin channels consist of five protomers , different numbers ( 0–5 ) of BEST1 mutant protomers could potentially be assembled to a BEST1 pentamer and impact the channel structure and function; ( 2 ) although the ratio of endogenous WT to mutant BEST1 proteins is key to determine the composition of pentameric BEST1 channels in patients , this critical factor cannot be determined by either western blot or immunostaining , as the BEST1-specific antibody cannot distinguish WT and mutant BEST1 proteins; ( 3 ) crystallographic studies with the BEST1 dominant mutant proteins only reflect homopentamers consisting of all five mutant protomers , but not heteropentamers with 1–4 mutant protomers; ( 4 ) it could be technically challenging to rescue phenotypes caused by dominant mutations , and thus hard to draw a clear conclusion . Nevertheless , we are actively investigating BEST1 dominant mutations using the pipelines established in this work with necessary modifications and cautions . Primary fibroblasts cells from donors were reprogrammed into pluripotent stem cells using the CytoTune-iPS 2 . 0 Sendai Reprogramming Kit ( Thermo Fisher Scientific , A16517 ) , and immunocytofluorescence assays were performed for scoring iPSC pluripotency following the previously published protocol ( Li et al . , 2016 ) . In brief , a panel of antibodies ( 1:200 , abcam , ab109884 ) against four standard pluripotency markers SOX2 , Tra-1–60 , SSEA4 and Nanog were applied to characterize the iPSCs from all the subjects enrolled in this study . Hoechst staining was applied to detect nuclei . Secondary antibodies were Alexa Fluor 488 conjugated goat anti-rabbit or Alexa Fluor 555 conjugated goat anti-mouse IgG ( 1:1 , 000; Life Technologies ) . Images for all antibody labels were taken under the same settings with fluorescence microscope ( NIKON , Eclipse , Ts2R ) . All iPSC lines were maintained in mTeSR-1 medium ( STEMCELL Technologies , 05850 ) and passaged every 3–6 days . The morphology and nuclear/cytoplasmic ratio of the iPSC lines were closely monitored to ensure the stability . To verify genome integrity , all the iPSC lines in this study were sent for karyotyping by G-banding at the Cell Line Genetics ( Wisconsin , USA ) . iPSC differentiation started at passage 4 for all iPSC lines . For differentiation , iPSC colonies were cultured to confluence in 6-well culture dishes ( Costar , Corning , Corning , NY ) pretreated with 1:50 diluted matrigel ( CORNING , 356230 ) in differentiation medium consisting of Knock-Out ( KO ) DMEM ( Thermo Fisher Scientific , 10829018 ) , 15% KO serum replacement ( Thermo Fisher Scientific , 10829028 ) , 1% nonessential amino acids ( Thermo Fisher Scientific , 11140050 ) , 2 mM glutamine ( Thermo Fisher Scientific , 35050061 ) , 50 U/ml penicillin-streptomycin ( Thermo Fisher Scientific , 10378016 ) , and 10 mM nicotinamide ( Sigma-Aldrich , N0636 ) for the first 14 days . During the 15th-28th days of differentiation , 100 ng/ml human Activin-A ( PeproTech , 120–14 ) was supplemented into differentiation medium . From day 29 , Activin-A supplementation was stopped until differentiation was completed . After 8–10 weeks , pigmented clusters were formatted and manually picked , then plated on matrigel-coated dishes in RPE culture medium as previous described ( Maminishkis et al . , 2006 ) . They were cultured for another 6–8 weeks to allow them to form a functional monolayer for function assay . Besides well-established classical mature RPE markers RPE65 , Bestrophin1 and CRALBP , two additional RPE markers , MITF and PAX6 , were used for RPE fate validation . All the iPSC-RPE cells used in this study were at their passage 1 . Mutations ( P274R and I201T ) in the mutant iPSC-RPEs were verified by sequencing . HEK293 cells were gifts from Dr . David Yule at University of Rochester . Although HEK293 is on the list of commonly misidentified cell lines maintained by the International Cell Line Authentication Committee , the HEK293 cells used in this study were authenticated by short tandem repeat ( STR ) DNA profiling . No mycoplasma contamination was found . Low-passage-number HEK293 cells were maintained in DMEM supplemented with 10% FBS and 100 μg/ml penicillin-streptomycin . Immunofluorescence staining was performed in all iPSC-RPE lines and human fetal RPE cells . Cells were washed with PBS and fixed in 4% paraformaldehyde for 45 min at room temperature . After washing with PBS twice , the cells were incubated in PBS with 0 . 1% Triton X-100% and 2% donkey serum for 45 min . Then , primary antibodies against BESTROPHIN-1 ( 1:200 , Novus Biologicals , NB300-164 ) , ZO-1 ( 1:500 , Invitrogen Life Technologies , 40–2200 ) and EEA1 ( 1:200 , Thermo Fisher Scientific , MA5-14794 ) were applied to each sample for 2 hr at room temperature . Alexa Fluor 488-conjugated and Alexa Fluor 555-conjugated IgG ( 1:1 , 000 , Thermo Fisher Scientific ) were used as secondary antibodies . Hoechst was used to detect the cell nuclei . Stained cells were observed by confocal microscopy ( Nikon Ti Eclipse inverted microscope for scanning confocal microscopy , Japan ) . Whole-cell recordings of RPE and HEK cells were conducted 48–72 hr after splitting the cells or transfection , respectively , using an EPC10 patch clamp amplifier ( HEKA Electronics ) controlled by Patchmaster software ( HEKA ) . Micropipettes were fashioned from 1 . 5 mm thin-walled glass with filament ( WPI Instruments ) and filled with internal solution containing ( in mM ) : 130 CsCl , 1 MgCl2 , 10 EGTA , 2 MgATP ( added fresh ) , 10 HEPES ( pH 7 . 4 ) , and CaCl2 to obtain the desired free Ca2+ concentration ( maxchelator . stanford . edu/CaMgATPEGTA-TS . htm ) . Series resistance was typically 1 . 5–2 . 5 MΩ . There was no electronic series resistance compensation . External solution contained ( in mM ) : 140 NaCl , 5 KCl , 2 CaCl2 , 1 MgCl2 , 15 glucose and 10 HEPES ( pH 7 . 4 ) . Whole-cell I-V curves were generated from a family of step potentials ( −100 to +100 mV from a holding potential of 0 mV ) . Currents were sampled at 25 kHz and filtered at 5 or 10 kHz . Traces were acquired at a repetition interval of 4 s ( Yang et al . , 2014a ) . Purified full length KpBest proteins were fused to planar lipid bilayers formed by painting a lipid mixture of phosphatidylethanolamine and phosphatidylcholine ( Avanti Polar Lipids ) in a 3:1 ratio in decane; across a 200 µm hole in polysulfonate cups ( Warner Instruments ) separating two chambers . The trans chamber ( 1 . 0 ml ) , representing the intra-SR ( luminal ) compartment , was connected to the head stage input of a bilayer voltage clamp amplifier . The cis chamber ( 1 . 0 ml ) , representing the cytoplasmic compartment , was held at virtual ground . Solutions were as follows ( in mM ) : 150 NaCl , and 10 HEPES ( pH 7 . 4 ) in the cis and trans solution . Purified proteins were added to the cis side and were fused with the lipid bilayer . Single-channel currents were recorded using a Bilayer Clamp BC-525D ( Warner Instruments , LLC , CT ) , filtered at 1 kHz using a Low-Pass Bessel Filter 8 Pole ( Warner Instruments , LLC , CT ) , and digitized at 4 kHz . All experiments were performed at room temperature ( 23 ± 2°C ) . Total cellular protein was extracted by M-PER mammalian protein extraction reagent buffer ( Pierce , 78501 ) with proteinase inhibitor ( Roche Diagnostics ) , and quantified by Bio-Rad protein reader . Protein samples ( 20 μg ) were then separated on 10% Tris–Cl gradient gel and electro-blotted onto nitrocellulose membrane . The membranes were incubated in blocking buffer for 1 hr at room temperature , washed three times in PBS with 0 . 1% Tween for 5 min each , and incubated with primary antibody in blocking buffer overnight at 4°C . Primary antibodies against the following proteins were used for western blots: RPE65 ( 1:1 , 000 Novus Biologicals , NB100-355 ) , BESTROPHIN-1 ( 1:500 Novus Biologicals , NB300-164 ) , CRALBP ( 1:500 Abcam , ab15051 ) , β-actin ( 1:2 , 000 Abcam , ab8227 ) , and GFP ( 1:5 , 000 Invitrogen , A6455 ) . Mouse and rabbit secondary antibodies were obtained from Santa Cruz and used at a concentration of 1: 5000 . WT BEST1-GFP expressed from a BacMam baculoviral vector was made as previously described ( Goehring et al . , 2014 ) , and was added into RPE culture 24 hr after splitting the cells ( MOI = 100 ) . P237R and L177T KpBestΔC11 have 11 residues truncated from the C-terminus of wild-type KpBest . The wild-type BEST1 ( synthesized by Genscript ) , was amplified using polymerase chain reaction ( PCR ) , and was subcloned into a pEGFP-N1 mammalian expression vector . C-terminus truncated KpBest and point mutations of KpBest and BEST1 were made using the In-fusion Cloning Kit ( Clontech ) . All clones were verified by sequencing . For electrophysiology experiments , HEK293 cells cultured in 6 cm tissue culture dishes were transiently transfected with the indicated BEST1 ( 6 μg ) and T antigen ( 2 μg ) , using the calcium phosphate precipitation method . Cells were washed with PBS 4–8 hr after transfection and maintained in supplemented DMEM , and replated onto fibronectin-coated glass coverslips 24 hr after transfection ( Yang et al . , 2013 ) . BL21 plysS cells were gifts from Dr . Wayne Hendrickson . For scaling up , transformed BL21 plysS cells were grown at 37°C in TB media to OD 0 . 6–0 . 8 after being inoculated with 1% of the overnight culture . The culture was induced with 0 . 4 mM IPTG and continued to grow at 37°C for another 4 hr . BL21 plysS cells expressing targeted proteins were harvested by centrifugation and stored at −80°C before use . Cells were resuspended in a buffer containing 50 mM HEPES ( pH 7 . 8 ) and 200 mM NaCl and lysed using a French Press with two passes at 15–20 , 000 psi . Cell debris was removed by centrifugation at 10 , 000 g for 20 min , and the membrane fraction was isolated from that supernatant by ultra-centrifugation at 150 , 000 g for 1 hr . The membrane fraction was homogenized in a solubilization buffer containing 50 mM HEPES ( pH 7 . 8 ) and 300 mM NaCl , and incubated with a final concentration of 0 . 05% ( w/v ) DDM for 1 hr at 4°C . The non-dissolved matter was removed by ultracentrifugation at 150 , 000 g for 30 min , and the supernatant was loaded to a 5 ml Hitrap Ni2+-NTA affinity column ( GE Healthcare ) , pre-equilibrated with the same solubilization buffer supplemented with 0 . 05% DDM . After 20 column volume buffer wash , the protein was eluted with 500 mM imidazole in the solubilization buffer . The 10-His tags were removed by adding super TEV at 1:1 mass ratio and incubating at 4°C for 30 min . Tag removal was confirmed by SDS-PAGE , and the resulting sample was concentrated to approximately 10 mg/ml . Preparative size-exclusion chromatography was carried out on a Superdex-200 column for further purification , including removal of TEV protease and the cleaved tag . The gel-filtration buffer contained 40 mM HEPES ( pH 7 . 8 ) , 200 mM NaCl , 0 . 1 mM Tris [2-carboxyethyl] phosphine ( TCEP ) , and 2 × CMC of detergent DDM . Purified protein was concentrated to ~10 mg/ml . Crystals were all grown at 20°C using the sitting-drop vapor diffusion method . The condition contained 0 . 05 M zinc acetate , 6% v/v ethylene glycol , 0 . 1 M sodium cacodylate , pH 6 . 0 , and 6 . 6 % w/v PEG 8000 . Cryoprotection was achieved by adding 20% ethylene glycol to the crystallization solution . High resolution native data set from a single L177T KpBestΔC11 crystal was collected at APS ( Argonne National Laboratory ) beamline 24-ID-E .	Mutations to the gene that encodes a protein called BESTROPHIN1 cause a number of human diseases that lead to a progressive loss of sight and even blindness . Over two hundred of these disease-causing mutations exist , but it is not understood how they affect BESTROPHIN1 . Furthermore , there are currently no treatments available to treat these diseases . BESTROPHIN1 is an ion channel found in cell membranes in the retinal pigment epithelium ( RPE ) , a layer of cells in the eye that is vital for vision . When BESTROPHIN1 is stimulated by calcium ions , it opens up to allow chloride ions to flow into and out of the cell . The health of human eyes can be assessed by measuring how well they respond to light – a response that is believed to be generated from the flow of calcium-stimulated chloride ions in the RPE . Patients with mutant BESTROPHIN1 channels have an abnormally low response to light , but it remains unclear whether these channels are responsible for maintaining the flow of chloride ions required for the light response . Indeed , it is not confirmed whether calcium-stimulated chloride flow occurs on the surface of normal human RPE cells at all . Human RPE cells are difficult to obtain . Instead , Li , Zhang et al . took human skin cells – some from patients who had disease-causing mutations that affect BESTROPHIN1 – and used stem cell technology to coax the cells to develop into RPE cells . Calcium-stimulated chloride ion flow could be recorded on the surface of these cells . Next , the impact of two disease-causing mutations on BESTROPHIN1 was examined . The mutation from the patient who displayed the more severe illness completely inactivated the channel , while the other associated with milder illness caused a partial loss of channel activity . Notably , introducing normal BESTROPHIN1 into the RPE cells developed from patients with mutant BESTRPOPHIN1 restored chloride ion flow to normal levels . Thus it appears that BESTROPHIN1 is essential for maintaining calcium-stimulated chloride ion flow in human RPE cells . The techniques developed by Li , Zhang et al . form a patient-specific ‘disease-in-a-dish’ approach that could be used to study the consequences of other mutations to the gene that produces BESTROPHIN1 . This work also suggests that gene therapy could potentially help to treat BESTROPHIN1-related diseases .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2017	Patient-specific mutations impair BESTROPHIN1’s essential role in mediating Ca2+-dependent Cl- currents in human RPE
Using a combination of genetic , biochemical , and structural approaches , we show that the cyclic-peptide antibiotic GE23077 ( GE ) binds directly to the bacterial RNA polymerase ( RNAP ) active-center ‘i’ and ‘i+1’ nucleotide binding sites , preventing the binding of initiating nucleotides , and thereby preventing transcription initiation . The target-based resistance spectrum for GE is unusually small , reflecting the fact that the GE binding site on RNAP includes residues of the RNAP active center that cannot be substituted without loss of RNAP activity . The GE binding site on RNAP is different from the rifamycin binding site . Accordingly , GE and rifamycins do not exhibit cross-resistance , and GE and a rifamycin can bind simultaneously to RNAP . The GE binding site on RNAP is immediately adjacent to the rifamycin binding site . Accordingly , covalent linkage of GE to a rifamycin provides a bipartite inhibitor having very high potency and very low susceptibility to target-based resistance . GE23077 ( GE ) is a cyclic-peptide antibiotic produced by the soil bacterium Actinomadura sp . DSMZ 13491 ( Figure 1A; Ciciliato et al . , 2004 ) . GE exhibits antibacterial activity against both Gram-negative and Gram-positive bacterial pathogens in culture , including Moraxella catarrhalis and Streptococcus pyogenes ( Supplementary file 1A; Ciciliato et al . , 2004 ) . GE inhibits both Gram-negative and Gram-positive bacterial RNA polymerase ( RNAP ) in vitro , but does not inhibit human RNAP I , II , or III in vitro ( Supplementary file 1B; Ciciliato et al . , 2004 ) . Analysis of the kinetics of inhibition suggests that GE inhibits RNAP at a stage subsequent to the formation of the RNAP-template complex ( Sarubbi et al . , 2004 ) . 10 . 7554/eLife . 02450 . 003Figure 1 . Mechanism of transcription inhibition by GE: inhibition of first nucleotide addition in transcription initiation . ( A ) Structure of GE . dmaDap , Nβ- ( Z-2 , 3-dimethylacryloyl ) -α , β-diaminopropionic acid; dhGln , β , γ-dihydroxy-glutamine; Ama , aminomalonic acid; aThr , allothreonine; iSer , isoserine . Wavy bonds , previously undefined stereochemistry . ( B ) GE does not inhibit formation of a transcription initiation complex . ( C ) GE inhibits nucleotide addition in transcription initiation ( primer-dependent transcription initiation ) . ( D ) GE does not inhibit nucleotide addition in transcription elongation ( elongation from halted TEC containing 29 nt RNA product ) . See Figure 1—figure supplements 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02450 . 00310 . 7554/eLife . 02450 . 004Figure 1—figure supplement 1 . GE inhibits nucleotide addition in transcription initiation ( de novo transcription initiation ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02450 . 00410 . 7554/eLife . 02450 . 005Figure 1—figure supplement 2 . GE does not inhibit nucleotide addition in transcription elongation ( reconstituted transcription elongation complexes ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02450 . 005 GE is a non-ribosomally-synthesized cyclic heptapeptide ( Figure 1A; Marazzi et al . , 2005 ) . The stereochemistry at four chiral centers of GE has been defined based on acid hydrolysis and gas chromatography , but the stereochemistry at five other chiral centers has not been defined ( Figure 1A; Marazzi et al . , 2005 ) . Analogs of GE having modifications of the dmaDap , dhGln , and Ama residues , have been prepared by semi-synthetic derivatization of GE ( Mariani et al . , 2005 ) . Here we report the target and mechanism of transcription inhibition by GE . In addition , we report a series of crystal structures—including the first crystal structure of a substrate complex for de novo transcription initiation by a multisubunit RNAP—that define the structural relationships between GE and RNAP , GE and promoter DNA , GE and NTPs , and GE and rifamycins . Our results show that GE inhibits RNAP through a novel binding site and novel mechanism . GE inhibits RNAP by binding to a site—the ‘GE target’—that overlaps the RNAP active-center ‘i’ and ‘i+1’ sites and that includes coordinating ligands of the RNAP active-center catalytic Mg2+ ion , Mg2+ ( I ) . Binding of GE sterically precludes binding of initiating NTPs to the i site , i+1 site , and Mg2+ ( I ) , and thereby blocks transcription initiation . GE is the first identified example of a non-nucleoside RNAP inhibitor that functions through direct interaction with the core catalytic components of the RNAP active-center: the i site , i+1 site , and Mg2+ ( I ) . Our results further show that the GE target has three features that make it an unusually attractive target—a ‘privileged target’—for antibacterial drug discovery involving RNAP . First , the GE target includes functionally critical residues of the RNAP active center that cannot be substituted without loss of RNAP activity , and , therefore , that cannot be substituted to yield resistant mutants . Accordingly , the target-based resistance spectrum for GE is unusually small . Second , the GE target does not overlap the rifamycin target ( the target of the most important RNAP inhibitors in current clinical use in antibacterial therapy; Ho et al . , 2009 ) . Accordingly , GE exhibits no or negligible cross-resistance with rifamycins . Third , the GE target is immediately adjacent to the rifamycin target . Accordingly , it is possible to link GE to a rifamycin to construct a bipartite inhibitor that binds simultaneously to the GE target and the rifamycin target and , therefore , that is exceptionally potent and exceptionally refractory to target-based resistance . To define the mechanism of transcription inhibition by GE , we assessed effects of GE on individual reaction steps in transcription initiation and transcription elongation . Figure 1B shows that GE does not inhibit steps in transcription initiation up to and including formation of a competitor-resistant RNAP-promoter open complex ( RPo ) . We infer that GE does not inhibit promoter binding , loading of promoter DNA into the RNAP active-center cleft , or promoter unwinding . The results in Figure 1C show that GE inhibits nucleotide addition in transcription initiation . GE inhibits both primer-dependent transcription initiation ( Figure 1C ) , and de novo transcription initiation ( Figure 1—figure supplement 1 ) . In primer-dependent transcription initiation , GE inhibits the first nucleotide-addition step , inhibiting the synthesis of a 3-nt RNA product from a 2-nt RNA primer and an NTP . In de novo transcription initiation , GE inhibits the first nucleotide-addition step , inhibiting the synthesis of a 2-nt RNA product from initiating NTPs . The results in Figure 1D show that GE does not inhibit nucleotide addition in transcription elongation . GE does not inhibit transcription elongation upon addition of NTPs to a halted elongation complex ( Figure 1D ) , and GE does not inhibit single nucleotide addition upon addition of an NTP to an elongation complex reconstituted from RNAP and a synthetic nucleic acid scaffold ( Figure 1—figure supplement 2 ) . We conclude that GE specifically inhibits nucleotide addition in transcription initiation . The observation that GE inhibits nucleotide addition in initiation but not in elongation suggests that GE functions through a binding site that is available in RPo but that is not available in an elongation complex—for example , a binding site that overlaps the RNAP active-center i and i+1 nucleotide binding sites , or the path of the RNA product from the i and i+1 nucleotide binding sites , and that therefore would be unoccupied in RPo but occupied by RNA in an elongation complex . The mechanism of transcription inhibition of GE is reminiscent of , but differs from , the mechanism of transcription inhibition by rifampin ( Rif ) and other members of the rifamycin class . Like GE , Rif does not inhibit formation of RPo ( Figure 1B; McClure and Cech , 1978 ) . Also like GE , Rif inhibits nucleotide addition in transcription initiation , but does not inhibit nucleotide addition in transcription elongation ( Figure 1C , D; Sippel and Hartmann , 1968 ) . However , in contrast to GE , Rif does not generally inhibit the first nucleotide-addition step in transcription initiation ( Figure 1C; McClure and Cech , 1978 ) . Rif generally only inhibits synthesis of >2–3-nt RNA products and does so by binding to a site along the path of RNA from the RNAP active-center and sterically blocking RNA extension ( Campbell et al . , 2001; Feklistov et al . , 2008 ) . The observation that GE inhibits synthesis of 2-nt RNA products , whereas Rif generally only inhibits synthesis of >2–3-nt RNA products , suggests that GE functions through a binding site located closer than the Rif binding site to the RNAP active-center . The mechanism of transcription inhibition by GE also differs from the mechanisms of transcription inhibition by other previously characterized RNAP inhibitors . Sorangicin ( Sor ) functions through the same binding site on RNAP as Rif and inhibits synthesis only of >2–3-nt RNA products ( Campbell et al . , 2005 ) . Myxopyronin ( Myx ) , corallopyronin ( Cor ) , ripostatin ( Rip ) , and lipiarmycin ( Lpm ) inhibit formation of RPo ( Ho et al . , 2009 ) . Streptolydigin ( Stl ) , CBR703 ( CBR ) , and microcin J25 ( MccJ25 ) inhibit nucleotide addition in both initiation and elongation ( Artsimovitch et al . , 2003; Mukhopadhyay et al . , 2004; Ho et al . , 2009 ) . We conclude that GE inhibits transcription through a novel mechanism . To define effects of GE on interactions of RNAP with promoter DNA , we determined a crystal structure of RPo in complex with GE at 2 . 8 Å resolution ( Figure 4 ) . The higher resolution of this structure ( 2 . 8 Å vs 3 . 35 Å ) enables confirmation of the inferred stereochemical assignments at stereocenters of GE ( Figure 4D , E ) and enables identification of additional water-mediated H-bonds , including additional water-mediated H-bonds in the network of water-mediated interactions connecting GE to Mg2+ ( I ) ( Figure 4D–E; Figure 4—figure supplement 1 ) . 10 . 7554/eLife . 02450 . 011Figure 4 . Structural basis of transcription inhibition by GE: crystal structure of RPo-GE . ( A ) Overall structure . ( B ) Crystallographic data and refinement statistics . ( C ) Electron density and atomic model for GE . ( D ) Contacts between RPo and GE ( stereodiagram ) . ( E ) Contacts between RPo and GE ( schematic ) . See Figure 4—figure supplements 1 , 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 02450 . 01110 . 7554/eLife . 02450 . 012Figure 4—figure supplement 1 . Structural basis of transcription inhibition by GE . Network of contacts to GE dhGln residue . Stereoview . Gray , RNAP carbon atoms . Green , GE carbon atoms . Red , oxygen atoms . Blue , nitrogen atoms . Violet sphere , Mg2+ ( I ) . Red spheres , water molecules . Dashed blue lines , H-bonds . Dashed orange lines , coordinate-covalent bonds . DOI: http://dx . doi . org/10 . 7554/eLife . 02450 . 01210 . 7554/eLife . 02450 . 013Figure 4—figure supplement 2 . Absence of effects of DNA on GE conformation and RNAP-GE interactions . Superimposition of crystal structures of RPo-GE ( Figure 4 ) and RNAP-GE ( Figure 3 ) . Blue mesh , blue sticks , red sticks , gray surface , yellow surface , and violet sphere: mFo-DFc omit map for GE , atomic model for GE , DNA , RNAP , σ , and Mg2+ ( I ) from crystal structure of RPo-GE ( Figure 4 ) . Cyan sticks , atomic model for GE from crystal structure of RNAP-GE ( Figure 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02450 . 01310 . 7554/eLife . 02450 . 014Figure 4—figure supplement 3 . Absence of effects of GE on DNA conformation and RNAP-DNA interactions . Superimposition of crystal structures of RPo-GE ( Figure 4 ) and RPo ( Zhang et al . , 2012 ) . Blue mesh , blue sticks , red sticks , gray surface , yellow surface , and violet sphere: mFo-DFc omit map for GE , atomic model for GE , DNA , RNAP , σ , and Mg2+ ( I ) from crystal structure of RPo-GE ( Figure 4 ) . Cyan sticks , DNA from crystal structure of RPo ( Zhang et al . , 2012 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02450 . 014 Comparison of the structures of RNAP-GE ( Figure 3 ) and RPo-GE ( Figure 4 ) shows that promoter DNA binding to form RPo does not change the conformation of GE or the interactions between RNAP and GE ( Figure 4—figure supplement 2 ) . Comparison of structures of RPo ( Zhang et al . , 2012 ) and RPo-GE ( Figure 4 ) show that GE does not change the conformation of DNA or the interactions between RNAP and DNA ( Figure 4—figure supplement 3 ) . The results provide a graphic confirmation of the results from Figure 1 , indicating that GE does not inhibit formation of RPo and , instead , inhibits a subsequent reaction required for the first nucleotide addition in transcription initiation . Our results establish GE inhibits RNAP through a novel mechanism and a novel target . Our results show that GE inhibits the first nucleotide-addition step in transcription initiation ( Figure 1 ) , show that GE functions through a binding site that overlaps the RNAP active-center i and i+1 sites ( Figure 2 ) , define the structural basis of RNAP-GE interaction and RPo-GE interaction ( Figures 3 , 4 ) , and show that GE prevents binding of initiating NTPs to the RNAP i and i+1 sites ( Figure 5 ) . Our results further establish that the binding site on RNAP for GE is adjacent to , but does not substantially overlap , the binding site on RNAP for the rifamycin antibacterial drugs ( Figure 2D–F ) , show that GE and a rifamycin can bind simultaneously to their adjacent binding sites in RNAP ( Figure 6 ) , and show that GE and a rifamycin can be covalently linked , through the GE dmaDap sidechain and the rifamycin C3-O4 region , to yield a bipartite RNAP inhibitor that binds to both the GE target and the rifamycin target ( Figure 7 ) . Three features of the GE target , identified in this work , indicate that the GE target is an unusually attractive target—a ‘privileged target’—for antibacterial drug discovery involving RNAP . First , since most residues of the GE binding site are functionally critical residues of the RNAP active center that cannot be substituted without loss of RNAP activity , the target-based resistance spectra of an antibacterial compound that functions through the GE binding site will be small ( ∼1/10 the size of the target-based resistance spectrum of Rif; ∼1/10 to ∼1/5 the size of the target-based resistance spectra of RNAP inhibitors; Figure 2D; Figure 2—figure supplement 2 ) . Second , since the GE binding site is different from the rifamycin binding site , an antibacterial compound that functions through the GE binding site will not exhibit target-based cross-resistance with rifamycins ( Figure 2E , F; Supplementary file 2D , E ) . Third , since the GE binding site is adjacent to , but does not substantially overlap , the rifamycin binding site ( Figures 2D and 6 ) , an antibacterial compound that functions through the GE binding site can be linked to a rifamycin or a sorangicin to construct a bipartite , bivalent inhibitor that binds to both the GE target and the rifamycin target and , therefore , that is exceptionally potent and exceptionally refractory to target-based resistance ( Figure 7 ) . GE23077 ( GE ) was prepared from cultures of Actinomadura sp . DSMZ 13491 as in Ciciliato et al . ( 2004 ) . Sorangicin ( Sor ) was prepared from cultures of Sorangium cellulosum strain So cel2 as in Irschik et al . ( 1987 ) . Lipiarmycin ( Lpm ) was prepared from cultures of Actinoplanes deccanensis as in Coronelli et al . ( 1975 ) . ( ± ) E , E-myxopyronin B ( Myx ) was synthesized as in Ebright and Ebright ( 2013 ) . Rifampin ( Rif ) , rifamycin SV ( RifSV ) , streptolydigin ( Stl ) , and CBR703 were purchased from Sigma–Aldrich ( St . Louis , MO ) , Sigma–Aldrich , Sourcon-Padena ( Tübingen , Germany ) , and Maybridge ( Tintagel , UK ) , respectively . Plasmid pRL706 encodes C-terminally hexahistidine-tagged E . coli RNAP β subunit under control of the trc promoter ( Severinov et al . , 1997 ) . Plasmid pRL663 encodes C-terminally hexahistidine-tagged E . coli RNAP β′ subunit under control of the tac promoter ( Wang et al . , 1995 ) . Plasmid pKD46 carries a temperature-sensitive replication origin , confers ampicillin-resistance , and encodes λ Exo , Beta , and Gam , under control of the ParaB promoter ( Datsenko and Wanner , 2000 ) . Plasmid pAKE604 confers kanamycin-resistance and sucrose-sensitivity ( El-Sayed et al . , 2001 ) . E . coli RNAP , [Asn516]β-RNAP , [Asp565]β-RNAP , and [Lys684]β-RNAP core and holoenzyme were prepared from E . coli strain XE54 ( Tang et al . , 1994 ) transformed with pRL706 , pRL706-516N , pRL706-565D , and pRL706-684K , respectively , using procedures essentially as in Niu et al . ( 1996 ) . E . coli RNAP holoenzyme derivatives site-specifically labelled with fluorescein at σ70 residue 517 ( [F517]σ70-RNAP holoenzyme derivatives ) were prepared as in Knight et al . ( 2005 ) . T . thermophilus RNAP holoenzyme was prepared as in Zhang et al . ( 2012 ) . Fluorescence-detected RNAP-inhibition assays were performed by a modification of the procedure of Kuhlman et al . ( 2004 ) . Reaction mixtures contained ( 20 μl ) : 0–100 μM test compound , bacterial RNAP holoenzyme ( 75 nM E . coli RNAP holoenzyme or E . coli RNAP holoenzyme derivative , 75 nM Staphylococcus aureus RNAP core enzyme and 300 nM S . aureus σA [prepared as in Srivastava et al . 2011] , 75 nM Mycobacterium tuberculosis RNAP core enzyme and 300 nM M . tuberculosis σA [prepared as in Srivastava et al . 2011] , or 75 nM T . thermophilus RNAP holoenzyme ) , 20 nM DNA fragment containing the bacteriophage T4 N25 promoter ( positions −72 to +367; prepared by PCR from plasmid pARTaqN25-340-tR2 [Liu , 2007] ) , 100 μM ATP , 100 μM GTP , 100 μM UTP , and 100 μM CTP , in TB ( 50 mM Tris–HCl , pH 8 . 0 , 100 mM KCl , 10 mM MgCl2 , 1 mM DTT , 10 μg/ml bovine serum albumin , 5% methanol , and 5 . 5% glycerol ) . Reaction components other than DNA and NTPs were pre-incubated 10 min at 37°C . Reactions were carried out by addition of DNA and incubation 15 min at 37°C , followed by addition of NTPs and incubation 60 min at 37°C . DNA was removed by addition of 1 μl 5 mM CaCl2 and 2 U DNase I ( Ambion , Grand Island , NY ) , followed by incubation 90 min at 37°C . RNA was quantified by addition of 100 μl Quant-iT RiboGreen RNA Reagent ( Life Technologies , Grand Island , NY; 1:500 dilution in 10 mM Tris–HCl , pH 8 . 0 , 1 mM EDTA ) , followed by incubation 10 min at 22°C , followed by measurement of fluorescence intensity ( excitation wavelength = 485 nm and emission wavelength = 535 nm; GENios Pro microplate reader [Tecan , Männedorf , Switzerland] ) . Radiochemical assays with human RNAP I/II/III were performed essentially as in Sawadogo and Roeder ( 1985 ) . Reaction mixtures contained ( 20 µl ) : 0–100 µM GE , 8 U HeLaScribe Nuclear Extract ( Promega , Madison , WI ) , 1 µg human placental DNA ( Sigma–Aldrich ) , 400 μM ATP , 400 μM [α32P]UTP ( 0 . 11 Bq/fmol ) , 400 μM CTP , 400 μM GTP , 50 mM Tris–HCl , pH 8 . 0 , 7 mM HEPES-NaOH , 70 mM ( NH4 ) 2SO4 , 50 mM KCl , 12 mM MgCl2 , 5 mM DTT , 0 . 1 mM EDTA , 0 . 08 mM phenylmethylsulfonyl fluoride , and 16% glycerol . Reaction components other than DNA and NTPs were pre-incubated 10 min at 30°C , DNA was added and reaction mixtures were incubated 15 min at 30°C , NTPs were added and reaction mixtures were incubated 60 min at 30°C . Reaction mixtures were spotted on DE81 filter discs ( Whatman , Kent , UK; pre-wetted with water ) and incubated 1 min at room temperature . Filters were washed with 3 × 3 ml Na2HPO4 , 2 × 3 ml water , and 3 ml ethanol , using a filter manifold ( Hoefer , Holliston , MA ) . Filters were placed in scintillation vials containing 10 ml Scintiverse BD Cocktail ( Thermo Fisher , Waltham , MA ) , and radioactivity was quantified by scintillation counting ( LS6500; Beckman–Coulter , Brea , CA ) . Half-maximal inhibitory concentrations ( IC50s ) were calculated by non-linear regression in SigmaPlot ( SPSS , Chicago , IL ) . Minimum inhibitory concentrations ( MICs ) were quantified using broth microdilution assays ( Clinical and Laboratory Standards Institute , 2009 ) , using a starting cell density of 3 × 104 cfu/ml , LB broth ( Sambrook and Russell , 2001 ) , and an air atmosphere for E . coli D21f2tolC ( tolC:Tn10 rfa lac28 proA23 trp30 his51 rpsL173 ampC tsx81; strain with cell-envelope defects resulting in increased susceptibility to hydrophobic agents , including GE; Fralick and Burns-Keliher , 1994; unpublished data ) , and using a starting cell density of 3 × 104 cfu/ml , Bacto Todd Hewitt broth ( TH broth; BD Biosciences , San Jose , CA ) , and a 7% CO2/6% O2/4% H2/83% N2 atmosphere for S . pyogenes and M . catarrhalis . Saturation mutagenesis of rpoB plasmid pRL706 and rpoC plasmid pRL663 was performed by use of PCR amplification with ‘doped’ oligodeoxyribonucleotide primers ( methods as in Mukhopadhyay et al . 2008 ) . ‘Doped’ oligodeoxyribonucleotide primers corresponding to codons 136-143 , 504-511 , 512-522 , 523-534 , 535-541 , 542-549 , 563-573 , 677-690 , 758-763 , 813-814 , 829-835 , 1054-1060 , 1064-1074 , 1102-1108 , and 1233-1242 of the rpoB gene of plasmid pRL706 , and codons 347-355 , 425-429 , 456-465 , 779-792 , and 934-943 of the rpoC gene of plasmid pRL663 , were synthesized on an Applied Biosystems 392/394 automated DNA/RNA synthesizer ( Foster City , CA ) using solid-phase β-cyanoethylphosphoramidite chemistry ( sequences in Supplementary file 2A ) . The level of ‘doping’ ( nucleotide misincorporation ) was selected to yield an average of 0 . 4–1 substitution ( s ) per molecule of oligodeoxyribonucleotide primer ( equations in Hermes et al . , 1989 , 1990 ) . Thus , the nucleotides corresponding to codons 758-763 and 813-814 of rpoB , and codons 425-429 of rpoC were synthesized using phosphoramidite reservoirs containing 92% of the correct phosphoramidite and 8% of a 1:1:1:1 mix of dA , dC , dG , and dT phosphoramidites ( i . e . , 94% total correct phosphoramidite and 6% total incorrect phosphoramidite ) . The nucleotides corresponding to codons 136-143 , 504-511 , 512-522 , 523-534 , 535-541 , 542-549 , 563-573 , 677-690 , 829-835 , 1054-1060 , 1064-1074 , 1102-1108 , and 1233-1242 of rpoB , and codons 347-355 , 456-465 , 779-792 , and 934-943 of rpoC were synthesized using phosphoramidite reservoirs containing 98% of the correct phosphoramidite and 2% of a 1:1:1:1 mix of dA , dC , dG , and dT phosphoramidites , ( i . e . , 98 . 5% total correct phosphoramidite and 1 . 5% total incorrect phosphoramidite . ) All other nucleotides were synthesized using phosphoramidite reservoirs containing 100% of the correct phosphoramidite . Mutagenesis reactions were performed using the QuikChange XL Site-Directed Mutagenesis Kit ( Agilent/Stratagene , La Jolla , CA ) with a “doped” oligodeoxyribonucleotide primer , a complementary oligodeoxyribonucleotide primer , and pRL706 or pRL663 as template ( primers at 75-150 nM; all other components at concentrations as specified by the manufacturer ) . Mutagenized plasmid DNA was introduced by transformation into E . coli XL1-Blue ( Agilent/Stratagene ) . Transformants ( 103-104 cells ) were applied to LB-agar plates ( Sambrook and Russell , 2001 ) containing 200 μg/ml ampicillin , plates were incubated 16 hr at 37°C , and plasmid DNA was prepared from the pooled resulting colonies . The resulting passaged mutagenized plasmid libraries for the 15 “doped” oligonucleotide primers targeting rpoB were pooled on an equimolar basis , and the resulting passaged mutagenized plasmid libraries for the five “doped” oligonucleotide primers targeting rpoC were pooled on an equimolar basis . Pooled , passaged mutagenized plasmid libraries for each gene were introduced by transformation into E . coli D21f2tolC . Transformants ( ∼103 cells ) were applied to LB-agar plates containing 200-500 μg/ml GE , 200 μg/ml ampicillin , and 1 mM IPTG; and plates were incubated 24-48 hr at 37°C . GE-resistant mutants were identified by the ability to form colonies on this medium , were confirmed by re-streaking on the same medium , were further confirmed by quantifying resistance levels in liquid cultures and accepting only isolates with >2-fold resistance ( procedures as described below ) , and were demonstrated to contain plasmid-linked GE-resistant mutations by preparing plasmid DNA , transforming E . coli D21f2tolC with plasmid DNA , and plating transformants on the same medium . For each confirmed mutant , nucleotide sequences of rpoB and rpoC were determined by Sanger sequencing ( eight primers per gene ) . Temperature-sensitive E . coli strain RL585 [rpoBamcI supDts43 , 74 Δ ( recA-srl ) 306 lacZam2110 galEKam leuam trpam sueA rpsL tsx srl301::Tn10-84; Landick et al . , 1990] was transformed with pRL706 or a pRL706 derivative , transformants ( 103-104 cells ) were applied to LB-agar plates containing 200 μg/ml ampicillin , 1 mM IPTG , and 10 μg/ml tetracycline , plates were incubated 22 hr at 43°C , and bacterial growth was scored . GE-resistant and Rif-resistant mutations were transferred from pRL706 derivatives to the chromosome of E . coli D21f2tolC by λ-Red-mediated recombineering ( procedures analogous to those in Datsenko and Wanner 2000 and Sawitzke et al . , 2007; but using chemical transformation rather than electroporation ) . DNA fragments ( 143 bp or 306 bp ) containing rpoB segments with GE-resistant or Rif-resistant mutations were prepared by PCR amplification using pRL706 derivatives carrying GE-resistant and Rif-resistant mutations as templates and 5’-CAGGTGGTATCCGTCGGTGCGTCCCTG-3’ and 5’-CGTTCCATACCAGTACCAACCAGCGGC-3’ ( for GE-resistant mutations ) or 5′-GGATATGATCAACGCCAAGCCGATTTCCGCAGC-3′ and 5′-CGATACGGAGTCTCAAGGAAGCCGTATTCG-3′ ( for Rif-resistant mutations ) as primers . DNA fragments were purified by isolation by electrophoresis on 0 . 8% agarose ( procedures as in Sambrook and Russell 2001 ) and extracted from gel slices using a Gel/PCR DNA Fragments Extraction Kit ( IBI Scientific , Peosta , IA; procedures as specified by the manufacturer ) . DNA fragments and co-selection/counter-selection plasmid pAKE604 ( 10 ng and 100 ng; for GE-resistant mutations ) or DNA fragments only ( 30 ng; for Rif-resistant mutations ) were introduced by transformation into chemically competent cells of E . coli D21f2tolC pKD46 ( prepared by culturing E . coli D21f2tolC pKD46 in LB broth containing 200 μg/ml ampicillin and 1 mM arabinose at 30°C until OD = 0 . 6 , pelleting cells , re-suspending cells in 85% LB , 10% PEG 3350 , 5% DMSO , and 50 mM MgCl2 , and flash freezing in dry-ice/ethanol ) , and transformants were cultured 3 . 5 hr at 37°C with shaking , applied to LB-agar plates containing 500 μg/ml GE and 40 μg/ml kanamycin ( for GE-resistant mutations ) or 1–2 μg/ml Rif ( for Rif-resistant mutations ) , and incubated 24-30 hr at 37°C . Isolates containing chromosomal GE-resistant or Rif-resistant mutations were identified by the ability to form colonies on media containing GE or Rif , were confirmed by re-streaking on the same media , and were verified to have lost temperature-sensitive plasmid pKD46 by re-streaking on LB-agar plates containing 0 or 200 μg/ml ampicillin . For GE-resistant isolates , segregants lacking sacB plasmid pAKE604 were identified and verified by plating on LB agar containing 5% sucrose . Isolates were demonstrated to contain the expected mutations by PCR amplification and nucleotide sequencing of rpoB . Resistance levels of GE-resistant mutants were quantified by performing broth microdilution assays . Single colonies were inoculated into 5 ml LB broth containing 200 μg/ml ampicillin , and 1 mM IPTG ( for E . coli plasmid-borne mutants and controls ) , 5 ml LB broth ( for E . coli chromosomal mutants and controls ) , or 5 ml TH broth ( for S . pyogenes mutants and controls ) and incubated at 37°C with shaking in air ( for E . coli ) or in 7% CO2/6% O2/4% H2/83% N2 ( for S . pyogenes ) until OD600 = 0 . 4–0 . 8 . Diluted aliquots ( ∼4 × 105 cells in 50 μl of the same medium ) were dispensed into wells of a 96-well plate containing 50 μl of the same medium or 50 μl of a twofold dilution series of GE in the same medium ( final concentrations = 0 and 8–8000 μg/ml ) , and were incubated 16 hr at 37°C with shaking under the same conditions . The MIC was defined as the lowest tested concentration of GE that inhibited bacterial growth by ≥90% . Cross-resistance levels were determined analogously to resistance levels . Liquid cultures were prepared as described above for determination of resistance levels . Diluted aliquots of cultures ( ∼2 × 105 cells in 97 μl growth medium ) were dispensed into wells of a 96-well plate , were supplemented with 3 μl methanol or 3 μl of a twofold dilution series of Rif , Sor , Stl , CBR703 , Myx , or Lpm in methanol ( final concentrations = 0 and 0 . 012–50 μg/ml ) , and were incubated 16 hr at 37°C with shaking . Reaction mixtures contained ( 20 μl ) : test compound ( 0 or 0 . 5 μM GE , or 2 . 2 μM Rif ) , 40 nM E . coli RNAP holoenzyme , 10 nM DNA fragment containing positions −42 to +426 of the lacUV5 ( ICAP ) promoter ( Naryshkin et al . , 2001 ) , and 100 μg/ml heparin , in TB . Reaction components other than DNA and heparin were pre-incubated 10 min at 37°C; DNA was added and reaction mixtures were incubated 15 min at 37°C; heparin was added and reactions were incubated 2 min at 37°C to disrupt non-specific RNAP-promoter complexes and RNAP-promoter closed complexes ( Cech and McClure , 1980 ) . Products were applied to 5% TBE polyacrylamide slab gels ( Bio-Rad , Hercules , CA ) , gels were electrophoresed in TBE ( 90 mM Tris-borate , pH 8 . 0 , and 2 mM EDTA ) , and gels were stained with SYBR Gold Nucleic Acid Gel Stain ( Life Technologies ) . Reaction mixtures contained ( 20 μl ) : test compound ( 0 or 0 . 5 μM GE , or 2 . 2 μM Rif ) , 5 nM E . coli RNAP holoenzyme [Epicentre] , 2 . 5 nM DNA fragment containing positions −49 to +30 of the lacCONS promoter ( Mukhopadhyay et al . , 2001 ) , 25 μg/ml heparin , 500 μM ApA , and 25 μM [α32P]UTP ( 0 . 9 Bq/fmol ) in TB . Reaction components other than DNA , heparin , ApA , and [α-32P]UTP were pre-incubated 10 min at 37°C; DNA was added and reaction mixtures were incubated 15 min at 37°C; heparin was added and reaction mixtures were incubated 2 min at 37°C; ApA and [α32P]UTP were added and reaction mixtures were incubated 10 min at 37°C . Reactions were terminated by adding 10 μl 80% formamide , 10 mM EDTA , 0 . 04% bromophenol blue , 0 . 04% xylene cyanol , and 0 . 08% amaranth red . Products were heated 5 min at 90°C , cooled 5 min on ice , applied to 16% polyacrylamide ( 19:1 acrylamide:bisacrylamide , 7 M urea ) slab gels , electrophoresed in TBE , and analyzed by storage-phosphor scanning ( Typhoon; GE Healthcare , Piscataway , NJ ) . Identities of tri- and tetranucleotide abortive products from transcription initiation at lacUV5 were defined as in Borowiec and Gralla ( 1985 ) . Reaction mixtures contained ( 20 μl ) : 0 or 0 . 5 μM GE , 100 nM E . coli RNAP holoenzyme , 20 nM DNA fragment containing positions −65 to +35 of the bacteriophage T7 A1 promoter ( prepared by PCR amplification of a synthetic nontemplate-strand oligodeoxyribonucleotide ) , 25 μg/ml heparin , 25 μM ATP , and 25 μM [α32P]UTP ( 0 . 7 Bq/fmol ) in TB . Reaction components other than DNA , heparin , and NTPs were pre-incubated 5 min at 23°C , DNA was added and reaction mixtures were incubated 15 min at 37°C , heparin and NTPs were added were added and incubated 5 min at 37°C . Reactions were terminated by adding 10 μl 80% formamide , 10 mM EDTA , 0 . 04% bromophenol blue , and 0 . 04% xylene cyanol . Products were heated 5 min at 95°C , cooled 5 min on ice , and applied to 16% polyacrylamide ( 19:1 acrylamide:bisacrylamide , 7 M urea ) slab gels , electrophoresed in TBE , and analyzed by storage-phosphor scanning ( Typhoon; GE Healthcare ) . Halted transcription elongation complexes ( halted at position +29 ) were prepared essentially as in Revyakin et al . ( 2006 ) . Reaction mixtures ( 18 μl ) contained: 40 nM E . coli RNAP holoenzyme , 10 nM DNA fragment N25-100-tR2 ( Revyakin et al . , 2006 ) , 100 μg/ml heparin , 5 μM ATP , 5 μM GTP , and 5 μM [α32P]UTP ( 4 Bq/fmol ) in TB . Reaction components except heparin and NTPs were pre-incubated 10 min at 37°C; heparin was added and reaction mixtures were incubated 2 min at 37°C; NTPs were added and reaction mixtures were incubated 5 min at 37°C . The resulting halted transcription elongation complexes were exposed to test compounds by addition of 1 μl 10 μM GE or 1 μl 44 μM Rif , incubated 5 min at 37°C , and were re-started by addition of 1 μl 1 mM CTP and incubation 5 min at 37°C . Reactions were terminated by adding 10 μl 80% formamide , 10 mM EDTA , 0 . 04% bromophenol blue , 0 . 04% xylene cyanol , and 0 . 08% amaranth red . Products were heated 5 min at 90°C , cooled 5 min on ice , applied to 16% polyacrylamide ( 19:1 acrylamide:bisacrylamide , 7 M urea ) slab gels , electrophoresed in TBE , and analyzed by storage-phosphor scanning ( Typhoon; GE Healthcare ) . Nucleic-acid scaffolds for assays were prepared as follows: nontemplate-strand oligodeoxyribonucleotide ( 5′-TCGCCAGACAGGG-3′; 1 μM ) , template-strand oligodeoxyribonucleotide ( 5′-CCCTGTCTGGCGATGGCGCGCCG-3′; 1 μM ) , and 32P-5′-end-labelled oligoribonuceotide ( 5′-32P-CGGCGCGCC-3′; 1 μM; 200 Bq/fmol ) in 25 μl 5 mM Tris–HCl , pH 7 . 7 , 200 mM NaCl , and 10 mM MgCl2 , were heated 5 min at 95°C and cooled to 4°C in 2°C steps with 1 min per step using a thermal cycler ( Applied Biosystems ) and then were stored at −20°C . Reaction mixtures for assays contained ( 15 μl ) : 0 or 0 . 5 μM GE or 0 or 2 . 2 μM Rif , 40 nM wild-type E . coli RNAP core enzyme ( Epicentre , Madison , WI ) , 10 nM 32P-labelled nucleic-acid scaffold ( 200 Bq/fmol ) , and 20 μM ATP in TB . Reaction components except inhibitors and ATP were pre-incubated 5 min at 37°C , GE or Rif was added and reaction mixtures were incubated 5 min at 37°C , and ATP was added and reaction mixtures were incubated 2 min at 37°C . Reactions were terminated by adding 15 μl 80% formamide , 10 mM EDTA , 0 . 04% bromophenol blue , and 0 . 04% xylene cyanol , and heating 2 min at 95°C . Products were applied to 20% polyacrylamide ( 19:1 acrylamide:bisacrylamide , 7 M urea ) slab gels , electrophoresed in TBE , and analyzed by storage-phosphor scanning ( Typhoon; GE Healthcare ) . RNAP-Rif interaction assays were performed as in Feklistov et al . ( 2008 ) . The assays monitored the quenching of the fluorescence emission of fluorescein incorporated into RNAP holoenzyme at σ70 residue 517 ( serving as a fluorescence-resonance-energy-transfer donor ) by the naphthyl moiety of Rif ( serving as a fluorescence-resonance-energy-transfer acceptor; Knight et al . , 2005; Feklistov et al . , 2008 ) . Fluorescence measurements were performed using a QuantaMaster QM1 spectrofluorometer ( PTI , Edison , NJ ) ( excitation wavelength = 480 nm; emission wavelength = 530 nm; and excitation and emission slit widths = 5 nm ) . For determination of association kinetics , 720 μl 2 nM [F517]σ70-RNAP holoenzyme and 0-2 μM GE in 40 mM Tris–HCl , pH 8 . 0 , 100 mM NaCl , 10 mM MgCl2 , 1 mM DTT , 0 . 02% Tween-20 , and 5% glycerol was incubated 15 min at 24°C and then mixed with 30 μl 0 . 01–0 . 5 μM Rif in the same buffer at 24°C in a cuvette chamber with a mixing dead time ∼0 . 5 s , and fluorescence emission intensities were monitored for 30 min at 24°C . On-rates for RNAP-Rif interaction , kon , were calculated by fitting data to:I= ( I0−I∞ ) exp ( −kobst ) +I∞where kobs is the observed association rate constant at a specified Rif concentration , I is the fluorescence emission intensity at time t , Io is the fluorescence emission intensity at t = 0 , and I∞ is the fluorescence emission intensity at t = ∞; followed by fitting the Rif-concentration-dependence of kobs to:kobs=kon[Rif]+koffwhere koff is ≥0 but is otherwise unconstrained . For determination of dissociation kinetics , 720 μl of 2 nM [F517]σ70-RNAP holoenzyme and 0 . 05 μM Rif in the same buffer was incubated 30 min at 24°C and then mixed with 30 μl of 0–50 μM GE and 12 . 5–50 μM Sor ( which binds to the same site as Rif but does not quench fluorescence emission and therefore serves as a ‘competitor trap’ for Rif dissociation kinetics; Feklistov et al . , 2008 ) in the same buffer at 24°C in a cuvette chamber with a mixing dead time ∼0 . 5 s; and fluorescence emission intensities were monitored for 5–300 min at 24°C . Dissociation kinetics were found not to depend on the concentration of Sor in the concentration range used in this work ( final concentrations of 0 . 5–2 μM ) , verifying that Sor in this concentration range does not compete with GE and does not actively displace Rif from RNAP . Off-rates for RNAP-Rif interaction , koff , were calculated as:I=I0+ ( I∞−I0 ) [1−exp ( −kofft ) ]where I is the fluorescence emission intensity at time t , Io is the fluorescence intensity at t = 0 , and I∞ is the fluorescence intensity at t = ∞ . Equilibrium dissociation constants for RNAP-Rif interaction , Kd , were calculated as koff/kon . The equilibrium dissociation constant for RNAP-GE interaction , Ki , was calculated from the association-kinetics data , by fitting the GE-concentration-dependence of I∞ to:I∞= ( I∞ , max[GE] ) / ( Ki+[GE] ) Crystallization and crystal handling were performed essentially as in Tuske et al . ( 2005 ) . A crystallization stock solution was prepared by adding 1 μl T . thermophilus RNAP holoenzyme ( 10 mg/ml ) in 20 mM Tris–HCl , pH 7 . 7 , 100 mM NaCl , and 1% glycerol to 1 μl 33 mM magnesium formate containing 40 μM ZnCl2 . The crystallization stock solution was equilibrated against a reservoir solution of 30 mM sodium citrate , pH 5 . 4 , and 35 mM magnesium formate in a vapor-diffusion hanging-drop crystallization tray ( Hampton Research , Aliso Viejo , CA ) at 22°C . Hexagonal crystals formed and grew to a final size of ∼0 . 4 × ∼0 . 4 × ∼0 . 2 mm within 6 d . GE was soaked into RNAP crystals by addition of 0 . 2 μl 10 mM GE in 60% ( vol/vol ) ( ± ) -2-methyl-2 , 4-pentanediol ( MPD; Hampton Research ) to the crystallization drop and incubation 15 min at 22°C . Crystals were transferred to solutions containing 0 . 5 mM GE , 20 mM MES , pH 6 . 0 , 13 mM magnesium formate , 2 mM spermine , 2 mM DTT , 5% PEG400 , and 15% ( vol/vol ) ( 2R , 3R ) - ( − ) -2 , 3-butanediol ( Sigma–Aldrich ) , and were flash-cooled with liquid nitrogen . Diffraction data for RNAP-GE were collected at Cornell High Energy Synchrotron Source ( CHESS ) beamline F1 and were processed and scaled using iMOSFLM and SCALA ( Battye et al . , 2011; Evans , 2006 ) . The structure of RNAP-GE was solved by molecular replacement with AutoMR in Phenix ( McCoy et al . , 2007 ) using a modified structure of T . thermophilus RNAP holoenzyme ( PDB 3DXJ; Mukhopadhyay et al . , 2008 ) as the search model . Early stages of refinement of the RNAP-GE complex included rigid-body refinement of subdomains ( ∼15–200 residue segments ) of the RNAP molecule . Cycles of rigid-body , individual-atom , and individual-B-factor refinement using Ramachandran and secondary structure restraints and optimized weights for stereochemistry and optimized atomic displacement parameters were carried out using Phenix ( Adams et al . , 2010 ) . Manual rebuilds against electron-density maps were performed using Coot ( Emsley et al . , 2010 ) and Molprobity ( Davis et al . , 2007; Chen et al . , 2010 ) . In addition , two refinement cycles were performed within Autobuster ( Bricogne et al . , 2011 ) . For GE backbone atoms and GE sidechain atoms with previously defined stereochemistry ( Marazzi et al . , 2005 ) , an initial atomic model was generated using Maestro ( Schrodinger , Portland , OR ) and was fit to mFo-DFc maps using Phenix ( Adams et al . , 2010 ) . For GE sidechain atoms with previously undefined stereochemistry , stereochemistry was deduced and atoms were added based on assessment of mFo-DFc maps and RNAP-GE interactions using PrimeX ( Schrodinger ) . All GE atoms could be fitted to density except atoms of the GE dmaDap residue distal to the sidechain carbonyl moiety . Subsequent cycles of refinement and model building were performed , leading to the current crystallographic model , with a standard crystallographic residual of Rwork = 0 . 21 and Rfree = 0 . 24 computed using all data from 38 . 97 to 3 . 35 Å resolution . Atomic coordinates and structure factors for RNAP-GE have been deposited in the PDB with accession code 4MQ9 . Crystals of T . thermophilus RPo were prepared using the same nucleic-acid scaffold as used for analysis of RPo in Zhang et al . ( 2012 ) , and were grown and handled essentially as in Zhang et al . ( 2012 ) . Crystallization drops contained 1 μl RPo in 20 mM Tris–HCl , pH 7 . 7 , 100 mM NaCl , and 1% glycerol , and 1 μl reservoir buffer ( RB; 100 mM Tris–HCl , pH 8 . 4 , 200 mM KCl , 50 mM MgCl2 , and 9 . 5% PEG4000 ) , and were equilibrated against 400 μl RB in a vapor-diffusion hanging-drop tray . Rod-like crystals appeared in 1 d , and were used to micro-seed hanging drops using the same conditions . GE was soaked into RPo crystals by addition of 0 . 2 μl 20 mM GE in RB to the crystallization drop and incubation 15 min at 22°C . Crystals were transferred in stepwise fashion to successive reservoir solutions containing 1 mM GE in 0 . 5% , 1% , 2 . 5% , 5% , 10% , 14% , and 17 . 5% ( v/v ) ( 2R , 3R ) - ( − ) -2 , 3-butanediol ( 20 s for first step and 2 s for each subsequent step ) and were flash-cooled with liquid nitrogen . Diffraction data were collected at CHESS beamline F1 and Brookhaven National Laboratory ( BNL ) beamline X29A and were processed using HKL2000 ( Otwinowski and Minor , 1997 ) . Structure factors were converted using the French-Wilson algorithm in Phenix ( French and Wilson , 1978 ) and were subjected to anisotropy correction using the UCLA MBI Diffraction Anisotropy server ( Strong et al . , 2006; http://services . mbi . ucla . edu/anisoscale/ ) . The structure was solved by molecular replacement with Molrep ( Vagin and Teplyakov , 1997 ) using one RNAP molecule from the structure of T . thermophilus RPo ( PDB 4 G7H; Zhang et al . , 2012 ) as the search model . Early-stage refinement included rigid-body refinement of the RNAP molecule , followed by rigid-body refinement of each subunit of RNAP molecule . Cycles of iterative model building with Coot ( Emsley et al . , 2010 ) and refinement with Phenix ( Adams et al . , 2010 ) were performed . Atomic models of the DNA nontemplate strand , the DNA template strand , and GE were built into mFo-DFc omit maps , and subsequent cycles of refinement and model building were performed . The final crystallographic model of RPo-GE , refined to Rwork and Rfree of 0 . 21 and 0 . 25 , has been deposited in the PDB with accession code 4OIN . ATP ( Sigma–Aldrich ) and CMPcPP ( Jena Biosciences , Jena , Germany ) were soaked into RPo crystals ( prepared as described above , using the nucleic-acid scaffold used for analysis of RPo-GpA in Zhang et al . 2012 ) by addition of 0 . 2 μl 30 mM ATP and 30 mM CMPcPP in 55% ( vol/vol ) RB to the crystallization drop , and incubation 15–20 min at 22°C . Crystals were transferred into reservoir solutions containing 2 mM ATP and 2 mM CMPcPP in 17 . 5% ( vol/vol ) ( 2R , 3R ) - ( − ) -2 , 3-butanediol and were flash-cooled with liquid nitrogen . Diffraction data were collected at BNL beamline X25 , processed and scaled using HKL2000 ( Otwinowski and Minor , 1997 ) , and subjected to anisotropic correction using the UCLA MBI Diffraction Anisotropy server ( Strong et al . , 2006; http://services . mbi . ucla . edu/anisoscale/ ) . The structure was solved and refined using procedures analogous to those described above for RPo-GE . The final crystallographic model contained RPo , ATP bound in the RNAP i site , and CMPcPP:Mg2+ bound in the RNAP i+1 site . The final crystallographic model of RPo-ATP-CMPcPP , refined to Rwork and Rfree of 0 . 21 and 0 . 26 , respectively , has been deposited in the PDB with accession code 4OIO . Crystals of RPo ( prepared as described above for RPo + ATP + CMPcPP ) first were soaked with GE ( addition of 0 . 2 μl 20 mM GE in RB to the crystallization drop and incubation 15 min at 22°C ) and then were soaked with ATP and CMPcPP ( addition of 0 . 2 μl 30 mM ATP and 30 mM CMPcPP in 55% [vol/vol] RB to the crystallization drop and incubation 15 min at 22°C ) . Crystals then were transferred to reservoir solutions containing 1 mM GE , 2 mM ATP , and 2 mM CMPcPP in 17 . 5% ( vol/vol ) ( 2R , 3R ) - ( − ) -2 , 3-butanediol and were flash-cooled with liquid nitrogen . Diffraction data were collected at BNL beamline X25 , and were processed , scaled , and corrected for anisotropy using HKL2000 ( Otwinowski and Minor , 1997 ) . The structure was solved and refined using procedures analogous to those described above for RPo-GE . The final crystallographic model contained RPo , GE bound to the GE target , and ATP:Mg2+ bound to the RNAP E site , and did not contain ATP in the RNAP i site or CMPcPP in RNAP i+1 site . The final crystallographic model , refined to Rwork and Rfree of 0 . 21 0 . 25 , respectively , has been deposited in the PDB with accession code 4OIP . Crystals of RPo ( prepared as described above for RPo + ATP + CMPcPP ) first were soaked with Rif ( addition of 0 . 1 μl 20 mM Rif in RB containing 40% [vol/vol] [2R , 3R]- ( − ) -2 , 3-butanediol to the crystallization drop and incubation 15 min at 22°C ) and then were soaked with GE ( addition of 0 . 2 μl 20 mM GE in RB to the crystallization drop and incubation 15 min at 22°C ) . Crystals then were transferred to reservoir solutions containing 1 mM GE and 0 . 4 mM Rif in 17 . 5% ( vol/vol ) ( 2R , 3R ) - ( − ) -2 , 3-butanediol and were flash-cooled with liquid nitrogen . Diffraction data were collected at CHESS beamline F1 , and were processed and scaled using HKL2000 ( Otwinowski and Minor , 1997 ) . The structure was solved and refined using procedures analogous to those described above for RPo-GE . The final crystallographic model contained RPo and GE bound to the GE target but did not contain Rif . The final crystallographic model , refined to Rwork and Rfree of 0 . 20 and 0 . 25 , respectively , has been deposited in the PDB with accession code 4OIQ . Crystals of RPo ( prepared as described above for RPo + ATP + CMPcPP ) first were soaked with RifSV ( addition of 0 . 2 μl 10 mM RifSV in RB to the crystallization drop and incubation 15 min at 22°C , or transfer of the crystal to 1 μl 10 mM RifSV in RB and incubation 15 min at 22°C ) and then were soaked with GE ( addition of 0 . 2 μl 20 mM GE in RB to the drop and incubation 15 min at 22°C ) . Crystals then were transferred in to reservoir solutions containing 1 mM GE and 1 mM RifSV in 17 . 5% ( vol/vol ) ( 2R , 3R ) - ( − ) -2 , 3-butanediol and were flash-cooled with liquid nitrogen . Diffraction data were collected at BNL beamline X25 , were processed and scaled using HKL2000 ( Otwinowski and Minor , 1997 ) , and were subjected to anisotropic correction using the UCLA MBI Diffraction Anisotropy server ( Strong et al . , 2006; http://services . mbi . ucla . edu/anisoscale/ ) . The structure was solved and refined using procedures analogous to those described above for RPo-GE . The final crystallographic model contained RPo , GE bound to the GE target , and RifSV bound to the Rif target . The final crystallographic model of RPo-GE-RifSV , refined to Rwork and Rfree of 0 . 21 and 0 . 25 , respectively , has been deposited in the PDB with accession code 4OIR . GE ( 20 mg 25 μmol ) , ammonium acetate ( 60 mg; 780 μmol; Aldrich ) , and perchloric-acid-impregnated silica ( 5 mg; prepared as in Singh et al . , 2009 ) , were mixed in 4 ml absolute ethanol in a screw-cap vial . The mixture was microwaved for 4 × 30 s ( 1000 W ) with intervals of 1 min for re-mixing contents of the vial . The mixture was allowed to incubate at room temperature for another 16 hr , evaporated to dryness , and resuspended in 2 ml 1% triethylamine-water . The mixture was centrifuged , and the supernatant was purified via HPLC ( Phenomenex C18 , semi-prep; 5 min 0% B , 20 min 5% B , 25 min 10% B , 30 min 30% B , 40 min 80% B; A = water , B = acetonitrile , 2 ml/min ) . The HPLC elution profile and mass spectrum of the product indicate that the product has undergone decarboxylation of the Ama sidechain ( Mariani et al . , 2005 ) . It is known that acid and heat induce decarboxylation of the GE Ama sidechain , and that decarboxylated GE exhibits ∼1/20 the RNAP-inhibitory activity and antibacterial activity of GE ( Mariani et al . , 2005 ) . Yield: 3 . 5 mg; 18% . MS ( MALDI ) : calculated: m/z 777 . 80 ( MH+ ) ; found: 778 . 20 , 800 . 59 ( M + Na+ ) . 3-Bromo-rifamycin S ( 2 . 7 mg; 3 . 47 μmol; prepared as in Marchi and Montecchi 1979 ) , compound 1 ( 2 . 7 mg; 3 . 47 μmol; Example 1a ) and triethylamine ( 0 . 5 μl; 3 . 47 μmol; Aldrich ) were mixed together in 200 μl DMF and allowed to react for 18 hr at 25°C . The reaction mixture was quenched with 100 μl water , centrifuged , and the supernatant was purified via HPLC ( Phenomenex C18 , semi-prep; 0 min 10% B , 35 min 100% B; A = water , B = acetonitrile , 2 ml/min ) . Yield: 1 . 51 mg; 30% . MS ( MALDI ) : calculated: m/z 1493 . 52 ( M + Na+ ) ; found: 1494 . 22 . Sodium ascorbate ( 2 . 38 mg; 12 μmol; Aldrich ) in 25 μl water was added to compound 2 ( 0 . 600 mg; 0 . 4 μmol; Example 1b ) in 100 μl water , mixed , and allowed to react for 10 min at 25°C . The product was isolated via HPLC ( Phenomenex C18 , analytical; 0’ 10% B , 35′ 100% B; A = water , B = acetonitrile , 1 ml/min ) . Yield: 0 . 1 mg; 17% . MS ( MALDI ) : calculated: m/z 1495 . 52 ( M + Na+ ) ; found: 1495 . 71 .	As increasing numbers of bacteria become resistant to antibiotics , new drugs are needed to fight bacterial infections . To develop new antibacterial drugs , researchers need to understand how existing antibiotics work . There are many ways to kill bacteria , but one of the most effective is to target an enzyme called bacterial RNA polymerase . If bacterial RNA polymerase is prevented from working , bacteria cannot synthesize RNA and cannot survive . GE23077 ( GE for short ) is an antibiotic produced by bacteria found in soil . Although GE stops bacterial RNA polymerase from working , and thereby kills bacteria , it does not affect mammalian RNA polymerases , and so does not kill mammalian cells . Understanding how GE works could help with the development of new antibacterial drugs . Zhang et al . present results gathered from a range of techniques to show how GE inhibits bacterial RNA polymerase . These show that GE works by binding to a site on RNA polymerase that is different from the binding sites of previously characterized antibacterial drugs . The mechanism used to inhibit the RNA polymerase is also different . The newly identified binding site has several features that make it an unusually attractive target for development of antibacterial compounds . Bacteria can become resistant to an antibiotic if genetic mutations lead to changes in the site the antibiotic binds to . However , the site that GE binds to on RNA polymerase is essential for RNA polymerase to function and so cannot readily be changed without crippling the enzyme . Therefore , this type of antibiotic resistance is less likely to develop . In addition , the newly identified binding site for GE on RNA polymerase is located next to the binding site for a current antibacterial drug , rifampin . Zhang et al . therefore linked GE and rifampin to form a two-part ( ‘bipartite’ ) compound designed to bind simultaneously to the GE and the rifampin binding sites . This compound was able to inhibit drug-resistant RNA polymerases tens to thousands of times more potently than GE or rifampin alone .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "microbiology", "and", "infectious", "disease" ]	2014	GE23077 binds to the RNA polymerase ‘i’ and ‘i+1’ sites and prevents the binding of initiating nucleotides
Insulin resistance in muscle , adipocytes and liver is a gateway to a number of metabolic diseases . Here , we show a selective deficiency in mitochondrial coenzyme Q ( CoQ ) in insulin-resistant adipose and muscle tissue . This defect was observed in a range of in vitro insulin resistance models and adipose tissue from insulin-resistant humans and was concomitant with lower expression of mevalonate/CoQ biosynthesis pathway proteins in most models . Pharmacologic or genetic manipulations that decreased mitochondrial CoQ triggered mitochondrial oxidants and insulin resistance while CoQ supplementation in either insulin-resistant cell models or mice restored normal insulin sensitivity . Specifically , lowering of mitochondrial CoQ caused insulin resistance in adipocytes as a result of increased superoxide/hydrogen peroxide production via complex II . These data suggest that mitochondrial CoQ is a proximal driver of mitochondrial oxidants and insulin resistance , and that mechanisms that restore mitochondrial CoQ may be effective therapeutic targets for treating insulin resistance . Insulin resistance is a major risk factor for several metabolic diseases including type two diabetes . This defect is found in all metabolic tissues most notably adipose tissue , muscle and liver . While insulin resistance in adipose and muscle tissue is likely to occur in a cell autonomous manner recent evidence suggests that liver insulin resistance may occur via a mechanism involving defects in adipose tissue ( Perry et al . , 2015; Titchenell et al . , 2016 ) . A range of perturbations have been shown to trigger insulin resistance including diets high in fat and/or sucrose ( Boden et al . , 2015; Samocha-Bonet et al . , 2012; Turner et al . , 2013 ) , hyperinsulinaemia , hyperlipidaemia ( Roden et al . , 1996 ) , inflammation ( Hotamisligil et al . , 1994 ) , corticosteroids ( Houstis et al . , 2006; Kusunoki et al . , 1995 ) . These insults may cause insulin resistance via distinct means indicating that insulin resistance maybe a heterogeneous disorder . For example , stresses such as lipotoxicity ( Chavez et al . , 2003; Griffin et al . , 1999 ) , endoplasmic reticulum stress ( Ozcan et al . , 2004 ) , mitochondrial dysfunction ( Kelley et al . , 2002; Montgomery and Turner , 2015 ) and mitochondrial oxidative stress ( Anderson et al . , 2009; Hoehn et al . , 2009; Houstis et al . , 2006 ) have all been reported to play a causal role in insulin resistance . However , we and others have shown that mitochondrial oxidative stress is a common feature of many in vitro insulin resistance models ( Hoehn et al . , 2008; Houstis et al . , 2006 ) and metabolic tissues both from mice and humans ( Anderson et al . , 2009; Paglialunga et al . , 2015 ) , and in humans ( Anderson et al . , 2009 ) . Further , many cellular stresses associated with insulin resistance such as ceramides ( García-Ruiz et al . , 1997 ) and endoplasmic reticulum stress ( Malhotra and Kaufman , 2007 ) may also increase mitochondrial oxidant production . Despite this evidence for mitochondrial oxidants being a common feature and cause of insulin resistance the molecular mechanisms that trigger increased oxidant production in mitochondria as well as the precise source of these oxidants remain unclear . In the present study , we have performed global analysis of the proteome and transcriptome in insulin-resistant adipose tissue from mice and humans and in a range of insulin resistance models in cultured adipocyte models including hyperinsulinaemia , inflammation , and glucocorticoids in an effort to identify changes that may contribute to mitochondrial oxidant production . The mevalonate/coenzyme Q ( CoQ ) biosynthesis pathway was altered in all models , and this was accompanied by a selective decrease in mitochondrial CoQ content in all models of adipocyte insulin resistance as well as in human adipose tissue and in insulin-resistant muscle from mice fed a high fat high sucrose diet . Loss of mitochondrial CoQ was both necessary and sufficient to drive complex II-dependent mitochondrial oxidant production and adipocyte insulin resistance . Our data provide evidence for decreased mitochondrial CoQ content and resultant generation of oxidants being a convergent pathway for many different models of insulin resistance . To identify pathways that may contribute to insulin resistance in adipose tissue we used unbiased proteomics to specifically look for factors or pathways that: ( a ) change across a range of insulin-resistant models including in humans , and ( b ) that have a demonstrable link to mitochondrial redox homeostasis . The models studied included adipose tissue from mice fed a high fat high sucrose diet ( HFHSD ) for different periods of time and three in vitro models of insulin resistance ( 3T3-L1 adipocytes treated with chronic insulin , dexamethasone or tumour necrosis factor-α ) . We initially focussed on adipose tissue as our tissue of interest for two main reasons . First , adipose tissue insulin resistance is found in mice and humans that display whole body insulin resistance and insulin resistance at this site can influence whole body insulin sensitivity ( Abel et al . , 2001; Sugii et al . , 2009 ) . Second , there are highly robust in vitro adipocyte models ( 3T3-L1 cells ) that accurately recapitulate both insulin action and the generation of insulin resistance using a range of insults that mimic perturbations implicated in insulin resistance in vivo such as hyperinsulinemia , inflammation and glucocorticoids . These models are invaluable since they provide a highly controlled system for manipulating insulin sensitivity in a cell autonomous manner . Mice fed a HFHSD for one day were glucose intolerant and the extent of glucose intolerance plateaued by 14 d ( Figure 1—figure supplement 1A–B ) . Adipose tissue insulin resistance was observed by 5 d and this degree of resistance was maintained to 42 d of HFHSD feeding ( Figure 1A , Figure 1—figure supplement 1C ) . Skeletal muscle exhibited a comparatively delayed onset of insulin resistance consistent with previous studies ( Turner et al . , 2013 ) , but also reached a maximal observed insulin resistance by 14 d ( Figure 1—figure supplement 1D ) . Insulin resistance in in vitro models was defined by impaired HA-GLUT4 translocation to the plasma membrane ( Figure 1B ) and insulin-mediated 2-deoxyglucose ( 2DOG ) uptake ( Figure 1—figure supplement 1E ) . Together , these models provided an ideal integrated platform with which to explore drivers of insulin resistance . Proteomic analyses of insulin-resistant adipose tissue and 3T3-L1 adipocytes provided quantitative data on 2981 and 3494 proteins , respectively ( Figure 1C–D , Supplementary file 1-tabs A , C ) , 98 of which were altered at both 5 and 14 d HFHSD time points ( Figure 1—figure supplement 1F , top right panel , Supplementary file 1- tab A ) and 491 in ≥2 in vitro models ( Figure 1—figure supplement 1F , left panel , Supplementary file 1-tab C ) . A small subset of these ( 19 ) were common to both analyses ( Figure 1—figure supplement 1F , bottom right panel , Figure 1—figure supplement 1 . L ) including the LRP1 chaperone LRPAP1 , which is of interest because LRP1 regulates GLUT4 trafficking in adipocytes ( Jedrychowski et al . , 2010 ) . From gene set enrichment analysis , 13 pathways were altered at both time points in adipose tissue from HFHSD mice ( Figure 1—figure supplement 1G , top right panel , Supplementary file 1-tab E ) . Similarly , there was a high degree of overlap in altered pathways across different in vitro models ( Figure 1—figure supplement 1G , left panel , Supplementary file 1-tab E ) . Ten pathways were overlapping between all models ( Figure 1—figure supplement 1G , bottom right panel , Figure 1—figure supplement 1 . M ) . Intriguingly , parallel analysis of gene expression in these models revealed minimal overlap between regulated transcripts and proteins in all models ( Figure 1—figure supplement 1I–K , Supplementary file 2 ) . For example , of the 98 proteins altered at both 5 and 14 d HFHSD only 12 were altered at the mRNA level ( Figure 1—figure supplement 1H , Supplementary file 2-tab A ) , and there was limited concordance between changes in proteins and transcript expression for the 19 proteins and 10 pathways found to be altered in both in vivo and in vitro models ( Figure 1—figure supplement 1L–M ) . This lack of concordance between gene and protein expression emphasises how crucial proteomic analyses are in identifying causal links to metabolic disease . To identify convergent changes in pathways at the proteome-level that correlated with insulin resistance across all adipocyte models , we generated a combined z-score for pathways across in vivo time points and in vitro models ( Figure 1E ) . There were 13 KEGG pathways ( excluding disease pathways ) that were highly altered ( z-score >4 ) in in vivo and in vitro analyses . Oxidative phosphorylation was most highly altered in both in vivo and in vitro models , and other pathways of interest included TCA cycle , branched chain amino acid metabolism ( valine , leucine and isoleucine degradation ) , proteasome , ribosome , spliceosome , N-glycan biosynthesis and terpenoid backbone biosynthesis/mevalonate pathway ( Figure 1E , Supplementary file 3- tab B ) . To further filter pathways that might be implicated in insulin resistance , we next performed proteomic analysis of adipose tissue from a cohort of obese subjects that have been extensively clinically phenotyped ( Chen et al . , 2015 ) . This cohort was matched for BMI and comprised insulin- sensitive and insulin-resistant subjects based on responses during a hyperinsulinaemic-euglycaemic clamp , meaning that we could identity pathways related to insulin sensitivity independent of obesity/BMI ( Chen et al . , 2015 ) . We quantified 4481 proteins across 22 subjects and correlated the expression of proteins ( Supplementary file 3- tab A ) and pathways ( Supplementary file 3- tab B ) with clinical features that are diagnostic of insulin sensitivity . For the purposes of this exercise , we focused on suppression of non-esterified fatty acids ( NEFAs ) during the clamp as this is likely to be more directly related to insulin action in adipose tissue than glucose infusion rate ( GIR ) , which is likely driven mainly by muscle . We identified 299 proteins ( Supplementary file 3- tab A ) and 26 pathways ( Supplementary file 3- tab B ) that were positively correlated with insulin sensitivity and 142 proteins and two pathways ( ribosome , spliceosome ) that were negatively correlated with insulin sensitivity ( r = <−0 . 423 or >0 . 423 ) ( Supplementary file 3- tabs A , B ) . Importantly , the PPAR signalling pathway , a known regulator of adipose insulin sensitivity ( Sugii et al . , 2009 ) , was positively associated with insulin sensitivity in this analysis . Of the 13 pathways of interest from the integrated proteomic analysis of insulin resistance models ( Figure 1E ) only five were positively associated with insulin sensitivity in human adipose tissue ( Figure 1F , Supplementary file 3-tab B ) . These comprised spliceosome , central carbon metabolism ( pyruvate metabolism , TCA cycle , glycolysis , pentose phosphate pathway ) , amino acid metabolism including branched chain amino acid synthesis/degradation and the terpenoid backbone biosynthesis/mevalonate pathway , which generates precursors for isoprenoids such as cholesterol and CoQ . This is of interest as branched chain amino acid metabolism ( Newgard et al . , 2009 ) and spliceosome function ( Vernia et al . , 2016 ) have been implicated in adipocyte or whole body insulin sensitivity , providing support for our analysis pipeline . We were particularly interested in the potential role of the mevalonate pathway in insulin resistance because this pathway feeds CoQ biosynthesis , an essential component of mitochondrial electron transport and so defects in this pathway might play a role in mitochondrial oxidative stress . Our data revealed changes in transcripts and proteins throughout the mevalonate pathway ( Figure 2 , Figure 2—figure supplement 1 , Supplementary file 1- tabs A-E , Supplementary file 2- tabs A-E , Supplementary file 3- tabs A-B ) , but when assessing different endpoints of the mevalonate pathway such as cholesterol , N-glycosylation ( dolichol ) and CoQ , we only observed altered protein expression in components of the CoQ pathway in both human and mouse adipose tissue ( Figure 2 ) . COQ7 and COQ9 were lower in insulin-resistant adipose tissue from humans and mice ( Figure 2 ) , while ADCK3/COQ8 was dysregulated at the protein and mRNA levels in vitro ( Figure 2—figure supplement 1 ) . Our integrated analysis of insulin-resistant proteomes from in vivo and in vitro models , and human , pointed toward a convergence upon dysregulated CoQ biosynthesis in insulin resistance . We next determined if the change in expression of mevalonate/CoQ pathway proteins translated into altered CoQ metabolism . To do this , we measured total CoQ content in in vivo and in vitro models of adipocyte insulin resistance . Whole-cell CoQ concentrations were decreased in in vitro models ( Figure 3A ) but not in insulin-resistant adipose tissue ( Figure 3B ) . We postulated that because CoQ is found in membranes throughout the cell that there might be a selective depletion of CoQ in specific locations , for example in mitochondria where it is synthesised . To investigate this , we analysed CoQ in subcellular fractions from 3T3-L1 adipocytes and adipose tissue . This revealed a selective depletion of CoQ in mitochondria across all models ( Figure 3C–D , Figure 3—figure supplement 1A–B ) . This decrease was not due to changes in mitochondrial content ( assessed via citrate synthase activity and OXPHOS subunit abundance; Figure 3—figure supplement 1C–J , Supplementary file 2- tabs A-D ) . Consistent with data from model systems , adipose tissue mitochondrial CoQ10 ( the major form of CoQ in humans ) was positively correlated with insulin-induced suppression of NEFAs ( Figure 3E–F , Table 1 ) and whole-body insulin sensitivity ( GIR ) in obese humans ( Figure 3—figure supplement 1K–L , Table 1 ) . Mitochondrial CoQ10 was also significantly and positively correlated with expression of proteins in the CoQ biosynthesis pathway ( r =+0 . 52 ) ( Supplementary file 3-tab B ) indicating that decreased biosynthesis may contribute to lower mitochondrial CoQ in insulin resistance . Stratification of participants by adipose tissue mitochondrial CoQ10 revealed no effect of age or BMI ( Table 1 ) . These findings reveal that decreases in mitochondrial CoQ are an obesity-independent feature of adipocyte insulin resistance . Intriguingly , our proteomic data indicated that the expression of proteins integral to the mevalonate pathway was decreased in fat from humans and mice and from 3T3-L1 adipocytes treated with dexamethasone or TNF-α whereas this was not the case in the chronic insulin 3T3-L1 adipocyte model ( Figure 2—figure supplement 1 ) . Thus , we next examined if the observed decrease in mitochondrial CoQ reflected changes in CoQ biosynthesis , which we measured by determining 13C6-CoQ9 in 3T3-L1 adipocytes incubated with 13C6-4-hydroxybenzoic acid . Consistent with pathway analysis and our intracellular measures of cholesterol content ( Figure 3—figure supplement 1M–P ) , CoQ biosynthesis rates were lower in cells treated with dexamethasone or TNF-α but elevated in response to chronic insulin ( Figure 3—figure supplement 1Q ) . Together , it appears probable that dexamethasone and TNF-α treatments lower mitochondrial CoQ largely via reduced biosynthesis , although increased CoQ in microsomal and PM subcellular fractions ( Figure 3—figure supplement 1A–B ) in these models point to additional dysregulation of CoQ trafficking . Since these models replicate the lower content of mevalonate/CoQ biosynthesis pathway proteins measured in mice and humans , it is likely that decreased CoQ biosynthesis contributes to loss of CoQ in these more physiological systems . This does not appear to be the case for adipocytes treated with chronic insulin , where additional pathway ( s ) likely contribute to dysregulated mitochondrial CoQ homeostasis . The above findings highlight loss of mitochondrial CoQ as a common feature of adipocyte insulin resistance so we next investigated if a similar phenomenon occurs in other insulin responsive tissues , most notably muscle in view of its major role in whole body glucose metabolism/insulin resistance . In muscle , we found decreased mitochondrial CoQ at 14 and 42 d HFHSD feeding ( Figure 3G ) . These time points correlate with the emergence of insulin resistance in muscle ( Figure 1—figure supplement 1D ) . In contrast to adipose tissue , total muscle CoQ was lower at all time points tested ( Figure 3—figure supplement 1R ) , potentially reflecting higher mitochondrial content of this tissue . Liver mitochondrial CoQ was unchanged in response to HFHSD feeding ( Figure 3—figure supplement 1S ) , despite changes in total CoQ at 5 d HFHSD feeding ( Figure 3—figure supplement 1T ) . Changes in cholesterol content in in vitro models ( Figure 3—figure supplement 1M–P ) , adipose tissue ( Figure 3—figure supplement 1U ) , muscle ( Figure 3—figure supplement 1V ) and liver ( Figure 3—figure supplement 1W ) were inconsistent with a causal role in insulin resistance across multiple tissues . These data suggest that a decrease in mitochondrial CoQ may be involved at an early stage in the development of insulin resistance in muscle and adipose tissue . To test whether loss of CoQ contributes to insulin resistance in adipocytes and muscle , we first examined whether restoration of mitochondrial CoQ could restore insulin sensitivity in 3T3-L1 adipocytes . Addition of CoQ9 had no effect in control cells but increased mitochondrial CoQ9 in insulin-resistant cells ( Figure 4A ) and both CoQ9 and CoQ10 improved insulin-stimulated HA-GLUT4 translocation and 2DOG uptake in all cell models ( Figure 4B , Figure 4—figure supplement 1A–B ) . In light of the importance of adipocyte lipolysis in whole body glucose homeostasis ( Perry et al . , 2015; Titchenell et al . , 2016 ) , we next tested whether CoQ could also improve insulin-regulated suppression of lipolysis in these models . Dex or TNFα treatment enhanced basal lipolysis as previously described ( Souza et al . , 1998; Xu et al . , 2009 ) and insulin-regulated inhibition of lipolysis was defective in all models ( Figure 4C , Figure 4—figure supplement 1C–E ) . Provision of CoQ9 had no effect in control cells but increased suppression of lipolysis by insulin in in vitro models ( albeit not significantly in the TNF model; p=0 . 11 ) . We next tested whether CoQ could alleviate insulin resistance in vivo by providing liposomal CoQ10 via intraperitoneal injection every second day . CoQ10 was used for these studies as it was not feasible to obtain sufficient CoQ9 and the doses of CoQ10 used in these studies were optimised so that we did not observe changes in body weight or adiposity , reported previously ( Xu et al . , 2017 ) , since such metabolic changes would likely directly affect insulin action in muscle and adipose . CoQ10 administration improved whole-body glucose tolerance in mice fed a HFHSD for 5 or 14 d ( Figure 4D–E , Figure 4—figure supplement 1F–G ) without altering insulin secretion ( Figure 4F , Figure 4—figure supplement 1H ) . Improved glucose tolerance was accompanied by increased suppression of NEFAs ( Figure 4G , Figure 4—figure supplement 1I ) and 2DOG clearance into epididymal ( Figure 4—figure supplement 1J ) and inguinal ( Figure 4—figure supplement 1K , p=0 . 06 ) adipose depots and quadriceps ( Figure 4—figure supplement 1L , p=0 . 08 ) during the GTT . In a more direct measure of insulin sensitivity in vivo , CoQ10 also increased insulin responsiveness of HFHSD-fed mice during an ITT as measured by blood glucose excursion ( Figure 4H ) and 2DOG uptake into epididymal and inguinal fat pads and quadriceps muscle ( Figure 4I–K ) . Adipose depots retain a high degree of insulin responsiveness and insulin resistance ex vivo ( Figure 1—figure supplement 1C , Figure 4L ) . Therefore , we used adipose tissue for ex vivo analyses rather than muscle tissue , for which ex vivo analyses are more technically challenging and give less robust responses ( data not shown ) . Administration of increasing doses of CoQ10 to HFHSD-fed mice for 14 d dose-dependently increased CoQ10 in mitochondria from epididymal adipose tissue ( Figure 4—figure supplement 1M ) , improved glucose tolerance ( Figure 4—figure supplement 1N–O ) and improved insulin-stimulated 2DOG uptake in an ex vivo assay using adipose tissue explants ( Figure 4L ) . CoQ10 did not alter whole body or epididymal fat pad mass ( Figure 4—figure supplement 1P–Q ) . Together , these data suggest that provision of exogenous CoQ improves insulin-regulated glucose uptake and that lower CoQ content may contribute to insulin resistance in adipose tissue and muscle . Our previous data shows that lower mitochondrial CoQ content is a common feature of insulin resistance and that replacing CoQ can overcome insulin resistance . We next investigated whether specific perturbation of CoQ biosynthesis was sufficient to induce insulin resistance in adipocytes . To achieve this , we incubated adipocytes with 4-nitrobenzoic acid ( NB ) or 4-cholorobenzoic acid ( CB ) ( Alam et al . , 1975; Forsman et al . , 2010 ) to competitively inhibit 4-hydroxybenzoate:polyprenyl transferase ( Coq2 ) ( Figure 5A ) . Both inhibitors decreased mitochondrial CoQ9 ( Figure 5A ) to a similar extent to that observed in insulin-resistant adipocytes ( Figure 3C ) , while CoQ9 supplementation restored normal mitochondrial CoQ9 concentrations ( Figure 5A ) . Importantly , both inhibitors caused insulin resistance in adipocytes ( Figure 5B–5C , Figure 5—figure supplement 1A–B ) which could be reversed with provision of CoQ . Similarly , siRNA knock down of the key regulatory proteins in CoQ biosynthesis that were down-regulated in insulin-resistant mouse and human adipose tissue ( Coq7 or Coq9 ) ( Figure 5—figure supplement 1C–D ) lowered mitochondrial CoQ9 ( Figure 5D ) and triggered insulin resistance ( Figure 5E–5F , Figure 5—figure supplement 1E ) . Insulin resistance triggered by pharmacological or genetic inhibition of CoQ biosynthesis occurred independently of consistent defects in insulin signalling to the key regulators of glucose transport ( Akt or the Akt substrate TBC1D4 ) , or changes in GLUT4 expression ( Figure 5—figure supplement 1F–M ) . Similarly , CoQ provision did not alter signalling responses or GLUT4 expression in any in vitro models of insulin resistance ( Figure 5—figure supplement 1N–Q ) despite improved insulin-stimulated HA-GLUT4 translocation , 2DOG uptake and suppression of lipolysis under these conditions ( Figure 4B–C , Figure 4—figure supplement 1A ) . These data are consistent with previous reports that insulin resistance is not driven by overt and consistent defects in proximal insulin signalling ( Hoehn et al . , 2008; Tan et al . , 2015 ) . The link between CoQ and insulin resistance is of interest in the context of statins that target the mevalonate pathway and have recently been shown to be associated with progression to type two diabetes in humans ( Cederberg et al . , 2015; Preiss et al . , 2011; Sattar et al . , 2010 ) . To begin to explore this , we incubated 3T3-L1 adipocytes with simvastatin or atorvastatin for up to 72 hr . Both statins lowered cellular cholesterol ( Figure 5—figure supplement 1R ) and CoQ content ( Figure 5—figure supplement 1S ) , providing proof-of-principle that lower mevalonate pathway activity influences cellular CoQ content in adipocytes . Statins induced insulin resistance ( Figure 5—figure supplement 1T ) , and this was reversed by providing CoQ9 or mevalonate ( Figure 5—figure supplement 1T ) . Together with the observation that more specific inhibitors of the CoQ biosynthetic pathway also trigger insulin resistance , these data provide convincing evidence that loss of CoQ is sufficient to induce adipocyte insulin resistance , and this may contribute to off-target effects of statin therapy . Mitochondrial CoQ is essential for cellular respiration , as it shuttles electrons from various membrane-bound/associated dehydrogenase complexes to complex III during oxidative phosphorylation . In addition , CoQ can regulate the formation of superoxide anion radicals from the various CoQ-interacting sites of complexes I , II and III . In this respect , excess mitochondrial CoQ above that required for maximal respiratory flux can be thought of as an ‘electron sink’ . Consistent with this concept , previous studies have shown that modest loss of mitochondrial CoQ can be tolerated for electron transport activity , but at the cost of increased mitochondrial oxidants ( Quinzii et al . , 2008 ) . We hypothesised that loss of CoQ in mitochondria may contribute to increased oxidants in insulin resistance . To test this possibility , we utilised peroxiredoxin ( PRDX ) dimerisation as an indicator of subcellular oxidant burden ( Perkins et al . , 2015 ) . Peroxiredoxins undergo homodimerisation as part of their mechanism to reduce hydroperoxides particularly hydrogen peroxide ( H2O2 ) . Therefore , the PRDX dimer:monomer ratio is a useful surrogate to assess subcellular H2O2 ( Bayer et al . , 2013 ) . There was no change in the total content of PRDX1-3 in insulin-resistant models ( Figure 6A , Figure 6—figure supplement 1A , C , J ) and the dimer:monomer ratio of cytosolic PRDX1 and PRDX2 also remained unchanged ( Figure 6—figure supplement 1A–D ) . In contrast , the dimer:monomer ratio of mitochondrial PRDX3 increased significantly in all in vitro models ( Figure 6A–6B ) . Increased dimerisation of PRDX3 was also observed under conditions of pharmacological ( Figure 6C , Figure 6—figure supplement 1E ) or genetic inhibition ( Figure 6—figure supplement 1G–H ) of CoQ biosynthesis , with limited or no changes in PRDX2 redox state ( Figure 6—figure supplement 1F , I ) . Restoration of normal mitochondrial CoQ9 content by provision of exogenous CoQ9 lowered the PRDX3 dimer:monomer ratio ( Figure 6D–6E ) . PRDX3 dimerisation was also enhanced in insulin-resistant adipose tissue at 5 and 14 d HFHSD feeding ( Figure 6F , Figure 6—figure supplement 1J ) . In vivo administration of CoQ10 under conditions that improved insulin sensitivity lowered the PRDX3 dimer-to-monomer ratio in a dose-dependent manner ( Figure 6G , Figure 6—figure supplement 1K ) . Together , these data place decreased mitochondrial CoQ upstream of increased mitochondrial oxidants , most likely in the form of H2O2 in adipocyte insulin resistance . To determine if increased mitochondrial H2O2 was necessary for loss of mitochondrial CoQ to cause insulin resistance , we over-expressed mitochondria-targeted catalase ( Figure 6—figure supplement 1L ) in the setting of CoQ deficiency . This lowered the PRDX3 dimer:monomer ratio in cells where CoQ biosynthesis was inhibited ( Figure 6H Figure 6—figure supplement 1M–O ) , and improved insulin responses in these conditions ( Figure 6I–6J ) , consistent with loss of mitochondrial CoQ causing insulin resistance via mitochondrial H2O2 . We next examined the effect of loss of mitochondrial CoQ on mitochondrial function and oxidant production in more detail . Although the relationship between impaired mitochondrial function and insulin resistance is controversial ( Montgomery and Turner , 2015 ) , we first examined whether insulin resistance or loss of CoQ was associated with bioenergetic defects since CoQ plays a key role in oxidative phosphorylation . We assessed mitochondrial respiration in all in vitro models of insulin resistance . In control and insulin-resistant 3T3-L1 adipocytes cultured in galactose , to force ATP-production via mitochondria ( Aguer et al . , 2011 ) , we observed no defect in basal or maximal ( uncoupler-induced ) respiration ( Figure 7A ) . Instead , basal oxygen consumption was increased in multiple models ( Figure 7A ) . To explore this further , we measured oxygen consumption in digitonin-permeabilised 3T3-L1 adipocytes to assess maximal respiratory function ( Figure 7B–D , Figure 7—figure supplement 1A ) . In these experiments , FAD-linked respiratory capacity assessed via succinate was the only activity compromised in all models ( Figure 7C ) . This defect was specific to succinate dehydrogenase , since oxidation of medium-chain fatty acid ( i . e . , octanoylcarnitine , which also donates electrons to CoQ via electron-transferring-flavoprotein dehydrogenase [Ruzicka and Beinert , 1977] ) , was not altered ( Figure 7—figure supplement 1A ) . Collectively , these data suggest that loss of mitochondrial CoQ decreases succinate-driven complex II capacity , but this does not limit overall mitochondrial respiration in cells . Although mitochondrial superoxide ( Hoehn et al . , 2009 ) and hydrogen peroxide ( H2O2 ) ( Anderson et al . , 2009; Paglialunga et al . , 2015 ) have been implicated in insulin resistance , the cause of increased oxidants has yet to be determined . Elevated mitochondrial H2O2 in response to lower CoQ content is likely caused by increased production of the superoxide anion radical , the precursor of H2O2 . To test this directly , we determined superoxide in specific in vitro models of adipocyte insulin resistance , using LC-MS to quantify the conversion of mito-hydroethidine to the superoxide-specific product mito-2-hydroxyethidium ( Zielonka et al . , 2008 ) . Mitochondrial superoxide was increased in the CI model ( Figure 7E ) and in adipocytes where CoQ biosynthesis was inhibited ( Figure 7E–7F ) , implying that increased mitochondrial H2O2 following loss of CoQ was due to increased superoxide production . We next used a series of mitochondrial poisons to determine the site of oxidant production . First , we assessed whether increased H2O2 was produced by the respiratory chain by incubating cells with chemical uncouplers to depolarise mitochondria ( Fisher-Wellman et al . , 2013 ) . BAM15 and FCCP had no effect on PRDX3 dimerisation in control cells but lowered the PRDX3 dimer/monomer status to control levels in cells treated with NB ( Figure 7G , Figure 7—figure supplement 1C ) , or in cells in which the expression of Coq7 and Coq9 were reduced using siRNA ( Figure 7—figure supplement 1D–E ) . This established that loss of CoQ increased H2O2 in a coupled respiration-dependent manner . Administration of the complex II inhibitors TTFA ( Figure 7—figure supplement 1B ) and malonate lowered the PRDX3 dimer/monomer ratio to near control values in cells treated to inhibit CoQ biosynthesis ( Figure 7G , Figure 7—figure supplement 1C–E ) and in other models of insulin resistance ( Figure 7H , Figure 7—figure supplement 1H–I ) , suggesting that increased H2O2 in insulin-resistant adipocytes was dependent on complex II . The majority of superoxide from succinate-driven respiration via complex II has been reported to result from reverse electron transport from CoQH2 into complex I , which can be inhibited with rotenone ( Quinlan et al . , 2012 ) . To test whether this may account for H2O2 generated in response to loss of CoQ we tested whether NB-responsive PRDX3 dimerisation was inhibited by rotenone . Rotenone had no effect on the PRDX3 redox state in NB-treated cells , similar to what was observed for antimycin A ( a complex III inhibitor ) and oligomycin ( a complex V inhibitor ) ( Figure 7—figure supplement 1F ) . Further , treatment of 3T3-L1 adipocytes with NB did not change the CoQ redox state , just as the CoQ redox state was not altered in other in vitro models of insulin resistance ( Figure 7—figure supplement 1G ) . Since a more reduced CoQ pool is required for reverse election transport from CoQH2 to complex I ( Murphy , 2009 ) , these data suggest that reverse electron transport was not involved in the observed increase in superoxide/H2O2 resulting from loss of mitochondrial CoQ , and that complex II itself was the likely origin the oxidants ( Quinlan et al . , 2012 ) . To test whether inhibition of oxidant production from complex II using TTFA and malonate could overcome insulin resistance we assessed insulin-stimulated HA-GLUT4 translocation in 3T3-L1 insulin-resistant adipocytes in the presence or absence of these complex II inhibitors . Treatment of control 3T3-L1 adipocytes with TTFA did not impair insulin-stimulated HA-GLUT4 translocation ( Figure 7I ) or insulin-regulated inhibition of lipolysis ( data not shown ) , suggesting that impaired complex II activity does not cause adipocyte insulin resistance per se . Both TTFA and malonate improved insulin-stimulated HA-GLUT4 translocation to the PM in all models of insulin resistance tested ( Figure 7I–7J ) , albeit not to the same level as observed in control cells . Further , TTFA improved insulin-stimulated 2DOG uptake in adipose explants isolated from mice fed a HFHSD ( Figure 7K ) . Taken together , these data suggest that lower mitochondrial CoQ accelerates superoxide generation , most likely from complex II , which in turn elevates the mitochondrial H2O2 burden , promotes PRDX3 dimerisation and drives insulin resistance . Mitochondrial oxidants have been reported to play an important role in the development of insulin resistance in adipose ( Hoehn et al . , 2009; Houstis et al . , 2006; Paglialunga et al . , 2015 ) and muscle tissue ( Anderson et al . , 2009; Hoehn et al . , 2009 ) . This is thought to occur primarily via increased production of superoxide or H2O2 in mitochondria ( Anderson et al . , 2009; Hoehn et al . , 2009; Paglialunga et al . , 2015 ) , yet the proximal mechanism that triggers mitochondrial oxidant production has remained elusive . Here , we provide insights into the sequence of events that lead to increased oxidant production and insulin resistance . Mass-spectrometry-based proteomic analysis of adipose tissue from mice fed a HFHSD , insulin-resistant 3T3-L1 adipocytes and adipose tissue from insulin-resistant humans , revealed down-regulation of the mevalonate/CoQ biosynthesis pathway in multiple models , and a separate targeted metabolite analysis revealed a common decrease in mitochondrial CoQ content in these models as well as in insulin-resistant muscle . Decreased mitochondrial CoQ was sufficient to cause insulin resistance via a mechanism requiring mitochondrial oxidants , while restoration of mitochondrial CoQ restored insulin sensitivity . These data suggest a novel pathway that may drive insulin resistance across a broad range of models including muscle and adipose tissue from mice and adipose tissue from obese humans . The pathway involves decreased expression of mevalonate pathway/CoQ biosynthetic enzymes , lower mitochondrial CoQ and insulin resistance as a result of oxidant production primarily from complex II . Based on the current study , many insults implicated in insulin resistance including hyperinsulinaemia , inflammation , corticosteroids and caloric excess converge upon loss of mitochondrial CoQ as a potential cause of insulin resistance . Mechanistically , this could be explained by decreased expression of CoQ biosynthetic enzymes , and lower CoQ synthesis rates , in a majority of models studied . Related to this , our finding that mevalonate pathway inhibiting statins lowered CoQ content and caused insulin resistance in a CoQ-dependent manner may shed new light on the link between statin therapy in humans and insulin resistance ( Cederberg et al . , 2015; Preiss et al . , 2011; Sattar et al . , 2010 ) . The lack of concordance between transcript and protein expression within the mevalonate/CoQ pathway ( Figure 2—figure supplement 1 ) across the different insulin-resistant models suggests that there may be multiple mechanisms by which this pathway is targeted in response to different upstream insults and this maybe mediated via either transcriptional or post-translational regulation . Furthermore , our subcellular analysis of CoQ content and data from the chronic insulin model indicate that other aspects of CoQ biology that we do not yet understand may be involved in regulating mitochondrial CoQ abundance in insulin-resistant conditions . These features could include the regulation of CoQ turnover and/or its trafficking between mitochondria and other parts of the cell . Together , this supports the notion that various insults act in different ways , all decreasing mitochondrial CoQ as a common means of inducing insulin resistance . Modest loss of CoQ has been reported to increase mitochondrial oxidants in a range of cellular systems ( Cornelius et al . , 2013; Duberley et al . , 2013; Quinzii et al . , 2013; Quinzii et al . , 2012; Rodríguez-Hernández et al . , 2009 ) , although the precise mechanism for this effect remains unknown . Our data are consistent with loss of mitochondrial CoQ increasing mitochondrial superoxide/H2O2 production because we observed increased superoxide in response to reduced CoQ biosynthesis using a highly specific mass-spectrometry-based assay for superoxide ( Figure 7E–F ) . Our data also suggest that this superoxide is , in part , derived from the flavin site of complex II ( IIF ) ( Quinlan et al . , 2012 ) since the complex II inhibitors TTFA and malonate lowered PRDX3 dimer/monomer status . Oxidation of succinate via complex II has been reported to generate large amounts of superoxide via reverse electron transport to complex I . However , rotenone had no effect on PRDX3 dimerisation in adipocytes treated with NB and we detected no difference in the overall CoQ redox state , suggesting that under these conditions superoxide/H2O2 does not originate from complex I via reverse electron transfer . Unlike electron transfer from complex I to CoQ , electron transfer from flavoproteins in complex II to CoQ is not limited by the energetic constraints established by the membrane potential . This means that electrons can be transferred to CoQ whenever additional substrate is made available to these flavoproteins , provided oxidized CoQ is available to receive the electrons . In addition , although there is some evidence that complex III is a site of superoxide production ( Quinlan et al . , 2013 ) , the majority of superoxide in the mitochondria is derived from flavin sites , including complex II ( Quinlan et al . , 2013; Starkov and Fiskum , 2003; Tretter et al . , 2007 ) . Therefore , we hypothesise that increased superoxide production from the IIF site is due to increased steady-state concentrations of the flavin radical as a result of impaired electron transfer to CoQ at the binding site IIQ , due to decreased CoQ . This interpretation is supported by our finding that maximal complex II activity was impaired in all models studied . Alternatively , lower CoQ may favour reverse electron transfer to complex II , and superoxide production from IIF , under conditions where other enzymes ( e . g . mitochondrial glycerol-3-phosphate dehydrogenase [Orr et al . , 2012] ) are feeding electrons into the CoQ pool ( Quinlan et al . , 2012 ) . However , there are likely additional sites of superoxide production since inhibition of complex II only partially rescued PRDX3 dimerisation . Despite knowledge of increased oxidants in insulin-resistant humans ( Boden et al . , 2015 ) and that scavenging mitochondrial oxidants benefits insulin sensitivity ( Anderson et al . , 2009; Hoehn et al . , 2009 ) , the mechanism for increased oxidants in mitochondria in insulin resistance has remained unclear . Our data address this question and place loss of mitochondrial CoQ as a common defect and cause of mitochondrial oxidants , via complex II-derived superoxide , and insulin resistance in adipocytes and perhaps muscle . An important question is how increased mitochondrial oxidants impair insulin action . Our data from cells supplemented with CoQ revealed that improvements in insulin-stimulated glucose transport and inhibition of lipolysis were not associated with improved insulin signalling to Akt or its downstream substrate TBC1D4 . This is consistent with previous reports that defects in insulin-stimulated glucose transport in insulin resistance are not due to obvious defects in proximal insulin signalling ( Hoehn et al . , 2008; Tan et al . , 2015 ) . Although retrograde signalling from the mitochondria to the nucleus is well described it seems unlikely that this mitochondrial oxidant-induced insulin resistance requires changes in transcription because induction of mitochondrial oxidants acutely impairs insulin action ( Hoehn et al . , 2009 ) . Thus , it is more likely that there is a presently undiscovered signal transduction pathway that communicates directly to mediators of insulin action in the cytoplasm . Future studies exploring specific targets involved in this pathway and their connection with mitochondria and oxidants are warranted . Coenzyme Q has received considerable attention as a supplement to ameliorate a range of medical conditions , including diabetes and cardiovascular disease , based on the observation that serum and tissue CoQ10 concentrations are decreased in individuals with these conditions . While there are reports of CoQ supplementation benefitting these conditions ( Amin et al . , 2014; Ayer et al . , 2015; Hodgson et al . , 2002; Mortensen et al . , 2014; Raygan et al . , 2016 ) , the efficacy of CoQ10 in the treatment of diabetes and cardiovascular diseases remains unclear ( Ayer et al . , 2015; Eriksson et al . , 1999; Suksomboon et al . , 2015 ) . Our study provides a reasonable rationale for targeting the CoQ biosynthesis pathway as a potential therapeutic target . Overall low bioavailability of orally administered CoQ10 or CoQ10H2 represents a substantial limitation , particularly in situations of modest CoQ deficiency such as those shown here to be sufficient to initiate insulin resistance , and where mitochondrial CoQ homeostasis needs to be restored in metabolic tissues such as adipose and muscle ( Zhang et al . , 1995 ) . We overcame this limitation in mice by intra-peritoneal administration of CoQ to provide proof-of-principle that restoration of mitochondrial CoQ improves insulin action and whole body glucose tolerance ( Figure 4 ) . Unfortunately , intra-peritoneal administration of CoQ10 is not likely a practical strategy for the treatment of insulin resistance in humans . Pharmacological inhibition of the CoQ biosynthesis pathway , e . g . , by polyisoprenoid epoxides ( Bentinger et al . , 2008 ) , or targeting additional processes that contribute to the regulation of mitochondrial CoQ content may represent potential options in the future . In the present study , we found that the protein levels , but not the corresponding levels of the mRNAs , of the CoQ biosynthetic enzymes COQ7 and COQ9 were decreased in insulin resistance . These proteins form a dimeric complex ( Lohman et al . , 2014 ) the formation of which may be regulated via a post-translational mechanism . Although little is known about the molecular regulation of COQ protein turnover , the CoQ biosynthetic protein complex has been reported to be stabilised by the atypical kinase ADCK3/COQ8 ( He et al . , 2014; Stefely et al . , 2016 ) ) , mitochondrial proteases ( Veling et al . , 2017 ) and CoQ itself ( He et al . , 2014 ) . Destabilisation of the complex lowers expression of many of the COQ proteins altered in insulin-resistant adipose tissue , including COQ3 , 7 and 9 . Therefore , it may be that loss of COQ7 and 9 in insulin resistance is a result of increased turnover due to complex instability . Another therapeutic option is based on our data implicating complex II as the site of increased oxidant production in response to loss of mitochondrial CoQ . Recent chemical screens have successfully identified compounds that prevent superoxide production from complex I ( Brand et al . , 2016 ) and III ( Orr et al . , 2015 ) , without impairing electron transport . Identifying similar compounds for complex II may be useful in mitigating superoxide/H2O2 production and overcoming insulin resistance . Eight-week-old male C57BL/6J mice were purchased from the Animal Resources Centre ( Perth , Australia ) or Australian BioResources ( Moss Vale , Australia ) . The animals were kept in a temperature-controlled environment ( 22 ± 1°C ) on a 12 hr light/dark cycle with free access to food and water . Mice were fed ad libitum for a period of 14 days with a standard lab diet ( CHOW ) ( 13% calories from fat , 22% calories from protein , and 65% calories from carbohydrate , 3 . 1 kcal/g; Gordon's Specialty Stock Feeds , Yanderra , Australia ) or with high fat high sucrose diet ( HFHSD; 47% of calories from fat ( 40% calories from lard ) , 21% calories from protein , and 32% calories from carbohydrates ( 16% calories from starch ) , 4 . 7 kcal/g ) . All experiments were carried out with the approval of the Garvan Institute/St . Vincent's Hospital Animal Experimentation Ethics Committee ( 09/46 ) or the approval of the University of Sydney Animal Ethics Committee ( 2014/694 ) , following guidelines issued by the National Health and Medical Research Council of Australia . All studies used at least five mice per treatment group . Mice from different cages were used to negate cage-specific effects . Epididymal fat depots were excised from mice , transferred immediately to warm DMEM/2% BSA/20 mM HEPES , pH 7 . 4 , and minced into fine pieces . Explants were washed twice and incubated in DMEM/2% BSA/20 mM HEPES , pH 7 . 4 for 2 hr . Explants were then rinsed and incubated in Krebs–Ringer phosphate buffer containing 2% bovine serum albumin ( BSA , Bovostar , Bovogen ) ( KRP buffer; 0 . 6 mM Na2HPO4 , 0 . 4 mM NaH2PO4 , 120 mM NaCl , 6 mM KCl , 1 mM CaCl2 , 1 . 2 mM MgSO4 and 12 . 5 mM Hepes ( pH 7 . 4 ) ) . Insulin was added for 20 min , and glucose transport was initiated by addition of 2-DOG ( 0 . 25 μCi , 50 μM ) and [14C]mannitol ( Source , 0 . 036 μCi/sample ) for the final 5 min of the assay to measure steady-state rates of 2DOG uptake . For experiments using TTFA ( Figure 7K ) , 100 µM TTFA or equivalent volume of vehicle ( EtOH ) was included during the 2 hr incubation period in DMEM/2% BSA and maintained throughout subsequent washes and incubation . Also , the 2DOG uptake assay was carried out in DMEM without glucose/2% BSA rather than KRP/2% BSA . Uptake was terminated with three rapid washes in ice-cold PBS , after which the cells were solubilised in radioimmune precipitation assay buffer ( RIPA; 50 mM Tris-HCl , pH 7 . 5 , 150 mM NaCl , 1% Triton X-100 , 0 . 5% sodium deoxycholate , 0 . 1% SDS , 1 mM EDTA , and 10% glycerol ) supplemented with protease inhibitors ( Roche ) . Samples were assessed for radioactivity by scintillation counting and the results were normalised for protein content determined by the bicichoninic acid assay . Cages were randomly assigned to diets and investigators were not blinded to experimental groups . Glucose tolerance tests ( GTTs ) and insulin tolerance tests ( ITTs ) were performed on mice following a 6 hr fast from 0700 to 1300 . For GTTs , mice were injected i . p . with 10% glucose solution at 1 g/kg ( Figure 1 ) or 2 mg/kg ( Figure 4 and Figure 4—figure supplement 1 ) per lean mass . For tracer uptake during the GTT , mice were also administered [3H]- 2DOG tracer ( 200 μCi/kg lean weight ) within the 10% glucose solution . For ITTs , mice were i . p . injected with 80 mg/kg pentobarbitone sodium ( Lethabarb Euthanasia Injection , Virbac , Australia ) , after 20 min the abdominal cavity was incised along the midline to reveal the liver to inject 1 U/kg lean weight of insulin and [3H]- 2-DOG ( 200 μCi/kg lean weight ) into the hepatic portal vein . At the times indicated , blood was sampled from the tail tip and blood glucose determined with an Accu-Chek II glucometer ( Roche ) . Clearance of the [3H]- 2DOG tracer from the blood during the GTT and ITT was assessed to allow calculation of tracer disappearance . [3H]- 2DOG tracer uptake into epididymal and inguinal adipose tissue and quadriceps muscle and conversion to glucose-6-phospate was determined as previously described ( Smith et al . , 2007 ) . Plasma NEFAs were measured using NEFA C ( Wako , Osaka , Japan ) according to the manufacturer’s instructions . Cages were randomly assigned to diets and investigators were not blinded to experimental groups . Blood samples were obtained via tail bleeds using 5 µL heparinised hematocrit tubes ( Drummond ) and ejected into a mouse ultra-sensitive insulin ELISA ( 90080 , Crystal Chem ) . ELISA performed as per manufacturer’s instructions . Mycoplasma-free 3T3-L1 fibroblasts obtained from 3T3-L1 Howard Green ( Harvard Medical School , Boston , MA ) were maintained in Dulbecco’s Modified Eagle Medium ( DMEM ) ( Thermo Fisher Scientific ) supplemented with 10% fetal calf serum ( FCS ) ( Thermo Fisher Scientific ) , 1% GlutaMAX ( Thermo Fisher Scientific ) in a humidified atmosphere with 10% CO2 . HA-GLUT4 overexpressing 3T3-L1 fibroblasts were generated by retroviral transduction as previously described ( Govers et al . , 2004 ) . Confluent 3T3-L1 cells were differentiated into adipocytes by the addition of DMEM containing 0 . 22 µM dexamethasone , 100 ng/mL biotin , 2 µg/mL insulin , 500 µM IBMX ( day 0 ) . After 72 hr , medium was replaced with DMEM/10% FCS/GlutaMAX containing 2 µg/mL insulin ( day three post differentiation ) . After a further 72 hr ( day six post differentiation ) , cells were cultured in DMEM/10% FCS/GlutaMAX . Medium was subsequently replaced every 48 hr . Cells were used between days 10 and 15 after the initiation of differentiation . For stable isotope labelling of amino acids in cell culture ( SILAC ) -based proteomics , 3T3-L1 fibroblasts were passaged for six cell divisions in DMEM ( Sigma Alrich ) /10% dialysed FCS ( Thermo Fisher Scientific ) containing L-arginine ( Arg 0 ) and L-lysine ( Lys 0 ) ( ‘light’ ) , L-arginine-U-13C614N4 ( Arg 6 ) and L-lysine-2H4 ( Lys 4 ) ( ‘medium’ ) or L-arginine-U-13C615N4 ( Arg 10 ) and L-lysine-U-13C615N2 ( Lys 8 ) ( ‘heavy’ ) . Final concentrations of arginine and lysine were 33 µg/mL and 76 µg/mL , respectively . This strategy generated three distinct SILAC populations . We periodically tested labelling efficiency by mass spectrometry analysis . SILAC-labelled fibroblasts were differentiated into adipocytes as above . Insulin resistance was induced by dexamethasone , tumour necrosis factor-α ( TNF ) , or hyperinsulinaemia as previously described ( Hoehn et al . , 2008 ) . The chronic insulin ( CI ) model of hyperinsulinemia was created by addition of 10 nM insulin to adipocytes at 1200 , 1600 and 2000 hr on day 1 and 0800 hr the following day . Glucocorticoid-induced insulin resistance was recreated with 20 nM dexamethasone ( Dex ) ( 0 . 01% ethanol carrier as control ) , starting on day seven post initiation of differentiation and media was changed every other day for 8 d . Chronic low-dose inflammation was mimicked in 3T3-L1 adipocytes by incubation with 2 ng/mL TNFα ( Calbiochem ) for 4 d . Medium was changed every 24 hr . For CoQ treatment , cells were incubated with 10 µM CoQ9 ( Sigma Aldrich ) or 10 µM liposomal CoQ10 ( LiQsorb , Tichson Corp . ) for 24 hr prior to assays . CoQ9 was dissolved in ethanol at 5 mM and diluted to 10 µM in pre-warmed DMEM/10% FCS , 1% GlutaMAX and incubated at 37°C for 30 min prior to addition to cells . Ethanol was used as a vehicle control ( 0 . 2% ) . Following 2 hr serum-starvation in DMEM/0 . 2% BSA/1% GlutaMAX , cells were washed and incubated in pre-warmed Krebs–Ringer phosphate buffer containing 0 . 2% bovine serum albumin ( BSA , Bovostar , Bovogen ) ( KRP buffer; 0 . 6 mM Na2HPO4 , 0 . 4 mM NaH2PO4 , 120 mM NaCl , 6 mM KCl , 1 mM CaCl2 , 1 . 2 mM MgSO4 and 12 . 5 mM Hepes ( pH 7 . 4 ) ) . Cells were stimulated with 100 nM insulin for 20 min . To determine non-specific glucose uptake , 25 μM cytochalasin B ( ethanol , Sigma Aldrich ) was added to the wells before addition of 2-[3H]deoxyglucose ( 2-DOG ) ( PerkinElmer ) . During the final 5 min 2-DOG ( 0 . 25 μCi , 50 μM ) was added to cells to measure steady-state rates of 2DOG uptake . Following three washes with ice-cold PBS , cells were solubilised in PBS containing 1% ( v/v ) Triton X-100 . Tracer uptake was quantified by liquid scintillation counting and data normalised for protein content . Data were further normalised to maximal insulin stimulation of control cells , set to 100% . Determination of plasma membrane HA-GLUT4 was performed as previously described ( Govers et al . , 2004 ) . Briefly , cells were serum-starved for 2 hr in DMEM/0 . 2% BSA/GlutaMAX . Cells were stimulated with 100 nM insulin for 20 min as indicated . Cells were fixed but not permeabilised , and the amount of HA-GLUT4 present at the plasma membrane determined by the accessibility of the HA epitope to anti-HA antibody ( Covance , clone 16B12 ) . Cells were incubated with 20 μg/mL goat anti-mouse Alexa-488-conjugated secondary antibody ( Thermo Fisher Scientific ) . Determination of total HA-GLUT4 was performed in a separate set of cells that underwent the same labelling procedure except that anti-HA staining was performed after permeabilisation of the cells with 0 . 1% ( w/v ) saponin . Total HA-GLUT4 was measured separately for each experimental treatment group . Fluorescence ( excitation 485 nm/emission 520 nm ) was measured using a fluorescent microtiter plate reader ( FLUOstar Galaxy , BMG LABTECH ) . Surface HA-GLUT4 was expressed as a percentage of total HA-GLUT4 . Following 2 hr serum-starvation in DMEM/0 . 2% BSA/1% GlutaMAX , cells were washed and incubated in DMEM containing 3 . 5% fatty acid-free BSA ( Sigma-Aldrich ) , GlutaMAX and 10 mM glucose . Cells were treated with or without 1 nM isoproterenol and/or indicated doses of insulin for 1 hr . Aliquots of medium were taken to assay for glycerol content using Sigma glycerol reagent ( Sigma-Aldrich ) according to the manufacturer's protocol . Following three washes with ice-cold PBS , cells were solubilised in PBS containing 1% ( v/v ) Triton X-100 . Glycerol release as a measure for lipolysis was normalized to cellular protein content . Complete datasets are presented in Figure 4—figure supplement 1 and Figure 5—figure supplement 1 . Insulin-stimulated inhibition of lipolysis ( Figures 4C , 5C and F ) was calculated as glycerol release from cells treated with 0 . 5 nM insulin and 1 nM isoproterenol as a percentage of the glycerol release from cells treated with 1 nM isoproterenol alone . Adipose mitochondrial CoQ content was assessed in a cohort of females studied using 2-stage hyperinsulinaemic-euglycaemic clamps with deuterated glucose tracers and adipose biopsies , as previously described ( Chen et al . , 2015 ) . Briefly , the study protocol was approved by St Vincent’s Hospital Human Research Ethics Committee ( Sydney , Australia ) ( HREC/10/SVH/133 ) and written consent obtained prior to the study . Volunteers were sedentary individuals with obesity ( BMI >30 kg/m2 ) and exclusion criteria were diabetes , treatment with medications known to affect glucose homeostasis , >20 g/d alcohol intake , body weight instability in previous 3 months and known cancer , cardiac , renal or liver disease . All studies were performed at the Clinical Research Facility at the Garvan Institute of Medical Research ( Sydney , Australia ) . The clamp started with a 2 hr primed ( 5 mg/kg ) , continuous ( 3 mg/kg/h ) infusion of [6 , 6-2H2]glucose , followed by a 2 hr infusion of low-dose insulin ( 15 mU/m2/min ) and a 2 hr infusion of high-dose insulin ( 80 mU/m2/min ) . The deuterated glucose infusion rate was halved ( 1 . 5 mg/kg/h ) during , and ceased at the end of , the low-dose insulin infusion . Glucose was infused to maintain whole-blood concentration of 5 mmol/L with variable rate infusion of dextrose ( 25% , enriched to 2 . 5% with deuterated glucose ) . Endogenous glucose production and non-esterified fatty acid ( NEFA ) suppression during the low- dose insulin clamp reflect hepatic and adipose insulin resistance , while glucose infusion rate during the last 30 min of the high-dose insulin clamp normalised to body fat free mass estimates muscle insulin resistance ( Chen et al . , 2015 ) . Periumbilical subcutaneous fat biopsy was performed during the basal clamp stage under sterile conditions using a trocar , as described ( Chen et al . , 2015 ) . Adipose tissue ( 50 mg ) was fixed , dehydrated , paraffin embedded and sectioned and adipocyte cell size measured as previously described ( Chen et al . , 2015 ) or snap frozen and stored in −80°C for processing and analysis of CoQ content . Body composition was evaluated using dual-energy X-ray absorptiometry and abdominal fat distribution and liver fat by magnetic resonance imaging as previously described ( Chen et al . , 2015 ) . To separate subjects into groups ( e . g . high vs low mitochondrial CoQ10 concentrations as in Table 1 , or insulin sensitive vs insulin resistant in Figure 3E and Figure 3—figure supplement 1K ) , subjects were divided into an upper tertile ( typically 11 subjects ) or lower two tertiles ( typically 22 subjects ) based on specified parameters ( e . g . mitochondrial CoQ10 concentration ( Table 1 ) , suppression of NEFAs ( Figure 3E ) , glucose infusion rates [Figure 3—figure supplement 1K] ) . Epididymal adipose tissue was excised from mice fed a HFHSD for the specified durations ( five mice per time point ) . Tissue was lysed in 6 M urea , 2 M thiourea , 25 mM triethylammonium bicarbonate , pH 7 . 9 containing phosphatase and protease inhibitor cocktails ( Roche ) by tip-probe sonication ( 2 × 30 s ) on ice . Lysates were centrifuged at 17 , 000 x g , 15 min , 4°C . The fat cake was removed and the supernate precipitated with 6 volumes of acetone , overnight at −20°C . Pelleted protein was re-suspended in 6 M urea , 2 M thiourea , 25 mM triethylammonium bicarbonate , pH 7 . 9 and quantified by Qubit fluorescence ( Thermo Fisher Scientific ) . 100 µg of protein was subjected to reduction with 10 mM DTT for 60 min at 25°C and alkylated with 25 mM iodoacetamide for 30 min at 25°C in the dark . Excess iodoacetamide was then removed by reaction 20 mM DTT and the sample digested with Lys-C ( Wako ) at 1:50 ( w/w ? ) enzyme to substrate ratio for 2 hr at 25°C . The mixture was diluted 5-fold with 25 mM triethylammonium bicarbonate and digested further with trypsin at 1:50 enzyme to substrate ratio for 12 hr at 30°C . The peptide mixture was acidified to a final concentration of 2% formic acid , 0 . 1% trifuoroacetic acid and centrifuged at 16 , 000 x g for 15 min . Peptides were desalted using hydrophilic lipophilic balance – solid phase extraction ( HLB-SPE ) cartridges ( Waters ) followed by elution with 50% acetonitrile , 0 . 1% trifuoroacetic acid and dried by vacuum centrifugation . SILAC 3T3-L1 adipocytes were left untreated ( insulin sensitive ) or treated to induce insulin resistance ( insulin resistant ) as described above . Four biological replicates were used for all SILAC experiments . SILAC mixes were made so that a SILAC doublet contained a constant untreated sample acting as a reference for each insulin-resistant model . A label switch was performed in biological replicates to ensure that identified changes in protein expression were due to experimental manipulation . For example , the light versus heavy doublet was analysed twice; we compared protein content in both insulin sensitive ( light ) versus insulin resistant ( heavy ) and insulin sensitive ( heavy ) versus insulin resistant ( light ) . Cells were serum-starved for 2 hr , washed three times with ice-cold PBS and lysed in RIPA containing protease inhibitors . After 15 min on ice , the protein concentration was determined by the BCA assay ( Thermo Fisher Scientific ) . Insulin-resistant samples were mixed in a 1:1 ratio based on total protein concentration with an insulin sensitive control to form a SILAC doublet . Per SILAC doublet , two samples were prepared as follows: mixed lysates were centrifuged at 18 , 000 x g for 15 min at 4°C to pellet insoluble material , and the resulting supernate ( RIPA-soluble fraction ) collected . The pellet was re-dissolved in RIPA buffer containing 4% SDS , sonicated ( Bandelin SONOPLUS ) ( 1 s × 12 ) , and centrifuged at 18 , 000 x g for 15 min , and the resulting supernate ( RIPA-insoluble fraction ) collected . The pH of the RIPA-soluble and –insoluble fraction was adjusted to ~8 by the addition of 50 mM Tris , pH 8 . 8 . Protein thiols were reduced by incubating samples with 1 mM dithiothreitol ( DTT ) at 95°C for 3 min , followed by incubation for 25 min at room temperature and alkylation with 5 . 5 mM iodoacetamide in the dark at room temperature for 20 min . Proteins were precipitated in 5 volumes of acetone at −20°C overnight . Precipitated proteins were pelleted at 15 , 000 x g , allowed to air dry before being re-suspended in 4% SDS , 50 mM Tris , pH 6 . 8 . The protein concentration of each sample was then determined by the BCA assay . RIPA-soluble and RIPA-insoluble samples for each SILAC doublet were prepared for SDS-PAGE by diluting in sample buffer ( Thermo Fisher Scientific ) . Samples were separated by SDS-PAGE on 4–20% gradient gels ( Thermo Fisher Scientific ) . Gels were stained with SYPRO Ruby Protein Gel Stain ( Thermo Fisher Scientific ) following the manufacturer's instructions . Lanes containing RIPA-soluble samples were cut into 11 slices , and lanes containing RIPA-insoluble samples were divided into eight slices . Each slice contained approximately the same level of SYPRO Ruby signal and was cut into 1 mm x 1 mm squares to assist protein extraction . Gel pieces were de-stained in 250 mM ammonium bicarbonate pH 8 , 50% acetonitrile under constant agitation for 30 min . Destain solution was removed and gel pieces dried in 100% acetonitrile . After 10 min , acetonitrile was removed and proteins were digested by the addition of modified trypsin ( Promega , 12 . 5 ng/μL ) in 100 mM NH4HCO3 . Digestion was carried out overnight at 37°C . The digestion was stopped by the addition of 5% formic acid and peptides eluted with acetonitrile . Finally , peptide solutions were dried , re-suspended in 1% trifluoroacetic acid and desalted on C18 Stagetips ( 3M , Empore ) . Adipose tissue homogenates were diluted 1:1 in 6 M guanidine in 100 mM Tris pH 7 . 5 containing 10 mM Tris ( 2-carboxyethyl ) phosphine and 40 mM chloroacetamide , and heated at 95°C for 5 min . The lysate was tip-probe sonicated and centrifuged at 20 , 000 x g for 30 min at 4°C . The supernatant was precipitated with 6 volumes of acetone , overnight at −20°C . Pelleted protein were re-suspended in 10% trifluoroethanol in 100 mM Tris , pH 7 . 5 and quantified by BCA . Seven µg of protein was digested with 140 ng sequencing grade Lys-C ( Wako ) for 2 hr at 25°C followed by 140 ng of sequencing grade trypsin ( Sigma Aldrich ) overnight at 37°C . The digest was acidified to a final concentration of 1% trifluroacetic acid ( TFA ) , and peptides purified using SDB-RPS solid-phase disks ( Sigma Aldrich ) and eluted with 1% ammonium hydroxide in 80% acetonitrile . Peptides were dried by vacuum centrifugation and resuspended in 0 . 1% TFA in 2% acetonitrile . For mouse adipose tissue experiments , peptides were loaded onto a 50 cm column with 75 µm inner diameter , packed in-house with 1 . 9 µM C18 ReproSil particles ( Dr Maisch GmbH ) . Reversed-phase chromatography was performed on an Easy nLC1000 HPLC using a binary buffer system of 0 . 5% acetic acid ( buffer A ) and 80% acetonitrile/0 . 5% acetic acid ( buffer B ) . Peptides were separated by linear gradients of buffer B from 5% to 35% over 240 min , at a flow rate of 250 nL/min and electrosprayed into the mass spectrometer by the application of 2 . 3 kV using a liquid junction connection . Ionised peptides were analysed on Q-Exactive ( Thermo Fisher Scientific ) . The Q-Exactive was operated in data-dependent acquisition mode , acquiring survey scans of 3 million ions at a resolution of 70 , 000 at 200 m/z . Twenty of the most abundant isotope patterns from each of the survey scans with charge state ≥2 were selected with an isolation window of 2 Th , and fragmented in the HCD cell with NCE of 25 . Maximum ion fill times for the MS/MS scans were 120 ms , target fill value was 1e5 ions with an under fill ratio of 20% . Fragmented ions were analysed with high resolution ( 35 , 000 at 400 m/z ) in the Orbitrap analyser . Dynamic exclusion was enabled with a duration of 60 s . For SILAC cell culture experiments , peptides were loaded onto a 20 cm column with 75 µm inner diameter , packed in-house with 3 µM C18 ReproSil particles ( Dr Maisch GmbH ) . Reversed-phase chromatography was performed on either the Dionex Ultimate 3000 or the Easy nLC II HPLC using a binary buffer system of 0 . 5% acetic acid ( buffer A ) and 80% acetonitrile/0 . 5% acetic acid ( buffer B ) . Peptides were separated by linear gradient of buffer B from 5% to 35% over 240 min , at a flow rate of 250 nL/min and electrosprayed into the mass spectrometer by the application of 1 . 9–2 . 3 kV using a liquid junction connection . Ionised peptides were analysed on an Orbitrap Velos ( Thermo Fisher Scientific ) . The Orbitrap Velos was operated in data-dependent acquisition mode , acquiring survey scans of 1 million ions at a resolution of 30 , 000 at 400 m/z . Three survey-scan ranges ( 350–1050 m/z , 850–1750 m/z and 350–1750 m/z ) were utilised to enhance dynamic range . Five ( for the first two mass windows ) or seven ( for the last mass window ) of the most abundant isotope patterns from each of the survey scans with charge state ≥2 were selected with an isolation window of 2 Th , and fragmented in the HCD cell with NCE of 40 . Maximum ion fill times for the MS/MS scans were 150 ms , target fill value was 4e4 ions , and ion selection thresholds were 1e4 . Fragmented ions were analysed with high resolution ( 7500 resolution at 400 m/z ) in the Orbitrap analyser . Dynamic exclusion was enabled with a duration of 60 s and a mass window of ±7 ppm . Lock-mass was enabled using 445 . 120025 . On the Q-Exactive , survey scans ( 300–1600 m/z ) were acquired at a resolution of 70 , 000 and up to 15 of the most intense isotope patterns were selected for fragmentation in the HCD cell with NCE 25 . Ion fill targets for MS/MS were 1E5 ions and fragmented ions were with high resolution ( 17 , 500 ) in the Orbitrap analyser . For human adipose tissue experiments , peptides were loaded onto a 50 cm column with 75 µm inner diameter , packed in-house with 1 . 9 µM C18 ReproSil particles ( Dr Maisch GmbH ) . Reversed-phase chromatography was performed on an Easy nLC1200 HPLC using a binary buffer system of 0 . 1% formic acid ( buffer A ) and 80% acetonitrile/0 . 5% acetic acid ( buffer B ) . Peptides were separated by linear gradients of buffer B from 2% to 35% over 180 min , at a flow rate of 300 nL/min and electrosprayed into the mass spectrometer by the application of 2 . 4 kV using a liquid junction connection . Ionised peptides were analysed on Q-Exactive HF ( Thermo Fisher Scientific ) operated in data-dependent acquisition mode , acquiring survey scans of 3 million ions at a resolution of 60 , 000 at 200 m/z . Fifteen of the most abundant isotope patterns from each of the survey scans with charge state ≥2 were selected with an isolation window of 1 . 4 Th , and fragmented in the HCD cell with NCE of 27 . Maximum ion fill times for the MS/MS scans were 125 ms , target fill value was 2e5 ions with a target threshold set to 1e5 Fragmented ions were analysed with high resolution ( 15 , 000 at 200 m/z ) in the Orbitrap analyser . Dynamic exclusion was enabled with a duration of 60 s . For mouse adipose tissue experiments , raw mass spectrometry data were processed using the MaxQuant software ( Cox and Mann , 2008; Cox et al . , 2009 ) version 1 . 4 . 0 . 8 using the default settings with minor changes: Oxidised Methionine ( M ) and Acetylation ( Protein N-term ) were selected as variable modifications , and carbamidomethyl ( C ) as fixed modification . A maximum of two missed cleavages was permitted , 10 peaks per 100 Da , MS/MS tolerance of 20 ppm , and a minimum peptide length of 6 . The ‘matching between runs’ algorithm was enabled with a time window of 2 min to transfer identifications between adjacent samples . Database searching was performed using the Andromeda search engine ( Cox et al . , 2011 ) integrated into the MaxQuant environment against the mouse Uniprot proteome database , concatenated with known contaminants and reversed sequences of all entries . Protein , peptide and site FDR was controlled at a maximum of 1% respectively . All contaminants and reverse sequenced peptides were removed and the data were log2 transformed . Proteins with 60% or more missing values within each experimental group were removed . Missing expression data within each group was imputed using nearest neighbour averaging or K-nearest neighbours ( k = 5 ) from R package impute ( Trevor Hastie et al . , 2018 ) . For SILAC cell culture experiments , raw mass spectrometry data were processed using the MaxQuant software ( Cox and Mann , 2008; Cox et al . , 2009 ) version 1 . 3 . 0 . 5 using the default settings with minor changes: oxidised methionine ( M ) and acetylation ( Protein N-term ) were selected as variable modifications , and carbamidomethyl ( C ) as fixed modification , as well as double SILAC labels ( either Arg 0/Lys 0 and Arg 10/Lys 8 or Arg 6/Lys 4 and Arg 10/Lys 8 ) . A maximum of two missed cleavages was permitted , 10 peaks per 100 Da , MS/MS tolerance of 20 ppm , and a minimum peptide length of 6 . The ‘matching between runs’ algorithm was enabled with a time window of 2 min to transfer identifications between adjacent gel slices , only for samples analysed using the same nanospray conditions . Database searching was performed using the Andromeda search engine integrated into the MaxQuant environment against the mouse Uniprot proteome database , concatenated with known contaminants and reversed sequences of all entries . Protein , peptide and site FDR was controlled at a maximum of 1% respectively . All contaminants and reverse sequenced peptides were removed , and the data were log2 transformed . This lead to the identification of 5908 protein groups , of which 2620 correspond to unique proteins and 3288 contained in multiple proteins . We subsequently applied Re-Fraction ( Yang et al . , 2012 ) on protein groups that contained multiple proteins and resolved a further 960 unique protein identifications . For groups that could not be identified as a single protein , proteins with the lowest posterior error probability were selected . For human adipose tissue experiments , raw mass spectrometry data were processed using the MaxQuant software ( Cox and Mann , 2008; Cox et al . , 2009 ) version 1 . 5 . 7 . 0 using the default settings with minor changes: Oxidised Methionine ( M ) and Acetylation ( Protein N-term ) were selected as variable modifications , and carbamidomethyl ( C ) as fixed modification . A maximum of two missed cleavages was permitted , 10 peaks per 100 Da , MS/MS tolerance of 20 ppm , and a minimum peptide length of 6 . The ‘matching between runs’ algorithm was enabled with a time window of 2 min to transfer identifications between adjacent samples . Database searching was performed using the Andromeda search engine ( Cox et al . , 2011 ) integrated into the MaxQuant environment against the human Uniprot proteome database , concatenated with known contaminants and reversed sequences of all entries . Protein , peptide and site FDR was controlled at a maximum of 1% respectively . All contaminants and reverse sequenced peptides were removed , and the data were log2 transformed . To normalise the protein quantifications across different samples , proteins were first ranked within each sample based on their MS intensities and then ranked across all samples to identify the most stable proteins for data normalisation . Using this approach , 40S ribosomal protein ( RPSA ) was selected for normalisation by calculating a log fold change of protein intensities with respect to RPSA within each sample , allowing relative protein abundance ( refer to as normalised ratio ) to be comparable across all samples for all proteins . RNA was prepared using the RNeasy protocol ( Qiagen , Valencia , CA ) . Quantity and quality of total RNA samples were determined using an ND-1000 spectrophotometer ( Thermo Fisher Scientific ) and Bioanalyzer 2100 ( Agilent Technologies , Palo Alto , CA ) , respectively . RNA with RNA integrity numbers > 8 , 260/280 values > 1 . 8 and 260/230 values > 1 . 8 was considered acceptable . Labelled cRNA was hybridised to Mouse Genome 430 2 . 0 arrays . Initial data analysis files were generated using manufacturer’s GeneChip Command Console Software . Relative mRNA contents were studied for 45 , 000 transcripts for cells and tissue . All statistical data analysis for the gene expression data were performed in the R-programming environment . AT transcripts in the mouse4302 Affymetrix array was pre-processed using R package affyPLM ( Bm , 2004; Bolstad BM et al . , 2005 ) . Background correction , quantile normalisation and summarisation was performed using RMA function ( Irizarry et al . , 2003 ) . Quality control assessments of the arrays were performed and problematic arrays were identified and down-weighted for ensuing statistical inference tests using function arrayWeights ( Ritchie et al . , 2006 ) within LIMMA ( Ritchie et al . , 2015 ) during linear model fitting of the microarray data . Probes were collapsed to gene level using the maximal average expression of probes across all samples . The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE ( Vizcaíno et al . , 2016 ) partner repository with the dataset identifiers PXD005128 and PXD006891 . The microarray discussed in this manuscript have been deposited in NCBI's Gene Expression Omnibus ( Edgar et al . , 2002 ) and are accessible through GEO Series accession numbers GSE87853 and GSE87854 . Differential expression analysis of proteins and transcripts within each model of insulin resistance were performed using moderated t-test from LIMMA package ( Goeman and Bühlmann , 2007; Michaud et al . , 2008; Wu and Smyth , 2012 ) . For the cell models and adipose tissue , proteins and transcripts with an absolute fold change ( FC ) >1 . 5 and controlled for 5% FDR were considered differentially expressed . Empirical Bayes was used for global variance shrinkage and Benjamini and Hochberg method ( Benjamini and Hochberg , 1995 ) was used to correct multiple hypothesis testing . The moderated t-statistics from differential expression analysis in each model was transformed to z-scores and used as the test statistics for the direction analysis ( Yang et al . , 2014 ) to identify genes/proteins altered ( up/down ) across multiple experimental conditions . For the cell models , we performed a 3-dimensional direction analysis ( across CI , Dex and TNF ) whereas in AT , we performed a 2-dimensional direction analysis ( across 5 and 14 d of HFHSD ) . Significance for transcripts was defined as p-value<0 . 001 and for proteins at 0 . 01 . Pathway enrichment for each condition in cell models and adipose tissue were analysed using by testing up- and down-regulation of 192 pathways from the KEGG database ( Kanehisa and Goto , 2000 ) as curated by the Molecular Signature Database C2: curated gene sets ( Liberzon et al . , 2011; Subramanian et al . , 2005 ) using the mean-rank gene-set enrichment test implemented in LIMMA R package ( Goeman and Bühlmann , 2007; Michaud et al . , 2008; Wu and Smyth , 2012 ) in our transcript and protein data . The fold change ( log2 scale ) of genes and proteins from the individual DE analysis in each model was used as the test-statistics for the GSEA analysis . Significance of pathways was defined at p-value<0 . 05 . CoQ pathway members were edited from the original KEGG annotation to include recent annotations and consisted of Hpd , Tat , Coq2/3/4/5/6/7/9 , Adck3 , and Adck4 . For human proteome data , we associated the normalised ratio of each protein with phenotypical information across all samples using pairwise Pearson’s correlation . To perform a similar analysis on pathway level , for each protein , we calculated a relative protein ratio in each sample by subtracting the average of normalised ratios across all samples . Subsequently , for each KEGG pathway , we summarised its overall trend within each sample by summing the relative ratios of all proteins belong to that pathway . The summary statistics for each pathway in each sample was then used to associate with phenotypical information using pairwise Pearson’s correlation . These results were subsequently visualised as heatmaps . To identify pathways enriched across multiple tissue and/or cell models , we used Fisher’s combined statistics to integrate p-values from individual pathway enrichment analysis conducted on each condition in cell models and adipose tissue . Fisher’s combined statistics follows a chi-square distribution and the integrated p-values from the chi-square distribution are converted into z-scores ( referred to as combined z-scores ) for cell model and adipose tissue , respectively , to visualise each pathway and their overall enrichment across multiple conditions in cell models and tissue . CoQ9 and CoQ10 content in 3T3-L1 adipocyte lysates , tissue homogenates , and membrane fractions from cells , and tissue were determined as described previously ( Gay and Stocker , 2004 ) . Investigators were blinded to experimental groups . Briefly , samples were thawed , mixed gently and 50–450 μL placed in a 15 mL screw top tube to which 2 mL methanol and 10 mL of water-washed hexane were added . The mixture was then mixed vigorously for 1 min , centrifuged ( 1430 x g , 5 min , 4°C ) and 9 mL of the top hexane layer collected then dried using a rotary evaporator . The resulting dried lipids were re-dissolved in 180 μL ice-cold mobile phase ( ethanol:methanol:isopropanol: ammonium acetate pH 4 . 4 , 65:30:3:2 , vol/vol/vol/vol ) and transferred into HPLC vials . Cholesterol , CoQ9 , CoQ9H2 , CoQ10 and CoQ10H2 were determined by HPLC using a Supelcosil LC-C18 column ( 5 μm , 250 × 4 . 6 mm ) eluted at 1 mL/min and connected to UV and electrochemical ( ESA CoulArray 5600A ) detectors . NEC was detected at 214 nm , while CoQ9 , CoQ9H2 , CoQ10 and CoQ10H2 were detected at −700 , +700 and +500 mV and quantified against authentic commercial standards obtained from Sigma Aldrich ( USA ) . CoQ9H2 and CoQ10H2 standards were generated by sodium borohydride-mediated reduction of commercial CoQ9 and CoQ10 , respectively . For accurate determination of CoQ9 and CoQ9H2 for calculation of CoQ redox status analyses were performed as above with several modifications to sample preparation to limit oxidation . Cells were washed in PBS containing 100 µM diethylenetriaminepentaacetic acid ( DTPA ) and homogenised in HES buffer pre-gassed with argon . Homogenisation was carried out under an argon atmosphere . Homogenates were processed one-at-a-time and samples dried in the rotary evaporator under argon . Samples were analysed immediately . Liquid chromatography/tandem mass spectrometry was used for the determination of CoQ9 , CoQ9H2 , CoQ10 and CoQ10H2 in human and mouse adipose tissue after in vivo supplementation with CoQ10 . Briefly , to 50–450 μL of adipose tissue homogenates , internal standard ( 200 pmol CoQ8; Avanti Polar Lipids ) were added and homogenates extracted with methanol/hexane as described above . Following evaporation of the hexane phase , the resulting dried lipids were re-dissolved in 150 μL ice-cold ethanol ( HPLC grade ) and 5 μL injected onto an Agilent 1290 UHPLC system connected to an Agilent 6490 triple-quadrupole mass spectrometer . Analytes were separated on a 2 . 6 μm Kinetex XB-C18 100 A column ( 50 × 2 . 10 mm; Phenomenex , USA ) by gradient elution using mobile phase A ( 2 . 5 mM ammonium formate in 95:5 methanol:isopropanol ) and mobile phase B ( 2 . 5 mM ammonium formate in 100% isopropanol ) at 0 . 8 mL/min . The gradient consisted of 0% mobile phase B from 0 to 1 . 5 min , 0–10% B from 1 . 5 to 2 min , 10% B from 2 to 3 min and back to 0% B from 3 to 5 min . Flow was then directed into the triple quadrupole mass spectrometer with parameters set as follows: gas temperature = 250°C; gas flow = 20 L/min; nebuliser pressure = 35 psi; sheath gas heater = 325°C; sheath gas flow = 12 L/min; capillary voltage = 3 , 500 V . Detection of CoQ8 , CoQ9 , CoQ9H2 , CoQ10 and CoQ10H2 was by multiple reaction monitoring ( MRM ) in positive ion mode using the above general mass spectrometry parameters with fragmentor voltage at 380 V and cell accelerator voltage at 5 V . In each case , the fragment ions generated by collision-induced dissociation of the [M + H]+ or the [M+NH4]+ ions were used for quantification . MRM settings for the target analytes were ( parent ion → fragment ion ) ; CoQ8 ( m/z 727 . 1 → 197 . 1 ) with collision energy ( CE ) = 33 V; CoQ9 ( m/z 795 . 5 → 197 . 1 ) with CE = 33 V; CoQ9H2-NH4 ( m/z 816 . 6 → 197 . 1 ) with CE = 25 V; CoQ10 ( m/z 863 . 6 → 197 . 1 ) with CE = 37 V; and CoQ10H2 ( m/z 882 . 7 → 197 . 1 ) with CE = 33 V . CoQ8 , CoQ9 and CoQ10 were quantified against authentic commercial standards obtained from Sigma Aldrich ( USA ) , while CoQ9H2 and CoQ10H2 standards were generated from CoQ9 and CoQ10 , respectively by sodium borohydride-mediated . For determination of CoQ9 synthesis rates , cells were incubated with 50 µM 13C6-4-hydroxybenzoic acid ( Cambridge isotope , USA ) in DMEM containing GlutaMAX and dialysed FCS for 12 hr following treatment to induce insulin resistance . Cells were washed with cold PBS and homogenised in HES-I buffer ( 10 mM HEPES , pH 7 . 4 , 1 mM EDTA , 250 mM sucrose , protease inhibitors ) by 12 strokes of a Dounce homogeniser . To 350 μL of cell homogenates , internal standard ( 200 pmol CoQ8; Avanti Polar Lipids ) was added and homogenates extracted with methanol/hexane as described above . Following evaporation of the hexane phase , the resulting dried lipids were re-dissolved in 150 μL ice-cold ethanol ( HPLC grade ) and 5 μL injected onto an Agilent 1290 UHPLC system connected to an Agilent 6490 triple-quadrupole mass spectrometer with column , mobile phases , gradient elution , flow rate and mass spectrometry parameters as above . MRM settings for 13C6-CoQ9 ( parent ion → fragment ion ) were m/z 801 . 5 → 203 with collision energy ( CE ) = 33 V . 13C6-CoQ9 was quantified against authentic CoQ9 commercial standard . To measure cholesterol and CoQ9 content is different cellular compartments ( Figure 3 and Figure 3—figure supplement 1 ) , we performed isolated plasma membrane , mitochondrial and microsomes by differential centrifugation . 3T3-L1 adipocytes were washed with ice-cold PBS , harvested in ice-cold HES-I , and subsequent steps were carried out at 4°C . Cells were homogenised by 12 strokes of a Dounce homogeniser to yield a whole cell homogenate prior to centrifugation at 700 × g for 10 min to pellet nuclei and unbroken cells . The resulting supernate was centrifuged at 13 , 550 × g for 12 min to pellet the plasma membrane ( PM ) and mitochondria , with the resulting supernate consisting of cytosol and microsomes . This supernate was then centrifuged at 235 , 200 × g for 75 min to obtain a cytosol fraction ( supernate ) and a microsomal fraction ( pellet ) . The PM/mitochondria pellet was re-suspended/washed in HES-I and re-centrifuged at 13 , 550 × g for 12 min . The PM/mitochondria pellet was re-suspended again in HES-I and layered over a high sucrose HES buffer ( 1 . 12 M sucrose , 0 . 05 mM EDTA , 10 mM HEPES , pH 7 . 4 ) and centrifuged at 111 , 160 × g for 60 min in a swing-out rotor . The pellet was the mitochondria fraction . The PM fraction was collected above the sucrose layer , and centrifuged again at 13 , 550 × g for 12 min to achieve a PM pellet . All fractions were re-suspended in HES-I . Protein concentration for each fraction was performed using the BCA assay . For specific assessment of mitochondrial CoQ content in cell culture ( Figure 4 ) , and for assessment of mitochondrial CoQ9 or CoQ10 in adipose tissue , skeletal muscle and liver from mice and adipose tissue from humans we used a protocol to directly enrich mitochondria ( Figure 3 , Figure 3—figure supplement 1 ) . Mitochondrial isolation of cultured adipocytes and tissues from mice and humans was performed as previously described ( Frezza et al . , 2007 ) . Briefly , cells/tissue was homogenised in mitochondrial isolation buffer ( 10 mM Tris-MOPS , pH 7 . 4 , 1 mM EGTA , 200 mM sucrose containing protease inhibitors ) by 12 strokes of a Dounce homogeniser at 4°C and samples kept at 4°C subsequently . Homogenates were centrifuged at 700 x g for 10 min and then the supernate centrifuged at 7000 x g for 10 min to obtain a pellet containing the mitochondria . The pellet was re-suspended in mitochondrial isolation buffer and re-centrifuged at 7000 x g for 10 min . The mitochondrial pellet was finally re-suspended in mitochondrial isolation buffer and protein concentration determined using the BCA assay . Cultured adipocytes or adipose tissue was homogenised in mitochondrial extraction buffer ( 50 mM potassium phosphate , 0 . 5 mM EGTA , 0 . 1% Triton , pH 7 . 4 ) by 12 strokes of a Dounce homogeniser . Homogenates were freeze-thawed three times and insoluble debris pelleted by centrifugation at 5000 x g for 10 min . Citrate synthase activity was assayed by adding diluted samples to citrate synthase assay buffer ( 100 mM Tris:HCl , pH 8 , 100 mM MgCl250 mM EDTA ) containing 0 . 5 mM dithionitrobenzoic acid and 80 µM acetyl-CoA and the reaction monitored spectrophotometrically at 37°C at 412 nm for background activity . Oxaloacetate was added to 1 mM to start the reaction . Background activity prior to the addition of oxaloacetate was subtracted to obtain rates specific to citrate synthase . Citrate synthase activity was calculated as mU/mg using the extinction coefficient dithionitrobenzoic acid for ( ε = 13 . 6 μmol/mL/cm ) and taking sample dilution into account . Cells were washed twice with PBS and lysed in 1% SDS in PBS-containing protease inhibitors . Cell lysates were sonicated ( Bandelin SONOPLUS ) for 12 s and centrifuged at 13 , 000 × g for 15 min at room temperature . The protein concentration of the resulting supernate was then determined by the BCA assay . Proteins ( 5–10 μg ) from cell/tissue lysates or homogenates were resolved by SDS-PAGE and transferred to polyvinylidene difluoride membranes . Membranes were blocked in 5% BSA or skim milk powder in Tris-buffered saline containing 0 . 1% ( v/v ) Tween 20 for 1 hr , followed by an overnight incubation at 4°C with specific primary antibody solutions . Membranes were incubated with an appropriate secondary antibody for 1 hr before signals were detected using ECL ( Thermo Fisher Scientific or Millipore ) on the Chemidoc MP ( Bio-Rad ) or on the Odyssey Fluorescence Detection System ( LiCOR ) . Antibodies detecting multiple mitochondrial complex subunits ( Cat . No . 45–8099 ) and PRDX1 were from Thermo Fisher Scientific ( Cat . No . PA3-750 ) , anti-PRDX2 ( Cat . No . ab109367 , clone: EPR5154 ) and anti-catalase ( Cat . No . ab52477 ) antibodies were from Abcam , anti-PRDX3 antibody from Ab Frontier ( Cat . No . LF-PA0030 ) , anti-14-3-3 antibody was from Santa Cruz ( Cat . No . sc-629 , clone K19 ) , anti-pT642 TBC1D4 ( Cat . No . 4288 ) , anti- TBC1D4 ( Cat . No . 2670 ) , anti-pT308 Akt ( Cat . No . 9275 ) , anti-pS473 Akt ( Cat . No . 4051 ) and anti-Akt ( Cat . No . 4685 ) antibodies were from Cell Signaling Technologies and anti-α-tubulin antibody was from Sigma Aldrich ( Cat . No . T9026 ) . Antibody to GLUT4 was generated in-house . Densitometry analysis was performed using Image Lab 5 . 2 . 1 ( Bio-Rad ) or Image Studio ( LiCOR ) . For studies in chow and HFHSD-fed mice , male C57BL/6J ( n = 9 per treatment ) fed ad libitum with normal chow or HFHSD for a period of 14 d . Mice were treated with 10 mg/kg the liposomal CoQ10 ( LiQsorb , Tichson Corp . ) diluted in saline by intraperitoneal injection every 48 hr for the duration of the feeding regimen . Control mice received saline . Mice were randomly assigned to a treatment group so groups were equally represented within each cage . For dose response studies , male C57BL/6J ( n = 5 per treatment ) fed ad libitum for a period of 14 d with HFHSD were treated with the specified dose of liposomal CoQ10 ( LiQsorb , Tichson Corp . ) ( 0 , 1 , 5 or 10 mg/kg ) diluted in saline by intraperitoneal injection every 48 hr for the duration of the feeding regimen . Saline was given to the control group . Mice were randomly assigned to a treatment group so that at least one mouse per cage was in each group . For each mouse , epididymal fat pads were assessed for mitochondrial content of CoQ9 and CoQ10 , PRDX3 dimer/monomer status and insulin sensitivity by the ex vivo 2DOG uptake assay . For ex vivo 2DOG uptake assay , saline-injected chow mice were included as a control . Differentiated 3T3-L1 adipocytes were treated with 10 µM simvastatin or 20 µM atorvatstin for indicated times or 2 . 5 mM 4-nitrobenzoic acid ( Alam et al . , 1975; Forsman et al . , 2010 ) or 4-chlorobenzoic acid ( Bour et al . , 2011 ) for 48 hr . These doses were selected to lower mitochondrial CoQ content to that observed in insulin-resistant 3T3-L1 adipocytes . Seven days post-differentiation , adipocytes were trypsinised ( 5 × trypsin , EDTA ) ( Thermo Fisher Scientific ) at 37°C , washed twice with PBS , and re-suspended in electroporation solution ( 20 mM HEPES , 135 mM KCl , 2 mM MgCl2 , 0 . 5% Ficoll 400 , 1% DMSO , 2 mM ATP , and 5 mM glutathione , pH 7 . 6 ) with 200 nM scrambled ( sense 5′-CAGTCGCGTTTGCGACTGGTT-3′ ) or pooled anti-Coq7 siRNA ( sense 5′-GGGAUCACGCUGGUGAAUAUTT-3′ , 5′-GGAUGACCUUAGACAAUAUTT-3′ , 5′- GCCUUGUUGAAGAGGAUUAUTT-3′ ) or pooled anti-Coq9 siRNA ( sense 5′-GCAGCAUUCUGAGACACAGTT-3′ , 5′-GCUGGUGAUGAUGCAGGAUTT-3′ , 5′- GCAAUGAACAUGGGCCACATT-3′ ) or ( Shanghai Genepharma ) . Cells were electroporated at 200 mV for 20 ms using an ECM 830 square wave electroporation system ( BTX Molecular Delivery Systems ) and seeded onto Matrigel ( Corning ) -coated plates . Cells were assayed 96 hr following electroporation . Total RNA was extracted from cells using TRIzol reagent ( Thermo Fisher Scientific ) . cDNA synthesis was carried out using PrimeScript first strand cDNA synthesis kit ( Clontech and Takara Bio Company ) . Polymerase chain reactions were performed on the LightCycler 480 II ( Roche Applied Science ) using FastStart Universal SYBR Green Master ( Roche Applied Science ) . Cyclophilin b was used as an internal control . The primer sets used were as follows: mCoq7_F; tttggaccatagctgcattg and mCoq7_R; tgaggcctcttccatactctg , mCoq9_F; tcagcagcattctgagacaca and mCoq9_R; gtgctgtagctgctcctcact , and mCypB-F; ttcttcataaccacagtcaagacc; mCypB-R , accttccgtaccacatccat . The bioenergetics of intact cells was assessed as described previously ( Krycer et al . , 2017 ) . Briefly , On Day 7–8 of differentiation , 3T3-L1 adipocytes were trypsinised with 5x Trypsin/EDTA ( Thermo Fisher Scientific ) in phosphate-buffered saline ( PBS ) and seeded onto XFp cell culture plates coated with Matrigel ( Corning , distributed by Sigma Aldrich ( Castle Hill , NSW , Australia ) ) . Following the indicated treatment , cells were washed three times with warm PBS , once with bicarbonate-free DMEM buffered with 30 mM Na-HEPES , pH 7 . 4 ( DMEM/HEPES ) , and then incubated in DMEM/HEPES supplemented with 0 . 2% ( w/v ) BSA , 25 mM glucose , 1 mM GlutaMAX and 1 mM glutamine , for 1 . 5 hr in a non-CO2 incubator at 37°C . Cells were then washed once with PBS , once with DMEM/HEPES , and assayed in the XFp Analyzer in DMEM/HEPES with 25 mM galactose , 1 mM GlutaMAX and 1 mM glutamine and 1 mM NaHCO3 . During the assay , respiration was assayed with mix/wait/read cycles of 3/0/2 min for 3T3-L1 adipocytes . Following assessment of basal respiration , the following compounds were injected sequentially: oligoymcin ( 10 µg/ml ) , BAM15 ( 10 µM ) ( Kenwood et al . , 2014 ) rotenone/antimycin A ( 5 µM / 10 µM ) . All of these reagents were obtained from Sigma Aldrich , except BAM15 ( Kenwood et al . , 2014 ) . After the assay , the medium was aspirated and DNA content was measured by Hoechst staining as described previously ( Krycer et al . , 2017 ) . 3T3-L1 adipocytes were seeded into Matrigel-coated XFp cell culture plates as described above . Following the indicated treatment , cells were washed with warm PBS and incubated in mitochondrial respiration buffer ( 20 mM K-HEPES pH 7 . 1 , 0 . 5 mM EGTA , 3 mM MgCl2 , 10 mM KH2PO4 , 110 mM D-sucrose , 0 . 1% ( w/v ) fatty-acid free BSA ) for 30 min at 37°C . Cells were permeabilised with 10 µg/mL digitonin prior to equilibration in the Seahorse XFp Analyzer ( Seahorse Biosciences ) . The following injections were used: Port A: ADP-regeneration system ( 5 mM deoxyglucose , 1 U/mL hexokinase ( Roche ) , 1 mM ADP ) , 2 mM malate and 10 mM glutamate; Port B: 10 µM rotenone; Port C: 10 mM succinate or 100 µM octanoylcarnitine; Port D: 10 µM antimycin A , 5 mM ascorbic acid and 0 . 5 mM tetramethyl-phenylenediamine . Each cycle included 3 min mixing time , 0 min waiting time and 2 min measuring time , repeated until a steady JO2 reading was obtained after each injection . Non-mitochondrial respiration was taken as JO2 following rotenone addition and this was subtracted from all substrate-driven rates ( ports A , C and D ) . State four respiration rates ( proton leak ) was determined by omission of the ADP-regeneration system . All of these reagents were obtained from Sigma Aldrich , unless otherwise stated . The resulting rate was normalised to DNA content , measured using Hoechst 33248 with a previously-described Hoechst-staining protocol ( Krycer et al . , 2017 ) . Sample preparation was carried out essentially as previously described ( Bayer et al . , 2013 ) . For in vitro studies , 3T3-L1 adipocytes were pre-treated as indicated and washed three times with ice-cold PBS that had been pre-treated with 10 µg/mL catalase for 1 hr . Cells were incubated with 100 mM N-ethylmaleimide in PBS for 10 min on ice to modify free cysteine residues . Cells were scraped in PBS-containing 1% SDS , protease inhibitors and 100 mM N-ethylmaleimide . For tissue preparation , epididymal adipose tissue was immediately lysed in 2 x RIPA buffer containing protease inhibitors and 100 mM N-ethylmaleimide upon collection buffer by sonication ( 20 s ) . Samples were centrifuged at 13 , 000 × g for 10 min at room temperature . Fat was removed , supernatant collected and protein concentration determined by BCA assay ( Thermo Fisher Scientific ) . Proteins were then separated by non-reducing SDS-PAGE . Intensities of PRDX dimer and monomer bands were determined by densitometry and ratios ( dimer/monomer ) expressed relative to control samples within each experiment . Conversion of MitoSOX to Mito-2-hydroxyethidium was detected by LC-MS/MS , as previously described ( Zielonka et al . , 2008 ) with modifications . 3T3-L1 adipocytes were treated as indicated and incubated with 5 µM MitoSOX ( Thermo Scientific ) for 60 min in DMEM/GlutMAX without FCS . Cells were scraped in ethanol and centrifuged at 17 , 000x g for 15 min , 4°C . 5 μL of supernate were then injected onto an Agilent 1290 UHPLC system connected to an Agilent 6490 triple-quadrupole mass spectrometer . Analytes were separated on a 2 . 6 μm Kinetex XB-C18 100 A column ( 50 × 2 . 10 mm; Phenomenex ) by gradient elution using mobile phase A ( 0 . 1% formic acid in water ) and mobile phase B ( 0 . 1% formic acid in acetonitrile ) at 0 . 4 mL/min . The gradient consisted of 25% mobile phase B from 0 to 2 min , 25–40% B from 2 to 7 min , 40–90% B from 7 to 8 min , 90% B from 8 to 11 min and back to 25% B from 11 to 12 min . Flow was then directed into the triple quadrupole mass spectrometer with parameters set as follows: gas temperature = 290°C; gas flow = 14 L/min; nebulizer pressure = 35 psi; sheath gas heater = 350°C; sheath gas flow = 11 L/min; capillary voltage = 3 , 500 V . Detection of MitoSOX , Mito-ethidium ( Mito-E+ ) and Mito-hydroxyethidium ( Mito-2OHE+ ) was by multiple reaction monitoring ( MRM ) in positive ion mode using the above general mass spectrometry parameters with fragmentor voltage at 380 V and cell accelerator voltage at 5 V . In each case , the fragment ions generated by collision-induced dissociation of the [M + H]+or the [M + 2 hr]+ ions were used for quantification . MRM settings for the target analytes were ( parent ion → fragment ion ) ; MitoSOX ( m/z 632 . 2 → 289 ) with collision energy ( CE ) = 25 V; Mito-E+ ( m/z 315 . 7 → 289 ) with CE = 25 V and Mito-2OHE+ ( m/z 323 . 7 → 289 ) with CE = 25 V . MitoSOX was quantified against authentic commercial standard obtained from Invitrogen while Mito-E+ and Mito-2OHE+ were quantified against standards synthesized from MitoSOX as previously described ( Zielonka et al . , 2008 ) . 3T3-L1 adipocytes were treated as indicated prior to treatment with 25 µM BAM15 , 10 µM FCCP ( Sigma Aldrich ) , 10 µM rotenone ( Sigma Aldrich ) , 50 nM antimycin A ( Sigma Aldrich ) , 10 µg/mL oligomycin ( Sigma Aldrich ) , 10 or 100 µM TTFA ( Sigma Aldrich ) or 10 mM malonate ( Sigma Aldrich ) for 2 hr in DMEM/10% FCS/GlutaMAX before determination of PRDX dimerisation status . For assessment of the effect of TTFA or malonate on HA-GLUT4 translocation in 3T3-L1 adipocytes , cells were incubated with 100 µM TTFA or 10 mM malonate for 2 hr in DMEM/10% FCS/GlutaMAX and for 2 hr during serum starvation in DMEM/0 . 2% BSA/GlutaMAX prior to assessment of insulin action . Mitochondrial-targeted human catalase ( human catalase with the 24 residue MnSOD mitochondrial localisation signal at its N-terminus ) ( Arita et al . , 2006 ) was inserted into the pBABE-puro vector . Mitochondrial-targeted catalase overexpressing 3T3-L1 fibroblasts were generated by retroviral transduction as previously described ( Govers et al . , 2004 ) . 3T3-L1 fibroblasts were transduced and selected using puromycin . Fibroblasts were differentiated into adipocytes as described . Data are expressed as means ± S . E . M . For each dataset , the sub-groups were initially compared with the relevant ANOVA depending on the levels of treatments . If row or column factors were significant , specific sub-groups were then compared using Student’s t-test ( adjusting for multiple comparisons using the Sidak method ) . Significant effects were defined as p<0 . 05 by these t-tests , as reported in the figures . Human data were assessed by Mann Whitney tests and significant effects were defined as p<0 . 05 .	After we eat , our blood sugar levels increase . To counteract this , the pancreas releases a hormone called insulin . Part of insulin’s effect is to promote the uptake of sugar from the blood into muscle and fat tissue for storage . Under certain conditions , such as obesity , this process can become defective , leading to a condition known as insulin resistance . This condition makes a number of human diseases more likely to develop , including type 2 diabetes . Working out how insulin resistance develops could therefore unveil new treatment strategies for these diseases . Mitochondria – structures that produce most of a cell’s energy supply – appear to play a role in the development of insulin resistance . Mitochondria convert nutrients such as fats and sugars into molecules called ATP that fuel the many processes required for life . However , ATP production can also generate potentially harmful intermediates often referred to as ‘reactive oxygen species’ or ‘oxidants’ . Previous studies have suggested that an increase in the amount of oxidants produced in mitochondria can cause insulin resistance . Fazakerley et al . therefore set out to identify the reason for increased oxidants in mitochondria , and did so by analysing the levels of proteins and oxidants found in cells grown in the laboratory , and mouse and human tissue samples . This led them to find that concentrations of a molecule called coenzyme Q ( CoQ ) , an essential component of mitochondria that helps to produce ATP , were lower in mitochondria from insulin-resistant fat and muscle tissue . Further experiments suggested a link between the lower levels of CoQ and the higher levels of oxidants in the mitochondria . Replenishing the mitochondria of the lab-grown cells and insulin-resistant mice with CoQ restored ‘normal’ oxidant levels and prevented the development of insulin resistance . Strategies that aim to increase mitochondria CoQ levels may therefore prevent or reverse insulin resistance . Although CoQ supplements are readily available , swallowing CoQ does not efficiently deliver CoQ to mitochondria in humans , so alternative treatment methods must be found . It is also of interest that statins , common drugs taken by millions of people around the world to lower cholesterol , also lower CoQ and have been reported to increase the risk of developing type 2 diabetes . Further research is therefore needed to investigate whether CoQ might provide the link between statins and type 2 diabetes .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2018	Mitochondrial CoQ deficiency is a common driver of mitochondrial oxidants and insulin resistance
G protein gated inward rectifier potassium ( GIRK ) channels are gated by direct binding of G protein beta-gamma subunits ( Gβγ ) , signaling lipids , and intracellular Na+ . In cardiac pacemaker cells , hetero-tetramer GIRK1/4 channels and homo-tetramer GIRK4 channels play a central role in parasympathetic slowing of heart rate . It is known that the Na+ binding site of the GIRK1 subunit is defective , but the functional difference between GIRK1/4 hetero-tetramers and GIRK4 homo-tetramers remains unclear . Here , using purified proteins and the lipid bilayer system , we characterize Gβγ and Na+ regulation of GIRK1/4 hetero-tetramers and GIRK4 homo-tetramers . We find in GIRK4 homo-tetramers that Na+ binding increases Gβγ affinity and thereby increases the GIRK4 responsiveness to G protein stimulation . GIRK1/4 hetero-tetramers are not activated by Na+ , but rather are in a permanent state of high responsiveness to Gβγ , suggesting that the GIRK1 subunit functions like a GIRK4 subunit with Na+ permanently bound . In the cardiovascular system , cardiac GIRK channels play a central role in parasympathetic slowing of the heart . Specifically , when the body is at rest , parasympathetic neurons convey signals from the central nervous system to cardiac pacemaker cells via cholinergic neurotransmission , activating the muscarinic acetylcholine receptor 2 ( M2R ) . Activated M2Rs release inhibitory G protein alpha subunits and Gβγ . Gβγ is a hetero-dimeric protein composed of tightly bound beta and gamma subunits . This free Gβγ , along with its lipid anchor , diffuses on the intracellular membrane surface and binds directly to GIRK to activate it ( Logothetis et al . , 1987; Whorton and MacKinnon , 2013; Sakmann et al . , 1983; Kurachi et al . , 1986 ) . Activation of GIRK shifts the resting membrane potential of pacemaker cells toward the equilibrium potential for K+ , lengthening the interval between cardiac action potentials and thereby slowing the heart ( Loewi and Navratil , 1926; Rayner and Weatherall M , 1959 ) . The critical role of parasympathetic regulation of cardiac GIRK channels is evident from the severe diseases that result from mutations in the GIRK gene such as Atrial Fibrillation ( Kovoor et al . , 2001; Voigt et al . , 2010 ) , and Long QT syndrome ( Yang et al . , 2010 ) . Mammals express four GIRK channel subunits ( GIRK1-4 ) , forming various homo-tetramers and hetero-tetramers . Cardiac GIRK channels are composed of GIRK1 and GIRK4 subunits ( Krapavinsky et al . , 1995 ) . Since the GIRK1 subunit does not form functional homo-tetramers , GIRK1 and GIRK4 subunits form functional GIRK1/4 hetero-tetramers and GIRK4 homo-tetramers in the heart ( Krapavinsky et al . , 1995; Chan et al . , 1996; Corey and Clapham , 1998 ) . GIRK1 and GIRK4 knockout mice show similar phenotypes in terms of heart rate ( Bettahi et al . , 2002 ) , suggesting that both subunits perform non-redundant tasks . However , little is known about whether or how GIRK1 influences cardiac GIRK channel behavior . Specifically , what are the functional differences between GIRK1/4 hetero-tetramers and GIRK4 homo-tetramers ? Although GIRK1 and GIRK4 subunits share ~44% sequence identity , one notable difference occurs in the Na+ binding site . The GIRK1 subunit has an aspartate to asparagine replacement in this Na+ binding site , presumably rendering it incapable of binding intracellular Na+ ( Ho and Murrell-Lagnado , 1999 ) . However , it is still unclear what influence this defective Na+ binding site has on the function of GIRK1/4 hetero-tetramers . Cellular electrophysiological experiments have not clarified this issue because it is difficult to control the concentration of GIRK ligands inside cells and it is also not possible to express GIRK1/4 hetero-tetramers without co-expression of GIRK4 homo-tetramers . To overcome these difficulties we have purified human GIRK1/4 hetero-tetramers and GIRK4 homo-tetramers and studied their ligand regulation by Na+ and Gβγ in the planar lipid bilayer system . Although the GIRK1 subunit does not form functional homo-tetrameric channels , it does form structural homo-tetramers similar to GIRK4 ( Figure 1 ) . Therefore , in order to isolate GIRK1/4 hetero-tetramers , GIRK1 and GIRK4 homo-tetramers had to be removed during purification . To remove both homo-tetramers two different tags , a deca-histidine tag and a 1D4 peptide tag , were fused to the GIRK1 and GIRK4 subunits , respectively . Two sequential affinity chromatography steps isolated only GIRK1/4 hetero-tetramer channels containing both tags ( Figure 2A ) . Equal bands in all lanes of an SDS-PAGE gel , corresponding to different elution fractions from a gel-filtration column , suggested that the predominant channel species purified contained two GIRK1 and two GIRK4 subunits ( Figure 2B ) . This suggestion is based on the different elution times of homo-tetramer GIRK1 and GIRK4 subunits ( Figure 1A ) . We cannot , however , exclude with certainty the possibility that some channels with 3:1 and/or 1:3 stoichiometry were present in the population of isolated channels . 10 . 7554/eLife . 15750 . 003Figure 1 . The GIRK4 subunit forms functional homo-tetrameric channels , whereas the GIRK1 subunit forms nonfunctional homo-tetramers . ( A ) HEK293T cells were transiently transfected with the GIRK1 or the GIRK4 subunit fused to GFP , and solubilized cell lysate was analyzed by fluorescent size-exclusion chromatography ( Superose 6 10/300 GL ) . Blue and red elution profiles show GIRK1 homo-tetramers and GIRK4 homo-tetramers , respectively . ( B ) HEK293T cells were transiently transfected with GIRK1 ( blue ) or GIRK4 ( red ) , and human M2Rs . Whole-cell voltage clamp recordings were performed . Membrane potential was held at -80 mV , and the extracellular solution was exchanged to high potassium buffer ( 100 mM KCl ) as indicated above the signal , followed by the application of 10 µM acetylcholine . DOI: http://dx . doi . org/10 . 7554/eLife . 15750 . 00310 . 7554/eLife . 15750 . 004Figure 2 . Purified cardiac GIRK channels are functional in reconstituted planar lipid bilayer membranes . ( A ) Schematic of GIRK1/4 hetero-tetramer purification procedure . 1D4-tagged GIRK4 homo-tetramers were removed with Co2+ affinity chromatography and His-tagged GIRK1 homo-tetramers were removed with subsequent 1D4 affinity chromatography . ( B ) Gel-filtration fractions of the GIRK1/4 hetero-tetramer peak were run on 12% SDS-PAGE . GIRK1 and GIRK4 monomers are 56 kDa and 46 kDa , respectively . ( C ) and ( D ) The top and bottom chambers are separated by the lipid bilayer formed on a transparency film . The same solution containing 10 mM potassium phosphate buffer pH 7 . 4 and 150 mM KCl was used in both chambers . Proteoliposomes containing GIRK channels were fused to the bilayer membrane . 32 µM C8-PIP2 and 2 mM MgCl2 were added to the intracellular side of the chamber , and proteoliposomes containing Gβγ were fused to the membrane , activating GIRK channels . ( C ) GIRK1/4 hetero-tetramer currents recorded in the lipid bilayer . Membrane potential was held at 0 mV , and 10 mV voltage steps from -80 mV to 80 mV were applied . ( D ) GIRK4 homo-tetramer currents recorded in the lipid bilayer . DOI: http://dx . doi . org/10 . 7554/eLife . 15750 . 004 Purified GIRK channels were reconstituted into liposomes and fused with planar lipid bilayer membranes . The channels were activated by fusing lipid-anchored Gβγ-containing vesicles with the membranes and adding the membrane-impermeable , short-chain PIP2 ( C8-PIP2 ) to one chamber of the planar bilayer . Although channels and Gβγ insert into the bilayer membrane randomly in both orientations , only channels with their intracellular surface facing the chamber to which PIP2 was added are activated ( Wang et al . , 2014 ) . The strong inward-rectification of current as a function of membrane voltage supports the uniform orientation of active channels ( Figure 2C ) . In contrast to GIRK1 homo-tetramers , GIRK4 homo-tetramers form functional channels that are activated by GPCR stimulation when expressed in HEK293T cells ( Figure 1B ) . To nullify any residual uncertainty that GIRK4 may actually form functional channels in cells by combining with native GIRK1 subunits that may be present , we purified and reconstituted GIRK4 homo-tetramers and found they produce robust inward-rectifier K+ currents in planar lipid membranes ( Figure 2D ) . To study the dependence of GIRK channel activity on Na+ and Gβγ concentrations , we used lipids with Ni-NTA modified head groups ( Ni-NTA-lipids ) as illustrated ( Figure 3A ) using a method described in the accompanying paper ( Wang et al . , 2016 ) . In this method , bilayer membranes containing specific mole fractions of Ni-NTA-lipids were formed . GIRK channel proteoliposomes , which also contained the same mole fraction of Ni-NTA-lipids as the bilayer membrane , were then fused to the membrane . C8-PIP2 and 2 µM soluble Gβγ ( sGβγ-His10 ) , which contained a deca-histidine-tag instead of its physiological lipid anchor , were applied to the intracellular side of the membrane . At 2 µM concentration sGβγ-His10 does not activate GIRK channels directly from solution , however , it saturates ( i . e . occupies nearly 100% of ) all available Ni-NTA-lipids in the membrane ( Wang et al . , 2016 ) . These membrane-bound sGβγ-His10 molecules are able to activate GIRK channels , which are present in the membrane at a much lower density than Ni-NTA-lipid molecules ( Figure 3 ) . This method permits the study of GIRK channel activation as a function of the membrane sGβγ-His10 density ( Gβγ concentration ) , which is controlled through the predetermined mole fraction of Ni-NTA-lipid molecules in the membrane ( Wang et al . , 2016 ) . In subsequent graphs , Gβγ concentration is quantified as Ni-NTA-lipid mole fraction , but for accounting purposes the stoichiometry of sGβγ-His10 to Ni-NTA-lipid is 1:3 ( i . e . a single sGβγ-His10 molecule binds to 3 Ni-NTA-lipid molecules ) , meaning the actual sGβγ-His10 density on the membrane is one third the density of Ni-NTA-lipid ( Wang et al . , 2016 ) . In order to compare currents from different membranes that generally contain different numbers of GIRK channels , at the end of each experiment proteolipsomes containing lipid-anchored Gβγ were fused to the membrane to maximally activate the GIRK channels in the membrane ( Figure 3B ) . Current activated at a specific Gβγ concentration ( determined by the density of Ni-NTA-lipids ) is referred to as normalized current . 10 . 7554/eLife . 15750 . 005Figure 3 . Schematic of the Na+ and Gβγ titration using Ni-NTA-lipids . ( A ) GIRK channels were incorporated into the membrane containing a known concentration of Ni-NTA-lipids . 32 µM C8-PIP2 and 2 µM sGβγ-His10 were added to the intracellular side of the membrane . Free sGβγ-His10 does not activate GIRK channels at the concentration applied , while Ni-NTA-lipids-bound sGβγ-His10 mimics lipid-anchored Gβγ and activates GIRK channels . Known concentrations of Na+ were subsequently added to study the effect of Na+ concentration on GIRK channel activity in the presence of known concentrations of Gβγ in the membrane . ( B ) Left and center traces show normalized GIRK4 currents before and after application of 2 µM sGβγ-His10 in the presence of 0 ( black ) or 0 . 002 ( red ) mole fraction of Ni-NTA-lipids in the membrane . At the end of each experiment , currents were fully activated by lipid-anchored Gβγ ( right signals ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15750 . 005 Figure 4A shows normalized GIRK4 current as a function of Gβγ concentration at 0 mM , 8 mM , and 32 mM Na+ ( Figure 4A ) . GIRK4 current increases as a sigmoid-shaped function , and Na+ concentration has a prominent effect on Gβγ activation . Specifically , Na+ increases GIRK4 current at all Gβγ concentrations , but notably , the increase is relatively largest at low Gβγ concentrations where , for example , at 0 . 001 Ni-NTA mole fraction 32 mM Na+ increases normalized current almost 20-fold , from 0 . 018 to 0 . 34 . These data suggest that GIRK4 is similar to the neuronal GIRK channel , GIRK2 , in its response to Gβγ and Na+ ( Wang et al . , 2016 ) . We therefore applied the same equilibrium gating model used to analyze GIRK2 ( Wang et al . , 2016 ) . The model has 25 states of ligand occupancy , corresponding to 0 to 4 of each ligand , Gβγ and Na+ , as illustrated ( Figure 4D ) . Parameters in the model include an equilibrium dissociation constant Kdb and cooperativity factor b for Gβγ binding , an equilibrium dissociation constant Kdn for Na+ binding ( the cooperativity factor for Na+ binding is 1 ) , a factor η for the effect that Gβγ and Na+ have on each other’s affinity , and a term θ relating conductivity to ligand occupancy ( Table 1 ) . The model adequately represents the data with values for the parameters given ( Table 1 ) . The errors on values for equilibrium dissociation constants and cooperativity factors are larger than those determined for GIRK2 ( Wang et al . , 2016 ) because the data set on GIRK4 is smaller . However , the overall conclusion is that GIRK4 is very similar to GIRK2 . Through model analysis the data support three conclusions: that 4 Gβγ molecules are required to open the channel ( the model yields higher residuals with less than 4 ) , that Gβγ binds cooperatively to GIRK4 , and that Na+ exerts its major effect by increasing the Gβγ affinity for the channel . 10 . 7554/eLife . 15750 . 006Figure 4 . GIRK channel activity as a function of Na+ and Gβγ concentrations . ( A ) , ( B ) , and ( C ) Plots of activity of GIRK4 homo-tetramers ( A ) , GIRK1/4 hetero-tetramers ( B ) , and GIRK1 ( N217D ) /4 hetero-tetramers ( C ) versus Ni-NTA-lipid mole fraction in the membrane at different Na+ concentrations . The same buffer ( 10 mM potassium phosphate pH 8 . 2 , 150 mM KCl ) was used in both chambers , and voltage across the lipid bilayer was held at -50 mV . GIRK proteoliposomes were fused to the bilayer membrane containing a known concentration of Ni-NTA-lipids . 2 mM MgCl2 and 32 µM C8-PIP2 were added to one side of the bilayer chamber and then 2 µM sGβγ-His10 was added to the same side of the chamber to activate GIRK channels . 8 mM and 32 mM Na+ were added to further activate GIRK channels . At the end of each experiment , channels were fully activated by fusing proteoliposomes containing lipid-anchored Gβγ and currents were normalized to the fully activated current ( mean ± SEM , n = 3 bilayers ) . The equilibrium model ( D ) was used to fit the data ( solid curves ) . Kdb: Equilibrium dissociation constant between Gβγ and ligand-free GIRK . Kdn: Equilibrium dissociation constant between Na+ and ligand-free GIRK ( mM ) . b: Cooperativity factor for Gβγ binding . η: Cross-cooperativity factor between Gβγ and Na+ binding . i: The number of Na+ ions bound to GIRK . For GIRK1/4 hetero-tetramers , the range of i was restricted to the range 2 to 4 . j: The number of Gβγ bound to GIRK . DOI: http://dx . doi . org/10 . 7554/eLife . 15750 . 00610 . 7554/eLife . 15750 . 007Table 1 . The fitting parameters for the Na+ and Gβγ titration . Kdb: Equilibrium dissociation constant for Gβγ in equilibrium with ligand-free GIRK . Kdn: Equilibrium dissociation constant for Na+ binding to ligand-free GIRK ( mM ) . b: Cooperativity factor for Gβγ binding . η: Cross-cooperativity factor between Gβγ and Na+ binding . θi , j: Normalized activity of i-Na+ and j-Gβγ-bound GIRK . R2: Adjusted R-squared . For fitting to GIRK1 ( N217D ) /4 hetero-tetramers , b and θ4 , 4 were fixed to the same parameters as GIRK1/4 hetero-tetramers . DOI: http://dx . doi . org/10 . 7554/eLife . 15750 . 007GIRK4GIRK1/4GIRK1 ( N217D ) /4Kdb0 . 004 ± 0 . 0050 . 004 ± 0 . 0060 . 0024 ± 0 . 0004kKdn ( mM ) 50 ± 4050 ± 30050 ± 30b0 . 6 ± 0 . 30 . 6 ± 0 . 30 . 6η0 . 7 ± 0 . 10 . 8 ± 0 . 40 . 71 ± 0 . 08θ0 , 40 . 6 ± 0 . 1-0 . 45 ± 0 . 04θ1 , 4θ0 , 4 + ( θ4 , 4- θ0 , 4 ) × 1/4-θ0 , 4 + ( θ4 , 4- θ0 , 4 ) × 1/4θ2 , 4θ0 , 4 + ( θ4 , 4- θ0 , 4 ) × 2/41 . 1 ± 0 . 1θ0 , 4 + ( θ4 , 4- θ0 , 4 ) × 2/4θ3 , 4θ0 , 4 + ( θ4 , 4- θ0 , 4 ) × 3/41 . 2 ± 0 . 8θ0 , 4 + ( θ4 , 4- θ0 , 4 ) × 3/4θ4 , 41 . 2 ± 0 . 11 . 1 ± 0 . 81 . 1R20 . 960 . 930 . 97 Figure 4B shows corresponding data for the GIRK1/4 channel . At all Na+ concentrations – even in the absence of Na+ – the response of the GIRK1/4 channel to Gβγ is similar to the GIRK4 channel at higher Na+ concentrations . Thus , the GIRK1/4 hetero-tetramer channel , compared to the GIRK4 homo-tetramer channel , behaves to a first approximation as if it remains permanently stuck in a Na+-activated state . That this influence of the GIRK1 subunit is related to its Na+ binding site is supported by the mutation N217D , which converts the GIRK1 Na+ binding locus to be more like that of GIRK4 by restoring its Na+ sensitivity to the hetero-tetramer ( Figure 4C ) ( Ho and Murrell-Lagnado , 1999 ) . To test the idea that Asn217 in GIRK1 mimics a Na+-bound Asp we fit the GIRK1/4 data to the same model used for the GIRK4 channel , but imposed the condition that two of the four sites are “permanently occupied” by Na+ , with the underlying idea that the two permanently occupied sites represent the GIRK1 subunits . This condition means GIRK1/4 is described by 15 states of ligand occupancy corresponding to 0 to 4 Gβγ and 0 to 2 Na+ . The model encodes this by collapsing the 0 , 1 and 2 Na+-occupied states of the GIRK4 model into a single state with affinity of Gβγ equal to Kdbη2 ( Table 1 ) . This model describes the data for the GIRK1/4 channel accurately with numerical values for Kdb , Kdn , b and η that are indistinguishable from those for the GIRK4 model ( Table 1 ) . Thus , the properties of the GIRK1/4 channel are consistent quantitatively with the GIRK1 subunits functioning as if they are GIRK4 subunits with Na+ ions permanently bound to them . In Figure 5 we ask how does intracellular Na+ affect GPCR-stimulated GIRK currents in mouse embryonic stem cell ( mESC ) -derived cardiac pacemaker cells . Whole-cell voltage clamp recordings show acetylcholine-activated K+ currents that are inhibited by tertiapin-Q ( TPNQ ) , a specific GIRK channel blocker ( Figure 5A ) . Such recordings were performed with 22 different cells with intracellular solutions containing either 0 mM or 30 mM Na+ . Pacemaker cells showed an average of 32 ± 4 pA of acetylcholine-activated K+ current in 0 mM Na+ and 47 ± 6 pA in 30 mM Na+ ( Figure 5B ) . We thus conclude that intracellular Na+ has essentially no influence on GPCR-stimulated GIRK current in these mESC-derived cardiac cells . This observation is consistent with the data recorded in bilayers if the cardiac cells express predominantly GIRK1/4 hetero-tetramer channels , which are only weakly Na+ sensitive , and not GIRK4 homo-tetramer channels , which are strongly Na+ sensitive ( Figure 4 ) . 10 . 7554/eLife . 15750 . 008Figure 5 . Intracellular Na+ does not significantly activate cardiac GIRK channels . ( A ) Whole-cell voltage clamp recordings on mESC-derived pacemaker cells . Membrane potential was held at -80 mV and the extracellular solution was exchanged to high potassium buffer ( 25 . 4 mM KCl ) as indicated above the signal . 10 µM acetylcholine ( Ach ) was then applied to activate GIRK channels and 100 nM tertiapin Q ( TPNQ ) was next applied to block cardiac GIRK currents . Acetylcholine-activated GIRK currents were measured by subtracting signals before and after acetylcholine application . ( B ) Acetylcholine induced GIRK currents were measured with the pipette solution containing 0 mM Na+ or 30 mM Na+ . Eleven recordings were performed with each pipette and average value was calculated ( mean ± SEM , n = 11 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 15750 . 008 In cardiac cells two different subunits , GIRK1 and GIRK4 , form G protein gated K+ channels . Homo-tetramers of GIRK4 and hetero-tetramers of GIRK1 and GIRK4 , GIRK1/4 , form functional K+ channels , while homo-tetramers of GIRK1 do not ( Krapavinsky et al . , 1995; Hedin et al . , 1996 ) . It is unclear to what extent GIRK4 homo-tetramers versus GIRK1/4 hetero-tetramers dominate in cardiac cells . It is also unclear to what extent the functional properties of these channels differ because it has not been possible to study GIRK1/4 channels in isolation , the reason being heterologous expression of both subunits naturally gives rise to a mixed population of homo- and hetero-tetramers . To overcome this problem we overexpressed and purified GIRK1/4 hetero-tetramers using sequential affinity columns and also expressed and purified GIRK4 homo-tetramers for comparative analysis . The composition of GIRK1/4 hetero-tetramers is reported to consist mainly of two GIRK1 and two GIRK4 subunits ( Silverman et al . , 1996; Corey et al . , 1998 ) . In this study purified GIRK1/4 hetero-tetramers are also most likely composed of two GIRK1 subunits and two GIRK4 subunits , as estimated from SDS-PAGE of fractions from a gel filtration column ( Figure 2B ) . However , we have no information on the arrangement of subunits within the tetramer either in cells or in our reconstitution experiments . We observe that GIRK4 homo-tetramers and GIRK1/4 hetero-tetramers exhibit distinctly different properties with respect to their activation by Gβγ and Na+ . It had been shown that the GIRK1 subunit has a defective Na+ site ( Ho and Murrell-Lagnado , 1999 ) , but the present study establishes the following new conclusions . First , that Na+ binding to the GIRK4 subunit increases affinity for Gβγ This effect is encoded in the model by the Gβγ-Na+ cross interaction term η . Second , the GIRK1 subunit behaves similarly to the GIRK4 subunit with Na+ permanently bound . Thus , while the GIRK1 subunit is unable to bind Na+ , it causes the channel to have high affinity for Gβγ even in the absence of Na+ . This effect is encoded in the model by enforcing permanent Na+ occupancy on the GIRK1 subunits . Taken together , these properties account for the functional differences we observe between GIRK4 and GIRK1/4 channels . GIRK4 channels are less sensitive to G protein stimulation at low Na+ concentrations ( Gβγ binds with lower affinity ) and more sensitive at high Na+ concentrations ( Gβγ binds with higher affinity ) . GIRK1/4 channels on the other hand are very sensitive to Gβγ stimulation at both low and high Na+ concentrations ( Gβγ binds with high affinity independent of Na+ concentration ) . We also find that GPCR-stimulated GIRK currents in mESC-derived cardiac pacemaker cells are nearly independent of intracellular Na+ concentration . Based on a comparison of these cellular data to the properties of isolated GIRK4 and GIRK1/4 channels in planar lipid bilayers , we conclude that GIRK channels in mESC-derived cardiac channels most likely are predominantly GIRK1/4 hetero-tetramers . In an accompanying paper we report that GIRK channels in mouse dopamine neurons are very sensitive to intracellular Na+: in experiments analogous to those in Figure 5B , eight fold amplification of GPCR-stimulated GIRK currents was observed ( Wang et al . , 2016 ) . This degree of Na+ sensitivity is consistent with neurons expressing GIRK2 homo-tetramers . GIRK2 , like GIRK4 , encodes an intact Na+ activation site . Our findings lead us to conclude that the GIRK1 subunit in a GIRK1/4 hetero-tetramer renders the channel relatively insensitive to Na+ but permanently in a state of high responsiveness to GPCR stimulation . We can only speculate as to why two kinds of GIRK channels exist , ones whose G protein sensitivity is regulated by intracellular Na+ ( i . e . homo-tetramer GIRK4 or GIRK2 channels ) and ones whose G protein sensitivity is not much regulated by Na+ but is always near maximum ( i . e . hetero-tetramer GIRK1/4 channels ) . In neurons , intracellular Na+ concentration increases during excitation because Na+ enters the cell through Na+ channels during action potentials and through glutamate receptor ion channels in response to excitatory neurotransmitters ( Lasser-Ross and Ross , 1992 ) . GIRK2 channels silence neurons in response to inhibitory neurotransmitters , which act through inhibitory GPCRs . The GIRK2 regulation by Na+ provides a way to modulate the inhibitory response according to the activity level . Such modulation would seem beneficial to a neuron that exhibits a wide range of electrical activity from near silent to high frequency spiking . Cardiac cells on the other hand appear to exhibit less activity-dependent variation in levels of intracellular Na+ ( Harrison et al . , 1992 ) . Thus , it seems reasonable that GIRK1/4 channels do not exhibit high Na+ sensitivity , but instead exhibit a permanent state of cholinergic responsiveness ( Ito et al . , 1994 ) . Human full-length GIRK1 and GIRK4 were cloned into pEG BacMam ( Goehring et al . , 2014 ) . At the C-terminus of the GIRK1 construct , a PreScission protease cleavage site , an enhanced green fluorescent protein ( eGFP ) , and a deca-histidine tag were attached for purification . A 1D4 peptide tag was used instead of a deca-histidine tag for the GIRK4 construct . These constructs were used for fluorescent size exclusion chromatography ( FSEC ) , and overexpression and protein purification . For FSEC , HEK293T cells were transiently transfected with GIRK1-His10-pEG BacMam or GIRK4-1D4-pEG BacMam , and incubated at 37°C for 48 hr . Cells were solubilized in 50 mM HEPES-KOH ( pH 7 . 35 ) , 150 mM KCl , 4% ( w/v ) n-decyl-η-D-maltopyranoside ( DM ) , and a protease inhibitor cocktail ( 1 mM PMSF , 0 . 1 mg/mL trypsin inhibitor , 1 µg/mL pepstatin , 1 µg/mL leupeptin , and 1 mM benzamidine ) . Lysed cells were centrifuged and supernatant was run on FSEC ( Superose 6 10/300 GL ) . For overexpression and protein purification , HEK293S GnTl- cells were grown in suspension , transduced with P3 BacMam virus of the GIRK1-His and the GIRK4-1D4 in 1:1 ratio , and incubated at 37°C 8-12 hr post-transduction , 10 mM sodium butyrate was added to the culture and cells were harvested 60 hr post-transduction . Cells were harvested by centrifugation , frozen in liquid N2 , and stored at -80°C until needed . Frozen cells were solubilized in 50 mM HEPES-KOH ( pH 7 . 35 ) , 150 mM KCl , 4% ( w/v ) DM , and protease inhibitor cocktail . 2 hr after solubilization , lysed cells were centrifuged and supernatant was incubated with the Talon metal affinity resin ( Clontech Laboratories , Inc . Mountain View , CA ) for 1 hr at 4°C with gentle mixing . The resin was washed in batch with 5 column volume ( cv ) of buffer A ( 50 mM HEPES-KOH [pH 7 . 0] , 150 mM KCl , 0 . 4% [w/v] DM ) , then loaded onto a column and further washed with 5 cv buffer A + 10 mM imidazole . The column was then eluted with buffer A + 200 mM imidazole . The peak fraction was collected and incubated with the 1D4 affinity resin for 1 hr at 4°C with gentle mixing . The resin was loaded onto a column and washed with buffer A . 5 mM DTT and 1 mM EDTA were added and eGFP and affinity tags were cut with PreScission protease overnight at 4°C . The cleaved protein was then concentrated to run on a Superose 6 10/300 GL gel filtration column in 20 mM Tris-HCl ( pH 7 . 5 ) , 150 mM KCl , 0 . 2% ( w/v ) DM , 20 mM DTT , and 1 mM EDTA . GIRK1 ( N217D ) /GIRK4 hetero-tetramers were purified using the same procedure . GIRK4 homo-tetramers were purified with a similar procedure . In brief , GIRK4 homo-tetramers were expressed in HEK293S GnTl- cells and purified using the 1D4 affinity chromatography and size-exclusion chromatography . Human lipid-anchored Gβγ and soluble deca-histidine tagged Gβγ were purified as described ( Wang et al . , 2016 ) . All lipids were purchased from Avanti Polar Lipids ( Alabaster , AL ) . Proteoliposomes were reconstituted as described ( Wang et al . , 2016 ) . In brief , 20 mg/mL of the lipid mixture ( 3:1 [wt:wt] = 1-palmitoyl-2-oleyl-sn-glycero-3-phosphoethanolamine [POPE]: 1-palmitoyl-2-oleyl-sn-glycero-3-phospho-[1’-rac-glycerol] [POPG] ) was dispersed by sonication and solubilized with 20 mM DM . In the Na+ and Gβγ titration experiment , 0–0 . 015 ( mole fraction ) of 1 , 2-dioleoyl-sn-glycero-3-[ ( N- ( 5-amino-1-carboxypentyl ) iminodiacetic acid ) succinyl] ( nickel salt ) ( DOGS-NTA ) were further added to the lipid mixture . Purified GIRK channels were combined with the lipid mixture in a GIRK: lipid ( wt:wt ) ratio of 1:10 . Protein-lipid mixtures were then diluted into the reconstitution buffer ( 10 mM potassium phosphate [pH 7 . 4] , 150 mM KCl , 1 mM EDTA , and 3 mM DTT ) to 1 mg/mL ( GIRK ) and 10 mg/mL ( lipid mixture ) . Detergent was removed by dialysis against the reconstitution buffer at 4°C for 4 days . Gβγ proteoliposomes were prepared as described ( Wang et al . , 2014 ) . Bilayer experiments were performed as described ( Wang et al . , 2016 ) . In brief , 20 mg/mL of a lipid solution in decane composed of 2:1:1 ( wt:wt:wt ) of 1 , 2-dioleoyl-sn-glycero-3-phosphoetanolamine ( DOPE ) , 1-palmitoyl-2-oleyl-sn-glycero-3-phosphocholine ( POPC ) , and 1-palmitoyl-2-oleoyl-sn-glycero-3-phospho-L-serine ( POPS ) was painted over a ~120 µm hole on a piece of transparency film . For the Na+ and Gβγ titration experiments , 0–0 . 015 ( mole fraction ) of DGS-NTA was added to a lipid solution in decane composed of 1:1 ( wt:wt ) of DOPE and POPC , and the lipid mixture was painted over a transparency film . The same buffer ( 10 mM potassium phosphate pH 7 . 4 or pH 8 . 2 for the Na+ and Gβγ titration experiment , 150 mM KCl ) was used in both chambers . Voltage across the lipid bilayer was clamped with an Axopatch 200B amplifier ( Molecular Devices , Sunnyvale , CA ) in whole-cell mode . The analog current signal was low-pass filtered at 1 kHz ( Bessel ) and digitized at 20 kHz with Digidata 1322A or Digidata 1440A digitizer ( Molecular Devices ) . Digitized data were recorded on a computer using the software pClamp ( Molecular Devices ) . Measurements were replicated on three membranes , and average and SEM values were calculated for each data point . Human M2R was cloned into the pIRES-mCherry vector for mammalian cell expression . HEK293T cells were transiently transfected with GIRK1-His10-pEG BacMam or GIRK4-1D4-pEG BacMam , and M2R-pIRES-mCherry were incubated at 37°C for 24–36 hr . Cells were dissociated and plated on PDL-coated glass coverslips for electrophysiological recordings . Whole-cell voltage clamp recordings were performed with an Axopatch 200 B amplifier in whole-cell mode . The analog current signal was low-pass filtered at 1 kHz ( Bessel ) and digitized at 20 kHz with a Digidata 1440 A digitizer . Digitized data were recorded on a computer using the software pClamp . Patch electrodes ( resistance 2 . 0–4 . 0 MΩ ) were pulled on Sutter P-97 puller ( Sutter Instrument Company , Novato , CA ) from 1 . 5 mm outer diameter filamented borosilicate glass . Membrane potential was held at -80 mV throughout the experiments , and the extracellular solution was exchanged with local perfusion with a 100 µM diameter perfusion pencil positioned beside the cell . The low potassium extracellular solution contained ( in mM ) : 150 NaCl , 3 KCl , 2 . 5 CaCl2 , 1 MgCl2 , 10 D-glucose , 10 HEPES-NaOH ( pH 7 . 4 ) ( ~320 mOsm ) . The high potassium extracellular solution contained ( in mM ) : 53 NaCl , 100 KCl , 2 . 5 CaCl2 , 1 MgCl2 , 10 D-glucose , 10 HEPES-NaOH ( pH7 . 4 ) ( ~311 mOsm ) and 10 µM acetylcholine was added . The pipette solution contained ( in mM ) : 150 KCl , 2 MgCl2 , 5 EGTA-K , 10 HEPES-KOH ( pH7 . 4 ) ( ~310 mOsm ) . W4 ( 129sv ) ES cell line was cultured in 2i/LIF medium ( Auerbach et al . , 2000; Ying et al . , 2008 ) . All ES culture reagents were purchased from Thermo Fisher Scientific ( Waltham , MA ) except for 2i and LIF ( EMD Millipore , Billerica , MA ) . ESCs were differentiated into spontaneously beating cardiomyocytes with the hanging drop method ( Maltsev et al . , 1993 ) . Embryoid bodies ( EBs ) were formed in hanging drops of ~20 µL from ~1000 cells in differentiation medium ( GMEM , 10% ES-FBS , 2 mM L-glutamine , 1 mM sodium pyruvate , 1x non-essential amino acids , 0 . 1 mM 2-mercaptoethanol ) and were cultivated in hanging drops for 5 days . Single EBs were transferred into gelatin-coated 48-well plates , and observed daily . Spontaneously contracting EBs were observed around day 8 . Contracting regions of day 16–18 EBs were dissected with micro knives , and collected into the solution containing ( in mM ) : 120 NaCl , 5 . 4 KCl , 5 MgSO4 , 20 Glucose , 10 HEPES-NaOH ( pH 6 . 9 ) , 20 Taurine . Collected cells were digested with 50 µM CaCl2 + 1 mg/mL type-II collagenase ( Sigma-Aldrich , St . Louis , MO ) for 30 min , and plated on 12 mm PDL-coated glass coverslips . Electrophysiological recordings were performed 24–48 hr after the dissociation . On average approximately three beating cells were identified per coverslip . Whole-cell voltage clamp recordings were performed with the same setup , pipettes , and perfusion system as described above . After the whole-cell configuration was formed , membrane potential was held at -80 mV in low potassium extracellular solution for about 3 min to equilibrate the intracellular solution with the pipette solution . The low potassium extracellular solution contained ( in mM ) : 140 NaCl , 5 . 4 KCl , 2 CaCl2 , 1 MgCl2 , 10 D-glucose , 10 HEPES-NaOH ( pH 7 . 4 ) ( ~300 mOsm ) . The high potassium extracellular solution contained ( in mM ) : 120 NaCl , 25 . 4 KCl , 2 CaCl2 , 1 MgCl2 , 10 D-glucose , 10 HEPES-NaOH ( pH7 . 4 ) and 10 µM acetylcholine and 100 nM TPNQ were added ( ~300 mOsm ) . 0 mM Na+ pipette solution contained ( in mM ) : 100 K-PO4 , 30 NMDG-Cl , 10 EGTA-K , 2 MgCl2 , 10 HEPES-KOH ( pH7 . 0 ) ( ~315 mOsm ) . 30 mM Na+ pipette solution contained ( in mM ) : 100 K-PO4 , 30 NaCl , 10 EGTA-K , 2 MgCl2 , 10 HEPES-KOH ( pH7 . 0 ) ( ~315 mOsm ) . 0 . 25 mM Na-GTP and 3 mM Mg-ATP were supplemented to pipette solutions just before the experiments .	Signals from outside of a cell can alter the activity inside the cell . This process often involves members of a large family of proteins called G protein-coupled receptors ( GPCRs ) that are found on the surface of many cells in the body . When these receptors are activated they release a G protein on the inside of the cell that then splits into two parts . One of these parts – called the Gβγ subunit – can directly bind to , and open , a protein called a GIRK channel that is found in the cell’s membrane . Once opened , these channels allow potassium ions to flow into the cell . GIRK channels are involved in a number of processes in the body . For example , when we are at rest , our brain sends nerve impulses to the heart and , via signals through GPCRs and Gβγ subunits , causes the GIRK channels to open . The flow of potassium into the heart muscle cells then helps to slow the heart rate . Heart cells produce two subtypes of GIRK channels , called GIRK4 and GIRK1/4 . However , it was not known whether these two channels play similar or distinct roles . Now , Touhara et al . report that GIRK4 and GIRK1/4 channels are distinct . In particular , the way that GIRK4 channels respond to the stimulation from the nervous system can be tuned by the concentration of sodium ions inside the heart cell . When there are more sodium ions in the cell , the Gβγ subunits bind more strongly to the GIRK4 channel; this means that the channel is more sensitive to the nerve impulses from the brain . In a related study , Wang et al . – who include all of the same researchers – also discovered that sodium ions affect GIRK2 channels from neurons in a similar way . By contrast , Touhara et al . found that the GIRK1/4 channel is unaffected by the sodium level inside the cell and is instead always sensitive to stimulation by nerve impulses that signal being at rest . Touhara et al . then looked at mouse heart cells that had been grown in the laboratory and found that they respond as if all of their GIRK channels were the GIRK1/4 type . That is to say , that a heart cell’s activity didn’t change much when extra sodium ions were present . This is likely because , unlike in neurons , the concentration of sodium ions inside a heart cell probably does not change much under normal conditions . These findings shed new light on G protein signaling , but there is still more that is not yet completely understood . For example , different GPCRs in cells will all release Gβγ subunits when stimulated but somehow produce specific responses . Touhara et al . are now interested in figuring out how this kind of specificity is achieved in heart cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2016	The GIRK1 subunit potentiates G protein activation of cardiac GIRK1/4 hetero-tetramers
Complete and robust human genome duplication requires loading minichromosome maintenance ( MCM ) helicase complexes at many DNA replication origins , an essential process termed origin licensing . Licensing is restricted to G1 phase of the cell cycle , but G1 length varies widely among cell types . Using quantitative single-cell analyses , we found that pluripotent stem cells with naturally short G1 phases load MCM much faster than their isogenic differentiated counterparts with long G1 phases . During the earliest stages of differentiation toward all lineages , MCM loading slows concurrently with G1 lengthening , revealing developmental control of MCM loading . In contrast , ectopic Cyclin E overproduction uncouples short G1 from fast MCM loading . Rapid licensing in stem cells is caused by accumulation of the MCM loading protein , Cdt1 . Prematurely slowing MCM loading in pluripotent cells not only lengthens G1 but also accelerates differentiation . Thus , rapid origin licensing is an intrinsic characteristic of stem cells that contributes to pluripotency maintenance . Metazoan DNA replication requires initiation at thousands of DNA replication origins during S phase of every cell cycle . Origins are genomic loci where DNA helicases first unwind DNA and DNA synthesis begins . Origins are made competent for replication during G1 phase of each cell cycle by the loading of minichromosome maintenance ( MCM ) complexes onto DNA . MCM is the core component of the replicative helicase , and the process of MCM loading is termed origin licensing . Total MCM levels remain constant throughout the cell cycle , but the levels of DNA-loaded MCM change as cells progress through the cell cycle . Cells can begin MCM loading as early as telophase and loading continues throughout G1 until the G1/S transition , the point of maximum DNA-loaded MCM ( Kimura et al . , 1994; Todorov et al . , 1995 ) . Throughout S phase , individual MCM complexes are activated for DNA unwinding as origins ‘fire’ . MCM complexes travel with replication forks and are progressively unloaded as replication forks terminate ( Figure 1a ) ( Deegan and Diffley , 2016; Remus and Diffley , 2009; Siddiqui et al . , 2013 ) . The control of origin licensing is critical for genome stability . Origins must not be re-licensed after S phase begins because such re-licensing can cause a genotoxic phenomenon known as re-replication which may result in double strand breaks , gene amplification , aneuploidy , and general genome instability ( Arias and Walter , 2007; Truong and Wu , 2011 ) . To avoid re-replication , MCM loading is tightly restricted to G1 phase by multiple overlapping mechanisms that destroy or inactivate MCM loading proteins to prevent any new origin licensing after S phase begins ( Remus and Diffley , 2009; Arias and Walter , 2007; Truong and Wu , 2011 ) . On the other hand , cells typically load 5- to 10-fold more MCM complexes in G1 than they strictly require under ideal circumstances , and the additional MCM loading ensures timely and complete genome duplication even if replication hurdles are encountered in S phase ( Ibarra et al . , 2008; Woodward et al . , 2006; Ge et al . , 2007 ) . It is possible for cells to proliferate with less than optimal MCM loading , but such cells are hypersensitive to DNA damage and replication stress ( Blow et al . , 2011; McIntosh and Blow , 2012 ) . MCM loading to license origins is restricted to G1 , but G1 length varies widely among different cell types . For example , specialized developmental and immune cell cycles have minimal G1 lengths of mere minutes ( O'Farrell et al . , 2004; Kinjyo et al . , 2015; Kermi et al . , 2017 ) . In cultured human embryonic stem cells , G1 is only 2–3 hr , and this short G1 is both a hallmark of and has been implicated in maintaining pluripotency ( Soufi and Dalton , 2016; Kareta et al . , 2015a ) . G1 lengthens early in differentiation , and in cultured somatic cells is often greater than 12 hr ( Calder et al . , 2013 ) . Thus , different proliferating cells have drastically different amounts of time available to complete MCM loading before making the G1-to-S phase transition . In addition , pluripotent stem cells respond to differentiation stimuli specifically in G1 phase , suggesting that the balance among cell cycle phases influences differentiation potential ( Gonzales et al . , 2015; Pauklin and Vallier , 2013 ) . Given that MCM loading is restricted to G1 and the wide variation of G1 lengths , we postulated that the absolute amount of loaded MCM in S phase is a product of both the time spent in G1 and the rate of MCM loading . The combination of these two parameters has implications for genome stability because loading more or less MCM in G1 influences S phase length and how effectively S phase cells can accommodate both endogenous and exogenous sources of replication stress ( Shima et al . , 2007; Pruitt et al . , 2007 ) . These implications are relevant both when cell cycle distributions change during development and during oncogenesis since many cancer cell lines also have short G1 phases . The actual rate of MCM loading in human cells has not yet been quantified , however , and little is known about how the amount , rate or timing of MCM loading varies between cells with different G1 lengths . Here , we utilized single-cell flow cytometry to measure MCM loading rates in asynchronous populations of pluripotent and differentiated cells . We discovered that rapid MCM loading is intrinsic to pluripotency , slows universally during differentiation , and rapid replication licensing suppresses differentiation . These findings demonstrate that the rate of MCM loading is subject to developmental regulation , and we suggest that rapid origin licensing is a new hallmark of pluripotency . We considered two possibilities for how cells with varying G1 lengths load MCM onto DNA . One possibility is that cells with short G1 phases load MCM at the same rate as cells with long G1 phases resulting in less total loaded MCM . Alternatively , cells with short G1 phases could load MCM faster than cells with long G1 phases and reach similar levels of loaded MCM . To distinguish between these scenarios , we developed an assay to measure DNA-loaded MCM in individual cells of asynchronously proliferating populations by adapting a previously reported flow cytometry method ( Håland et al . , 2015; Moreno et al . , 2016 ) . We extracted immortalized epithelial cells ( RPE1-hTERT ) with nonionic detergent to remove soluble MCM . We then fixed the remaining chromatin-bound proteins for immunofluorescence with anti-MCM2 antibody as a marker of the MCM2-7 complex , and for DNA content ( DAPI ) and DNA synthesis ( EdU ) to measure cell cycle phases ( Figure 1b , left ) . We used a sample without primary antibody as a control ( relevant flow cytometry gating schemes are shown in Figure 1—figure supplement 1 ) . For chromatin flow cytometry , MCM signal below the antibody threshold is colored grey ( ‘G1/G2/M-MCMDNA neg’ ) , whereas detectable MCM signal is colored either blue in G1 cells ( ‘G1-MCMDNA pos’ ) or orange in S phase cells ( ‘S-MCMDNA pos’ ) . As expected , total MCM protein levels do not substantially change during the cell cycle ( Figure 1b , right ) ( Todorov et al . , 1995 ) . For comparison to commonly used cell fractionation methods to assess MCM dynamics , we also probed immunoblots of chromatin-enriched fractions , and noted a similar MCM expression , G1 loading , and S phase unloading pattern ( Figure 1c ) ( Cook et al . , 2002; Méndez and Stillman , 2000 ) . Interestingly , individual G1 cells ( blue stripe ) have a very broad range of DNA-loaded MCM levels with a more than 100-fold difference between minimum and maximum ( Figure 1b , left ) . MCMs are unloaded during S phase , ending in G2/M with undetectable MCM on DNA ( Figure 1b , left ) . Moreover , loaded MCM is resistant to extraction in high-salt buffer which removes peripherally bound chromatin proteins ( Figure 1—figure supplement 2f–h ) , similar to yeast MCM complexes loaded in vitro ( Bowers et al . , 2004; Randell et al . , 2006 ) . We validated MCM2 antibody specificity using quiescent G0 synchronized cells ( MCM unloaded ) , and we also observed the same broad G1 signal distribution using MCM3 antibody ( Figure 1—figure supplement 2a–d ) . Loaded MCM complexes are extremely stable on DNA , both in vivo and in vitro ( Cocker et al . , 1996; Evrin et al . , 2009; Remus et al . , 2009; Bowers et al . , 2004 ) . In human cells , MCMs can persist on DNA for more than 24 hr during a cell cycle arrest and are typically only unloaded during S phase ( Kuipers et al . , 2011 ) . These properties result in MCM loading that occurs unidirectionally throughout G1 phase ( Symeonidou et al . , 2013 ) . The unidirectional nature of MCM loading means that G1 cells with low MCM levels are in early G1 , and G1 cells with high MCM levels are in late G1 . Since we observed a broad distribution of MCM loading throughout G1 including many cells with low levels of loaded MCM , we conclude that RPE1-hTERT cells load MCM relatively slowly during their ~ 9 hr G1 . We then used this method to analyze MCM loading in asynchronous cells with different G1 lengths . H9 human embryonic stem cells ( hESCs ) have a short G1 phase and spend most of the cell cycle in S phase . In contrast to the differentiated epithelial cells , the majority ( ~80% ) of G1 hESCs had high levels of loaded MCM; very few G1 cells had low levels of loaded MCM ( blue cells , Figure 1d ) . This difference suggests that hESCs load MCM rapidly to achieve abundant DNA-loaded MCM in a short time . To test if MCM loading varies in differentiated cells , we differentiated hESCs into neural progenitor cells ( NPCs ) to generate an isogenic pair of pluripotent and differentiated cells . In contrast to hESCs , differentiated NPCs had a longer doubling time and a wide distribution of DNA-loaded MCM in G1 ( blue cells , Figure 1d ) ; they also spend approximately five time longer in G1 ( Figure 1e; e . g . 15% of a 17 hr hESC cell cycle is 2 . 5 hr in G1 vs 45% of a 30 hr NPC cell cycle is 13 . 6 hr in G1 ) . Since the NPCs had many cells with low levels of DNA-loaded MCM , we conclude that these differentiated cells load MCM more slowly than hESCs . To generate another isogenic pair of pluripotent and differentiated cells , we reprogramed ARPE-19 primary retinal pigmented epithelial cells ( RPE ) into induced pluripotent stem cells ( iPSCs ) . The iPSCs had hallmark features of pluripotency as measured by microscopy , bisulfite sequencing , gene expression , and teratoma formation ( Figure 1—figure supplement 3 ) , and their G1 phases were typically seven times shorter than their differentiated parents ( Figure 1e ) . Like hESCs , the pluripotent iPSCs had predominantly high levels of DNA-loaded MCM in G1 ( Figure 1d ) . Importantly , both of the pluripotent cell lines reached approximately equal levels of DNA-loaded MCM at the start of S phase as their differentiated counterparts did , but in in less time ( the absolute MCM loading intensities are comparable when samples are processed and analyzed with identical instrument settings ) . Taken together , these data demonstrate that pluripotent cells load MCMs rapidly in G1 , but differentiated cells load MCMs slowly . We then quantified the relative MCM loading rates in pluripotent and differentiated cells using ergodic rate analysis , a mathematical method that can derive rates from fixed , steady state populations ( Kafri et al . , 2013 ) . Ergodic analysis can measure any unidirectional rate parameter from a steady state distribution and is not limited to the cell cycle ( e . g . car traffic jams ) ( Gray and Griffeath , 2001 ) . The ergodic analysis as applied to the cell cycle means that within a steady state population with a constant doubling time and cell cycle distribution , the number of cells at any point in the cell cycle is inversely related to the rate they move through that point . For any measured parameter , the density of cells indicates rate: low cell density on a flow cytometry plot indicates a fast rate passing through that cell cycle state , whereas high cell density indicates a slow rate . This phenomenon is analogous to a high density of slow-moving cars observed at a given point on a road in a traffic jam compared to a low density of fast-moving cars on an open highway . We visualized MCM loading as histograms of the MCMDNA intensities in only the G1 cells for ergodic rate analysis ( G1-MCMDNA , Figure 2a , b and Figure 2—figure supplement 1 ) . To compute MCM loading rate per hour , we experimentally determined the cell cycle distributions and doubling times of each cell population ( Figure 2—figure supplement 1 ) . Pluripotent cells reached near equal levels of loaded MCM at the G1/S transition in less time than differentiated cells . To quantify the actual MCM loading rate difference , we subdivided the G1-MCMDNA population into 10 equally-sized bins , calculated the MCM loading rate for each bin , then the overall average MCM loading rate for each G1 population . These calculations revealed that pluripotent hESCs loaded MCM 6 . 5 times faster per hour than differentiated NPCs and pluripotent iPSCs loaded MCM 3 . 9 times faster per hour than differentiated RPEs ( Figure 2c ) . Thus , pluripotent cells with short G1 phases load MCMs significantly faster than differentiated cells with long G1 phases . We hypothesized that MCM loading is fundamentally linked to pluripotency because MCM loading rate decreased during differentiation and increased during reprogramming . This idea predicts that slowed MCM loading is a phenomenon common to differentiation towards all germ layers . To test that hypothesis , we initiated differentiation in hESCs toward the three embryonic germ layers ( neuroectoderm , mesoderm and endoderm ) , collecting cells at 24 hr and 48 hr after inducing differentiation ( Figure 3—figure supplement 1 ) . We confirmed progress toward each lineage by the expected gene expression changes , particularly induction of lineage-specific markers and modest reduction of pluripotency markers – even at these very early time points ( Figure 3c ) . We assessed MCM loading rates during differentiation by flow cytometry as before . The MCM loading rate clearly decreased for all germ layers rapidly within the first 48 hr of initiating differentiation ( Figure 3a , compare the grey histograms for undifferentiated G1 cells to the green and blue histograms ) . The decrease in MCM loading rate also coincided with the increase in the proportion of G1 cells for each lineage . For example , within 24 hr of neuroectoderm differentiation , G1 had already lengthened and MCM loading had slowed , but during mesoderm ( BMP4 ) differentiation both G1 lengthening and slowed MCM loading took 48 hr ( Figure 3a and b ) . The closely coordinated changes that we universally observed during differentiation suggest that MCM loading rate is coupled to G1 length . Importantly , these results demonstrate that the rate of origin licensing by MCM loading is developmentally regulated . We next asked if G1 length and MCM loading rate are obligatorily coupled , or if the link can be short-circuited by artificially advancing the G1/S transition . To distinguish between these possibilities , we constructed an RPE1-hTERT derivative with a CYCLIN E1 cDNA under doxycycline-inducible control . Cyclin E1 overproduction reproducibly shortened G1 length , consistent with previous studies ( Figure 4a , b ) ( Resnitzky et al . , 1994; Ekholm-Reed et al . , 2004 ) . Strikingly , cells overproducing Cyclin E1 ( designated as ‘↑Cyclin E1’ ) not only spent less time in G1 but also began S phase with much lower amounts of loaded MCM compared to the control; this new subpopulation appeared in the central triangular region of the plots that is typically clear of S phase cells ( Figure 4c , d orange S-MCMDNApos ) . Cyclin E1 overproduction dramatically increased the proportion of these MCM-low early S phase cells by sixfold from 9 . 9% of control S phase cells to 63 . 6% of ↑Cyclin E1 , S phase cells ( Figure 4c-e ) . We also conclude that the MCM loading rate did not increase to accommodate the shorter G1 because the MCM loading pattern in G1 remained constant and the ↑Cyclin E1 cells had on average at least two-fold less DNA-loaded MCM in early S phase than control cells ( Figure 4f–h ) . Furthermore , the ↑Cyclin E , MCM-low cells incorporated significantly less EdU per unit time than MCM-high cells did ( 1 . 6 fold lower mean , 1 . 8 fold lower median ) , indicating that low levels of loaded MCM are insufficient for normal S phase progression ( Figure 4i ) . The early S phase cells with the least MCM loaded also had the least DNA synthesis by EdU intensity ( data not shown ) . Thus , shortening G1 length without increasing MCM loading rate causes G1 cells to enter S phase prematurely without the full complement of DNA-loaded MCM . Previous studies have shown that CDKs can inhibit MCM loading by directly inhibiting MCM loading factors , such as by stimulating Cdt1 degradation ( Cdc10-dependent transcript 1 ) , a protein essential for MCM loading ( Ekholm-Reed et al . , 2004; Sugimoto et al . , 2004; Tanaka and Diffley , 2002 ) . Cdt1 levels in lysates of asynchronous cells indeed decreased upon Cyclin E1 overproduction ( Figure 4—figure supplement 1a ) . On the other hand , since Cdt1 is stable in G1 phase and degraded in S phase , the lower Cdt1 signal could have reflected less Cdt1 in the lysate due to the higher proportion of S phase cells; this indirect effect could apply to any cell cycle-regulated protein in cell populations with different cell cycle distributions . To test that idea , we measured total Cdt1 protein levels specifically in G1 by flow cytometry ( Figure 4—figure supplement 1b ) . Cyclin E overproduction did not significantly reduce Cdt1 G1 levels relative to control ( 1 . 1-fold higher mean , 1 . 2 higher median , Figure 4—figure supplement 1c ) . We validated the Cdt1 antibody for immunofluorescence flow cytometry ( Figure 4—figure supplement 1d–f ) . We conclude that Cyclin E/Cdk2 inhibits MCM loading indirectly , at least in part , by shortening G1 and decreasing the time available for MCM loading . Loading MCM complexes onto DNA requires the six subunit Origin Recognition Complex ( ORC ) , Cdc6 , and Cdt1 . We hypothesized that fast MCM loading in pluripotent stem cells is achieved by increased levels of the loading proteins . To test this idea , we probed protein lysates of asynchronous cells to compare the amount of MCM loading proteins between isogenic cell lines . Total Mcm2 and ORC protein levels remained constant ( Figure 5a ) . The other MCM loading factors normally change in their abundance during the cell cycle due to regulated proteolysis . Cdc6 protein levels are low in G1 and high in S phase ( Figure 5b ) . Conversely , Cdt1 protein levels are high in G1 and low in S phase ( Figure 5b ) ( Mailand and Diffley , 2005; Pozo and Cook , 2016 ) . Since an asynchronous population of pluripotent cells spends significantly more time in S phase than differentiated cells do , we expected Cdc6 levels to be higher in asynchronous pluripotent cells compared to their isogenic counterparts . Cdc6 was indeed higher in pluripotent cells , as was Geminin , a protein regulated in a similar manner as Cdc6 ( Figure 5a ) ( McGarry and Kirschner , 1998 ) . To our surprise , even though the majority of asynchronous pluripotent cells were in S phase , a time when Cdt1 is degraded , Cdt1 levels were higher in asynchronous pluripotent cells than in isogenic differentiated cells ( Figure 5a ) . A similar observation was reported for mouse embryonic stem cells ( Ballabeni et al . , 2011 ) . These data imply that Cdt1 levels are higher in G1 phase of pluripotent cells than G1 of differentiated cells , providing a potential explanation for fast MCM loading in pluripotent cells . To directly measure Cdt1 levels specifically in G1 , we collected asynchronous hESCs and NPCs then fixed and stained them for Cdt1 , EdU and DAPI . S phase Cdt1 degradation in hESCs is similar to differentiated cells with very low levels in S phase ( purple , Figure 5c ) . In contrast , the hESCs had a large population of cells with high Cdt1 levels in G1 ( green cells ) and significant amounts of Cdt1 in G2/M phase ( grey cells ) , whereas the NPCs had a broad and overall lower Cdt1 distribution in G1 and very little Cdt1 in G2/M ( Figure 5d ) . The G1 hESCs consistently harbored significantly more Cdt1 than G1 NPCs did ( 2 . 9-fold higher median , 2 . 2-fold higher mean , three replicates [Figure 5d] ) . We note that CDT1 mRNA is modestly but consistently higher in asynchronous hESCs compared to differentiated derivatives , and that Cdt1 protein levels decrease during early differentiation coincident with the slowdown in licensing rate , but before loss of Oct4 ( Figure 3c and Figure 3—figure supplement 1 ) . We postulate that the higher amount of this essential MCM loading protein specifically in G1 contributes to the fast MCM loading rate in hESCs . Cdt1 is essential for MCM loading ( Pozo and Cook , 2016 ) ; therefore , reducing Cdt1 levels should slow MCM loading . If MCM loading rate is linked to G1 length , then slowing MCM loading by reducing Cdt1 levels could also lengthen G1 . To test this prediction , we used siRNA to reduce Cdt1 in hESCs and measured changes in both MCM loading rate and G1 length ( Figure 6a , b ) . As expected , Cdt1 depletion reduced MCM loading rate in hESCs ( Figure 6c ) . Strikingly , G1 length increased coincidentally with the decrease in MCM loading rate ( Figure 6d ) . These data corroborate the close link between MCM loading rate and G1 length in hESCs . As a complement to slowing MCM loading in hESCs with a short G1 , we also attempted to accelerate origin licensing in cells with a long G1 by overproducing essential licensing proteins . We first constructed an RPE_hTERT derivative with ~two-fold inducible Cdt1 overproduction , but this manipulation was insufficient to accelerate MCM loading ( Figure 6—figure supplement 1b ) . We also tested ectopic myc-tagged Cdc6 expressed constitutively in RPE cells ( Figure 6e ) ; this addition also had only minimal effects on MCM loading rate ( Figure 6g , compare the grey and green histograms ) . We considered , however , that human Cdc6 is unstable throughout much of G1 phase because it is targeted for degradation by APCCdh1 Mailand and Diffley , 2005; Petersen et al . , 2000 . We therefore expressed a previously-described Cdc6 mutant that is resistant to APCCDH1-mediated destruction both alone and combination with inducible Cdt1 ( Figure 6e and Figure 6—figure supplement 1a ) ( Mailand and Diffley , 2005; Petersen et al . , 2000 ) . We have previously demonstrated that tagged Cdc6 and Cdt1 are functional ( Cook et al . , 2002; Coleman et al . , 2015; Chandrasekaran et al . , 2011 ) . Expression of the stable Cdc6-mut was sufficient to increase MCM loading rates ( Figure 6g , compare the blue histogram to the grey and green histograms ) ; Cdt1 overproduction had little additive effect on MCM loading rate in RPE cells ( Figure 6—figure supplement 1b ) . Interestingly accelerating MCM loading by this method did not shorten G1 in RPEs ( Figure 6f ) , further demonstrating that the length of G1 phase and the rate of MCM loading to license origins can be uncoupled . Slow MCM loading may delay S phase entry through the licensing checkpoint ( Teer et al . , 2006; Shreeram et al . , 2002; Nevis et al . , 2009; Ge and Blow , 2009 ) , but rapid MCM loading itself is not sufficient to trigger S phase entry . Our demonstration that slower MCM loading occurs universally during early differentiation suggested a functional link between the rate of MCM loading and pluripotency maintenance . We considered that slowing MCM loading might promote differentiation . To explore this idea , we prematurely slowed MCM loading in hESCs by Cdt1 depletion prior to inducing their differentiation ( Figure 7e ) . After Cdt1 depletion , we stimulated differentiation toward mesoderm with BMP4 ( Bernardo et al . , 2011 ) . After 48 hr , we quantified Oct4 and Cdx2 by immunostaining ( Figure 7a , Figure 7—figure supplement 1 ) . The pluripotency transcription factor Oct4 and the homeobox transcription factor Cdx2 reciprocally repress one another’s expression , creating a clear distinction between Oct4-positive Cdx2-negative pluripotent cells and Oct4-negative Cdx2-positive differentiating cells ( Niwa et al . , 2005 ) . We quantified the mean fluorescence intensity of both Oct4 and Cdx2 in >18 , 000 cells per condition with a customized , automated CellProfiler pipeline , plotting the signal intensities for each cell in a density scatter plot ( Figure 7b , c ) . Stimulating control hESCs with 10 ng/mL of BMP4 slightly shifted the population toward differentiation , but most cells remained pluripotent with high Oct4 levels at this time point . Strikingly , hESCs pretreated with Cdt1 siRNA to prematurely slow MCM loading gained a substantial population of Oct4 negative-Cdx2 positive differentiating cells relative to controls that were treated similarly . To quantify the extent of differentiation , we divided the Cdx2 intensity of each cell by its Oct4 intensity , creating a single differentiation score ( Figure 7d ) . After 10 ng/ml BMP4 treatment , Cdt1-depleted hESCs had significantly higher scores , indicating that prematurely slowing MCM loading promoted differentiation ( p<0 . 0001 , two-tailed Mann-Whitney test ) . Both control cells and Cdt1-depleted cells differentiated more fully at a higher concentration of 50 ng/mL BMP4 , but the Cdt1-depleted cells still differentiated further than the controls ( p<0 . 0001 , two-tailed Mann-Whitney test , Figure 7b and Figure 7—figure supplement 2a ) . Other combinations of BMP4 concentrations or treatment times also resulted in a consistent , significant increase in differentiation in cells pretreated to slow MCM loading ( p<0 . 0001 , two-tailed Mann-Whitney test , data not shown ) . Importantly , the phenotype was conserved across multiple differentiation lineages , as prematurely slowing MCM loading prior to endoderm differentiation also increased the number of cells positive for the endoderm transcription factor Sox17 relative to controls at the same time point ( Figure 7—figure supplement 1b ) . To test if the pluripotency maintenance was due to Cdt1’s role in origin licensing and not its mitotic or other functions ( Varma et al . , 2012 ) , we slowed licensing by depleting the orthogonal MCM loading protein , Cdc6 ( Figure 7—figure supplement 3a–d ) . A more modest Cdc6 knockdown correlated with a weaker , but detectable effect on MCM loading . Interestingly , this degree of licensing inhibition had no effect on G1 length . Despite the short G1 length , slowing MCM loading by depleting Cdc6 significantly promoted differentiation ( Figure 7—figure supplement 2b–f , p<0 . 0001 , two-tailed Mann-Whitney test . Thus , we conclude that slow MCM loading generally promoted differentiation and by extension , that rapid MCM loading preserves pluripotency . In this study , we demonstrate that rapid MCM loading to license replication origins is an intrinsic property of pluripotent cells . Human embryonic stem cells have a remarkably fast MCM loading rate , and reprogramming to create induced pluripotent stem cells increases MCM loading rate . Moreover , MCM loading slows concurrently with the G1 lengthening and extensive cell cycle remodeling that accompany the early stages of differentiation ( Figure 7f ) . To our knowledge , this is the first demonstration that the rate of MCM loading is developmentally regulated . Developmental regulation of MCM loading rate is consistent with previous work showing higher levels of total Cdt1 in asynchronous mouse ESCs than in differentiated cells ( Ballabeni et al . , 2011 ) . The regulated decrease in MCM loading rate is critical during differentiation , as rapid MCM loading protects pluripotency , and prematurely slowing MCM loading promotes differentiation . Pluripotent stem cells load MCM complexes rapidly to reach similar total amounts of loaded MCM at the G1/S transition in less time than their isogenic differentiated counterparts . Although we did not detect substantial MCM loading in telophase , as suggested previously ( Dimitrova et al . , 2002 ) , it is clear that telophase loading is an option in some cells and a requirement in cells with no detectable G1 such as S . pombe and in the first nuclear divisions in D . melanogaster ( Nishitani et al . , 2000; Farrell and O'Farrell , 2014 ) . Stem cells achieve faster MCM loading , at least in part , by particularly high Cdt1 levels in G1 . These high levels are achieved not only by a modest difference in CDT1 mRNA ( Figure 3c ) but also post-transcriptionally by specific re-accumulation of Cdt1 protein in the preceding G2 phase ( Figure 5c ) . Cdt1 stability in G2 phase has been attributed to geminin-mediated protection from the SCFSkp2 E3 ubiquitin ligase in non-stem cells ( Tsunematsu et al . , 2013; Ballabeni et al . , 2004; Clijsters et al . , 2013 ) . It is not clear , however , that geminin levels are particularly high in stem cells relative to differentiated cells ( aside from differences expected from cell cycle distribution changes ( Figure 5a ) ( Ballabeni et al . , 2011 ) ) , so it seems unlikely that geminin drives higher Cdt1 levels in stem cells . Skp2 levels also do not change in the earliest stages of stem cell differentiation ( Egozi et al . , 2007 ) . Cdt1 is protected in late S phase and G2 by cyclin A/Cdk1 activity ( Rizzardi et al . , 2015 ) , and we thus consider it likely that the documented high CDK activity in stem cells contributes to Cdt1 stabilization in G2 ( Sela et al . , 2012 ) . The anticipatory Cdt1 accumulation to promote MCM loading in G1 was originally proposed from experiments in cancer-derived cell lines ( Ballabeni et al . , 2004; Clijsters et al . , 2013 ) . Our observations in stem cells suggest this strategy is employed by non-transformed cells during developmental stages that require short G1 . Other factors besides Cdt1 accumulation may also accelerate MCM loading . The hESCs we assayed have 2–3 fold greater Cdt1 protein levels in G1 relative to NPCs yet load MCM 6 . 5 times faster per hour than NPCs . One Cdt1 molecule can ( in vitro ) load multiple MCM complexes since Cdt1 is released into the soluble phase immediately after completing a loading reaction ( Ticau et al . , 2015 ) . Stem cells may experience less Cdc6 degradation in early G1 due to nearly constitutive Cyclin E/Cdk2 activity and/or attenuated APCCdh1 activity , corroborated by our observation that a Cdc6 mutant that is not targeted by APCCdh1 increases MCM loading rate in cells with long G1 phases ( Figure 6g ) ( Ballabeni et al . , 2011; Neganova et al . , 2009; Filipczyk et al . , 2007 ) . Additionally , stem cells are enriched for euchromatin , an environment that may be particularly permissive for rapid MCM loading ( Chen and Dent , 2014 ) . Rapid MCM loading may itself contribute to mechanisms that maintain short G1 phases in pluripotent cells . The origin licensing checkpoint links the amount of loaded MCM to G1 length by controlling Cdk2 activity . In that regard , overproducing Cyclin E ‘short-circuited’ the licensing checkpoint in slow loading differentiated cells . This checkpoint has thus far only been demonstrated in p53-proficient differentiated mammalian cells ( Ge and Blow , 2009 ) , but the G1 lengthening of hESCs after Cdt1 depletion suggests that pluripotent stem cells also have a functioning licensing checkpoint . Cells with fast MCM loading could satisfy this checkpoint quickly , activate Cyclin E/Cdk2 , and thus spend less time in G1 . Mechanisms that support short G1 length preserve pluripotency in hESCs ( and promote reprogramming to iPSCs ) since cells are most sensitive to differentiation cues in G1; in that regard , extending G1 phase in hESCs can increase differentiation propensity ( Soufi and Dalton , 2016; Pauklin and Vallier , 2013; Filipczyk et al . , 2007; Coronado et al . , 2013 ) . Recent work with quintuple knockout mice lacking all D and E type cyclins also reported that Cyclin E/Cdk2 further contributes to maintaining pluripotency by stabilizing the Oct4 , Sox2 , and Nanog transcription factors ( Liu et al . , 2017 ) . Fast MCM loading may have evolved as an intrinsic property of pluripotent cells to maintain high Cdk2 activity and keep G1 phase short . Cyclin/Cdk activity is not the sole connection between the cell cycle and pluripotency . Non-CDK cell-cycle-associated proteins regulate expression of key pluripotency genes including SOX2 and NANOG ( Gonzales et al . , 2015; Pauklin et al . , 2016; Li et al . , 2012 ) . Pluripotency transcription factors themselves regulate expression of cell cycle genes including those encoding cyclins , CDK inhibitors , and E2F3a ( Lee et al . , 2010; Kanai et al . , 2015; Choi et al . , 2012 ) . On the other hand , pluripotency and cell cycle functions can be genetically uncoupled in experiments where manipulating the cell cycle did not alter pluripotency and vice versa ( Scognamiglio et al . , 2016; Kareta et al . , 2015b ) . We observe that licensing inhibition can accelerate differentiation even without greatly lengthening G1 ( Figure 7—figure supplements 2 , 3 ) which may point to an additional direct and cell cycle-independent link between MCM loading rate and differentiation . We note that early differentiation is not the only setting in which rapid MCM loading during a short G1 may be relevant . Like hESCs , activated T cells have very fast cell cycles with short G1 phases ( Kinjyo et al . , 2015 ) . Oncogenic transformation is also frequently associated with G1 shortening . It may be that the pathways linking differentiation to MCM loading rate are also coopted in some cancers to induce rapid licensing . On the other hand , a subset of cancers may proliferate in a perpetually underlicensed state that contributes to the genome instability characteristic of transformed cells . Future investigations will elucidate the molecular relationships among developmental signaling pathways , MCM loading rate , and cell cycle remodeling . Cell lines were authenticated by STR profiling ( ATCC , Manassas , VA ) and confirmed to be mycoplasma negative . T98G , HEK293T , and RPE1-hTERT were cultured in Dulbecco's Modified Eagle Medium ( DMEM ) supplemented with 2 mM L-glutamine and 10% fetal bovine serum ( FBS ) and incubated in 5% CO2 at 37°C . ARPE-19 ( male ) were cultured in 1:1 DMEM:F12 supplemented with 2 mM L-glutamine and 10% fetal bovine serum and incubated in 5% CO2 at 37°C . T98G , HEK293T , RPE1-hTERT , and ARPE-19 cells were from the ATCC and were passaged with trypsin and not allowed to reach confluency . WA09 ( H9 hESCs ) were cultured in mTeSR1 ( StemCell Technologies ) with media changes every 24 hr on Matrigel ( Corning , New York , NY ) coated dishes and incubated in 5% CO2 at 37°C . H9s had normal diploid karyotype at passage 32 and were used from passage 32–42 . ARPE-iPSCs were cultured in Essential 8 ( Life Technologies , Grand Island , NY ) with media changes every 24 hr on Matrigel ( Corning ) coated dishes and incubated in 5% CO2 at 37°C . iPSCs were used from passage 20–25 and had normal karyotype . Both hESCs and iPSCs were routinely passaged every 4 days as aggregates using ReLeSR , according to manufacturer’s instructions ( StemCell Technologies , Canada ) . The hESCs and iPSCs were only passaged as single cells in 10 μM Y-27632 2HCl ( Selleck Chemicals , Houston , TX ) for experiments , as described previously ( Watanabe et al . , 2007 ) . NPCs were cultured in Neural Progenitor Medium ( StemCell Technologies ) with media changes every 24 hr on poly-L-ornithine/Laminin ( Sigma Aldrich , St . Louis , MO ) coated dishes and incubated in 5% CO2 at 37°C . NPCs were passaged with StemPro Accutase ( Gibco , Waltham , MA ) weekly . Cells were collected via trypsinization . For total protein lysates , cells were lysed on ice for 20 min in CSK buffer ( 300 mM sucrose , 100 mM NaCl , 3 mM MgCl2 , 10 mM PIPES pH 7 . 0 ) with 0 . 5% triton x-100 and protease and phosphatase inhibitors ( 0 . 1 mM AEBSF , 1 µg/ mL pepstatin A , 1 µg/ mL leupeptin , 1 µg/ mL aprotinin , 10 µg/ ml phosvitin , 1 mM β-glycerol phosphate , 1 mM Na- orthovanadate ) . Cells were centrifuged at 13 , 000 xg at 4°C for 5 min , then the supernatants were transferred to a new tube for a Bradford Assay ( Biorad , Hercules , CA ) using a BSA standard curve . Chromatin fractionation for immunoblotting was performed as described previously ( Cook et al . , 2002; Méndez and Stillman , 2000 ) , using CSK buffer with 1 mM ATP , 5 mM CaCl2 , 0 . 5% triton x-100 and protease and phosphatase inhibitors to isolate insoluble proteins and S7 nuclease ( Roche ) to release DNA bound proteins . A Bradford Assay ( Biorad ) was performed for equal loading . For 100 mM or 300 mM NaCl soluble/pellet fractionation , cells were lysed in standard CSK ( 100 mM NaCl ) or high-salt CSK ( 300 mM NaCl ) with 0 . 5% triton X-100 with protease and phosphatase inhibitors for 5 min on ice . Then cells were centrifuged at 2000 xg for 3 min , supernatants transferred to a new tube as the soluble fraction . The remaining pellet was suspended in 2x SDS loading buffer ( 2% SDS , 5% 2-mercaptoethanol , 0 . 1% bromophenol blue , 50 mM Tris pH 6 . 8 , 10% glycerol ) as the pellet fraction . Bradford assay was performed on the soluble fraction for equal loading . Samples were diluted with SDS loading buffer ( final: 1% SDS , 2 . 5% 2-mercaptoethanol , 0 . 1% bromophenol blue , 50 mM Tris pH 6 . 8 , 10% glycerol ) and boiled . Samples were run on SDS-PAGE gels , then the proteins transferred onto polyvinylidene difluoride membranes ( Thermo Fisher , Waltham , MA ) or nitrocellulose ( GE Healthcare , Chicago , IL ) . Membranes were blocked at room temperature for 1 hr in either 5% milk or 5% BSA in Tris-Buffered-Saline-0 . 1%-tween-20 ( TBST ) . After blocking , membranes were incubated in primary antibody overnight at 4°C in either 1 . 25% milk or 5% BSA in TBST with 0 . 01% sodium azide . Blots were washed with TBST then incubated in HRP-conjugated secondary antibody in either 2 . 5% milk or 5% BSA in TBST for 1 hr , washed with TBST , and then membranes were incubated with ECL Prime ( Amersham , United Kingdom ) and exposed to autoradiography film ( Denville , Holliston , MA ) . Equal protein loading was verified by Ponceau S staining ( Sigma Aldrich ) . Antibodies used for immunoblotting were: Mcm2 , ( BD Biosciences , San Jose , CA , Cat#610700 ) , Mcm3 , ( Bethyl Laboratories , Montgomery , TX , Cat#A300-192A ) , Cdt1 , ( Cell Signaling Technologies , Beverly , MA , Cat#8064S ) , Cdc6 , ( Santa Cruz Biotechnology , Santa Cruz , CA , Cat#sc-9964 ) , Oct4 , ( Abcam , Cambridge , MA , Cat#ab19857 ) , Cyclin E1 , ( Santa Cruz Biotechnology , Cat#sc-198 ) , Orc1 , ( Bethyl Laboratories , Cat#A301-892A ) , Orc6 , ( Santa Cruz Biotechnology , Cat#sc-32735 ) , geminin , ( Santa Cruz Biotechnology , Cat#sc-13015 ) , Histone H3 , ( Cell Signaling Technologies , Cat#4499S ) , TRA-1–60 , ( Invitrogen , Cat#41–1000 ) , nestin , ( Abcam , Cat#ab22035 ) , p27 ( Santa Cruz Biotechnology , Cat#sc-528 ) , α-tubulin ( Sigma Aldrich , Cat#9026 ) . For EdU-labeled samples , cells were incubated with 10 uM EdU ( Santa Cruz Biotechnology ) for 30 min prior to collection . For total protein flow cytometry , cells were collected with trypsin and resuspended as single cells , washed with PBS , and fixed with 4% paraformaldehyde ( Electron Microscopy Sciences , Hatfield , PA ) in PBS for 15 min at room temperature , then 1% BSA-PBS was added , mixed and cells were centrifuged at 1000 xg for 7 min ( and for all following centrifuge steps ) then washed with 1% BSA-PBS and centrifuged . Fixed cells were permeabalized with 0 . 5% triton x-100 in 1% BSA-PBS at room temperature for 15 min , centrifuged , then washed once with 1% BSA , PBS and centrifuged again before labeling . For chromatin flow cytometry , cells were collected with trypsin and resuspended as single cells , washed with PBS , and then lysed on ice for 5 min in CSK buffer with 0 . 5% triton x-100 with protease and phosphatase inhibitors . Next , 1% BSA-PBS was added and mixed , then cells were centrifuged for 3 min at 1000 xg , then fixed in 4% paraformaldehyde in PBS for 15 min at room temperature . For 100 mM NaCl vs 300 mM NaCl CSK , cells were processed as in chromatin flow cytometry above , except CSK containing 300 mM NaCl was used instead of the normal 100 mM NaCl . Next , 1% BSA-PBS was added , mixed and cells were centrifuged then washed again before labeling . The labeling methods for total protein samples and chromatin samples were identical . For DNA synthesis ( EdU ) , samples were centrifuged and incubated in PBS with 1 mM CuSO4 , 1 μM fluorophore-azide , and 100 mM ascorbic acid ( fresh ) for 30 min at room temperature in the dark . 1% BSA-PBS +0 . 1% NP-40 was added , mixed and centrifuged . Samples were resuspended in primary antibody in 1% BSA-PBS +0 . 1% NP-40 and incubated at 37°C for 1 hr in the dark . Next , 1% BSA-PBS +0 . 1% NP-40 was added , mixed and centrifuged . Samples were resuspended in secondary antibody in 1% BSA-PBS +0 . 1% NP-40 and incubated at 37°C for 1 hr in the dark . Next , 1% BSA-PBS +0 . 1% NP-40 was added , mixed and centrifuged . Finally , cells were resuspended in 1% BSA-PBS +0 . 1% NP-40 with 1 μg/mL DAPI ( Life Technologies ) and 100 μg/mL RNAse A ( Sigma Aldrich ) and incubated overnight at 4°C in the dark . Samples were run on a CyAn ADP flow cytometer ( Beckman Coulter , Brea , CA ) and analyzed with FCS Express six software ( De Novo , Glendale , CA ) . The following antibody/fluorophore combinations were used: ( 1 ) : Alexa 647-azide ( Life Technologies ) , primary: Mcm2 ( BD Biosciences , Cat#610700 ) , secondary: Donkey anti-mouse-Alexa 488 ( Jackson ImmunoResearch ) , DAPI . ( 2 ) : Alexa 488-azide ( Life Technologies ) , primary: Cdt1 ( Abcam , Cat#610700 ) , secondary: Donkey anti-rabbit-Alexa 647 ( Jackson ImmunoResearch ) , DAPI . ( 3 ) : Alexa 647-azide ( Life Technologies ) , primary: Mcm3 ( Bethyl Cat#A300-192A ) , secondary: Donkey anti-rabbit-Alexa 488 ( Jackson ImmunoResearch ) , DAPI . ( 4 ) : primary: Mcm3 ( Bethyl Cat#A300-192A ) , Mcm2 ( BD Biosciences , Cat#610700 ) , secondary: Donkey anti-mouse-Alexa 488 , Donkey anti-rabbit-Alexa 647 ( Jackson ImmunoResearch , West Grove , PA ) , DAPI . Cells were gated on FS-area vs SS-area . Singlets were gated on DAPI area vs DAPI height . The positive/negative gates for EdU and MCM were gated on a negative control sample , which was treated with neither EdU nor primary antibody , but incubated with 647-azide and the secondary antibody Donkey anti-mouse-Alex 488 and DAPI to account for background staining ( Figure 1—figure supplement 1 ) . Doubling time was calculated by plating equal number of cells as described above and counting cell number over time using a Luna II automated cell counter ( Logos Biosystems , South Korea ) at 24 , 48 , and 72 hr after plating . Three or four wells were counted as technical replicates at each timepoint . GraphPad Prism’s regression analysis was used to compute doubling time , and multiple biological replicates were averaged for a final mean doubling time . ARPE-19s were counted four times , hESCs and NPCs three times , and iPSCs two times . To synchronize cells in G1 , T98G cells were grown to 100% confluency , washed with PBS , and incubated for 72 hr in 0 . 1% FBS , DMEM , L-glutamine . After serum-starvation , cells were re-stimulated by passaging 1:3 with trypsin to new dishes in 20% FBS , DMEM , L-glutamine , collecting cells 10 hr and 12 hr post-stimulation . To synchronize T98G cells in early S phase , cells were treated as for G1 , except 1 mM Hydroxyurea ( Alfa Aesar , Haverhill , MA ) was added to the media upon re-stimulating and cells were collected 18 hr post-stimulation . To synchronize cells in mid-late S , cells were treated as in early S , then at 18 hr post-stimulation cells were washed with PBS and released into 10% FBS , DMEM , L-Glutamine , collecting 6 hr , 8 hr post release . To synchronize RPE1-hTERT cells in G0 , cells were grown to 100% confluency , then incubated for 48 hr in 10% FBS , DMEM , L-glutamine . For RPE1 in G1/S , G0 cells were trypsinized and passaged 1:6 with trypsin to new dishes in 10% FBS , DMEM , L-glutamine , and collected with trypsin 22 hr later . For cycloheximide ( Sigma ) treatment , asynchronous RPE cells were treated with 10 ug/mL for 4 hr or 8 hr as indicated . For UV irradiation , asynchronous RPE cells were treated with 20 J/m2 of UV with a Stratalinker ( Stratagene , San Diego , CA ) and collected 1 hr later . The pInducer20-Cyclin E plasmid was constructed using the Gateway cloning method ( Invitrogen ) . The attB sites were added to Cyclin E1 cDNA by PCR using Rc/CMV cyclin E plasmid as a template and BP-cycE-F ( 5' GGGGACAAGTTTGTACAAAAAAGCAGGCTACCATGAAGGAGGACGGCGGC ) and BP-cycE-R primers ( 5' GGGGACCACTTTGTACAAGAAAGCTGGGTTCACGCCATTTCCGGCCCGCT ) ( Hinds et al . , 1992 ) . The PCR product was recombined with pDONR221 plasmid using BP clonase ( Invitrogen ) according to the manufacturer’s instructions and transformed into DH5α to create pENTR221-Cyclin E1 . Then the LR reaction was performed between pInducer20 and pENTR221-Cyclin E1 using LR Clonase ( Invitrogen , Carlsbad , CA ) according to manufacturer’s instructions and transformed into DH5α to create pInducer20-Cyclin E1 . The pInducer20 plasmid was converted to blasticidin resistance ( pInducer20-blast2 ) by Gibson Assembly ( New England Biolabs , Ipswitch , MA ) following manufacturer’s protocol . pInducer20 was cut with AgeI and assembled with PCR products with the following primers: AgeI-rta3-F: 5- gctcggatctccaccccgtaccggtcctgcagtcgaattcac AgeI-IRES-blast-R: 5- ACAAAGGCTTGGCCATGGTT TAAGCTTATCATCGTGTTTTTCA Blast-F:5- tgaAaaacacgatgataagcttaaaccatggccaagcctttgt Blast-AgeI-Ind-R: 5- GTTCAATCATGGTGGACCGG CTATTAGCCCTCCCACACATAACCA The pLenti CMV blast plasmid was a template for the blasticidin resistance gene . A tagged Cdt1-HA was cloned into pInducer20-blast using Gateway cloning as described above . The Cdc6 mutant unable to bind APCCDH1 ( 5myc-Cdc6-mut ) was described previously: R56A , L59A , K81A , E82A , N83A ( Petersen et al . , 2000 ) . pCLXSN-5myc-Cdc6-wt was cloned to 5myc-Cdc6-mut by two sequential Gibson assemblies ( NEB ) according to manufacturer’s instructions . Primers used: CDC6-KEN-F: 5- ctccaccaaagcaaggcaaggcggccgcaggtccccctcactcacatacac CDC6-KEN-R: 5- GTGTATGTGAGTGAGGGGGACCTGCGGCCGCCTTGCCTTGCTTTGGTGGAG CDC6-DBOX-F: 5- aagccctgcctctcagccccgccaaacgtgccggcgatgacaacctatgcaa CDC6-DBOX-R: 5- TTGCATAGGTTGTCATCGCCGGCACGTTTGGCGGGGCTGAGAGGCAGGGCTT To package retrovirus , pCLXSN 5myc-Cdc6 wt or mut were co-transfected with pCI-GPZ and VSVG plasmids into HEK293T using 50 μg/mL Polyethylenimine-Max ( Aldrich Chemistry ) . To package lentivirus , pInducer20-Cyclin E1 or pInducer20-blast2-Cdt1-HA wwere co-transfected with ΔNRF and VSVG plasmids into HEK293T using 50 μg/mL Polyethylenimine-Max ( Aldrich Chemistry ) . Viral supernatant was transduced with 8 ug/mL Polybrene ( Millipore , Burlington , MA ) onto RPE1-hTERT cells overnight . Cells were selected with 500 ug/mL neomycin ( Gibco ) or 5 μg/mL blasticidin ( Research Products International , Mount Prospect , IL ) for 1 week . To overproduce Cyclin E1 , cells were treated with 100 ng/mL doxycycline ( CalBiochem , San Diego , CA ) for 72 hr in 10% FBS , DMEM , L-glutamine . Control cells were the Inducer20-Cyclin E1 without doxycycline . To overproduce Cdt1 , cells were treated with 100 ng/mL doxycycline for 48 hr in 10% FBS , DMEM , L-glutamine , control cells were without doxycycline . For siRNA treatment , Dharmafect 1 ( Dharmacon , Lafayette , CO ) was mixed in mTeSR1 with the appropriate siRNA according to the manufacturer’s instructions , then diluted with mTeSR1 and added to cells after aspirating old media . The final siRNA concentrations were: 100 nM siControl ( Luciferase ) , 25 or 100 nM siCdt1 , or a mixture of two siCdc6 ( 2144 and 2534 at 50 nM each ) . The Cdt1 siRNA mix was incubated on cells for either 20 or 24 hr , then changed to new mTeSR1 without siRNA . The Cdc6 siRNA mix was incubated on cells for 24 hr , then changed to new mTeSR1 without siRNA for 8 hr ( 32 total hours ) . The Cdt1 , Cdc6 and Luciferase siRNA were described previously ( Coleman et al . , 2015; Nevis et al . , 2009 ) . For siRNA treatment of RPE cells , Dharmafect 1 ( Dharmacon ) was mixed in Optimem ( Gibco ) with the appropriate siRNA according to manufacturer’s instructions , then diluted with DMEM , 10% FBS , L-glutamine and added to cells after aspirating old media . The next day , the siRNA mix was aspirated and replaced with fresh DMEM , 10% FBS , L-glutamine , collecting samples 72 hr after the start of siRNA treatment . The siRNA were siControl ( Luciferase ) at 100 nM or a mixture of two MCM3 siRNA ( 2859 and 2936 at 50 nM each ) . siMCM3-2859 5’- augacuauugcaucuucauugdTdT siMCM3-2936 5’- aacauaugacuucugaguacudTdT Mesoderm ( BMP4 ) : hESCs were passaged as single cells at 7 × 103/ cm2 in mTeSR1 with 10 μM Y-27632 2HCl onto Matrigel-coated plates . 24 hr later , the media was changed to start differentiation with fresh mTeSR1 with 100 ng/mL BMP4 ( R and D Systems , Minneapolis , MN ) , and 24 hr later the media was changed to fresh mTeSR1 with 100 ng/mL BMP4 for 48 total hours of differentiation . Neuroectoderm: hESCs were differentiated using a monolayer-based protocol in Neural Induction Medium: hESCs were passaged as single cells at 5 . 2 × 104/ cm2 in STEMdiff Neural Induction Medium ( StemCell Technologies ) with 10 μM Y-27632 2HCl onto Matrigel coated plates , and plating started the differentiation . 24 hr later , the media was changed to fresh Neural Induction Medium for another 24 hr for 48 hr total differentiation . To derive NPCs , hESCs were differentiated in Neural Induction medium ( StemCell Technologies ) using the Embryoid Body Neural Induction protocol according to manufacturer’s instructions , similar to previous reports ( Robinson et al . , 2016 ) . Once generated , NPCs were maintained in Neural Progenitor Medium ( StemCell Technologies ) . Mesoderm ( GSK3βi ) : hESCs were passaged as single cells at 3 × 104/ cm2 in mTeSR1 with 10 μM Y-27632 2HCl onto Matrigel-coated plates . 24 hr later , the media was changed to start differentiation . Cells were washed with Advanced RPMI 1640 ( Gibco ) , then incubated in Advanced RPMI 1640 with B27 minus insulin ( Gibco ) , 2 mM L-glutamine , and 8 μM CHIR-99021 ( Selleck Chemicals ) . At 24 hr after changing the media , cells were washed with Advanced RPMI 1640 then incubated in Advanced RPMI 1640 with B27 minus insulin ( Gibco ) , 2 mM L-glutamine , without CHIR-99021 for 24 hr for a total of 48 hr of differentiation . Endoderm: hESCs were passaged as single cells at 4 × 103/ cm2 in mTeSR1 with 10 μM Y-27632 2HCl onto Matrigel-coated plates . The next day , the media was changed to fresh mTeSR1 without Y-27632 2HCl . 24 hr later , the media was changed to start differentiation . The cells were washed with Advanced RPMI 1640 ( Gibco ) , then incubated in Advanced RPMI 1640 with 0 . 2% FBS , 2 mM L-glutamine , 100 ng/mL Activin A ( R and D Systems ) and 2 . 5 μM CHIR-99021 . At 24 hr after changing the media , cells were washed with Advanced RPMI 1640 then incubated in Advanced RPMI 1640 with 0 . 2% FBS , 2 mM L-glutamine , 100 ng/mL Activin A , without CHIR-99021 for 24 hr for a total of 48 hr of differentiation . Phase contrast images were acquired with an Axiovert 40 CFL inverted microscope , 20x objective ( Zeiss , Germany ) . For immunofluorescence microscopy , hESCs were plated as single cells in mTeSR1 with 10 μM Y-27632 2HCl in Matrigel-coated , 24 well , #1 . 5 glass bottom plates ( Cellvis ) at 7 × 103/ cm2 for siCdt1 , Mesoderm ( BMP4 ) , at 5 × 103 for siCdc6 , Mesoderm ( BMP4 ) , and at 4 × 103/ cm2 for siCdt1 , Endoderm . Cells were incubated with siCdt1 for 20 hr ( Mesoderm ( BMP4 ) ) , 24 hr ( Endoderm ) or siCdc6 for 32 hr ( Mesoderm [BMP4] ) all in parallel with siControl as described above ( siRNA transfections ) . After siRNA treatment , cells were differentiated as described above ( Differentiation ) with the following modifications: For Mesoderm ( BMP4 ) , multiple BMP4 concentrations and treatment times were used as indicated ( Figure 7 , Figure 7—figure supplement 1 ) . For treatment less than 48 hr , cells were incubated in mTeSR1 after siRNA treatment until starting differentiation . ( Example: 12 hr of mTeSR1 then 36 hr of BMP4 , for a total of 48 hr ) . For endoderm , the first RPMI/Activin/CHIR-99021 was immediately after siRNA , without a day of incubation in mTeSR1 . After differentiation , cells were fixed in 4% paraformaldehyde in PBS for 15 min at room temperature , washed with PBS , and permeabalized with 5% BSA , PBS , 0 . 3% triton x-100 at 4°C overnight . Next , cells were incubated in primary antibody in 5% BSA , PBS , 0 . 3% triton x-100 at 4°C overnight . Cells were washed with PBS at room temperature , then incubated in secondary antibody in 5% BSA , PBS , 0 . 3% triton x-100 at room temperature for 1 hr . Cells were washed with PBS , then incubated in 1 μg/mL DAPI in PBS for 10 min at room temperature , then washed with PBS . For Mesoderm ( BMP4 ) the primary antibodies were Oct4 ( Millipore , Cat#MABD76 ) and Cdx2 rabbit ( Abcam , Cat#ab76541 ) , the secondary antibodies were goat anti-mouse-Alexa 594 , donkey anti-rabbit-Alexa 488 . For endoderm the primary antibody was Sox17 ( R and D Systems , Cat#AF1924 ) , the secondary antibody was donkey anti-goat-Alexa 594 ( Jackson ImmunoResearch ) . Cells were imaged in PBS on a Nikon Ti Eclipse inverted microscope with an Andor Zyla 4 . 2 sCMOS detector . Images were taken as 3 × 3 scan of 20x fields with a 0 . 75 NA objective , stitched with 15% overlap between fields using NIS-Elements Advanced Research Software ( Nikon , Japan ) . Shading correction was applied within the NIS-Elements software before acquiring images . Raw images were quantified using a custom CellProfiler pipeline . RNA lysates were prepared using Norgen Biotek’s Total RNA Purification Kit ( Cat . 37500 ) . Lysates were first treated with Promega RQ1 RNase-Free DNase ( Promega , Madison , WI ) , and then converted to cDNA using Applied Biosystem’s High-Capacity RNA-to-cDNA Kit ( Cat . 4387406 ) . Quantitative real-time PCR ( qPCR ) with SYBR Green ( Bio-Rad; SsoAdvanced Universal SYBR Green Supermix , Cat . 1725271 ) was carried out to assess gene expression . All results were normalized to ACTB . Primers for qPCR were ordered from Eton Bioscience , San Diego , CA . Primers: POU5F1-F: 5'-CCTGAAGCAGAAGAGGATCACC , POU5F1-R 5'-AAAGCGGCAGATGGTCGTTTGG , CDX2-F 5'-ACAGTCGCTACATCACCATCCG , CDX2-R 5'-CCTCTCCTTTGCTCTGCGGTTC , T-F 5'-CTTCAGCAAAGTCAAGCTCACC , T-R 5'-TGAACTGGGTCTCAGGGAAGCA , SOX17-F 5'-ACGCTTTCATGGTGTGGGCTAAG , SOX17-R 5'-GTCAGCGCCTTCCACGACTTG , CDT1-F 5'-GGAGGTCAGATTACCAGCTCAC , CDT1-R , 5'-TTGACGTGCTCCACCAGCTTCT , SOX2-F 5'-CTACAGCATGATGCAGGACCA , SOX2 -R 5'-TCTGCGAGCTGGTCATGGAGT , PAX6-F 5'-AATCAGAGAAGACAGGCCA , PAX6-R 5'-GTGTAGGTATCATAACTC , ACTB-F 5'-CACCATTGGCAATGAGCGGTTC , ACTB-R 5'-AGGTCTTTGCGGATGTCCACGT . ARPE-19-iPS cells were derived from the human retinal pigment epithelial cell line ARPE-19 by reprogramming with CytoTune-iPS 2 . 0 Sendai reprogramming kit ( Invitrogen ) following the manufacturer's instructions . Briefly , two days before Sendai virus transduction , 100 , 000 ARPE-19 cells were plated into one well of a 6-well plate with ATCC-formulated DMEM:F12 medium and were transduced with the CytoTune 2 . 0 Sendai reprogramming vectors at the MOI recommended by the manufacturer 48 hr later ( d0 ) . The medium was replaced with fresh medium every other day starting from one day after transduction ( d1 ) . At day 7 , transduced cells were replated on Matrigel-coated six-well plates . Cells were fed with Essential eight medium every day . Colonies started to form in 2–3 weeks and were ready for transfer after an additional week . Undifferentiated colonies were manually picked and transferred to Matrigel-coated six-well plates for expansion . After two rounds of subcloning and expansion ( after passage 10 ) , RT-PCR was used to verify whether iPS cells were vector-free with the primer sequences published in the manufacturer’s manual . After iPS cells became virus-free , they were submitted to the University of Minnesota Cytogenomic Laboratory for karyotype analysis . This analysis indicated that the ARPE-19-iPS cells have normal karyotypes . To examine pluripotency markers , iPS cells were fixed with 4% paraformaldehyde for 20 min . If nuclear permeation was required , cells were treated with 0 . 2% triton-x-100 in phosphate-buffered saline ( PBS ) for 30 min , blocked in 3% bovine serum albumin in PBS for 2 hr , and incubated with the primary antibody overnight at 4˚C . Antibodies targeting the following antigens were used: TRA1-60 ( MAB4360 , 1:400 ) , TRA1-81 ( MAB4381 , 1:400 ) , stage-specific embryonic antigen-4 ( MAB4304 , 1:200 ) , and stage- specific embryonic antigen-3 ( MAB-4303 , 1:200 ) , all from Millipore/Chemicon ( Billerica , MA ) , OCT3/4 ( AB27985 , 1:200 ) from Abcam ( Cambridge , MA ) , and NANOG ( EB068601:100 ) from Everest ( Upper Heyford , Oxfordshire , UK ) . Cells were incubated with secondary Alexa Fluor Series antibodies ( all 1:500 , Invitrogen ) for 1 hr at room temperature and then with DAPI for 10 min . Images were examined using an Olympus FluoView 1000 m IX81 inverted confocal microscope and analyzed with Adobe Photoshop CS6 . Direct alkaline phosphatase ( AP ) activity was analyzed as per the manufacturer’s recommendations ( Millipore ) . Genomic DNA was isolated using the DNeasy Blood and Tissue kit ( Qiagen , Germany ) per manufacturer’s recommendations for isolation from mammalian cells . Bisulfite conversion was performed using the Epitect Bisulfite kit ( Qiagen ) according to the manufacturer’s protocol for low amounts of DNA . Single-step PCR amplification of the NANOG and OCT4 promoter regions were conducted using Accuprime Supermix II ( Invitrogen ) . Amplification products were visualized by gel electrophoresis and bands were excised and purified using the QIAquick Gel Extraction kit ( Qiagen ) . Purified PCR products were inserted into the PCR4-TOPO vector ( Invitrogen ) and individual clones were sequenced . Alignment and methylation analysis were performed using the online QUMA program ( http://quma . cdb . riken . jp/ ) . Sequenced clones with at least 90% non-CpG cytosine conversion and at least 90% sequence homology were retained for analysis . RNA was isolated using RNeasy Mini kit ( Qiagen ) and treated with TURBO DNA-free ( Ambion , Austin , TX ) . First-strand cDNA was synthesized using a Superscript III First-Strand Synthesis SuperMix ( Invitrogen ) . Reverse transcriptase-PCR was performed using TaqMan Gene Expression Assays and TaqMan Universal PCR Master Mix , No AmpErase UNG ( Applied Biosystems , Carslbad , CA ) as per the manufacturer’s protocol . TaqMan gene expression assays used were OCT4 ( Hs04260367-gH ) , SOX2 ( Hs01053049-sl ) , NANOG ( Hs04399610-g1 ) , KLF4 ( Hs00358836-m1 ) , MYC ( Hs00153408-m1 ) , LIN28 ( Hs00702808-s1 ) , REXO1 ( Hs00810654-m1 ) , ABCG2 ( Hs1053790-m1 ) , DNMT3 ( Hs00171876-m1 ) , with GAPDH ( Hs99999905-m1 ) used as an endogenous control . Expression levels were measured in duplicate . For genes with expression below the fluorescence threshold , the cycle threshold ( Ct ) was set at 40 to calculate the relative expression . Analysis was performed using an ABI PRISM 7500 sequence detection system ( Applied Biosystems ) . ARPE-19-iPS cells contained in a mixture of DMEM/F12 , Matrigel and collagen were implanted onto the hind flank of NSG mice ( n = 5 ) until a palpable mass formed . Teratoma tissue was excised for histological examination following embedding and staining by hematoxylin and eosin . Experiments were conducted with the approval of the Institutional Animal Care and Research Committee at the University of Minnesota . Ergodic Rate Analysis for cell cycle data was based on previously published work ( Kafri et al . , 2013 ) . First , the raw data from flow cytometry files were extracted using FCSExtract Utility ( Earl F Glynn ) to comma separated value ( . csv ) files . The data in the . csv files were then gated in FCS Express 6 ( De Novo Software ) and the data for only G1-MCMDNA were exported . MCM negative cells were excluded based on the negative control sample ( see Flow Cytometry ) . The mean MCM loading rate was calculated in MATLAB ( MathWorks ) . To calculate the mean MCM loading rate , the G1-MCMDNA were subdivided into 10 equal sized bins , with rate calculated for each bin , and all 10 rates were averaged together for a mean MCM loading rate . The rate calculation was based on the formula from Kafri et al: wn=α2−Ffnwn = MCM loading rate in bin n α = ln ( 2 ) /doubling time ( Figure 2—figure supplement 1 ) F = number of G1-MCMDNAcells/total number of cells in sample . F was calculated from FCS Express and entered into MATLAB manually ( Figure 2—figure supplement 1 ) . fn = number of cells in bin n/number of G1-MCMDNA cells The bins were created in MATLAB ( Figure 2—figure supplement 1 ) . To control for small day to day differences in raw data from staining intensities , the histogram edges were defined with the first bin starting at the lowest MCM value and the last bin ending at the highest MCM value , divided into 10 equal sized bins between the lowest and highest MCM value . The 10 wn were then averaged for a final mean w per sample . Sample MATLAB code: alpha_iPSC = log ( 2 ) /15 . 64; F_iPSC_1 = 0 . 0801; %calculate lowest and highest MCM values% maxMCM_iPSC_1 = max ( iPSC_1 ( : , 1 ) ) ; minMCM_iPSC_1 = min ( iPSC_1 ( : , 1 ) ) ; %create histogram with 10 bins and specified first and last bin limits% h10_iPSC_1 = histogram ( iPSC_1 ( : , 1 ) , 'NumBins' , 10 , 'BinLimits' , [minMCM_iPSC_1 , maxMCM_iPSC_1] ) ; %calculate fn within each bin% totalf_iPSC_1 = size ( iPSC_1 ) ; fn10_iPSC_1= ( h10_iPSC_1 . Values ) /totalf_iPSC_1 ( 1 , 1 ) ; %calculate mean w% w10mean_iPSC_1 = mean ( alpha_iPSC . * ( 2> F_iPSC_1 ) . /fn10_iPSC_1 ) ; The mean MCM loading rate was calculated for three biological replicates for each cell line , and the replicates were averaged using GraphPad Prism for further statistical analysis . We cannot use ergodic rate analysis on actively differentiating cells ( e . g . Figure 3 ) because they are not at steady state . Statistical analysis was performed with GraphPad Prism seven using unpaired , two-tailed t test ( displayed as mean ±SD ) or two-tailed Mann-Whitney test as indicated in figure legends . Significance levels were set at p≤0 . 05 , p≤0 . 01 , p≤0 . 001 , **p≤0 . 0001 . All experiments were performed a minimum of two times , and representative data are shown in figures .	From red blood cells to nerve cells , animals’ bodies contain many different types of specialized cells . These all begin as stem cells , which have the potential to divide and make more stem cells or to specialize . All dividing cells must first unwind their DNA so that it can be copied . To achieve this , cells load DNA-unwinding enzymes called helicases onto their DNA during the part of the cell cycle known as G1 phase . Cells must load enough helicase enzymes to ensure that their DNA is copied completely and in time . Stem cells divide faster than their specialized descendants , and have a much shorter G1 phase too . Yet these cells still manage to load enough helicases to copy their DNA . Little is known about how the amount , rate and timing of helicase loading varies between cells that divide at different speeds . Now Matson et al . have measured how quickly helicase enzymes are loaded onto DNA in individual human cells , including stem cells and specialized or “differentiated” cells . Stem cells loaded helicases rapidly to make up for the short time they spent in G1 phase , while differentiated cells loaded the enzymes more slowly . Measuring how the loading rate changed when stem cells were triggered to specialize showed that helicase loading slowed as the G1 phase got longer . Matson et al . found that the levels of key proteins required for helicase loading correlated with the rates of loading . Altering the levels of the proteins changed how quickly the enzymes were loaded and how the cells behaved – for example , slowing down the loading of helicases made the stem cells specialize quicker . These findings show that the processes of cell differentiation and DNA replication are closely linked . This study and future ones will help scientists understand what is happening during early animal development , when specialization first takes place , as well as what has gone wrong in cancer cells , which also divide quickly . A better understanding of this process also helps in regenerative medicine – where one of the challenges is to make enough specialized cells to transplant into a patient with tissue damage without those cells becoming cancerous .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2017	Rapid DNA replication origin licensing protects stem cell pluripotency
We demonstrate that it is feasible to determine high-resolution protein structures by electron crystallography of three-dimensional crystals in an electron cryo-microscope ( CryoEM ) . Lysozyme microcrystals were frozen on an electron microscopy grid , and electron diffraction data collected to 1 . 7 Å resolution . We developed a data collection protocol to collect a full-tilt series in electron diffraction to atomic resolution . A single tilt series contains up to 90 individual diffraction patterns collected from a single crystal with tilt angle increment of 0 . 1–1° and a total accumulated electron dose less than 10 electrons per angstrom squared . We indexed the data from three crystals and used them for structure determination of lysozyme by molecular replacement followed by crystallographic refinement to 2 . 9 Å resolution . This proof of principle paves the way for the implementation of a new technique , which we name ‘MicroED’ , that may have wide applicability in structural biology . X-ray crystallography depends on large and well-ordered crystals for diffraction studies . Crystals are solids composed of repeated structural motifs in a three-dimensional lattice ( hereafter called ‘3D crystals’ ) . The periodic structure of the crystalline solid acts as a diffraction grating to scatter the X-rays . For every elastic scattering event that contributes to a diffraction pattern there are ∼10 inelastic events that cause beam damage ( Henderson , 1995 ) . Therefore , large crystals are required to withstand the high levels of radiation damage received during data collection ( Henderson , 1995 ) . Despite the development of highly sophisticated robotics for crystal growth assays and the implementation of microfocus beamlines ( Moukhametzianov et al . , 2008 ) , this important step remains a critical bottleneck . In an attempt to alleviate this problem , researchers have turned to femtosecond X-ray crystallography ( Chapman et al . , 2011; Boutet et al . , 2012 ) , in which a very intense pulse of X-rays yields coherent signal in a time shorter than the destructive response to deposited energy . While this technique shows great promise , the current implementation of the technology requires an extremely large number of crystals ( millions ) and access to sources is still in developmental stages . Electron crystallography is a bona fide method for determining protein structure from crystalline material but with important differences . The crystals that are used must be very thin ( Henderson and Unwin , 1975; Henderson et al . , 1990; Kuhlbrandt et al . , 1994; Kimura et al . , 1997 ) . Because electrons interact with materials more strongly than X-rays ( Henderson , 1995 ) , electrons can yield meaningful data from relatively small and thin crystals . This technique has been used successfully to determine the structures of several proteins from thin two-dimensional crystals ( 2D crystals ) ( Wisedchaisri et al . , 2011 ) . High energy electrons result in a large amount of radiation damage to the sample , leading to loss in resolution and destruction of the crystalline material ( Glaeser , 1971 ) . As each crystal can usually yield only a single diffraction pattern , structure determination is only possible by merging data originating from hundreds of individual crystals . For example , electron diffraction data from more than 200 individual crystals were merged to generate a data set for aquaporin-0 at 1 . 9 Å resolution ( Gonen et al . , 2005 ) . While electron crystallography has been successful with 2D crystals , previous attempts at using electron diffraction for structure determination from protein 3D crystals were not successful . A number of studies detail the difficulties associated with data collection and processing of diffraction data that originates from several hundreds of 3D crystals , limiting the ability to integrate and merge the data in order to determine a structure in such a way ( Shi et al . , 1998; Jiang et al . , 2011 ) . We show here that atomic resolution diffraction data can be collected from crystals with volumes up to six orders of magnitude smaller than those typically used for X-ray crystallography . The technique , which we call ‘MicroED’ , uses equipment standard in most cryo-EM laboratories and facilities . We developed a strategy for data collection with extremely low electron dose and procedures for indexing and integrating reflections . We processed the diffraction data and determined the structure of lysozyme at 2 . 9 Å resolution . Thus , a high-resolution protein structure can be determined from electron diffraction of three-dimensional protein crystals in an electron microscope . Lysozyme was chosen as a model protein because it is a well-behaved and well-characterized protein that readily forms well-ordered crystals . From the time its structure was first analyzed ( Blake et al . , 1962 , 1965 ) , lysozyme has been a well-studied protein and the model protein of choice for many new methods in crystallography ( Boutet et al . , 2012; Cipriani et al . , 2012; Nederlof et al . , 2013 ) . Small microcrystals of lysozyme were grown by slightly modifying the crystal growth conditions as detailed in the ‘Materials and methods’ section . Figure 1A shows a typical crystallization drop containing microcrystals , which appear as barely visible specks ( arrows ) alongside the larger crystals that are typically used for X-ray crystallography . These specks are up to 6 orders of magnitude smaller in volume than the larger crystals in the drop . The solution containing these microcrystals was applied to an electron microscopy holey-carbon grid with a pipette and plunged into liquid ethane . The grids were then imaged using a 200 kV TEM under cryogenic conditions ( Figure 1B ) . More than 100 microcrystals were typically observed per grid preparation , and these ranged in size from several microns to sub micron . The crystals typically appeared as electron dense rectangular or triangular forms with very sharp edges . 10 . 7554/eLife . 01345 . 003Figure 1 . Images of lysozyme microrystals . ( A ) Light micrograph showing lysozyme microcrystals ( three examples indicated by arrows ) in comparison with larger crystals of the size normally used for X-ray crystallography . Scale bar is 50 μm . ( B ) Lysozyme microcrystals visualized in over-focused diffraction mode on the cryo-EM prior to data collection . The length and width of the crystals varied from 2 to 6 µm with an estimated thickness of ∼0 . 5–1 µm . Scale bar is 1 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 003 Electron diffraction was used to assess the quality of the cryo-preparations . Crystals that appeared thick ( estimated as >3 μm ) did not yield diffraction data because the electron beam could not penetrate the sample . Crystals that appeared slightly thinner , estimated at ∼1 . 5 μm , did show diffraction , but because the quality of the pattern varied depending on the sample tilt ( Figure 2A ) , we did not use crystals of this thickness and size for data collection . Approximately 50% of the crystals in our preparations appeared much thinner , estimated at ∼0 . 5 μm , and showed a distribution of attainable resolutions with the best diffracting to ∼1 . 7 Å resolution ( Figure 2B , C ) . Generally , we were only able to obtain high quality diffraction data from the very thin crystals , ∼0 . 5–1 μm thick and 1–6 µm long and wide . While these crystals are exceptionally small , they still contain approximately 55 × 106 unit cells . Moreover , we found that for such thin crystals the tilt had no significant adverse affect on the diffraction quality ( Figure 2D ) . 10 . 7554/eLife . 01345 . 004Figure 2 . Resolution and data quality of lysozyme microcrystals . ( A ) Analysis of the effects of crystal thickness on maximum resolution of observed reflections from thick crystals . The analysis shows adverse effects of crystal thickness on the obtainable resolution as large crystals are tilted . ( B ) For assessing the quality of our cryo preparations , diffraction data were obtained from 100 lysozyme microcrystals . 43/100 were thin crystals that showed reflections in the 2–4 Å range , with the best crystal in this set yielding data to ∼1 . 7 Å resolution . ( C ) An example of lysozyme diffraction data collected at 0 . 01e−/Å2/second and a 10 s exposure . The pattern shows strong and sharp spots surpassing 2 Å resolution . This diffraction pattern was processed with ImageJ and despeckled for ease of viewing . ( D ) Analysis of the effects of crystal thickness on maximum resolution of observed reflections from thin crystals . The small crystal shows a relatively constant maximum resolution that does not appear to be affected by crystal tilt . DOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 004 For 2D electron crystallography , the electron dose that is typically used in diffraction causes significant radiation damage to the sample , leading to a rapid loss in resolution and destruction of the crystal ( Glaeser , 1971; Unwin and Henderson , 1975; Taylor and Glaeser , 1976 ) . As a result , each crystal exposed to high dose usually only yields a single diffraction pattern , and structure determination requires the merging of data originating from a large number of individual crystals . However , 3D crystals can deliver electron diffraction data to atomic resolution with very low doses . A recent study documents ∼3 Å resolution diffraction data from catalase 3D crystals after a single exposure of less than 10e−/Å2 ( Baker et al . , 2010 ) . We reasoned that one way to overcome the difficulties of indexing and merging data from hundreds of crystals is to collect a complete diffraction data set from a single crystal while keeping the total dose below ∼10e−/Å2 . Because all the data would originate from a single crystal , indexing , integration and merging should be straightforward and structure determination possible . We used a sensitive CMOS based detector ( Tietz Video and Image Processing Systems GmbH ) , previously shown to be beneficial for electron diffraction studies ( Tani et al . , 2009 ) and modified our data collection procedure . We found that even with extremely low electron dose of <0 . 01 e−/Å2 per second , we could record diffraction data from lysozyme microcrystals showing strong and sharp diffraction spots extending well beyond the 2 Å resolution mark ( Figure 2C ) . As a dataset containing multiple exposures from a single crystal is collected , energy transferred by inelastic scattering will damage the crystalline matrix , negatively affecting both the resolution limit and intensities of observed reflections . Although the overall damage from electron scattering is much lower than that for X-rays ( approximately 60 eV deposited per elastic scattering event vs 80 keV per elastic X-ray scattering event [Henderson , 1995] ) , accumulating radiation damage will eventually contribute significant error to the recorded intensities . We performed an experiment to quantify the effects of increasing electron dosage on recorded intensities ( Figure 3 ) . A single protein microcrystal was subjected to sequential 10 s exposures , each delivering ∼0 . 1 e−/Å2 , until a total accumulated dose of ∼12 e−/Å2 was reached . The intensities of three diffraction spots , ranging from resolutions of 2 . 9 to 4 . 6 Å , were measured on each of the 120 resulting diffraction patterns and compared . There were no observable adverse effects on resolution ( Figure 2D ) or intensity until the accumulated dose had reached ∼9 e−/Å2 ( Figure 3 ) . We therefore optimized the data collection protocol to keep the total accumulated electron dosage below this critical value . 10 . 7554/eLife . 01345 . 005Figure 3 . Effects of cumulative electron dose on diffraction data quality . A single lysozyme microcrystal was subjected to 120 sequential exposures without tilting , each of a dose of ∼0 . 1 e−/Å2 for a total accumulated dose of ∼12 e−/Å2 . Normalized intensity vs total accumulated dose for three diffraction spots observed over all 120 sequential frames was plotted . A decrease in diffraction intensity becomes apparent at a dosage of ∼9 e−/Å2 ( ‘critical dose’ ) . Bars represent standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 005 By using such a low dose , we could limit the radiation damage to the crystal , allowing us to collect multiple diffraction patterns from a single crystal instead of just a single pattern . Using this modified procedure , we were able to collect up to 90 individual diffraction patterns from a single crystal ( Video 1 ) . Each pattern was recorded following a 1° tilt to cover ∼40–90° ( begin with the stage tilted at −45° and proceed to collect data to +45° in order to cover a 90° wedge ) . 0 . 1 and 0 . 2 degree increments were also applied to sample the reciprocal space at higher resolution . Each exposure lasted up to 10 s at a dosage of approximately 0 . 01 e−/Å2 per second , for a cumulative dose of no more than ∼9 e−/Å2 per data set . 10 . 7554/eLife . 01345 . 006Video 1 . An example of a complete three-dimensional electron diffraction data set from a single lysozyme microcrystal . In this example , diffraction patterns were recorded at 1° intervals from a single crystal , tilted over 47° . Cumulative dose was ∼5 e−/Å2 in this example . DOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 006 The lattice parameters were determined and the lattice indexed with software based on previous studies ( Shi et al . , 1998 ) . By collecting multiple frames from the same crystal , it was possible to determine the orientation and magnitude of the reciprocal unit cell vectors a* , b* , and c* as described in the ‘Materials and methods’ section . These vectors were calculated for each data set , allowing the prediction of the position of the reflections in each diffraction pattern ( Figure 4; Video 2 ) and indexing of the entire data set . The unit cell dimensions were calculated as a = b = 77 Å , c = 37 Å , α = β = γ = 90° and P43212 symmetry . This space group symmetry and unit cell dimensions are consistent with previous lysozyme X-ray diffraction data ( Diamond , 1974; Sauter et al . , 2001; Cipriani et al . , 2012 ) . 10 . 7554/eLife . 01345 . 007Figure 4 . Prediction of reflections and indexing in the diffraction patterns . ( A and B ) Two examples of diffraction patterns obtained from a single crystal at tilt angles of 0° and 20° respectively . Locations indicated by circles were predicted to contain diffraction spots by our spot prediction algorithm . Additional examples from the same crystal are presented in Video 2 . The resolution limit was set at 2 . 9 Å resolution for this study . DOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 00710 . 7554/eLife . 01345 . 008Video 2 . An example of spot prediction in diffraction data from a single crystal . Reflections predicted on representative diffraction patterns obtained from a single crystal tilted over 39° sampled every 2° in this video . Predictions were made to 2 . 9 Å resolution using our spot prediction algorithm . DOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 008 The electron diffraction data were collected on a microscope operating at 200 kV and equipped with a field emission gun ( FEG ) electron source . The FEG can generate a very coherent beam with an energy-spread function of <1 eV at 200 kV acceleration voltage . The electron beam wavelength is 0 . 025 Å at 200 kV compared with ∼1 Å for X-rays . Under such conditions , the Ewald sphere in our experiments is nearly flat ( the sphere is off the reciprocal plane by only 0 . 003 Å−1 at 2 Å resolution ) even in the high-resolution range . Measurements of the full width at half maximum intensity for the strongest reflections indicate that the reflections in our experiments are very tight , spreading less than a 6 pixel sphere that corresponds to ∼1/1000 Å . ( Figure 5 ) In our experiments , the shortest unit cell dimension for lysozyme in reciprocal space is a* = b* = 1/77 Å . Therefore , without beam oscillation or mechanical oscillation of the crystal ( microscope compustage ) , the lattice points on a single projection that are not exactly at the Ewald sphere surface will give partial intensities . 10 . 7554/eLife . 01345 . 009Figure 5 . Three-dimensional profiles of the intensity of a single reflection over three consecutive diffraction patterns at −0 . 1° , 0° , and 0 . 1° degree tilts . The plots show the approximate dimensions of the full reflection with a width ( full width at half maximum height ) of 3–5 pixels in the x , y , and z direction . DOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 009 Because we densely sampled the reciprocal space , we recorded multiple observations for every lattice point ( Table 1 ) . Therefore , we could sample the observed intensity values for each reflection multiple times ( multiplicity value = 34 ) , and we made the assumption that the strongest intensity roughly approximated the complete intensity . Therefore , we kept only the maximum intensity and treated it as a unique reflection in the final structure factor file . All other recorded intensities were presumed to be partial reflections and were therefore discarded . The merging of data in P422 symmetry from three separate crystals processed in this manner resulted in a final data set with 2490 unique reflections with ∼92% cumulative completeness at 2 . 9 Å resolution ( Table 1 , Video 3 ) . The measured intensities were converted to amplitudes by assuming Ihkl ≈ \|Fhkl\|2 ( Drenth , 1994 ) and an mtz file generated . 10 . 7554/eLife . 01345 . 010Table 1 . MicroED crystallographic dataDOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 010Data collection Excitation voltage200 kV Electron sourceField emission gun Wavelength ( Å ) 0 . 025 Total electron dose per crystal∼9 e−/Å2 Number of patterns per crystal40–90 No . crystals used3 Total reflections to 2 . 9 Å84 , 889Data refinement Space groupP43212 Unit cell dimensions a = b77 Å c37 Å α = β = γ90° Resolution2 . 9–20 . 0 Å Total unique reflections2490 Reflections in working set2240 Reflections in test set250 Multiplicity34 Completeness ( 2 . 9–3 . 1 ) 92% ( 57% ) Rwork/Rfree ( % ) 25 . 5/27 . 8 RMSD bonds0 . 051 Å RMSD angles1 . 587° Ramachandran ( % ) † ( allowed , generous , disallowed ) 99 . 1; 0 . 9; 0Multiplicity is defined as total measured reflections divided by number of unique reflections . †Statistics given by PROCHECK ( Laskowski et al . , 1993 ) . 10 . 7554/eLife . 01345 . 011Video 3 . Three-dimensional representation of merged intensity values . 2490 total unique reflections are present for an overall completeness of 92% at 2 . 9 Å resolution . Video begins with a* axis horizontal , b* axis vertical , and the c* axis normal to the image plane . DOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 011 The structure of lysozyme was solved at 2 . 9 Å resolution by molecular replacement ( MR ) using the lysozyme PDB 4AXT ( Cipriani et al . , 2012 ) as a search model . The initial MR 2Fobs−Fcalc map prior to refinement is presented in Figure 6 . The map shows well-defined density around the model , indicating high quality phases from MR ( Figure 6A , B ) . Likewise , a composite-omit map that was calculated by omitting 5% at a time showed good agreement with the original map obtained by MR ( Figure 6C ) . When a poly-alanine ( polyA ) model of lysozyme was used for MR , the resulting map showed significant density beyond the alanine side chains ( indicated by arrows in Figure 6E , F ) , into which the correct side chains could be built . These results indicated that our solution from MR was not dominated by model bias . 10 . 7554/eLife . 01345 . 012Figure 6 . Results of phasing by molecular replacement prior to crystallographic refinement . Molecular replacement was performed with both the full model of lysozyme ( PDB 4AXT , top panels ) as well as a poly-alanine model ( bottom panels ) and the resulting 2Fobs−Fcalc maps around residues 1–20 are shown . ( A and B ) The phases following molecular replacement with the full model were of good quality demonstrated by how well the density surrounding the model fits , even before any refinement is performed . ( C ) A composite-omit map calculated by omitting 5% at a time showed good agreement with the unrefined structure indicating the phases were not dominated by model bias . ( D–F ) As an additional test of model bias , phasing was done with a poly-alanine homology search model of lysozyme . The resulting 2Fobs−Fcalc map is of good quality ( D ) and shows density extending beyond the poly-alanine model ( E and F , arrows ) . ( F ) The same density map as E but with the structure of lysozyme fit . Arrows in D and E show examples of clear side chain density from the poly-alanine map . All maps are contoured at 1 . 0σ . DOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 012 Following refinement that included the use of electron scattering factors , rigid body , simulated annealing , and B-factor refinement , a solution was found with acceptable statistics ( Rwork/Rfree = 25 . 5%/27 . 8% ) and good geometry at 2 . 9 Å resolution ( Table 1 ) . The density map obtained by electron diffraction shows good agreement with the refined model ( Figure 7A , Video 4 ) . The Fobs−Fcalc difference map shows no interpretable features ( Figure 7B ) . Additionally , the final structure has a very low RMSD ( 0 . 475 Å for Cα , 0 . 575 Å for all atoms ) when compared to the previously published high-resolution structure of lysozyme ( Cipriani et al . , 2012 ) . 10 . 7554/eLife . 01345 . 013Figure 7 . MicroED structure of lysozyme at 2 . 9 Å resolution . ( A ) The 2Fobs−Fcalc ( contoured at 1 . 5σ ) map covers protein residues 5–45 of lysozyme . ( B ) Fobs−Fcalc difference map contoured at +3 . 0σ ( green ) and −3 . 0σ ( red ) for the same protein region . The map ( A ) shows well-defined density around the vast majority of side chains and the difference map ( B ) shows no large discrepancies between the observed data ( Fobs ) and the model ( Fcalc ) . The final structure of lysozyme is shown in panel C and the complete three-dimensional map is presented in Video 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 01310 . 7554/eLife . 01345 . 014Video 4 . 2Fobs−Fcalc density around the complete lysozyme model at 2 . 9 Å resolution ( contoured at 1 . 5σ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 014 To further validate the method and test for model bias , we performed a number of tests on the data to check whether a good solution could be obtained from random noise as has been demonstrated for electron micrographs ( Shatsky et al . , 2009 ) . We created multiple randomized datasets to test the robustness of the phasing and model building procedure . The test datasets were generated as follows:All measured intensities were replaced with random numbers ranging between the minimum and maximum of the actual observed experimental values . The experimental intensity values were kept but the Miller indices were randomized . All experimental intensities were replaced with an actual intensity value that was measured by X-ray crystallography of an unrelated structure ( Calmodulin PDB ID:3SUI [Lau et al . , 2012] ) . Each experimental intensity was increased or decreased randomly by up to 35% . In addition , the correct experimental dataset was also used and labeled as dataset ‘5’ . These five datasets were treated as ‘blind test cases’ , in which the user did not know the identities of the various test datasets . Each test dataset was used for molecular replacement with the lysozyme model ( Cipriani et al . , 2012 ) , followed by a single round of refinement in PHENIX ( Adams et al . , 2010 ) . Only dataset 5 , which contained the correct observed experimental intensities , yielded a solution that could be further refined to acceptable Rwork/Rfree and geometry . Datasets 1–4 , which contained the random errors described above , either did not yield MR solutions or would not allow refinement to produce an acceptable structure ( Table 2 ) . 10 . 7554/eLife . 01345 . 015Table 2 . Results of model validation and bias testsDOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 015Data setMolecular replacement resultTFZFinal Rfree ( % ) ¶1No solutionN/AN/A2†Solution19 . 154 . 93‡No solutionN/AN/A4§Solution12 . 635 . 25#Solution14 . 727 . 8Random intensities . †Shuffled Miller indices . ‡Calmodulin replaced intensities . §Intensities ± 35% . #Original data . ¶Final Rfree after a minimum of two cycles of refinement . *Solution was found; however , the space group was incorrect ( P4121 ) . We also tested the robustness of the MR procedure by using a number of unrelated structures , chosen from the PDB for their similar unit cell dimensions and protein molecular weights , as search models against our experimental data . The unrelated structures were: T4 lysozyme , calmodulin , dodecin , and αA crystallin ( Table 3 ) . None of these structures gave an acceptable MR solution . 10 . 7554/eLife . 01345 . 016Table 3 . Models for molecular replacement validationDOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 016ProteinPDB IDMolecular weight ( kDa ) SymmetryUnit cell dimensionsMR solutionHen Egg White Lysozyme4AXT14 . 3P43212a = b = 78 . 24 ÅYesc = 37 . 47 Åα = β = γ = 90°T4 Lysozyme†2LZM18 . 7P3212a = b = 61 . 20 ÅNoc = 96 . 80 Åα = β = 90°γ = 120°Calmodulin‡3CLN16 . 7P1a = 29 . 71 Å , Nob = 53 . 79 Å , c = 24 . 99 Åα = 94 . 13° , β = 97 . 57° , γ = 89 . 46°Dodecin§4B2J8 . 5F4132a = b = c = 142 . 90 ÅNoα = β = γ = 90°αA Crystallin#3L1E11 . 9P41212a = b = 56 . 22 Å , Noc = 68 . 66 Åα = β = γ = 90°Cipriani et al . ( 2012 ) . †Weaver and Matthews ( 1987 ) . ‡Babu et al . ( 1988 ) . §Staudt et al . ( 2013 ) . #Laganowsky et al . ( 2010 ) . Together , these experiments indicate that the extracted intensities are accurate enough to yield a reliable structure and that model bias originating from MR did not skew our results . Data sets collected in electron crystallography of 2D crystals suffer from a missing cone due to the limitation of the maximum achievable tilt angle in the TEM . Previous reports estimate that with tilt angles up to 60° , the missing cone is roughly 13% ( Glaeser et al . , 1989 ) , and the resolution in plane is typically higher than the resolution perpendicular to the tilt axis ( z ) . In our experiments , because the data from 3 crystals were eventually used , and the orientation of each crystal on the grid varied , we could cover the full reciprocal space ( Video 3 ) . Dynamic scattering likely introduces inaccuracies in the electron diffraction data . In electron diffraction , dynamic scattering ( multi scattering events ) could redistribute primary reflection intensities , reducing the accuracy of the intensity measurements by randomly contributing to the observed intensities ( Grigorieff et al . , 1996 ) . The lysozyme crystals have P43212 , symmetry and systematic absences are expected at ( 2n+1 , 0 , 0 ) . However , very weak reflections were observed at the positions where absences were expected ( Figure 8 ) . It is likely that these reflections originate from dynamic scattering events . We plotted the intensities along the a* and b* axes and compared the intensity values . The intensities of Miller indices ( 2n+1 , 0 , 0 ) and ( 0 , 2n+1 , 0 ) were measured and compared to the intensities of the four immediately adjacent reflections ( 2n+2 , 1 , 0 ) , ( 2n+2 , −1 , 0 ) , ( 2n−2 , 1 , 0 ) , and ( 2n−2 , −1 , 0 ) . On average , the intensity in the systematic absences was found to be 4 . 9% of the total intensity of the adjacent spots . ( Standard deviation 2 . 7% , Max 12 . 4% , n = 17 ) . Moreover , comparison of our experimental intensities with intensities obtained by X-ray diffraction of lysozyme of the same crystal form indicates that our data follow a similar trend and are not dominated by intensity randomness . A Pearson correlation coefficient between the two data sets was 0 . 63 from 6 . 0 to 13 . 5 Å ( 0 . 56 from 2 . 0–13 . 5 Å ) , indicating conservation of reflection hierarchy—strong intensities remain strong and weak intensities remain weak . Together , our analyses suggest that multiple scattering contributes at maximum roughly 10% to the intensity value and that at least for structure determination at 2 . 9 Å resolution such an error in intensity appears to be tolerable . It is possible that dynamic scattering will become a significant source of error at higher resolutions and some correction algorithm will then have to be developed . 10 . 7554/eLife . 01345 . 017Figure 8 . Dynamic scattering in lysozyme 3D crystals . Intensity measurement along the a* axis of a raw diffraction pattern illustrating the relatively small contributions due to dynamic scattering . ( A ) Diffraction pattern from the major plane of a lysozyme crystal with visible intensity in the ( 2n+1 , 0 , 0 ) and ( 0 , 2n+1 , 0 ) Miller indices . ( B ) ( 2n+1 , 0 , 0 ) reflections ( starred ) are expected to be systematically absent and observed intensities at these indices are assumed to be the result of dynamic scattering . Image contrast was enhanced for clarity using ImageJ . DOI: http://dx . doi . org/10 . 7554/eLife . 01345 . 017 We present a method , ‘MicroED’ , for structure determination by electron crystallography . It should be widely applicable to both soluble and membrane proteins as long as small , well-ordered crystals can be obtained . We have shown that diffraction data at atomic resolution can be collected and a structure determined from crystals that are up to 6 orders of magnitude smaller in volume than those typically used for X-ray crystallography . For difficult targets such as membrane proteins and multi-protein complexes , screening often produces microcrystals that require a great deal of optimization before reaching the size required for X-ray crystallography . Sometimes such size optimization becomes an impassable barrier . Electron diffraction of microcrystals as described here offers an alternative , allowing this roadblock to be bypassed and data to be collected directly from the initial crystallization hits . While our proof of principle is an important first step , further optimization of the method is required . Better programs need to be developed for accurately determining lattice parameters , indexing all reflections , extracting the intensities and correcting for incomplete intensities , dynamic scattering , and Ewald sphere curvature . Specifically , developing procedures for postrefinement ( unit cell refinement , estimating the mosaic spread , rocking curve , etc ) should allow for the proper correction and scaling of partially recorded reflections , leading to improved estimation of full intensities . Relatively minor modifications to existing programs such as MOSFLM ( Leslie and Powell , 2007 ) should allow the handling of electron diffraction data from 3D crystals and take advantage of the large body of work already dedicated to processing X-ray diffraction data . The accuracy of the microscope compustage can be improved and procedures for crystal or beam oscillation implemented . Our method of using the maximum intensity measurement as an approximation of the full intensity of any given spot is admittedly crude , as it depends on the intersection of the Ewald sphere through the center of each spot at some point in the tilt series . As the resolution increases , this event becomes increasingly unlikely . Crystal oscillation or related methods such as precession of the electron beam ( Gjønnes et al . , 1998 ) would allow more accurate determination of spot intensities , especially at very high resolutions . Further development of various methods for phasing the diffraction data are also required and could possibly include heavy metal phasing . Such phasing methods are standard in X-ray crystallography and rely on differences in intensity values between a native data set and heavy metal derivative data sets . It is possible that in electron crystallography dynamic scattering could hinder phasing by such methods , and new algorithms will need to be developed to make this possible . Phase extension from projection maps or from low-resolution density maps can also be used for direct phasing ( Gipson et al . , 2011; Wisedchaisri and Gonen , 2011 ) . It is also possible that single particle cryo-EM could be used for direct phasing as previously demonstrated where a low-resolution single particle map was used to phase X-ray diffraction data ( Speir et al . , 1995; Dodson , 2001; Xiong , 2008 ) . Moreover , a double tilt cryo holder as well as newly developed goniometer-based grid holders could be used to cover more of the Fourier space . Finally this method could benefit from automation in data collection . This first study serves as a proof of principle that three-dimensional electron diffraction can yield an accurate protein structure from microcrystals . As additional protocols and programs are developed , MicroED promises to advance the field of structural biology and open the door to many exciting new studies . Lysozyme was purchased from Fisher Scientific and a 200 mg/ml solution was prepared in 50 mM sodium acetate pH 4 . 5 . Lysozyme solution was mixed 1 to 1 with precipitant solution ( 3 . 5M sodium chloride; 15% PEG 5 , 000; 50 mM sodium acetate pH 4 . 5 ) and crystals were grown by the hanging drop method . Following the crystal formation , the sample was diluted three to five times in 5% PEG 200 . A 5 μl drop of the crystal solution was applied to a quantifoil 2/2 holey-carbon copper EM grid . The grid was then blotted and vitrified by plunging into liquid ethane using a Vitrobot Mark IV ( FEI ) . The frozen-hydrated grid was loaded onto a Gatan 626 cryo-holder and transferred to a cryo-TEM . All electron microscopy was performed on a FEI Tecnai F20 TEM equipped with a field emission electron source ( FEG ) and operating at 200 kV . Electron diffraction pattern tilt series data were recorded with a bottom mount TVIPS F416 4 k × 4 k CMOS camera with pixel size 15 . 6 μm using built in series exposure mode . The electron dose was kept below 0 . 01 e−/Å2 per second , and each frame of a data set was taken with an exposure time of up to 10 s per frame . The electron dosage was calibrated with the use of a Faraday cage as well as by calibrating the counts on the CMOS detector in bright field mode . Each data set consisted of up to 90 still frames taken at 0 . 1–1° intervals with a maximum total dose of ∼9e−/Å2 per crystal . The camera length was optimized for the desired resolution as described previously ( Gonen , 2013 ) . Although our original intent was to perform all data analysis with existing X-ray crystallography software various incompatibilities and logistical roadblocks necessitated the development of some additional tools . Diffraction patterns were indexed and background subtracted intensities extracted and merged with in-house developed software implemented in python using methods adapted from those developed by Shi et al . ( 1998 ) . Briefly , measurements were made on images identified as major planes of the crystal with ImageJ and used to determine the approximate magnitudes of the unit cell vectors a* , b* , and c* and the angles between them ( α , β , and γ ) . Subsequently , 100 to 350 spots were chosen across several images from each set of diffraction patterns . Vectors in reciprocal space were calculated for all of the selected spots . Difference vectors between spot vectors were calculated allowing vectors approximating the estimated unit cell lengths to be identified . The angles between potential unit cell vectors were calculated and ‘orthogonal triplets’ identified . Orthogonal triplets are defined as sets of vectors that contain a predicted a* , b* , and c* , which are all 90° from each other ( α = β = γ = 90° for this crystal ) . All sets of the orthogonal triplets were averaged to yield estimated a* , b* , and c* vectors . The estimated a* , b* , and c* vectors were then refined by identifying parallel difference vectors derived from the original selected spots with lengths that were multiples of the unit cell lengths . The calculated unit cell vectors were then used to predict the spots in each diffraction pattern . Two reference spots were chosen for each image and their Miller indices calculated using the previously determined unit cell vectors . For every diffraction pattern , the vector normal to the detector plane was calculated as:r1× r2=nwhere r1 and r2 are the vectors defined by the Miller indices from reference spots one and two , respectively , and n is the resulting vector normal to the detector plane . Any reflection that appears on a given diffraction pattern will satisfy:n·v=0where v is any set of Miller indices . For any h , k , l that satisfied the above equation , within a defined threshold , that particular reflection was predicted to appear on the diffraction image , and its x , y detector coordinates on the diffraction pattern image were calculated . Intensities for each predicted reflection were integrated by first drawing both a square and a circular mask centered on the reflection , with the diameter of the circle identical to the length of the square . The mean pixel intensity outside the circle but within the square was calculated yielding the mean background intensity . The mean background was then subtracted from each pixel within the circle , and the resulting pixel intensities were summed . All related intensities from three data sets were grouped based on P422 symmetry . The maximum value for each group of equivalent reflections was assumed to best approximate the full intensity and was used for that reflection in the final data set . Because each intensity measurement ultimately originated from a single observation , SigI and SigF values were estimated as the square root of the intensity and square root of the structure factor , respectively . The final mtz file contains columns h , k , l , F , SIGF , I , SIGI . The final data set contained 2490 unique reflections from 2 . 9–20 Å with cumulative completeness of 92% ( Table 1 ) . Phaser ( McCoy et al . , 2007 ) was used to obtain phases with lysozyme structure 4AXT ( Cipriani et al . , 2012 ) as a MR search model ( LLG = 372 and TFZ = 14 . 7 ) . The structure was then refined using CNS ( Brünger et al . , 1998 ) and PHENIX ( Adams et al . , 2010 ) by rounds of rigid body , simulated annealing , and B-factor refinement . The Rfree data set represented 10% of the total data set . The data were subjected to twinning analysis; however , twinning with this symmetry group is forbidden and therefore we ruled out twinning in our crystals . Electron scattering factors ( Gonen et al . , 2005 ) were used during refinement . The structure factors and coordinates of the final model were deposited in the Protein Data Bank with accession code 3J4G . The in house developed program that was used for processing the MicroED data is available for download at http://www . github . com/gonenlab/microED . git .	X-ray crystallography has been used to work out the atomic structure of a large number of proteins . In a typical X-ray crystallography experiment , a beam of X-rays is directed at a protein crystal , which scatters some of the X-ray photons to produce a diffraction pattern . The crystal is then rotated through a small angle and another diffraction pattern is recorded . Finally , after this process has been repeated enough times , it is possible to work backwards from the diffraction patterns to figure out the structure of the protein . The crystals used for X-ray crystallography must be large to withstand the damage caused by repeated exposure to the X-ray beam . However , some proteins do not form crystals at all , and others only form small crystals . It is possible to overcome this problem by using extremely short pulses of X-rays , but this requires a very large number of small crystals and ultrashort X-ray pulses are only available at a handful of research centers around the world . There is , therefore , a need for other approaches that can determine the structure of proteins that only form small crystals . Electron crystallography is similar to X-ray crystallography in that a protein crystal scatters a beam to produce a diffraction pattern . However , the interactions between the electrons in the beam and the crystal are much stronger than those between the X-ray photons and the crystal . This means that meaningful amounts of data can be collected from much smaller crystals . However , it is normally only possible to collect one diffraction pattern from each crystal because of beam induced damage . Researchers have developed methods to merge the diffraction patterns produced by hundreds of small crystals , but to date these techniques have only worked with very thin two-dimensional crystals that contain only one layer of the protein of interest . Now Shi et al . report a new approach to electron crystallography that works with very small three-dimensional crystals . Called MicroED , this technique involves placing the crystal in a transmission electron cryo-microscope , which is a fairly standard piece of equipment in many laboratories . The normal ‘low-dose’ electron beam in one of these microscopes would normally damage the crystal after a single diffraction pattern had been collected . However , Shi et al . realized that it was possible to obtain diffraction patterns without severely damaging the crystal if they dramatically reduced the normal low-dose electron beam . By reducing the electron dose by a factor of 200 , it was possible to collect up to 90 diffraction patterns from the same , very small , three-dimensional crystal , and then—similar to what happens in X-ray crystallography—work backwards to figure out the structure of the protein . Shi et al . demonstrated the feasibility of the MicroED approach by using it to determine the structure of lysozyme , which is widely used as a test protein in crystallography , with a resolution of 2 . 9 Å . This proof-of principle study paves the way for crystallographers to study protein that cannot be studied with existing techniques .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "structural", "biology", "and", "molecular", "biophysics" ]	2013	Three-dimensional electron crystallography of protein microcrystals
Macrophages are myeloid-derived phagocytic cells and one of the first immune cell types to respond to microbial infections . However , a number of bacterial pathogens are resistant to the antimicrobial activities of macrophages and can grow within these cells . Macrophages have other immune surveillance roles including the acquisition of cytosolic components from multiple types of cells . We hypothesized that intracellular pathogens that can replicate within macrophages could also exploit cytosolic transfer to facilitate bacterial spread . We found that viable Francisella tularensis , as well as Salmonella enterica bacteria transferred from infected cells to uninfected macrophages along with other cytosolic material through a transient , contact dependent mechanism . Bacterial transfer occurred when the host cells exchanged plasma membrane proteins and cytosol via a trogocytosis related process leaving both donor and recipient cells intact and viable . Trogocytosis was strongly associated with infection in mice , suggesting that direct bacterial transfer occurs by this process in vivo . All intracellular pathogens enter and replicate inside some type of host cell . At the earliest stage of disease only a limited number of host cells will be infected . In order to successfully continue propagation intracellular pathogens must continually infect new susceptible cells . Most of these organisms are thought to infect a cell , replicate , re-enter the extracellular space and start the process over again . However , re-entering the extracellular space exposes the pathogen to antibodies , complement , and other extracellular antimicrobial factors that can inhibit their growth or block their entry into new cells . It is therefore not surprising that certain intracellular pathogens have evolved mechanisms to transfer directly from infected to uninfected cells . The majority of intracellular bacterial pathogens that are known to transfer directly from cell to cell do so through a process known as actin based motility . While there are modest variations in the specific mechanisms employed by individual species , in general the process is pathogen driven through the expression of effector proteins that nucleate and polymerize host cell actin in a manner that physically propels the bacteria into a neighboring cell ( Ireton , 2013 ) . There are , however , natural host cell processes that transfer cytosolic material that could be exploited by intracellular pathogens to facilitate direct cell to cell spread . Many recent studies have demonstrated that host cells can exchange cytosolic or membrane materials with neighboring cells through contact-dependent mechanisms ( Joly and Hudrisier , 2003; Rogers and Bhattacharya , 2013 ) . The exchange of cytosolic components occurs in different contexts across a wide range of distinct cells types , and there are several morphologically distinct mechanisms that exchange cytosolic material , including nanotubes , gap junctions , cytonemes and synapses ( Onfelt et al . , 2006; Rogers and Bhattacharya , 2013; Kanaporis et al . , 2011; Roy et al . , 2014 ) . The different exchange mechanism morphologies are associated with the transfer of specific types of material . For example , gap junctions are selectively permeable to ions and small molecules while nanotubes can transfer functional organelles from a donor to a recipient cell ( Onfelt et al . , 2006; Kanaporis et al . , 2011 ) . Certain viral pathogens are known to transfer directly from cell to cell by exploiting one or more of these natural cellular processes . For example , human immunodeficiency virus ( HIV ) transfers between cells via tunneling nanotubes ( Sowinski et al . , 2008 ) , whereas Human T-lymphotophic virus ( HTLV-1 ) can spread directly from infected to uninfected T-cells through virological synapses ( Igakura et al . , 2003 ) . The exchange of plasma membrane proteins between eukaryotic cells occurs through a mechanism termed trogocytosis ( trogo = Greek for nibble ) ( Joly and Hudrisier , 2003 ) . For trogocytosis to occur two cells form a transient intimate interaction during which the membranes appear to fuse . The cells eventually separate , with each participant cell having acquired plasma membrane components from the partner cell . The transferred membrane proteins retain their orientation and their function until they are recycled via normal membrane turnover . In certain mouse tissues , over half of the cells have undergone detectable trogocytosis at any given time ( Yamanaka et al . , 2009 ) . In immune cells , trogocytosis leads to a variety of acquired functions that likely impact infection and immunity . For example , trogocytosis improves T cell signaling in response to antigens and dendritic cells can activate T cells after acquiring antigens from neighboring cells ( Osborne and Wetzel , 2012; Rosenits et al . , 2010; Wakim and Bevan , 2011 ) . Trogocytosis has been implicated as a critical factor in several pathologies including cancer biology , tissue engraftment , and vaccination efficacy ( Li et al . , 2012; Chow et al . , 2013; Chung et al . , 2014; Zhang et al . , 2008 ) . Trogocytosis can occur without the transfer of cytosolic material ( Puaux et al . , 2006 ) , but it is unclear if the presumptive transient membrane fusion that occurs during certain types of cytosolic transfer also results in trogocytosis . Although mammalian cells exchange intracellular and membrane material , there is a major gap in our knowledge about how these transfer mechanisms impact the infectious process of intracellular pathogens . Foreign material including beads and Mycobacterium bovis have been shown to transfer directly between macrophages ( Onfelt et al . , 2006 ) . But it is unclear how prevalent these transfer events are , how they influence pathogenesis , and if these transfer events benefit the pathogen ( through cell to cell spread ) , the host ( via immune detection or pathogen destruction upon transfer ) , or some combination of each . These questions are important because direct cell to cell transfer via cytosolic exchange could be a critical part of the infectious life-cycle for certain intracellular pathogens . For example , an estimated 60% of cells infected by HIV occur through direct viral transfer ( Iwami et al . , 2015 ) . To investigate how cytosolic transfer affects intracellular pathogens , we used the macrophage-tropic , facultative intracellular bacterium Francisella tularensis as a model pathogen . Importantly , F . tularensis has not been shown to transfer between cells and lacks homologs of proteins that other bacteria use to transfer between cells . But the rapid spread of F . tularensis to new cells and cell types during infection suggests that direct cell to cell transfer may occur ( Hall et al . , 2008; Roberts et al . , 2014; Lindemann et al . , 2011 ) . Here , we demonstrate that live F . tularensis bacteria transfer directly from infected cells to macrophages via a contact and cytosolic exchange dependent mechanism . Direct bacterial transfer appears to occur frequently both in vitro and in a mouse infection model . Bacterial transfer was cell type specific and correlated strongly with trogocytosis , specifically the exchange of functional major histocompatibility complex I ( MHC-I ) . Lastly , we observed similar transfer events during infections with Salmonella enterica or fluorescent beads , suggesting that trogocytosis-associated cell to cell transfer may be a commonly exploited phenomenon . Francisella tularensis is a highly infectious zoonotic bacterial pathogen that is capable of invading and replicating in numerous cell types including , but not limited to , epithelial cells and macrophages ( Hall et al . , 2008 ) . In a mouse model of infection the number of Francisella infected cells increases dramatically over a short period of time ( Hall et al . , 2008; Roberts et al . , 2014 ) . This result suggested to us that F . tularensis could spread directly from infected to uninfected cells . To test this hypothesis we monitored GFP- F . tularensis infected J774A . 1 macrophage-like ( J774 ) cells by live cell imaging . We found that the bacteria transferred from infected to uninfected macrophages upon cell to cell contact ( Video 1 and 2 , Figure 1A ) . After bacterial transfer , both donor and recipient macrophages were typically motile following separation suggesting that both cells remained viable after bacterial transfer ( Videos 1 and 2 , Figure 1A ) . From these results , we concluded that F . tularensis bacteria could transfer directly between J774 cells without entering the extracellular space . 10 . 7554/eLife . 10625 . 003Video 1 . F . tularensis bacteria transfer from infected macrophage to neighboring cells . Time lapse video of an F . tularensis infected J774 macrophage ( top middle in opening frame ) that migrates to neighboring cells , infects those macrophages , and then migrates away . F . tularensis is depicted in green , bright field in red . The time is hours: minutes post inoculation . Images were acquired every 5 min . DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 00310 . 7554/eLife . 10625 . 004Video 2 . F . tularensis bacteria transfer from infected macrophage to neighboring cells . Time lapse video from an experiment separate from Video 1 depicting an F . tularensis infected J774 macrophage migrating toward neighboring cells , infects those macrophages , and then migrating away . F . tularensis is depicted in green , bright field in red . The time is hours: minutes post inoculation . Images were acquired every 5 min . DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 00410 . 7554/eLife . 10625 . 005Figure 1 . F . tularensis transfers between macrophages during cytosolic transfer . ( A ) Representative panels from live cell imaging of F . tularensis infected J774 cells transferring bacteria . Time- hour: minutes post inoculation; * - initially infected cell; White arrow- first bacterial transfer event; Orange arrow- second bacterial transfer event . Movie available as Video 1 . ( B ) The proportion of recipient macrophages infected after a 6 hr co-incubation with infected cells of the same type ( 3 independent experiments performed in triplicate ) . ( C ) A representative histogram of the amount of calcein that transferred to recipient cells ( log10 fluorescence ) after 6 hr co-incubation . ( D ) The percent of infected or uninfected cells that exchanged cytosolic content ( positive for both Cell Trace Red and calcein ) after 6 hr co-incubation . The uninfected population represents cells in the infected well that did not become infected . DC refers to a doublet control ( 2 independent experiments performed in triplicate ) ( E ) Bacterial transfer to uninfected cells is significantly higher with direct cell to cell contact . Infected BMDMs on a transwell filter were suspended over uninfected BMDMs . The percent of total cells infected on the bottom chamber ( bottom ) and top filter ( transwell ) were determined by FACS 6 and 24 hr after suspending the transwell over uninfected cells . Side brackets indicate the change in numbers of infected cells in each chamber from 6 to 24 hr . Transfer to BMDMs separated from the initially infected cells was significantly lower than transfer to BMDMS in contact with the infected cell population . ( 3 independent experiments performed in triplicate ) . ( Mean +/- SD ) . ( *p<0 . 001 ) DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 00510 . 7554/eLife . 10625 . 006Figure 1—figure supplement 1 . The extracellular space is not a major source of infectious bacteria . ( A ) The total number of BMDMs infected at 6 hr intervals compared to the number of extracellular bacteria in 1 milliliter of media . All samples were initially treated with gentamicin between 2 and 6 hr post inoculation to destroy extracellular bacteria from the inoculum . The antibiotic containing media was replaced with antibiotic free media at 6 hr post infection . ( B ) BMDMs were assessed for the number of infected cells at 6 or 24 hr post inoculation in the presence of specific inhibitors . Z-Vad ( OMe ) -FMK or necrostatin-1 were added at 6 hr post inoculation . ( All experiments from 3 independent experiments performed in triplicate ) ( Mean +/- SD ) . ( ND- not determined , bacterial counts below limit of detection , ns p>0 . 05 , p<0 . 05 , p<0 . 01 , p<0 . 001 ) DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 00610 . 7554/eLife . 10625 . 007Figure 1—figure supplement 2 . Experimental design and bacterial motility for transwell assay . ( A ) Experimental design of the transwell assay performed in Figure 1E . ( B ) Extracellular F . tularensis can penetrate and infect BMDMs on both sides of the transwell membrane . F . tularensis was added to the indicated chamber for 2 hr and both chambers were assessed for the number of BMDMs that became infected . ( 3 independent experiments performed with a single replicate ) ( Mean +/- SD ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 007 We next developed a flow cytometry assay to quantify the transfer event to determine if the direct transfer of bacteria from cell to cell that we observed by live cell imaging occurred at a sufficient frequency to be biologically relevant . In these experiments , we infected cells with F . tularensis , added the antibiotic gentamicin to kill extracellular bacteria and allowed the intracellular bacteria to proliferate for 18 hr . We then added uninfected recipient cells labeled with Cell Trace Red to these infected cells . The mixed cell population was co-cultured for 6 hr in the presence of gentamicin and then quantified infected recipient cells by flow cytometry based on double staining for Cell Trace Red and intracellular bacteria . Under these conditions F . tularensis transferred from infected to uninfected recipient J774 cells , mouse bone marrow derived macrophages ( BMDMs ) and primary human monocyte derived macrophages ( hMDMs ) ( Figure 1B ) . To validate that these transfer events occurred through direct cytosolic exchange rather than from extracellular bacteria , we tracked the transfer of F . tularensis with the cytosolic dye calcein-AM which becomes membrane impermeable after entering the cell . F . tularensis transfer strongly correlated with the transfer of the cytosolic dye between cells ( Figure 1C , D ) . These data indicate that the majority of newly infected cells were infected through the exchange of cytosolic material rather than from extracellular bacteria . It is important to note that there are very few extracellular F . tularensis bacteria in media containing gentamicin . We found that the number of cells infected at each tested interval was significantly higher than the total number of extracellular bacteria in a milliliter of media ( Figure 1—figure supplement 1A ) . Further decreasing the number of extracellular bacteria by inhibiting infected cell lysis through apoptosis or necrosis had no detectable effect on the number of cells infected ( Figure 1—figure supplement 1B ) . These results further support our conclusion that most of the newly infected macrophages become infected via cell to cell spread . The live cell images suggested that F . tularensis was transferred upon cell to cell contact with no obvious infection of new cells through bacteria – containing exosomes . We verified that cell contact dependent transfer was the predominant method for new cells to become infected in the following experiment . We compared the amount of cells that became infected when the cells could come into physical contact with cells that were physically separated by a bacteria permeable membrane . After 12 hr of co-incubation with no antibiotics in the media , there was a roughly 17% increase in infected cells that were able to physically touch compared to a roughly 4% increase in infected cells that were separated by the membrane ( Figure 1E , Experimental design in Figure 1—figure supplement 2A ) . The reciprocal set-up yielded similar results ( data not shown ) . From these results , we conclude that the majority of bacterial transfer events that we observed occurred from contact-dependent transfer . F . tularensis bacteria transferred between cells , but it is unclear if the transferred bacteria were viable or could sustain growth in the newly infected cells . To assess bacterial viability after transfer , we permeabilized the host cell and measured bacterial viability by propidium iodide exclusion . The percent of viable bacteria was similar between the donor and recipient populations , indicating that bacteria were not killed during transfer between cells ( Figure 2A , B and Figure 2—figure supplement 1 ) . 10 . 7554/eLife . 10625 . 008Figure 2 . Live bacteria transfer to macrophages during bacterial transfer . ( A ) The percent of viable bacteria ( propidium iodide negative ) in donor and recipient BMDMs ( 2 independent experiments , 50 fields of view each ) ( B ) Micrographs of propidium idodide treated permeabilized F . tularensis infected BMDMs . Arrow- propidium iodide positive bacterium . Scale bar- 10 uM . ( C ) The number of cells infected in untreated or soy lecithin treated BMDMs . Infected BMDMs in soy lecithin treated ( grey ) and untreated ( black ) were quantified by FACS at indicated times post inoculation . Data are presented as number of infected cells regardless of number of bacteria per infected cell . Soy lecithin was added to the treated populations after initial infection with F . tularensis . ( D ) Soy lecithin does not inhibit F . tularensis intracellular replication . The number of viable bacteria in untreated or soy lecithin treated BMDMs was quantified at indicated times by dilution plating and calculation of colony forming units . Lecithin was added at 6 hr post inoculation ( 3 independent experiments performed in triplicate for both lecithin experiments ) . ( Mean +/- SD ) . ( ns p>0 . 05 , p<0 . 05 , p<0 . 01 , p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 00810 . 7554/eLife . 10625 . 009Figure 2—figure supplement 1 . Propidium iodide can access and bind to dead intracellular bacteria following saponin treatment . Fluorescence microscopy images of BMDMs infected with ( A–C ) wild type Schu S4-GFP or ( D–F ) intracellular growth defective mutant △FTT_0924-GFP , permeabilized with saponin and treated with propidium iodide ( PI ) . ( A , D ) GFP positive bacteria ( green ) , ( B , E ) PI positive bacteria ( white ) , ( C , F ) merged images . ( G ) Proportion of PI positive ( dead ) intracellular wild type ( WT ) and growth deficient mutant bacteria ( △FTT_0924 ) . The data were analyzed from n = 100 macrophages per strain from 3 independent experiments . ( Mean +/- SD ) ( *P<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 009 The best way to accurately assess the contribution of cell to cell transfer on the overall intracellular proliferation of F . tularensis would be to compare bacterial growth between conditions that permit and inhibit cell to cell transfer . We therefore screened numerous membrane altering factors for an inhibitor that blocked bacterial transfer and found that the addition of soy lecithin after infecting BMDM effectively blocked the transfer of F . tularensis to uninfected cells ( Figure 2C ) . To test if the transferred bacteria could sustain infection , we infected approximately 1% of a BMDM population then added soy lecithin to inhibit cell to cell contact dependent bacterial transfer . We monitored bacterial viability over 3 days . Cells that cannot directly transfer bacteria will lyse after peak infection ( ~24 hr ) , releasing their bacteria into antibiotic containing media . If the bacteria in the untreated cells survive transfer and are able to proliferate , bacterial viability should be higher in these samples than samples treated with soy lecithin at time points after peak infection . We found that both wild-type and transfer-inhibited cells reached peak infection at 24 hr post inoculation , but the lecithin treated samples had significantly fewer viable bacteria compared to untreated samples at 48 and 72 hr post inoculation ( Figure 2D ) . Thus , F . tularensis exploits cell to cell transfer to extend infection by invading and replicating in previously uninfected cells without entering the extracellular space . Many bacterial species transfer from cell to cell through bacteria mediated processes , such as actin based motility . These bacteria can spread between several different host cell types because the transfer mechanisms are driven by bacterial effectors , ( Tilney et al . , 1990; Makino et al . , 1986; Heinzen et al . , 1993 ) . To address if F . tularensis transfer occurs through a host or bacterial mediated process , we tested F . tularensis transfer between different host cell types . Specifically , we compared bacterial transfer between TC-1 epithelial cells , bacterial transfer between macrophages , and bacterial transfer from TC-1 epithelial cells to macrophages . Although F . tularensis replicates well in TC-1 epithelial cells ( Fuller et al . , 2008 ) , F . tularensis did not detectably transfer from infected to uninfected TC-1 cells ( Figure 3A , B ) . However , when we added uninfected BMDMs to the infected TC-1 cells , the BMDMs became infected ( Figure 3C ) . The number of infected epithelial cells did not change when BMDMs were added , indicating that the bacteria did not transfer from BMDMs to uninfected epithelial cells ( Figure 3C ) . These data indicate that F . tularensis transfer is limited to specific recipient cell types and suggest that F . tularensis transfer is likely a host mediated process . 10 . 7554/eLife . 10625 . 010Figure 3 . Bacterial transfer is cell type specific . ( A ) Percentage of infected J774 macrophages or TC-1 epithelial cells at the indicated time post inoculation ( log10 fluorescence ) . ( B ) A compilation of the number of J774 or TC-1 cells infected over time . Statistics represent test for a significant increase in the number of cells infected compared to the previous 6 hr time point . ( C ) TC-1 to TC-1 transfer vs TC-1 to BMDM transfer after a 0 or 18 hr co-incubation . ( All results from 3 independent experiments performed in triplicate ) ( Mean +/- SD ) . ( ns p>0 . 05 , p<0 . 05 , *p<0 . 01 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 01010 . 7554/eLife . 10625 . 011Figure 3—figure supplement 1 . F . tularensis does not transfer via actin based motility or autophagy . Representative micrographs of ( A ) uninfected , ( B ) F . tularensis infected , or ( C ) Listeria monocytogenes infected BMDMs 16 hr post inoculation . DAPI is depicted in blue , phalloidin ( actin ) in red and the bacteria in green . The scale bar represents 10 uM . ( D ) Transfer of bacteria to recipient cells in the presence of host ( cycloheximide [CHX] ) or bacteria ( chloramphenicol [Chlor] ) protein synthesis inhibitors . ( E ) F . tularensis transfer to recipient BMDMs after a 6 hr co-incubation in the presence or absence of 3-methyladenine ( 3 MA ) . ( All experiments from 3 independent experiments performed in triplicate ) ( Mean +/- SD ) . ( ns p>0 . 05 , p<0 . 05 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 011 Recipient cell type specificity suggests that F . tularensis does not use the transfer mechanisms described in other bacterial pathogens . We further tested this conclusion by comparing F . tularensis transfer to other known bacterial mechanisms . We found that F . tularensis did not form actin tails that are characteristic of bacterial pathogens such as L . monocytogenes , suggesting that F . tularensis does not use actin based motility ( Figure 3—figure supplement 1A–C ) . Likewise , actin based motility requires continual bacterial protein synthesis ( Tilney et al . , 1990 ) ; but , bacterial protein synthesis was not required for F . tularensis transfer ( Figure 3—figure supplement 1D ) . A proposed alternative form of F . tularensis spread is through an autophagy related mechanism termed the Francisella containing vacuole ( Checroun et al . , 2006; Starr et al . , 2012 ) , but inhibiting autophagy with 3-methyladenine ( 3MA ) or using ATG5 knockout BMDMs did not block bacterial transfer ( Figure 3—figure supplement 1E , data not shown ) . Altogether these data are consistent with F . tularensis exploiting host-mediated cytosolic transfer for cell to cell spread . One mechanism for cytosolic exchange observed in cytotoxic T cells ( CTL ) occurs when pores connecting the cytosol form between the CTL and the target cell ( Stinchcombe et al . , 2001 ) . During this cytosolic intermingling , the cells also exchange specific plasma membrane proteins ( Stinchcombe et al . , 2001 ) . The cell to cell exchange of intact and functional plasma membrane proteins that retain their orientation is termed trogocytosis ( Joly and Hudrisier , 2003 ) . We noted a similar phenomenon of plasma membrane transfer following bacterial transfer . Newly infected recipient BMDMs frequently acquired plasma membrane proteins as well as cytosolic material from the initially infected cell ( Figure 4A , B ) . Interestingly , transferred plasma membrane proteins retained their orientation; so membrane proteins that were surface exposed on the initially infected cell were also surface exposed on the newly infected recipient cell . In the presented images , the cells were stained with a biotin succinimidyl ester prior to mixing the cells with differentially labelled BMDMs . The mixed population was then labeled with a fluorescent conjugated streptavidin immediately before fixation . As a result , the protein must be surface exposed before and after transfer to be labelled ( Figure 4A , B ) . These data are consistent with trogocytosis and imply that trogocytosis occurs at the same time as bacterial transfer . Importantly , these results indicate that trogocytosis can be used as a marker for bacterial transfer and differentiate direct F . tularensis transfer from more conventional infection mechanisms such as actin based motility or reinfection by extracellular bacteria . 10 . 7554/eLife . 10625 . 012Figure 4 . Plasma membrane protein transfer correlates with bacterial transfer . ( A ) Fluorescence micrographs of BMDMs before , during , and after trogocytosis . ( B ) A donor [white plasma membrane] and trogocytosis positive recipient BMDM [red cell] exchanging cytosolic material and bacteria . The bottom right panel is a 3D rendering of the Z-stack from the cells in the top panel . Percent of trogocytosis positive recipient cells that are ( C ) BMDMs or ( D ) hMDMs . ( E ) The percent of Balb/c recipient BMDMs that acquired SIINFEKL peptide bound MHC-I from B6 BMDMs . ( F ) The percent of infected splenocytes that underwent trogocytosis in a mouse infection model ( 8 or 9 mice per group from 4 independent experiments ) . DC refers to a doublet control . ( All tissue culture data are from 3–4 independent experiments performed in triplicate ) ( Scale bar- 10 um ) ( Mean +/- SD ) . ( ns p>0 . 05 , p<0 . 05 , p<0 . 01 , *p<0 . 001 ) DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 01210 . 7554/eLife . 10625 . 013Figure 4—figure supplement 1 . Plasma membrane protein exchange increases during infection . ( A ) The percent of Balb/c recipient BMDMs that acquired B6 MHC-I from F . tularensis infected or uninfected donor B6 BMDMs . ( 3 independent experiments performed in triplicate ) ( B ) The percent of splenocytes that exchanged MHC-I in infected or uninfected mice ( 8 or 9 animals per group from 4 independent experiments ) . ( Mean +/- SD ) . ( p<0 . 05 , **p<0 . 001 ) DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 01310 . 7554/eLife . 10625 . 014Figure 4—figure supplement 2 . Trogocytosis does not require de novo protein synthesis , but is inhibited by lecithin . ( A ) The change in MHC-I transfer induced by infection when host ( cycloheximide [CHX] ) or bacteria ( chloramphenicol [Chlor] ) protein synthesis is inhibited . The uninfected cells were averaged for each experiment and each infected or uninfected sample was compared to this average because cycloheximide and chloramphenicol altered the basal level of MHC-I exchange . ( Mean +/- SD ) . ( n=2 independent experiments performed in triplicate ) ( ns p>0 . 05 , p<0 . 05 , p<0 . 01 ) . ( B ) The percent of J774 recipient cells that underwent plasma membrane transfer in the presence or absence of 0 . 5% soy lecithin . DC refers to doublet control . Graph is representative of 3 independent experiments performed in triplicate ( Mean +/- SD ) . ( p=0 . 001 ) DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 01410 . 7554/eLife . 10625 . 015Figure 4—figure supplement 3 . Trogocytosis in various cell types in mouse splenocytes . The percent of Balb/c H2-Kd positive cells that had surface exposed CD45 . 1 in each represented population . Macrophage: F4/80+; Monocyte: F4/80- , CD11b+; Dendritic Cell: F4/80- , CD11c+; Other: F4/80- , CD11b- , CD11c- . Results for each cell type were normalized to the ‘Other’ population in that mouse , so fold change in the ‘Other’ population is always 1 . Dendritic cells fold change: Uninfected = 1 . 3 fold +/- 0 . 3 , Infected 1 . 6 fold +/- 0 . 3 ( average +/- SD ) . Macrophages and monocytes were significantly more likely to undergo trogocytosis in both the infected and uninfected mice than the ‘Other’ cell types based on the raw data ( p<0 . 05 ) . Bars represent the mean . ( 8 or 9 animals per group from 4 independent experiments ) DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 015 To quantify how often bacterial transfer resulted in detectable levels of trogocytosis , we monitored major histocompatibility complex I ( MHC-I ) transfer between infected donor and uninfected recipient BMDMs ( Wakim and Bevan , 2011; Smyth et al . , 2008 ) . We infected C57BL/6 ( B6 ) BMDMs ( MHC-I H2-Kb ) and added uninfected Balb/c BMDMs ( MHC-I H2-Kd ) to the infected B6 cells . After 6 hr of co-incubation , we assayed the Balb/c BMDMs for both F . tularensis infection and the acquisition of B6 MHC-I . We found that infection increased the amount of Balb/c BMDMs that acquired B6 MHC-I ( Figure 4—figure supplement 1A ) . Likewise , newly infected Balb/c cells were significantly more likely to acquire B6 MHC-I than neighboring Balb/c cells that did not become infected ( Figure 4C ) . As with bacterial transfer , trogocytosis did not require de novo host or bacterial protein synthesis and was inhibited by treatment with lecithin ( Figure 4—figure supplement 2 ) . We also observed MHC-I exchange during bacterial transfer when monitoring hMDMs ( Figure 4D ) . The surface exposed MHC-I likely remained functional after transfer because it was capable of binding the ovalbumin derived peptide SIINFEKL ( Figure 4E ) . Taken together , these data indicate that trogocytosis occurred concurrently with bacterial transfer . We found that trogocytosis is a marker for cell to cell transfer , so we assessed the exchange of plasma membrane proteins in infected splenocytes to track bacterial transfer in vivo . We generated chimeric mice by injecting irradiated F1 B6 and Balb/c mice with wild type Balb/c and transgenic CD45 . 1+ B6 bone marrow . In these mice , no cells have genes for both CD45 . 1 and the MHC-I H2-Kd . Thus , cells must undergo trogocytosis if both CD45 . 1 and H2-Kd are present on the surface of an individual cell . We infected these mice with F . tularensis for 3 days and assayed their splenocytes for infected cells and trogocytosis . Consistent with our in vitro data , F . tularensis infection increased trogocytosis ( Figure 4—figure supplement 1B ) . Furthermore , infected cells were significantly more likely than uninfected splenocytes from the same mouse to possess both CD45 . 1 and H2-Kd ( Figure 4F ) . Combined with our in vitro data , these results suggest that cell to cell bacterial transfer occurs in a mouse infection model . The proportion of cells that underwent detectable trogocytosis varied widely between different cell types . Of the cell types we tested , macrophages and monocytes underwent significantly more trogocytosis than dendritic cells or a compilation of all of the other cell types ( Figure 4—figure supplement 3 ) . These data further indicate that the rate of trogocytosis , and likely bacterial transfer , are cell type specific . Recipient cell type specificity suggests that trogocytosis-associated bacterial transfer is a host mediated event . If true then other bacterial species that can survive in macrophages should also exhibit cell to cell spread by this mechanism . To test this hypothesis , we assessed bacterial transfer and trogocytosis with Salmonella enterica serovar Typhimurium ( S . typhimurium ) infected cells . Similar to F . tularensis cell to cell transfer , S . typhimurium infection increased trogocytosis and bacterial transfer correlated with the exchange of MHC-I ( Figure 5A , Figure 5—figure supplement 1 ) . 10 . 7554/eLife . 10625 . 016Figure 5 . Trogocytosis – associated bacterial transfer is not restricted to F . tularensis . The percent of recipient cells that underwent plasma membrane protein transfer in response to ( A ) Salmonella typhimurium or ( B ) fluorescent beads . The recipient BMDMs that acquired bacteria or beads grouped separately from recipient cells in the same well that did not acquire foreign material . DC refers to a doublet control . ( All results from 3–4 independent experiments performed in triplicate ) ( Mean +/- SD ) . ( ns p>0 . 05 , p<0 . 01 , p<0 . 001 ) DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 01610 . 7554/eLife . 10625 . 017Figure 5—figure supplement 1 . Plasma membrane protein exchange increases during infection . The percent of Balb/c recipient BMDMs that acquired B6 MHC-I from uninfected or ( A ) S . typhimurium or ( B ) bead infected donor B6 BMDMs . ( Mean +/- SD ) . ( 3 independent experiments performed in triplicate ) ( p<0 . 01 , ***p<0 . 001 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10625 . 017 We also measured the transfer of beads between BMDMs to test if trogocytosis-associated transfer was specific to bacterial infections or occurred with general foreign material . Unlike infections , beads did not increase the level of trogocytosis above baseline ( Figure 5—figure supplement 1 ) , suggesting that a general bacterial factor increased the rate of trogocytosis and possibly bacterial transfer . But similar to bacterial transfer , the Balb/c macrophages that acquired beads also acquired B6 MHC-I at a significantly higher rate than macrophages that did not acquire beads ( Figure 5B ) . The lower rate of total trogocytosis as well as the correlation between trogocytosis and transfer is likely due to phagocytosis of extracellular beads in the well . Taken together , our data demonstrate trogocytosis-associated transfer of intracellular bacteria is a mechanism that potentially any macrophage-tropic intracellular pathogen can exploit . Our study demonstrates that intracellular bacteria can exploit a host cell cytosolic exchange mechanism to transfer directly from infected cells to macrophages . This cytosolic exchange mechanism is associated with trogocytosis , which is classically defined as the exchange of plasma membrane proteins between two cells . Given that lecithin inhibited both plasma membrane protein and bacterial transfer suggests that trogocytosis and cytosolic exchange are linked with respect to the mechanism of bacterial cell to cell transfer . The bacteria are viable after transfer and can use direct cell to cell transfer to sustain infection without entering the extracellular space . During infections , trogocytosis-associated transfer is a likely mechanism for F . tularensis dissemination . Alveolar macrophages are essentially the only cell type initially infected by F . tularensis following intranasal inoculation in mice ( Roberts et al . , 2014 ) . Prior to peak infection dendritic cells become infected with F . tularensis and these newly infected cells traffic to the draining lymph node ( Bar-Haim et al . , 2008 ) . We postulate that direct cell to cell transfer is the mechanism for the bacteria to transfer from alveolar macrophages to these dendritic cells . Subsequent transfer events may also contribute to systemic dissemination from the lymph node . An important question is why certain bacterial species evolved mechanisms to transfer between cells if bacteria can already transfer through cytosolic exchange . In a confluent monolayer of primary cells , only 10–20% of macrophages became infected with F . tularensis via trogocytosis-associated transfer in a 6 hr interval ( Figure 1B ) . The efficiency of transfer is probably far lower in an infected host because fewer cells are infected and the concentrations of macrophages are much lower . Bacteria that encode mechanisms such as actin based motility likely increase the rate of cell to cell spread . Additionally , certain bacterial species use actin based motility to transfer between epithelial and endothelial cells , whereas F . tularensis does not ( Makino et al . , 1986; Heinzen et al . , 1993; Reed et al . , 2014 ) . It is possible these transfer mechanisms evolved so that these bacteria can transfer between cells types that do not undergo trogocytosis-associated transfer or to increase the rate of transfer . Bacterial transfer through a trogocytosis-like process is limited to specific recipient cell types , suggesting that this transfer mechanism is a host mediated event . The spread of bacteria aids in expanding the replicative niche and possibly dissemination . So why would the host undergo a process that is so potentially deleterious ? In cancer biology , trogocytosis of pMHC-I and pMHC-II results in a cytotoxic T cell response to the tumor ( Dolan et al . , 2006; Zhang et al . , 2008 ) . The immune system may use a similar tactic during infection . Epithelial cells transfer whole antigen to macrophages and dendritic cells to initiate a T cell response via cytosolic exchange ( Ramirez and Sigal , 2002 ) . Trogocytosis-associated transfer may be the mechanism for this antigenic dissemination . Our work focused on macrophages because F . tularensis primarily infects this cell type . However , macrophages are not an ideal cell type to stimulate an adaptive immune response . Future work on trogocytosis-associated transfer in dendritic cells or in the context of adaptive immunity may reveal an important immunological function for this process . An unexpected observation from our work was that the rate of trogocytosis increased during infection in primary cells and in a mouse infection model ( Figure 4—figure supplement 1 , Figure 5—figure supplement 1 ) . These results suggest that infected cells expressed a signal of some kind that initiated , enhanced , or stabilized trogocytosis . This signal is likely not soluble or generalizable because the frequency of trogocytosis only increased in infected cells , not neighboring uninfected cells in the same experimental sample . Trogocytosis is an important immunological process with broad consequences on host engraftment , vaccine efficacy , immune regulation and tumor recognition ( Chow et al . , 2013; Zhang et al . , 2008; Li et al . , 2012; Gu et al . , 2012 ) . Our results indicate that a bacterial stimulus increases the rate of trogocytosis . Future efforts to discern the bacterial products or processes responsible for trogocytosis up-regulation may lead to a specific tool to manipulate trogocytosis . Trogocytosis-associated bacterial transfer is likely beneficial to both the host and macrophage-tropic pathogens depending on the context . Based on our results , we postulate that trogocytosis-associated transfer benefits certain pathogens early during infection by enhancing dissemination , but could also help initialize or propagate a T cell response that eventually clears the pathogen . Future studies on how this process impacts pathogenesis will likely improve our understanding of how bacteria spread in the host and how the innate immune system acquires antigen to initiate the adaptive immune response . Francisella tularensis subsp . tularensis Schu S4 was obtained from Biodefense and Emerging Infectious Research Resources Repository ( BEI Resources ) and Francisella tularensis subsp . holartica live vaccine strain ( LVS ) expressing GFP was generated as described ( Hall et al . , 2008 ) . Schu S4 was used for all experiments shown except live cell imaging . Prior to infection , F . tularensis was grown overnight in Chamberlin’s defined media . L . monocytogenes and S . typhimurium were grown overnight in Luria broth . The clone numbers for the antibodies used in these experiments: anti- F . tularensis lipopolysaccharide ( 1 . B . 288 , US Biologicals; Salem , MA ) , anti-MHC I H2-Kd ( SF1-1 . 1 . 1 , eBioscience; San Diego , CA ) , anti-MHC I H2-Kb ( AF6-88 . 5 . 5 . 3 , eBioscience ) , anti-MHC I HLA-A2 ( BB7 . 2 , eBioscience ) , anti-CD45 . 1 ( A20 , eBioscience ) , anti-CD45 ( 30-F11 , eBioscience ) , anti-MHC I H2-Kb-SIINFEKL ( 25-D1 . 16 , eBioscience ) . The catalog number and company for critical reagents used in these experiments: Cell Trace Red DDAO-SE ( C34553 , Life Technologies; Carlsbad , CA ) , Calcein-AM ( C3099 , Life Technologies ) , Soy Lecithin ( Cas number 8002-43-5 , Acros; Waltman , MA ) , phalloidin ( A22287 , Life Technologies ) , 3 um pore Transwells ( 3402 Costar; Corning , NY ) , gentamicin ( 15750-060 , Gibco; Carlsbad , CA ) The beads ( M-1002-010 , Solulink; San Diego , CA ) used in these experiments were labeled with AF488 succinimidyl ester ( A-20100 , Life Technologies ) to make fluorescent beads . TC-1 lung epithelial cells ( ATCC CRL-2785; Manassas , VA ) were maintained in RPMI supplemented with sodium pyruvate , L-glutamine and non-essential amino acids in 10% fetal bovine serum ( FBS , Gibco ) . J774A . 1 macrophage-like cells ( ATCC TIB-67 ) were maintained in DMEM containing 10% FBS supplemented with sodium pyruvate and L-glutamine . All cell types were kept at 37°C and 5% CO2 . All cell types were checked for proper morphology prior to every experiment and consistently monitored for changes in cell replication that might indicate Mycoplasma contamination . For the BMDM , TC-1 and J774 transfer experiments , cells were seeded the night before the experiment at 250 , 000 cells per well in non-tissue culture treated 12 well dishes or 500 , 000 cells per well in a 6 well dish on coverslips for microscopy . BMDMs were generated as previously described ( Mortensen et al . , 2010 ) . Unless otherwise indicated , cells were infected with F . tularensis at a multiplicity of infection ( MOI ) of 100 , S . typhimurium at an MOI of 10 or beads at an MOI of approximately 1 . 10 ug/ml of gentamicin was added at 2 hr post inoculation when BMDMs or J774s were infected or 3 hr post inoculation for TC-1 cells . For co-incubation experiments , the indicated recipient cell type was added to the infected cells at 18 hr post inoculation and harvested at 24 hr post inoculation unless otherwise indicated . Primary human monocyte derived macrophages were generated by acquiring human blood in heparin tubes and isolating the peripheral blood mononuclear cells ( PBMC ) and serum on a ficoll gradient . The cells were plated in Iscove’s modified Dulbecco’s medium ( IMDM ) for 2 hr . The non-adherent cells were washed away and the media was replaced with IMDM containing 5% autologous human serum . Primary human cells were cultured for 7 days prior to infection . The blood was isolated from several healthy volunteers who gave informed , written consent following an approved protocol by the Institutional Review Board for human volunteers at the University of North Carolina at Chapel Hill . Blood was obtained specifically for these experiments . Different donors were used for each experiment . The infected cells were seeded onto a coverslip for all experiments involving primary human cells . The coverslip was inverted in a well of uninfected cells so that the infected cells were in contact with the uninfected cells . The reciprocal setup was used for TC-1 to BMDM transfer experiments . Other methods to transfer the cells resulted in large amounts of cell lysis . For live cell imaging , J774 cells were infected at an MOI of 500 with GFP-expressing F . tularensis LVS bacteria in a synchronous infection . Briefly , the J774 cells were chilled on ice for 30 min , the media was exchanged with media containing the bacteria , centrifuged for 5 min and then the bottom of the plate was placed in a 37°C water bath for 2 min . The cells were incubated for 15 min in an incubator at 37°C and 5% carbon dioxide and then the media was replaced with media containing gentamicin . The cells were then imaged every 5 min for 24 hr using a 40x objective on an Olympus IX70 microscope in a temperature and carbon dioxide contained chamber . All data were analyzed using ImageJ ( Schneider et al . , 2012 ) BMDMs were seeded at 500 , 000 cells the night before infection . Cells were infected with an MOI of 0 . 5 bacteria and 10 ug/ml of gentamicin was added at 2 hr post inoculation . 0 . 5 mg/ml of soy lecithin ( Acros ) was added with gentamicin at 6 hr post inoculation . 50% of each sample was used for viable bacteria quantification through serial dilutions and plating on chocolate agar . The remaining 50% of the sample was used to determine the number of cells infected as previously described . Soy lecithin is a common emulsifier often used in food preparation that significantly blocks bacterial transfer ( Figure 2C ) and trogocytosis ( Figure 4—figure supplement 2 ) . We were unable to ascertain the precise mechanism behind this inhibition , but suspect that it is due to its properties as an emulsifier because other complex phospholipid mixtures such as bovine lung surfactants ( Survanta ) also decreased bacterial transfer , albeit to a lesser extent ( data not shown ) . Treating infected cells with individual phospholipid components of soy lecithin , such as phosphatidylcholine , did not affect bacterial transfer ( data not shown ) . When analyzing surface markers ( CD45 , H2-KD , H2-KB , or H2-KB-SIINFEKL ) , cells were stained in the wells in which they were infected . We added 2 . 4G2 cell supernatant ( Fc blocking buffer ) to infected cells for 5 min . The 2 . 4G2 supernatant was removed and antibodies were added . After 5 min , the cells were washed twice in PBS containing 2% fetal bovine serum ( FBS ) , re-suspended , and fixed in 4% paraformaldehyde . F . tularensis within infected cells were detected by permeabilizing the plasma membrane with 0 . 1% saponin ( Millipore ) in PBS and 2% FBS ( Gibco ) . The cells were stained with an anti-F . tularensis lipopolysaccharide antibody ( US biological ) conjugated to either Pacific blue , AF488 , or AF647 by combining the antibody with a succinimidyl ester of the dye . The conjugated antibody was separated from unbound dye by a 30 , 000 molcular weight filter and repeated washes with PBS and glycine . Conjugation efficiency was then assayed for each batch . We were able to detect bacteria at 1 hr post-inoculation when as few bacteria as 1 bacteria per cell were present ( data not shown ) . We stained for both extracellular and intracellular bacteria and found that 1% or less of the infected BMDMs were positive due to surface bound extracellular bacteria ( data not shown ) . Due to the low number of false-positive events , we did not stain specifically for extracellular bacteria in the majority of assays so that we could minimize spectral overlap of our panel . All mouse plasma membrane protein transfer experiments included a doublet control . Uninfected cells from both populations were each stained with all antibodies . Each population was removed from the plate and combined in 4% paraformaldehyde . BMDMs were infected for 18 hr and then stained with calcein-AM following the manufacturer’s protocol ( Invitrogen , Grand Island , NY ) . Uninfected BMDMs were concurrently stained with Cell Trace Red ( Invitrogen ) following the manufacturer’s protocol . The different populations were either fixed immediately for controls or combined and co-incubated for 6 hr . The cells were then stained for F . tularensis as described above . The day before infection , BMDMs were seeded either in a 12 well plate or in the chamber of 12 mm , 3 . 0 uM pore transwell . Each chamber ( transwell and plate ) was kept separate . One chamber per pair was infected and 10 ug/ml of gentamicin was added at 2 hr post inoculation to kill any extracellular bacteria . At 6 hr post-inoculation , the gentamicin was removed and the infected and uninfected chambers were combined . We then separated and harvested each chamber at either 6 or 18 hr post inoculation . To test for bacteria traversing the membrane , we combined the chambers , added bacteria directly to the media of the indicated chamber ( MOI 100 ) and tested for the number of infected cells in each chamber 2 hr later ( Figure 1—figure supplement 1B ) . BMDMs were infected for 2 hr and then gentamicin was added . At 6 hr post inoculation , the media was exchanged for media with or without gentamicin . At 6 hr intervals , the cells were harvested and stained for intracellular F . tularensis and the media was serially diluted and plated on chocolate agar . To approximate the number of cells infected every 6 hr , we used the change in infection percentage between intervals and assumed the number of BMDMs doubled overnight . We made this assumption based on previous observation of chromosomal segregation in infected BMDMs by microscopy ( data not shown ) . The conclusions , however , would remain the same even if no cell division is assumed . BMDMs were infected and gentamicin was added at 2 hr post inoculation . At 6 hr post inoculation , the media was exchanged for media containing gentamicin and the indicated treatment . Z-Vad ( OMe ) -FMK ( Cayman Chemical , Ann Arbor , MI ) was used at 20 uM and Necrostain-1 ( Cayman Chemical ) at 10 uM . At 6 or 24 hr , samples were harvested and analyzed for intracellular bacteria . Autophagy inhibition experiments were performed in the same manner , with 10 μM 3-methyladenine ( Cayman Chemical ) added at 18 hr post inoculation . Cell Trace Red BMDMs were added to GFP-expressing F . tularensis infected BMDMs 18 hr post inoculation . At 24 hr post inoculation , the cells were treated with 0 . 1% saponin in PBS and 2% FBS for 15 min at room temperature ( wash buffer ) . 3 uM propidium iodide ( PI ) was added to the cell for 12 min in wash buffer . The cells were washed 3 times and then fixed in paraformaldehyde . GFP positive , PI negative live bacteria and GFP , PI double positive dead bacteria were enumerated using fluorescence microscopy . To ensure that PI could access and bind to dead bacteria following saponin treatment BMDMs infected with the GFP expressing intracellular growth impaired mutant △FTT_0924 ( Brunton et al . , 2015 ) were subjected to the same procedure ( Figure 2—figure supplement 1 ) . TC-1 epithelial cells were infected for 6 hr as described above . A cover slip seeded with BMDMs was inverted on top of the infected TC-1 cells and the cells were co-incubated for 18 hr in media containing gentamicin . At 24 hr post inoculation , the slide was removed and the TC-1 and BMDM cells that migrated from the cover slip to the bottom of the plate were stained for CD45 to determine cell type , fixed , and then stained with F . tularensis LPS antibody as described above . The 0 hr co-incubation represents TC-1 cells that were infected for 24 hr but did not have BMDMs added to the well . All mice were obtained from Jackson Laboratory ( Bar Harbor , ME ) and were housed in specific pathogen free housing at the University of North Carolina- Chapel Hill . All mouse experiments were performed under approved protocols from the University of North Carolina- Chapel Hill Institutional Animal Care and Use Committee . All mice used were female . The age of mice for bone marrow macrophage production varied ( 6 weeks to 6 months old ) . All mice used to generate chimeric mice were 6 weeks old at the time of irradiation or bone marrow harvest . F1 mice from a mating of C57Bl/6 and Balb/c mice were irradiated with 1000 cGY using an X-ray irradiator . About 5 hr after irradiation , the irradiated mice were reconstituted by intravenous injection of 10 million T cell depleted bone marrow cells per mouse ( T cells depleted using Miltenyi CD3e Microbead Kit following the manufacturers protocol ) . The bone marrow cells were approximately a 1:1 mixture of cells from wild-type Balb/c mice and CD45 . 1 C57bl/6 mice ( B6 . SJL-PTprca Pepcb/ BojJ ) . No blinding was performed in these studies . Five to seven weeks after irradiation , half of the bone marrow chimera mice in each irradiation group were infected intranasally with approximately 500 colony forming units of GFP-expressing F . tularensis Schu S4 . Mice were randomly assigned to each group . At 3 days post inoculation , the spleens were harvested and made into a single cell suspension . The cells were treated with ammonium chloride lysing buffer to removed red blood cells . The splenocytes were then stained with anti-CD45 . 1 and anti-H2-KD ( Balb/c MHC I ) antibodies , washed , fixed in 4% paraformaldehyde , stained for intracellular F . tularensis and analyzed by flow cytometry . C57BL/6 BMDMs were infected and gentamicin was added at 2 hr post inoculation . At 18 hr post inoculation , Balb/c BMDMs were added to the infected B6 cells in the presence of gentamicin . For select experiments , 0 . 5 ug of the ovalbumin peptide SIINFEKL ( ova 257–264 ) ( AnaSpec Inc ) was also added at 18 hr post inoculation . At 24 hr , the cells were stained and harvested for flow cytometry . All flow cytometry experiments included a doublet control , where stained and paraformaldehyde fixed B6 and Balb/c cells were mixed at approximately a 1 to 1 ratio with a similar cell concentration as the rest of the samples . The doublet control sample represents the background level of false positives for plasma membrane protein transfer due to doublets . Experiments with S . typhimurium or magnetic beads were performed by infecting B6 BMDMs with an MOI of 10 GFP expressing S . typhimurium bacteria or an MOI of 1 streptavidin coated magnetic bead ( Solulink ) conjugated to AF488 . At 2 hr post inoculation , the cells were washed and media containing 25 ug/ml of gentamicin was added . At 10 hr , Balb/c BMDMs were added and the samples were harvested at 16 hr . The samples were surface stained as previously described . For microscopy , infected BMDMs were biotinylated at 18 hr post inoculation ( Thermo Scientific; EZ-Link Sulfo-NHS-LC-biotin following the manufacturer’s protocol ) . Cell Trace Red labeled BMDMs were added to the infected cells immediately following biotinylation . 1 to 2 hr later , the samples were stained with AF568 or PE conjugated streptavidin , fixed in 4% paraformaldehyde , and mounted using DAPI containing mounting media . Images were acquired using the 63x objective on a Zeiss CLSM 700 Confocal Laser Scanning Microscope . Images were acquired using Zen software ( Zeiss ) . All data were analyzed using ImageJ ( Schneider et al . , 2012 ) . 3D images were generated using Imaris software ( Bitplane ) . For human samples , HLA-A2 negative , biotinylated MDMs were added to infected HLA-A2+ MDMs at 18 hr post inoculation . The cells were co-incubated for 6 hr and then the recipient cell population was stained with PE-streptavidin and HLA-A2 to assess plasma membrane protein transfer . Recipient BMDMs and the indicated treatment ( 0 . 1 ng/ml cycloheximide or 50 ug/ml chloramphenicol ) were added to infected BMDMs at 18 hr post inoculation . The samples were assessed as described above . At these concentrations , cycloheximide increased the basal rate of plasma membrane protein transfer while chloramphenicol decreased the basal rate of plasma membrane protein transfer . Cells were infected with an MOI of 1 for L . monocytogenes or 100 for F . tularensis . Cells were harvested at 16 hr post inoculation , fixed , permeabilized and stained with AF647 conjugated phalloidin . All statistics were performed by a 2 tailed , unpaired Student t-tests using raw data values . Confocal microscopy experiments represent all cells from 100 total fields of view from 2 independent experiments . For statistics , each field of view was treated as an independent sample . Chimeric mouse experiments were performed with 2 mice per group in 4 independent experiments . We estimated the size for these animal studies based on our results in tissue culture . All other experiments were performed in triplicate for each group in at least 3 independent experiments unless otherwise indicated .	Many of the bacteria that are able to cause disease in humans and other animals are able to grow inside their host’s cells . In doing so , these bacteria can avoid being recognized and killed by the host’s immune system . However , the ability of the bacteria to grow within the cell is constrained by the limited space and nutrients that are available inside the infected cell . The current theory is that most of these bacteria eventually kill the cell they have infected and are released into the body so that they can infect other host cells . However , since some host cells can exchange material with their neighbors , it is also possible that the bacteria may be able to travel directly between host cells without leaving the safety of the cell environment . Macrophages are immune cells that patrol the body to identify and destroy damaged host cells , bacteria and other microbes . Macrophages are also able to interact with neighboring healthy cells through a process called trogocytosis ( “trogo” is essentially Greek for nibble ) . During this process , the membranes of the two participating cells briefly fuse and some of the proteins in the membranes are transferred from one cell to the other . Afterwards , the two cells separate but retain the membrane proteins they acquired from the other cell . The purpose of trogocytosis is poorly understood , but it is thought to help the host to develop immune responses against microbes and tumors . Steele et al . investigated whether infected mouse and human cells can transfer bacteria to healthy macrophages during trogocytosis . The experiments show that two types of bacteria – called Francisella tularensis and Salmonella enterica – can transfer from infected cells to macrophages via trogocytosis . Furthermore , the cells of mice infected with F . tularensis were more likely to undergo trogocytosis , which suggests that the bacterium may promote and use this process to spread throughout tissues in the body . Together , Steele et al . ’s finding show that some bacteria can hijack a naturally occurring cellular process to move between host cells without re-entering the space that surrounds cells , or damaging either the donor or recipient cell . The next steps following on from this work are to find out how much trogocytosis contributes to the spread and progression of disease . A future goal is to understand the molecular mechanism of trogocytosis so it may be possible to develop drugs that can inhibit the spread of the bacteria in patients .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease", "immunology", "and", "inflammation" ]	2016	Trogocytosis-associated cell to cell spread of intracellular bacterial pathogens
Although it is clear that trisomy 21 causes Down syndrome , the molecular events acting downstream of the trisomy remain ill defined . Using complementary genomics analyses , we identified the interferon pathway as the major signaling cascade consistently activated by trisomy 21 in human cells . Transcriptome analysis revealed that trisomy 21 activates the interferon transcriptional response in fibroblast and lymphoblastoid cell lines , as well as circulating monocytes and T cells . Trisomy 21 cells show increased induction of interferon-stimulated genes and decreased expression of ribosomal proteins and translation factors . An shRNA screen determined that the interferon-activated kinases JAK1 and TYK2 suppress proliferation of trisomy 21 fibroblasts , and this defect is rescued by pharmacological JAK inhibition . Therefore , we propose that interferon activation , likely via increased gene dosage of the four interferon receptors encoded on chromosome 21 , contributes to many of the clinical impacts of trisomy 21 , and that interferon antagonists could have therapeutic benefits . Trisomy 21 ( T21 ) is the most common chromosomal abnormality in the human population , occurring in approximately 1 in 700 live births ( Alexander et al . , 2016 ) . The extra copy of chromosome 21 ( chr21 ) impacts human development in diverse ways across every major organ system , causing the condition known as Down syndrome ( DS ) . One of the most intriguing aspects of T21 is that it causes an altered disease spectrum in the population with DS , protecting these individuals from some diseases ( e . g . solid tumors , hypertension ) , while strongly predisposing them to others ( e . g . Alzheimer’s disease , leukemia , autoimmune disorders ) ( Alexander et al . , 2016; Sobey et al . , 2015; Bratman et al . , 2014; Roberts and Izraeli , 2014; Anwar et al . , 1998; Malinge et al . , 2013; Hasle et al . , 2016 ) . Despite many years of study , the molecular , cellular , and physiological mechanisms driving both the protective and deleterious effects of T21 are poorly understood . A few chr21-encoded genes have been implicated in the development of specific comorbidities , such as APP in Alzheimer’s disease ( Wiseman et al . , 2015 ) , and DYRK1A and ERG in hematopoietic malignancies ( Stankiewicz and Crispino , 2013; Malinge et al . , 2012 ) . Therefore , research in this area could inform a wide range of medical conditions affecting not only those with DS , but also the typical population . The clinical manifestation of DS is highly variable among affected individuals , with various comorbidities appearing in a seemingly random fashion , suggesting the presence of strong modifiers , genetic or otherwise , of the deleterious effects of T21 . Even conserved features , such as cognitive impairment , display wide quantitative variation ( de Sola et al . , 2015 ) . Collectively , our understanding of the mechanisms driving such inter-individual variation in the population with DS is minimal . More specifically , it is unclear what gene expression changes are consistently caused by T21 , versus those that are context-dependent . Integrated analyses of a large body of studies have indicated that the changes in gene expression caused by T21 involve various signaling pathways ( Scarpato et al . , 2014 ) , however , these studies vary widely in cell type , number of samples , and even analysis platform , among other variables ( Volk et al . , 2013; Costa et al . , 2011 ) . More recently , gene expression analysis of cells derived from discordant monozygotic twins , only one of which was affected by T21 , concluded that global gene expression changes in T21 cells are driven by differences in chromatin topology , whereby affected genes are clustered into large chromosomal domains of activation or repression ( Letourneau et al . , 2014 ) . However , independent re-analysis of these data has challenged this conclusion ( Do et al . , 2015 ) . Therefore , there remains a clear need to identify the consistent gene expression changes caused by T21 and to characterize how these programs are modified across cell types , tissue types , genetic backgrounds , and developmental stages . In order to identify consistent signaling pathways modulated by T21 , defined as those that withstand the effects of inter-individual variation , we employed two complementary genomics approaches , transcriptome analysis and shRNA loss-of-function screening , in both panels of cell lines and primary cell types from individuals of diverse genetic background , gender , and age , with and without T21 . Our RNA-seq transcriptome analysis identified consistent gene expression signatures associated with T21 in all cell types examined . Interestingly , the fraction of this gene expression signature that is not encoded on chr21 is dominated by the interferon ( IFN ) transcriptional response , an observation that is reproducible in skin fibroblasts , B cell-derived lymphoblastoid cell lines , as well as primary monocytes and T cells . In parallel , we performed a kinome-focused shRNA screen that identified the IFN-activated kinases JAK1 and TYK2 as strong negative regulators of T21 cell proliferation in fibroblasts . Importantly , pharmacological inhibition of JAK kinases improves T21 cell viability . Taken together , our results identify the IFN pathway as consistently activated by T21 , which could merely be a result of increased gene dosage of four IFN receptor subunits encoded on chr21 . We hypothesize that IFN activation could contribute to many of the effects of T21 , including increased risk of leukemia and autoimmune disorders , as well as many developmental abnormalities also observed in interferonopathies ( Yao et al . , 2010; Zitvogel et al . , 2015; Crow and Manel , 2015; McGlasson et al . , 2015 ) . In order to investigate consistent gene expression signatures associated with T21 , we performed RNA-seq on a panel of 12 age- and gender-matched human fibroblasts from euploid ( disomic , D21 ) and T21 individuals ( Figure 1—figure supplement 1A–C ) . T21 was confirmed by PCR analysis of the chr21-encoded RCAN1 gene ( Figure 1—figure supplement 1D ) . We included samples from different genetic backgrounds , ages , and genders , in order to avoid identifying differences that are specific to a particular pair of isogenic or genetically related cell lines and which would not withstand the effects of inter-individual variation . To illustrate this point , comparison of one pair of disomic male individuals of similar age yielded thousands of differentially expressed genes ( DEGs ) , with similar numbers of upregulated and downregulated DEGs ( Figure 1A–B , Male 1 vs . Male 2 ) . However , when the 12 samples are divided into two groups with roughly balanced age , sex , and T21 status , very few consistent changes were identified , thus demonstrating the impact of inter-individual variation within our sample set ( Figure 1A–B , Figure 1—figure supplement 1C , Group 1 vs . Group 2 ) . In contrast , comparison of all T21 versus all D21 cells identified 662 consistent DEGs , with a disproportionate number of these upregulated in T21 cells ( 471 of 662 , Figure 1A , T21 vs . D21 , Supplementary file 1A ) . We also observed an uncharacteristic spike of DEGs at ~1 . 5-fold overexpression in T21 cells on a volcano plot , consistent with many chr21 genes being overexpressed solely due to increased gene dosage ( Figure 1B ) . For comparison purposes , we also analyzed samples by sex which expectedly yielded DEGs encoded on chrX ( e . g . XIST ) and chrY ( Figure 1 A–B; Female vs . Male ) . Sex causes fewer significant changes than T21 , with roughly equal numbers of upregulated and downregulated genes . Taken together , these data indicate that T21 produces consistent changes in a gene expression signature that withstands differences in genetic background , age , sex , and site of biopsy . Of note , when we performed RNA-seq analysis using increasing numbers of T21 vs . D21 pairs , the fraction of chr21-encoded DEGs increased steadily with sample size , accounting for ~12% of the core gene expression signature in the 12 cell line panel . However , 88% of DEGs are located on other chromosomes , indicating the existence of conserved mechanisms driving these genome-wide changes in gene expression ( Figure 1—figure supplement 1E ) . 10 . 7554/eLife . 16220 . 003Figure 1 . Transcriptome analysis identifies consistent changes in global gene expression between euploid ( D21 ) and trisomy 21 ( T21 ) fibroblasts . ( A ) MA plots displaying the results of RNA-seq analysis for the indicated comparisons ( see Figure 1—figure supplement 1A–C ) . Differentially expressed genes ( DEGs ) , as defined by DEseq2 ( FDR < 10% ) , are labeled in red . ( B ) Volcano plots of comparisons in A highlight changes in chr21 gene expression ( green ) consistent with increased gene dosage effects . ( C ) Manhattan plots displaying DEGs ( red ) and all genes ( black ) for individual chromosomes do not show obvious domains of contiguous upregulation or downregulation . Shaded areas highlight regions of overlapping upregulation and downregulation ( see Figure 1—figure supplements 2A and 3 ) . ( D ) Violin plots of chr21 and non-chr21 DEGs displaying the distribution of fold changes of DEGs in each category . p-values were calculated with the Kolmogorov-Smirnov test . ( E ) Heatmap of all significant DEGs showing clustering of chr21 DEGs ( green ) around 1 . 5 fold upregulation in T21 cells . ( F ) Kernel density estimate plot highlighting the probabilities of chr21 DEGs ( green , green dashed line indicates median ) , non-chr21 DEGs ( black , black dashed line indicates median ) , and all genes ( gray ) , of having a given fold change . ( G ) Box and whisker plot of standard deviations of fold changes in DEGs for six pairwise comparisons of age- and gender-matched T21 versus D21 fibroblasts showing greater variation in fold change for non-chr21 DEGs . p-values were calculated with the Kolmogorov-Smirnov test . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 00310 . 7554/eLife . 16220 . 004Figure 1—figure supplement 1 . T21 and D21 fibroblast RNA-seq . ( A ) Description of fibroblast cell lines used in this study . ( B ) Principal component analysis ( PCA ) of fibroblast RNA-seq samples demonstrates tight grouping of biological replicates . ( C ) Schematic of group comparisons . ( D ) PCR of genomic DNA for the RCAN1 gene encoded on chr21 confirms T21 status . RPLP0 is a control gene encoded on chr12 . ( E ) Bar graph displaying how numbers of differentially expressed genes ( DEGs ) encoded on chr21 increase with sample size . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 00410 . 7554/eLife . 16220 . 005Figure 1—figure supplement 2 . Amplification of changes in gene expression emanating from T21 . ( A ) Manhattan plot showing most DEGs ( red ) are not encoded on chr21 . ( B ) Example box and whisker plots of chr21 ( green ) and non-chr21 ( black ) DEGs . mRNA expression values are displayed in reads per kilobase per million ( RPKM ) . Benjamini-Hochberg adjusted p-values were calculated using DESeq2 . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 00510 . 7554/eLife . 16220 . 006Figure 1—figure supplement 3 . Differentially expressed genes in trisomy 21 fibroblasts are not organized into obvious chromatin domains . Manhattan plots for individual chromosomes indicating differentially expressed genes in red . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 006 A recent report concluded that changes in gene expression caused by T21 between a single pair of discordant monozygotic twins were due to dysregulation of chromosomal domains ( Letourneau et al . , 2014 ) . Thus , we next asked where the ~88% of core DEGs not encoded on chr21 are located across the genome . This exercise revealed broad distribution across all chromosomes , with no obvious contiguous domains of up- or downregulation ( see Figure 1—figure supplement 2A for a whole genome Manhattan plot , and Figure 1—figure supplement 3 for individual chromosomes ) . In fact , mere visual analysis of DEGs from the individual chromosomes previously claimed by Letourneau et al . to harbor large dysregulated domains ( e . g . chr3 , chr11 , chr19 ) did not reveal such domains in our dataset , showing instead obvious regions of overlapping activation and repression ( shaded gray boxes in Figure 1C ) . Thus , our analysis is more consistent with the report that re-analyzed the data in Letournaeu et al . and questioned the existence of these chromosomal domains ( Do et al . , 2015 ) . In fact , the only region of the genome at which there was clear contiguous upregulation of DEGs was chr21 itself ( Figure 1C , Figure 1—figure supplements 2A and 3 ) . In order to characterize the mechanism driving the consistent changes caused by T21 , we examined the regulatory differences between DEGs encoded on chr21 and those not encoded on chr21 . Several lines of evidence indicate that , while chr21 DEGs are regulated mostly by increased gene dosage , non-chr21 DEGs may be driven by specific pathways that are subject to signal amplification , with a bias toward upregulation , and greatly affected by inter-individual variation . First , violin plots display the relatively small number of chr21 DEGs , showing mostly upregulation clustered around 1 . 5 fold , versus a much larger number of non-chr21 DEGs , showing both up- and downregulation with no obvious clustering of fold changes ( Figure 1D , Figure 1—figure supplement 2B ) . Second , the obvious effect of gene dosage on the expression of chr21 DEGs is apparent in the violin plots and heatmaps ( Figure 1D , E ) , where the median fold change centers around 1 . 5 fold ( e . g . APP , ETS2 ) , while a few genes show greater induction ( e . g . MX1 , MX2 ) . In fact , chr21 genes exhibit more than an 80% probability of a ~1 . 5-fold change as calculated by kernel density estimation analysis ( Figure 1F ) . Third , the bias toward upregulation among non-chr21 DEGs is evident in the violin plots , heatmaps , and density estimation analysis ( Figure 1D–F ) , where a larger fraction of these genes is upregulated . Finally , we measured the inter-individual variation of chr21 DEGs versus non-chr21 DEGs by calculating the standard deviation for each DEG across each age- and gender-matched pair of fibroblasts . As shown in Figure 1G , the median standard deviation for chr21 DEGs is much smaller than for all DEGs . Altogether , these results suggest the existence of consistent signaling pathways activated by increased dosage of chr21 genes , which in turn cause global changes in gene expression , with a bias toward upregulation and displaying strong inter-individual variation . Next , we subjected T21 DEGs to upstream regulator analysis using Ingenuity Pathway Analysis ( IPA ) to identify putative factors contributing to consistent changes in gene expression . This analysis tool includes both a hypergeometric test for overlapping sets of genes and a directional component to predict activation or inactivation of factors that control gene expression ( e . g . transcription factors , protein kinases ) ( Krämer et al . , 2014 ) . We confirmed the effectiveness of this tool using published RNA expression datasets from our lab for cells treated with an inhibitor of the p53-MDM2 interaction , hypoxia , and serum stimulation ( Sullivan et al . , 2012; Donner et al . , 2010; Galbraith et al . , 2013 ) . IPA effectively identified p53 , the Hypoxia Inducible Factor 1A ( HIF1A ) , and growth factor receptors and downstream kinases ( PDGF , ERK ) as the top upstream regulators in each scenario , respectively ( Figure 2—figure supplement 1A ) . Strikingly , the top 13 upstream regulators predicted to be activated in T21 cells are all IFN-related factors , including IFN ligands ( e . g . IFNA2 , IFNB , IFNG ) and IFN-activated transcription factors ( e . g . IRF3 , IRF5 , IRF7 , STAT1 ) ( Figure 2A ) . Importantly , most of these signals are derived from non-chr21 DEGs , and would be missed by analyses focused specifically on chr21-encoded genes ( Figure 2A ) . This analysis also identified two known repressors of IFN signaling , MAPK1 and TRIM24 , as upstream regulators inactivated in T21 cells , consistent with activation of the IFN pathway ( Huang et al . , 2008; Tisserand et al . , 2011 ) . As an example of how the RNA-seq data supports the upstream regulator prediction by IPA , Figure 2B shows the gene network centered on the ligand IFNA2 as a potential driver of consistent gene expression changes . Strong activation of the IFN pathway was also predicted using a different tool , the Pathway Commons Analysis in WebGestalt ( Zhang et al . , 2005; Wang et al . , 2013; Cerami et al . , 2011 ) , where 4 of the top 15 pathways identified were IFN-related ( Figure 2—figure supplement 1B ) . 10 . 7554/eLife . 16220 . 007Figure 2 . The interferon ( IFN ) transcriptional response is activated in trisomy 21 ( T21 ) fibroblasts . ( A ) Upstream regulator analysis of the T21-associated gene expression signature using Ingenuity Pathway Analysis ( IPA ) predicts numerous IFN-related factors as activated in T21 cells . ( B ) Representative results of the upstream regulator analysis for the Type I IFN ligand IFNA2 . ( C ) Graphical summary of the observed deregulation of the IFN pathway in T21 fibroblasts , showing the six IFN receptor subunits , four of which are encoded on chr21 and significantly upregulated in T21 fibroblasts; the predicted upstream regulators ( orange ) , including the Type I , II , and III IFN ligands , as well as the IFN-activated transcription factors ( IRFs and STATs ) ; and select examples of Interferon Stimulated Genes ( ISGs ) upregulated in T21 fibroblasts , either encoded on chr21 ( green ) or elsewhere in the genome ( gray ) . ( D ) Box and whisker plots showing RNA expression for the six IFN receptor subunits and select ISGs . chr21-encoded genes are highlighted in green . mRNA expression values are displayed in reads per kilobase per million ( RPKM ) . Benjamini-Hochberg adjusted p-values were calculated using DESeq2 . ( E ) Western blot analysis confirming upregulation of IFN receptors , STAT1 phosphorylation , and ISGs , in T21 fibroblasts . ( F ) Box and whisker plots showing protein expression of select IFN-related genes as measured by SOMAscan assay . chr21-encoded genes are highlighted in green . Protein expression values are displayed in relative fluorescence units ( RFU ) . Adjusted p-values were calculated using the Empirical Bayes method in QPROT . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 00710 . 7554/eLife . 16220 . 008Figure 2—figure supplement 1 . Network analysis confirms IFN activation signature in T21 cells . ( A ) IPA upstream regulator analysis of genes activated upon MDM2 inhibition with Nutlin-3 , hypoxia ( 1% O2 ) , and serum stimulation in HCT116 colorectal cancer cells correctly identifies the transcription factor p53 , the transcription factor HIF1A , and the growth factor PDGF , as the key upstream regulators in each scenario . ( B ) Top 15 deregulated pathways in T21 cells identified by Pathway Commons Analysis in WebGestalt . IFN-related pathways are highlighted in red . ( C ) Pie charts showing the percentage of chr21 and non-chr21 upregulated genes in the interferon pathway . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 008 Notably , activation of IFN signaling in T21 cells could be explained by the fact that four of the six IFN receptors , IFNAR1 , IFNAR2 , IFNGR2 , and IL10RB , ( representing each IFN class , Type-I , -II , and -III ) , are chr21-encoded DEGs ( Figure 2C , D ) . Using a combination of IPA upstream regulator predictions and our RNA-seq data , we clearly identified the canonical IFN pathways –from ligands through receptors and kinases and down to transcription factors and IFN-stimulated genes ( ISGs ) – as activated in T21 cells ( Figure 2C ) . Whereas IFN receptors are upregulated ~1 . 5 fold with relatively low levels of inter-individual variation , as expected for increased gene dosage in T21 cells , the downstream ISGs exhibit larger fold changes , greater variation between samples , and tend to have low expression levels in D21 cells , in accord with activation of IFN only in T21 cells ( Figure 2D ) . We confirmed the elevated basal expression of three of the IFN receptors ( IFNAR1 , IFNGR2 , and ILR10RB ) , enhanced basal phosphorylation of STAT1 , as well as increased basal expression of several ISGs at the protein level in T21 cells , with noticeable inter-individual variation ( Shuai et al . , 1994; Waddell et al . , 2010; Schoggins et al . , 2011 ) ( Figure 2E ) . We next analyzed protein lysates from the 12 fibroblast lines using SOMAScan technology , which employs DNA aptamers to monitor epitope abundance ( Gold et al . , 2012; Mehan et al . , 2014; Hathout et al . , 2015 ) . This assay confirmed elevated protein levels for many of the IFN-related genes found to be induced at the mRNA level in the RNA-seq experiment ( Figure 2F ) . Finally , we examined the fraction of our upregulated DEGs linked to IFN signaling using IPA , Pathway Commons , and a list of 387 validated ISGs curated by Schoggins and colleagues ( Schoggins et al . , 2011 ) . Our analysis revealed that 21% ( 101/471 ) of DEGs upregulated in T21 cells are linked to IFN signaling , with contributions from both chr21 ( 17% , 14/81 ) and non-chr21 ( 22% , 87/390 ) DEGs , pointing to IFN activation as a potential mechanism for the larger number of upregulated versus downregulated DEGs ( Figure 2—figure supplement 1C ) . Altogether , these results indicate that the IFN pathway is consistently induced by trisomy 21 in fibroblasts , and that the IFN transcriptional response accounts for a considerable fraction of the transcriptome changes caused by trisomy 21 across the genome . We next investigated whether T21 cells produce a stronger response to specific IFN ligands than their D21 counterparts . To test this , we treated three pairs of fibroblasts –roughly matched by age and sex– with various doses of the Type I ligands IFN-α or -β , or with the Type II ligand IFN-γ , and monitored the expression of key ISGs via western blot . We also monitored phosphorylation of STAT1 . Overall , these efforts revealed that trisomy 21 cells show stronger induction of ISGs upon treatment with all three ligands , albeit with variation across specific cell lines and ligands ( Figure 3 ) . For example , stimulation with IFN-α led to stronger induction in the T21 cell line for MX1 in pairs 1 and 2 , stronger induction of IDO1 in pairs 1 and 3 , and stronger induction of ISG15 in pairs 1 and 2 ( Figure 3A ) . Similar results were observed for the other Type I ligand , IFN-β . However , ligand-specific differences were also observed . For example , IDO1 was more strongly induced by IFN-α and -β in the T21 cell line in pair 1 , but this was not the case when using IFN-γ ( Figure 3A–C ) . Thus , these results confirm the notion of strong inter-individual variation in the downstream signaling effects of T21 . Of note , all three IFN ligands consistently induced STAT1 phosphorylation ( pSTAT1 ) both in D21 and T21 cells , but the levels of pSTAT1 did not correlate precisely with the expression levels of the various ISGs . For example , the obviously different levels of ISG15 in pair 2 upon treatment with the three ligands do not correlate with dissimilar levels of pSTAT1 ( Figure 3A–C ) . This suggests that STAT1 phosphorylation is not a robust predictor of ISG expression , which is ultimately defined by the orchestrated action of multiple IFN-activated transcription factors . 10 . 7554/eLife . 16220 . 009Figure 3 . T21 fibroblasts are more sensitive to IFN stimulation than D21 fibroblasts . ( A ) Western blots showing that three T21 cell lines are more sensitive to IFN-α treatment ( 24 hr ) than age- and gender-matched D21 control cells as measured by induced expression of the ISGs MX1 , IDO1 and ISG15 . Elevated pSTAT1 levels confirm effective induction of the IFN pathway in response to ligand exposure . ( B ) Western blots as in A for IFN-β treatment . ( C ) Western blots as in A for IFN-γ treatment . * indicates non-specific bands . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 009 In a parallel unbiased approach to identify signaling cascades deregulated by T21 , we employed an shRNA screen to identify protein kinases that may have a differential impact on the viability ( i . e . proliferation and/or survival ) of T21 cells relative to D21 cells . We hypothesized that core gene expression changes in T21 cells lead to a rewiring of signaling cascades , creating differential requirements for specific kinases during cell survival and proliferation . In order to identify such kinases , we introduced a library of 3 , 075 shRNAs targeting 654 kinases into each of the 12 fibroblast cell lines we subjected to transcriptome analysis . We then propagated these cells for 14 days to allow for selection of cells harboring shRNAs targeting kinases that differentially affect survival and/or proliferation of T21 cells versus D21 cells , henceforth referred to as DMT21 kinases ( Differential Modulators of T21 cells ) ( Figure 4A ) . In this screen , relative enrichment of a given shRNA in the T21 population could result from the targeted kinase being a negative regulator of T21 cellular fitness , a positive regulator of D21 cellular fitness , or a combination of both . To minimize the possibility of shRNA off-target effects , we required at least three independent shRNAs targeting a given kinase to score as significantly enriched or depleted , with no more than one shRNA against each kinase scoring in the opposite direction ( see Materials and methods for details ) . This analysis identified a total of 25 and 15 kinases that negatively and positively affect the fitness of T21 cells relative to D21 cells , respectively ( Figure 4B , Figure 4—figure supplement 1 , Supplementary file 2 ) . The top scoring enriched kinase was mTOR , indicating that this kinase differentially decreases the fitness of T21 cells ( and/or differentially increases the fitness of D21 cells ) . This could be consistent with previous reports showing hyperactivation of mTOR signaling in the brains of individuals with DS and mouse models of trisomy 21 and consequent impairments in autophagy ( Ahmed et al . , 2013; Perluigi et al . , 2015; Troca-Marín et al . , 2014; Iyer et al . , 2014 ) . Importantly , among DMT21 kinases predicted to hinder T21 cell viability were the IFN-activated kinases JAK1 and TYK2 ( Müller et al . , 1993; Stahl et al . , 1994 ) ( Figure 4B , C , Figure 4—figure supplement 1A , B ) . To confirm that JAK1 signaling negatively affects the relative viability of T21 cells , we treated two pairs of D21/T21 fibroblasts with increasing doses of the JAK1/2 inhibitor ruxolitinib ( Rux ) ( Tefferi et al . , 2011 ) . Rux treatment led to decreased levels of pSTAT1 , decreased protein expression of MX1 –an ISG encoded on chr21– , and decreased mRNA expression of several ISGs found to be upregulated in T21 fibroblasts in our RNA-seq experiment ( Figure 4D and Figure 4—figure supplement 1C , D ) . To assess the impact of Rux treatment on cell viability , we seeded equal numbers of D21 and T21 fibroblasts in the absence or presence of increasing doses of the inhibitor , and counted the number of viable cells 3 days post-seeding . Notably , the number of viable T21 cells was much lower in all conditions tested ( Figure 4E and Figure 4—figure supplement 1E ) . However , whereas Rux treatment led to a dose-dependent increase in the number of viable T21 cells , it also produced a decrease in the number of viable D21 cells at the highest concentration . When the cell counts are represented as T21/D21 ratios , it is clear that JAK inhibition has a differential effect on cell proliferation between T21 and D21 cells ( Figure 4F , G and Figure 4—figure supplement 1F , G ) . This is consistent with shRNAs targeting JAK1 ( and TYK2 ) being differentially enriched in T21 cells during the 14-day course of the screen . Ultimately , these data support the notion of differential signaling requirements in T21 relative to D21 cells and identify two IFN-related kinases as negative regulators of T21 fibroblast viability . 10 . 7554/eLife . 16220 . 010Figure 4 . An shRNA screen identifies the interferon ( IFN ) -activated kinases JAK1 and TYK2 as negative regulators of trisomy 21 ( T21 ) cellular fitness . ( A ) Schematic of kinome-focused shRNA screen to identify Differential Modulators of T21 ( DMT21 ) cellular fitness . ( B ) Volcano plot highlighting shRNAs targeting DMT21 genes that differentially inhibit T21 ( blue ) or euploid ( D21 , yellow ) cellular fitness . Top hits were filtered by a FDR < 5% and at least three shRNAs to the same gene scoring in one direction with no more than one shRNA scoring in the opposite direction . NRBP1 and JAK1 shRNAs are indicated with arrows . ( C ) Bar graphs of the screen results for the IFN-related kinases JAK1 and TYK2 , as well as mTOR , NRBP1 , MAPK9 and TSSK6 . ( D ) Western blot analysis confirming downregulation of STAT1 phosphorylation and MX1 expression upon inhibition of JAK kinases with ruxolitinib ( Rux ) at the indicated concentrations in the GM02036-GM02767 cell pair . ( E ) Absolute cell numbers grown for 72 hr in their respective conditioned media with the indicated doses of Rux . ( F ) Relative cell numbers from ( E ) . ( G ) Ratio of T21:D21 relative cell numbers demonstrates the overall differential effect of Rux on the number of viable cells from this T21-D21 pair . Results from a second cell line pair are shown in Figure 4—figure supplement 1D–G . All data shown are an average of three experiments ± standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 01010 . 7554/eLife . 16220 . 011Figure 4—figure supplement 1 . An shRNA screen identifies differential modulators of T21 ( DMT21 ) cellular fitness . ( A , B ) shRNAs targeting DMT21 genes that differentially inhibit T21 ( blue ) or D21 ( yellow ) cellular fitness . ( C ) Q-RT-PCR demonstrating that Ruxolitinib ( Rux ) treatment downregulates mRNA expression for many ISGs in a dose-dependent manner . ( D ) Western blots demonstrating the effect of Rux treatment on pSTAT1 and MX1 on the cell line pair GM05659 ( D21 ) and AG05397 ( T21 ) ( pair 2 ) . ( E ) Absolute cell numbers from pair 2 grown for 72 hr in their respective conditioned media with the indicated doses of Rux . ( F ) Relative cell numbers from ( E ) . ( G ) Ratio of T21:D21 relative cell numbers demonstrates the overall differential effect of Rux on the number of viable cells from this T21-D21 pair . All data shown are an average of three experiments ± standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 011 To test whether consistent changes in gene expression programs elicited by trisomy 21 are conserved across cell types , we performed RNA-seq on a panel of six age-matched , female lymphoblastoid cell lines from D21 and T21 individuals ( Figure 5—figure supplement 1A–B ) . These cell lines were generated by immortalizing B cells with Epstein Barr Virus ( EBV ) , thus enabling us to compare a cell type of lymphocytic origin with the fibroblasts of mesenchymal origin . Analysis of DEGs associated with T21 identified 1538 genes both up and downregulated with more upregulated DEGs ( 861 out of 1538 ) , as was seen in the fibroblasts ( Figure 5A , Supplementary file 1B ) . Similarly , a peak of highly significant DEGs with ~1 . 5-fold change , comprised of chr21-encoded genes , is observed in a volcano plot ( Figure 5B ) . Furthermore , most DEGs are distributed across the genome , and not arranged into obvious chromosomal domains outside of chr21 ( Figure 5C and Figure 5—figure supplement 2 ) . IPA revealed that the top upstream regulators of the consistent gene expression signature driven by T21 in lymphoblastoids are also IFN-related , and that this prediction is powered by non-chr21 DEGs ( Figure 5D ) . Comparison of DEGs from fibroblasts and lymphoblastoids demonstrates that many of the same upstream regulators are predicted to be activated and are IFN-related factors ( Figure 5E ) . All four chr21-encoded IFN receptors are significantly upregulated in lymphoblastoids ( Figure 5F ) , as they are in fibroblasts . In fact , the most significant DEG encoded on chr21 is IFNAR1 ( Figure 5B ) . Increased basal protein expression was confirmed by western blot for IFNAR1 and IL10RB , as well as for the interferon-related genes TBX21 , GBP5 and BCL2L11 ( BIM ) ( Figure 5G ) . STAT1 phosphorylation was also elevated in the T21 lymphoblastoids ( Figure 5G ) . 10 . 7554/eLife . 16220 . 012Figure 5 . Activation of the interferon ( IFN ) transcriptional response is conserved in trisomy 21 ( T21 ) lymphoblastoid cell lines . ( A ) MA plot displaying the gene expression signature associated with T21 in a panel of six lymphoblastoid cell lines , three of which harbor T21 . Differential expressed genes ( DEGs ) , as defined by DEseq2 ( FDR < 10% ) , are labeled in red . ( B ) Volcano plot of DEGs with those encoded on chr21 highlighted in green . ( C ) Manhattan plot of chr21 with DEGs in red and all other genes in black . ( D ) Upstream regulator analysis reveals activation of the IFN transcriptional response in T21 lymphoblastoid cell lines . ( E ) Comparative analysis between fibroblasts and lymphoblastoids highlights conserved upstream regulators within the IFN pathway . ( F ) Box and whisker plots of RNA expression for the four IFN receptor subunits encoded on chr21 ( green ) and three interferon-related genes ( black ) . mRNA expression values are displayed in reads per kilobase per million ( RPKM ) . Benjamini-Hochberg adjusted p-values were calculated using DESeq2 . ( G ) Western blot analysis confirming upregulation of IFN receptors , pSTAT1 , and interferon related genes , at the protein level in T21 lymphoblastoids . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 01210 . 7554/eLife . 16220 . 013Figure 5—figure supplement 1 . Biological replicates of lymphoblastoid samples are highly related . ( A ) Table of lymphoblastoid cell lines used in this study . All lymphoblastoid lines used are female . ( B ) Principal component analysis ( PCA ) of RNA-seq samples from lymphoblastoid cell lines . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 01310 . 7554/eLife . 16220 . 014Figure 5—figure supplement 2 . Differentially expressed genes in trisomy 21 lymphoblastoid cell lines are not organized into obvious chromatin domains . Manhattan plots for individual chromosomes indicating differentially expressed genes in red . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 01410 . 7554/eLife . 16220 . 015Figure 5—figure supplement 3 . Components of the IFN response are activated in a mouse model of Down syndrome . ( A ) Principal component analysis ( PCA ) of RNA-seq samples produced from lineage negative , Sca1 positive , c-kit positive ( LSK ) cells from Dp16 mice and matched littermate controls . ( B ) Box and whisker plots of RNA expression for the four IFN receptor subunits encoded on chr16 ( green ) and representative IFN-related genes from Dp16 LSK cells . mRNA expression values are displayed in reads per kilobase per million ( RPKM ) . Benjamini-Hochberg adjusted p-values were calculated using DESeq2 . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 015 We next wanted to determine if the IFN signature was conserved in a mouse model of Down syndrome . Dp16 mice were selected because they contain a region of mouse chromosome 16 syntenic to human chromosome 21 that includes the IFN receptor cluster , without triplication of non-syntenic regions ( Li et al . , 2007 ) . RNA-seq was performed on the LSK ( Lineage negative , Sca1 positive , c-Kit positive ) population of multipotent hematopoietic stem and progenitor cells obtained from the bone marrow of Dp16 mice and matched littermate controls . These results confirmed that three of the four IFN receptors are upregulated in Dp16 mice ( Ifnar1 , Ifnar2 , and Ifngr2 ) , along with several canonical ISGs ( Figure 5—figure supplement 3 , Supplementary file 1C ) . Our results demonstrate that IFN activation by trisomy 21 is conserved in the hematopoietic lineage . In order to determine whether our findings are applicable to living human individuals with T21 , we isolated monocytes , T cells , and B cells , from 10 individuals with T21 and seven D21 individuals . As for our cell line work , we included samples from both sexes with varying ages and genetic backgrounds ( Figure 6—figure supplement 1A , B ) . Monocytes and T cells were subjected to transcriptome analysis by RNA-seq , and B cells used for IFN receptor surface expression analysis by flow cytometry . The transcriptome analyses identified hundreds of consistent gene expression changes associated with T21 in both cell types , with the expected ~1 . 5x fold increase in chr21 gene expression ( Figure 6—figure supplement 1C , D ) . The IFN receptors encoded on chr21 are significantly upregulated in circulating blood cell types from individuals with T21 , with the sole exception of IFNGR2 in T cells ( Figure 6A , B , Supplementary file 1D ) . Flow cytometry detected a minor increase in surface expression of IFNAR1 , IFNGR2 , and IL10RB , in the B cell population , but not for IFNGR1 , which is not encoded on chr21 ( Figure 6—figure supplement 2 ) . Once again , upstream regulator analysis identified IFN ligands and IFN-activated transcription factors as predicted drivers of gene induction in T21 monocytes and T cells ( Figure 6C and Figure 6—figure supplement 3 ) with many canonical ISGs scoring among the most significantly induced genes ( Figure 6A , B ) . 10 . 7554/eLife . 16220 . 016Figure 6 . IFN signaling is activated in circulating blood cells from individuals with T21 . ( A ) Box and whisker plots of RNA expression for the four IFN receptor subunits encoded on chr21 and representative IFN-related genes in circulating monocytes . mRNA expression values are displayed in reads per kilobase per million ( RPKM ) . Benjamini-Hochberg adjusted p-values were calculated using DESeq2 . ( B ) Box and whisker plots of RNA expression as in ( A ) for circulating T cells . ( C ) Upstream regulator analysis reveals activation of the IFN transcriptional response in T21 monocytes and T cells , as well as downregulation of the MYCN-driven transcriptional program . ( D ) Canonical pathway analysis reveals activation of the IFN signaling pathway in T21 monocytes and T cells , as well as downregulation of the EIF2 signaling pathway . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 01610 . 7554/eLife . 16220 . 017Figure 6—figure supplement 1 . Effects of T21 on the transcriptome of circulating monocytes and T cells from individuals with T21 and typical controls . ( A ) Description of samples from individuals with T21 and typical controls used in this study . ( B ) Principal component analysis ( PCA ) of monocyte and T cell RNA-seq samples . ( C ) MA plots displaying the results of RNA-seq analysis for monocytes and T cells . Differentially expressed genes ( DEGs ) , as defined by DEseq2 ( FDR < 10% ) , are labeled in red . ( D ) Volcano plots of data from monocytes and T cells highlight changes in chr21 gene expression ( green ) consistent with increased gene dosage effects . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 01710 . 7554/eLife . 16220 . 018Figure 6—figure supplement 2 . Surface expression of IFN receptors is increased in B cells from individuals with T21 . Flow cytometric analysis of surface expression of three chr21-encoded IFN receptors ( in green ) and one encoded on another chromosome ( IFNGR1 ) in B cells isolated from the same individuals as monocytes and T cells used in Figure 6 . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 01810 . 7554/eLife . 16220 . 019Figure 6—figure supplement 3 . The IFN gene signature from monocytes and T cells is largely encoded by non-chr21 genes . IPA upstream regulator analysis of all DEGs , non-chr21 DEGs , and chr21 DEGs , for monocytes and T cells . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 019 A comparison of the upstream regulator analyses of the four cell types included in this study revealed both conserved and cell type-specific features . The upstream regulator analysis shows that IFN activation is conserved , as is predicted inactivation of the IFN repressors MAPK1 and TRIM24 ( Figure 6C ) . However , a unique feature of the primary cell types -monocytes and T cells- is a predicted inactivation of the gene expression program driven by the transcription factor MYCN ( Figure 6C ) . Comparison of the canonical pathways deregulated in all four cell types confirms that IFN signaling is the top activated pathway , but also reveals that monocytes and T cells , and to a lesser degree lymphoblastoids , show strong repression of the EIF2 pathway ( Figure 6D ) . Since both MYCN and EIF2 are potent regulators of protein synthesis , we decided to investigate this observation in more detail . A well-established aspect of the IFN response is the selective control of protein translation , purportedly to prevent the synthesis of viral proteins during the course of infection ( Johnson et al . , 1968 ) . Mechanistically , it has been shown that IFN signaling impairs processing of rRNAs and controls the activity and/or expression of specific translation factors ( Walsh et al . , 2013; Maroun , 1978 ) . On the other hand , the MYC family of transcription factors are known drivers of ribosome biogenesis , protein synthesis and cell growth ( van Riggelen et al . , 2010; Boon et al . , 2001; Kim et al . , 2000; Arabi et al . , 2005 ) . Similarly , the EIF2 pathway is a key driver of protein translation , with eIF2 itself being an essential translation initiation factor ( Hinnebusch , 2014 ) . Analysis of the gene signatures identified by IPA that predicted inactivation of both the MYCN transcriptional program and the EIF2 pathway showed a substantial degree of overlap ( Figure 7A , C , Supplementary file 1E ) . In monocytes and T cells , the genes common between the two repressed programs encode components of both the small and large ribosome subunits ( i . e . RPS proteins in the 40S complex and RPL proteins in the 60S complex ) ( Figure 7A , C , Figure 7—figure supplements 1 and 2 ) . Genes exclusive to the MYCN signature are enriched for metabolic enzymes and translation elongation factors ( EEFs ) . Genes exclusive to the EIF2 signature are enriched for translation initiation factors ( EIFs ) and additional ribosomal proteins . Examples of RPSs , RPLs , EEFs and EIFs downregulated in trisomy 21 cells are shown in Figure 7B and D ( see also Figure 7—figure supplements 1 and 2 ) . This result is consistent with reports that interferon treatment results in a global decrease in expression of the translation machinery in primary PBMCs ( Taylor et al . , 2007; Gupta et al . , 2012 ) . Altogether , these results indicate that T21 causes a general downregulation of dozens of components of the protein synthesis machinery in circulating monocytes and T cells . 10 . 7554/eLife . 16220 . 020Figure 7 . Trisomy 21 globally downregulates the translation machinery in monocytes and T cells . ( A ) Venn diagram demonstrating the overlap in DEGs comprising the MYCN upstream regulator and EIF2 signaling pathway gene signatures identified by IPA in monocytes . Prominent components of each group are indicated with arrows . See also Figure 7—figure supplement 2 . ( B ) Box and whisker plots of RNA expression for representative translation-related genes from monocytes . mRNA expression values are displayed in reads per kilobase per million ( RPKM ) . Benjamini-Hochberg adjusted p-values were calculated using DESeq2 . ( C ) Venn diagram demonstrating the overlap in DEGs as in ( A ) for T cells . ( D ) Box and whisker plots of RNA expression as in ( C ) for T cells . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 02010 . 7554/eLife . 16220 . 021Figure 7—figure supplement 1 . The MYCN transcriptional program is downregulated by T21 . A heatmap demonstrates downregulation of numerous components of the translational machinery associated with MYCN-driven transcription in monocytes from individuals with T21 . The data presented are the fold change of the RPKM of each sample relative to the mean RPKM of all D21 individuals . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 02110 . 7554/eLife . 16220 . 022Figure 7—figure supplement 2 . The EIF2 Signaling pathway is downregulated by T21 . A heatmap demonstrates downregulation of numerous components of the translation machinery associated with EIF2 Signaling in monocytes from individuals with T21 . The data presented are the fold change of the RPKM of each sample relative to the mean RPKM of all D21 individuals . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 022 Having performed transcriptome analysis of cell types of different origins , we investigated to what degree the gene expression changes caused by T21 are affected by cell type-specific regulatory landscapes . A principal component analysis ( PCA ) shows the fibroblast transcriptomes segregating strongly ( PC1 80 . 5% ) from those of the cell types of hematopoietic origin ( Figure 8A ) . B cell-derived lymphoblastoids and T lymphocytes cluster together , yet they segregate away from the monocytes of myeloid origin ( PC2 , 11 . 3% ) . Within this background , the global impact of the trisomy on the transcriptome is secondary to the effects of the cell type of origin ( Figure 8B ) . Next , we asked to what degree genes encoded on chr21 could be affected by these cell type-specific regulatory landscapes . Indeed , it was easy to identify many chr21 genes displaying obvious differences in relative expression among cell types . For example , APP is relatively more highly expressed in fibroblasts , U2AF1 more highly expressed in lymphoblastoids , ETS2 more highly expressed in monocytes , and DYRK1A more highly expressed in T cells ( Figure 8C , Supplementary file 1F ) . The IFN receptors on chr21 also showed some degree of cell type-specific expression ( e . g . IFNAR2 lowly expressed in fibroblasts , IFNGR2 lowly expressed in T cells , Figure 8D ) . Furthermore , relative differences in cell type-specific expression is also evident for canonical ISGs ( Figure 8E ) . These observations led us to ask to what degree the IFN transcriptional response elicited by T21 is conserved across cell types . To address this , we compared the DEGs comprising the T21-induced Interferon alpha signature identified by IPA in each cell type ( Figure 6C ) . Remarkably , this exercise revealed a large degree of cell type-specificity , with most IFN-related genes being differentially expressed in only one cell type ( Figure 8F ) . In fact , the only common genes among all four signatures are three IFN-α-related genes encoded on chr21: IFNAR1 , IFNAR2 , and MX1 . Expectedly , lymphoblastoids and T cells showed a greater degree of overlap than other pairwise comparisons . Overall , these results indicate that while T21 operates within , and is modulated by , cell type-specific regulatory landscapes , it nonetheless activates the IFN transcriptional response consistently by inducing different gene sets within this program . This is in stark contrast to the notion that T21 affects gene expression either stochastically or through large rearrangements of chromatin domains . In fact , Manhattan plots of the DEGs in monocytes and T cells derived from the same individuals not only confirm the absence of large domains of chromatin deregulation , but also highlight the high degree of cell type-specific changes caused by the trisomy ( Figure 8G ) . 10 . 7554/eLife . 16220 . 023Figure 8 . Trisomy 21 activates the IFN gene expression program in a cell type-specific manner . ( A ) Principal component analysis ( PCA ) of all RNA-seq samples from this study colored by cell type . ( B ) PCA analysis as in ( A ) colored by chr21 copy number . ( C ) Box and whisker plots of RNA expression for representative chr21-encoded genes from all samples . mRNA expression values are displayed in reads per kilobase per million ( RPKM ) . Benjamini-Hochberg adjusted p-values were calculated using DESeq2 by comparing all T21 samples to all D21 samples . Individual data points are colored by cell type . ( D , E ) Box and whisker plots as in ( C ) for chr21-encoded IFN receptors and representative ISGs . ( F ) Venn diagram showing the cell type-specificity of the Interferon alpha gene expression programs identified by IPA for each cell type . ( G ) Manhattan plots for chromosomes 19 and 21 comparing the DEGs from monocytes and T cells derived from the same individuals . DOI: http://dx . doi . org/10 . 7554/eLife . 16220 . 023 We report here that T21 leads to consistent activation of the IFN pathway . As discussed below , IFN hyperactivation could explain many of the developmental and clinical impacts of T21 . In fact , we posit that Down syndrome can be understood largely as an interferonopathy , and that the variable clinical manifestations of T21 could be explained by inter-individual differences in adaptation to chronic IFN hyperactivity . The link between IFN signaling and T21 is not entirely unprecedented . More than 40 years ago , it was found that human T21 fibroblasts , but not those trisomic for chr13 or chr18 , have increased sensitivity to IFN exposure and are more resistant to viral infection ( Tan et al . , 1974a , 1974b ) . In fact , somatic cell hybrid experiments showed that chr21 is sufficient to confer sensitivity to human IFN in mouse cells ( Slate et al . , 1978 ) . Pioneering work by Maroun and colleagues using an early mouse model of DS carrying an extra copy of chr16 that harbors orthologues of many human chr21 genes , including the four IFN receptors , clearly implicated IFN as a contributor to the deleterious effects of the trisomy . For example , treatment of pregnant female mice with anti-IFN antibodies resulted in the partial rescue of embryonic growth defects and embryonic lethality ( Maroun , 1995 ) . Furthermore , partial normalization of gene dosage for the IFN receptor subunits via gene knockout was shown to improve embryonic development and survival of T21 cortical neurons in vitro ( Maroun et al . , 2000 ) . More recently , a study found global disruption of IFN-related gene networks in the brains of the Ts1Cje mouse model of DS , which also carries triplication of the IFN receptor subunits ( Ling et al . , 2014 ) . However , deeper investigations of IFN signaling in human T21 cells and tissues are largely absent from the literature of the past 30 years , with a few exceptions , such as the description of IFN signaling as a contributor to periodontal disease in DS ( Tanaka et al . , 2012; Iwamoto et al . , 2009 ) . Collectively , these reports and the genomics analyses reported here demonstrate that activation of the IFN pathway in T21 cells is a widespread phenomenon that occurs in diverse tissues , and that is relevant to human Down syndrome as well as the various mouse models of DS with triplication of IFN receptors . Constitutive activation of IFN signaling could conceivably explain a large number of comorbidities associated with DS , such as the increased risk of transient myeloproliferative disorder , diverse leukemias , several autoimmune disorders ( Richardson et al . , 2011 ) , and perhaps even the lower rate of solid tumors ( Zitvogel et al . , 2015; Hasle et al . , 2016 ) . Importantly , several JAK inhibitors are either approved or being tested in clinical trials for the treatment of several conditions associated with DS –albeit in the typical population– , including myeloproliferative , inflammatory and autoimmune disorders , as well as leukemia ( Padron et al . , 2016; Spaner et al . , 2016; Tefferi et al . , 2011; Quintás-Cardama et al . , 2010; Shi et al . , 2014; Keystone et al . , 2015; Jabbari et al . , 2015 ) . It should be noted , however , that the dose limiting toxicities of JAK inhibitors , like ruxolitinib , are anemia and thrombocytopenia ( McKeage , 2015; Plosker , 2015 ) . Therefore , rigorous clinical investigations will be required to define if there is a therapeutic window in which these drugs would benefit individuals with DS before the appearance of toxicity . Additional research will also be required to elucidate the interplay between hyperactive IFN signaling in DS with other important factors encoded on chr21 ( e . g . DYRK1A , APP ) ( Malinge et al . , 2012; Wiseman et al . , 2015 ) or elsewhere in the genome , that have been involved in the development of the specific comorbidities . For example , the Sonic Hedgehog ( SHH ) pathway has been implicated in the etiology of structural and cognitive defects in a mouse model of DS , including cerebellar atrophy ( Das et al . , 2013 ) . Interestingly , IFN signaling has been shown to crosstalk with the SHH pathway , and cerebellar atrophy is also a hallmark of Type I Interferonopathies ( Moisan et al . , 2015; Sun et al . , 2010; McGlasson et al . , 2015; Crow and Manel , 2015 ) . Increased JAK/STAT signaling has been postulated to contribute to some of the neurological features of DS ( Lee et al . , 2016 ) . Notably , it has been reported that therapeutic exposure to interferons can produce diverse types of neurological dysfunction , including depression , cerebral palsy and spastic diplegia ( Wichers et al . , 2005; Grether et al . , 1999; Wörle et al . , 1999; Barlow et al . , 1998 ) . Furthermore , a large number of neurological conditions have been linked to deregulated IFN signaling , most prominent among them the so called Type I Interferonopathies ( McGlasson et al . , 2015; Crow and Manel , 2015 ) . Therefore , we propose that constitutive activation of the IFN pathway in the central nervous system of individuals with DS is responsible for many of the neurological problems caused by the trisomy . In particular , IFN-mediated activation of microglia could lead to neurotoxicity by several mechanisms , including serotonin depletion , generation of reactive oxygen species , and excitatory toxicity , which could potentially be ameliorated with inhibitors of the IDO1 enzyme , a key ISG ( Wichers and Maes , 2004; Wichers et al . , 2005 ) . Although much research remains to be done , it is now possible to envision early intervention strategies to ameliorate the variable ill effects of T21 by using pharmacological inhibitors of the IFN pathway . Six human fibroblast lines from individuals with trisomy 21 ( T21 ) and six approximately age- and sex-matched fibroblast lines from typical individuals ( D21 ) were obtained from the Coriell Cell Repository ( Camden , NJ ) and immortalized with hTERT as described ( Lindvall et al . , 2003 ) . EBV-immortalized lymphoblastoid lines , three T21 and three D21 , were obtained from the Nexus Clinical Data Registry and Biobank at the University of Colorado . Fibroblasts were maintained in DMEM and lymphoblastoids were maintained in RPMI medium in a humidified 5% CO2 incubator at 37°C . The media was supplemented with 10% fetal bovine serum and 1% antibiotic/antimycotic and was changed every 3–6 days . Fibroblast monolayers were serially passaged by trypsin-EDTA treatment , and lymphoblastoids were serially passaged via dilution in fresh media . Fibroblast lines used in this study are described in Figure 1—figure supplement 1 . All cell lines were confirmed mycoplasma negative by PCR as previously described ( Uphoff and Drexler , 2002 ) . T21 status was authenticated as described in Figure 1—figure supplement 1D . Research Resource Identifiers ( RRIDs ) for fibroblast cell lines are: LineRRID #GM08447CVCL_7487GM05659CVCL_7434GM00969CVCL_7311GM02036CVCL_7348GM03377CVCL_7384GM03440CVCL_7388GM04616CVCL_V475AG05397CVCL_L780AG06922CVCL_X793GM02767CVCL_V469AG08941CVCL_X871AG08942CVCL_X872 Recombinant human interferons alpha 2A ( 11101–2 , R&D Systems ) , beta ( 300-02BC , Peprotech ) and gamma ( PHC4031 , Gibco ) were obtained from Thermo Fisher Scientific ( Waltham , MA ) , aliquoted , and stored at −80°C . Three T21 fibroblast lines and their age- and sex-matched D21 fibroblast counterparts were plated at equivalent densities and grown 72 hr to ensure similar cycling of the cells , then re-plated at equivalent densities and incubated overnight . Media was removed the following day and replaced with media containing the indicated doses of interferon ligands dissolved in PBS or vehicle ( PBS alone ) . All media were normalized for final PBS concentration at highest interferon dose . Cells were grown an additional 24 hr after interferon application , then the media removed , cells washed with PBS and harvested via cell scraping . The harvested cells were pelleted and lysed in RIPA buffer with protease and phosphatase inhibitors . Ruxolitinib ( INCB018424 ) was obtained from Selleck Chemicals ( Houston , TX , S1378 ) and dissolved in DMSO to make a 5 mM stock solution and stored at −20°C . Fibroblast lines were plated at equivalent cell numbers and allowed to grow for 72 hr in order to condition the media with secreted factors . The conditioned media was harvested and stored at 4°C for 3–7 days prior to use . One T21 fibroblast line and its age- and sex-matched D21 fibroblast counterpart were plated at equivalent cell numbers in their respective conditioned media and incubated overnight . Plating media was removed the following day and replaced with conditioned media containing the indicated doses of ruxolitinib or DMSO . All conditioned drug media was normalized for DMSO concentration . Cells were grown an additional 72 hr after drug application , harvested with trypsin-EDTA , and counted with 0 . 2% trypan blue using a hemocytometer . Cells were plated at equal densities and allowed to grow 72 hr before harvesting cell pellets . Pellets were washed with PBS and resuspended in RIPA buffer containing 1 μg/mL pepstatin , 2 μg/mL aprotonin , 20 μg/mL trypsin inhibitor , 10 nM leupeptin , 200 nM Na3VO4 , 500 nM phenylmethylsulfonyl fluoride ( PMSF ) , and 10 μM NaF . Suspensions were sonicated at six watts for 15 s two times and clarified by centrifugation at 21 , 000 g for 30 min at 4°C . Supernatants were quantified in a Pierce BCA Protein Assay and diluted in complete RIPA with 4x Laemmli sample buffer . Tris-glycine SDS-polyacrylamide gel electrophoresis was used to separate 20–40 μg protein lysate , which was transferred to a 0 . 2 μm polyvinylidene fluoride ( PVDF ) membrane . Membranes were blocked in 5% non-fat dried milk or 5% bovine serum albumin ( BSA ) in Tris-buffered saline containing 0 . 1% TWEEN ( TBS-T ) at room temperature for 30–60 min before probing overnight with primary antibody in 5% non-fat dried milk or 5% BSA in TBS-T at 4°C while shaking . Membranes were washed 3x in TBS-T for 5–15 min before probing with a horseradish peroxidase ( HRP ) conjugated secondary antibody in 5% non-fat dried milk or 5% BSA at room temperature for one hour . Membranes were again washed 3x in TBS-T for 5–15 min before applying enhanced chemiluminescence ( ECL ) solution . Chemiluminensence signal was captured using a GE ( Pittsburgh , PA ) ImageQuant LAS4000 . Antibodies used in this study: AntibodyManufacturerProduct #RRID #anti-mouse IgG-HRPSanta Cruz Biotechnology ( Dallas , TX ) sc-2005AB_631736anti-rabbit IgG-HRPSanta Cruz Biotechnologysc-2317AB_641182BIMCell Signaling Technology ( Danvers , MA ) 2819AB_659953GAPDHSanta Cruz Biotechnologysc-365062AB_10847862GBP5Abcam ( Cambridge , MA ) ab96119AB_10678091IFI27Abcamab171919N/AIFNAR1R&D SystemsAF245AB_355270IFNGR2R&D SystemsAF773AB_355589IL10RBR&D SystemsAF874AB_355677ISG15Cell Signaling Technology2743AB_2126201MX1Abcamab95926AB_10677452pSTAT1Cell Signaling Technology7649AB_10950970TBX21Cell Signaling Technology5214AB_10692112 Total RNA was isolated using Trizol ( Thermo Fisher ) according to manufacturer’s instructions . cDNA was synthesized using the qScript kit from Quanta Biosciences ( Beverly , MA ) . PCR was performed using SYBR Select on a Viia7 from Life Technologies ( Thermo Fisher ) . Oligonucleotides used in this study: Gene IDAccession #ForwardReverseIFI27NM_001130080TCTGCAGTCACTGGGAGCAACTAACCTCGCAATGACAGCCGCAAIFITM1NM_003641TTCGCTCCACGCAGAAAACCAACAGCCACCTCATGTTCCTCCTMX1NM_001144925TCCACAGAACCGCCAAGTCCAAATCTGGAAGTGGAGGCGGATCAMX2NM_002463TCGGACTGCAGATCAAGGCTCTCGTGGTGGCAATGTCCACGTTAOAS1NM_001032409CCGCATGCAAATCAACCATGCCTTGCCTGAGGAGCCACCCTTTAOAS2NM_001032731AGGTGGCTCCTATGGACGGAAACGAGGATGTCACGTTGGCTTCT Biological replicates for each cell line were obtained by independently growing cells in duplicate . Total RNA was purified from ~1 × 107 logarithmically growing cells using Qiagen ( Valencia , CA ) RNeasy columns per manufacturer’s instructions including on-column DNAse digestion . RNAs were quantified using a Take3 Micro-Volume plate in a Biotek ( Winooski , VT ) Synergy2 plate reader and their integrity confirmed using the Agilent RNA 6000 Pico Kit and the Agilent ( Santa Clara , CA ) 2100 Bioanalyzer System . 500 ng of total RNA with an RNA Integrity Number ( RIN ) greater than 7 were used to prepare sequencing libraries with the Illumina ( San Diego , CA ) TruSeq Stranded mRNA Library Prep Kit . Libraries were sequenced with an Illumina HiSeq 2000 System at the UCCC Genomics Core . Peripheral blood was collected in EDTA vacutainer tubes from 10 individuals with T21 and seven D21 controls . Blood was centrifuged at 500 g for 15 min to separate plasma , buffy coat and red blood cells ( RBCs ) . Peripheral Blood Mononuclear Cells ( PBMCs ) were isolated from the buffy coat fraction by RBC lysis and 1x PBS wash according to manufacturer’s instructions ( BD , 555899 ) . After RBC lysis and PBS wash , PBMCs were stained for sorts at 10–20 × 107 cells/ml then diluted to approximately 5 × 107 cells/ml in flow cytometry sorting buffer ( 1x PBS , 1 mM EDTA , 25 mM HEPES pH 7 . 0 , 1% FBS ) . All staining was performed in flow cytometry sorting buffer with fluorochrome-conjugated antibodies for at least 15 min on ice while protected from light . Single cell suspensions were stained with CD45 ( eBioscience , San Diego , CA , HI30 , RRID:AB_467273 ) , CD14 ( Biolegend , San Diego , CA , 63D3 , RRID:AB_2571928 ) , CD3 ( Biolegend , OKT3 , RRID:AB_571907 ) , CD16 ( Biolegend , B73 . 1 , RRID:AB_2616914 ) , CD19 ( Biolegend , HIB19 , RRID:AB_2073119 ) , CD56 ( Biolegend , 5 . 1H11 , RRID:AB_2565855 ) and CD34 ( Biolegend , 561 , RRID:AB_343601 ) antibodies . CD45+CD14+CD19-CD3-CD56- Monocytes and CD45+CD3+CD14-CD19-CD56- T cells were FAC-sorted into Dulbecco’s Modified Eagle Medium ( DMEM ) supplemented with 4 . 5 g/L D-Glucose , L-Glutamine , and 5% FBS , on the MoFlo Astrios ( Beckman Coulter , Brea , CA ) at the CU-SOM Cancer Center Flow Cytometry Shared Resource . FAC-sorted cells were centrifuged at 500 g for 5 min and the media removed . Cells were resuspended in 350 μl RLT plus ( Qiagen ) and Beta-mercaptoethanol ( BME ) lysis buffer ( 10 μL BME:1 mL RLT plus ) for downstream RNA isolation . Lysed cells were immediately stored at −80°C and RNA was later extracted using the AllPrep DNA/RNA/Protein Mini Kit according to manufacturer’s instructions ( Qiagen , 80004 ) . RNA quality was determined by BioAnalyzer ( Agilent ) and quantified by Qubit ( Life Technologies ) . Samples with RIN of 7 or greater and a minimum of 500 ng total RNA were used for library prep and sequencing . Analysis of library complexity and high per-base sequence quality across all reads ( i . e . q > 30 ) was performed using FastQC ( v0 . 11 . 2 ) software ( Andrews , 2010 ) . Low quality bases ( q < 10 ) were trimmed from the 3’ end of reads and short reads ( <30 nt after trimming ) and adaptor sequences were removed using the fastqc-mcf tool from ea-utils . Common sources of sequence contamination such as mycoplasma , mitochondria , ribosomal RNA were identified and removed using FASTQ Screen ( v0 . 4 . 4 ) . Reads were aligned to GRCh37/hg19 using TopHat2 ( v2 . 0 . 13 , --b2-sensitive --keep-fasta-order --no-coverage-search --max-multihits 10 --library-type fr-firststrand ) ( Kim et al . , 2013 ) . High quality mapped reads ( MAPQ > 10 ) were filtered with SAMtools ( v0 . 1 . 19 ) ( Li et al . , 2009 ) . Reads were sorted with Picardtools ( SortSAM ) and duplicates marked ( MarkDuplicates ) . QC of final reads was performed using RSeQC ( v2 . 6 ) ( Wang et al . , 2012 ) . Gene level counts were obtained using HTSeq ( v0 . 6 . 1 , --stranded=reverse –minaqual=10 –type=exon –idattr=gene --mode= intersection-nonempty , GTF-ftp://igenome:G3nom3s4u@ussd-ftp . illumina . com/Homo_sapiens/UCSC/hg19/Homo_sapiens_UCSC_hg19 . tar . gz ) ( Anders et al . , 2015 ) . Differential expression was determined using DESeq2 ( v1 . 6 . 3 ) and R ( 3 . 10 ) ( Love et al . , 2014 ) . Volcano plots , manhattan plots , and violin plots , were made using the Python plotting library 'matplotlib' ( http://matplotlib . org ) . A pool of plasmids encoding 3 , 075 shRNAs targeting 654 kinases ( kinome library ) in the pLKO . 1 backbone produced by The RNAi Consortium ( TRC , Sigma-Aldrich , St . Louis , MO ) were obtained from the University of Colorado Cancer Center Functional Genomics Shared Resource , as were the pΔ8 . 9 and pCMV-VSV-G lentiviral packaging plasmids . 2 μg of kinome library plasmid DNA at 100 ng/μL was mixed with 2 μg of packaging plasmid mix ( at a 9:1 ratio of pΔ8 . 9:pCMV-VSV-G ) at 100 ng/μL and incubated with 12 μg of Polyethylenimine for 15 min at RT . The entire mixture was then added to 3 × 105 HEK293FT packaging cells to give 100X coverage . 16 hr after transfection , media on cells was replaced with complete DMEM . 24 hr after media replacement , target cells were seeded at 1 × 105 cells/ well in a 6-well plate . Three wells for each line were combined at the time of harvest to reach a starting number of 3 × 105 cells per condition ( again 100X coverage of the kinome library ) . 24 hr after seeding , the media from each well of packaging cells ( now containing lentiviral library particles ) was filtered through 0 . 45 μm cellulose acetate filters , diluted 1:3 into 6 mL of DMEM , and mixed with 6 μL of 8 mg/mL polybrene to facilitate transduction . This mixture was then used to transduce 3 wells ( one total replicate ) of each target cell line . 24 hr after transduction viral transduction , the media was replaced with fresh media . Finally , after an additional 24 hr , selection began by adding fresh DMEM with 1 μg/mL puromycin . Cells were then propagated for 14 days and genomic DNA harvested from all remaining cells using the Qiagen DNeasy Blood and Tissue kit with the optional RNAse A treatment step . Genomic DNA was quantified by A260 using a Take3 micro-volume plate on a Synergy2 Microplate Reader . The quality of the genomic DNA was confirmed via electrophoresis on a 0 . 5% TAE agarose gel . Screens were performed in three independent biological replicates for each of the 12 fibroblast cell lines . The library preparation strategy uses genomic DNA and two rounds of PCR in order to isolate the shRNA cassette and prepare a single strand of the hairpin for sequencing by means of an XhoI restriction digest in the stem loop region . This is critical as the hairpin secondary structures of shRNAs are not amenable to NGS and the TRC shRNAs do not have a long enough loop to allow PCR amplification of one shRNA arm in a single step . The first step in sequencing library preparation is to calculate how much genomic DNA must be used for PCR1 which isolates and amplifies the shRNA cassettes from genomic DNA using Phusion Polymerase . The oligonucleotides for PCR1 anneal to regions inside of the LTRs that are common to all clones in the library and should , therefore , amplify all shRNA cassettes with equal efficiency . Each reaction mixture for PCR1 consisted of 10 μL 5X Phusion HF buffer , 1 μL dNTPs ( 10 mM each ) , 2 . 5 μL pLKO Forward and Reverse primers ( 10 μM ) , 1 μL of 2 unit/μl Phusion Polymerase , 500 ng genomic DNA , and dH2O to 50 μL . The cycling conditions were as follows: 1 cycle of 98°C for 5 min , 15 to 25 cycles of 98°C for 30 s , 70°C for 30 s , 72°C for 30 s , and 1 cycle of 72°C for 7 min . 5 μL of each PCR1 were run on a 2% TAE agarose gel in order to visualize the expected band of 497 bp . It should be noted that the optimal PCR1 cycle number must be empirically determined for each library and to limit cycle numbers to minimize the effects of amplification bias . The correct product of PCR1 is 497 bp; however , excessive cycle numbers can result in the appearance of a slower migrating band . This band represents an annealing event between two amplification products with different shRNA sequences . As the majority of the 497 bp amplicon is common to all products , denatured PCR products can anneal to one another when not out-competed by an excess of primer in later cycles . This aberrant product does not correctly anneal within the central shRNA-containing sequence , therefore disrupting the double-stranded XhoI site required for the subsequent restriction digestion . Carefully determining the appropriate number of cycles prevents the appearance of this undesired product . After establishing an optimal cycle number , we performed 12 identical PCR1 reactions in order to amplify sufficient amounts of genomic DNA and pooled them all prior to cleanup with a QIAquick PCR Purification Kit . 1 μg of the resulting DNA was digested with XhoI overnight at 37°C . Digest reactions consisted of 3 . 5 μL 10X FD buffer , 1 μL of 20 , 000 units/mL XhoI , 1 μg of DNA and dH2O to 35 μL . Heat inactivation of XhoI is not recommended , as the high temperatures result in reappearance of the spurious annealing products mentioned above , leading to a disruption of the XhoI overhang required for ligation . For the TRC1 and TRC1 . 5 libraries , there are two XhoI sites within the product of PCR1 , resulting in fragments of 271 , 43 and 183 bp . In order to purify the desired fragment , the entire digest was run on a 2% TAE agarose gel and purified the 271 bp fragment using a QIAquick Gel Extraction Kit . Once the band was excised , three volumes of buffer QG were added and the mixture heated at 30°C to dissolve the agarose . Lower melting temperatures are recommended so as not to denature the complementary double-stranded shRNA cassettes , which may not reanneal to their cognate strand . After the agarose was dissolved , one volume of isopropanol was added and protocol resumed following the manufacturer’s instructions including the optional addition of NaOAc . We prepared the barcoded linkers required for ligation by resuspending the lyophilized oligonucleotides in ST buffer ( 10 mM Tris pH 8 . 0 , 50 mM NaCl ) to 200 μM and combining 25 μL of each for a final concentration of 100 μM . The mixture was heated to 94°C for 10 min and gradually cooled to ensure proper annealing . Single-stranded oligonucleotides were removed from annealed oligonucleotides using Illustra MicroSpin G-25 columns . The sense ( S1-S4 ) oligonucleotides are 5’-phosphorylated and the antisense oligonucleotides ( AS1-AS4 ) each contain a single phosphorothioate bond at the 3’ end to stabilize them and are designed to prevent the reformation of a functional XhoI site . The barcodes within these linkers are used for multiplexing and their length ensures they are compatible with the Illumina HiSeq 2000 . Shorter barcode sequences may be compatible with other sequencing platforms . The selected barcoded linkers were added to ligation reactions with 100 ng of each purified 271 bp XhoI fragment , 3 . 5 μL 10X T4 DNA ligase buffer , 4 μL of 1 μM barcoded linker , 1 μL T4 DNA ligase and dH2O to 35 μL . Ligations were performed overnight at 16°C . The entire ligation was run on a 2% TAE agarose gel and the resulting 312 bp band purified using the QIAquick Gel Extraction Kit in the same manner as previously described . The final step in the preparation of the sequencing library is a second PCR with oligonucleotides that contain the Illumina adaptors required for bridge amplification and sequencing . In this PCR , the number of cycles is minimized in order to avoid PCR bias as well as errors that could affect sequencing . The reaction for PCR2 was as follows: 10 μL 5X Phusion HF buffer , 1 μL dNTPs ( 10 mM each ) , 2 . 5 μL Forward adapter primer ( 10 μM ) 2 . 5 μL , Reverse adapter primer ( 10 μM ) , 1 μL Phusion DNA polymerase 10 ng barcoded DNA , and dH2O to 50 μL . The cycling program consisted of 1 cycle of 98°C for 2 min , 2 cycles of 98°C for 30 s , 62°C for 30 s , 72°C for 30 s , 7 cycles of 98°C for 30 s , 72°C for 30 s and 1 cycle of 72°C for 3 min . The final 141 bp product was purified on a 2% TAE-agarose gel followed by QIAquick Gel Extraction as described above . We assessed the purity of our sequencing library using the Bioanalyzer High Sensitivity DNA Kit ( Agilent-5067-4626 ) and confirmed the presence of a single 141 bp peak , indicating one PCR product at the appropriate size . We utilized a multiplexing strategy consisting of four different barcodes with each nucleotide represented at each position of the barcode , allowing us to sequence four samples in each lane on a HiSeq 2000 Illumina instrument . To accomplish this , each sample was quantified and mixed together at a final concentration of 10 ng/μL and using Illumina-specific oligonucleotides and qPCR , we determined the cluster formation efficiency ( i . e . effective concentration ) of our library to be slightly greater than that of a known library . Accordingly , we loaded the flow cell at 5 pM and included a 10% ΦX-174 spike-in , which aids in quality control of cluster formation and sequencing on the Illumina platform . Cluster formation efficiency and the concentration of library to be loaded on the flow cell needs to be determined empirically for each library preparation . These loading conditions yielded cluster densities between 733 , 000 clusters/mm2 and 802 , 000 clusters/mm2 and between 203 and 222 million reads per lane . shRNA data were analyzed in a similar fashion to RNA-seq data . Briefly , quality control was performed with FastQC , reads were trimmed to include only shRNA sequences using FASTQ trimmer , and filtered with the FASTQ Quality Filter . Reads were then aligned to a custom reference library of shRNA sequences using TopHat2 . Three out of 36 samples were removed based on poor performance in unsupervised hierarchical clustering and/or principal component analysis , but each fibroblast cell line retained at least two biological replicates and nine of 12 retained all three replicates . Count tables were generated using HTSeq and differential expression determined by DESeq2 . Cell lysates from all 12 fibroblast cell lines were analyzed using SOMAscan v4 . 0 according to manufacturer’s instructions and as previously reported ( Hathout et al . , 2015; Mehan et al . , 2014 ) . Data were analyzed using the QPROT statistical package ( Choi et al . , 2015 ) . The whole bone marrow was harvested from the long bones of Dp16 mice ( RRID:IMSR_JAX:013530 ) and matched littermate controls . Cells were first purified using hemolysis to remove RBCs and then stained and sorted for LSK cells ( CD3- , Ter119- , Mac1- , Gr1- , B220- , Sca1+ , cKit ) using the Moflo XDP 70 FACS sorter . RNA was then isolated from these cells using the RNeasy Kit from Qiagen .	Our genetic information is contained within structures called chromosomes . Down syndrome is caused by the genetic condition known as trisomy 21 , in which a person is born with an extra copy of chromosome 21 . This extra chromosome affects human development in many ways , including causing neurological problems and stunted growth . Trisomy 21 makes individuals more susceptible to certain diseases , such as Alzheimer’s disease and autoimmune disorders – where the immune system attacks healthy cells in the body – while protecting them from tumors and some other conditions . Since cells with trisomy 21 have an extra copy of every single gene on chromosome 21 , it is expected that these genes should be more highly expressed – that is , the products of these genes should be present at higher levels inside cells . However , it was not clear which genes on other chromosomes are also affected by trisomy 21 . Sullivan et al . aimed to identify which genes are affected by trisomy 21 by studying samples collected from a variety of individuals with , and without , this condition . Four genes in chromosome 21 encode proteins that recognize signal molecules called interferons , which are produced by cells in response to viral or bacterial infection . Interferons act on neighboring cells to regulate genes that prevent the spread of the infection , shut down the production of proteins and activate the immune system . Sullivan et al . show that cells with trisomy 21 produce high levels of genes that are activated by interferons and lower levels of genes required for protein production . In other words , the cells of people with Down syndrome are constantly fighting a viral infection that does not exist . Constant activation of interferon signaling could explain many aspects of Down syndrome , including neurological problems and protection against tumors . The next steps are to fully define the role of interferon signaling in the development of Down syndrome , and to find out whether drugs that block the action of interferons could have therapeutic benefits .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "short", "report" ]	2016	Trisomy 21 consistently activates the interferon response
During meiosis , crossover recombination is essential to link homologous chromosomes and drive faithful chromosome segregation . Crossover recombination is non-random across the genome , and centromere-proximal crossovers are associated with an increased risk of aneuploidy , including Trisomy 21 in humans . Here , we identify the conserved Ctf19/CCAN kinetochore sub-complex as a major factor that minimizes potentially deleterious centromere-proximal crossovers in budding yeast . We uncover multi-layered suppression of pericentromeric recombination by the Ctf19 complex , operating across distinct chromosomal distances . The Ctf19 complex prevents meiotic DNA break formation , the initiating event of recombination , proximal to the centromere . The Ctf19 complex independently drives the enrichment of cohesin throughout the broader pericentromere to suppress crossovers , but not DNA breaks . This non-canonical role of the kinetochore in defining a chromosome domain that is refractory to crossovers adds a new layer of functionality by which the kinetochore prevents the incidence of chromosome segregation errors that generate aneuploid gametes . The formation of haploid reproductive cells during meiosis relies on the accurate segregation of chromosomes during two meiotic divisions ( meiosis I and II ) . Faithful segregation of homologous chromosomes during meiosis I is contingent on inter-homologue linkages that are established during the preceding G2/prophase . These linkages ( chiasmata ) are the final outcome of programmed DNA break formation and crossover ( CO ) repair . Improper placement of COs in the vicinity of centromeres negatively influences meiotic chromosome segregation ( Hassold and Hunt , 2001; Koehler et al . , 1996; Rockmill et al . , 2006 ) , so CO formation close to centromeres is infrequent in many species , including humans ( Centola and Carbon , 1994; Copenhaver et al . , 1999; Ellermeier et al . , 2010; Gore et al . , 2009; Lambie and Roeder , 1986; Mahtani and Willard , 1998; Nakaseko et al . , 1986; Puechberty et al . , 1999; Saintenac et al . , 2009; Tanksley et al . , 1992 ) . However , the mechanisms that control DNA break formation and CO repair close to centromeres remain poorly understood ( Choo , 1998; Talbert and Henikoff , 2010 ) . Centromeres are functionally conserved but structurally diverse , ranging from the simple so-called “point” centromeres of budding yeasts to a variety of more complex “regional” or even holocentric centromeres in other eukaryotes ( Allshire and Karpen , 2008 ) . Point centromeres are defined by a short ~125bp sequence upon which the kinetochore assembles , while regional centromeres are typically comprised of specialized centromeric chromatin interspersed with blocks of heterochromatin . In fission yeast and Drosophila , the integrity of pericentromeric heterochromatin was found to repress double-strand break ( DSB ) formation and centromere-proximal recombination ( Ellermeier et al . , 2010; Westphal and Reuter , 2002 ) . Although budding yeast lack pericentromeric heterochromatin , suppression of centromere-proximal recombination is also observed in this organism ( Lambie and Roeder , 1986 , Chen et al . , 2008 ) . Moreover , pericentromeric CO suppression has been observed in situations where centromere position is uncoupled from its associated heterochromatin , such as in Drosophila strains carrying translocated chromosomes ( Mather , 1939 ) . These observations suggest the existence of a fundamental mechanism of recombination suppression that functions independently of associated heterochromatin . Genome-wide DSB maps in budding yeast have inferred that the centromere exerts a zone of inhibition of meiotic DSB formation , the activity of which decreases over a distance of approximately 10 kb ( Blitzblau et al . , 2007; Buhler et al . , 2007; Pan et al . , 2011 ) . Excision of a centromere relieved this DSB suppression , indicating that the centromere , or its associated factors , exert this effect ( Robine et al . , 2007 ) . The synaptonemal complex component , Zip1 ( Chen et al . , 2008 ) , and the Bloom’s helicase , Sgs1 ( Rockmill et al . , 2006 ) , which influences repair pathway choice , are also known to minimize centromere recombination . However , both proteins affect recombination globally , acting at a step after DSB formation , and are not specifically localized at centromeres . Instead , centromere-bound factors are likely to dictate the region of recombination suppression in the surrounding pericentromere through mechanisms that remain unclear . Candidate centromere-bound factors for the repression of pericentromeric recombination are components of the kinetochore , a sophisticated multi-subunit protein complex nucleated by centromeric chromatin ( reviewed in Biggins ( 2013 ) ; Cheeseman , ( 2014 ) ) . Within kinetochores , multiple generally conserved sub-complexes can be recognized that perform specific roles . Outer kinetochore sub-complexes together form an interface with microtubules and serve as a platform for spindle assembly checkpoint signaling , coupling chromosome-microtubule interactions with cell cycle progression . Inner kinetochore sub-complexes direct assembly of the outer kinetochore . Several kinetochore subcomplexes together assemble into a Constitutive Centromere-Associated Network ( CCAN; also known as the Ctf19 complex in budding yeast ) ( reviewed in McAinsh and Meraldi ( 2011 ) ; Westermann and Schleiffer ( 2013 ) ) As its name implies , the CCAN/Ctf19 complex is bound to centromeric chromatin throughout the mitotic or meiotic cell division program . In meiotic G2/prophase of budding yeast , when recombination occurs , only the Ctf19 and Mis12/MIND ( Mtw1 including Nnf1-Nsl1-Dsn1 ) kinetochore complexes are bound to the centromere ( Meyer et al . , 2015; Miller et al . , 2012 ) . The Ctf19 complex exerts long range effects by promoting cohesin enrichment throughout the ~20–50 kb surrounding pericentromere despite being restricted to the core ~125 bp centromere sequence ( Eckert et al . , 2007; Fernius and Marston , 2009; Ng et al . , 2009 ) . It does so by targeting the Scc2/4 cohesin loader to the centromere , from where cohesin spreads into the pericentromere ( Fernius et al . , 2013 ) . These characteristics make the Ctf19 complex a particularly good candidate for mediating kinetochore-derived recombination suppression . Here we show that both cohesin-independent suppression of DSB formation and cohesin-dependent repair pathway choice underlie a central role for the Ctf19 complex in suppression of CO formation in the pericentromere . To understand how pericentromeric COs are prevented , we used a fluorescent CO reporter assay ( Thacker et al . , 2011 ) ( Figure 1A ) to measure recombination rates within a pericentromere ( around CEN8 , the centromere of chromosome VIII ) or , as a control , on a chromosome arm interval of equivalent size on chromosome VIII of budding yeast ( Figure 1B , C ) . In wild-type cells , map distance , a measure of CO frequency , was 7 . 5 cM within the arm interval but only 0 . 04 cM within the pericentromere interval . In cells lacking the synaptonemal component , Zip1 , map distance within the pericentromeric interval rose to ~2 cM ( Figure 1B ) , in agreement with previous observations ( Chen et al . , 2008 ) , while we observed a modest decrease in map distance within the chromosomal arm region ( Figure 1C ) . Thus , the fluorescent reporter assay can report on pericentromeric CO formation . 10 . 7554/eLife . 10850 . 003Figure 1 . The Ctf19 kinetochore sub-complex represses pericentromeric meiotic recombination . ( A ) Scheme of meiosis and the live cell reporter assay to measure CO recombination . Homologous chromosomes are shown in light and dark blue with Green fluorescent protein ( GFP ) , tdTomato ( RFP ) and m-Cerulean ( CFP ) reporters represented in green , red and cyan , respectively . Expression of reporters in spores leads to segregation of coloured markers as indicated in the images of live tetrads . ( B , C ) Map distances ( centiMorgans ( cM ) ) and standard error ( bars ) were determined for a ~10 kb pericentromeric ( B ) or chromosomal arm ( C ) interval as described in Materials and methods . p-values were obtained using Fisher’s exact test ( * p<0 . 05; ** p<0 . 0001 ) . ( D ) Schematic representation of the kinetochore showing yeast Ctf19 sub-complex components with superscripts indicating the centomere protein ( CENP ) equivalent in humans . Proteins essential for vegetative growth or proper spore viability after meiosis are shown in dark and light blue , respectively . CO , crossover; n . d . , not determined . DOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 00310 . 7554/eLife . 10850 . 004Figure 1—figure supplement 1 . Partial inter-dependence of the Ctf19 complex and MIND subunits . ( A ) The Ctf19 component , Mcm21 , is required for proper levels of Dsn1 at the centromere during prophase I . Strains AM11633 ( no tag ) , AM20078 ( DSN1-6HIS-3FLAG ) and AM20080 ( DSN1-6HIS-3FLAG mcm21△ ) all carrying ndt80△ were harvested 5 hr after resuspension in sporulation medium to induce a prophase I arrest and Dsn1 levels at CEN4 and a chromosomal arm site were measured by anti-FLAG ChIP-qPCR . ( B ) Effect of depleting the MIND components , Mcm21 or Dsn1 , on centromeric levels of the Ctf19 component , Mcm21 . Strains AM11633 ( no tag ) , AM20296 ( MCM21-yeGFP ) , AM20295 ( MCM21-yeGFP pCLB2-3HA-DSN1 ) , AM20294 ( MCM21-yeGFP pCLB2-3HA-MTW1 ) all carrying ndt80△ were treated as in ( A ) except that anti-GFP ChIP-qPCR was performed . In ( A ) and ( B ) the mean of four biological replicates is shown with error bars indicating standard error . p values were obtained using a paired t test . * p<0 . 05 . ( C–G ) Nuclear division and efficiency of protein depletion in pCLB2-3HA-MTW1 and pCLB2-3HA-DSN1 strains . Wild type ( AM1835 ) , pCLB2-3HA-MTW1 ( AM20084 ) and pCLB2-3HA-DSN1 ( AM20082 ) strains were induced to sporulate and the percentages of binucleate and tetranucleate cells were determined after scoring 100 cells at each of the indicated times ( C–E ) . ( F , G ) The amount of 3HA-Dsn1 ( F ) or 3HA-Mtw1 ( G ) was analysed by anti-HA western immunoblot at the indicated times . Pgk1 is shown as a loading control . ChIP , Chromatin immunoprecipitation; MIND , Mtw1 including Nnf1-Nsl1-Dsn1; qPCR , quantitative polymerase chain reactionDOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 004 Next , we tested whether the kinetochore ( Figure 1D ) affects CO formation at pericentromeres . During meiotic prophase , when recombination occurs , only the MIND and Ctf19 complexes are assembled on centromeres ( Meyer et al . , 2015; Miller et al . , 2012 ) and their components associate with kinetochores at least partially independently ( Figure 1—figure supplement 1 ) . We were unable to test the requirement for the MIND complex in preventing pericentromeric recombination using the fluorescent reporter assay because depletion of two components of the essential MIND complex , Dsn1 or Mtw1 , prevented proper execution of the meiotic divisions and tetrad formation ( Figure 1—figure supplement 1 ) . Therefore , we focused on the conserved Ctf19/CCAN kinetochore complex . Using the live cell reporter assay , we observed a significantly increased frequency of pericentromeric COs in cells lacking the Ctf19 complex components Iml3CENP-L , Chl4CENP-N , Mcm21CENP-O and Ctf19CENP-P ( Figure 1B , Supplementary file 1; the gene names of the human homologues are indicated in superscript ) . This effect appeared to be specific to the pericentromere , as no significant changes in recombination were observed within the chromosomal arm interval in the absence of Iml3CENP-L , Chl4CENP-N , Mcm21CENP-O and Ctf19CENP-P ( Figure 1C , Supplementary file 2 ) . Other kinetochore subunits ( Cnn1CENP-T , Wip1CENP-W , Nkp1 , Nkp2 ) had a more modest effect on pericentromeric COs , while we found no evidence that Mhf1CENP-S and Mhf2CENP-X , which have additional roles in meiotic DNA repair together with Mph1 ( Osman and Whitby , 2013 ) , are required for suppression of pericentromeric COs . Thus , the Ctf19 inner kinetochore subcomplex affects pericentromeric meiotic recombination . We corroborated these findings by analyzing the effect of the Ctf19 complex on global meiotic recombination patterns , using high throughput sequencing to identify single nucleotide polymorphisms in the haploid progeny generated from meiosis of a hybrid yeast strain ( Oke et al . , 2014 ) ( Figure 2A ) . This method allows the detection of CO and non-crossover ( NCO ) repair products ( i . e . gene conversions ) , which can inform on altered regulation of meiotic DNA break repair ( Figure 2B ) . Poor spore viability ( typically <30% ) precluded us from using Ctf19 complex deletion mutants . Instead , we used a meiosis-specific hypomorphic depletion allele of IML3 ( pCLB2-3HA-IML3; Figure 2—figure supplement 1 ) and were able to isolate 8 four-spore-viable tetrads for high-throughput sequencing and global recombination analysis ( Supplementary file 3 ) . Although we observed no global change in the average number of COs or NCOs in the pCLB2-3HA-IML3 strain compared to wild type ( Figure 2C , D ) , the distribution of recombination was affected: the frequency of both COs and NCOs within 20 kb of centromeres was overall significantly increased ( Figure 2E , F ) . These events were detected on most , but not all , chromosomes , although we are unable to test significance for individual chromosomes due to an insufficient number of events analyzed ( Figure 2—figure supplement 2A; Supplementary file 2 ) . We note that , in the live cell recombination reporter assay , the increase in centromere-proximal COs on chromosome VIII was more modest in pCLB2-3HA-IML3 cells than iml3△ cells ( Figure 1B , C ) , indicating that analysis of the hypomorphic pCLB2-3HA-IML3 strain likely underestimates the importance of the Ctf19 complex in suppressing pericentromeric recombination events . Nevertheless , the observed effect of pCLB2-3HA-IML3 on pericentromeric recombination was greater than sgs1△ ( Figure 2—figure supplement 2B , C ) , which is known to affect meiotic recombination near centromeres ( Rockmill et al . , 2006; Oke et al . , 2014 ) . Together , these experiments demonstrate that the Ctf19 complex shapes the meiotic recombination landscape by minimizing pericentromeric CO recombination . 10 . 7554/eLife . 10850 . 005Figure 2 . Genome-wide analysis shows that a functional Ctf19 complex is required to prevent both pericentromeric COs and NCOs . ( A ) Assay to measure meiotic recombination genome-wide by analysis of SNPs after high-throughput sequencing of germinated spores resulting from hybrid meiosis . ( B ) Scheme of meiotic recombination showing CO and NCO outcomes . ( C , D ) Overall recombination levels are not affected by the pCLB2-3HA-IML3 mutation . The average numbers of COs ( C ) and NCOs ( D ) per tetrad are not significantly different between wild type and pCLB2-3HA-IML3 cells . Error bars represent standard deviation . A two-tailed t test indicated non-significance ( p>0 . 05 ) . ( E , F ) Both COs ( E ) and NCOs ( F ) within 20 kb of the pericentromere are increased in pCLB2-3HA-IML3 cells . Data for wild type is from ( Oke et al . , 2014 ) . Number of meioses scored was 8 for pCLB2-3HA-IML3 and 52 for wild type . p-values were calculated using chi-square test with Yates correction . CO , crossover; DSB , double strand break; NCO , non-crossover; SNPs , single nucleotide polymorphisms . DOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 00510 . 7554/eLife . 10850 . 006Figure 2—figure supplement 1 . Depletion of Iml3 during meiosis . Placement of IML3 under the control of the CLB2 promoter results in depletion of Iml3 during meiosis . Cells carrying pCLB2-3HA-IML3 ( strain AM17552 ) were induced to undergo meiosis and samples extracted for anti-HA western immunoblot at the indicated times . A cycling , vegetative ( V ) sample is also shown . Pgk1 is shown as a loading control . DOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 00610 . 7554/eLife . 10850 . 007Figure 2—figure supplement 2 . Depletion of Iml3 in meiosis increases the frequency of recombination events within 20 kb of the centromere . ( A ) The distribution of the observed events on each chromosome is shown for wild type and pCLB2-3HA-IML3 cells . The percentages of CO or NCO events within 20 kb of the pericentromere were calculated as a fraction of the combined total of CO and NCO events on that chromosome . ( B , C ) Comparison of the effect of Iml3 depletion on centromeric recombination with other mutants known to affect CO ( B ) and NCO ( C ) frequency . Events are shown as a percentage of the combined total of the CO and NCO events per meiosis . Significance was determined using a 2x×2 contingency table and p values were calculated using a Chi square test with Yates correction ( ** p<0 . 0001; * p<0 . 05; n . s . , not significant ) . Data for zip3△ , sgs1△ and tel1△ is from ( Oke et al . , 2014 ) . CO , crossover; NCO , non-crossover . DOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 007 We next addressed how the Ctf19 complex prevents pericentromeric COs and NCOs . All meiotic recombination events begin with the programmed introduction of DSBs by the topoisomerase-related protein Spo11 ( Keeney et al . , 1997 ) . DSBs form throughout the genome , but are strongly repressed within ~3 kb of budding yeast centromeres ( Blitzblau et al . , 2007; Buhler et al . , 2007; Pan et al . , 2011 ) . We generated genome-wide DSB maps of the mcm21△ mutant by high-throughput sequencing of oligonucleotides that remain covalently bound to Spo11 as a by-product of DSB processing ( Pan et al . , 2011 ) ( Figure 3A ) . Interestingly , increased centromere-proximal DSBs were detected in the mcm21△ mutant compared with wild type , revealing a general role for the Ctf19 complex in suppression of pericentromeric DSBs ( Figure 3B , C ) . Pericentromeric DSBs in the mcm21△ mutant reached a level similar to the genome average ( dotted lines , Figure 3C ) , indicating that there is no residual DSB suppression in the absence of Mcm21 . Strikingly , the increase in DSBs occurred over a narrower domain ( ~3 kb on each side of the centromere , i . e . a total region of ~6 kb ) than CO formation ( ~20 kb each side of the centromere , ~40 kb in total; Figure 2E , F ) , indicating that suppression of COs at centromeres is not solely due to a reduction in DSB levels . To compare the effect on DSBs and COs more directly , we examined DSB formation in the intervals on chromosome VIII analyzed in the live cell recombination assay ( Figure 3D ) . As expected , we observed an increase in DSBs within the same pericentromeric interval in which we measured CO frequency using the live cell reporter assay . However , DSBs were increased only ~5-fold over wild type in mcm21△ within this region ( Figure 3E ) , while we observed an ~21-fold increase in COs within the same region ( Figure 3F; Figure 1B , C ) . Therefore , the increase in DSBs can only account for approximately 24% of the increase in pericentromeric COs . 10 . 7554/eLife . 10850 . 008Figure 3 . The kinetochore protects the centromere-proximal domain from DSBs . ( A ) Sequencing of Spo11-oligos allows DSBs to be mapped genome wide . ( B ) Fold change in average Spo11-oligo density ( RPM per kb ) in 3 kb segments in mcm21△ cells compared to wild type ( from Zhu and Keeney ( 2015 ) ) , over a 36 kb region surrounding all 16 centromeres . Boxes show median and interquartile range , whiskers enclose data points within 1 . 5 times the interquartile range; outliers are not shown . Red dashed line , fold change of one . ( C ) Mean Spo11 signal as a function of distance from the centromere . Spo11 oligo density within 500 bp bins starting from the centromere and moving up to 75 kb away , averaged across the 32 chromosome arms was determined . The horizontal dotted line indicates genome average . The red ( wild type ) and blue ( mcm21△ ) lines indicate loess smoothing , and the shading indicates the 95% confidence interval . ( D ) DSBs in the region examined in the CO assay . Spo11 oligo counts smoothed with a 201-bp Hann window are shown . The black circle indicates the centromere , filled triangles indicate the midpoints of coordinates where RFP ( red ) and GFP ( green ) cassettes were targeted to for CEN8 analysis in the live cell recombination assay; open triangles indicate the locations where the cassettes were targeted to for ARM8 analysis . ( E and F ) Fold change in the number of DSBs ( E ) or COs ( F ) in mcm21△ vs . wild type within the same pericentromeric or arm intervals on chromosome 8 that were analyzed in the live cell recombination assay ( Figure 1B and C ) . CO , crossover; DSBs , double strand breaks; RPM , reads per million mapped . DOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 00810 . 7554/eLife . 10850 . 009Figure 3—figure supplement 1 . Genome-wide view of meiotic recombination initiation in the mcm21△ mutant . ( A ) Fold change in mcm21△ over wild type of Spo11-oligo counts ( RPM ) in the 10 kb encompassing each centromere . Red dashed line indicates fold change of one . ( B ) Whole-chromosome view of changes in the Spo11-oligo distribution in mcm21△ . Each point is the fold change ( plotted on log2 scale ) of mcm21△ over wild type Spo11 oligos ( RPM ) summed in 5-kb bins . Blue lines , smoothed fit ( loess ) ; black triangles , centromeres; yellow shading , centromere ± 20 kb; black solid line , 0 ( log2 scale ) indicating no change over wild type; black dotted lines , 2-fold change ( log2 scale ) . RPM , reads per million mapped . DOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 009 The DSB effect varied per individual chromosome ( Figure 3—figure supplement 1 ) . The variability in susceptibility of the different pericentromeres to DSBs is likely explained by their underlying features , since it is well-established that DSB formation is influenced by chromatin and genome organization ( e . g . the availability of gene promoters that serve as a preferred target of Spo11 ( Blitzblau et al . , 2007; Pan et al . , 2011 ) ) . Chromosome I in particular , showed emergence of a very prominent DSB hotspot in a promoter region immediately adjacent to the centromere ( CEN1; Figure 4A ) . We exploited the appearance of this hotspot to monitor ectopic DSBs close to centromeres by Southern blotting in the repair-deficient dmc1△ background , where DSBs persist . Close to CEN1 , DSBs were detected in the absence of Mcm21CENP-O or Ctf19 CENP-P ( Figure 4B ) . Importantly , DSBs that formed near CEN1 in the absence of Mcm21CENP-O or Ctf19 CENP-P were dependent on the catalytic activity of Spo11 , demonstrating that these were genuine programmed DSBs ( Figure 4B ) . Screening additional Ctf19 complex subunits showed a striking correlation between increased DSB formation at CEN1 and increased CO formation as measured in our live cell recombination reporter assay ( Figure 4B , Figure 4—figure supplement 1A ) . Interestingly , depletion of the MIND complex component , Dsn1 ( Figure 1—figure supplement 1 ) also resulted in the appearance of CEN1-proximal DSBs , suggesting that the overall integrity of the kinetochore might be generally important for repressing DSB formation within pericentromeres . In conclusion , one likely mechanism by which the Ctf19 complex prevents pericentromeric recombination is via the inhibition of DSB formation close to centromeres . 10 . 7554/eLife . 10850 . 010Figure 4 . Analysis of DSB formation in Ctf19 complex mutants . ( A ) Appearance of a strong DSB hotspot proximal to CEN1 in mcm21△ cells . Spo11-oligo density in a 20 kb region surrounding the centromere of chromosome I in mcm21△ ( top ) and wild type ( bottom ) . Spo11-oligo counts ( RPM ) were smoothed with 201-bp Hann window . ( B ) Detection of CEN1-proximal DSBs in Ctf19 complex mutants by Southern blotting and their dependence on SPO11 catalysis ( using a catalytic dead mutant allele of SPO11 , spo11-Y135F ) . Repair-deficient ( dmc1△ ) cells were harvested at defined times after inducing sporulation ( ( t=0 , 3 , 5 , 8 hr ) and faster migrating DNA species ( indicative of DSBs ) were detected using a probe to CEN1 or the control YCR047C locus . Arrowheads , Spo11-dependent DSBs . Strains used were GV48 ( dmc1△ ) , GV1912 ( dmc1△ ctf19△ ) , GV2128 ( dmc1△ ctf19△ spo11-Y135F-HA ) , GV2050 ( dmc1△ mcm21△ ) and GV2205 ( dmc1△ mcm21△ spo11-Y135F-HA ) . ( C ) Summary of tested mutants and their importance for suppression of DSBs and inhibition of COs close to centromeres . n . d . , not determined . CO , crossover; DSB , double strand break; RPM , reads per million mapped . DOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 01010 . 7554/eLife . 10850 . 011Figure 4—figure supplement 1 . Ctf19 complex components are required to prevent DSB formation close to CEN1 . Repair-deficient ( dmc1△ ) cells ( strains GV48 ( dmc1△ ) , GV1912 ( dmc1△ ctf19△ ) , GV1870 ( dmc1△ iml3△ ) and GV2139 ( dmc1△ chl4△ ) ) were harvested at defined times after inducing sporulation ( t=0 , 3 , 5 , 8 hr ) and faster migrating DNA species ( indicative of DSBs ) were detected using a probe to CEN1 or the control YCR047C locus . Arrowheads , Spo11-dependent DSBs . DSBs , doublestrand breaksDOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 011 During mitotic growth , the Ctf19 complex targets loading of the sister-chromatid-linking complex , cohesin , to the centromere prior to S phase to enrich cohesin in the surrounding pericentromere ( Eckert et al . , 2007; Fernius and Marston , 2009; Fernius et al . , 2013; Ng et al . , 2009 ) . This enrichment provides the basis for robust sister chromatid cohesion , the establishment of which is coupled to DNA replication in S-phase ( Eckert et al . , 2007; Fernius and Marston , 2009; Fernius et al . , 2013; Hinshaw et al . , 2015; Ng et al . , 2009; Uhlmann and Nasmyth , 1998 ) . Pericentromeric enrichment of the meiotic cohesin subunit , Rec8 , at G2/prophase I also depended on the Ctf19 complex ( Figure 5A–D ) . Since cohesin has been implicated in influencing meiotic DNA break formation and repair ( Ellermeier and Smith , 2005; Klein et al . , 1999; Kugou et al . , 2009 ) , we used the live cell recombination reporter assay to test the requirement for cohesin in preventing pericentromeric COs . To prevent pleiotropic phenotypes and sporulation failure associated with total cohesin loss in meiosis , we employed a mutation in the Scc4 subunit of the cohesin loader ( scc4-m35 ) , which in vegetative cells specifically abolishes pericentromeric cohesin enrichment ( Hinshaw et al . , 2015 ) . In meiotic prophase , Rec8 levels were indeed reduced at centromeric and pericentromeric sites in scc4-m35 cells . However , chromosomal arm sites were also affected ( Figure 5E ) , suggesting that the scc4-m35 mutations might influence cohesin loading at non-centromeric sites during meiosis . Nevertheless , scc4-m35 cells underwent meiosis to produce spores , analysis of which revealed an increased frequency of pericentromeric ( Figure 5F ) , but not chromosomal arm ( Figure 5G ) COs , though this increase was more modest than in the absence of Ctf19 complex subunits ( Figure 1B , C ) . This finding supports the notion that pericentromeric cohesin enrichment by the Ctf19 complex contributes to the suppression of centromere-proximal COs . 10 . 7554/eLife . 10850 . 012Figure 5 . Ctf19-dependent cohesin enrichment prevents pericentromeric COs . ( A–D ) , The Ctf19 complex enriches meiotic cohesin in the pericentromere during prophase I . ( A–C ) Wild type ( AM4015 ) , iml3△ ( AM4016 ) and mcm21△ ( AM13833 ) strains carrying REC8-3HA and ndt80△ were harvested 5h after resuspension in sporulation medium and Rec8 association was analyzed by ChIP-Seq . Rec8 association with chromosome V and a close up of the 50 kb pericentromeric interval is shown ( A ) . The median Rec8 level for all 16 pericentromeric regions is shown over a 25 kb region on each side of the centromere for iml3△ ( B ) and mcm21△ ( C ) compared to wild type . ( D ) Strains as in ( A ) together with chl4△ ( AM4017 ) , ctf19△ ( AM20086 ) and a no tag control ( AM11633 ) carrying ndt80△ were arrested in prophase I by harvesting 5h after being induced to sporulate . The level of Rec8 at the indicated sites was determined by anti-HA ChIP-qPCR . Primer sets used corresponded to sites on the arm of chromosome IV ( arm1 , 2 , 3 ) , within the 20 kb pericentromere ( pericen1 , 2 ) or ~150bp from CEN4 ( CEN4 ) and sequences and coordinates are given in Supplementary file 4B . Error bars represent standard error ( n=4 biological replicates for iml3△ , chl4△ , ctf19△ and mcm21△; n=8 for no tag and wild type ) . p<0 . 05 , unpaired t test . ( E ) Chromosomal Rec8 levels are reduced in scc4-m35 cells . Wild type ( AM4015 ) , scc4-m35 ( AM18211 ) cells and a no tag control ( AM11633 ) carrying ndt80△ were arrested in prophase I by harvesting 5 hr after being induced to sporulate . The level of Rec8 at the indicated sites was determined by anti-HA ChIP-qPCR . Error bars represent standard error ( n=4 biological replicates ) . p<0 . 05 , paired t test . ( F , G ) Map distances ( in cM ) in the pericentromere ( F ) or a chromosome arm ( G ) interval in wild type , a control SCC4 replacement strain , or the scc4-m35 mutant were determined and their significance analysed as described in Figure 1 . ChIP-Seq , chromatin immunoprecipitation with sequencing . DOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 012 We next asked whether the Ctf19 complex influences DSB patterns near centromeres by acting before and during S-phase , when it is known to promote the pericentromeric enrichment of cohesin . To test this , we used the “anchor away” system to selectively and conditionally deplete Ctf19 from the nucleus after S-phase ( via timed addition of rapamycin ( Haruki et al . , 2008 ) ; Figure 6A , Figure 6—figure supplement 1 ) . We reasoned that anchoring away Ctf19 before S phase ( t=0 ) would prevent pericentromeric cohesin establishment , similar to a ctf19△ strain ( Marston et al . , 2005; Fernius et al . , 2009 ) . In contrast , because cohesin establishment is coupled to DNA replication ( Uhlmann and Nasmyth , 1998 ) , addition of rapamycin after S phase ( t=3 ) would allow cohesion establishment , thereby allowing us to test cohesin-independent functions of the Ctf19 complex ( Figure 6B ) . First , we confirmed the successful removal of Ctf19-Frb-GFP from the centromere by addition of rapamycin either before ( t=0 ) or after ( t=3 ) DNA replication ( Figure 6C ) . Next , we examined the effect on centromeric cohesin . Rec8 chromatin immunoprecipitation ( ChIP ) in prophase I cells revealed that centromeric cohesin levels were more greatly reduced by anchoring away Ctf19 before ( t=0 h ) rather than after DNA replication ( t=3 h ) ( Figure 6D ) . Because this assay does not allow cohesin loaded before or after DNA replication to be distinguished , we sought to test the functionality of pericentromeric cohesin in the two conditions . Since only functional cohesin is expected to be retained at centromeres during anaphase I ( Klein et al . , 1999 ) , we examined Rec8 on chromosome spreads in binucleate cells ( Figure 6E , F ) . Although centromeric Rec8 was detected in only 4% of binucleate cells where Ctf19 was anchored away before DNA replication ( Rapa t=0 ) , centromeric Rec8 was observed in 48% of binucleate cells where Ctf19 was anchored away after DNA replication ( Rapa t=3 ) ( Figure 6F ) . Therefore , the presence of Ctf19 before and during S phase allows for the establishment of functional pericentromeric cohesin . Having established conditions that allowed us to uncouple pericentromeric cohesin establishment from post-S phase functions of the Ctf19 complex , we asked whether the role of the Ctf19 complex in suppressing pericentromeric DSBs is linked to its role in cohesin establishment during S phase . As expected , anchoring away Ctf19 before ( Rapa t=0 ) DNA replication led to the appearance of CEN1-proximal DSBs with comparable timing and intensity to those of ctf19△ cells ( Figure 6G–I ) . DSBs were also observed following rapamycin addition at 3 hr , i . e . after DNA replication ( Figure 6H ) . This suggests that the Ctf19 complex is required throughout meiotic prophase to prevent pericentromeric DSB formation and that it does so in a manner independent of its role in cohesin establishment . 10 . 7554/eLife . 10850 . 013Figure 6 . Pericentromeric cohesin does not prevent DSBs , but ensures their repair does not form COs . ( A–D ) Anchoring away Ctf19 after DNA replication and cohesion establishment is not sufficient to prevent the appearance of centromere-proximal DSBs . ( A , B ) Scheme of the anchor away system and experimental setup used to deplete Ctf19 during meiosis . ( C ) Addition of Rapamycin leads to Ctf19 removal from the pericentromere . Three cultures of strain AM18978 ( CTF19-FRB-GFP RPL13A-2XFKBP12 tor1-1 fpr1△ ndt80△ REC8-3HA ) were induced to sporulate . Either DMSO or Rapamycin were added to two of the cultures ( t=0 ) . Rapamycin was added 3 hr after inducing sporulation to the third culture ( t=3 ) . A fourth culture was a no tag ndt80△ control ( AM11633 ) to which DMSO was added . All cultures were harvested 5 hr after inducing sporulation ( prophase I arrest ) and Ctf19 levels were analyzed by anti-GFP ChIP-qPCR at the indicated sites . Error bars represent standard error ( n=4 biological replicates ) . p<0 . 05 , paired t-test . See Figure 5 and Supplementary file 4B for details of primer sets used . ( D–F ) Anchoring away Ctf19 after DNA replication allows establishment of centromeric Rec8 . ( D ) Cells treated as in ( C ) were processed for anti-HA ChIP-qPCR at the indicated sites . Error bars represent standard error ( n=4 biological replicates ) . p<0 . 05 , paired t-test . ( E and F ) Three cultures of strain AM20138 ( CTF19-FRB-GFP NDC10-6HA pGAL-NDT80 pGPD1-GAL4 ( 848 ) -ER RPL13A-2XFKBP12 tor1-1 fpr1△ ndt80△ REC8-3HA ) were resuspended in sporulation medium ( t=0 ) . DMSO or Rapamycin were immediately added to the first and second cultures ( t=0 ) , while the third culture received Rapamycin after 3 hr incubation ( t=3 ) . After 6 hr total , β-estradiol was added to release cells from the prophase I arrest , samples were harvested at 15 min intervals and chromosome spreads were prepared and stained with anti-HA and anti-Myc antibodies . ( E ) Examples of binucleate cells with centromeric or no Rec8 . ( F ) Percentages of binucleate cells with centromeric Rec8 are shown for indicated conditions . ( G–I ) Timed depletion of Ctf19 to test DSB formation dependencies . ( G ) DNA replication is largely complete prior to anchoring Ctf19 away in cultures where Rapamycin was added at 3 hr . A control strain GV2367 ( RPL13A-2XFKBP12 tor1-1 fpr1△ dmc1△ ) and equivalent experimental strain carrying CTF19-FRB-GFP ( GV2354 ) were induced to undergo meiosis together with a ctf19△ dmc1△ mutant ( GV1912 ) . Rapamycin or DMSO were added at the indicated times ( red circles , addition of DMSO at t=0; blue squares , addition of Rapamycin at t=0; green squares , addition of Rapamycin at t=3 ) and samples were processed for FACS analysis ( timepoints t=0 , 3 , 5 , 8 hr ) . ( H ) Analysis of DSB formation in the experiment shown in ( B ) . Southern blot shows that DSB formation close to CEN1 occurs either when Ctf19 is anchored away early ( t=0 ) or after DNA replication and cohesin enrichment ( t=3 ) . Red circles , addition of DMSO at t=0; blue squares , addition of Rapamycin at t=0; green squares , addition of Rapamycin at t=3; Arrowheads , Spo11-dependent DSBs; asterisks , cross-hybridizing species . ( I ) Quantification DSBs shown in ( H ) . ( J , K ) Inhibition of centromere-proximal DSBs does not depend on cohesin . ( J ) CEN1-proximal DSBs are observed in rec8△ cells only upon deletion of Ctf19 complex components . Strains GV48 ( dmc1△ ) , GV2050 ( dmc1△ mcm21△ ) , GV2403 ( dmc1△ rec8△ ) , GV2286 ( dmc1△ mcm21△ rec8△ ) were analyzed by Southern blotting as described in Figure 4B . ( K ) Cohesin impairment does not allow CEN1-proximal DSB formation . Strains used were GV48 ( dmc1△ ) , GV1912 ( dmc1△ ctf19△ ) , GV2305 ( dmc1△ SCC4 ) and GV2533 ( dmc1△ scc4-m35 ) . ( L ) Anchoring away Ctf19-FRB after DNA replication and cohesin establishment leads to only a modest increase in pericentromeric COs . Three cultures of strain AM19543 [carrying heterozygous pericentromeric RFP/GFP reporters separated by ~10 kb , homozygous chromosomal arm CFP reporters ( Figure 1B ) , estradiol-inducible Ndt80 to allow release from prophase I arrest ( pGAL-NDT80 pGPD1-GAL4 ( 848 ) -ER ) and the Ctf19 anchor away system ( CTF19-FRB RPL13A-2XFKBP12 tor1-1 fpr1△ ) ] were resuspended in sporulation medium ( t=0 ) . DMSO or Rapamycin were immediately added to the first and second cultures ( t=0 ) , while the third culture received Rapamycin after 3 hr incubation ( t=3 ) . All cultures were incubated for 6 hr total before being released from the prophase I arrest by addition of β-estradiol and tetrads were scored after incubation for 48 hr total . ChIP , chromatin immunoprecipitation; DMSO , dimethyl sulfoxide; DSBs , double strand breaks; FACS , fluorescence-activated cell sorting; qPCR , quantitative polymerase chain reaction . DOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 01310 . 7554/eLife . 10850 . 014Figure 6—figure supplement 1 . Anchoring Ctf19 away early in meiosis results in reduced sporulation efficiency and spore viability . RPL13A-2xFKBP12 tor1-1 fpr1△ strains carrying FRB-tagged ( CTF19-FRB-GFP; GV2275 ) or untagged Ctf19 ( CTF19; GV1853 ) were treated with Rapamycin or DMSO , as a control , upon resuspension in sporulation medium ( t=0 ) . ( A ) The percentages of cells producing monads ( 1 spore ) , dyads ( 2 spores ) or tetrads ( 4 spores ) was scored ( n=200 ) . ( B ) The number of spores germinating per tetrad was scored after dissection . At least 60 spores were analysed per condition . DOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 014 To further test the requirement for cohesin in preventing pericentromeric DSB formation , we asked whether DSBs are increased near CEN1 in cells defective for cohesin . Consistent with our findings above , CEN1-proximal DSBs were not observed in rec8△ or scc4-m35 cells , unless a Ctf19 complex component ( Mcm21CENP-O ) was also absent ( Figure 6J , K ) . Thus , DSB inhibition near centromeres does not depend on cohesin , but requires the continuous presence of the Ctf19 complex . These findings provide an explanation for our observation that Ctf19 complex mutants exhibit a higher frequency of pericentromeric COs than scc4-m35 mutant cells ( Figure 1C , Figure 5F ) , despite a comparable reduction in pericentromeric cohesin ( Figure 5D , E ) . Loss of MCM21 relieved DSB suppression over a domain ( ~6 kb; Figure 3B ) that is much larger than the 125 bp where the kinetochore resides but smaller than the cohesin-rich pericentromere ( ~20 kb; Figure 5A ) . We therefore speculate that the large multi-subunit Ctf19 complex may exert DSB suppression near centromeres by altering local chromosome structure such that accessibility of DSB-promoting factors is prevented . Our findings suggest that pericentromeric cohesin might provide a safeguarding mechanism to channel residual centromere-proximal DNA breaks towards repair pathways that do not promote CO formation . If this is the case , DSBs arising after DNA replication and pericentromeric cohesion establishment would not be expected to give rise to pericentromeric COs . To test this idea , we anchored away Ctf19 either before ( t=0 ) or after ( t=3 ) DNA replication , both of which induce CEN1-proximal DSBs soon thereafter ( Figure 6H ) and measured CO formation using the live cell reporter assay ( Figure 6L ) . Anchoring Ctf19 away before DNA replication ( t=0 ) increased pericentromeric COs to a similar extent to ctf19△ cells ( compare Figure 1B and Figure 6L ) , as expected . However , anchoring Ctf19 away after DNA replication ( t=3 ) led to a more modest increase in pericentromeric COs ( Figure 6L ) , despite comparable levels of DNA break formation in these cells ( Figure 6H , I ) . Therefore , pericentromeric cohesin acts at a step after DSB formation to direct repair , probably through a pathway avoiding the homolog , to ensure CO suppression near centromeres . Because the cohesin complex is known to promote inter-sister recombination in mitosis and meiosis ( Covo et al . , 2010; Ellermeier and Smith , 2005; Klein et al . , 1999; Kugou et al . , 2009; Sjogren and Nasmyth , 2001 ) the simplest explanation is that increased pericentromeric cohesin shunts meiotic DSBs into inter-sister-specific recombinational repair , although we cannot rule out pericentromere-specific , cohesin-dependent activation of alternative repair pathways . Rec8 globally influences the localization of the synaptonemal complex ( SC ) component Zip1 along chromosomes ( Chuong and Dawson , 2010; Brar et al . , 2009 ) . Because Zip1 , like cohesin ( Kim et al . , 2010 ) has been suggested to confer a bias on DSBs to be repaired from the sister chromatid , rather than the homolog , and because loss of Zip1 increases pericentromeric COs ( Figure 1B ) , but not DSBs ( Chen et al . , 2008; Figure 7—figure supplement 1 ) , we reasoned that a critical role of pericentromeric cohesin might be Zip1 recruitment . Consistently , ChIP-qPCR indicated that Zip1 localization was impaired in scc4-m35 ( Figure 7A ) and Ctf19 complex mutants ( Figure 7B ) . Furthermore , analysis of spread meiotic chromosomes revealed that the “dotty” Zip1 localization pattern , representative of centromeres , was rarely observed in Ctf19 complex mutants ( Figure 7C ) . Unexpectedly , “full” Zip1 localization along chromosomes was also impaired in Ctf19 complex mutants . Although the underlying reasons for this are currently unclear , possible explanations are delayed G2/prophase progression and/or a requirement for pericentromeric Zip1 in producing the “full” Zip1 tracts observed in cytological analyses . Nevertheless , ChIP-Seq of prophase I-arrested cells confirmed that Ctf19 complex components are specifically required for Zip1 association with core centromeres and the pericentromere ( Figure 7D–F ) . We measured comparable pericentromeric CO frequencies of 2 . 4 and 2 . 0 cM in scc4-m35 and zip1△ cells , respectively ( Figure 1B , Figure 5F ) suggesting that pericentromeric cohesin establishment by the Ctf19 complex directs Zip1 association to suppress pericentromeric COs , although we do not rule out Zip1-independent functions of cohesin in CO suppression . 10 . 7554/eLife . 10850 . 015Figure 7 . Cohesin enables centromeric Zip1 recruitment . ( A ) Zip1 enrichment on chromosomes is reduced in scc4-m35 mutants . Wild type ( AM11633 ) , scc4-m35 ( AM18881 ) and zip1△ ( AM10913 ) cells carrying ndt80△ were induced to sporulate and harvested after 5 hr ( prophase I ) arrest for anti-Zip1 ChIP-qPCR . Error bars represent standard error ( n=4 biological replicates ) . p<0 . 05 , paired t test . See Figure 5 and Supplementary file 4B for details of primer sets used . ( B-–F ) The Ctf19 complex is required for Zip1 localization at centromeres . ( B ) ChIP-qPCR analysis of Zip1 localization in prophase I in Ctf19 complex mutants . Wild type ( AM11633 ) , iml3△ ( AM10686 ) , chl4△ ( AM10658 ) , ctf19△ ( AM10660 ) , mcm21△ ( AM10664 ) and zip1△ ( AM10913 ) cells carrying ndt80△ were induced to sporulate and harvested after 5 hr for anti-Zip1 qPCR . Error bars represent standard error ( n=3 biological replicates ) . p<0 . 05 , paired t test . See Figure 5 and Supplementary file 4B for details of primer sets used . ( C ) Analysis of Zip1 localization on chromosome spreads as cells progress into prophase I . Examples of Zip1 localization on chromosome spreads are shown with Zip1 in green and DNA in blue . Categories of Zip1 localization were scored in 100 spread nuclei at each of the indicated times after resuspension in sporulation medium . Strains used carried NDC10-6HA , pGAL-NDT80 , pGPD1-GAL4-ER and were AM8769 ( wild type ) , AM8772 ( iml3△ ) , AM8770 ( chl4△ ) , AM9049 ( ctf19△ ) and AM8861 ( mcm21△ ) . ( D–F ) ChIP-Seq analysis of Zip1 localization during prophase I in wild type , iml3△ and mcm21△ strains carrying ndt80△ ( B ) and harvested 5 hr after resuspension in sporulation medium . ( D ) Zip1 localization along chromosome V is shown as an example with the 50 kb region around the centromere amplified . ( E , F ) Median Zip1 localization over a 50 kb domain surrounding all 16 centromeres is shown compared to wild type for iml3△ ( E ) and mcm21△ ( F ) . ChIP , chromatin immunoprecipitation; qPCR , quantitative polymerase chain reaction . DOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 01510 . 7554/eLife . 10850 . 016Figure 7—figure supplement 1 . Zip1 does not induce pericentromeric DSB formation . Repair-deficient ( dmc1△ ) cells ( strains GV48 ( dmc1△ ) , GV2050 ( dmc1△ mcm21△ ) , GV2734 ( dmc1△ zip1△ ) were harvested at defined times after inducing sporulation ( ( t=0 , 3 , 5 , 8 hr ) and faster migrating DNA species ( indicative of DSBs ) were detected using a probe to CEN1 or the control YCR047C locus . Arrowheads , Spo11-dependent DSBs . DSBs , double strand breaks . DOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 016 Centromeric Zip1 mediates homology-independent pairing of homologous chromosomes early in meiotic prophase; the subsequent conversion to homologous pairing requires Spo11 ( Tsubouchi and Roeder , 2005; Tsubouchi et al . , 2008 ) . The Ctf19 complex was required for the centromeric localization of Zip1 in spo11△ cells ( Figure 8A ) and homology-independent centromere coupling ( Figure 8B ) but not for the SPO11-dependent transition to homologous pairing ( Figure 8C , D ) . To test whether Zip1 exerts its role in suppression of pericentromeric COs through homology-independent centromere coupling , we analyzed the synaptonemal complex assembly-proficient but centromere coupling-defective zip1-S75E mutant ( Falk et al . , 2010 ) , using the live cell recombination reporter assay . Pericentromeric COs were not significantly increased in the zip1-S75E mutant ( Figure 8E , F ) , indicating that Zip1 suppresses pericentromeric COs independently of its role in centromere coupling . 10 . 7554/eLife . 10850 . 017Figure 8 . Recruitment of Zip1 to centromeres by the Ctf19 complex independently promotes centromere coupling and suppression of pericentromeric COs . ( A , B ) Analysis of Zip1 and kinetochore foci ( Ndc10-6HA ) on chromosome spreads in spo11△ cells progressing into meiotic prophase I . The percentages of cells with “dotty” Zip1 foci , representing kinetochores , were scored in 100 spread nuclei at each of the indicated times after resuspension of strains spo11△ ( AM9018 ) , spo11△ iml3△ ( AM9288 ) , spo11△ chl4△ ( AM9287 ) , spo11△ ctf19△ ( AM9017 ) and spo11△ mcm21△ ( AM8861 ) in sporulation medium ( A ) . The number of Ndc10 kinetochore foci per spread nucleus was scored in the indicated strains at the 6 hr time point with the average and standard deviation indicated . n=48 ( spo11△ ) , 63 ( spo11△ iml3△ ) and 60 ( spo11△ mcm21△ ) . ( B ) The inset shows Ndc10-6HA staining ( red ) in an example nucleus . DNA is shown in blue . ( C ) The Ctf19 complex is not required for SPO11-dependent homologous pairing . Pairing of homozygous CEN5-GFP ( C ) or LYS2-GFP ( D , arm ) foci was scored for 100 cells at each of the indicated times after resuspension in sporulation medium in ndt80△ cells . Heterologous CEN5-GFP and LYS2-GFP foci labels were used as a control for spurious interactions ( strain AM12823 , red filled squares ) . Strains used in ( C ) were AM12829 ( wild type ) , AM13348 ( iml3△ ) , AM12466 ( chl4△ ) , AM13346 ( ctf19△ ) and AM12837 ( mcm21△ ) . Strains used in ( D ) were AM12469 ( wild type ) , AM12978 ( iml3△ ) , AM12980 ( chl4△ ) , AM12831 ( ctf19△ ) and AM12825 ( mcm21△ ) . A representative experiment is shown . ( E , F ) The centromere-coupling function of Zip1 is separable from its role in suppression of recombination in the pericentromere . Recombination frequency in the pericentromere ( E ) or chromosomal arm ( F ) interval on chromosome VIII for the centromere-coupling defective zip1-S75E mutant is shown together with data for wild type and zip1△ reproduced from Figure 1B , C . COs , crossovers . DOI: http://dx . doi . org/10 . 7554/eLife . 10850 . 017 The placement of meiotic DSBs is influenced by factors acting on different levels of chromosome and chromatin organization . On a global scale , the assembly of a meiotic chromosome axis dictates the spatial distribution of the meiotic DSB machinery along chromosomes ( Kim et al . , 2010; Kugou et al . , 2009; Pan et al . , 2011; Panizza et al . , 2011 ) . On a smaller scale , genome organization and histone modifications have been shown to allow Spo11 activity ( reviewed in de Massy ( 2013 ) ) . Spo11-dependent DSB formation is strongly associated with chromatin regions that are enriched for histone H3 Lysine 4 ( H3K4 ) methylation ( reviewed in de Massy ( 2013 ) ) and in budding yeast , these modifications are found near transcriptional start sites ( Tischfield and Keeney , 2012 ) ) . Indeed , within the budding yeast genome , Spo11 prefers to cleave DNA in open , nucleosome-free regions that are most often found in active , divergent promoters ( Blitzblau et al . , 2007; Pan et al . , 2011 ) . Superimposed on these global determinants of the DSB landscape are specific and spatial controls , which create zones of inhibition within at-risk genomic regions such as telomeres , repetitive DNA arrays and centromeres ( reviewed in de Massy ( 2013 ) ) . We find that the kinetochore actively minimizes DSB formation within a region of ~6 kb , surrounding all budding yeast centromeres . The underlying genome organization of these regions is not obviously different from the rest of the genome , harboring a similar density of genes and regulatory elements , which would ordinarily be expected to contain several DSB-permissive regions , were they not located close to a centromere . Indeed , in the absence of the Ctf19 kinetochore sub-complex , DSB formation is strongly increased , and these DSBs have features typical of preferred DSB sites genome-wide . Accordingly , one of the strongest pericentromeric DSBs that we identified here , at CEN1 , falls within a divergent promoter ( of the genes TFC3 and NUP60; Figure 4A ) . The Ctf19 complex suppresses DSB formation surrounding all centromeres , though to different extents depending on the chromosome ( Figure 3—figure supplement 1 ) , suggesting that it overcomes the intrinsic features of the chromatin organization in the pericentromere to dampen DSB activity . This could be achieved by locally shaping a higher order chromosome organization and/or by influencing the recruitment of meiotic chromosome axis components . Intriguingly , the chromosomal region protected from DSBs ( ~6 kb ) extends far beyond the binding of the Ctf19 complex , which is restricted to the 125 bp point centromere , as defined by the assembly of a nucleosome containing the centromere-specific histone variant Cse4CENP-A ( Biggins , 2013 ) . Therefore , the Ctf19 complex protects a region about 50 times larger than that occupied by the centromeric nucleosome from DSBs , suggesting that the effect is not merely due to a local disturbance of Cse4CENP-A . A precedent for the idea that the Ctf19 complex can exert long-range effects on the pericentromere through its role in driving cohesin loading at the centromere to enrich the surrounding 20–50 kb is well-documented ( Eckert et al . , 2007; Fernius and Marston , 2009; Fernius et al . , 2013; Ng et al . , 2013 ) . However , we found that the Ctf19 complex must prevent DSB formation independently of its role in promoting cohesin enrichment within pericentromeres . Cohesin enrichment was neither sufficient nor required to prevent DSBs near CEN1 . Furthermore , DSB suppression acts on shorter chromosomal distances ( ~6 kb ) , than the ~20–50 kb sized regions within which the Ctf19 complex influences cohesin recruitment . Taken together , these findings suggest the existence of an additional , distinct Ctf19 complex-dependent effect on meiotic DSB formation within regions adjacent to centromeres . By analogy to the effect on pericentromeric cohesin enrichment , we speculate that the Ctf19 complex may enable the centromeric recruitment of factors that alter chromatin organization in the surrounding region . Minimizing the initiating event of meiotic recombination , DSB formation , is an efficient way to shield against unwanted pericentromeric CO formation . However , the prevention of pericentromeric DSBs by the Ctf19 complex is not absolute , as DSBs are observed near centromeres in wild type cells , although at reduced levels ( Blitzblau et al . , 2007; Buhler et al . , 2007; Pan et al . , 2011 ) . We found that DSBs that escape the repressive control of the Ctf19 complex are diverted from repair pathways that would lead to potentially deleterious CO formation by cohesin , which is established at high levels within pericentromeres . These observations are in agreement with previous conclusions that pericentromeric CO formation is more strongly suppressed than DSB formation ( Blitzblau et al . , 2007; Buhler et al . , 2007; Chen et al . , 2008; Pan et al . , 2011 ) . We found that forced removal of the Ctf19 complex from kinetochores after S-phase triggered increased DSB formation , but led to only a relatively modest increase in CO formation . Conversely , a cohesin-loader mutant ( scc4-m35 ) that disrupts the proper establishment of cohesin showed increased levels of CO formation , whereas no increased DSB formation could be detected . Thus , DSBs that occur within the pericentromeric regions when high levels of cohesion are present are shunted away from inter-homologous recombinational repair ( that eventually can yield a CO or NCO ) . Taking into account the established roles for cohesin in promoting inter-sister repair ( Covo et al . , 2010; Ellermeier and Smith , 2005; Klein et al . , 1999; Kugou et al . , 2009; Sjogren and Nasmyth , 2001 ) , we consider it is most likely that these breaks will preferentially be repaired using the sister chromatid as a repair template . How would the enrichment of cohesin minimize inter-homologue repair ? One possibility is that high levels of cohesin turn the sister chromatid within pericentromeres into a preferred repair template . Indeed , it has been shown that template choice ( sister or homologue ) for DSB repair during budding yeast vegetative growth is dictated by the levels of cohesin ( Covo et al . , 2010 ) . Globally , on meiotic chromosomes , repair of DSBs is normally biased towards use of the homologous chromosome instead of the sister chromatid to ensure at least one COs is formed to link each homolog pair ( Hollingsworth , 2010; Humphryes and Hochwagen , 2014 ) . In budding yeast , this homologue bias is established by the Red1/Hop1/Mek1 signaling axis ( Hong et al . , 2013; Kim et al . , 2010; Niu et al . , 2007; Niu et al . , 2005 ) . This signaling pathway is thought to locally antagonize cohesin to enable the use of the more distant homologous chromosome as a repair template ( Hong et al . , 2013; Kim et al . , 2010 ) . Potentially , the Red1/Hop1/Mek1 system is not capable of counteracting cohesin within pericentromeres , where much higher levels of cohesin are present as compared to elsewhere in the genome . Alternatively , one could envision that , within pericentromeric regions , Red1/Hop1/Mek1 are incapacitated via active inhibition or removal . Our findings point to a second mechanism through which cohesin steers repair of meiotic DSBs away from inter-homologue repair: by promoting the local recruitment of Zip1 . Zip1 has previously been implicated in preventing pericentromeric CO and NCO repair , and was suggested to promote inter-sister repair ( Chen et al . , 2008 ) . We found that proper recruitment of Zip1 to the pericentromere requires the Ctf19 complex and kinetochore-targeted Scc2/4 , consistent with previous studies demonstrating a requirement for cohesin in Zip1 recruitment ( Chuong and Dawson , 2010; Brar et al . , 2009 ) . Zip1 performs specific functions at centromeres during prophase , which include non-homologous centromere-coupling ( Tsubouchi and Roeder , 2005; Tsubouchi et al . , 2008 ) , repression of pericentromeric CO formation ( Chen et al . , 2008 ) and the bi-orientation of homologous chromosomes during meiosis I ( Gladstone et al . , 2009; Newnham et al . , 2010 ) . Here we have shown that the Ctf19 complex , presumably through its role in directing pericentromeric cohesin enrichment , enables the dedicated recruitment of a centromere-localized pool of Zip1 to perform these specialized functions . Overall , we conclude that the kinetochore , and specifically the Ctf19 complex , promotes the establishment of an inter-homologue recombination-suppressed zone surrounding centromeres . To do so , the Ctf19 complex recruits high levels of cohesin within pericentromeres , which in turn triggers efficient recruitment of the synaptonemal complex component Zip1 , effectively leading to strong , local inhibition of inter-homologue directed repair . CO formation in the vicinity of centromeres negatively influences meiotic chromosome segregation in diverse organisms , and is associated with the incidence of Trisomy 21 , or Down’s syndrome , in humans ( Hassold and Hunt , 2001; Koehler et al . , 1996; Rockmill et al . , 2006 ) . Mechanistically , pericentromeric CO formation has been suggested to lead to a local disturbance of sister chromatid cohesion , which could lead to precocious separation sister chromatid , causing meiosis II non-disjunction and aneuploidy ( Rockmill et al . , 2006 ) . While meiosis II mis-segregation is a characteristic feature of Ctf19 complex mutants ( Fernius and Marston , 2009 ) , the impact of pericentromeric COs on this phenotype is currently unclear because of the requirement for this complex for proper pericentromeric cohesion . The kinetochore is a sophisticated machine that couples chromosomes to microtubules and drives their segregation in meiosis and mitosis . The kinetochore also serves as a signaling platform that monitors and responds to the state of kinetochore–microtubule attachment in the context of the cell cycle . During meiosis , in addition to the canonical events that also take place during mitosis , the kinetochore takes on additional roles to bring about the specialized segregation pattern . First , sister chromatids need to attach to microtubules that emanate from the same pole in a mono-oriented fashion . Second , sister chromatid cohesion at pericentromeres needs to be protected from removal during meiosis I . In both cases , the kinetochore controls these events by coordinating the recruitment of specific protein complexes ( i . e . monopolin and Sgo1/PP2A , respectively ) ( reviewed in Duro and Marston ( 2015 ) ) . Our data add a hitherto unknown , additional level of functionality to the kinetochore in meiosis , in which it impacts meiosis-specific CO formation by influencing both meiotic DSB formation and recombinational break repair . As such , it effectively prevents the formation of unwanted chiasma within pericentromeres . Finally , we note that our data might also provide an additional rationale for why much higher cohesin levels need to be established around centromeres , namely to minimise CO formation near centromeres . The Ctf19 complex is a generally conserved component of eukaryotic kinetochores . Suppression of meiotic CO recombination within pericentromeres is a widespread feature of meiotic recombination in many diverse organisms , whether or not their centromeres are surrounded by large blocks of heterochromatinized DNA . We therefore suggest that Ctf19 complex-driven suppression of meiotic CO formation serves as a universal component of the mechanisms that shape the meiotic recombination landscape in order to promote the faithful propagation of the genome from generation to the next . Yeast strains used in this study are derivatives of SK1 and genotypes are given in Supplementary file 4A except for the strains used for analysis of recombination genome-wide where AM15182 and AM15183 haploid strains derived from YJM789 and S96 , respectively , were used . Standard techniques were used to generate gene deletions , promoter replacements and epitope-tagged proteins . zip1-S75E was described in Falk et al . ( 2010 ) . scc4-m35 and SCC4::HIS3 were generated in SK1 as described by Hinshaw et al . ( 2015 ) . SPO11-6HIS-3FLAG-loxP-KanMX-loxP was provided by K . Ohta ( Kugou et al . , 2009 ) and DSN1-6HIS-3FLAG was described in Sarangapani et al . , ( 2014 ) . For the anchor away system , parental SK1 strains were generated as described by Haruki et al . ( 2008 ) harbouring tor1-1 , fpr1Δ , and RPL13A-2xFKBP12 . CTF19-FRB-GFP was made using standard polymerase chain reaction ( PCR ) -based transformation using pFA6a-FRB-GFP-KanMX6 ( GVp584 ) as a template ( Haruki et al . , 2008 ) . Prophase I block-release experiments used strains carrying pGAL-NDT80 pGPD1-GAL4 ( 848 ) -ER ( Benjamin et al . , 2003 ) . For the live cell recombination reporter assay , pYKL050c-CFP/RFP/GFP* constructs ( Thacker et al . , 2011 ) were integrated at specific loci . Plasmids AMp1005 and AMp1048 were generated by cloning an ~500 bp region corresponding to SGD coordinates 115024–115572 and 150521–151070 into pSK726 and pSK691 , respectively . pYKL050-CFP was introduced at the THR1 locus by integration of plasmid pSK695 and pYKL050-RFP was integrated at the CEN8 locus by integration of plasmid pSK693 or at SGD coordinates 150521–151070 by integration of plasmid AMp1048 . pYKL050c-GFP* was introduced at the ARG4 locus by integration of the plasmid pSK729 and at SGD coordinates 115024–115572 locus by integration of plasmid AMp1005 . Diploid yeast strains were placed on Yeast peptone glycerol ( YPG ) agar plates ( 1% yeast extract , 2% Bacto peptone , 2 . 5% glycerol , and 2% agar ) and grown for 16 hr at 30oC before transferring to YPD 4% agar plates ( 1% yeast extract , 2% Bacto peptone , 4% glucose , and 2% agar ) and incubated for 24 hr at 30oC . Strains were inoculated in YPD media ( 1% yeast extract , 2% Bacto peptone , and 2% glucose ) and cultured for 24 hr before being transferred to YPA ( 1% yeast extract , 2% Bacto peptone , and 1% potassium acetate ) or BYTA ( 1% yeast extract , 2% Bacto tryptone , 1% potassium acetate , 50 mM potassium phthalate ) at an OD600 = 0 . 2–0 . 3 for ~16 hr . Cells were washed once with sterile distilled water and re-suspended in SPO media ( 0 . 3% potassium acetate , pH 7 ) at an OD600 = 1 . 8–1 . 9; t=0 . Cells were incubated at 30oC for the duration of the experiment . Prophase I block-release experiments were performed as described by Carlile and Amon ( 2008 ) . Western blotting was performed as previously described ( Clift et al . , 2009 ) , with the exception that proteins were visualized using a fluorophore-conjugated antibody and the Odyssey system ( LI-COR Biosciences , Lincoln , Nebraska ) . To visualise 3HA-Iml3 , 3HA-Dsn1 and 3HA-Mtw1 , mouse anti-HA ( 12CA5 , Roche , Basel , Switzerland ) was used at a dilution of 1:1000 and anti-mouse IRDye 800CW ( LI-COR Biosciences ) at a dilution of 1:10 , 000 . To visualise Pgk1 ( loading control ) , rabbit anti-Pgk1 ( Marston lab stock ) and anti-rabbit IRDye 680RD ( LI-COR Biosciences ) were used at a dilution of 1:10000 . Proteins tagged with the FKBP12-rapamycin-binding ( FRB ) domain of mTOR1 were depleted from the nucleus by Rpl13A-2xFKBP12 upon addition of rapamycin to a final concentration of 1 μM , as previously described ( Haruki et al . , 2008 ) . ChIP-qPCR and ChIP-Seq were performed as described in Verzijlbergen et al . ( 2014 ) using mouse anti-HA ( 12CA5 , Roche ) , rabbit anti-Zip1 ( Santa Cruz Biotechnology , Dallas , Texas ) , mouse anti-FLAG ( Mono M2 , Sigma Aldrich , St Louis , Missouri ) or mouse anti-GFP ( Sigma Aldrich ) . Primers used for qPCR analysis are given in Supplementary file 4B . ChIP-Seq samples were analysed on a HiSeq2000 instrument ( Illumina , San Diego , California ) by the EMBL Core Genomics Facility ( Heidelberg , Germany ) . Using BWA ( Version: 0 . 7 . 5a-r405 ) ( Li and Durbin , 2010 ) , single reads were mapped to the sacCer3 reference genome . Duplicate reads were removed for parallel analysis using SAMtools ( Version 1 . 2 ) ( Li et al . , 2009 ) . The data shown were normalized to the number of reads per million of total mapped reads . The total mapped reads were established after any processing . Additional scripts for processing data around the pericentromeres can be found at https://github . com/AlastairKerr/Vincenten2015 . ChIP-Seq data sets have been deposited with the NCBI Gene Expression Omnibus under the accession number GSE70032 . Chromosome spreading was performed as described previously ( Loidl et al . , 1998 ) . Ndc10-6HA was detected using a mouse anti-HA antibody ( Mono HA . 11 , Covance , Princeton , New Jersey ) at 1:250 dilution and an anti-mouse Cy3 antibody ( Jackson ImmunoResearch ) at 1:300 dilution . Rec8-13 Myc was detected with a rabbit anti-MYC antibody ( Gramsch Laboratories ) and an anti-rabbit fluorescein isothiocyanate ( FITC ) -conjugated antibody ( Jackson Immunoresearch , West Grove , Pennsylvania ) , both at 1:300 dilution . Zip1 was detected with a rabbit anti-Zip1 antibody ( Santa Cruz Biotechnology ) at 1:500 dilution and an anti-rabbit FITC antibody ( Jackson ImmunoResearch ) at 1:300 dilution . Chromosome spread samples were analysed on a DeltaVision Elite system ( Applied Precision , Isaaquah , Washington ) using an inverted Olympus IX-71 microscope with a 100x UPlanSApo NA 1 . 4 oil lens . Images were acquired using the Photometrics Cascade II EMCCD camera . The camera , shutters , and stage were operated through SoftWorx software ( Applied Precision ) . Samples for studying GFP-labelled chromosomes were prepared as described previously ( Klein et al . , 1999 ) . For the recombination and GFP-labelled chromosome assays , microscopy analysis was performed using a Zeiss Axioplan 2 microscope with a 100x Plan ApoChromat NA 1 . 4 oil lens . Images were acquired using the Photometrics Evolve EMCCD camera operated through Axiovision software . Images were processed and analysed using ImageJ software ( National Institutes of Health ) . Yeast strains were placed on SPO agar plates ( 0 . 8% KAc and 2% agar ) and allowed to sporulate at 300°C . After 4 days , images were captured in three channels and the pattern of fluorescence scored in the tetrads . To prevent confounding effects due to chromosome mis-segregation ( a common occurance in kinetochore mutants ) , only tetrads where all 4 spores had acquired CFP ( blue ) fluorescence were included in the final analysis . Recombination frequency , expressed as map distance in Morgans , and standard error , was calculated using online tools ( http://molbio . uoregon . edu/~fstahl/compare2 . php ) . Power analysis was performed to determine sample size required for >0 . 87 confidence in differences from wild type and Fishers exact test was used to determine significance . All raw data and statistical analysis is given in Supplementary files 1 and 2 . Southern blotting was performed as previously described ( Vader et al . , 2011 ) . The following probes ( SGD coordinates ) were used: YCR047C;III , 209 , 361–210 , 030 . CEN1:I , 145 , 305–145 , 650 . DSB intensities were analysed using ImageJ . Flow cytometry was performed as described ( Vader et al . , 2011 ) . Identification of single nucleotide polymorphisms by high-throughput sequencing was carried out as described by Oke et al . ( 2014 ) . Spo11-oligo maps of mcm21Δ were generated as described by Zhu and Keeney , 2015 with modifications in sporulation culture cell density and Spo11-Flag immunoprecipitation ( IP ) . Briefly , after 14 hr pre-sporulation in YPA media , cells were transferred to sporulation media ( SPM ) described in Neale and Keeney ( 2009 ) to a cell density ( OD600 ) of 6 . 0 . The Spo11-oligo maps were generated from samples harvested after 4 hr in SPM . Spo11-Flag IP was carried out as described , except with protein G Dynabeads ( Life Technologies , Carlsbad , California ) instead of protein G agarose beads ( 400 μl protein G Dynabeads per 25 ml whole-cell extract in 50 ml IP volume for first round of IP; 125 μl protein G Dynabeads in 800 μl IP volume for second round of IP ) .	The cells of animals , plants and many other organisms store most of their DNA inside the cell nucleus , packaged into structures called chromosomes . Most cells contain two copies of each chromosome – one inherited from each parent . However , sex cells ( such as egg cells and sperm cells ) contain just one copy of every chromosome , so that when they fuse , the new cell that is formed contains the full set . Sex cells form in a process called meiosis , where a cell containing two copies of every chromosome duplicates its genetic material and then divides to form four new cells , each of which contains one copy of each chromosome . During meiosis , different versions of the same chromosome are able to swap sections of their DNA in a process called crossover . However , if a crossover occurs in the wrong part of the chromosome , the chromosome copies may not segregate correctly during cell division . This can lead to the formation of sex cells that contain the wrong number of chromosome copies , which can cause developmental conditions such as Down’s syndrome . Crossovers tend not to occur at a region of the chromosomes called the centromere , which is where copies of the same chromosome from the same parent are joined together until it is time for them to separate . If a crossover does occur in this region , segregation problems are more likely to occur . However , exactly how crossovers are suppressed at centromeres is not understood . Vincenten et al . examined crossover positioning in budding yeast cells , which are often used as a model to investigate processes such as cell division . This revealed that a protein complex called Ctf19 stops DNA breaks from occurring near the centromere , and so prevents crossovers . Ctf19 also promotes the enrichment of another protein complex called cohesin near centromeres . This does not prevent DNA breaks from occurring , but also prevents crossovers . Identifying this role of the Ctf19 complex has paved the way for understanding exactly how DNA breaks are prevented near centromeres . This will allow researchers to determine the impact of misplaced crossovers on how chromosomes segregate into sex cells .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2015	The kinetochore prevents centromere-proximal crossover recombination during meiosis
Transcription at individual genes in single cells is often pulsatile and stochastic . A key question emerges regarding how this behaviour contributes to tissue phenotype , but it has been a challenge to quantitatively analyse this in living cells over time , as opposed to studying snap-shots of gene expression state . We have used imaging of reporter gene expression to track transcription in living pituitary tissue . We integrated live-cell imaging data with statistical modelling for quantitative real-time estimation of the timing of switching between transcriptional states across a whole tissue . Multiple levels of transcription rate were identified , indicating that gene expression is not a simple binary ‘on-off’ process . Immature tissue displayed shorter durations of high-expressing states than the adult . In adult pituitary tissue , direct cell contacts involving gap junctions allowed local spatial coordination of prolactin gene expression . Our findings identify how heterogeneous transcriptional dynamics of single cells may contribute to overall tissue behaviour . Gene expression in single living cells is often pulsatile and heterogeneous between cells ( Sanchez and Golding , 2013; Coulon et al . , 2013 ) . In a complex three-dimensional ( 3D ) tissue , dynamic and heterogeneous single cell behaviour gives rise to the overall tissue-level gene expression state and determines the ability of the tissue to respond appropriately to acute and chronic stimuli . Transcriptional bursting , defined by periods of RNA synthesis followed by usually longer silent periods , occurs at many genes with characteristic gene-specific timing ( Suter et al . , 2011 ) . These dynamics have been proposed to be influenced by intrinsic factors that appear stochastic and extrinsic factors that reflect the state of the cell . Thus far , the key to identifying these processes has been single cell analysis ( Raj and van Oudenaarden , 2009; Spiller et al . , 2010 ) . In situ hybridisation techniques have revealed non-equivalent activity at gene alleles within individual cells ( Wijgerde et al . , 1995; Raj et al . , 2006 ) , but only provide snap-shot measurements of activity . The analysis of gene expression in single living cells using real-time direct RNA imaging systems confirms these pulsatile kinetics ( Chubb et al . , 2006; Larson et al . , 2013; Martin et al . , 2013 ) . However , direct RNA analysis is technically challenging and relatively low throughput , even over short time periods . An alternative is the use of reporter gene analysis . Whilst being indirect , this can be combined with mathematical modelling ( Suter et al . , 2011; Harper et al . , 2011 ) to give a quantitative description of the dynamics of single cell gene expression . The pituitary gland is an excellent model system in which to address how dynamic changes in gene activity are regulated in vivo . The gland is composed of multiple cell lineages that are regulated by external signalling inputs from the hypothalamus and circulation ( Featherstone et al . , 2012 ) , as well as through complex paracrine signalling within the gland ( Denef , 2008 ) . The spatial positioning of cells is organised with cell networks facilitating the propagation of signals across the gland ( Le Tissier et al . , 2012; Mollard et al . , 2012 ) . The distinct cell types of the pituitary secrete specific hormones in an organised manner in response to developmental and environmental cues , which has proved to be a useful model system for investigating pituitary-tissue-specific and regulated gene transcription dynamics ( Featherstone et al . , 2012 ) . Prolactin ( PRL ) is an important pituitary-derived hormone with multiple functions that is secreted from pituitary lactotrophic cells and controlled in a complex way in response to both acute and long-term signals ( Featherstone et al . , 2012; Ben-Jonathan et al . , 2008 ) . The human PRL ( hPRL ) gene displays bursting activity in cell lines and primary cells , with variable periods of active and inactive transcriptional states , including a refractory period in the inactive state ( Harper et al . , 2011 ) . For hPRL , the dynamics observed in dispersed cells are compatible with a binary mathematical model in which there is an 'off' and an 'on' state together with a preparatory or 'primed' state . The transcription machinery transitions between these states as off – primed – on – off , with the time in each state exponentially distributed ( Suter et al . , 2011; Harper et al . , 2011 ) . We call this the telegraph process with priming . Quantitative imaging of hPRL reporter gene expression ( Spiller et al . , 2010 ) has been used to describe PRL gene activity in different physiological states of the pituitary gland ( Featherstone et al . , 2011; Harper et al . , 2010 ) . Here , we employ a new mathematical and statistical model to estimate not only the timing but also different levels of transcriptional activity . This provides a quantitative framework by which to explore temporal and spatial PRL gene expression in single cells within the anterior pituitary gland . We show that shorter durations of activity occur at high transcription rates in immature pituitary glands compared to the adult pituitary; however , we were unable to detect differences in the distribution of transcription rates in different pituitary states . These results suggest that dynamics of gene activity may have a common mechanistic basis in tissues and single cells at all stages of development , which are directly regulated by the developmental state of the tissue . Moreover , these dynamics demonstrate that transcription does not occur as a simple binary 'on-off' process with a single transcription rate in the 'on' state . In this study we also investigated how the spatial organisation of lactotroph cells within the pituitary affect the transcription dynamics of the hPRL gene . Transcription activity was coordinated between closely localised cells within the adult gland , but in developing pituitaries this transcriptional coordination was not evident , potentially due to an immature network of cellular communication . Perturbation of cell communication in the adult gland through trypsin-mediated digestion of extracellular proteins or pharmacological inhibition of intercellular gap junctions reduced transcriptional coordination between cells . These studies provide insight into how transcription dynamics throughout development and adult life relate to the state and structure of a complex and physiologically important tissue . To quantify PRL gene transcription dynamics within living pituitary tissue , we used transgenic rats that contain a destabilised EGFP ( d2EGFP ) reporter gene expressed under the control of the hPRL gene locus ( hPRL-d2EGFP ) ( Semprini et al . , 2009 ) . Imaging analyses were initially performed on adult male pituitary slices ( 300 µm coronal sections ) ( as depicted in Figure 1A ) for up to 48 hr in basal culture medium , with fluorescence measured from all actively transcribing cells within a field of approximately 100 cells . Tissue slices were subsequently stimulated with forskolin ( 5 µM ) , inducing at least a twofold increase in signal , showing that tissue slices remain viable in culture for long periods ( Figure 1B ) . Cells showed differing profiles of fluorescence activity during the imaging period ( Figure 1C ) . Generally cells increased hPRL-d2EGFP reporter gene expression , with the mean fluorescence activity reaching maximal levels at approximately 24 hr , potentially due to the removal of the pituitary from the inhibitory effects of hypothalamic dopamine ( Harper et al . , 2010 ) . Autocorrelation analyses did not identify a regular homogeneous period , and fluorescence activity showed a clear deviation from a white noise process ( Figure 1D ) . Further quantitative analysis using a stochastic switch model ( described in [Hey et al . , 2015] ) uncovered a clear statistical structure indicating a pulsatile transcriptional behaviour compatible with a pulsed telegraph process . The mathematical analysis of this is addressed in the modelling studies described below . Overall , we found that PRL transcription is dynamic within individual cells , with heterogeneous activity across the cell population . 10 . 7554/eLife . 08494 . 003Figure 1 . Patterns of prolactin gene transcription activity in pituitary tissue . ( A ) Schematic of the reporter construct and experimental approach used . ( B ) Images of d2EGFP expression in lactotroph cells of adult male pituitary tissue during a 46-hr imaging period , in basal culture media , and a subsequent 18-hr imaging period following stimulation with forskolin ( FSK ) . Images shown are a maximum intensity projection of a z-stack over 242 µm . Bar represents 100 µm . ( C ) Graph of d2EGFP fluorescence ( average intensity , arbitrary units ) from 20 individual cells ( representative of 101 cells analysed , from one experiment ) from adult male pituitary tissue . The black line represents the mean activity from all the cells analysed . The dark blue line represents the background fluorescence intensity ( mean from five areas ) . ( D ) Autocorrelation analysis of d2EGFP fluorescence intensity from cells shown in ( C ) , with the 95% confidence interval representing a white noise process shown between black lines . Data are representative of three independent experiments . For validation of single cell imaging see Figure 1—figure supplement 1 . d2EGFP , destabilised EGFP . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 00310 . 7554/eLife . 08494 . 004Figure 1—figure supplement 1 . Validation of single cell imaging . ( A ) Cell diameters were measured in two planes by confocal microscopy , as shown ( i-ii ) and compared to estimates of cell diameters from electron microscopy ( iii ) . Estimates of cell diameter were slightly smaller by electron microscopy , which may be accounted for by the processing of tissues for electron microscopy . Estimates of cell diameter are within the range of a typical eukaryotic cell , indicating that settings used for fluorescence microscopy gave good image resolution . ( B ) Areas of cells tracked from live-cell imaging are very similar to estimated cell areas calculated from cell diameters in ( A , iii ) ( 99–160 μm2 quartile range ) . The similar distribution of cell areas across three independent data sets shows the consistency of our cell tracking approach . Box plots show the median , interquartile range , min , and max values . Bar in image represents 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 004 The measurement of fluorescent protein activity by confocal microscopy enables the quantification of real-time gene transcription dynamics through mathematical modelling . The original transcription rate of the reporter gene is determined by considering the contribution of mRNA translation rates and mRNA and protein degradation rates to the observed fluorescent protein activity . Binary modelling of PRL transcription dynamics using this approach identified important parameters of gene activity , including a refractory period in the 'off' state ( Harper et al . , 2011 ) . We have extended this approach through the development of a stochastic switch model ( Hey et al . , 2015 ) , which enables the inference of different levels of transcription rate as well as the timing of switches between different levels of activity . The model uses a reversible jump Markov chain Monte Carlo ( Green , 1995 ) algorithm to produce a probability distribution over all possible transcriptional profiles for each cell . Post-processing of this distribution enables the extraction of a number of candidate transcriptional profiles , each with an associated probability of occurrence ( as depicted in Figure 2A ) . Consequently , the transcriptional analysis for each cell is a weighted analysis of all possible transcriptional profiles taking into account the probability of occurrence of each profile ( further details of the processing of the data and algorithm output are given in 'Materials and methods' ) . The fit of the model was tested through calculation of recursive residuals as a way of comparing the prediction from the model and the observed data ( for more information see Appendix G of [Hey et al . , 2015] ) . These showed no departure from the model assumptions , indicating that the stochastic switch model fitted the data well . 10 . 7554/eLife . 08494 . 005Figure 2 . Characterisation of prolactin transcription dynamics . ( A ) i ) Schematic of the parameters ( translation rate , mRNA and protein degradation rates ) used by the stochastic switch model to back-calculate to the original transcription rate , β ( t ) , of the reporter construct . For each fluorescence time series , there exist several mutually exclusive possible transcriptional profiles based on the sampled switch times . ii ) The graph shows the relationship between the measured fluorescence activity ( blue ) and possible switch times ( red , dashed lines represent ± 1 . 96 SD ) . iii ) The four possible profiles associated with the estimation of two transcriptional switches are shown along with their predicted probability of occurrence . ( B–E ) Characterisation of transcription rate activity of the reporter construct in cells maintained within adult pituitary tissue . ( B ) The number of switches in transcriptional states estimated during the imaging time-course . The frequency of switches is weighted by the probability of occurrence of each profile . Data are represented as the mean + SD from three independent experiments . ( C ) Distribution of the frequency of transcription rates estimated from fluorescence activity , with a weighted density calculated from all possible transcriptional profiles for each cell . The cumulative distribution of transcription rates from three independent experiments is shown in Figure 7B . ( D ) The duration of transcriptional states , with the caveat that the duration is the minimum duration as complete periods of activity were not detectable for most states . Transcription rates were binned into deciles and the associated minimum durations calculated . Boxplots are calculated by random sampling of the weighted transcription rate distributions for each cell . There is some evidence for longer durations at mid-range transcriptional rates ( 4th–8th decile ) , although longer observation windows may yield further trends . Data shown were pooled from three independent experiments . Boxplots represent the median and interquartile range ( IQR ) , with whiskers drawn 1 . 5xIQR away from the lower and upper quartile . ( E ) Heatmap of cell transcription rate patterns ordered by the timing of the first switch event . A single transcriptional profile was selected at random for each cell taking into account the probability of occurrence . Data shown in ( C ) and ( E ) are from a single experiment on adult pituitary , representative of three independent experiments . SD , standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 00510 . 7554/eLife . 08494 . 006Figure 2—figure supplement 1 . Processing of fluorescence time-series prior to transcription rate estimation . ( A ) The range of fluorescence activity resulting from hPRL gene expression exceeded the detection range of the microscope systems used . Data saturation would affect the transcription rates and timing of switches between different transcription rates estimated from the stochastic switch model . To overcome this issue , we collected d2EGFP signal at 492–544 nm and 546–611 nm wavelengths simultaneously using two detectors at different levels of sensitivity . Comparisons of signal intensity are shown in images i ) and ii ) and graph of expression patterns measured using the insensitive detector in iii ) , which is comparable to the graph shown in Figure 1C for the sensitive detector . Bars represent 100 µm . ( B ) i ) and ii ) Fluorescence intensity measurements from the two detectors were combined using linear regression analysis . iii ) The combined data ( black line ) consists of data from the sensitive detector ( green line ) until a threshold value ( red line ) , after which data from the less sensitive detector , rescaled using the linear regression equation , was used . hPRL , human prolactin . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 006 We applied the stochastic switch model to the time-lapse fluorescence imaging data to understand the switches in hPRL gene expression in the intact tissue and investigate whether multiple levels of transcriptional rate were employed during phases of active transcription . The model also estimates the timings of switches between statistically different transcriptional states . The fluorescence data were processed to ensure that they were linear and quantitative as described in Figure 2—figure supplement 1 . In adult pituitary tissue , we found that the majority of cells switch between different levels of PRL gene transcription rate up to two times within the 46-hr imaging period ( Figure 2B ) , consistent with the autocorrelation analysis . The transcription rates displayed by individual cells showed a continuous range of activity and were not restricted to simple binary on-off behaviour ( Figure 2C ) . Durations of activity varied with different transcription rates ( Figure 2D ) , and switches in activity were heterogeneous in both their timing and amplitude ( Figure 2E ) . A key aim was to establish the extent and mechanism of cell communication in the control of PRL gene expression in pituitary tissue . We characterised the organisation of lactotroph cells in pituitary tissue and found it to be indistinguishable from a random distribution ( Figure 3A ) . This does not preclude the presence of direct communication , or of an organised interaction network , between the cells ( as previously described [Hodson et al . , 2012] ) . We assessed the level of correlation between the fluorescence profiles of individual cells . The increase in PRL gene transcription observed in adult pituitary tissue enabled us to investigate whether cells would coordinate their activity in response to a release from the dopaminergic inhibition that would have occurred in vivo . Similar activity would be expected in vivo when changes in dopaminergic tone facilitate PRL expression and secretion , such as during circadian regulation and lactation ( Ben-Jonathan and Hnasko , 2001; Romano et al . , 2013; Le Tissier et al . , 2015 ) . For any two cells , we calculated the correlation between their fluorescence activity over the time-series as described in Figure 3B . Correlation was calculated as a function of the distance between individual cells , and showed that cells within approximately 35 µm ( estimates from all datasets were 25–35 µm ) of each other were more correlated than cells located further apart . The level of correlation over short distances was also significantly greater than the correlation profile obtained when fluorescence profiles were randomised between the cells , but the positioning information was maintained ( Figure 3B ) . This pattern of spatial correlation was not an artefact caused by limitations in image resolution or signal saturation ( Figure 3—figure supplement 1A , B ) . Spatial correlation of transcription was maintained when the number of cell pairs in each bin was normalised , indicating that this was not an artefact caused by small numbers of cells in the analysis ( Figure 3—figure supplement 1C ) . Spatial correlation also persisted throughout the time-course , indicating that coordination of PRL transcription was an ongoing process and not just a transient phenomenon facilitated by cellular stimulation ( Figure 3—figure supplement 1D ) . 10 . 7554/eLife . 08494 . 007Figure 3 . Spatial organisation of transcription activity in pituitary tissue . ( A ) Assessment of the spatial distribution of lactotroph cells with PRL gene transcriptional activity . i ) The median position of the cells over the time-course along with the field or convex hull occupied by the cells . ii ) Ripley’s k function was used to test whether the cellular distribution was random . The observed distribution ( Kobs ( r ) ) deviates from the theoretical prediction ( Ktheo ( r ) ) in the short r range due to the finite size of the cells , above which the cell distribution is found to be indistinguishable from random and within the 95% confidence interval ( Klo ( r ) – Khi ( r ) ) . ( B ) Correlation vs distance analysis of cells from adult male pituitary tissue . The correlation between two time series xi and yi measured at N times is defined as N−1∑i=1Nxiyi . i ) The distribution of distances between cell pairs . Correlation vs distance plots ii ) all cell data and iii ) cell pairs within 50 µm; indicate that proximally located cells have more similar activity than cells located further apart . The median and 90% confidence interval ( CI ) are shown along with the profile obtained from the 90% confidence interval of randomised of cellular transcription patterns . The range shown indicates the distances over which the randomised data were significantly greater ( paired t-test , p<0 . 001 ) to the non-randomised data . ( C ) Analysis of correlation mediated by a potential lactotroph cell network in adult male pituitary tissue . i ) Cell connectivity map with cells classified as connected or unconnected according to the threshold distance used . Connections between cells at a single time point are portrayed . ii ) Correlation was calculated between connected ( red ) and unconnected cells ( black ) in parallel with randomised data . iii ) The number of connected and unconnected cell pairs at various threshold distances . Connected cells were more correlated in their transcription activity than unconnected cells . Cell connectivity in ii ) -iii ) are based on median cell distances across the time-course to take into account cell movement . Data shown are representative of three independent experiments . PRL , prolactin . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 00710 . 7554/eLife . 08494 . 008Figure 3—figure supplement 1 . Characterisation of spatial organisation of prolactin transcription activity . ( A ) Spatial correlation between proximally positioned cells was not due to restricted resolution in the imaging data . The regions of interest ( ROI ) analysed were collapsed to 25% of the original area to ensure correlation was not due to overlap of signal between cells . i ) An example cell shows that this method caused the fluorescence signal to show large fluctuations in intensity but with the pattern of activity maintained . This is probably due to less averaging of signal intensity due to the smaller number of pixels . ii ) The pattern of correlation over distance was maintained for both sizes of ROIs ( 100% ROI in black and 25% ROI in red ) showing that imaging resolution was not a factor in causing increased correlation between proximally located cells . ( B ) Spatial correlation was maintained when imaging was performed at differing levels of sensitivity ( as discussed in Figure 2—figure supplement 1 ) . The correlation vs distance analyses show similar results for the two levels of signal detection ( sensitive detector shown in black , insensitive detector shown in red ) , although the absolute value of correlation changes due to the inclusion of either saturated values or zeros . Ranges shown indicate the distances over which the randomised data were significantly greater ( paired t-test , p<0 . 001 ) to the non-randomised data . Data were calculated in bins of 5 µm . ( C ) Spatial organisation of transcription was maintained when the number of cell pairs in each bin was normalised . Correlation vs distance plots are shown ( i ) along with the number of cell pairs in each bin ( ii ) . Graphs are coloured as in ( B ) . ( D ) Spatial organisation of transcription activity is maintained over 48 hr . Correlation of transcription patterns was not solely due to the stimulation of cell activity by the removal of dopamine as this correlation was maintained between proximally located cells over 48 hr and after 24 hr when hPRL-d2EGFP reporter gene activity declined . Data shown are from the sensitive detector , although similar results were obtained with data from the insensitive detector ( not shown ) . Correlation vs distance plots are shown as described in Figure 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 00810 . 7554/eLife . 08494 . 009Figure 3—figure supplement 2 . Correlation of transcription profiles within a cellular network structure . ( A ) Schematic showing the classification of cells into directly connected , indirectly connected and unconnected cell groups . ( B ) . Analysis of correlation between different cell groups ( classified as pictured in A ) with connections calculated using a defined threshold distance . Correlation was calculated in parallel with randomised data . ( C ) The number of directly connected , indirectly connected , and unconnected cell pairs at various threshold distances . Directly connected cells were more correlated in their transcription activity than indirectly connected or unconnected cells . Cell connectivity in ( B ) and ( C ) are based on median cell distances across the time-course to take into account cell movement . Data shown are from the same experiment in Figure 3C and are representative of three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 009 The range over which cells would be able to coordinate their transcription activity profiles through a cellular network structure was assessed . We found that cells connected directly , as defined by a threshold distance , had the greatest correlation in activity ( Figure 3—figure supplement 2 ) , but also that cells connected together via a potential cellular network structure were more correlated than cells that were unconnected ( Figure 3C ) . Overall , these data suggest that the mechanism by which lactotroph cells coordinate their functional activity may propagate across the pituitary through a cellular network . We analysed the timing and direction ( up or down ) of switches in transcription rate between cells , determined using the stochastic switch model . We tested the hypothesis that cells positioned close together , and that switch activity in the same direction , would do so more synchronously than cells located further apart ( scenarios and analysis illustrated in Figure 4 ) . Cells located closer together had a tendency to switch activity within a smaller time interval , but only if they switched activity in the same direction ( Figure 4B , C ) . 10 . 7554/eLife . 08494 . 010Figure 4 . Spatial organisation of stochastic switch model derived prolactin transcription dynamics . ( A ) Schematic outlining the hypothesis that was used to assess the spatial organisation of PRL transcription dynamics . The hypothesis was that two cells located closer together will tend to switch transcription in the same direction with more synchronous timing than cells that are located further apart . Moreover , a similar co-ordination in the timing of switches will not be observed if switches occur in the opposite direction . Comparisons are made to the index cell ( black ) . Red denotes cells that switch transcription rate in the same direction , blue denotes cells that switch transcription rate in the opposite direction . T1 is the time interval between cells located within 30 μm ( and dashed lines ) and T2 is the time interval between cells located more than 30 μm apart ( and solid lines ) . ( B ) Graph showing boxplots of switch timing intervals in cells that switch in the same direction and cells that switch in different directions , binned by the distance between cells . A rising trend is seen in the time interval between transcription rate switch events in cells that switch activity in the same direction ( red ) , but not in cells that switch activity in the opposite direction ( blue ) . Specifically , the median time interval between switch events is smallest in cells that are located within 30 μm and that switch activity in the same direction . Cumulative distributions and significance testing of these differences are shown in ( C ) . All pairwise switches are considered . Boxplots represent the median and interquartile range ( IQR ) , with whiskers drawn 1 . 5xIQR away from the lower and upper quartile . ( C ) The cumulative distribution of the time interval between switch events shows that cells within 30 µm that switch activity in the same direction ( red dashed line ) do so within a smaller time frame than cells located greater than 30 µm apart ( red solid line ) , the unsorted population ( black ) and cells that switch activity in opposite directions ( blue dashed and blue solid lines ) ( confirmed by significant p-value <0 . 01 of Kolmogorov-Smirnov tests ) . These were calculated by sampling at random , a pair of possible transcriptional profiles for each pair of cells . Data shown were pooled from three independent experiments . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 010 We investigated the involvement of signalling through cell junctions in the coordination of PRL transcription . We tested the hypothesis that the lower lactotroph cell density in developing pituitary tissue would provide less opportunity for cell junction signalling and thereby result in less transcriptional coordination between cells . We characterised the potential for cell junction signalling in E18 . 5 and P1 . 5 pituitaries by immunofluorescence ( Figure 5A , B ) and electron microscopy ( Figure 5C–K ) . Clustering of lactotroph cells increased in P1 . 5 pituitaries compared to E18 . 5 pituitaries ( Figure 5B , D and G ) . In the P1 . 5 pituitary gland lactotroph cells preferentially made contacts with other lactotroph cells , whereas in the E18 . 5 pituitary gland lactotroph cells were mainly isolated from each other ( Figure 5D , G ) . We specifically determined the presence of gap junctions and adherens junctions between lactotroph cells , as these have been implicated in the organisation and function of cell networks in the adult pituitary gland ( Morand et al . , 1996; Chauvet et al . , 2009; Kikuchi et al . , 2006 ) . In E18 . 5 pituitaries , only a small number of adherens junctions could be detected ( Figure 5E ) and the expression levels of the adherens junction proteins ( E- , N-cadherin and β-catenin ) were very low . E- , N-cadherin , and β-catenin were more abundantly expressed in P1 . 5 pituitaries than in E18 . 5 pituitaries ( Figure 5B ) , which coincided with increased numbers of adherens junctions as well as gap junctions and tight junctions ( Figure 5E , H ) . In addition to the presence of visually normal junctions ( Figure 5I , J ) , we also detected abnormal junctions where cadherin expression at the membrane could be detected but the characteristic thickening of the membrane was absent ( Figure 5K ) . These data indicate that , although the potential for communication between lactotroph cells increases during development , cell junction communication in P1 . 5 pituitaries may still be immature or atypical of the communication that occurs in the adult gland . 10 . 7554/eLife . 08494 . 011Figure 5 . Characterisation of cell signalling potential in immature pituitary glands . ( A , B ) Immunofluorescence analysis of lactotroph cell connectivity in developing pituitaries . ( A ) Immunofluorescence images show co-expression of luciferase antibody ( green ) , used to identify lactotroph cells , with PRL , N-Cadherin , E-Cadherin , and β-catenin ( red ) in paraffin-embedded sections from P1 . 5 PRL-Luc49 rats . Nuclei were counterstained with DAPI ( blue ) . Right: 8x crop images . Bars in images represent 50 µm . ( B ) Table showing the proportion of lactotroph cell clustering in E18 . 5 and P1 . 5 pituitary tissue and the level of co-expression of the luciferase protein and the protein of interest indicated , counted from immunofluorescence images . ( C–K ) Electron microscopy analysis of lactotroph cell connectivity in E18 . 5 ( C-E ) and P1 . 5 ( F-K ) pituitaries . ( C , F ) Representative image of two luciferase positive cells detected through immunogold labelling of luciferase antibody . Bars represent 1 µm . ( D , G ) Graph of the number of cells contacting an individual luciferase-positive cell . Undiff , undifferentiated cell . Luc , luciferase-expressing cell . FS , folliculostellate cell . PRL , lactotroph . GH , somatotroph . LH , gonadotroph . ( E , H ) Graph quantifying the different types of cell junctions present between luciferase-positive cells . Data in D-E and G-H are represented as mean + SEM . ( I , J ) Electron micrograph of a visually normal adherens junction ( AdJ ) , a visually normal tight junction ( TJ ) and a visually normal gap junction ( GJ ) in P1 . 5 pituitaries . Bars represent 200 nm . ( K ) Electron micrograph of an abnormal adherens junction in P1 . 5 pituitary . Bar represents 200 nm . Information and validation of antibodies used are presented in Figure 5—figure supplement 1 . SEM , standard error of the mean . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 01110 . 7554/eLife . 08494 . 012Figure 5—figure supplement 1 . Description and validation of antibodies used . Information and validation of antibodies used in analyses presented in Figure 5 and Figure 8—figure supplement 1 . ( A ) Table describing antibodies used . ( B ) Antibodies were validated by western blot to determine the specificity of the antibody and to ensure a protein of the appropriate size was detected . NS = non-specific binding . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 012 Profiles of hPRL-d2EGFP reporter gene activity in developing pituitaries were different to those seen in the adult pituitary . In E18 . 5 pituitaries , d2EGFP signal was initially very low , but showed a sustained increase leading to a greater number of cells being detected at the end of the imaging period ( Figure 6A , B ) . In P1 . 5 pituitaries , we again detected pulsatile transcription , with signal increasing after approximately 20 hr of imaging ( Figure 6E , F ) . Autocorrelation analysis of hPRL-d2EGFP fluorescence profiles from both E18 . 5 and P1 . 5 pituitaries indicated that the majority of cells showed only a single episode of PRL transcription with no dominant period being evident ( Figure 6—figure supplement 1 ) . Spatial correlation analyses of fluorescence activity , showed that PRL transcribing lactotroph cells in developing pituitaries were more sparsely distributed in comparison to adult pituitaries ( Figure 6C , G ) , as expected , and that the correlation between closely localised cells was reduced in comparison to the correlation profiles detected in adult tissues ( Figure 3B and 6D , H ) . 10 . 7554/eLife . 08494 . 013Figure 6 . Patterns and spatial organisation of prolactin gene transcription activity in immature pituitary tissue . ( A , B ) Activity of the hPRL-d2EGFP reporter construct in single cells in E18 . 5 pituitary tissue over 46 hr . ( A ) Images of d2EGFP expression in lactotroph cells in E18 . 5-day-old pituitary tissue ( male ) . ( B ) Fluorescence profiles from 20 individual cells , representative of 136 cells analysed ( average intensity , arbitrary units ) . The black line represents the mean average activity from all the cells analysed ( 136 cells ) , the dark blue line represents the background fluorescence intensity ( mean from five areas ) . ( C , D ) Spatial correlation between fluorescence profiles of PRL transcription activity is reduced in E18 . 5 pituitary tissue in comparison to adult tissue . ( C ) The distribution of cell pairs in E18 . 5 tissue is compared to the adult cellular distribution . ( D ) Correlation vs distance plots show no large differences between randomised and non-randomised data over short cell-to-cell distances . ( E , F ) Activity of the hPRL-d2EGFP reporter construct in single cells in P1 . 5 pituitary tissue over 46 hr . ( E ) Images of d2EGFP expression in lactotroph cells in P1 . 5-day-old pituitary tissue ( female ) . ( F ) Fluorescence profiles from 20 individual cells , representative of 115 cells analysed . Lines are coloured as in ( A ) with the mean representing the activity from all cells analysed . ( G , H ) Spatial correlation between fluorescence profiles of PRL transcription activity was reduced in P1 . 5 pituitary tissue in comparison to adult tissue . ( G ) The distribution of cell pairs in P1 . 5 tissue is compared to the adult cellular distribution . ( H ) Correlation vs distance plots show no large differences between randomised and non-randomised data over short cell-to-cell distances . Correlation vs distance plots are shown as described in Figure 3B . Images shown are maximum intensity projections of zstacks ( E18 . 5: 165 µm; P1 . 5: 196 µm ) . Bar in images represents 100 µm . E18 . 5 ( 181 cells ) and P1 . 5 ( 217 cells ) data shown were taken from two independent representative experiments . hPRL , human prolactin . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 01310 . 7554/eLife . 08494 . 014Figure 6—figure supplement 1 . Autocorrelation analysis of fluorescence profiles from developing pituitary tissue . Autocorrelation analysis of d2EGFP fluorescence intensity from cells shown in Figure 6B and 6F , with the 95% confidence interval representing a white noise process shown between black lines . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 01410 . 7554/eLife . 08494 . 015Figure 6—figure supplement 2 . Comparison of prolactin transcription activity in adult pituitary tissue in different medium . Graph showing the population transcriptional response from cells within adult pituitary tissue slices cultured in either 10% FBS or 50% rat serum supplemented medium in a closed system showing that differences in transcription profiles between immature and mature pituitaries was not due to the different culture media used . Prl-Luc49 rats were used and photon counts were collected using a photomultiplier tube ( PMT ) system . Data are from three independent pituitary cultures from two different animals ( n = 6 ) for each condition , with the mean and SD shown . SD , standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 015 Analysis of fluorescence imaging profiles using the stochastic switch model was used to infer the underlying hPRL transcription dynamics in single cells at different stages of pituitary development . The transcriptional switch data ( using data from cells as shown in Figure 2A iii ) were visualised for all cells in each tissue sample ( Figure 7A ) . Inspection of these data suggested that periods of active transcription tend to be of longer duration in adult compared to immature tissue . Direct analyses found no evidence for changes in the distribution of transcription rates at different stages of development ( Figure 7B and Figure 7—figure supplement 2A ) , even when transcription rates were grouped into low and active states ( Figure 7—figure supplement 1A and Figure 7—figure supplement 2B ) . However , the number of switches between different rates of activity appeared to be lower in cells in P1 . 5 pituitary tissue compared to in adult and E18 . 5 tissues ( Figure 7C ) . We also found that pulses of transcription in the highest quartile of transcription rates in E18 . 5 tissues were clearly of shorter duration than those in the bottom 75% , and in comparison to durations of activity in P1 . 5 and adult tissue ( Figure 7D and Figure 7—figure supplement 2A ) . This was also apparent when transcription rates were divided into low and active states and indicates that transcription occurs in a more pulsatile manner in embryonic pituitaries than in more mature tissues ( Figure 7—figure supplement 1B and Figure 7—figure supplement 2B ) . Where there was more than one switch ( and thus the full duration of an interswitch transcriptional state could be determined ) , the time to the next switch was shorter in immature tissue compared to adult tissue ( Figure 7E ) . These data indicate that transcription dynamics are more stable in the adult tissue . No evidence was obtained for spatial coordination of transcription rates in developing pituitaries ( Figure 7—figure supplement 3 ) . 10 . 7554/eLife . 08494 . 016Figure 7 . Comparison of prolactin transcription dynamics in different pituitary states . ( A ) The stochastic switch model enables comparison of transcription dynamics in different pituitary states . Example plots of transcription rates estimated using the stochastic switch model . Analyses of populations of cells are shown from different states of pituitary tissue development . Each cell is associated with a single transcriptional profile obtained by random sampling of all the possible profiles for that cell ( shown as in Figure 2A iii , with transcription rate and switch timings depicted ) . Data shown were from a single experiment representative of independent experiments ( i: E18 . 5 n = 181 cells , ii: P1 . 5 n = 217 cells , iii: Adult n = 305 cells ) . ( B–E ) Characterisation of transcription dynamics estimated using the stochastic switch model . ( B ) The cumulative distribution function of the frequency of different transcription rates was estimated using a weighted kernel . No clear difference between pituitary developmental states was evident . ( C ) Weighted histogram of the number of switches for each developmental state . Data , pooled from independent experiments , show the mean + SD , calculated by weighting the number of switches in each profile by the probability of occurrence . P1 . 5 has significantly fewer switches to Adult and E18 . 5 ( p<0 . 01 Mann-Whitney U test of the pooled distributions ) . ( D ) The duration of transcription rates for each tissue state . Transcription rates were binned into quartiles , with the first three lower quartiles grouped together ( <75 ) as one bin and the highest quartile as the other bin ( >75 ) . The duration of transcriptional states was calculated for each bin . The duration represents the minimum duration spent in each transcriptional state , as not all transcriptional states were completely observed within the time-frame of the imaging experiment . Both E18 . 5 datasets showed a significant difference in the duration spent in low transcriptional states ( lower 3 quartiles ) to high transcriptional states ( upper quartile ) as did one P1 . 5 dataset and two Adult datasets . Moreover , the duration spent in the upper quartile of the E18 . 5 ( pooled data ) was significantly different to the duration spent in the upper quartile of either the pooled data of P1 . 5 or Adult . Significance was assessed through the distribution of the p-values calculated from a Mann-Whitney bootstrap ( Figure 7—figure supplement 2 ) . Boxplots were calculated by a random sampling of the weighted transcription rate distributions and the associated durations . ( E ) The graph shows a kernel density estimate of the time between any two switches using data from panel D , for cells that displayed more than one switch so that the total duration spent in a single transcriptional state was estimated . The inter-switch time in the adult tissues is longer than the inter switch times of either the E18 . 5 or P1 . 5 tissues . E18 . 5 = orange , P1 . 5 = red , Adult = black . All boxplots represent the median and interquartile range ( IQR ) , with whiskers drawn 1 . 5xIQR away from the lower and upper quartile . SD , standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 01610 . 7554/eLife . 08494 . 017Figure 7—figure supplement 1 . Characterisation of prolactin transcription dynamics . ( A ) Characterisation of transcription rates , determined using the stochastic switch model , in E18 . 5 , P1 . 5 , and adult pituitaries . Rates were classified as ‘Low’ if they were preceded and followed by higher rates; all other rates were classified as ‘Active’ even if they were preceded by a higher level of activity , given that they were always followed by a further down-switch . Cells with no switches were removed from the analysis . Graphs are of the kernel density estimate of the cumulative distribution of transcription rates and show that low transcription rates were reduced compared to the active transcription rates . There is some evidence of reduced transcriptional rates in the ‘active’ states ( lower rates relative to fold change of overall tissue ) in the P1 . 5 datasets , but no clear difference in the distribution of rates employed by cells can be seen between Adult and E18 . 5 . ( B ) Durations spent at 'low' ( L ) and 'active' ( A ) transcription rates . Boxplots of the duration spent in either an 'active' or 'low' transcriptional state . These are obtained by sampling from the weighted density of transcriptional states and calculating the associated duration . These show that 'active' transcription in E18 . 5 and P1 . 5 pituitaries occurs over shorter time periods than in adult pituitaries , as calculated through bootstrap Mann-Whitney U-tests . All boxplots represent the median and interquartile range ( IQR ) , with whiskers drawn 1 . 5xIQR away from the lower and upper quartile . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 01710 . 7554/eLife . 08494 . 018Figure 7—figure supplement 2 . Significance testing of transcriptional state durations . The boxplots in Figure 7D and Figure 7—figure supplement 1B are a single sample obtained from the weighted distribution of transcriptional rates . In order to test for significance difference between groups , bootstrap samples were obtained and a Mann-Whitney U-test was performed on each sample . The histograms of the p-values of each of these tests are shown in Figure 7—figure supplement 2A ( corresponding to the tests on Figure 7D ) and in Figure 7—figure supplement 2B ( corresponding to the tests on Figure 7—figure supplement 1B ) . The vertical line indicates a p-value of 0 . 5% ( an overall significance level of 5% with Bonferroni correction to account for multiple testing ) . Significance is indicated when the empirical histogram of the sampled p-values lies to the left of the red line . ( A ) Significant differences were found when comparing the duration spent in the lower transcriptional rates ( lower three quartiles ) or higher transcriptional rates ( upper quartile ) for datasets; Adult-2 , Adult-3 , E18 . 5-1 and E18 . 5-2 . Significant differences were also found between the duration spent in the high transcriptional states of pooled datasets for E18 . 5 vs Adult and E18 . 5 vs P1 . 5 . ( B ) Significant differences were found when comparing the duration spent in low ( L ) to active ( A ) transcriptional rates for datasets; Adult-1 , Adult-3 , P1 . 5-1 , P1 . 5-2 , E18 . 5-1 and E18 . 5-2 . In addition , there is a significant difference in the duration spent in the high transcriptional states of the pooled datasets for E18 . 5 vs Adult and P1 . 5 vs Adult . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 01810 . 7554/eLife . 08494 . 019Figure 7—figure supplement 3 . Spatial organisation of transcription switch profiles in developing pituitaries . The spatial organisation of transcriptional switch profiles determined by the stochastic switch model in developing pituitaries was performed as outlined in Figure 4 . ( A ) Graph of boxplots of switch timing intervals in cells that switch in the same direction and cells that switch in different directions , binned by the distance between cells . All pairwise switches are considered . An increase in the time interval between switches in transcription , which occur in the same direction , is seen with increasing distance between the cells in adult pituitary tissues but not in E18 . 5 or P1 . 5 pituitary tissues . ( B ) The cumulative distribution of the time interval between switch events . In adult pituitary tissue , cells within 30 µm that switch activity in the same direction do so within a smaller time frame than cells located greater than 30 µm apart , the unsorted population and cells that switch activity in opposite directions ( confirmed by significant p-value <0 . 01 of Kolmogorov-Smirnov tests ) . In E18 . 5 and P1 . 5 pituitary tissues , this trend is not seen , as cells within 30 µm that switch activity in the same direction do so within a similar time frame to the whole cell population . In pooled data from P1 . 5 tissues , a significant difference ( p-value <0 . 01 of Kolmogorov-Smirnov tests ) is seen in the distribution of time intervals between cells that switch transcription activity in the same direction that are less than 30 µm apart and those that are greater than 30 µm apart , but this was not consistent across all data sets . All boxplots represent the median and interquartile range ( IQR ) , with whiskers drawn 1 . 5xIQR away from the lower and upper quartile . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 019 We next investigated the role of cell junctions in the spatial coordination of transcription in adult pituitary tissue . Trypsin was used as a non-specific protease to digest extracellular proteins and thereby abolish outside-in cell signalling , without tissue disaggregation so that cells were maintained in a tissue environment . Trypsin reduced protein levels of adherens junction proteins E- and N-Cadherin and the gap junction protein Connexin 43 , whilst β-catenin , an intracellular component of adherens junctions was unaffected ( Figure 8—figure supplement 1 ) . Fluorescence profiles of hPRL gene expression from cells in trypsin-treated tissue showed an overall increase in expression levels during the time-course , as did control tissue ( Figure 8A ) . Lactotroph cells in trypsin-treated tissue appeared less connected and had a greater intercellular distance than lactotroph cells in untreated tissue ( Figure 8B , C ) . Spatio-temporal analyses showed a reduction in correlation between the fluorescence profiles of closely localised cells ( Figure 8D ) , indicating that cell communication is important for the coordination of PRL transcription dynamics between lactotroph cells in pituitary tissue . 10 . 7554/eLife . 08494 . 020Figure 8 . Cell communication influences the spatial organisation of prolactin transcription dynamics . ( A ) Comparison of fluorescence profiles of hPRL-d2EGFP reporter gene activity from individual cells in control and trypsin-treated tissue . Cells in trypsin-treated tissue appeared less synchronised over time , but still showed an overall rise in activity as shown by the mean activity ( black ) . The level of background fluorescence is shown in dark blue ( mean from five areas ) . ( B–D ) Spatial correlation between fluorescence profiles of hPRL transcription activity is reduced in trypsin-treated tissue . ( B ) Images of cells within control and trypsin-treated tissue show that the distribution of cells and contacts between d2EGFP-expressing cells appeared altered following trypsin treatment . Bar represents 100 µm . ( C ) The intercellular distance between cells from control and trypsin-treated tissue was calculated as the median distance over the fluorescence imaging time-course ( shown in A ) . ( D ) Correlation vs distance analyses showed a reduction in the difference between non-randomised and randomised data in trypsin-treated tissue compared to control , indicating a reduction in the spatial influence on transcription . ( E–H ) Inhibitors of gap junction signalling were used to assess whether juxtacrine signalling is influential in coordinating PRL transcription activity . ( E ) Real-time luminescence activity from populations of cells in primary cultures show that gap junction inhibitors ( 18α-glycyrrhetinic acid , AGA , and palmitoleic acid , PA ) had little effect on overall PRL gene expression . Forskolin ( FSK ) was used as a positive control . ( F ) Fluorescence profiles of single cells in AGA treated tissue . Data are represented as described in ( A ) . Transcription activity increased during the time-course ( mean activity , black ) , similarly to control tissue ( A ) . ( G ) The intercellular distance between cells from control and AGA-treated tissue was calculated as the median distance over the fluorescence imaging time-course ( shown in A , F ) . ( H ) Correlation vs distance analyses showed a reduction in the difference between non-randomised and randomised fluorescence profiles in AGA treated tissue compared to control tissue indicating a reduction in the spatial coordination of transcription . Correlation vs distance plots are shown as described in Figure 3B . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 02010 . 7554/eLife . 08494 . 021Figure 8—figure supplement 1 . Effect of trypsin on cell junction proteins . ( A ) Schematic of trypsin treatment regime . Tissues were incubated in trypsin , at 37°C for 2 hr without disaggregation and then cultured for defined time periods before analysis by western blot . ( B ) Representative western blot of E-Cadherin expression following trypsin treatment . The graph shows the quantification of E-Cadherin following trypsin treatment , expressed as the fold change from the initial untreated sample ( 0 hr ) , represented by the mean and SD . ( C ) Representative western blot of N-Cadherin expression following trypsin treatment . Protein expression levels were quantified as described in ( B ) and shown in the graph . ( D ) Representative western blot of Connexin43 expression following trypsin treatment . Protein expression levels were quantified as described in ( B ) and shown in the graph . ( E ) Representative western blot of β-Catenin expression following trypsin treatment . Protein expression levels were quantified as described in ( B ) and shown in the graph . SD , standard deviation . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 021 We assessed the transcription response of cells exposed to gap junction inhibitors ( 18alpha-glycyrrhetinic acid , AGA , and palmitoleic acid , PA [Juszczak and Swiergiel , 2009] ) . Real-time luminometry on pituitary primary cell cultures showed little change in transcription activity at the population level ( Figure 8E ) . Fluorescence profiles of hPRL-d2EGFP reporter gene activity in AGA-treated tissue were similar to fluorescence profiles measured in untreated tissue ( Figure 8A , F ) indicating that the gap junction inhibitor AGA had no effect on PRL gene transcription dynamics . AGA also had no effect on the spatial distribution of the reporter-gene-expressing cells ( Figure 8G ) . However , correlation between closely localised cells was reduced ( Figure 8H ) , indicating a decrease in the coordination of PRL transcription between cells . Overall , these data indicate that cellular communication , at least in part through gap junctions facilitates the coordination of transcriptional activity between lactotroph cells . Quantitative studies of transcription dynamics have added new levels of complexity to our understanding of gene expression and its regulation . Direct RNA counting techniques ( MS2 , or RNA-FISH ) and the imaging of indirect fluorescence and luminescence reporter gene expression has shown that mammalian gene expression is often pulsatile ( or 'bursty' ) with different genes showing varying characteristics of activity ( Sanchez and Golding , 2013; Coulon et al . , 2013; Suter et al . , 2011; Spiller et al . , 2010 ) . These studies , performed in single cell systems , raise the question as to how apparently uncoordinated and heterogeneous dynamics enable integrated tissue-level responses to physiological stimuli . We have assessed PRL transcription dynamics within pituitary tissue , utilising newly derived mathematical models to define transcription activity . Cells displayed a continuous distribution of transcription rates with heterogeneous patterns of activity across the cell population . Embryonic pituitary glands displayed shorter durations of high transcription rates compared to adult pituitary tissue , which could reflect epigenetic changes during tissue development . We also characterised the spatial organisation of PRL gene expression within lactotroph cells of the pituitary gland and found evidence for the local coordination of transcription dynamics that is potentially mediated by intercellular signalling providing insights into how transcriptional timing is organised in tissue systems . Over the past decade , efforts have been made to mathematically model transcription activity to provide a better mechanistic understanding of gene regulation ( Sanchez et al . , 2013; Larson et al . , 2009 ) . Poissonian distributions of mRNA production , where mRNAs are produced in random , uncorrelated events , have been described ( Zenklusen et al . , 2008; So et al . , 2011 ) . However , the prevailing model for mammalian transcription dynamics is the Random Telegraph Model , which describes 'bursty' gene expression , where the gene exists in two states , either 'on' or 'off' , with transcripts produced at a defined rate in the 'on' period ( Larson et al . , 2009; Peccoud and Ycart , 1995 ) . Using binary modelling , we and others have shown that there is a refractory period in the 'off' state , but not in the 'on' state , indicating that there are significant differences between the kinetics of gene activation and inactivation ( Suter et al . , 2011; Harper et al . , 2011 ) . Binary modelling of transcription dynamics is likely to represent an oversimplification of the true transcription process . Therefore , we developed a stochastic switch model , which allowed us to estimate transcription rates at differing levels and defined the timing of switches between different rates ( Hey et al . , 2015 ) . This model enabled us to characterise transcription activity in cells maintained in tissue , where individual transcriptional states were sustained for long periods and complete cycles of activity were not often detected . Using the stochastic switch model , we found a continuous distribution of transcription rates , that is differential 'on' states , across the cell population , with heterogeneous timing in transcriptional switches between cells . This heterogeneity in transcriptional activity is likely to result from intrinsic factors previously shown to influence PRL transcription ( Harper et al . , 2011 ) , together with extrinsic factors , which include the specialisation of lactotroph cells into distinct subtypes ( Christian et al . , 2007 ) . To understand the role of stochastic transcriptional processes in tissue biology , it is important to determine how the properties of transcription dynamics are modulated to facilitate changes in gene expression under different physiological states . Modelling of transcriptional activity using a random telegraph process has suggested that levels of gene expression can be controlled through digital responses where the burst frequency or duration is altered , but not the transcription rate ( Larson et al . , 2013 ) . Similarly , digital responses in transcription activity have been detected in single cell systems where the probability that cells are recruited to an expressing population changes under different conditions ( Chubb et al . , 2006; Walters et al . , 1995; Kar et al . , 2012 ) . In contrast , different kinetic transcriptional responses including changes to transcription rate and durations of 'on' activity have been found for the connective tissue growth factor ( ctgf ) gene in response to different stimuli ( Molina et al . , 2013 ) , and for housekeeping genes in Dictyostelium ( Muramoto et al . , 2012 ) . Overall , the total level of transcription in a given pulse will depend not only on the length of the pulse but also on the rate of transcription during the pulse . Different rates of transcription will depend on levels of RNA polymerase II loading , which may be controlled by different chromatin and promoter states . We observed a continuous distribution of transcription rates within cell populations , indicating that different levels of activity are attainable . However , at the population level similar distributions of activity were detected in all developmental states analysed . Thus , differences in transcription rate contribute to heterogeneous activity at the population level and may be important in maintaining tissue function . In different developmental states , we found changes in the duration of high transcription rates between embryonic and more mature pituitary glands , indicative of a more pulsatile activity in immature tissues . Thus , changes to the duration of activity appear more prominent in facilitating changes in the level of gene expression than changes to transcription rate . Transcriptional stochasticity within cellular populations may be advantageous in maintaining population fitness to changing environments ( Thattai , 2004 ) , or facilitate cell fate choices ( Chang et al . , 2008; Wernet et al . , 2006 ) . However , the role of stochastic gene expression in tissue systems where integrated responses to physiological demand are required is less clear . It has been proposed that heterogeneous responses may facilitate robust tissue-level responses and potentially avoid inappropriate amplification of signals through feedback mechanisms ( Paszek et al . , 2010 ) . In contrast , mechanisms to reduce expression level heterogeneity have been described in processes such as patterning and specification in other species ( Little et al . , 2013; Raj et al . , 2010 ) . A recent study used single-molecule RNA-FISH at single points in time to define bursting transcriptional behaviour in fixed liver tissue and identified polyploidy as a mechanism to reduce intrinsic variability between cells ( Bahar Halpern et al . , 2015 ) . The pituitary gland is an excellent model system in which to investigate cellular population responses to physiological signals . The gland is composed of multiple cell types that are spatially organised within the pituitary , several of which have been suggested to form interdigitated cellular networks ( Le Tissier et al . , 2012; Mollard et al . , 2012; Hodson et al . , 2012; Fauquier et al . , 2001; Bonnefont et al . , 2005 ) . Lactotroph cells coordinate their calcium signalling in basal physiological states and more substantially during increased demand such as lactation ( Hodson et al . , 2012 ) . In this study , we have provided a quantitative analysis of lactotroph cell connectivity and shown that PRL transcription is coordinated between lactotroph cells over short distances ( 25–35 µm ) and propagated through a network structure . Transcriptional coordination was actively facilitated by intercellular signalling , and we have shown that this could be via juxtacrine signalling including gap junctions . Intercellular signalling has been shown to be important for coordinating other oscillatory systems such as the circadian clock in the suprachiasmatic nucleus ( Liu et al . , 2007 ) , the somite segmentation clock ( Horikawa et al . , 2006; Masamizu et al . , 2006 ) , and electrical coupling of and insulin secretion from pancreatic β cells ( Smolen et al . , 1993 ) . The global picture that arises is that transcription is highly stochastic but has some coordination of bursting at distances up to approximately 35 µm in adult pituitary tissue , but not at greater distances . In contrast there was no coordination at any intercellular distance in earlier developmental states . The limited short distance coordination between lactotroph cells in the adult tissue is not sufficiently strong to lose the key characteristic of cell-to-cell heterogeneity . However , it can be hypothesised that the global system of short range cell-to-cell communication may stabilise longer term changes in the expression level of the tissue , such as those associated with the oestrus cycle or lactation . Thus far the gland as a whole prolactin transcription is essentially random in that for the vast majority of cell pairs , the temporal pattern of their transcription is uncoordinated . Therefore , the law of large numbers guarantees stable long-term results in terms of the global response . Our work addresses how an endocrine tissue , such as the pituitary gland , generates a controlled output from a diverse set of intermingled cell types . The acute , medium , and long-term outputs of endocrine cells must be regulated across different time scales and to diverse environmental signals , while achieving accurate control of hormone expression . A key question has been whether cells behave similarly to each other or whether they operate in a heterogeneous autonomous fashion . Previous data have suggested the latter in isolated cells and cell lines . Our data now indicate that developing adult tissue structure exerts a coordinating effect on cell behaviour ( see Figure 9A ) . Our observation that cells display multi-state transcriptional behaviour ( 'off' , 'primed' and 'on' at various levels; outlined in Figure 9A ) , as opposed to simple binary 'on'/'off' behaviour , is suggestive of mechanisms that tune the overall output from the gland through the generation of graded responses . Changes in the duration of particular transcription states , as we observed with shorter 'on' periods in immature pituitary glands , also help to refine the overall output from the gland . A simulation ( Figure 9B ) shows how decreasing the duration of the 'primed' period allows a population of cells to switch to a new stable state of mRNA production . Such modification of activity may facilitate the differentiation of the tissue response without a risk of overshooting behaviour . This offers a new mechanism to explain how tissues integrate different signals with varying durations into appropriate responses . Examples of dynamic control of the pituitary gland include the acute suppression or activation by hypothalamic regulators , circadian response , and long-term behavioural changes through development , puberty , pregnancy , and ageing . 10 . 7554/eLife . 08494 . 022Figure 9 . Multi-state transcription dynamics achieve robust tissue level responses . ( A ) Organised cell networks and gap junction cell-cell signalling in adult pituitary glands ( lower panels ) , but not in immature pituitary glands ( upper panels ) , leads to local correlation of transcription between cells . Across a large population of cells transcription dynamics are essentially random and uncoordinated . Multi-state transcription dynamics ( 'off' , 'primed' and 'on' at various levels ) contribute to transcriptional heterogeneity between cells as the time spent in each state is exponentially distributed . The schematic indicates putative regional signalling and signalling gradients across the pituitary , for example from hypothalamic dopamine ( DA ) and systemic estradiol ( E2 ) . Segments of pituitary tissue are represented with lactotroph cells ( red ) connected via gap junctions ( black rectangles ) . Examples of single cell transcription profiles ( left plots ) and their relation to multi-state transcription dynamics ( right plots ) , with 'off' , 'primed' ( dashed lines ) and different levels of transcription in the 'on' state are shown . ( B ) Simulation of the telegraph model with priming demonstrates how decreasing the 'primed” period ( at the dashed line ) gives rise to a rapid robust response across a population of cells showing heterogeneous activity . The simulation was performed on fluorescence profiles from 100 cells , with the mean of these responses shown ( black line ) along with 5 representative cell responses ( coloured lines ) . DOI: http://dx . doi . org/10 . 7554/eLife . 08494 . 022 In summary , the results presented provide a quantitative framework of cellular transcriptional activity within living tissue . We demonstrate that there is local transcriptional coordination between prolactin-expressing cells in adult , but not in developing pituitary tissue . We have described quantitative analyses of transcription kinetics in different developmental states of the pituitary gland . In all states , we showed that transcription occurs at different levels of activity separated by statistically definable switches . We also detected changes in transcription dynamics between different stages of pituitary development . In the adult tissue , the cells show longer lived active states , associated with an altered environment in which there is greater cell-to-cell connectivity . We have identified potential mechanisms by which the combined dynamics of single cells within a tissue may achieve both acute and long-term functional adaptation of the expression of secreted regulatory proteins . Animal studies were performed under UK Home Office License and subject to local ethical committee review . The generation and characterisation of Fischer 344 transgenic rats with the luciferase ( PRL-Luc49 ) or destabilised GFP ( PRL-d2eGFP455 ) reporter genes under the control of the hPRL locus has been described previously ( Semprini et al . , 2009 ) . The BAC transgenes are integrated into the genome at a single site with high copy number in the PRL-Luc49 line and low copy number ( ≤5 ) in the PRL-d2eGFP455 line ( Harper et al . , 2010; Semprini et al . , 2009 ) . Animals were housed in humidity- ( 50 ± 10% ) and temperature- ( 20 ± 1°C ) controlled conditions in a 12 hr light:dark cycle with food ( rat and mouse standard diet; Special Diet Services , Witham , UK ) and water ad libitum . Timed matings were set-up with transgenic males and wild-type females with the detection of a vaginal plug the morning after mating considered as E0 . 5 . Animals were genotyped using DNA extracted from ear biopsies ( adults ) or tail biopsies ( fetal and neonates ) . Young animals were sexed using PCR conditions described in ( Featherstone et al . , 2011 ) . Genotyping was performed on PRL-Luc49 DNA as described in ( Featherstone et al . , 2011 ) . Genotyping of PRL-d2eGFP455 DNA was performed with the same conditions as for PRL-Luc49 DNA but with d2eGFP primers substituted for luciferase primers; d2eGFP forward , 5’-GACGACGGCAACTACAAGAAC -3’ and d2eGFP reverse , 5’-ACTCCAGCAGCACCATGTGAT -3’ . Animals were sacrificed by a schedule 1 method ( exposure to a rising concentration of CO2 followed by cervical dislocation ) followed by resection of pituitary glands . Coronal slices of adult pituitary glands ( 300 µm ) were cultured on 0 . 4-µm filter stages ( Greiner Bio-One , UK ) in 35-mm glass-coverslip-based dishes with access to air and medium ( DMEM + 4 . 5 g/l glucose , 10% ( v/v ) FBS , 1 mM sodium pyruvate , 100 U/ml penicillin , 100 µg/ml streptomycin , and 2 mM ultraglutamine ) ( Figure 1A ) . Whole pituitary glands from fetal ( E18 . 5 ) ( n = 2 ) or neonatal ( P1 . 5 ) ( n = 2 ) animals were treated as described in ( Featherstone et al . , 2011 ) except that luciferin was omitted from the medium and pituitaries were cultured on filter stages as described above . Adult pituitary tissue was either untreated ( n = 3 ) , treated with Trypsin ( 0 . 1% ( w/v ) Trypsin ( Sigma UK ) , 0 . 0045% ( w/v ) DNase I , 0 . 325% ( w/v ) BSA in HBSS ) ( n = 2 ) , or AGA ( 20 µM in 10% FBS medium ) ( n = 2 ) for 2 hr at 37°C prior to imaging . Pituitaries were imaged using Carl Zeiss laser scanning confocal microscopes ( LSM ) : Pascal , 710 and 780 , maintained at 37°C in PeCon XL incubators ( PeCon , Germany ) with a humidified atmosphere of 95% air and 5% CO2 and with a Fluar 10X magnification 0 . 5 NA objective . Excitation of d2EGFP was performed using an argon ion laser at 488 nm with emitted light captured through appropriate filters or a selected portion of the spectrum . All imaging was acquired as z-stacks with images captured in 15 min intervals for 48 hr in basal culture medium and then for 24 hr following forskolin stimulation ( Adult: 5 µM or Immature Pituitaries: 1 µM ) . Fluorescence from tissues was analysed as maximum intensity projections using ZEN 2010b ( Zeiss , UK ) or CellTracker software ( http://www2 . warwick . ac . uk/fac/sci/systemsbiology/staff/bretschneider/celltracker/ ) . Spatio-temporal analyses were performed by measuring the fluorescence intensity from all cells within an area , encompassing approximately 100 cells . The area analysed was always taken from the lateral edge of the pituitary to minimise differences between cellular activities that may exist across the gland . Regions of interest were drawn around cells and mean intensity data collected using CellTracker software . Cell areas analysed were consistent with the size of a typical eukaryotic cell ( Figure 1—figure supplement 1 ) . Matlab R2014a software ( MathWorks , UK ) including Bioinformatics and Statistical toolboxes , or the R programming language ( www . r-project . org ) were used for mathematical analyses . In all analyses , the positioning of cells was taken as the median x , y coordinates of the centroid . The spatial distribution of cells was tested using a 2D Poission process and Ripley’s K function ( Ripley , 1976 ) . Correlation analyses of transcription patterns from fluorescence data were performed in two ways: either using the Euclidean distance between cells or by a network analysis . Correlation based on the Euclidean distance was performed in two parts . In the first part , cell pairs were partitioned into bins according to the distance between the cells starting at 5 µm with 5 µm intervals . Correlation analysis was performed 99 times for each bin , with n cell pairs sampled with replacement , and with the 5 and 95 percentiles reported . In the second part , n time-series are randomly permuted while the spatial information is unaltered following which correlation coefficients are calculated as described above . A paired t-test was used to identify significant differences between space-time correlations from randomised and non-randomised data . Network correlation analysis was performed similarly to Euclidean distance correlation analyses except that cell pairs were divided into two bins , connected and unconnected , which was defined in a network approach as follows . Cells were assumed to be circular with the same diameter ( D ) . If the distance between two cells was smaller than D , they are considered to be connected . Furthermore cell connectivity is transitive so that if cells a and b are connected and so are cells b and c , then cells a and c are considered to be connected being part of a cell cluster , even if separated by a distance greater than D . Analysis is performed with a varying D . The inference of transcription rates from fluorescence data using the stochastic switch model was performed as described ( Hey et al . , 2015 ) , using the following stochastic reaction network:∅→β ( t ) mRNAmRNA→δm∅mRNA→αmRNA+ProteinProtein→δp∅ ( * ) where δm and δp denote the degradation rates of reporter mRNA and reporter protein respectively , α denotes the rate of translation and β ( t ) denotes the time varying rate of transcription . Specifically , the transcription function is given by , β ( t ) = βi for t∈[si−1 , si ) for i=1 , . . . , K , where K is the number of transcriptional switches , occurring at times s1 , s2 , . . . , sK and β1 , β2 , . . . , βK are the corresponding transcriptional rates . We impose no restriction to the form of the transcriptional levels but note that the conventional binary switch behaviour can be seen as a specific example where βi=βLOW if the gene is inactive in the time period [si−1 , si ) or βi = βHIGH if the gene is active . Assuming light intensity measurements are related to reporter protein levels through the equation , Y ( t ) =κ P ( t ) +ε ( t ) , ε ( t ) ~N ( 0 , σ2 ) , inference is performed through the linear noise approximation to the system in ( * ) to obtain the posterior transcriptional function for each single cell . To ensure model identifiability , we impose informative prior distributions about the degradation parameters , obtained from independent experiments as described in ( Finkenstädt et al . , 2013 ) . In addition , we specify a hierarchical framework over each dataset , as individual parameters are unlikely to change substantially . We confirmed that the degradation parameters in tissue samples were similar to estimates obtained using cell lines . In order to estimate both the number and positioning of transcriptional switches , we employ a reversible jump Markov chain Monte Carlo ( MCMC ) algorithm ( Green , 1995 ) . Consequently , the posterior distribution consists of all possible transcriptional profiles . In order to extract the information regarding the estimated transcriptional dynamics , the posterior samples go through a post-processing procedure outlined below: Thus , this post-processing procedure associates each single cell to a set of mutually exclusive transcriptional profiles . The analysis presented in the main paper has been calculated from the set of all possible transcriptional profiles , weighted by their probability of occurrence . Data of d2EGFP fluorescence measured in single cells and subsequent stochastic switch modelling of transcription activity have been deposited in the Dryad data repository ( Featherstone et al . , 2015 ) Immunofluorescence was performed on Prl-Luc49 immature pituitary glands maintained in situ to enable sectioning in the coronal orientation . Prl-Luc49 rats were used in immunofluorescence and electron microscopy analyses as multiple copies of the Prl-Luciferase transgene enabled the detection of low expressing PRL cells . Exposed pituitaries were fixed in Bouin’s solution ( Sigma , UK ) for 24 hr , washed with water and 70% ethanol and then embedded in wax and 5-µm sections cut and mounted onto poly-L-lysine slides ( Thermo Fisher , UK ) . Immunofluorescence staining was performed as previously described ( Semprini et al . , 2009 ) except that slides were pretreated with 3% ( v/v ) H2O2 , in methanol for 30 min following antigen retrieval . Pituitaries were stained with mouse anti-luciferase ( P . pyralis ) antibody ( Life Technologies , UK ) ; co-stained either with: mouse anti- ACTH ( Abcam , UK ) , goat anti-Pit1 ( Santa Cruz , UK ) , rabbit anti- PRL ( R51 , A McNeilly , Edinburgh , UK ) , mouse anti-E-Cadherin ( BD Biosciences , UK ) , rabbit anti- N-Cadherin ( Calbiochem , UK ) , mouse anti- β-Catenin ( BD Biosciences , UK ) and counterstained with DAPI . Briefly , all slides were blocked with blocking buffer ( 10% ( v/v ) donkey serum , 5% BSA ( w/v ) in PBS ) and stained with primary antibody in blocking buffer at 4°C overnight . The primary antibody was detected with anti-mouse-HRP ( Vector Labs , UK ) and Tyramide Signal Amplification- FITC ( PerkinElmer , UK ) . Slides were boiled in sodium citrate ( 10 mM pH6 ) for 2 min and left to stand in hot buffer for 30 min before sequentially blocking with biotin , streptavidin ( Vector Labs , UK ) and blocking buffer and application of second primary antibodies at 4°C , overnight . Detection of the second primary antibodies was with either anti- rabbit biotin , anti- goat biotin , or anti-mouse biotin ( Vector Labs , UK ) and subsequently with streptavidin- alexa546 ( Life Technologies , UK ) . Specificity of antibody labelling was confirmed in negative control sections in which the primary antibody was replaced with the appropriate non-immune serum . Slides were examined and images taken using a Zeiss Excitor confocal microscope with Fluar 20X magnification 0 . 75 NA objective . For each developmental stage three pituitaries were analysed ( two males and one female were analysed from two independent litters ) with co-localisation between antibodies counted across at least one whole pituitary slice . No sexual dimorphism was detected . Electron microscopy was performed on dissected immature pituitaries , which were processed and immunogold-labelled as previously described ( Abel et al . , 2013 ) . Briefly , the tissue was contrasted with uranyl acetate ( 2% ( w/v ) in distilled water ) , dehydrated in methanol and embedded in LR Gold resin . Ultrathin sections ( 50–80 nm ) were prepared using a Reichart-Jung ultracut microtome and mounted on nickel grids ( Agar Scientific , UK ) . For identification of adherens junctions , sections were immunogold-labelled for E-cadherin ( mouse anti- E-Cadherin: BD Biosciences , UK ) or β-catenin ( mouse anti-β-catenin: BD Biosciences , UK ) . In order to identify lactotroph cells , sections were immunogold labelled for PRL ( rabbit anti- PRL R51: A McNeilly , Edinburgh , UK ) and to identify luciferase-positive cells , sections were immunogold labelled for luciferase ( mouse anti-luciferase: LifeTechnologies , UK ) . Sections were counterstained with lead citrate and uranyl acetate and examined on a JOEL 1010 transmission electron microscope ( JOEL USA , USA ) . Specificity of antibody labelling was confirmed in negative control sections in which the primary antibody was replaced with the appropriate non-immune serum . For each pituitary ( n = 3 ) , 10 luciferase-immunogold labelled cells were identified and the adjacent cells and intercellular junctions identified on the basis of morphological criteria and counted . Primary cultures of adult female pituitary glands were prepared as described ( Featherstone et al . , 2011 ) . Cells were resuspended in medium ( DMEM + 4 . 5 g/l glucose , 10% FBS , 1 mM sodium pyruvate , 100 U/ml penicillin , 100 µg/ml streptomycin , and 2 mM ultraglutamine ) and plated ( 1 × 105/well ) in white plastic 96-well plates pre-treated with poly-L-Lysine at . Cells were allowed to recover for 24 hr ( 37°C , 5% CO2 ) after which the medium was replaced with serum-deprived medium ( DMEM + 4 . 5 g/l glucose , 0 . 25% ( w/v ) BSA , 1 mM sodium pyruvate , 100 U/ml penicillin , 100 µg/ml streptomycin and 2 mM ultraglutamine ) supplemented with 1 mM luciferin ( Biosynth , Switzerland ) and cells were incubated for a further 24 hr ( 37°C , 5% CO2 ) . Cells were then treated with either: DMSO ( control ) , 20 µM α-glycyrrhetinic acid ( Sigma , UK ) , 50 µM Palmitoleic acid ( Sigma , UK ) , or 5 µM forskolin ( Sigma , UK ) . Cell responses were measured using the FLUOstar Omega CO2- and temperature-controlled luminometer plate reader ( BMG Labtech ) with photon counts collected for 10s per well every 15 min for 24 hr . Three independent cultures were analysed with 3-4 replicates per treatment group . Pituitary tissue slices from Prl-Luc49 adult male rats were prepared as described in 'Preparation and culture of pituitary tissue' . Slices were cultured on filter stages in 35-mm dishes in either FBS supplemented recording medium ( DMEM + 4 . 5 g/l glucose , 10% ( v/v ) FBS , 1 mM sodium pyruvate , 100 U/ml penicillin , 100 µg/ml streptomycin , and 2 mM ultraglutamine , 10 mM HEPES , 1 mM luciferin ) or rat serum supplemented recording medium ( DMEM and F12 , 50% ( v/v ) serum from sacrificed animal , 100 U/ml penicillin , 100 µg/ml streptomycin , 2 mM ultraglutamine , 10 mM HEPES , 1 mM luciferin ) in a closed system at 37°C , to show that differences in transcriptional activity seen between adult and immature tissues was not due to the medium used ( Figure 6—figure supplement 2 ) . Bioluminescence emissions were recorded by photon multiplier assemblies ( Hamamatsu Photonics , UK ) with counts collected over a 1-min period . The mean and standard deviation were calculated from data collected over hourly periods . Two independent adult male rat pituitaries were analysed with each condition performed in triplicate . Pituitary tissue slices were prepared and either untreated or treated with trypsin for 2 hr at 37°C , as described in 'Preparation and culture of pituitary tissue' . Slices were washed three times in medium ( DMEM + 4 . 5 g/l glucose , 10% ( v/v ) FBS , 1 mM sodium pyruvate , 100 U/ml penicillin , 100 µg/ml streptomycin and 2 mM ultraglutamine ) and then cultured on filter stages as described previously at 37°C , 5% CO2 for either 0 , 24 , or 48 hr . Slices were lysed in RIPA buffer ( 50 mM Tris-HCl pH8 , 150 mM NaCl , 1% ( v/v ) Nonidet P40 , 0 . 5% ( w/v ) sodium deoxycholate , 0 . 1% ( w/v ) SDS ) with Complete Mini EDTA-free Protease Inhibitors ( Roche , UK ) . Three independent cultures of trypsin-treated and untreated tissue were analysed by western blot . Lysates prepared from cell lines or whole tissue samples were generated by lysis in RIPA buffer with Complete Mini EDTA-free Protease Inhibitors ( Roche , UK ) following washing with cold HBSS . All samples were subjected to SDS-PAGE ( 10% ) before transfer to nitrocellulose membrane . Primary antibodies ( rabbit anti- α-tubulin ( Proteintech , UK ) , mouse anti-E-Cadherin ( BD Biosciences , UK ) , mouse anti-β-catenin ( BD Biosciences , UK ) , rabbit anti-N-Cadherin ( Calbiochem , UK ) , mouse anti-connexin43 ( Santa Cruz Biotechnology , USA ) were applied overnight at 4°C in blocking buffer ( Tris buffered saline , 5% ( w/v ) dried milk , 0 . 1% ( v/v ) Tween20 ) , and species-specific horseradish peroxidase conjugated secondary antibodies ( GE Healthcare , UK ) were applied for 1 hr at room temperature . Staining was detected with Clarity Western ECL Substrate ( Biorad ) using Biomax XAR film ( Kodak , UK ) . Protein levels were standardised to α-tubulin expression following quantification of protein expression by determining the relative band size using ImageJ software . For further information on antibodies used see Figure 5—figure supplement 1 .	Although humans have thousands of genes , only a fraction of these are expressed in any given cell . Each cell type expresses only the genes that are relevant to its particular job or that are necessary for general cell maintenance . Even these genes are not expressed all the time: most cells express genes in bursts , and the cells that make up a tissue produce these bursts at different times . This makes it easier for the tissue to respond to new conditions . The pituitary gland , found at the base of the brain , is often studied to investigate changes in gene expression . The pituitary gland is found in all animals that have a backbone , and it makes and releases many different hormones . For example , one type of pituitary cell expresses the gene that encodes a hormone called prolactin . This hormone has a range of roles , including stimulating milk production and regulating fertility in mammals . The coordinated production of prolactin by pituitary cells is important for reproduction , but it is not clear how ( or whether ) individual prolactin-producing cells in the gland communicate to coordinate bursting patterns of expression of the prolactin gene . Featherstone et al . marked the prolactin-encoding gene in the pituitary cells of rats with a gene that encodes a fluorescent protein; this enabled the gene’s activity to be observed in thin slices of living tissue using a microscope . Mathematical models were then used to analyse the recorded expression patterns . The results showed that in a single cell , the bursts of expression of the prolactin gene are randomly timed . This means that although the expression activity of an individual cell is unpredictable , the overall activity of a group of cells can be precisely determined . The model also showed that cells coordinate when they express the prolactin gene to a greater extent with their near neighbours than with cells that are further away in the tissue . Featherstone et al . found that this coordination depends on structures ( called gap junctions ) that connect the cells and allow signalling between them , and this tissue organisation is established during early development . The mechanisms underlying the timing of the bursts remain to be discovered . The timing for the prolactin gene seems to be dominated by a minimum delay that must occur before the next burst can be reactivated . Future challenges also include determining whether coordinated gene expression occurs in other tissues and whether this coordination is disrupted in disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "cell", "biology", "computational", "and", "systems", "biology" ]	2016	Spatially coordinated dynamic gene transcription in living pituitary tissue
Gram-negative bacteria import essential nutrients such as iron and vitamin B12 through outer membrane receptors . This process utilizes proton motive force harvested by the Ton system made up of three inner membrane proteins , ExbB , ExbD and TonB . ExbB and ExbD form the proton channel that energizes uptake through TonB . Recently , crystal structures suggest that the ExbB pentamer is the scaffold . Here , we present structures of hexameric complexes of ExbB and ExbD revealed by X-ray crystallography and single particle cryo-EM . Image analysis shows that hexameric and pentameric complexes coexist , with the proportion of hexamer increasing with pH . Channel current measurement and 2D crystallography support the existence and transition of the two oligomeric states in membranes . The hexameric complex consists of six ExbB subunits and three ExbD transmembrane helices enclosed within the central channel . We propose models for activation/inactivation associated with hexamer and pentamer formation and utilization of proton motive force . Many essential biological processes are coupled to ion potentials across the lipid membrane . These include , amongst others , ATP synthesis ( Junge et al . , 2009 ) , uptake of nutrients ( Faraldo-Gómez and Sansom , 2003; Noinaj et al . , 2010; Krewulak and Vogel , 2011; Abramson et al . , 2003 ) , protein secretion ( Tsukazaki et al . , 2011 ) , multi-drug excretion ( Murakami et al . , 2006 ) , and rotation of the flagellar motor ( Kojima , 2015 ) . The Ton complex ( Faraldo-Gómez and Sansom , 2003; Noinaj et al . , 2010; Krewulak and Vogel , 2011 ) in gram-negative bacteria is an interesting example of a molecular machine that depends on proton motive force ( PMF ) . While other systems use this transmembrane ( TM ) energy at source ( that is at the membrane with the proton gradient ) the Ton system conveys it to the distant outer membrane , ~20 nm away from the energy source sited at the inner membrane ( Du et al . , 2014; Matias et al . , 2003 ) ( Figure 1—figure supplement 1 ) . The Ton complex is composed of three proteins ExbB , ExbD and TonB located in the inner membrane . It utilizes PMF to take up diverse compounds such as iron-loaded siderophores , haem , and vitamin B12 through outer membrane receptors ( Faraldo-Gómez and Sansom , 2003; Noinaj et al . , 2010; Krewulak and Vogel , 2011; Schauer et al . , 2008 ) . Toxins such as colicin and some antibiotics , as well as phages , hijack the Ton system to gain entry to these cells ( Noinaj et al . , 2010; Krewulak and Vogel , 2011; Schauer et al . , 2008; Lloubès et al . , 2012 ) . It could perhaps be used , through drug targeting , to attack pathogenic gram-negative bacteria ( Neelapu et al . , 2015 ) . In response to proton flux through the Ton complex , the C-terminal domain of TonB interacts with TonB-dependent outer membrane receptors , which bind specific substrates . A conformational change releases the bound substrate from the receptor into the periplasmic space . Each substrate is then transported into the cell by its specific inner membrane transporters and related proteins ( Figure 1—figure supplement 1 ) . In all cases , the Ton complex initiates the process and works as the energizer for outer membrane substrate uptake . ExbB has three TM helices with a large cytoplasmic domain ( Celia et al . , 2016 ) , and both ExbD and TonB are predicted to have single TM helices and a compact periplasmic domain ( Garcia-Herrero et al . , 2007; Ködding et al . , 2005 ) . ExbB and ExbD together form a proton channel , and isolated ExbBD complexes show ion conductivity and cation selectivity in vitro ( Celia et al . , 2016 ) . The ExbBD complex can be thought of as the engine and TonB a drive shaft connecting the engine to the gate in the outer membrane . Other channel complexes such as MotAB/PomAB ( Kojima , 2015 ) , TolQR ( Lloubès et al . , 2012; Godlewska et al . , 2009 ) , and AglRQS ( Agrebi et al . , 2015 ) exhibit sequence homology to ExbBD and also utilize ion motive force , but have different physiological functions . MotAB and PomAB generate torque for flagellar rotation , TolQR maintains outer membrane integrity , and AglRQS energizes surface gliding and sporulation in slime bacteria . Although structural information on this family is limited , recently crystal structures of the ExbB pentamer with and without a short segment of ExbD including the TM region have been solved , giving valuable insights into the architecture of the ExbB complex . However , a pentameric functional unit seems not to fit with previous studies ( Higgs et al . , 2002; Jana et al . , 2011; Pramanik et al . , 2011; Sverzhinsky et al . , 2014; Sverzhinsky et al . , 2015 ) . Here , we report the structures of complexes of hexameric ExbB and ExbD revealed by X-ray crystallography and single particle cryo-EM . Image analyses show hexameric and pentameric complexes coexisting in detergent micelles and within lipid bilayers . Formation of the hexamer is promoted at higher pH , in line with an increased macroscopic channel conductance . The architecture of the ExbB hexamer and ExbD trimer complex suggests a model for utilization of PMF . We expressed and purified Escherichia coli full-length ExbB-ExbD complexes . The structure of crystals grown at pH 9 . 0 was solved to 2 . 8 Å by molecular replacement . A starting hexamer model was constructed by docking the coordinates of the ExbB monomer cut out from the pentamer ( Celia et al . , 2016 ) into a low-resolution cryo-EM map of the hexameric complex ( See Materials and methods ) . The asymmetric unit in the lattice with space group P21 ( Supplementary file 1-Table S1 ) contains two ExbB hexamers , one disposed upside down with respect to the other , and interacting laterally with each other ( Figure 1—figure supplement 2 ) . The overall hexamer structure resembles a bell with a vertical dimension of ~110 Å and a maximal horizontal dimension on the cytoplasmic side of ~85 Å ( Figure 1 ) . Throughout this paper ‘top’ and ‘bottom’ correspond to the periplasmic and cytoplasmic ends , respectively . We also define the periplasmic side ‘up’ and the cytoplasmic side ‘down’ and describe the structure accordingly . The proportion of ExbD tended to decrease upon crystallization , as deduced from SDS-PAGE patterns of crystal samples ( Figure 1—figure supplement 3 ) . Yet , peptide mass finger printing analysis showed that some ExbD is retained in the crystals . A Fourier difference map revealed untraceable masses , probably parts of ExbD TM helices , inside the channel of the hexamer . Low occupancy and possibly flexibility may account for the lack of clarity . The ExbB monomers in the hexamer show similar structures to those in the pentamer ( the average RMSD of Cα atoms between two monomers is 1 . 4 Å ) , despite the different subunit number ( Figure 2A ) . The structure has been described previously ( Celia et al . , 2016 ) , and , briefly , consists of three long α-helices ( 80 ~ 100 Å ) , one of which is broken at its middle in the cytoplasm , extending across the cytoplasmic and periplasmic spaces , and connected with short α-helices and loops; in total 7 α-helices ( α1 - α7 from the N-terminus ) are in each monomer . The TM region is well conserved over species ( Figure 2A ) . The hexamer and pentamer form similar funnel structures with large cytoplasmic and smaller periplasmic domains connected through the TM region ( Figures 1 and Figure 2—figure supplement 1 ) . The overall volume is ~60 , 400 Å3 in the hexamer with an outermost diameter of ~85 Å on the cytoplasmic side and ~43 , 900 Å3 and ~75 Å respectively in the pentamer . The channel at center is surrounded by TM helices α6 and α7 in the membrane region and α5 and α7 further down ( Figures 1 , 2A and Figure 2—figure supplement 1 ) . The channel in the hexamer is significantly larger than that in the pentamer ( Figure 1D ) . The diameter in the hexamer is ~17 Å and that in the pentamer ~10 Å through the TM region . The pentamer structure of the crystal grown at pH 4 . 5 resolved one rod-like density located in the central channel towards the periplasm , and a part of the rod was assigned to a TM segment of ExbD ( Celia et al . , 2016 ) . The channel of the pentamer can accommodate only one helix , whereas that of the hexamer has multiple helices . This is confirmed in cryo-EM maps seen below . The larger central channel in the hexamer would fit better with a multimerizing nature of ExbD ( Higgs et al . , 2002; Ollis et al . , 2009; Gresock et al . , 2015 ) . In both complexes , the large cytoplasmic domain has a cluster of positively charged residues surrounding the channel close to the membrane surface and a group of negatively-charged residues at the end on the cytoplasmic side ( Figure 2—figure supplement 1C ) . Neighboring monomers of ExbB in the hexamer and in the pentamer bury approximately the same surface areas ( 3000 Å2 ) and the residues contributing to interactions are also similar . It is not obvious why hexamers crystallized at pH 9 . 0 , but the negatively charged amino acids in the acidic cluster at the cytoplasmic end are further apart in the hexamer , and the channel radius is particularly large here ( Figures 1D , 2B , Figure 2—figure supplement 1A and B ) . Lower pH increases chances for hydration of ionised acidic residues , which may allow the denser packing here of the pentamer . This region is important for function , as deletion of residues 100–109 , which include four acidic residues ( Asp 102 , Asp 103 , Glu 105 , and Glu 109 ) decreases iron transport activity ( Bulathsinghala et al . , 2013 ) . It should be noted that in the previous structure study , well-diffracting crystals of pentamers at pH 7 . 0 or lower were obtained following treatment of ExbB with a methyl-reducing reagent to modify the amide in lysine side chains ( Celia et al . , 2016 ) . Indeed , the modified lysine near Glu 109 and a calcium ion bound to all five Glu 105 side chains appeared to be needed for pentamer formation in the crystals ( Figure 2d of Celia et al . , 2016 ) . Our crystals do not need such treatment . We measured the channel conductance of ExbBD complexes reconstituted into planar lipid bilayers . The behaviour of the channel is similar to that reported in Celia et al . , 2016: The conductance shows high pH dependence , poor at low pH and greater at neutral pH or above ( Figure 3; Celia et al . , 2016 ) . We then examined frozen-hydrated specimens of ExbBD complexes ( Figure 4—figure supplement 1A and B ) purified at various pHs with cryo-EM . Reference-free 2D class average of molecular images produced characteristic projections ( Figures 4A , B , Figure 4—figure supplement 1B and E ) . They include hexameric and pentameric structures corresponding to projections normal to the membrane plane . The hexameric and pentameric images resolve axial projections of the α-helices and well resemble the crystal structures viewed from the same direction ( Figure 4A and B ) . Complexes of ExbB pentamer and hexamer coexist in the sample solution and the proportion of each depends on pH , being 3:1 at pH 5 . 4 , 1:3 at pH 8 . 0 , and almost entirely hexamers at pH 9 . 0 ( Figure 4C and Supplementary file 1-Table S2 ) . The number of ExbB and ExbD subunits in the functional unit has been argued for years ( Higgs et al . , 2002; Jana et al . , 2011; Pramanik et al . , 2011; Sverzhinsky et al . , 2014; Sverzhinsky et al . , 2015 ) . Estimation of the subunit number has been elusive likely due to this pH dependent pentamer/hexamer equilibrium . A previous MS analysis showed that the main fraction of the sample in the weak basic condition used corresponded to ExbB hexamers and small amounts of the pentamer were also found ( Pramanik et al . , 2011 ) . This proportion is consistent with our results ( Figure 4 ) . Hexamers and pentamers also coexist within lipid bilayers . We prepared 2D crystals at pH 5 . 4 by reconstituting purified ExbBD complexes into liposomes . Fourier transform of the crystals showed pseudo-hexagonal lattice patterns as in 3D crystals in P1 lattices of another crystal form found by X-ray diffraction experiments ( See Materials and methods ) . Fourier filtering of different patches from negatively-stained 2D crystals revealed pentagonal and hexagonal images ( Figure 5 ) . The coexistence of pentamers and hexamers probably prevents growth of well-diffracting crystals at pH 5 . 4 ( see Materials and methods ) . The 2D crystals contained both full-length ExbB and ExbD ( Figure 5E ) . In contrast , loss of ExbD occurred in 3D crystals ( Figure 1—figure supplement 3C ) . The 3D crystals appeared to be built by stacking of 2D crystal layers ( Figure 1—figure supplement 2A and B ) . The periplasmic part of ExbD , which includes a long loop and a compact peptidoglycan-binding motif , would be difficult to fit between the 2D crystal layers in both P21 and P1 3D crystal lattices without deforming these domains ( Figure 1—figure supplement 2A and B ) . Thus , stacking of 2D crystal layers might exclude ExbD because of these steric effects . The measurement of the single channel conductance showed that there are two major conductance states at both pH 7 . 5 and at pH 4 . 5 , exhibiting conductances of 140–150 pS and ~60 pS ( Figure 3A and B ) . The frequency of the high conductance state was about 2 . 6 times that of the low at pH 7 . 5 ( 47 events versus 18 ) , and was about half that of the low at pH 4 . 5 ( 19 events versus 37 ) . Thus two distinct states co-exist and frequency distributions depend on pH , tying in with the EM data if the hexamer and pentamer represent channels with high and low conductance , respectively . The two states/channels appear interchangeable in the lipid bilayer as macroscopic currents are reversible on adjusting pH ( Figure 3C and D ) . As shown , the electron density map of the crystal of the ExbB hexamer and ExbD complex does not allow modelling of ExbD ( Figure 1—figure supplement 2C and D ) , even though the crystal retains a small amount of ExbD ( Figure 1—figure supplement 3C ) . Purified samples actually contain a significant amount of ExbD ( Figure 1—figure supplement 3B ) . To identify ExbD in the complex , we carried out 3D reconstruction from the cryo-EM images of the samples at pH 8 . 0 by single particle analysis ( Supplementary file 1-Table S2 ) . Starting from a reference map filtered to 30 Å resolution from the atomic model of the ExbB hexamer , a 3D structure was determined at 6 . 7 Å resolution ( Figure 4—figure supplement 1C ) with Bayesian-based algorisms implemented in RELION ( Scheres , 2012 ) . The heterogeneity ( Figure 4B ) as well as the relatively low molecular weight and low-symmetry did not allow reconstruction at higher resolutions and densities corresponding to side chains were not visible . Nonetheless , the map clearly resolves all α-helices and loops in the ExbB hexamer except for small parts of the loops at the periplasmic and at the cytoplasmic ends ( Figures 6 and Figure 6—figure supplement 1A ) . Densities corresponding to bound detergent around the TM region are visible at a lower contour level ( Figure 6—figure supplement 1A ) . The map also reveals three rod-like densities inside the channel , which likely correspond to the TM helices of three ExbD molecules ( Figure 6 ) . The helices fit snugly into the hexamer channel , with the arrangement off-centre and asymmetrical . Indeed , the rod-like densities disappear if 3-fold symmetry is enforced . The local resolutions were estimated to be 4 ~ 5 . 5 Å for most of the α-helices both in the ExbB hexamer and the central channel ( Figure 4—figure supplement 1D ) . Several lines of evidence show that ExbD has a propensity for dimerization ( Celia et al . , 2016; Higgs et al . , 2002; Ollis et al . , 2009 ) , but can be monomeric as in an NMR experiment at low pH ( Garcia-Herrero et al . , 2007 ) . Also , ExbD forms a heterodimer with TonB ( Ollis et al . , 2009; Gresock et al . , 2015 ) , further suggesting more than one oligomerization possibility for ExbD . Another interesting feature in the cryo-EM map is that a single mass in the periplasm becomes visible at ~ a half-contour level ( Figure 6—figure supplement 1B ) of that in Figure 6—figure supplement 1A . The mass is ~ 50 Å above the membrane surface . ExbD is predicted to consist of a compact periplasmic domain , one TM helix , and a loop region connecting these two . The compact periplasmic domain in the NMR structure of ExbD consists of several short β-strands and α-helices facing each other , and this arrangement is seen in other peptidoglycan binding proteins ( Garcia-Herrero et al . , 2007; Wojdyla et al . , 2015 ) . We docked this domain of ExbD in the cryo-EM map , and the dimension of the mass in the periplasm is such that it accommodates three domain structures well ( Figure 6—figure supplement 1B ) . We also reconstructed a 3D structure of the pentameric complex at pH 5 . 4 in the same way . The 3D map resolved a pentameric structure with clear three long α-helices per subunit in the cytoplasmic domain but poorer masses in the TM region except for a rod-like density inside the channel . This is probably due to the lower molecular weight of the complex compared with the hexamer and more preferential orientations for top views in cryo-EM images of samples at pH 5 . 4 . Five-fold symmetrization , however , gave a better map . Although the quality of this pentamer map , particularly the continuity of α-helices at the periplasm , is still not quite as good as those of the hexameric complex , a single rod-like density appears again inside the channel in the TM region ( Figure 6—figure supplement 1C ) . The rod probably corresponds to the TM helix of ExbD . There is insufficient space to accommodate more helices in the channel . The findings are consistent with the crystal structure of the pentamer at pH 4 . 5 ( Celia et al . , 2016 ) . The crystal structure of the ExbB hexamer needed slight modifications to be fitted into the cryo-EM map of the hexameric complex ( Figure 6 ) . This was done by rigid-body refinement of the atomic model of the ExbB hexamer and three ExbD TM helices ( ExbDTM ) against the cryo-EM map in real space ( Supplementary file 1-Table S3 ) . In the crystal , each ExbB subunit is settled on a flat plane parallel to the membrane and well follows the six-fold symmetry normal to this plane . In contrast , each subunit in the fitted model ( ExbB6ExbD3TM ) is gradually dislocated by 1 ~ 3 Å from the membrane plane downwards towards the cytoplasmic side in the cryo-EM structure ( Figure 6C ) . The arrangement of the ExbB models in the ExbB pentamer with one ExbDTM ( ExbB5ExbD1TM ) shows little change from the crystal structure ( Figure 6—figure supplement 1C ) . Our results show that pentameric and hexameric complexes coexist in detergent micelles and within lipid bilayers ( Figures 4 and Figure 4—figure supplement 1 ) , and the ratio of the hexamer to the pentamer increases with pH ( Figure 4 and Supplementary file 1-Table S2 ) , similar to the pH dependence of the macroscopic channel conductance ( Figure 3; Celia et al . , 2016 ) . This suggests that the pentamer is less active and the hexamer more active . The crystal structure at pH 4 . 5 ( Celia et al . , 2016 ) and our cryo-EM reconstruction of the pentameric complex ( Figure 6—figure supplement 1C ) show a central single TM helix of ExbD , which makes a ~2 Å pore at the shortest diameter ( Figure 7C ) . The ExbBD complex has been shown to be a cation-selective channel ( Celia et al . , 2016 ) , but this pore size is too small for permeation of cations unless there are large fluctuations of the central helix . Thus , the structure of this pentameric complex suggests a low efficiency , blocked or semi-blocked , energy transducing state . The ExbB hexamer and ExbD trimer complex has a ~5 . 5 Å pore ( Figure 7C ) , which fits with the reported pore size of cation channels . The inner surface of the hexameric subunits of ExbB in the TM region is lined with hydrophobic residues and polar residues such as Thr 148 and Thr 181 ( Figures 2A , Figure 1—figure supplement 2C and D ) . A thorough mutational study has indicated that there are no critical residues in ExbB directly involved in proton transport ( Baker and Postle , 2013 ) . In the case of ExbD , Asp 25 in the lower part of the TM helix is critical for transport mediated by the Ton system ( Braun et al . , 1996 ) , and conserved over this family , and is likely part of the proton conductance pathway . Significantly , the cryo-EM structure shows that each of the subunits of the hexamer are slightly shifted downwards with respect to the prior adjacent molecule , resulting in a downward spiral of the hydrophobic ExbB wall that leads to the Asp 25 residues of the trimer ( Figure 7D ) . These aspartates may form a cation selective filter ( Celia et al . , 2016 ) . In our model , the aspartates reside at the lower part of the TM region and two of them face the pore , producing an electronegative field in the ion path ( Figure 7B ) . The aspartates likely allow only cations to pass , and those cations may be funnelled by the electropositive circle of residues lower down and propelled towards the electronegative toroid at the exit ( Figures 7B and Figure 2—figure supplement 1C ) . The asymmetric disposition of the three TM helices of ExbD within the channel may be important for utilizing PMF . Many biological molecular machines depend on asymmetry . Vacuolar-type ATPase is an example , where a part of the single c’ subunit is located off-centre of the symmetric integral membrane c-ring , resulting in rotation and proton pumping using the energy of ATP hydrolysis ( Mazhab-Jafari et al . , 2016 ) . A similar mechanism may operate in the ExbBD complex , as dynamic motion of TonB , possibly rotation , has been observed by fluorescent optical microscopy ( Jordan et al . , 2013 ) . The proton gradient through the spiral of ExbB subunits could generate a torsional force ( Chang et al . , 2001 ) to induce rotational movement of the ExbD TM helices relative to the ExbB ring ( Figure 8A ) and step-wise participation of two of the three Asp 25 residues as they become exposed in the channel ( Figures 6C and 7B and D ) . The force could be transduced to TonB – it is known that ExbD interacts with TonB through their periplasmic domains ( Ollis et al . , 2009; Gresock et al . , 2015 ) . TonB resides around the ExbB complex , as TonB TM residues interact with the outward-facing side of the first TM helix ( α2 ) of ExbB ( Figures 1 and 6C; Larsen et al . , 1994; Larsen et al . , 1999; Larsen and Postle , 2001 ) . We have tried to obtain stable complexes of TonB-ExbB-ExbD , TonB-ExbB and TonB-ExbD for structural studies , but without success . The difficulties may reflect the peripheral location of TonB , outside of the hexameric complex , and a tendency to dissociate from it . In this model , TonB would have little effect on the complexation of ExbB and ExbD ( Celia et al . , 2016 ) . TonB may move around the outer rim of the hexamer , and also promote lateral movement of the whole TonB complex in the inner membrane through transient contact of the periplasmic C-terminal domain of TonB with the PG layer as proposed by Klebba , 2016 . This movement could facilitate linking up with outer membrane receptors ( Klebba , 2016 ) to induce a conformational change that releases receptor-bound substrate into the periplasmic space ( Chang et al . , 2001; Klebba , 2016 ) . The equilibrium of hexamer and pentamer may depend on the protonation state of acidic residues such as Glu 105 , Glu 109 , Asp 103 , Asp 225 at the cytoplasmic end of the channel ( Figures 2B and 7D ) , in accordance with PMF . If the hydronium concentration rises in this vicinity , protonation of these residues and easing of electrostatic repulsion may allow denser packing among ExbB subunits , which repels one ExbB subunit from the hexamer ring and excludes two ExbD from the central space to form the pentamer ring ( Figure 8B ) . Reversely , a hydronium concentration decrease and deprotonation of the acidic residues would shift to looser packing , which recruits two ExbD to the ring and one ExbB to form the hexameric complex . The channel in the hexameric complex accommodates three but not four TM helices . Also , a lower number of TM helices in the hexamer channel , as well as a vacant pentamer channel , produce larger pores and unproductive and high proton influx . Thus , the channel activity may be controlled by joining/ejecting two ExbD molecules and one ExbB through partial complex disruption . The ExbD dimer is stable ( Celia et al . , 2016; Jana et al . , 2011; Ollis et al . , 2009 ) and excluded ExbB molecules may merge into smaller multimers according to cross linking ( Jana et al . , 2011 ) . The low conductance of the pentamer would not only stop import of compounds , but decreased proton influx could also prevent excessive hydronium accumulation within the cytoplasm of the bacterium ( Figure 8B ) . Oligomeric variation is displayed by the mechano-sensitive channel , MscL ( Walton et al . , 2015 ) and other homologous oligomeric membrane proteins such as F0-ATPases ( Watt et al . , 2010 ) and light harvesting complexes ( Cogdell et al . , 1997 ) . It is , however , not clear whether these various oligomeric states relate to function or are artefacts due to detergent solubilisation , or deletion mutation ( Walton et al . , 2015 ) . Unlike other examples , we observe formation of hexamer and pentamer complexes in detergent micelles and within lipid bilayers , and record two distinct ion conductivities in a pH dependent ratio in planar lipid membranes for the same ExbBD sample . These characteristics could be relevant to the biological nature of the energizer in vivo . Oligomeric transformation may be facilitated by a highly fluid lipid membrane . Indeed , high fluidity allows diffusion of MotB molecules within the cell membrane ( Leake et al . , 2006 ) and expedites large movements of TM helices in Ca2+-ATPase during the reaction cycle ( Norimatsu et al . , 2017 ) . The observations reported here may provide a new vision of dynamic membrane biology and spur further studies to explore the processes involved . The E . coli exbB and exbD genes were subcloned into vector pET-21b ( Novagen ) . A hexahistidine tag was added at the C-terminus of ExbB , but not on ExbD . The plasmid was transformed in E . coli BL21 ( DE3 ) Gold cells ( Novagen ) , and the cells were grown at 37°C in LB medium . To induce ExbB and ExbD , IPTG was added to a final concentration of 0 . 5 mM when the culture reached an optical density at 600 nm ( OD600nm ) of ~1 . 0 . After additional culture for 3 hr at 30°C , the cells were harvested by centrifugation at 4000 g for 30 min at 4°C . The pellet was resuspended in 50 mM Tris-HCl , pH 8 . 0 , 0 . 3 M NaCl , 10% ( v/v ) glycerol , 10 mM imidazole , protease inhibitor cocktail cOmplete ( Roche ) , DNase I ( Roche ) , 5 mM MgCl2 , and 0 . 2 mg/ml lysozyme and disrupted with ~7 passages at ~10 , 000 psi through a high pressure homogenizer EmulsiFlex C-5 ( AVESTIN ) . Cell debris was removed by centrifugation at 10 , 000 g for 30 min at 4°C , and the membrane fraction was collected by ultracentrifugation at 125 , 000 g for 90 min at 4°C . The ExbBD complex was purified by immobilized metal affinity ( IMAC ) and size-exclusion chromatography ( SEC ) . The membrane fraction was first solubilized with 1–2% ( w/v ) n-dodecyl-β-D-maltoside ( DDM; Dojindo ) in 50 mM Tris-HCl , pH 8 . 0 , 0 . 3 M NaCl , 10 mM imidazole , 10% glycerol , and cOmplete . The insoluble membrane fraction was removed by centrifugation at 125 , 000 g for 60 min at 4°C . The supernatant was then loaded onto a column filled with a metal affinity resin COSMOGEL His-Accept ( Nacalai tesque ) at 4°C . The column was washed with the solubilization buffer except for containing 0 . 018% ( w/v; × 2 critical micelle concentration; cmc ) DDM , and developed with a stepwise gradient of imidazole concentration . Fractions eluting at ~0 . 3 M imidazole were collected and concentrated using a spin concentrator with a molecular weight cut-off of 100 , 000 . The concentrate was then applied to a SEC column Superdex 200 10/300 GL ( GE Healthcare ) , and run in 10 mM glycine , pH 9 . 0 , 0 . 2 M NaCl , 10% glycerol , 1 mM TCEP , 0 . 5% ( v/v; × 2 cmc ) tetraethylene glycol monooctyl ether ( C8E4; Anatrace ) or 0 . 07% ( v/v; × 2 cmc ) pentaethylene glycol monodecyl ether ( C10E5; Anatrace ) . The peak was collected and the purity was checked by SDS-PAGE . The protein concentration was measured by the BCA assay ( Pierce ) . We also prepared a construct of the ExbBD complex with a TEV protease recognition site ( Glu-Asn-Leu-Tyr-Phe-Gln-Gly-Ser ) inserted between the C-terminus of ExbB and a hexahistidine tag . The protein complex was induced and purified in the same way except for cleavage of the histidine tag by 3 hr incubation with TEV protease at r . t . Then , the sample , passing through a desalting column to remove imidazole , was mixed with the metal affinity resin and stirred for 1 hr at 4°C . The unbound fraction was run on SEC with the buffer . Samples were concentrated to ~10 mg/ml and subjected to extensive screening over sparse matrix conditions with a Mosquito crystallization robot ( TTP Labtech ) . Crystals were grown by hanging-drop vapor diffusion at 20°C in a mother liquor containing 0 . 1 M glycine , pH 9 . 0 , 0 . 15 M CaCl2 , ~40% PEG 350 MME and 0 . 05–0 . 2 M L-arginine . Plate-like crystals of approximately 100 μm × 100 μm × 10 μm grew over 1–2 months . Hexagonal crystals grew in mother liquors at pH 5 . 4 . To identify the crystal content , crystals were collected with a nylon loop and washed with the crystallization buffer . The crystals were dissolved in SDS sample buffer , heated at ~95°C for 5 min , and then run on a 15% homogeneous polyacrylamide gel . Bands cut out from the gel stained with Coomassie brilliant blue were digested with trypsin . The digestion mixtures were subjected to MALDI-LIFT TOF/TOF MS on an Ultraflextreme mass spectrometer ( Bruker Daltonics ) and/or LC MS/MS on a Q Exactive ( Thermo Fisher Scientific ) followed by peptide mass finger printing analysis . The obtained MS spectra were used to search the SwissProt database using the Mascot program ( Perkins et al . , 1999 ) . ExbB and ExbD were identified with remarkably high identity scores . No other candidates were flagged in the database . Crystals were flash-frozen in liquid nitrogen and diffraction data were collected on the synchrotron radiation beam at BL26 , BL32XU and BL41XU of SPring-8 at a wavelength of 1 Å . The datasets from crystals grown at pH ~9 were processed with XDS ( Kabsch , 2010 ) . The crystals belonged to space group P1 or P21 . Crystals grown at pH 5 . 4 yielded anisotropic patterns with high diffuse scattering backgrounds , where diffraction spots were limited to 7–8 Å resolution from their hexagonal crystal plane . The datasets from these crystals could not be processed . To phase the diffraction data , we used a cryo-EM map and the atomic coordinates of the ExbB pentamer ( PDB accession code: 5SV0; Celia et al . , 2016 ) . The atomic model of the ExbB monomer cut-out from the pentamer was fitted into a low-resolution cryo-EM map of the hexamer ( see below ) using UCSF Chimera ( Pettersen et al . , 2004 ) . Starting from the constructed hexamer model , molecular replacement was carried out using PHASER ( McCoy et al . , 2007 ) . The solutions were unique and of high scores , achieved for the diffraction data from both crystals in P1 and in P21 lattices ( Supplementary file 1-Table S1 ) . The asymmetric unit contains two ExbB hexamers disposed upside down with respect to each other . The two hexamers interacted mostly through the cytoplasmic end in the P1 lattices along the c-axis , but laterally through long α-helices over the TM and cytoplasmic domains in the P21 lattice ( Figure 1—figure supplement 2A and B ) . The latter probably made the crystal packing tight . Since crystals in the P1 lattices showed incomplete data statistics ( Supplementary file 1-Table S1 ) and poorer isomorphism , structure refinement was carried out only against the data from the crystals in the P21 lattice . The structure of the ExbB hexamer was refined using BUSTER ( Bricogne et al . , 2016 ) , and manually adjusted with COOT ( Emsley et al . , 2010 ) . The electron density map resolved amino acids 10–234 ( subunit A ) , 20–234 ( subunit B ) , 10–232 ( subunit C ) , 19–233 ( subunit D ) , 11–232 ( subunit E ) and 19–234 residues ( subunit F ) for one hexamer , and 10–234 ( subunit G ) , 19–234 ( subunit H ) , 10–232 ( subunit I ) , 20–233 ( subunit J ) , 11–234 ( subunit K ) and 20–234 residues ( subunit L ) for the other . In σA-weighted 2 \|Fobs\| - \|Fcalc\| and \|Fobs\| - \|Fcalc\| maps , significant densities appeared in the central channel , but assignment of any model for ExbD was not possible . Data collection and refinement statistics are shown in Supplementary file 1-Table S1 . The channels were analysed by HOLE ( Smart et al . , 1996 ) and buried surface areas and volumes were calculated by PDBePISA ( Krissinel and Henrick , 2007 ) . To achieve the optimal contrast of molecular images , we used lauryl maltose neopentyl glycol ( LMNG; Anatrace ) , which is a detergent with a low cmc , and glycerol was not included in the buffer for all experiments . Samples were solubilized with 1% ( w/v ) LMNG and purified in 0 . 002% ( v/v; × 2 cmc ) LMNG by IMAC and SEC as described . To examine samples at different pH values , SEC was carried out in buffers containing either 10 mM Na-citrate , pH 5 . 4 , 10 mM HEPES , pH 7 . 0 , 10 mM Tris-HCl , pH 8 . 0 , or 10 mM glycine , pH 9 . 0 . The SEC patterns showed a small peak consisting of ExbB and ExbD at a lower molecular weight than those in the main fraction . The corresponding peak became smaller when run at pH 8 . 0 and disappeared at pH 9 . 0 , which is consistent with an equilibrium of pentamer and hexamer ( Figure 4C and Supplementary file 1-Table S2 ) . For frozen-hydrated sample preparation , gold was manually sputtered on holey carbon film-coated copper grids ( Quantifoil R1 . 2/1 . 3 , Quantifoil Micro Tools GmbH ) . This reduced charging of samples in ice and thereby minimized beam-induced movement ( Russo and Passmore , 2014 ) , when the samples were exposed to the electron beam . Three μl of ExbBD samples at a protein concentration of 0 . 25 - 0 . 5 mg/ml were applied onto the grid , deposited manually with filter paper from the reverse side of the grid , and rapidly frozen in liquid ethane with a home-made plunger in a cold chamber with a humidifier . The sample grids were screened with a JEOL-2100 electron microscope ( JEOL ) equipped with a LaB6 gun operated at an accelerating voltage of 200 kV . The grids frozen in seemingly good conditions were first examined on Falcon II detectors ( FEI ) in Titan Krios electron microscopes ( FEI ) at the Research Center for UHVEM , Osaka University and NeCEN , Leiden University . We found that the modulation transfer function ( MTF ) of the Falcon II ( FEI ) was not sufficiently high enough to obtain high-resolution information from molecular images in this sample size . Hence , all the data used for image analysis were collected on the K2 Summit direct electron detector ( GATAN ) as below . Images of samples at pH 8 . 0 and pH 5 . 4 were collected with a Tecnai Polara scope ( FEI ) operated at an accelerating voltage of 300 kV at UCSF and a Titan Krios operated at 300 kV in eBIC , Diamond Light Source , respectively . Dose-fractionated images were recorded on a K2 summit in super-resolution counting mode . At eBIC , inelastically scattered electrons were removed through a GATAN Quantum energy filter with an energy slit width of 20 eV . The dose rate was six electrons per pixel per s , the total exposure time was 20 s , and one physical pixel corresponded to 1 . 22 Å at UCSF . The dose rate was 7 . 2 electrons per pixel per s , the total exposure time was 6 s , and one physical pixel corresponded to 1 . 06 Å at eBIC . Each frame accumulated electrons for 0 . 2 s . Defocus values ranged from 0 . 6 to 3 . 1 μm both at UCSF and at eBIC . Collected images were used for 3D reconstruction ( Supplementary file 1-Table S2 ) . For only 2D classification analysis in this study , images of samples at pH 7 . 0 and 9 . 0 were collected on a K2 summit with a Tecnai Arctica ( FEI ) operated at 200 kV . The dose rate was 7 . 5 electrons per pixel per s , the total exposure time was 6 . 4 s with an accumulation time of 0 . 2 s for each frame , and one physical pixel corresponded to 0 . 99 Å . Defocus values ranged from 0 . 5 to 4 . 9 μm ( Supplementary file 1-Table S2 ) . We also collected images of the complex at pH 8 . 0 with the histidine tag cleaved off . Images were collected with a Titan Krios operated at 300 kV in eBIC . The dose rate was ~ 5 electrons per pixel per s , the total exposure time was 10 s with an accumulation time of 0 . 2 s for each frame , and one physical pixel corresponded to 1 . 06 Å . Defocus values ranged from 0 . 6 to 2 . 9 μm ( Supplementary file 1-Table S2 ) . Image stacks were × 2 binned in Fourier space , drift-corrected , dose-weighted , and summed with MotionCor2 ( Zheng et al . , 2017 ) . Contrast transfer function ( CTF ) parameters were estimated with CTFFIND ( Rohou and Grigorieff , 2015 ) and a few thousand particles were manually picked up with EMAN2 ( Tang et al . , 2007 ) for initial templates . Automatic particle picking and reference-free 2D classification were carried out with RELION ( Scheres , 2012 ) . This yielded clear hexameric and pentameric 2D class averages . An initial 3D structure of the hexamer was reconstructed from well-defined class averages at pH 8 . 0 through probabilistic orientation assignment by PRIME ( Elmlund et al . , 2013 ) . GPU-accelerated RELION ( Kimanius et al . , 2016 ) was used for all the following steps unless noted otherwise . Particles belonging to the selected classes were refined against the initial structure by 3D auto refinement . This gave a ~ 8 . 5 Å-resolution map based on the gold standard Fourier shell correlation ( FSC ) criteria ( Chen et al . , 2013 ) , where the FSC between two volumes , each independently generated from half the data set , drops to 0 . 143 . The map resolved some α-helices running through the cytoplasmic , TM and periplasmic domains . The map was used to construct the initial 3D atomic model of the hexamer for solving the crystal structure as described . For final reconstructions , initial 3D references were generated from the crystal structures of the pentamer ( accession code: 5SV0; Celia et al . , 2016 ) and the hexamer in this study using EMAN2 ( Tang et al . , 2007 ) , and then low-pass filtered to 30 Å for the hexamer and 20 Å for the pentamer . Particles at pH 8 . 0 and 5 . 4 were subjected to 3D classification and 3D auto refinement . No symmetry was applied during these procedures . The numbers of particles for final 3D reconstruction of the hexameric complex was 38 , 323 from a total of 276 , 526 picked-up particles for datasets at pH 8 . 0 . Reconstructions from dataset at pH 5 . 4 resolved a pentameric structure with clear α-helices in the cytoplasmic domain but poorer masses in the TM region except for a rod-like density inside the channel . To improve the structure , 5-fold symmetry was enforced for 3D classification and 3D auto refinement . The numbers of particles for final 3D reconstruction of the pentameric complex was 22 , 243 from 122 , 908 picked-up particles . A soft mask was estimated and applied to the two half-maps in the post-processing process of RELION ( Scheres , 2012 ) . B-factor estimation and map sharpening were also carried out in the post-processing step . The resolutions based on the gold standard FSC 0 . 143 criteria ( Chen et al . , 2013 ) were measured to be 6 . 69 Å for the hexamer at pH 8 . 0 and 7 . 11 Å for the pentamer at pH 5 . 4 . Local resolution was estimated using ResMap ( Kucukelbir et al . , 2014 ) and Blocres in the Bsoft package ( Heymann and Belnap , 2007 ) from unfiltered half maps . Both programs gave similar estimates , but the latter values appeared to be of slightly lower resolution . Details related to cryo-EM SPA are summarized in Supplementary file 1-Table S2 . The ExbD TM helix ( ExbDTM ) was determined by hydrophobicity calculation along the sequence . Atomic models of the ExbB hexamer and ExbDTM trimer ( ExbB6ExbD3TM ) and the ExbB pentamer and ExbDTM monomer ( ExbB5ExbD1TM ) were constructed based on the cryo-EM maps of the hexameric and pentameric complexes , respectively , by using COOT ( Emsley et al . , 2010 ) . Phe 23 and Pro22 in ExbDTM were adjusted to the TM level of aromatic residues such as Tyr 195 , Tyr 132 , Phe 198 , Phe 41 , Phe 42 , and Trp 38 in ExbB . The detergent belt encloses these residues near the cytoplasmic surface ( Figure 6—figure supplement 1A ) . Then , the models were fitted into the maps with rigid-body refinement of individual ExbB subunits and ExbDTM in real space using Phenix . refine ( Adams et al . , 2010 ) . This gave good geometries and high cross correlation ( CC ) values against the cryo-EM maps for both the models ( Supplementary file 1-Table S3 ) . Structure figures in Figures 1 , 2 , 7A , B and D , Figure 1—figure supplement 2 and Figure 2—figure supplement 1 were prepared with PyMol ( The PyMOL Molecular Graphics System , Schrödinger , LLC ) , and those in Figure 6A–C , Figure 4—figure supplement 1D and Figure 6—figure supplement 1 were prepared with UCSF Chimera ( Pettersen et al . , 2004 ) . Electrostatic distributions were calculated using PDB2PQR and displayed using APBS ( Unni et al . , 2011 ) . The lipid mixture of DOPG , DOPC and DOPE at a molar ratio 2:3:5 ( Celia et al . , 2016 ) was spread over a 200 μm diameter hole in a 0 . 5 mm thick Teflon sheet partition . A planar lipid bilayer formed at the hole separated two chambers filled with 2 mM KPi at pH 7 . 5 . Purified ExbBD complexes were mixed with liposomes of the same lipid mixture at a lipid-protein weight ratio of 20:1 or 10:1 . After 2 hr incubation at room temperature , the detergent was removed with a Bio-Spin six column ( Bio-Rad ) equilibrated with 10 mM KPi , pH 7 . 5 , 50 mM KCl , and 0 . 3 M sucrose . Proteoliposomes were then injected from the cis chamber and fused into the bilayers . The reference electrode was placed in the cis chamber . Current data were recorded and stored in a PC using pCLAMP software ( Molecular Devices , Sunnyvale , CA ) through Axopatch 200B amplifier and Digidata 1550A digitizer ( Molecular Devices ) . The low pass filter was set to 1 kHz for the cut-off frequency , and data were sampled at 10 kHz . To change pH , 0 . 5 M succinic acid , pH 4 . 0 or 0 . 5 M Tris-HCl , pH 8 . 95 was added to both chambers . ExbBD complexes purified with C8E4 were mixed with E . coli total lipid extract ( Avanti Polar Lipids ) . The samples were dialyzed over a semipermeable membrane with a molecular weight cut-off of 10 , 000 to remove the detergent in 150 mM NaCl , 1 mM TCEP , 0 . 01% NaN3 , and 25 mM Na-citrate , pH 5 . 4 or 25 mM HEPES , pH 7 . 0 or 25 mM glycine , pH 9 . 0 at 4°C . The samples were negatively-stained with 2% ( w/v ) uranyl acetate and examined with the JEOL-2100 electron microscope ( JEOL ) . Many 2D crystals were formed at pH 5 . 4 , but few at pH 7 . 0 and none at pH 9 . 0 . Crystal images showing diffraction spots were selected and analysed by Fourier filtering of diffraction spots with a modified version of the MRC package ( Crowther et al . , 1996 ) . Crystals were centrifuged and the pellet was washed with the buffer . This process was repeated twice . Then , the sample was dissolved in SDS sample buffer , heated at ~ 95°C for 5 min , and run on a polyacrylamide gel .	Many biological processes that are essential for life are powered by the flow of ions across the membranes of cells . Similar to how energy is stored in the water behind a dam , energy is also stored when the concentration of ions on one side of a biological membrane is higher than it is on the other . When these ions then flow down this concentration gradient , the energy can be harnessed to power other processes . In many bacteria , the concentration of hydrogen ions , or protons , is higher on the outside of the cell . When the protons flow down the concentration gradient , a protein complex called the Ton system in the bacteria’s inner membrane harnesses the energy to transport various compounds , including essential nutrients , across the outer membrane , which is about 20 nanometres away . Toxins , and viruses that infect bacteria , can also hijack the Ton system to gain entry into these cells . This means that the Ton system could perhaps be targeted via drugs to treat bacterial infections . Though the Ton system is important , structural information on this protein family is limited . The Ton complex is composed of three proteins – ExbB , ExbD and TonB – located in the bacteria’s inner membrane . ExbB and ExbD together form a channel for the protons and the complex made from these two proteins can be thought of as the system’s engine . Maki-Yonekura et al . wanted to understand how the ExbB / ExbD complex works , which was challenging because the complex was not well suited to any single structural biology technique . To get around this issue , a combination of two techniques called X-ray crystallography and single particle cryo-EM were used . This approached revealed that the two proteins form complexes made up of either five or six ExbB subunits with one or three ExbD subunits , respectively . It also showed that the proteins transition between the two forms in a cell’s membrane . More of the larger six-unit complex ( also called a “hexamer” ) formed at higher pH . This is consistent with the increased flow of protons through the channel when the local conditions inside the cell become less acidic . Based on these results , Maki-Yonekura et al . propose that some subunits in the core of the complex rotate to harness the energy from the flow of protons , and the number of subunits in the complex changes when it switches to become active or inactive . The discoveries may provide a new vision of dynamic membrane biology . Further studies are now needed to see how general this mechanism is in biology , and the new structural information could also be used to help develop more anti-bacterial drugs .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "structural", "biology", "and", "molecular", "biophysics" ]	2018	Hexameric and pentameric complexes of the ExbBD energizer in the Ton system
Long lasting pyrethroid treated bednets are the most important tool for preventing malaria . Pyrethroid resistant Anopheline mosquitoes are now ubiquitous in Africa , though the public health impact remains unclear , impeding the deployment of more expensive nets . Meta-analyses of bioassay studies and experimental hut trials are used to characterise how pyrethroid resistance changes the efficacy of standard bednets , and those containing the synergist piperonyl butoxide ( PBO ) , and assess its impact on malaria control . New bednets provide substantial personal protection until high levels of resistance , though protection may wane faster against more resistant mosquito populations as nets age . Transmission dynamics models indicate that even low levels of resistance would increase the incidence of malaria due to reduced mosquito mortality and lower overall community protection over the life-time of the net . Switching to PBO bednets could avert up to 0 . 5 clinical cases per person per year in some resistance scenarios . It is estimated that 68% of the 663 million cases of malaria that have been prevented since the year 2000 have been through the use of long-lasting insecticide treated bednets ( LLINs ) ( Bhatt et al . , 2015 ) . However , there is a growing realisation that insecticide resistance is putting these advances under threat ( WHO , 2012 ) , with mosquitoes reporting widespread resistance to pyrethroids , the only class of insecticides currently approved for use in bednets ( Ranson and Lissenden , 2016 ) . The public health impact of pyrethroid resistance in areas of LLIN use is hard to quantify as a comparison between sites is complicated by multiple epidemiological factors making it difficult to ascribe differences in malaria metrics solely to mosquito susceptibility ( Kleinschmidt et al . , 2015 ) . The efficacy of LLINs against mosquitoes is typically measured in experimental hut trials ( WHO , 2013a ) . These experiments are time consuming , relatively expensive , and geographically limited and by themselves they do not fully account for all effects of the LLIN as they do not show the community impact ( herd effects ) caused by the insecticide killing mosquitoes ( Killeen et al . , 2007; Magesa et al . , 1991 ) . Mathematical models can be used to translate entomological endpoint trial data into predictions of public health impact . Currently this has only been done for a small number of sites ( Briët et al . , 2013 ) making it difficult for malaria control programmes to understand the problems caused by insecticide resistance in their epidemiological setting . There are no easy to use genetic markers that can reliably predict the susceptibility of mosquitoes to pyrethroid insecticide ( Weetman and Donnelly , 2015 ) . The current most practical phenotypic method for assessing resistance is the use of bioassays which take wild mosquitoes and measures their mortality after exposure to a fixed dose of insecticide ( WHO , 2013a ) . However the discriminating doses used in the assay are unrelated to the field exposure and so the predictive value of these bioassays for assessing the problems of pyrethroid resistance is unknown . A meta-analysis has shown that insecticide treated bednets still outperform untreated nets in experimental hut trials even against pyrethroid resistant populations ( Strode et al . , 2014 ) though the community impact ( herd effects ) of the LLIN was not assessed ( Killeen et al . , 2007 ) . The population prevalence of pyrethroid resistance is known to be changing at a fast rate ( Toé et al . , 2014 ) making it important to regularly re-evaluate the efficacy of LLINs in order to guide current vector control and resistance management strategies ( WHO , 2012 ) . There are limited tools available for tackling pyrethroid resistance and protecting the advances made in malaria control . Until new LLINs containing alternative insecticide are available the only alternative bednet are those containing pyrethroids plus the insecticide synergist piperonyl butoxide ( PBO ) . Studies have shown that PBO LLINs are substantially better at killing insecticide resistant mosquitoes in some locations but not others ( Ngufor et al . , 2014a , 2014b; Kitau et al . , 2014; Asale et al . , 2014; Ngufor et al . , 2014c; Koudou et al . , 2011; Corbel et al . , 2010; Tungu et al . , 2010; Malima et al . , 2008; Adeogun et al . , 2012a; Agossa et al . , 2014; Malima et al . , 2013 ) . PBO LLINs are more expensive than standard LLINs , with one manufacturer’s 2012 price for PBO LLIN being US$4 . 90 compared to a comparable standard LLIN price of US$3 . 25 ( Briët et al . , 2013 ) . This makes it unclear where and when their use would be beneficial over standard LLINs given constrained public health budgets . A mathematical modelling study used results from 6 experimental hut trials comparing a standard LLIN ( PermaNet 2 . 0 ) with a PBO LLIN ( PermaNet 3 . 0 ) against Anopheles gambiae sensu lato mosquitoes ( Briët et al . , 2013 ) . It predicted that the more expensive PBO LLIN was still cost effective compared to a threshold of US$150/DALY averted ( not comparing against standard LLINs ) in 4 of the 6 sites , though these results are not generalisable beyond the specific sites chosen by the manufacturer , population prevalence of resistance , the type of LLIN or mosquito species . The WHO has recognised the increased bio-efficacy of PermaNet 3 . 0 in some areas ( WHO , 2015 ) but there is a lack of clear consensus on when and where these should be deployed . Defining the added public health benefit expected by a switch to PBO LLINs is essential to guide decisions on pricing , purchasing and deployment . Here we propose that information on the current malaria endemicity , mosquito species and population prevalence of pyrethroid resistance ( as measured by bioassay mortality ) can be used to predict the public health impact of pyrethroid resistance and choosing the most appropriate LLIN for the epidemiological setting . Firstly ( 1 ) a meta-analysis and statistical model are used to determine whether mosquito mortality in a bioassay can be used to predict the proportion of mosquitoes , which die in experimental hut trials and to define the shape of this relationship . Secondly ( 2 ) , another meta-analysis of experimental hut trial data is analysed to characterise the full impact of pyrethroid resistance on LLIN effectiveness . Thirdly , information from ( 1 ) and ( 2 ) is used to parameterise a widely used malaria transmission dynamics mathematical model to estimate the public health impact of pyrethroid resistance in different settings taking into account the community impact of LLINs . An illustration of model predictions showing how different malaria metrics change over time is given in Figure 1 . The figure also indicates how LLIN coverage and variables such as malaria endemicity are incorporated in the model . Finally ( 4 ) this model is combined with bioassay and experimental hut trial results to predict the epidemiological impact of switching from mass distribution of standard to PBO LLIN . 10 . 7554/eLife . 16090 . 003Figure 1 . Scenario under investigation: timings for the introduction of LLINs , insecticide resistance and PBO LLINs for different malaria metrics . The figure illustrates how insecticide resistance is incorporated into the mathematical model . Panel ( A ) shows parasite prevalence by microscopy in 2–10 year olds , ( B ) clinical incidence in the entire population ( cases per 1000 people per year ) and ( C ) the annual entomological inoculation rate ( EIR ) . In all three panels 4 different scenarios are run: black line shows a situation with no insecticide resistance whilst red line illustrates resistance arriving at year 6 ( moderate , 50% survival measured in a bioassay ) ; solid lines show non-PBO LLIN whilst dashed lines show PBO LLINs introduced at year 9 ( vertical dotted-dashed grey line ) . There is no vector control in the population up until time zero ( vertical dashed grey line ) at which time there is a single mass distribution of non-PBO LLINs to 80% of the population . LLINs are redistributed every 3 years to the same proportion of the population . Mosquitoes are entirely susceptible up until resistance arrives overnight at the start of year 6 ( vertical grey dotted line ) . Endemicity ( a variable in Figures 4 and 5 ) is changed by varying the slide prevalence in 2–10 year olds at year 6 ( by changing the vector to host ratio ) and in this plot takes a value of 10% ( as illustrated by the horizontal green dashed line in A ) . The impact of insecticide resistance is predicted ( in Figure 4 ) by averaging the clinical incidence and EIR for the solid red lines ( resistance ) and solid black lines ( no resistance ) between the years 6 and 9 ( period ) . Similarly , the impact of switching to PBO LLINs ( in Figure 5 ) is estimated by averaging the clinical incidence and EIR for the solid red line ( standard LLINs ) and dashed red lines ( switch to PBO LLINs ) lines between the years 9 and 12 ( period ) . Different scenarios with a low and high prevalence of pyrethroid resistance are shown in Figure 1—figure supplements 1 and 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 00310 . 7554/eLife . 16090 . 004Figure 1—figure supplement 1 . Scenario under investigation: example of a mosquito population with a low population prevalence of resistance . The figure illustrates how insecticide resistance is incorporated into the mathematical model . Panel ( A ) shows parasite prevalence by microscopy in 2–10 year olds , ( B ) clinical incidence in the entire population ( cases per 1000 people per year ) and ( C ) the annual entomological inoculation rate ( EIR ) . In all three panels 4 different scenarios are run: black line shows a situation with no insecticide resistance whilst red line illustrates resistance arriving at year 6 ( 20% survival measured in a bioassay ) ; solid lines show non-PBO LLIN whilst dashed lines show PBO LLINs introduced at year 9 ( vertical dotted-dashed grey line ) . There is no vector control in the population up until time zero ( vertical dashed grey line ) at which time there is a single mass distribution of non-PBO LLINs to 80% of the population . LLINs are redistributed every 3 years to the same proportion of the population . Mosquitoes are entirely susceptible up until resistance arrives overnight at the start of year 6 ( vertical grey dotted line ) . Endemicity ( a variable in Figures 4 and 5 ) is changed by varying the slide prevalence in 2–10 year olds at year 6 ( by changing the vector to host ratio ) and in this plot takes a value of 10% ( as illustrated by the horizontal green dashed line in A ) . The impact of insecticide resistance is predicted ( in Figure 4 ) by averaging the clinical incidence and EIR for the solid red lines ( resistance ) and solid black lines ( no resistance ) between the years 6 and 9 ( period ) . Similarly , the impact of switching to PBO LLINs ( in Figure 5 ) is estimated by averaging the clinical incidence and EIR for the solid red line ( standard LLINs ) and dashed red lines ( switch to PBO LLINs ) lines between years 9 and 12 ( period ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 00410 . 7554/eLife . 16090 . 005Figure 1—figure supplement 2 . Scenario under investigation: example of a mosquito population with a high population prevalence of resistance . The figure illustrates how insecticide resistance is incorporated into the mathematical model . Panel ( A ) shows parasite prevalence by microscopy in 2–10 year olds , ( B ) clinical incidence in the entire population ( cases per 1000 people per year ) and ( C ) the annual entomological inoculation rate ( EIR ) . In all three panels 4 different scenarios are run: black line shows a situation with no insecticide resistance whilst red line illustrates resistance arriving at year 6 ( 80% survival measured in a bioassay ) ; solid lines show non-PBO LLIN whilst dashed lines show PBO LLINs introduced at year 9 ( vertical dotted-dashed grey line ) . There is no vector control in the population up until time zero ( vertical dashed grey line ) at which time there is a single mass distribution of non-PBO LLINs to 80% of the population . LLINs are redistributed every 3 years to the same proportion of the population . Mosquitoes are entirely susceptible up until resistance arrives overnight at the start of year 6 ( vertical grey dotted line ) . Endemicity ( a variable in Figures 4 and 5 ) is changed by varying the slide prevalence in 2–10 year olds at year 6 ( by changing the vector to host ratio ) and in this plot takes a value of 10% ( as illustrated by the horizontal green dashed line in A ) . The impact of insecticide resistance is predicted ( in Figure 4 ) by averaging the clinical incidence and EIR for the solid red lines ( resistance ) and solid black lines ( no resistance ) between the years 6 and 9 ( period ) . Similarly , the impact of switching to PBO LLINs ( in Figure 5 ) is estimated by averaging the clinical incidence and EIR for the solid red line ( standard LLINs ) and dashed red lines ( switch to PBO LLINs ) lines between years 9 and 12 ( period ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 005 The population prevalence of pyrethroid resistance is defined from the percentage of mosquitoes surviving a pyrethroid bioassay performed according to standardised methodologies . Data from all bioassay types ( such as the WHO tube susceptibility bioassay ( WHO , 2013b ) , WHO cone bioassay ( WHO , 2013a ) or CDC tube assay [Brogdon , 2010] ) are combined to produce a simple to use generalisable metric . Note that this pyrethroid resistance test does not differentiate between varying levels of resistance within an individual mosquito as only single discriminating doses are used . It is assumed that the ability of a mosquito to survive insecticide exposure is not associated with any other behavioural or physiological change in the mosquito population which influences malaria transmission . For example , an increased propensity for mosquitoes to feed outdoors ( subsequently referred to as behavioural resistance ) would limit their exposure to LLINs though there is currently insufficient field evidence to justify its inclusion in the model ( Briët and Chitnis , 2013; Gatton et al . , 2013 ) . Table 1 summarises the datasets used in the different meta-analyses . Meta-analysis M1 shows that mosquito mortality in experimental hut trials can be predicted by the percentage of mosquitoes surviving a simple pyrethroid bioassay ( Figure 2A ) . There is a substantial association between pyrethroid resistance in a bioassay and mortality measured in a standard LLIN experimental hut trial ( Figure 2A , Deviance Information Criteria , DIC , with resistance as an explanatory variable = 2544 . 0 , without = 2649 . 0 ( lower value shows more parsimonious model ) , best fit parameters α1 = 0 . 634 ( 95% Credible Intervals , 95%CI , 0 . 012–1 . 29 ) and α2 = 3 . 99 [95%CI 3 . 171–5 . 12] ) . This indicates that bioassay survival can be used as a quantitative test to assess how the population prevalence of pyrethroid resistance influences LLIN efficacy . The number of studies identified in M1 is relatively small ( only 21 data-points ) so the predictive ability of the bioassay was further validated using the A . gambiae s . l . PBO data ( Figure 2B , C ) . 10 . 7554/eLife . 16090 . 006Table 1 . Summary of data collated in the three meta-analyses . The number of data points is subdivided according to the insecticides or LLIN tested and the predominant mosquito species in each population tested . Studies which did not determine species in the Anopheles gambiae complex are shown separately . All Published Data can be downloaded from Dryad Digital Repository whilst a list of the studies included their geographical range are given in the Material and methods . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 006Meta-analysis descriptionDetailsNo . StudiesNumber data pointsAnopheles gambiae s . s . Anopheles arabiensisAnopheles gambiae s . l . Anopheles funestusTotalM1Bioassay and experimental hut trial mortalityDeltamethrin52110013Permethrin821306Other100112Total134214121M2Impact of PBO in pyrethroid bioassaysDeltamethrin1615529857Permethrin2022730968Other4204612Total2439126323137M3Experimental hut trials of standard and PBO LLINSOlyset66010016PermaNet61846028Total122441604410 . 7554/eLife . 16090 . 007Figure 2 . The ability of the pyrethroid resistance test ( the percentage mosquito survival in a bioassay ) to predict the results of experimental hut trials and the increase in mosquito mortality caused by the synergist PBO . Panel A: The relationship between mosquito mortality measured in non-PBO WHO tube bioassay and experimental hut trials ( the percentage of mosquitoes , which enter the house that die within the next 24 hr ) . Solid grey line shows the best fit model for all mosquito species combined . Panel B: Differences in mosquito mortality caused by adding PBO to a pyrethroid bioassay . Panel C: Best fit models from Panel A and Panel B were combined to predict the change in mortality seen by adding PBO to a pyrethroid LLIN for mosquito populations with different levels of insecticide resistance . Points show the different mortalities measured from the limited number of experimental hut trials where PBO and non-PBO nets were simultaneously tested . Overall the model appears to be a good predictor of these data , both visually and statistically ( Analysis of Variance test shows there was no significant difference between model predictions and observed data p-value=0 . 25 ) . No experimental hut trial data were available for validation of the Anopheles funestus model . Throughout all panel colour denotes mosquito species , either Anopheles gambiae sensu lato ( red ) or A . funestus ( blue ) , whilst the shape of points indicates the type of pyrethroid used: permethrin ( circle ) , deltamethrin ( square ) , or other pyrethroid ( diamond ) . In panels A and B the fill of the points indicates the type of bioassay used ( filled points = WHO cone; no fill = WHO tube; light fill = CDC bottle ) . Solid line shows the best fit model whilst the shaded areas indicate the 95% credible intervals around the best fit line . In all panels the dashed lines show no difference between the x and y axes . Pre-defined search string used in the meta-analyses are listed in Figure 2—source data 1 whilst raw data from panels A , B and C are provided in Figure 2—source data 2 , Figure 2—source data 3 , and doi:10 . 5061/dryad . 13qj2 respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 00710 . 7554/eLife . 16090 . 008Figure 2—source data 1 . Summary of the different predefined search strings used for meta-analysis M1 , M2 and M3 . Different search engines were used on different dates ( each with their own search numbering ) DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 00810 . 7554/eLife . 16090 . 009Figure 2—source data 2 . Summary of data from meta-analysis M1 presented in Figure 2A . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 00910 . 7554/eLife . 16090 . 010Figure 2—source data 3 . Summary of data from meta-analysis M2 presented in Figure 2B . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 010 The increased mortality observed by adding the synergist PBO to a pyrethroid bioassay was assessed for Anopheles funestus and Anopheles gambiae senu lato mosquitoes with different levels of pyrethroid resistance ( M2 , Figure 2B ) . Data suggests that for the A . gambiae complex PBO has the greatest benefit in mosquito populations with intermediate levels of pyrethroid resistance ( including pyrethroid resistance as an explanatory variable DIC = 2544 . 0 , without DIC = 4748 . 0 ) . In A . funestus adding PBO appears to kill all mosquitoes irrespective of the prevalence of pyrethroid resistance ( including resistance as an explanatory variable improved model fit , with DIC = 2544 . 0 , without DIC = 2547 . 0 , though the gradient of the line was so shallow as to effectively make the PBO synergised pyrethroid mortality independent of the population prevalence of pyrethroid resistance ) . The relationships identified in Figure 2A and B are used to predict the added benefit of a PBO LLIN over a standard LLIN ( Figure 2C ) . These predictions are consistent with the observed results from all published experimental hut trials directly comparing both LLIN types ( M3 ) ( see overlap of data points with model predictions on Figure 2C ) providing further independent evidence that the population prevalence of pyrethroid resistance measured by a bioassay can be used to predict LLIN induced mortality in a hut trial for both standard and PBO LLINs . Mortality in experimental huts was shown to be a useful predictor of LLIN induced deterrence , exiting and the rate of pyrethroid decay ( Figure 3A–C ) . Figure 3A indicates that the number of mosquitoes deterred from entering the experimental hut substantially decreases in areas of higher pyrethroid resistance ( where LLIN induced mortality inside the hut is low ) though the variability around the best fit line is high suggesting the precise shape of the relationship is uncertain . As the population prevalence of pyrethroid resistance increases ( and mortality inside the hut decreases ) an increasing proportion of mosquitoes entering the house exit without blood-feeding ( Figure 3B ) . Only when there is a very high population prevalence of pyrethroid resistance does the probability that a mosquito will successfully feed start to increase ( Figure 3C ) . Changing behaviour of a host seeking mosquito with different levels of pyrethroid resistance is shown in Figure 3D . 10 . 7554/eLife . 16090 . 011Figure 3 . Meta-analysis of how the different outcomes of experimental hut trials which impact LLIN efficacy change with the percentage of mosquitoes which survive after entering the hut . ( A ) The probability that mosquitoes will be deterred away from a hut with an LLIN , ( B ) once entered the hut the mosquito will exit without feeding , or ( C ) will successfully feed . Panel ( D ) shows how the average probability that a bloodfeeding mosquitoes will be killed , deterred from entering , exit without feeding or successfully feed and survive during a single feeding attempt and how this changes with the population prevalence of pyrethroid resistance ( as measured as the percentage survival in a pyrethroid bioassay ) . The lines are drawn using the best fit estimates from ( A–C ) . Panel ( E ) shows how the longevity of the insecticide activity ( estimated from washed nets ) is longer in mosquito populations with high mosquito mortality in experimental hut trials . A possible hypothesis for this change is proposed in ( F ) where the black line indicates how insecticide concentration might decay over time . The time taken for a hypothetical resistant mosquito to survive the insecticide concentration ( pink arrow ) may be shorter than a susceptible mosquito ( purple arrow ) . In Panels ( A ) , ( B ) , ( C ) and ( E ) the points show data from experimental hut trials with standard ( green ) or PBO ( purple ) LLINs . In ( A ) points which fell below the line ( i . e . mosquitoes were more likely to enter huts with LLINs ) were set to zero . The black line shows the best fit model to these data whilst the shaded area denotes the 95% credible interval estimates for the best fit line . Graphical assessment of the validity of the distributional assumptions and the posterior distributions for each parameter are shown in Figure 3—figure supplement 1A ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 01110 . 7554/eLife . 16090 . 012Figure 3—figure supplement 1 . Justification of normality distributed errors in the deterrence dataset ( A ) and posterior distributions of parameter estimates ( B ) . ( A ) shows a normal quantile-quantile plot for the residuals of the data for the relationship between deterrence and mosquito survival in experimental hut trials ( Figure 3A , Equation [9] ) . The linearity of the residuals ( the proximity of the blue dots to the red dotted line ) indicates that the error in these data are adequately described by the normal distribution ( Equation [9] ) . Panel ( B ) shows a kernel density plot for the posterior distributions for all model parameters . Line colours match legend colours ( with values indicating median and 95% credible intervals for all parameters ) . In panel ( B ) all x-axes values are shown on the absolute scale . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 012 The overall efficacy of an LLIN depends on its initial efficacy and the rate at which this changes over the life-time of the net . Since there are currently no published durability studies in areas of high pyrethroid resistance or with PBO LLINs we estimate the loss of insecticidal activity from experimental hut trials using washed nets . Results indicate that washing decreases efficacy fastest in areas of higher pyrethroid resistance . Figure 3E shows estimates of the decay in pyrethroid activity assuming that the loss of efficacy due to washing is proportional to the change in activity seen over time ( i . e . if the rate of decay over subsequent washes is twice as fast in a resistant mosquito population than the decay of pyrethroid activity over time will also be twice as fast ) . Mosquitoes with high pyrethroid resistance appear to overcome the insecticide activity of the LLIN faster than susceptible mosquitoes . A hypothesis for the cause of this relationship is outlined in Figure 3F . The transmission dynamics model predicts that the higher the population prevalence of pyrethroid resistance the greater impact it will have on both the number of clinical cases ( Figure 4A and B ) and the force of infection ( as measured by the EIR , Figure 4C ) . This is due to the lower initial killing efficacy of the LLIN but also because of the higher rate of decay of insecticidal activity ( it gets less effective more quickly ) . The absolute increase in EIR caused by resistance increases in areas of high endemicity ( Figure 4C ) , though the model predicts that the number of clinical cases caused will peak at intermediate parasite prevalence because high levels of clinical immunity will mask increased infection rates in hyper-endemic areas . Understandably the impact of resistance will depend on the current LLIN coverage , with the total public health impact of resistance being greatest in areas where bednets were having the highest impact ( i . e . areas of lower , 50% , coverage , see Figure 4—figure supplement 1 ) . Equally the impact of resistance will be higher in areas with mosquito species which are more amenable to control through the use of LLINs ( i . e . greater in Anopheles gambiae sensu stricto than Anopheles arabiensis , Figure 4—figure supplements 2 and 3 ) . The transmission dynamics model predicts that the public health impact of pyrethroid resistance will be high . For example with as little as 30% resistance ( 70% mortality in discriminating dose assay ) in a population with 10% slide prevalence ( in 2–10 year olds ) the model predicts that pyrethroid resistance would cause an additional 245 ( 95%CI 142–340 ) cases per 1000 people per year ( Figure 4A , averaged over the 3 year life-expectancy of the net ) . Similar increases in the number of cases are seen in those with or without LLINs ( Figure 4A ) . 10 . 7554/eLife . 16090 . 013Figure 4 . The predicted impact of pyrethroid resistance on the clinical incidence of malaria ( Panels A and B ) and the force of infection ( Panel C ) . Panel ( A ) shows how the number of clinical cases in the population increases with the population prevalence of pyrethroid resistance ( as assessed by the percentage survival in a pyrethroid bioassay ) for a single setting ( with 10% slide prevalence ) . Black lines show the full resistance model whilst the brown lines give predictions for mosquito populations where the rate of change in insecticide activity over time is the same for all mosquitoes ( i . e . resistance has no impact on LLIN longevity ) . Solid lines show the average for the population , shaded grey area indicates the 95% credible intervals around this best fit line , dashed lines denote those using bednets whilst dotted-dashed lines show those who do not . Panel ( B ) shows the 3D relationship between prevalence of resistance ( x-axis ) , endemicity ( y-axis ) and the absolute increase in the number of clinical cases ( contours , see colour legend ) per 1000 people ( all ages ) . Panel ( C ) presents the same model as ( B ) though showing the absolute increase in the entomological inoculation rate ( EIR , the average number of infectious bits per person per year ) . In this figure it is assumed that the mosquito species is Anopheles gambiae sensu stricto and that there is 80% LLIN coverage . Figure 4—figure supplement 1 shows the same figure with 50% LLIN coverage . Further secondary figures indicate how the impact of resistance changes with mosquito species , be it Anopheles arabiensis ( Figure 4—figure supplement 2 ) or Anopheles funestus ( Figure 4—figure supplement 3 ) . Panel ( A ) shows the importance of the rate of change in insecticide activity over time . Figure 4—figure supplement 4 shows how Panels B and C would change if the rate of decay in insecticide activity was the same for resistant and susceptible mosquitoes . The uncertainty in the three LLIN efficacy parameters used to generate the confidence interval estimates in Panel ( A ) are shown in ( Figure 4—figure supplement 5 ) for different levels of pyrethroid resistance . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 01310 . 7554/eLife . 16090 . 014Figure 4—figure supplement 1 . The predicted impact of pyrethroid resistance on the clinical incidence of malaria ( Panels A and B ) and the force of infection ( Panel C ) in an area with A . gambiae s . s . mosquitoes and 50% LLIN coverage . Panel ( A ) shows how the number of clinical cases in the population increases with the population prevalence of pyrethroid resistance ( as assessed by the percentage survival in a pyrethroid bioassay ) for a single setting ( with 10% slide prevalence ) . Solid lines show the average for the population whilst shaded grey area indicates the 95% credible intervals around this best fit line . Panel ( B ) shows the 3D relationship between prevalence of resistance ( x-axis ) , endemicity ( y-axis ) and the absolute increase in the number of clinical cases ( contours , see colour legend ) per 1000 people ( all ages ) . Panel ( C ) presents the same model as ( B ) though showing the absolute increase in the entomological inoculation rate ( EIR , the average number of infectious bits per person per year ) . In all figures it is assumed that the mosquito species is Anopheles gambiae sensu stricto and that there is 50% LLIN coverage . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 01410 . 7554/eLife . 16090 . 015Figure 4—figure supplement 2 . The predicted impact of pyrethroid resistance on the clinical incidence of malaria ( Panels A and B ) and the force ofinfection ( Panel C ) in an area with A . arabiensis mosquitoes and 80% LLIN coverage . Panel ( A ) shows how the number of clinical cases in the population increases with the population prevalence of pyrethroid resistance ( as assessed by the percentage survival in a pyrethroid bioassay ) for a single setting ( with 10% slide prevalence ) . Solid lines show the average for the population whilst shaded grey area indicates the 95% credible intervals around this best fit line . Panel ( B ) shows the 3D relationship between prevalence of resistance ( x-axis ) , endemicity ( y-axis ) and the absolute increase in the number of clinical cases ( contours , see colour legend ) per 1000 people ( all ages ) . Panel ( C ) presents the same model as ( B ) though showing the absolute increase in the entomological inoculation rate ( EIR , the average number of infectious bits per person per year ) . In all figures it is assumed that the mosquito species is Anopheles arabiensis and that there is 80% LLIN coverage . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 01510 . 7554/eLife . 16090 . 016Figure 4—figure supplement 3 . The predicted impact of pyrethroid resistance on the clinical incidence of malaria ( Panels A and B ) and the force of infection ( Panel C ) in an area with A . funestus mosquitoes and 80% coverage . Panel ( A ) shows how the number of clinical cases in the population increases with the population prevalence of pyrethroid resistance ( as assessed by the percentage survival in a pyrethroid bioassay ) for a single setting ( with 10% slide prevalence ) . Solid lines show the average for the population whilst shaded grey area indicates the 95% credible intervals around this best fit line . Panel ( B ) shows the 3D relationship between prevalence of resistance ( x-axis ) , endemicity ( y-axis ) and the absolute increase in the number of clinical cases ( contours , see colour legend ) per 1000 people ( all ages ) . Panel ( C ) presents the same model as ( B ) though showing the absolute increase in the entomological inoculation rate ( EIR , the average number of infectious bits per person per year ) . In all figures it is assumed that the mosquito species is Anopheles funestus and that there is 80% LLIN coverage . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 01610 . 7554/eLife . 16090 . 017Figure 4—figure supplement 4 . The predicted impact of pyrethroid resistance on ( A ) the clinical incidence of malaria and ( B ) the force of infection when pyrethroid resistance does not influence the rate of decay in LLIN insecticide activity over time ( i . e . resistance has no impact on LLIN longevity ) . Panel ( A ) shows the 3D relationship between population prevalence of resistance ( x-axis ) , endemicity ( y-axis ) and the absolute increase in the number of clinical cases ( contours , see legend ) per 1000 people ( all ages ) . Panel ( B ) presents the same model as ( A ) though showing the absolute increase in the entomological inoculation rate ( EIR , the average number of infectious bits per person per year ) . These panels can be directly compared to panels ( 4B ) and ( 4C ) of the main text where pyrethroid resistant mosquitoes overcome the actions of the insecticide earlier . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 01710 . 7554/eLife . 16090 . 018Figure 4—figure supplement 5 . Estimates in the uncertainty of the three LLIN efficacy parameters for different levels of pyrethroid resistance . Panels ( A–C ) show values for Anopheles gambiae senu lato whilst ( D–F ) show Anopheles funestus . ( A ) and ( C ) predict the proportion of mosquitoes dying per feeding attempt ( dp ) whilst ( B ) and ( C ) show the proportion of mosquitoes which successfully feed and survive ( sp ) . Panels ( C ) and ( F ) show how the estimated half-life of insecticide activity in years changes ( Hy ) with the pyrethroid resistance test . Green lines denote standard LLINs whilst purple lines indicate PBO LLINs . Solid line represents the best fit estimates whilst the shaded region gives the 95% credible intervals generated by sampling from the individual parameter posterior distributions used within the equation . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 018 The impact of the addition of the synergist , PBO , on pyrethroid induced mortality appears to depend on mosquito species and the population prevalence of pyrethroid resistance . In mosquito populations with moderate to high resistance results indicate PBO is an effective synergist of pyrethroids ( Figure 5A ) . For example in an area with 10% endemicity and 80% resistance ( 20% mortality in discriminating dose assay ) the model predicts that switching to PBO LLINs would avert an additional 501 ( 95%CI 319–621 ) cases per 1000 people per year ( Figure 5A ) compared to the same level of standard LLIN coverage . The absolute number of cases averted by switching to PBO LLINs is predicted to be greater in areas with intermediate endemicity as human immunity is likely to partially buffer the added benefit of PBO LLINs in areas of highest malaria prevalence ( Figure 5B ) . However , due to the non-linear relationship between incidence of clinical infection and endemicity , the greatest percentage reduction in clinical cases and EIR is seen in areas of low endemicity ( Figure 5CF ) . The exact change in clinical cases will vary between settings . For example switching from 80% coverage with standard LLINs to 80% coverage with PBO LLINs in an area with 30% endemicity and a mosquito population with 60% pyrethroid resistance is predicted to reduce the number of clinical cases by ~60% whereas the same switch in the type of nets used in an area with 30% endemicity and 20% pyrethroid resistance would only reduce the number of clinical cases by ~20% ( Figure 5C ) . Greater percentage reductions are likely to be seen in EIR than the number of clinical cases due to human immunity ( Figure 5E ) . 10 . 7554/eLife . 16090 . 019Figure 5 . Predicting the added benefit of switching from standard LLINs to combination PBO nets . Panels ( A–C ) show clinical incidence ( per 1000 people per year , all ages ) whilst Panels ( D–F ) gives the entomological inoculation rate ( EIR , infectious bites received per person per year ) . ( A ) and ( D ) show how malaria incidence and the force of infection increase with the population prevalence of pyrethroid resistance ( as assessed by the percentage survival in a pyrethroid bioassay ) in a single setting ( with 10% slide prevalence ) for standard LLINs ( green line ) and PBO LLINs ( purple line ) . Shaded region denotes the 95% credible intervals around the best fit lines . Panels ( B ) and ( E ) show the 3D relationship between the prevalence of resistance ( x-axis ) , endemicity ( y-axis ) and the absolute number of cases ( and EIR ) averted by switching to PBO LLINs . ( C ) and ( F ) give 3D relationship for the percentage reduction in cases and EIR ( respectively ) caused by switching from standard to PBO LLINs . The non-linear relationship between endemicity , clinical incidence and EIR means that the greatest percentage reduction is seen at low endemicities despite the greatest absolute reduction being in higher transmission settings . In all Panels it is assumed that the mosquito species is Anopheles gambiae sensu stricto and that there is 80% LLIN coverage . Figure 5—figure supplement 1 shows the same figure with 50% LLIN coverage . Further secondary figures indicate how the impact of resistance changes with mosquito species , be it Anopheles arabiensis ( Figure 5—figure supplement 2 ) or Anopheles funestus ( Figure 5—figure supplement 3 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 01910 . 7554/eLife . 16090 . 020Figure 5—figure supplement 1 . Predicting the added benefit of switching from standard LLINs to combination PBO nets in an area with A . gambiae s . s . mosquitoes and 50% LLIN coverage . Panels ( A–C ) show clinical incidence ( per 1000 people per year , all ages ) whilst Panels ( D–F ) gives the entomological inoculation rate ( EIR , infectious bites received per person per year ) . ( A ) and ( D ) shows how malaria incidence and the force of infection increases with the population prevalence of pyrethroid resistance ( as assessed by the percentage survival in a pyrethroid bioassay ) in a single setting ( with 10% slide prevalence ) for standard LLINs ( green line ) and PBO LLINs ( purple line ) . Panels ( B ) and ( E ) show the 3D relationship between the prevalence of resistance ( x-axis ) , endemicity ( y-axis ) and the absolute number of cases ( and EIR ) averted by switching to PBO LLINs . ( C ) and ( F ) give 3D relationship for the percentage reduction in cases ( and EIR ) caused by switching from standard to PBO LLINs . The non-linear relationship between endemicity , clinical incidence and EIR means that the greatest percentage reduction is seen at low endemicities despite the greatest absolute reduction being in higher transmission settings . In all figures it is assumed that the mosquito species is Anopheles gambiae sensu stricto and that there is 50% LLIN coverage . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 02010 . 7554/eLife . 16090 . 021Figure 5—figure supplement 2 . Predicting the added benefit of switching from standard LLINs to combination PBO nets in an area with A . arabiensis mosquitoes and 80% LLIN coverage . Panels ( A–C ) show clinical incidence ( per 1000 people per year , all ages ) whilst Panels ( D–F ) gives the entomological inoculation rate ( EIR , infectious bites received per person per year ) . ( A ) and ( D ) shows how malaria incidence and the force of infection increases with the population prevalence of pyrethroid resistance ( as assessed by the percentage survival in a pyrethroid bioassay ) in a single setting ( with 10% slide prevalence ) for standard LLINs ( green line ) and PBO LLINs ( purple line ) . Panels ( B ) and ( E ) show the 3D relationship between the prevalence of resistance ( x-axis ) , endemicity ( y-axis ) and the absolute number of cases ( and EIR ) averted by switching to PBO LLINs . ( C ) and ( F ) give 3D relationship for the percentage reduction in cases ( and EIR ) caused by switching from standard to PBO LLINs . The non-linear relationship between endemicity , clinical incidence and EIR means that the greatest percentage reduction is seen at low endemicities despite the greatest absolute reduction being in higher transmission settings . In all figures it is assumed that the mosquito species is Anopheles arabiensis and that there is 80% LLIN coverage . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 02110 . 7554/eLife . 16090 . 022Figure 5—figure supplement 3 . Predicting the added benefit of switching from standard LLINs to combination PBO nets in an area with A . funestus mosquitoes and 80% LLIN coverage . Panels ( A–C ) show clinical incidence ( per 1000 people per year , all ages ) whilst Panels ( D–F ) gives the entomological inoculation rate ( EIR , infectious bites received per person per year ) . ( A ) and ( D ) shows how malaria incidence and the force of infection increases with the population prevalence of pyrethroid resistance ( as assessed by the percentage survival in a pyrethroid bioassay ) in a single setting ( with 10% slide prevalence ) for standard LLINs ( green line ) and PBO LLINs ( purple line ) . Panels ( B ) and ( E ) show the 3D relationship between the prevalence of resistance ( x-axis ) , endemicity ( y-axis ) and the absolute number of cases ( and EIR ) averted by switching to PBO LLINs . ( C ) and ( F ) give 3D relationship for the percentage reduction in cases ( and EIR ) caused by switching from standard to PBO LLINs . The non-linear relationship between endemicity , clinical incidence and EIR means that the greatest percentage reduction is seen at low endemicities despite the greatest absolute reduction being in higher transmission settings . In all figures it is assumed that the mosquito species is Anopheles funestus and that there is 80% LLIN coverage . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 022 Pyrethroid resistance is widespread across Africa though its public health impact is unknown . Here we show that the simple bioassay can be used to predict how pyrethroid resistance is changing the efficacy of different types of LLIN and how this would be expected to influence malaria morbidity . The bioassay is a crude tool for measuring pyrethroid resistance , though its simplicity makes it feasible to use on a programmatic level . Figure 2A and C indicate that on average bioassay mortality is able to predict the results of standard and PBO LLIN experimental hut trials for A . gambiae s . l . mosquitoes . There is a high level of measurement error in the bioassay ( as seen by the wide variability in points in Figure 2A and B ) so care should be taken when interpreting the results of single assays as differences in mosquito mortality may have been caused by chance . Multiple bioassays could be conducted on the same mosquito population and the results averaged to increase confidence . However the exact cause of the measurement error remains unknown so increased repetition many not necessarily generate substantially more accurate results as possible causes of variability , such as mosquito husbandry techniques or environmental conditions ( Kleinschmidt et al . , 2015 ) , may be repeated . Further work is therefore needed to determine whether assay repetition substantially improves overall accuracy or whether further standardisation or more complex assays are required . The majority of data are for A . gambiae s . l . so the analysis needs to be repeated for other species once data becomes available . More advanced methods of measuring insecticide resistance ( such as the intensity bioassay [Bagi et al . , 2015] or the use of genetic markers [Weetman and Donnelly , 2015] ) are likely to be a more precise way of predicting resistance . However , since there are insufficient data to repeat these analyses with these other assays their predictive ability remains untested . Similarly , this analysis has grouped WHO tube , WHO cone and CDC bottle assays together when the use of a single assay type might be more predictive . The meta-analysis of experimental hut trials in areas with different levels of resistance has important implications for our understanding of how pyrethroid resistance influences LLIN efficacy . This analysis suggests that the probability that a mosquito will feed on someone beneath an LLIN only increases substantially at high levels of pyrethroid resistance ( Figure 3C ) . People under bednets exposed to mosquito populations with intermediate levels of resistance still have a high degree of personal protection whilst in bed as those mosquitoes , which do not die are likely to exit the hut without feeding . It is only when mosquito populations are highly resistant ( >60% survival ) that an increasing proportion of mosquitoes appear to successfully feed through the LLIN ( Figure 3D ) . This may explain why a previous meta-analysis on the impact of pyrethroid resistance on LLIN efficacy in experimental hut trials failed to find a substantial effect ( Strode et al . , 2014 ) as resistance was categorised into broad groups ( partially based on highly variable bioassay data ) unlike here where resistance is treated as a continuous variable ( as measured using experimental hut trial mortality data which are less variable than bioassay data ) . This earlier study also only analysed papers published or presented prior to May 2013 and so it did not include the recent experimental hut trials which had the lowest mosquito mortality ( Toé , 2015; Pennetier et al . , 2013 ) . The meta-analysis revealed that the number of mosquitoes deterred from entering a hut with an LLIN , decreases with increasing pyrethroid resistance . LLIN efficacy is therefore reduced as mosquitos enter huts where they have both a higher chance of feeding and a lower chance of being killed . These parallel changes in behaviour increase the resilience of mosquito populations to LLINs as in a susceptible mosquito population , high deterrence will reduce LLIN efficacy by preventing mosquitoes entering houses where they have a high chance of being killed ( relative to susceptible populations ) . Importantly the loss of deterrence suggests that those sleeping in a house with an LLIN though not sleeping under the net themselves ( a phenomenon particularly common in older children [Nankabirwa et al . , 2014] ) will lose an additional degree of protection ( on top of the community impact of mosquito killing ) . The overall effectiveness of LLINs depends on the duration of insecticidal activity . Evidence suggests that multiply washed LLINs lose their ability to kill mosquitoes more in areas of high pyrethroid resistance . Washing is seen as an effective method of aging LLINs ( WHO , 2013a ) . Repeatedly washing a net ( and presumably reducing the concentration of the insecticide ) appears to have little impact on its ability to kill a susceptible mosquito whilst significantly reducing the lethality of the LLIN against more resistant mosquitoes ( Figure 2E ) . The difference in mortality is likely to be caused by mosquitoes with a higher population prevalence of resistance being able to tolerate a higher concentration of insecticide ( WHO , 2013a ) . If so , then the higher longevity of LLINs against susceptible mosquitoes observed in the washed net data may be explained by the longer time it takes for the insecticide concentration on the LLIN to drop below this critical level ( Figure 2F ) . This analysis assumes that the decay in pyrethroid activity over time is proportional to its decay following washing and this needs to be confirmed by durability studies in areas of high pyrethroid resistance . Nevertheless the results seem to be confirmed by two recent studies which evaluated mosquito mortality in older ( standard ) LLINs ( Toé et al . , 2014; Wanjala et al . , 2015 ) . Durability studies should be prioritised as the model predicts that , even at low levels of pyrethroid resistance , the loss of insecticide activity over the three year bednet life-expectancy , has a bigger epidemiological impact on malaria , than the initial efficacy of new LLINs . If confirmed then more regular net distribution could be considered as a temporary , albeit expensive , method to mitigate the public health impact of high pyrethroid resistance . Transmission dynamics mathematical models are a useful tool for disentangling the different impacts of LLINs . Though a person under an LLIN requires high pyrethroid resistance before LLINs start to fail ( Figure 3C ) , the models predict that at a population level even low pyrethroid resistance can increase the number of malaria cases over the life-time of the net ( Figure 4A ) . Hut trials measure feeding when the volunteer is underneath a bednet whilst in reality ( and in the mathematical model ) a percentage of mosquito bites are taken when people are not in bed . The loss of LLIN induced mosquito mortality is likely to decrease the community impact of LLINs , increasing average mosquito age and the likelihood that people are infected whilst unprotected by a bednet . This is primarily due to the shorter duration of insecticide potency of LLINs in mosquito populations with a higher prevalence of resistance ( Wanjala et al . , 2015 ) . Without this change in the duration of pyrethroid activity , the epidemiological impact of pyrethroid resistance will only become evident once it reaches a high level ( Figure 4A ) . The change in the community impact of LLINs can be seen in the increase in the number of cases in people who do not use nets . This change is substantial , reinforcing the need to consider community effects in any policy decision . Detecting an epidemiological impact of a low population prevalence of resistance may be challenging for local health systems ( for example , see < 20% resistance prevalence Figure 1—figure supplement 1 , Figure 4 ) especially in an area where LLIN coverage , local climatic conditions and the use of other malaria control interventions are changing over time . These simulations also assume that resistance arrives overnight , when in reality it will spread through a mosquito population more gradually and therefore may be harder to detect . Mosquitoes exposed to LLINs may have reduced fitness ( Viana et al . , 2016 ) . Currently the model assumes that mosquitoes which survive 24 hr after LLIN exposure are indistinguishable from unexposed mosquitoes . If this is not the case then hut trials data alone will be insufficient to predict the public health impact of pyrethroid resistance as current models will over-estimate its impact . Similarly , if the mosquito population exhibits additional behavioural mechanisms to avoid LLINs , such as earlier biting times , in tandem to the increased tolerance of pyrethroid insecticide then the predictions presented here will likely underestimate the public health impact as this behaviour change has not been incorporated . Currently a mosquito population is defined as being pyrethroid resistant if there is < 90% bioassay mortality ( WHO , 2013b; Mnzava et al . , 2015 ) . Though useful , this entomological measure should not be considered as a measure of the effectiveness of pyrethroid LLINs . The personal protection provided by sleeping under an LLIN is likely to be substantial even at very high levels of resistance ( Strode et al . , 2014; Randriamaherijaona et al . , 2015 ) . Any reduction in mosquito mortality will likely reduce the community impact of LLINs though it may be hard to detect , especially in areas with new LLINs ( the public health impact of resistance is likely to be greater in older nets , Figure 3E ) . As with all transmission dynamic mathematical models , these predictions need to be validated in particular locations with well-designed studies combining epidemiological and entomological data . We are currently unaware of any published data with sufficient information to test the model against though a thorough validation exercise should be carried out as soon as such studies become available . Currently the meta-analyses and transmission dynamics models concentrated on malaria in Africa and give predictions for the three primary mosquito vector species found there . Each meta-analyses has data from multiple countries but these sites are not geographically representative of the whole of malaria endemic Africa . Though the principles outlined here may apply to other mosquito species in different care settings should be taken when extrapolating the results beyond the areas where the data were collated . The bioassay data indicate that the ability of PBO to synergise pyrethroid induced mortality depends on the mosquito species . In A . funestus PBO always appears to restore near 100% mortality whilst for mosquitoes from the A . gambiae complex the greatest additional benefit of PBO being seen at intermediate levels of pyrethroid resistance ( Figure 2B ) . The exact causes of this are unknown but is likely related to the predominant resistance mechanisms in each species . PBO’s primary synergistic effect of pyrethroids is thought to be due to the inhibition of the cytochrome P450 enzymes which catalyse the detoxification of the insecticides ( Farnahm , 1998 ) . Elevated P450 levels are the primary resistance mechanism in A . funestus whereas in A . gambiae s . l . both increased detoxification and alterations in the target site contribute to pyrethroid resistance with the latter mechanism being largely unaffected by PBO ( Mulamba et al . , 2014; Riveron et al . , 2013 ) . For A . gambiae s . l . populations this result was verified by experimental hut trial data which directly compare standard and PBO LLINs ( Figure 2C ) . Both bioassay and hut trial data suggest a minimal additional benefit of PBO in areas with very high levels of pyrethroid resistance . Unfortunately , there are currently no published studies where PBO LLINs have been tested in experimental hut trials in areas with A . funestus so these bioassay results should be treated with caution until they can be further verified . Additional data would also allow the differences between species in the A . gambiae complex to be assessed . A previous analysis comparing PermaNet 2 . 0 and 3 . 0 was unable to test whether the increase in efficacy of the PBO LLIN was solely due to the addition of PBO as this net has a higher concentration of insecticide ( Briët et al . , 2013 ) . The results presented here show a consistent pattern between PermaNet 2 . 0 and 3 . 0 and Olyset and Olyset Plus . As both Olyset nets have the same concentration of insecticide , this suggests that PBO is causing the enhancement of efficacy . The WHO recommends that countries routinely conduct non-PBO pyrethroid bioassays as part of their insecticide resistance management plan ( WHO , 2012 ) . In areas with A . gambiae s . l . the evidence presented here suggests that the results of bioassays with and without PBO can be used to predict the additional public health benefit of PBO LLINs . If there is a greater mortality in the PBO bioassay and the relative mortalities broadly agree with the red curve in Figure 2B , then Figure 5B can be used to predict the approximate number of cases that will be saved by switching from standard to PBO LLINs ( for a given level of endemicity and LLIN coverage ) . Areas with 40–90% survival ( 10–60% mortality ) in a non-PBO standard bioassay ( of any type ) should consider conducting PBO synergism bioassays to determine the suitability of PBO LLINs . We would suggest that either the WHO cone , WHO tube or CDC bottle assay ( conducted in triplicate and averaged to improve precision ) should be sufficient evidence to justify the need to switch to PBO LLINs . The decision to recommend PBO nets over standard LLINs requires information on the relative cost effectiveness and affordability of PBO nets . If both net types cost the same and resistance has been detected then this work indicates that PBO LLINs should always be deployed as evidence suggests that they are always more effective . However , if PBO nets are more expensive , then cost effectiveness analysis will be required . The results of such analysis are likely to be context specific ( depending on price , resistance level , endemicity and coverage ) and interpreting them will require information on decision makers’ willingness and ability to pay for additional effectiveness . In many situations , malaria control budgets are likely to be fixed and therefore switching to more expensive PBO LLINs may cause a reduction in overall bednet coverage . The impact of reduced coverage must therefore be off set against the benefits of introducing PBO nets , taking into consideration any additional factors such as changed programmatic costs , and equity issues . Rapid deployment of new vector control products saves lives and gives incentives for industry to invest in new methods of vector control . New methods are likely to have a higher unit price than existing tools so it is important to be able to determine where and when they should be deployed in an efficient and transparent manner . Frameworks such as those presented here offer a relatively straightforward method of comparing two different products to determine whether the increased effectiveness justifies the higher unit price . Much of the debate over the impact of pyrethroid resistance on LLIN effectiveness has focused on the loss of personal protection provided by new nets and does not fully take into account their community impact . A large body of evidence has shown how widespread use of LLINs can cause considerable community protection , both to those who use bednets and non-users ( Killeen , 2007 and references therein ) . Therefore the community impact should be considered in any study investigating the consequences of pyrethroid resistance ( Briët et al . , 2013; Killeen , 2014 ) , as any reduction in mosquito killing is likely to increase malaria cases even in areas with mildly resistant mosquito populations where LLINs are still providing good personal protection . The evidence presented here suggests that high levels of pyrethroid resistance are likely to have a bigger public health impact than previously thought and therefore could represent a major threat to malaria control in Africa . To generate results which are broadly applicable to all mathematical models were fit to data compiled by systematic meta-analyses of the published literature . Where possible meta-analyses were extended to the grey literature by including unpublished information . These include unpublished bioassay data from Liverpool School of Tropical Medicine , submissions to the World Health Organisation Pesticide Evaluation Scheme ( WHOPES ) and results from unpublished experimental hut trials ( collated by contacting LLIN manufacturers Vestergaard-Frandsen and Sumitomo Chemicals Ltd ) . The meta-analyses followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines ( Moher et al . , 2009 ) for study search , selection and inclusion criteria though the study was not registered . The predefined inclusion criteria of each of the meta-analyses are presented in Table 2 whilst the pre-defined search strings and the databases searched are outlined in full in Figure 2—source data 1 . Extraction was done by N . L . into piloted forms . Study corresponding authors were contacted for raw data when this information was unavailable ( all contacted investigators responded with the requisite information ) . 10 . 7554/eLife . 16090 . 023Table 2 . Inclusion and exclusion criteria used when conducting literature searches of published and grey literature . Pre-defined search string used are listed in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 023Inclusion criteriaExclusion criteriaGeneral criteria across all meta-analyses – Mosquito belong to the A . gambiae complex or A . funestus group – Study conducted in Africa – Bioassay must be of the standard dose for the particular pyrethroid ( WHO , 2013a , 2013b; Brogdon , 2010 ) – Net must be a pyrethroid LLIN – Studies which report percentage mortality but not the numbers tested / caught† – Experimental hut trials which do not have adequate design to reduce bias ( i . e . treatments arms were not rotated between huts; sleeper bias unaccounted for by preliminary testing; randomisation or rotation; huts were not cleaned between treatments ) – Experimental huts of the Ifakara design‡M1 – Bioassay and experimental hut trial mortality – Mosquito mortality measured in both an experimental hut study and separate bioassay ( e . g . WHO tube assay , WHO cone assay , CDC bottle assay ) – Cone assays where the net had been washedM2 – Impact of PBO in pyrethroid bioassays – adult mosquito stage exposure to PBOM3 – Experimental hut trials of standard and PBO LLINS – Study compares a combination LLIN ( PermaNet 3 . 0 or Olyset Plus ) with a conventional LLIN ( PermaNet 2 . 0 or Olyset Net ) * – LLINs should be holed ( Six 4 × 4 cm holes ) – Studies without both standard and PBO LLINs as non-parallel studies as studies from different sites may bias the difference between LLINs – Trials without untreated control nets – Studies which did not include feeding success* currently there are only two commercially available LLINs with PBO , PermaNet 3 . 0 ( Vestergaard-Frandsen ) and Olyset Plus ( Sumitomo Chemicals Ltd ) . To limit the difference between LLIN types only nets made by the same manufacturer are directly compared . † to increase the size of the bioassay dataset the authors of papers which failed to give sample sizes were contacted directly . ‡ The probability that a mosquito will die in an experimental hut will depend on the hut design . To minimise the difference between studies , the most common design of hut is used , excluding the small number of studies which use the new Ifakara design ( eg . Okumu et al . , 2013 ) . To determine whether simple pyrethroid bioassays can be used to infer the outcome of experimental LLIN hut trials a meta-analysis ( summarised as Meta-analysis 1 , M1 ) was conducted to identify studies where both were carried out concurrently . To test whether this relationship changed with the population prevalence of insecticide resistance simple functional forms were fit to the raw data using a mixed-effect logistic regression ( summarised as Relationship 1 , R1 ) . There has been an attempt to standardise bioassay and experimental hut trial procedures to enable data from different studies to be directly compared . These include using standard concentrations of insecticide , mosquito exposure time and mosquito husbandry in bioassays , hut design , trap type and the use of human baits in experimental hut trials . Nevertheless , some procedural discrepancies remain between studies , for example , in bioassays the age and sex of mosquitoes and how they were collected ( e . g . F1 progeny of wild caught mosquitoes or wild caught larvae reared in insectary and tested as adults ) . These co-variates and others ( for example information on genetic markers associated with insecticide resistance ) , could be included within the analysis , though their addition would increase data needs of future studies and complicate the use of study results . Instead a mixed-effects binomial regression is adopted which allows mosquito mortality to vary at random between studies . This statistical method enables a wider selection of studies to be included within the analysis , produces more generalizable results and reduces problems caused by data autocorrelation . Mosquito mortality in an experimental hut trial is defined as the proportion of mosquitoes , which enter the hut which die , either within the hut or within the next 24 hr . Meta-analysis 1 ( M1 ) identified only 7 studies where concurrent bioassays and experimental hut trials were carried out ( Table 3 ) . Given the paucity of data results from all types of bioassay and mosquito species were combined and a simple , functional form was used to describe the relationship ( the fixed-effect ) . Let x denote the proportion of mosquitoes dying in a standard ( non-PBO ) pyrethroid bioassay then the population prevalence of pyrethroid resistance ( expressed as a percentage , denoted I ) is described by the following equation , [1]I=100 ( 1−x ) . 10 . 7554/eLife . 16090 . 024Table 3 . List of studies identified in meta-analysis M1 - Predicting LLIN effectiveness from bioassay mortality . Pre-defined search string used in the meta-analyses are listed in Figure 2—source data 1 whilst raw data from are provided in Figure 2—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 024StudyReferenceTestCountry1Ngufor et al . ( 2014a ) WHO tubeCôte d'Ivoire2Ngufor et al . ( 2014b ) WHO tubeBenin3Kitau et al . ( 2014 ) WHO tubeTanzania4Asale et al . ( 2014 ) WHO tubeEthiopia5Ngufor et al . ( 2014c ) WHO tubeBurkina Faso6Agossa et al . ( 2014 ) WHO tubeBenin7Malima et al . ( 2013 ) WHO tubeTanzania8Adeogun et al . ( 2012b ) WHO tubeNigeria9Koudou et al . ( 2011 ) WHO tubeCôte d'Ivoire10Corbel et al . ( 2010 ) WHO tubeBenin , Burkina Faso , Cameroon11Tungu et al . ( 2010 ) WHO tubeTanzania12Malima et al . ( 2008 ) WHO tubeTanzania13Kétoh ( 2016 ) WHO tubeTogo14Toé ( 2015 ) WHO tubeBurkina Faso Extending the notation of Griffin et al . ( 2010 ) the proportion of mosquitoes , which died in a hut trial is denoted lp , where subscript p indicates the net type under investigation , be it a no-net control hut ( p=0 ) , a standard non-PBO LLIN ( p=1 ) , or a PBO LLIN ( p=2 ) . For a standard LLIN it is assumed to be explained by the equation , [2]logit ( l1 ) =α1+α2 ( x−τ ) , where parameters α1 and α2 define the shape of the relationship and τ is a constant used to centre data to aid the fitting process . More sophisticated functional forms could be used for R1 ( Equation [2] ) though they were not currently warranted given the limited dataset . Let Np indicate the number of mosquitoes entering a hut in an experimental hut trial . If the number of these mosquitoes which enter the hut and subsequently die ( L1 ) follows a binomial distribution then parameters α1 and α2 can be estimated for a non-PBO net by fitting the following equation to M1 , [3]L1∼B ( l1 , N1 ) +ϵα . The random-effects component is included by allowing mortality to vary at random between sites by adding the error term ϵα which has a mean of zero and a constant variance . The number of experimental hut trials investigating the difference between standard and PBO nets is limited . Instead a meta-analysis of all bioassay data investigating the impact of PBO on pyrethroid induced mosquito mortality is undertaken incorporating all published and unpublished literature ( M2 , Table 4 ) . Bioassay mortality can be influenced by a multitude of factors including assay type , temperature and relative humidity ( Kleinschmidt et al . , 2015 ) . To account for this difference between studies , the relationship between the benefit of adding PBO and the population prevalence of pyrethroid resistance was estimated using a mixed-effect logistic regression ( R2 ) . Preliminary analysis suggests that the shape of the relationship is relatively complex and cannot simply be described by the use of a standard linear function typically used in regression . Since the added benefit of PBO in a given population will ultimately be determined by the shape of this relationship a variety of different functional forms are tested statistically . It was initially intended to include the type of assay used ( e . g . WHO tube assay , WHO cone assay or CDC bottle assay ) as an additional fixed effect , though the paucity of data ( especially comparing bioassay mortality to experimental hut trial mortality ) meant that data from all assays were combined and this covariate was excluded . As the same type of assay are used for both non-PBO and PBO tests this should not bias the results and will generate recommendations that are generalizable across all three assay types . The proportion of mosquitoes killed by pyrethroid insecticide in a bioassay with the addition of PBO is denoted f and is given by the equation:[4]logit ( f ) =β1+β2 ( x−τ ) 1+β3 ( x−τ ) 10 . 7554/eLife . 16090 . 025Table 4 . List of studies identified in meta-analysis M2 - Estimating the impact of PBO in pyrethroid bioassays . Bioassays run using laboratory strains are denoted . * Pre-defined search string used in the meta-analyses are listed in Figure 2—source data 1 whilst raw data from are provided in Figure 2—source data 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 025StudyReferenceTestCountry1Matowo et al . ( 2015 ) CDC tubeTanzania2Farnahm ( 1998 ) WHO tubeUganda & Kenya3Choi et al . ( 2014 ) WHO tubeZambia & Zimbabwe4Edi et al . ( 2014 ) WHO tubeCôte d'Ivoire5Jones et al . ( 2013 ) WHO tubeZanzibar6Chouaïbou et al . ( 2014 ) WHO tubeCôte d'Ivoire7Koffi et al . ( 2013 ) WHO tubeCôte d'Ivoire8Witzig et al . ( 2013 ) WHO tubeChad9Darriet and Chandre ( 2013 ) WHO tube10Mawejje et al . ( 2013 ) WHO tubeUganda11Adeogun et al . ( 2012b ) WHO tubeNigeria12Nardini et al . ( 2012 ) WHO tubeNigeria13Darriet and Chandre ( 2011 ) WHO tubeSouth Africa & Sudan14Kloke et al . ( 2011 ) WHO cone15Awolola et al . ( 2009 ) WHO tubeMozambique16Brooke et al . ( 2001 ) WHO tubeNigeria17N’Guessan et al . ( 2010 ) WHO tubeMozambique18Ranson ( 2015 ) Personal CommunicationWHO tubeBurkina Faso/Benin19Ranson ( 2015 ) Personal CommunicationWHO tubeChad colony20Morgan ( 2015 ) Personal CommunicationWHO tubeCôte d'Ivoire21Ranson ( 2015 ) Personal CommunicationWHO tubeBenin22Koudou & Malone ( 2015 ) Personal CommunicationWHO coneCôte d'Ivoire23PMI ( 2014 ) . Personal CommunicationCDC tubeMali24Toé ( 2015 ) WHO tubeBurkina Faso25Abílio et al . ( 2015 ) WHO coneMozambique26Riveron et al . ( 2015 ) WHO coneMalawi27Awolola et al . ( 2014 ) WHO coneNigeria28Yewhalaw et al . ( 2012 ) WHO coneEthiopia where x is the proportion of mosquitoes dying in a non-PBO bioassay , parameters , β1 , β2 and β3 define the shape of the relationship and τ is a constant supporting the fitting process ( this relationship is referred to as R2 ) . Let Ai be the number of mosquitoes used in a bioassay and Di the number which died , with subscript i denotes whether or not PBO was added to the bioassay ( i = 1 pyrethroid alone , i = 2 pyrethroid plus PBO ) . If it is assumed that the number of mosquitoes that die in the bioassay follows a binomial distribution then parameters , β1 , β2 and β3 can be estimated by fitting the following equations to the dataset from ( M1 ) , [5]D1∼B ( x , A1 ) +ϵβ , [6]D2∼B ( f , A2 ) +ϵβ . Parameter ϵβ represents a normally distributed random error with a mean of zero and a constant variance and is estimated from the fitting procedure . Relationships R1 and R2 can be used to predict the effectiveness of PBO LLINs in experimental hut trials . When bioassay data are unavailable the current population prevalence of insecticide resistance can be predicted from mosquito mortality measured in a standard LLIN experimental hut trial by rearranging Equation [2] , [7]x^=[ ( exp⁡ ( l1 ) 1−exp⁡ ( l1 ) ) −α1]/α2+τ , where the section in round brackets is the inverse logit function . This equation together with Equations [2] and [4] can be then used to predict the relationship between hut trial mortality in standard and PBO LLINs for a range of areas with different levels of pyrethroid resistance using the following steps ( a ) to ( c ) below . To test the predictive ability of R1 and R2 a third meta-analysis was carried out for all experimental hut trials which directly compare standard and PBO pyrethroid LLINs ( M3 , Table 5 ) . The accuracy of these predictions can then be examined by comparing them visually ( Figure 2C ) or statistically using an Anaylsis of Variance . 10 . 7554/eLife . 16090 . 026Table 5 . List of studies identified in meta-analysis M3 - Estimating the impact of PBO in experimental hut trials . Pre-defined search string used in the meta-analyses are listed in Figure 2—source data 1 whilst raw data from published studies are provided at doi:10 . 5061/dryad . 13qj2 . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 026StudyReferenceCountry1Pennetier et al . ( 2013 ) Benin , Cameroon2Adeogun et al . ( 2012a ) ) Nigeria3Corbel et al . ( 2010 ) Benin , Burkina Faso , Cameroon4Tungu et al . ( 2010 ) Tanzania5N'Guessan et al . ( 2010 ) Benin6Kétoh et al . , UnpublishedTogo7Tungu et al . , Personal CommunicationTanzania8Toé et al . , Personal CommunicationBurkina Faso The impact of insecticide resistance on mosquito interactions with LLINs is systematically investigated by analysing the experimental hut trials identified in M3 . Restricting the analysis to the two most commonly used standard LLINs minimises the inter-study variability and allows the different behaviours of mosquitoes exposed to standard and PBO LLINs to be directly assessed . Following a widely used transmission dynamics model of malaria ( Griffin et al . , 2010; Walker et al . , 2015 ) it is assumed that an LLIN can alter a host-seeking mosquito behaviour in one of three ways: firstly it can deter a mosquito from entering a hut ( an exito-repellency effect ) ; secondly the mosquito can exit the hut without taking a bloodmeal; and thirdly it could kill a mosquito ( with the mosquito either being fed or unfed ) . A mosquito that isn’t deterred , exited or killed will successfully blood-feed and survive . The public health benefit of LLINs depends not only on their initial effectiveness but also on how the properties of the net changes over its life-time . The ability of a net to kill a mosquito will decrease over time as the quantity of insecticide active ingredient declines . The non-lethal protection provided by the LLIN may also decrease with the decay of the active ingredient and the physical degradation of the net ( i . e . the acquisition of holes ) . It is assumed that the underlying difference in hut trial mortality between sites for standard LLINs is caused by the mosquito population having a different population prevalence of pyrethroid resistance . Pyrethroid resistance may also influence the relative strength of LLIN deterrence and exiting and it is important to characterise these modifications of behaviour as they contribute substantially to the population level impact of mass LLIN distribution . Visual inspection of these data indicates that mosquito deterrence and exiting can be described by the degree of mosquito mortality seen in the same hut trial . The proportion of mosquitoes not deterred from entering a hut by the LLIN is estimated using mp , the ratio of the number of mosquitoes entering a hut with an LLIN ( N1or N2 ) to the number entering a hut without a bednet ( N0 , here assumed to be the same as a hut with an untreated bed net ) . A statistical model is used to determine whether there is an association between the number of mosquitoes entering a hut with a standard LLIN and the proportion of mosquitoes which die when they do ( which is assumed to be a proxy for mosquito susceptibility , i . e . m1 is described by l1 and m2 is described by l2 ) . It is assumed that the shape of the relationship between the proportion of mosquitoes entering a hut with an LLIN relative to a hut with an untreated net ( 1-deterrence ) and mortality is described by the flexible third order polynomial , [8]mp=1−[ δ1+δ2 ( lp−τ ) +δ3 ( lp−τ ) 2 ][9]Np~N ( mpN0 , δ4 ) Though there is no a priori reason to assume an inflection point in the relationship between mp and lp the polynomial function is chosen as it is highly flexible and would allow such a curve should it exist ( which is necessary given the variability in the raw data ) . The shape parameters δ1 , δ2 and δ3 are estimated assuming that the number of mosquitoes caught has a normal distribution ( verified using a and deterrence is allowed to vary at random between sites ( with variance δ4 ) . The proportion of mosquitoes entering the hut which exit without feeding is denoted jp whilst the proportion which successfully feed upon entering is kp . Once entered the hut mosquitoes have to either exit , die or successfully feed ( i . e . 1=jp+lp+kp ) . Visual inspection of these data indicates that kp increases with decreasing mortality at an exponential rate ( Figure 3C ) . Therefore , if the number of mosquitoes which feed and survive ( Sp ) follows a binomial distribution then , [10]Sp~B ( kp , Np ) +ϵθ [11]kp=θ1exp[ θ2 ( 1−lp−τ ) ] where θ1 and θ2 determine the shape of the relationship and ϵθ is a normally distributed random error which varies between sites . Estimates of jp , lp and mp can be used to determine the proportion of mosquitoes repeating ( a combination of deterrence and exiting , rp0 ) , dying ( dp0 ) and feeding successfully ( sp0 ) during a single feeding attempt in a hut with a new LLIN relative to those successfully feeding in a hut without an LLIN ( i . e . p=1 or 2 ) , [12]rp0= ( 1−kp'k0 ) ( jp'jp'+lp' ) [13]dp0= ( 1−kp'k0 ) ( lp'jp'+lp' ) [14]sp0=kp'k0 where jp=1−lp−kp , jp'=mpjp+ ( 1−mp ) , kp'=mpkp and lp'=mplp ( Griffin et al . , 2010 ) . Not all mosquitoes which enter a house will successfully feed even if there are no vector control interventions inside . The experimental hut trials used in this analysis did not include a no-net control ( k0 ) so historical studies are used for this parameter ( Curtis et al . , 1996; Lines et al . , 1987 ) . Though theoretically sp0 could have values > 1 for practical purposes , it is constrained between zero and one as on average mosquitoes entering a hut with an LLIN are less likely to feed than a mosquito entering a hut without a bednet ( as shown by all estimates of kp being substantially lower than k0 , see Figure 3C and Table 6 ) . The majority of experimental hut trials in M3 are in areas where the dominant vector is A . gambiae s . s . and no studies were conducted in areas with A . funestus . As there is insufficient information to generate these functions for each species separately it is assumed that the relationship between rp0 , sp0 and dp0 are consistent across all species . The average effectiveness of LLINs in an entirely susceptible mosquito population identified in M3 is slightly higher than those analysed by Griffin et al . ( 2010 ) which included a wider range of LLIN types . Values of mp ( the propensity of mosquitoes to enter a hut with an LLIN relative to one without ) greater than one are truncated at one as there is insufficient evidence to justify that mosquitoes preferentially enter huts with LLINs ( in part because the number of studies with very low mortality are low and the metric has high measurement error ) . 10 . 7554/eLife . 16090 . 027Table 6 . Parameter definitions and fitted values . Unless otherwise stated , all other parameters used were taken from Griffin et al . ( 2010 ) . Some parameters are mosquito species-specific whilst others are constant within a species complex ( denoted * ) or universal ( species independent $ ) . DOI: http://dx . doi . org/10 . 7554/eLife . 16090 . 027Parameter definitionsAnopheles gambiae s . s . Anopheles arabiensisAnopheles funestusxproportion mosquitoes dying in a discriminating dose pyrethroid bioassay-Ipopulation prevalence of pyrethroid resistance ( percentage survival ) estimated using x ( Equation [1] ) -pnet type under investigation in experimental hut trials: untreated ( p=0 ) ; standard LLIN ( p=1 ) ; PBO LLIN ( p=2 ) . -dpprobability a mosquito dies during single feeding attempt ( Equation [18] ) Estimated from parameters belowrpprobability a mosquito exits the hut during single feeding attempt ( Equation [17] ) Estimated from parameters belowspprobability a mosquito feeds during single feeding attempt ( Equation [19] ) Estimated from parameters belowNpthe number of mosquitoes entering a hut with net type p ( Equation [3] ) -mpproportion of mosquitoes entering a hut with an LLIN to relative to a hut with an untreated bed net ( Np/N0 , Equation [8] ) $ . δ1 = 0 . 071 δ2 = 1 . 26 δ3 = 1 . 52lpproportion of mosquitoes that enter a hut with net type p that die ( Equation [2] ) $α1 = 0 . 63 α2 = 4 . 00kpproportion of mosquitoes that enter a hut with net type p that successfully feed and survive ( Equation [11] ) $θ1 = 0 . 02 θ2 = 3 . 32jpproportion of mosquitoes that enter a hut with net type p that exit without feeding1−lp−kpγprate of decay in insecticide activity ( in washes ) for net type p ( Equation [16] ) $μp = −2 . 36 ρp = −3 . 05fproportion of mosquitoes killed in pyrethroid + PBO bioassay ( Equation [4] ) *β1 = 3 . 41 , β2 = 5 . 88 , β3 = 0 . 78β1 = 2 . 53 β2 = 0 . 89τconstant used to centre the data to aid the fitting process0 . 5Relevant parameters previously estimated by Griffin et al . ( 2010 ) † and Walker et al . ( 2015 ) ‡k0proportion of mosquitoes that enter a hut with no bednet that successfully feed$0 . 70†Hysinsecticide activity half-life in years for a susceptible mosquito population$2 . 64†rMproportion of mosquitoes which exit the hut when LLIN has no insecticidal activity0 . 24†0 . 24‡0 . 24†-mean life expectancy ( days ) 7 . 6†7 . 6‡8 . 9†-proportion blood meals taken on humans without LLINs ( human blood index ) 0 . 92†0 . 71‡0 . 94†-proportion of bites taken on humans whilst they are in bed0 . 89†0 . 83‡0 . 90† The ability of a net to kill a mosquito will decrease over time as the quantity of insecticide active ingredient declines . The non-lethal protection provided by the LLIN may also decrease with the decay of the active ingredient and the physical degradation of the net ( i . e . the acquisition of holes ) . To fully capture the loss of efficacy of an LLIN requires a net durability survey to be carried out over multiple years . To our knowledge , no durability studies have been published in areas of high pyrethroid resistance nor using the new generation of LLINs with the addition of PBO . In the absence of these data , we use the results from experimental hut trials that washed the net prior to its use . These experimental huts give some indication of how mosquitoes react to the change in insecticide concentration , though they do not provide information on the physical durability of the net ( as holes in the net are artificially generated ) . For simplicity and following ( Griffin et al . , 2010 ) it is assumed that the killing activity of pyrethroid over time ( the half-life in years , denoted Hy ) is proportional to the loss of morbidity caused by washing ( the half-life in washes , Hw ) . A prior estimate of the half-life in years ( Mahama et al . , 2007 ) from a durability study of a non-PBO LLIN with susceptible mosquitoes ( Hys ) is then used to reflect changes caused by pyrethroid resistance by , [15] Hy=Hw/HwsHys where superscript s indicates the half-life in a fully susceptible mosquito population ( i . e . l1 = 1 ) . Note that if the newer PBO nets have better durability than standard LLINs then this will under estimate their additional benefit . Following Griffin et al . , 2010 it is assumed that the activity of the insecticide decays at a constant rate according to a decay parameter γp , which is related to the half-life by Hw=ln ( 2 ) /γp . To test whether the rate of decay changes with lp ( i . e . mosquito mortality caused by new standard and PBO LLINs ) the following equation was fit to M3 , [16]logit ( γp ) =μp+ρp ( lp−τ ) . Shape parameters μp and ρp are allowed to vary between net types . The proportion of mosquitoes repeating due to the LLIN decreases from a maximum , rp0 , to a non-zero level rM , reflecting the protection still provided by an LLIN that no longer has any insecticidal activity . For simplicity , it is assumed that the rate of decay from rp0 to rM is given by γp ( as the degradation of the net over time is unlikely to be recreated by washing ) . The full equations for the proportion of mosquitoes repeating , dying and successfully feeding at time t following LLIN distribution ( rp , dp and sp , respectively ) is given by , [17]rp= ( rp0−rM ) exp ( −γpt ) +rM[18]dp=dp0exp ( −γpt ) [19]sp=1−rp−dp . All models were fit using a Markov chain Monte Carlo sampling algorithm implemented in the programme OPENBUGS ( Lunn et al . , 2009 ) . This Bayesian method enabled measurement error to be incorporated in both the dependent and independent variables according to the number of mosquitoes sampled ( both in bioassays and hut trials ) . Uninformative priors were used for all parameters with the exception of the random effects variance parameters which were constrained to be positive ( though were still uninformative , see Source code 1 in the Supplementary Information for a full list of priors ) . Three Markov chains were initialized to assess convergence and the first 5000 Markov chain Monte Carlo iterations were discarded as burn in . Convergence was assessed visually and a total of 10 , 000 iterations were used to derive the posterior distribution for all parameters and to generate 95% Bayesian credible interval estimates for model fits . The models were compared using the deviance information criterion ( DIC ) where the smaller value indicates a better fit , and a difference of five deviance information criterion units is considered to be substantial ( Spiegelhalter et al . , 2002 ) . Equations [8] to [19] were fit simultaneously to M3 enable the impact of washed nets to contribute to the relationship between rp , dp and sp , through the decay function , γp , doubling the number of datapoints in the analysis . A direct comparison between net types is beyond the scope of this study . Only one study compared PermaNet 2 . 0 and PermaNet 3 . 0 at the same time and place as Olyset and Olyset Plus and this study did not conduct hut trials with washed LLINs . As the different nets were tested in areas with different levels of pyrethroid resistance ( in part because the low overall number of studies ) then the impact of resistance and net type cannot currently be disentangled . The public health benefit of PBO-LLINs will depend on the epidemiological setting in which they are deployed . This includes the baseline characteristics of the setting ( e . g . mosquito species , abundance and seasonality ) , history of malaria control interventions ( e . g . prior use of bednets , management of clinical cases ) and prevalence of insecticide resistance . The rate at which pyrethroid resistance has evolved is highly uncertain . It is likely that it first became evident through its use in agriculture and the relative contribution of vector control to the selection of resistance is unknown and will vary between sites . This makes it impossible to recreate the spread of resistant phenotypes in a particular setting and predict its cumulative public health impact without detailed longitudinal studies spanning decades ( which do not exist for malaria endemic regions ) . Instead the impact of pyrethroid resistance is estimated by assuming it arrives instantaneously at a given level . To generate a broadly realistic history of LLIN usage it is assumed that LLINs were introduced at a defined coverage at year zero and redistributed every three years to the same percentage of the human population ( Figure 1 ) . The mosquito population is assumed to be either A . gambiae s . s . , A . arabiensis or Anopheles funestus ( the three major vectors in Africa ) which are entirely susceptible to pyrethroids up until year 6 when pyrethroid resistance arrives instantaneously . The public health impact of resistance is then measured over the subsequent three years ( the average clinical incidence or entomological inoculation rate ( EIR ) between the years 6 and 9 ) and compared to a population where resistance did not arise . The impact of PBO LLINs is predicted by introducing them into the resistant population at the year 9 and then measuring over the subsequent 3 years . For simplicity , it is assumed that there is perennial transmission , no other type of vector control and that once introduced pyrethroid resistance remains constant . Though perennial transmission is unrealistic it is necessary in order to produce simple guidelines ( as there is a very high number of combinations of seasonal patterns , relative mosquito species abundance and timings of LLIN distribution campaigns ) . A sensitivity analysis with more realistic seasonal patterns shows the change in clinical incidence compared to the perennial transmission is relatively minor , in part because the LLINs are used over 3 yearly cycles and their decay in effectiveness is relatively slow . LLINs are initially distributed at time zero at random ( i . e . there was no targeting to those with the highest infection ) and from then on the same people receive them every campaign to ensure that coverage remains at the defined level ( i . e . the number of people with an LLIN would go up if the distribution was random each round ) . Realistic usage patterns are adopted to reflect higher coverage immediately after LLIN distribution . No other vector control is incorporated whilst 35% of clinical cases are assumed to receive treatment , 36% which receive an ACT ( estimated by averaging across Africa using data collated by Cohen et al . , [2012] ) . A full list of the parameters , their definitions and estimated values are given in Table 6 whilst all other parameters are taken from Griffin et al . ( 2010 ) and White et al . ( 2011 ) . To investigate how the uncertainty in mosquito behaviour and the impact of PBO influence model predictions , a full sensitivity analysis is carried out for the parameters determining LLIN efficacy . A thousand parameter sets for α1 , α2 , β1 , β2 , , δ1 , δ2 , θ1 , θ2 , μp and ρp are sampled from the posterior distribution and are used to generate a range of possible values for rp0 , sp0 , dp0 and γp ( Figure 4—figure supplement 5 ) . This allows uncertainty in all measurements ( such as the relationship between resistance and hut trial mortality ) to be propagated throughout the equations . These parameter sets are then included as runs within the full transmission dynamics model to unsure the full uncertainty in these data is represented and the 95% credible intervals for model outputs are then shown . Figure 2—source data 1–3 . Figure 2—source data 4 is hosted on Dryad ( doi:10 . 5061/dryad . 13qj2 )	In recent years , widespread use of insecticide-treated bednets has prevented hundreds of thousands cases of malaria in Africa . Insecticide-treated bednets protect people in two ways: they provide a physical barrier that prevents the insects from biting and the insecticide kills mosquitos that come into contact with the net while trying to bite . Unfortunately , some mosquitoes in Africa are evolving so that they can survive contact with the insecticide currently used on bednets . How this emerging insecticide resistance is changing the number of malaria infections in Africa is not yet clear and it is difficult for scientists to study . To help mitigate the effects of insecticide resistance , scientists are testing new strategies to boost the effects of bednets , such as adding a second chemical that makes the insecticide on bednets more deadly to mosquitoes . In some places , adding this second chemical makes the nets more effective , but in others it does not . Moreover , these doubly treated , or “combination” , nets are more expensive and so it can be hard for health officials to decide whether and where to use them . Now , Churcher et al . have used computer modeling to help predict how insecticide resistance might change malaria infection rates and help determine when it makes sense to switch to the combination net . Insecticide-treated bednets provide good protection for individuals sleeping under them until relatively high levels of resistance are achieved , as measured using a simple test . As more resistant mosquitos survive encounters with the nets , the likelihood of being bitten before bed or while sleeping unprotected by a net increases . This is expected to increase malaria infections . As bednets age and are washed multiple times , they lose some of their insecticide and this problem becomes worse . Churcher et al . also show that the combination bednets may provide some additional protection against resistant mosquitos and reduce the number of malaria infections in some cases . The experiments show a simple test could help health officials determine which type of net would be most beneficial . The experiments and the model Churcher et al . created also may help scientists studying how to prevent increased spread of malaria in communities where mosquitos are becoming resistant to insecticide-treated nets .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "epidemiology", "and", "global", "health" ]	2016	The impact of pyrethroid resistance on the efficacy and effectiveness of bednets for malaria control in Africa
Hippocampal oscillations are dynamic , with unique oscillatory frequencies present during different behavioral states . To examine the extent to which these oscillations reflect neuron engagement in distinct local circuit processes that are important for memory , we recorded single cell and local field potential activity from the CA1 region of the hippocampus as rats performed a context-guided odor-reward association task . We found that theta ( 4–12 Hz ) , beta ( 15–35 Hz ) , low gamma ( 35–55 Hz ) , and high gamma ( 65–90 Hz ) frequencies exhibited dynamic amplitude profiles as rats sampled odor cues . Interneurons and principal cells exhibited unique engagement in each of the four rhythmic circuits in a manner that related to successful performance of the task . Moreover , principal cells coherent to each rhythm differentially represented task dimensions . These results demonstrate that distinct processing states arise from the engagement of rhythmically identifiable circuits , which have unique roles in organizing task-relevant processing in the hippocampus . Neural oscillations arise from the temporal coordination of activity in organized networks of neurons ( Buzsáki and Draguhn , 2004 ) . The unique connectivity of a network constrains the number of distinct rhythmic profiles that its local circuits can manifest , and the input to the network at a given time dictates the rhythmic circuits that are engaged ( Cannon et al . , 2014 ) . The dynamics of rhythms can thus reflect fast-paced changes in the coordination of activity within local circuits during information processing . Changes in the oscillatory activity of the hippocampus , a brain structure important for memory function , occur as it processes information it receives from multiple brain regions ( Buzsáki and Draguhn , 2004; Cannon et al . , 2014; Colgin et al . , 2009; Schomburg , et al . , 2014; Lee et al . , 1994; Igarashi et al . , 2014 ) . By studying the interactions of hippocampal neurons with their rhythmic circuits , we gain insight into how single neuron activity is coordinated into the local circuit and systems level processes that support memory . Although great advances have been made in describing both single cell and rhythmic correlates of memory in hippocampal circuits , relatively few studies examine the interaction of these phenomena . The hippocampus exhibits a diversity of rhythms ( Cannon et al . , 2014; Buzsáki , 2002; Buzsáki and Freeman , 2015; Colgin and Moser , 2010 ) . The theta ( 4–12 Hz ) rhythm is a dominant rhythm in the hippocampus that engages both principal and interneuron cell types , and depends on inputs from the medial entorhinal cortex ( MEC ) and the medial septum ( Lee et al . , 1994; Buzsáki , 2002; Kocsis et al . , 1999; Montgomery et al . , 2009; Kubie et al . , 1990 ) . The hippocampus also exhibits oscillations in the beta and gamma frequency ranges that span from 15-150Hz ( Buzsáki and Freeman , 2015; Colgin and Moser , 2010; Kay and Freeman , 1998; Martin et al . , 2007; Gourevitch et al . , 2010; Buzsáki and Schomburg , 2015 ) . Changes in the prominence of these higher frequency oscillations can reflect changes in input from converging afferents . Specifically , slow and fast gamma oscillations in the CA1 region of the hippocampus are thought to arise from the influence of CA3 and MEC inputs , respectively ( Colgin et al . , 2009; Buzsáki and Schomburg , 2015; Schomburg et al . , 2014 ) . In addition , an intermediate beta frequency range in CA1 has been hypothesized to reflect inputs from the lateral entorhinal cortex ( LEC ) ( Igarashi et al . , 2014 ) . These higher frequency oscillations often occur concurrently with the theta4-12Hz rhythm , and previous studies suggest that coordination of cell activity within co-occurring rhythms produces nested levels of organization in the hippocampal network ( Colgin et al . , 2009; Harris et al . , 2003; Mizuseki et al . , 2009; Buzsáki , 2010 ) . Thus , the diverse rhythmic states observed in the hippocampus can reflect the coordination of distinct information processing . The rhythms in the hippocampus are governed by the neurons that constitute its circuits . The diffuse , local projections of the interneuron population place them in an ideal position to shape the rhythmic organization of the network in response to inputs received from a diverse array of afferents ( Freund and Buzsáki , 1996; Sik et al . , 1995 ) . Interneurons in the CA1 region differ greatly according to their thresholds of excitability , the decay time of their inhibition , and the subcellular compartments where they preferentially target principal cells ( Cannon et al . , 2014; Royer et al . , 2012; Roux et al . , 2014; Roux and Buzsáki , 2015 ) . This diversity enables the interneuron population to flexibly sculpt the oscillatory profile of the hippocampal network while simultaneously shaping principal cell activity ( Freund and Buzsáki , 1996; Sik et al . , 1995; Sik et al . , 1997 ) . As the hippocampus integrates dynamic input during behavior , the interneurons can flexibly engage the appropriate circuits , dictating how the hippocampus processes information . Thus , changes in oscillations can indicate that hippocampal circuits have undergone a shift in processing state . Such shifts in processing state can be observed through distinctive rhythmic dynamics in the hippocampus as it processes information during memory tasks . Transient increases in the amplitude of higher frequency beta and low gamma activity can be observed during the presentation of conditioned stimuli , suggesting that the hippocampus undergoes a change in processing state ( Igarashi et al . , 2014; Kay and Freeman , 1998; Gourevitch et al . , 2010; Berke et al . , 2008; Rangel et al . , 2015 ) . In addition , cross-frequency coupling in the hippocampus develops while learning context-guided odor-reward associations ( Tort et al . , 2009; 2010 ) , which occurs concurrently with the development of odor-place conjunctive encoding in hippocampal principal neurons ( Komorowski et al . , 2009 ) . Since the hippocampus exhibits distinctive rhythmic states during memory tasks , and several of them are tied to the onset of learning , these changes in oscillatory profiles could reflect circuit level processes supporting memory function . However , it remains unknown how the rhythmicity of hippocampal circuits relates to the activity of the constitutive neurons during memory processing . We investigated the extent to which rhythmic engagement of distinct cell types during a memory task could support the ability of the hippocampus to represent associations . In previous studies , it has been shown that single neurons in the CA3 and CA1 regions of the hippocampus develop activity that is selective for odors , odor port locations , and conjunctions of particular odors at specific locations ( odor-position selectivity ) ( Komorowski et al . , 2009 ) . We designed a novel task to spatially and temporally isolate the sampling of an olfactory cue from its behavioral outcome during a context-guided odor-reward association task . We then performed in vivo recordings of single cell and local field potential activity in the CA1 region of the rat hippocampus to characterize the relationship between individual neurons and local circuit dynamics . We observed changes in theta ( 4–12 Hz ) , beta ( 15–35 Hz ) , low gamma ( 35–55 Hz ) , and high gamma ( 65–90 Hz ) frequency power during odor sampling epochs when task-relevant information must be integrated for successful performance . Theta4-12Hz , beta15-35Hz , low gamma35-55Hz , and high gamma65-90Hz rhythms differentially recruited principal cells and interneurons during successful performance of the task , suggesting that the different frequency bands represent functionally distinct processing states . Notably , principal cell and interneuron entrainment to beta15-35Hz frequency oscillations were the most correlated with correct performance . We propose that the beta15-35Hz rhythm instigates a processing of information in the hippocampus that is distinct from the processing that occurs in theta4-12Hz , low gamma35-55Hz , and high gamma65-90Hz and that the presence of the beta15-35Hz rhythm signals a recruitment of cell activity that may be critical for memory function . We observed dynamic rhythmic activity during the nose poke interval . Prominent changes in amplitude were observed in the theta ( 4–12 Hz ) , beta ( 15–35 Hz ) , low gamma ( 35–55 Hz ) , and high gamma ( 65–90 Hz ) frequency ranges ( Figure 1a ( middle , bottom ) , b-c ) . For each frequency band , we determined whether amplitude changed over the course of the nose poke or differed according to behavioral outcome ( correct or incorrect ) . We performed a two-factor repeated measures ANOVA and found a significant main effect of time during the nose poke for all frequencies ( Figure 1e–h; time: repeated measures ANOVAtheta: d . f . = 5 , F= 10 . 32 , p<0 . 00001; repeated measures ANOVAbeta: d . f . = 5 , F= 23 . 87 , p<0 . 00001; repeated measures ANOVAlow gamma: d . f . = 5 , F= 17 . 34 , p<0 . 00001; repeated measures ANOVAhigh gamma: d . f . = 5 , F= 63 . 78 , p<0 . 00001 ) , and no main effect for outcome ( correct or incorrect ) in any frequency ( outcome: repeated measures ANOVAtheta: d . f . = 1 , F= 1 . 19 , p=0 . 2797 , n . s . ; repeated measures ANOVAbeta: d . f . = 1 , F ≈ 0 , p=0 . 9746 , n . s . ; repeated measures ANOVAlow gamma: d . f . = 1 , F = 1 . 32 , p=0 . 2546 , n . s . ; repeated measures ANOVAhigh gamma: d . f . = 1 , F = 0 . 08 , p=0 . 7747 , n . s . ) . These results indicate that while all four frequencies demonstrated significant changes in amplitude over the course of the nose poke , mean amplitudes were not significantly different across correct and incorrect trial types . However , we observed a significant interaction effect in the low gamma35-55Hz frequency range , due to increased low gamma35-55Hz amplitude during correct trials during the last second of the odor-sampling interval ( time x outcome: repeated measures ANOVAtheta: d . f . = 5 , F= 0 . 34 , p=0 . 8886 , n . s . ; repeated measures ANOVAbeta: d . f . = 5 , F = 1 . 46 , p=0 . 2008 , n . s . ; repeated measures ANOVAlow gamma: d . f . = 5 , F = 4 . 32 , p=0 . 0008; repeated measures ANOVAhigh gamma: d . f . = 5 , F = 0 . 40 , p=0 . 8513 , n . s . ) . This increase in low gamma35-55Hz amplitude at the end of the nose poke during Correct Trials Only is evident in the ratio of the spectrograms for correct and incorrect trials ( Figure 1d ) . This indicates that there is a change in processing over the course of the nose poke within low gamma35-55Hz rhythmic circuits that differentiates between correct and incorrect trials . Together , these results indicate that the nose poke interval contains a shift in processing state in the hippocampus , which is observable through the onset of changes in rhythmic circuits . Populations of interneurons exhibited strong spike-phase coherence to the rhythms present during odor sampling . To test whether single cell entrainment to theta4-12Hz , beta15-35Hz , low gamma35-55Hz , or high gamma65-90Hz frequency ranges was related to successful performance of the associative memory task , we first examined whether interneuron spike-phase coherence to each frequency range during the odor sampling interval was selective to correct or incorrect trial types . This interval was initiated by a nose poke , and continued as the poke was sustained for 1 . 5 s , when the rat committed to a decision . The interneurons ( N = 67 , 45 sessions with each half-session analyzed separately , 6 rats , see Materials and methods ) were categorized as exhibiting significant spike-phase coherence to a given frequency range during Correct Trials Only , Incorrect Trials Only , or All ( both correct and incorrect ) Trials ( Figure 2a ) . If single cell engagement in a rhythm in the form of spike-phase coherence is important for successful processing during the task , one might expect a larger number of cells to be coherent during Correct Trials Only . The converse might be true if single cell spike-phase coherence was to interfere with successful performance , resulting in a larger number of cells exhibiting significant spike-phase coherence during Incorrect Trials Only . Lastly , cells that exhibit significant spike-phase coherence to a rhythm on All Trials ( correct and incorrect ) might instead be engaged in underlying processes that are not task-specific . For each rhythm , the proportions of interneurons in each category were compared to the proportions that would be expected if the cells were equally distributed across all three categories . This comparison thus asks whether the number of cells exhibiting significant spike phase coherence to a rhythm is different across the three performance categories . 10 . 7554/eLife . 09849 . 005Figure 2 . Interneuron and principal cell engagement in rhythmic circuits is related to task performance . ( a ) Proportions of interneurons demonstrating significant spike-phase coherence to theta4-12Hz ( far left ) , beta15-35Hz ( middle left ) , low gamma35-55Hz ( middle right ) , and high gamma65-90Hz ( far right ) during Correct Trials Only ( green ) , Incorrect Trials Only ( red ) , or All Trials ( gray ) . The largest proportion of theta4-12Hz coherent interneurons ( far left ) was coherent during All Trials , regardless of outcome . In contrast , the largest proportion of beta15-35Hz coherent interneurons ( middle left ) was coherent selectively during Correct Trials Only . ( b ) Same as in a , for the principal cell population . For each rhythm , the largest proportions of principal cells were coherent during Correct Trials Only . ( c ) The number of interneurons coherent during Correct Trials Only as a ratio of the total number of interneurons coherent during correct trials ( # coherent during Correct Trials Only + # coherent during All Trials ) . ( d ) Same as in c , for the principal cell population . ( e ) The proportions of interneurons and principal cells coherent to each rhythm during Correct Trials Only , subdivided into the proportions exhibiting coherence to a single rhythm or multiple rhythms . While the interneuron population demonstrates flexible engagement into multiple rhythmic circuits during successful performance , principal cells are more often engaged in single rhythmic circuits . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 00510 . 7554/eLife . 09849 . 006Figure 2—source data 1 . The number of interneurons within each rhythmic category that were coherent to each possible combination of the four rhythms . The interneurons categorized first by significant spike-phase coherence to a given rhythm and then by coherence during a given performance category ( Correct Trials Only , Incorrect Trials Only , All Trials ) were further divided by their coherence to each possible combination of the four rhythms examined in this study . For the interneurons that exhibited significant spike-phase coherence to a given rhythm during All Trials , the distribution of their coherence to all possible combinations of rhythms is shown separately for correct and incorrect trials . Interneurons coherent during All Trials often exhibited different profiles of engagement across the four rhythms during correct trials compared to incorrect trials . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 00610 . 7554/eLife . 09849 . 007Figure 2—source data 2 . The number of principal cells within each rhythmic category that were coherent to each possible combination of the four rhythms . The principal cells categorized first by significant spike-phase coherence to a given rhythm and then by coherence during a given performance category ( Correct Trials Only , Incorrect Trials Only , All Trials ) were further divided by their coherence to each possible combination of the four rhythms examined in this study . For the principal cells that exhibited significant spike-phase coherence to a given rhythm during All Trials , the distribution of their coherence to all possible combinations of rhythms is shown separately for correct and incorrect trials . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 007 Of the interneurons that exhibited significant spike-phase coherence to beta15-35Hz ( Figure 2a , middle left ) , the number that exhibited coherence to beta15-35Hz during Correct Trials Only was greater than the numbers coherent during Incorrect Trials Only or All Trials ( χ2beta ( 2 , N=66 ) = 51 . 54 , p<0 . 00001; post hoc pairwise comparisons with Bonferroni adjusted alpha: χ2correct v incorrect ( 1 , N=53 ) = 38 . 21 , p<0 . 00001; χ2correct v all ( 1 , N=62 ) = 20 . 90 , p<0 . 00001; χ2incorrect v all ( 1 , N=17 ) = 4 . 77 , p=0 . 029 , n . s . ) . Similarly , the number of interneurons coherent to high gamma65-90Hz ( Figure 2a , far right ) during Correct Trials Only was greater than the numbers coherent during Incorrect Trials Only or All Trials , with a larger number of interneurons coherent during All Trials than during Incorrect Trials Only as well ( χ2high gamma ( 2 , N=107 ) = 59 . 23 , p<0 . 00001 ) , post hoc pairwise comparisons with Bonferroni adjusted alpha: χ2correct v incorrect ( 1 , N=71 ) = 59 . 51 , p<0 . 00001; χ2correct v all ( 1 , N=104 ) = 9 . 85 , p=0 . 00017; χ2incorrect v all ( 1 , N=39 ) = 27 . 92 , p<0 . 00001 ) . In contrast , the largest number of theta4-12Hz coherent interneurons ( Figure 2a , far left ) were coherent during All Trials , although a greater number of cells still exhibited coherence during Correct Trials Only compared to Incorrect Trials Only ( χ2theta ( 2 , N=126 ) = 80 . 19 , p<0 . 00001 , post hoc pairwise comparisons with Bonferroni adjusted alpha: χ2correct v incorrect ( 1 , N=42 ) = 34 . 38 , p<0 . 00001; χ2correct v all ( 1 , N=124 ) = 15 . 61 , p=0 . 00007; χ2incorrect v all ( 1 , N=86 ) = 78 . 19 , p<0 . 00001 ) . Lastly , the numbers of interneurons coherent to low gamma35-55Hz ( Figure 2a , middle right ) during Correct Trials Only and All Trials were greater than the number coherent during Incorrect Trials Only , but were not significantly different from each other ( χ2low gamma ( 2 , N=91 ) = 37 . 21 , p<0 . 00001 ) , post hoc pairwise comparisons with Bonferroni adjusted alpha: χ2correct v incorrect ( 1 , N=49 ) = 37 . 74 , p<0 . 00001; χ2correct v all ( 1 , N=88 ) = 0 . 18 , p=0 . 6697 , n . s . ; χ2incorrect v all ( 1 , N=45 ) = 33 . 80 , p<0 . 00001 ) . In summary , while the proportion of interneurons exhibiting coherence during Correct Trials Only or All Trials varies across each of the four rhythms , coherence exclusively during incorrect trials is quite rare . Moreover , the heterogeneity across rhythms indicates that each rhythmic circuit uniquely engages interneurons in processing states that differentially contribute to task performance . To determine whether any of the rhythms are unique in their ability to engage interneuron activity during specific trial types , we also compared the distribution of interneurons across the three performance categories for all rhythms . The interneurons coherent to theta4-12Hz were distributed differently across the three performance categories than the interneurons coherent to beta15-35Hz , low gamma35-55Hz , or high gamma65-90Hz ( χ2theta-beta ( 2 , N=192 ) = 38 . 56 , p<0 . 00001; χ2theta-low gamma ( 2 , 217 ) = 9 . 21 , p=0 . 009; χ2theta-high gamma ( 2 , N=233 ) = 25 . 28 , d . f . = 2 , p<0 . 00001 ) . Post hoc pairwise comparisons revealed that these differences were driven by the relative proportions of interneurons in the Correct Trials Only and All Trials categories , while similar proportions were observed in the Incorrect Trials Only category across rhythms ( theta-beta: χ2correct ( 1 , N=192 ) = 31 . 46 , p<0 . 00001 , χ2incorrect ( 1 , N=192 ) = 2 . 86 , p=0 . 0906 , n . s . , χ2all ( 1 , N=192 ) = 38 . 23 , p<0 . 00001; theta-low gamma: χ2correct ( 1 , N=217 ) = 7 . 81 , p=0 . 0052 , χ2incorrect ( 1 , N=217 ) = 0 . 69 , p=0 . 4075 , n . s . , χ2all ( 1 , N=217 ) = 9 . 13 , p=0 . 0025; theta-high gamma: χ2correct ( 1 , N=233 ) = 23 . 54 , p<0 . 00001 , χ2incorrect ( 1 , N=233 ) = 0 . 41 , p=0 . 5230 , n . s . , χ2all ( 1 , N=233 ) = 25 . 26 , p<0 . 00001; Bonferroni adjusted alpha ) . Thus , interneuron coherence during All Trials occurs more often in the theta4-12H rhythm , distinguishing it from other rhythms . In addition , interneurons coherent to low gamma35-55Hz were distributed differently across the three performance categories than the interneurons coherent to beta15-35Hz ( χ2beta-low gamma ( 2 , N=157 ) = 11 . 85 , p=0 . 003 ) , due to a greater degree of selectivity in the beta15-35Hz coherent population for engagement during Correct Trials Only ( χ2correct ( 1 , N=157 ) = 8 . 99 , p=0 . 003 , χ2incorrect ( 1 , N=157 ) = 0 . 69 , p=0 . 4075 , n . s . , χ2all ( 1 , N=157 ) = 11 . 77 , p=0 . 0006; Bonferroni adjusted alpha ) . The distributions across the three performance categories were not significantly different between beta15-35Hz and high gamma65-90Hz coherent interneurons ( χ2beta-high gamma ( 2 , N=173 ) = 4 . 56 , p=0 . 1021 , n . s . ) or between low gamma35-55Hz and high gamma65-90Hz coherent interneurons ( χ2low gamma-high gamma ( 2 , N=198 ) = 3 . 44 , d . f . = 2 , p=0 . 1793 , n . s . ) . To better illustrate differences observed across rhythms ( Figure 2c ) , we plotted the ratio of the number of interneurons coherent during Correct Trials Only to the total number that exhibited coherence during correct trials ( the combined Correct Trials Only and All Trials categories ) . These results indicate that interneuron engagement in certain rhythms can be differentially dependent upon task performance . Notably , for each of the four rhythms , the smallest number of interneurons exhibited significant spike-phase coherence during Incorrect Trials Only . This decrease in interneuron spike-phase coherence during incorrect trials can also be observed by comparing the magnitude of coherence for the interneurons during correct and incorrect trials . Adjusting for firing rate differences between trial types ( Figure 3a , b , see Materials and methods ) , we observed significant decreases in the strength of interneuron spike-phase coherence to each rhythm during incorrect trials when compared to correct trials ( Median ( Mdn ) theta-correct = 0 . 1884 , Mdntheta-incorrect = 0 . 0998 , Wilcoxon signed-rank test Z = 4 . 76 , p<0 . 00001; Mdnbeta-correct = 0 . 0883 , Mdnbeta-incorrect = 0 . 0223 , Wilcoxon signed-rank test Z = 9 . 25 , p<0 . 00001; Mdnlow gamma-correct = 0 . 0906 Mdnlow gamma-incorrect = 0 . 0317 , Wilcoxon signed-rank test Z = 7 . 27 , p<0 . 00001; Mdnhigh gamma-correct = 0 . 0830 Mdnhigh gamma-incorrect = 0 . 0221 , Wilcoxon signed-rank test Z = 8 . 28 , p<0 . 00001 ) . Since higher firing rates in phase-modulated cells can increase estimates of spike-phase coherence strength , we determined whether firing rate differences between correct and incorrect trials could explain the differences in selective coherence . If the decrease in coherence during Incorrect Trials Only is due to lower firing rates during incorrect trials , then we would observe significantly lower firing rates during incorrect trials compared to correct trials . To the contrary , we observed that interneurons exhibited significantly higher firing rates during incorrect trials than correct trials ( Figure 3c; Mdncorrect = 12 . 78 Hz , Mdnincorrect = 14 . 20 Hz , Wilcoxon signed-rank test Z = -3 . 22 , p=0 . 0013 ) . Thus , the lack of interneuron engagement during Incorrect Trials Only is not due to firing rate differences between correct and incorrect trial types . Instead , unsuccessful processing during the task coincides with a unique inability to engage the interneuron population in rhythmic circuits . Taken together , these results suggest that trial outcome is strongly related to interneuron engagement in each of the four rhythms . 10 . 7554/eLife . 09849 . 008Figure 3 . The strength of interneuron coherence to each rhythm is greater during correct trials than incorrect trials . ( a ) The proportions of interneurons exhibiting a given magnitude of coherence on the x-axis during correct ( green ) and incorrect ( red ) trials with respect to the theta4-12Hz , beta15-35Hz , low gamma35-55Hz , or high gamma65-90Hz rhythm . Greater proportions of interneurons exhibit larger magnitudes of coherence during correct trials compared to incorrect trials . ( b ) The magnitude of coherence during correct trials plotted against the magnitude of coherence during incorrect trials for all interneurons that were coherent to each rhythm during either correct or incorrect trials . ( c ) The average firing rate during correct trials plotted against the average firing rate during incorrect trials for all interneurons that were coherent to each rhythm during either correct or incorrect trials . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 00810 . 7554/eLife . 09849 . 009Figure 3—figure supplement 1 . The phase of interneuron coherence to each rhythm during correct and incorrect trials . ( a ) The magnitude and phase of coherence during correct trials , for every interneuron that exhibited coherence to a given rhythm during either correct or incorrect trials . Each cell is represented by a single arrow , with the length of the arrow representing the magnitude of coherence and the direction indicating the preferred phase of spiking . Interneurons that exhibited significant spike-phase coherence during correct trials ( Correct Trials Only and All Trial categories ) are shown in red , while cells that exhibited significant spike-phase coherence during Incorrect Trials Only are shown in black . ( b ) Same as in a , for incorrect trials . In this case , the interneurons that exhibited significant spike-phase coherence during incorrect trials ( Incorrect Trials Only and All Trial categories ) are shown in red , while cells that exhibited significant spike-phase coherence during Correct Trials Only are shown in black . ( c ) The preferred phase of spiking during correct trials plotted against the preferred phase of spiking during incorrect trials for interneurons that exhibited significant spike-phase coherence during Correct Trials Only ( red ) . Gray points indicate circular repetitions of the data in red , to better visualize the circular nature of the data . Circular correlations were performed on the preferred phases observed across the population during correct and incorrect trials and are indicated at the top of each panel ( d ) Same as in c , for the interneurons exhibiting significant spike-phase coherence during All Trials . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 009 To examine whether performance dependent engagement of the interneurons coincides with a rhythmic phase preference , we compared their average phase of spiking during correct and incorrect trial types ( Figure 3—figure supplement 1 , see Materials and methods ) . If engagement in a rhythm during Correct Trials Only represents participation in a rhythmic processing state that occurs uniquely during successful task performance , then the phase of interneuron spiking activity may change between correct and incorrect trials . In addition , interneurons exhibiting coherence to a rhythm during All Trials might exhibit the same phase preference during correct and incorrect trials because their participation is not related to successful task performance . We first tested the interneurons that exhibited significant spike-phase coherence to a specific rhythm during Correct Trials Only by performing circular correlations on their preferred ( average ) phase during correct and incorrect trials . Interneurons that were coherent to theta4-12Hz , beta15-35Hz , and high gamma65-90Hz during Correct Trials Only did not exhibit a consistent phase preference between correct and incorrect trials ( Figure 3—figure supplement 1c; Rtheta-correct = 0 . 26 , p=0 . 0823; Rbeta-correct = 0 . 12 , p=0 . 4003; Rhigh gamma-correct = 0 . 16 , p=0 . 2012 ) . Interneurons that were coherent to low gamma35-55Hz during Correct Trials Only exhibited only a weak correlation in phase preference between correct and incorrect trials ( Rlow gamma-correct = 0 . 31 , p=0 . 0258 ) . Overall , the interneurons that were coherent during Correct Trials Only did not exhibit similar engagement across trial types as measured by spike-phase coherence . Given that the magnitude of coherence is greater across this population during correct trials , together these results suggest that there is a critical reorganization of spike timing in these interneurons during successful processing in the hippocampus . In stark contrast , interneurons that were coherent to theta4-12Hz , beta15-35Hz , low gamma35-55Hz , and high gamma65-90Hz during All Trials exhibited relatively strong correlations between the average phases observed across the population during correct and incorrect trials ( Figure 3—figure supplement 1d; Rtheta-all = 0 . 82 , p<0 . 00001; Rbeta-all = 0 . 55 , p=0 . 0402; Rlow gamma-all = 0 . 41 , p=0 . 0131; Rhigh gamma-all = 0 . 66 , p=0 . 0002 ) . Thus the interneurons that are coherent during All Trials also have similar phases of entrainment during correct and incorrect trials , suggesting that they are similarly engaged in processing within local rhythmic circuits irrespective of whether the rat successfully performs the task . Combined with evidence that each rhythm demonstrates a unique ability to engage interneurons across correct and incorrect trials ( Figure 2a ) , these results demonstrate that each rhythmic circuit differentially participates in task-related processing as demonstrated by the selective entrainment of interneuron spike timing . Spike-phase coherence analyses were then performed on the principal cell population . All principal cells ( N = 1301 , 45 sessions with each half-session analyzed separately , 6 rats ) were categorized as exhibiting significant spike-phase coherence to a given frequency range during Correct Trials Only , Incorrect Trials Only , or All Trials ( Figure 2b ) . For all rhythms , principal cells preferentially exhibited significant spike-phase coherence during Correct Trials Only ( theta: χ2theta ( 2 , N=349 ) = 266 . 86 , p<0 . 00001 , post hoc pairwise comparisons with Bonferroni adjusted alpha: χ2correct v incorrect ( 1 , N=298 ) = 165 . 38 , p<0 . 00001; χ2correct v all ( 1 , N=311 ) = 140 . 45 , p<0 . 00001; χ2incorrect v all ( 1 , N=89 ) = 1 . 90 , p=0 . 1682 , n . s . ; beta: χ2beta ( 2 , N=91 ) = 92 . 40 , p<0 . 00001 , post hoc pairwise comparisons with Bonferroni adjusted alpha: χ2correct v incorrect ( 1 , N=88 ) =38 . 23 , p<0 . 00001; χ2correct v all ( 1 , N=76 ) = 64 . 47 , p<0 . 00001; χ2incorrect v all ( 1 , N=18 ) = 8 . 00 , p=0 . 0047; low gamma: χ2low gamma ( 2 , N=120 ) = 132 . 65 , p<0 . 00001 , post hoc pairwise comparisons with Bonferroni adjusted alpha: χ2correct v incorrect ( 1 , N=116 ) = 57 . 97 , p<0 . 00001; χ2correct v all ( 1 , N=101 ) = 85 . 63 , p<0 . 00001; χ2incorrect v all ( 1 , N=21 ) = 8 . 05 , p=0 . 0045; high gamma: χ2high gamma ( 2 , N=134 ) = 132 . 65 , p<0 . 00001 , post hoc pairwise comparisons with Bonferroni adjusted alpha: χ2correct v incorrect ( 1 , N=128 ) = 66 . 13 , p<0 . 00001; χ2correct v all ( 1 , N=116 ) = 93 . 24 , p<0 . 00001; χ2incorrect v all ( 1 , N=24 ) = 6 . 00 , p=0 . 0143 ) . In addition , for every rhythm except theta4-12Hz ( Figure 2b , far left ) , the number of principal cells coherent during incorrect trials was significantly greater than the number coherent during All Trials . Since the entrainment of principal cells in rhythmic circuits occurs most often during Correct Trials Only , these data suggest that principal cell engagement is a unique feature of successful processing in the hippocampus . To determine whether any of the rhythms are unique in their ability to engage principal cell activity during specific trial types , we also compared the distributions of principal cells across the three performance categories for all pairs of rhythms . The distributions of principal cells across the three performance categories were not significantly different in beta15-35Hz , low gamma35-55Hz , or high gamma65-90Hz coherent principal cells ( χ2beta-low gamma ( 2 , N=211 ) = 0 . 216 , p=0 . 8976 , n . s . ; χ2beta-high gamma ( 2 , N=225 ) = 0 . 556 , p=0 . 7573 , n . s . ; χ2low gamma-high gamma ( 2 , N=254 ) = 0 . 237 , p=0 . 8883 , n . s . ) . However , the distribution of theta4-12Hz coherent principal cells was significantly different than the distributions in all other rhythms ( χ2theta-beta ( 2 , N=440 ) = 9 . 72 , p=0 . 0078; χ2theta-low gamma ( 2 , N=469 ) = 11 . 25 , p=0 . 0035; χ2theta-high gamma ( 2 , N=483 ) = 9 . 69 , p=0 . 0078 ) . Post hoc pairwise comparisons revealed that this is due to a larger number of theta coherent principal cells that were coherent during All Trials ( theta-beta: χ2correct ( 1 , N=440 ) = 1 . 28 , p=0 . 2573 , n . s . , χ2incorrect ( 1 , N=440 ) = 2 . 13 , p=0 . 1442 , n . s . , χ2all ( 1 , N=440 ) = 8 . 59 , p=0 . 0033; theta-low gamma: χ2correct ( 1 , N=469 ) = 3 . 18 , p=0 . 0744 , n . s . , χ2incorrect ( 1 , N=467 ) = 0 . 93 , p=0 . 3356 , n . s . , χ2all ( 1 , N=469 ) = 10 . 98 , p=0 . 0009; theta-high gamma: χ2correct ( 1 , N=483 ) = 3 . 11 , p=0 . 0777 , n . s . , χ2incorrect ( 1 , N=483 ) = 0 . 61 , p=0 . 4340 , n . s . , χ2all ( 1 , N=483 ) = 9 . 56 , p=0 . 0020 ) . To better illustrate this difference in the theta4-12Hz coherent population compared to other rhythms ( Figure 2d ) , we plotted the ratio of principal cells coherent during Correct Trials Only to the total number coherent during correct trials ( the combined Correct Trials Only and All Trials categories ) . We demonstrate that principal cells exhibiting coherence to correct trials are almost exclusively coherent to Correct Trials Only for beta15-35Hz , low gamma35-55Hz , and high gamma65-90Hz , while the theta4-12Hz coherent population exhibits a significantly greater number of neurons that are coherent to All Trials . In summary , these results suggest that rhythmic entrainment of principal neurons to beta15-35Hz , low gamma35-55Hz , and high gamma65-90Hz is closely related to processing that is specific to the memory task , while theta4-12Hz rhythmic entrainment serves a less performance-specific function . Similar to the interneuron population , the smallest number of principal cells exhibited significant spike-phase coherence during Incorrect Trials Only in each of the four rhythms . This decrease in principal cell spike-phase coherence during incorrect trials was also observed through significant decreases in the strength of principal cell spike-phase coherence to each rhythm during incorrect trials when compared to correct trials , adjusted for firing rate differences between trial types ( Figure 4a and b , Median ( Mdn ) theta-correct = 0 . 0499 , Mdntheta-incorrect = 0 . 0356 , Wilcoxon signed-rank test Z = 6 . 69 , p<0 . 00001; Mdnbeta-correct = 0 . 0351 , Mdnbeta-incorrect = 0 . 0172 , Wilcoxon signed-rank test Z = 9 . 85 , p<0 . 00001; Mdnlow gamma-correct = 0 . 0231 , Mdnlow gamma-incorrect = 0 . 0137 , Wilcoxon signed-rank test Z = 7 . 67 , p<0 . 00001; Mdnhigh gamma-correct = 0 . 0190 , Mdnhigh gamma-incorrect = 0 . 0126 , Wilcoxon signed-rank test Z = 7 . 16 , p<0 . 00001 ) . To ensure that these changes in coherence strength were not due to decreases in principal cell firing rates during incorrect trials , we compared the firing rates across trial types and found no significant differences between correct and incorrect trials ( Figure 4c; Mdncorrect = 0 . 6000 Hz , Mdnincorrect = 0 . 4444 Hz , Wilcoxon signed-rank test = 0 . 5608 , p=0 . 5750 ) . These results further demonstrate that the manner in which principal cells are engaged in each of the four rhythms is strongly related to task performance . 10 . 7554/eLife . 09849 . 010Figure 4 . The strength of principal cell coherence to each rhythm is greater during correct trials than incorrect trials . ( a ) The proportion of principal cells exhibiting a given magnitude of coherence to theta4-12Hz , beta15-35Hz , low gamma35-55Hz , or high gamma65-90Hz during correct ( green ) and incorrect ( red ) trials . Greater proportions of principal cells exhibit larger magnitudes of coherence during correct trials compared to incorrect trials . ( b ) The magnitude of coherence during correct trials plotted against the magnitude of coherence during incorrect trials for all principal cells that were coherent to each rhythm during either correct or incorrect trials . ( c ) The average firing rate during correct trials plotted against the average firing rate during incorrect trials for all principal cells that were coherent to each rhythm during either correct or incorrect trials . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 01010 . 7554/eLife . 09849 . 011Figure 4—figure supplement 1 . The phase of principal cell coherence to each rhythm during correct and incorrect trials . ( a ) The magnitude and phase of coherence during correct trials , for every principal cell that exhibited coherence to a given rhythm during either correct or incorrect trials . Each cell is represented by a single arrow , with the length of the arrow representing the magnitude of coherence and the direction indicating the preferred phase of spiking . Principal cells that exhibited significant spike-phase coherence during correct trials ( Correct Trials Only and All Trial categories ) are shown in red , while cells that exhibited significant spike-phase coherence during Incorrect Trials Only are shown in black . ( b ) Same as in a , for incorrect trials . In this case , the principal cells that exhibited significant spike-phase coherence during incorrect trials ( Incorrect Trials Only and All Trial categories ) are shown in red , while cells that exhibited significant spike-phase coherence during Correct Trials Only are shown in black . ( c ) The preferred phase of spiking during correct trials plotted against the preferred phase of spiking during incorrect trials for principal cells that exhibited significant spike-phase coherence during Correct Trials Only ( red ) . Gray points indicate circular repetitions of the data in red , to better visualize the circular nature of the data . Circular correlations were performed on the preferred phases observed across the population during correct and incorrect trials and are indicated at the top of each panel ( d ) Same as in c , for the principal cells exhibiting significant spike-phase coherence during All Trials . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 011 To examine whether performance dependent engagement of the principal cells coincides with a rhythmic phase preference , we compared the average phase of spiking during correct and incorrect trial types for each neuron ( Figure 4—figure supplement 1 , see Materials and methods ) . The principal cells coherent to theta4-12 Hz and high gamma65-90 Hz during Correct Trials Only did not exhibit a consistent phase preference across correct and incorrect trial types ( Figure 4—figure supplement 1c; Rtheta-correct = -0 . 05 , p=0 . 4107; Rhigh gamma-correct = -0 . 10 , p=0 . 2239 ) , suggesting that these principal cells are engaged by theta4-12 Hz and high gamma65-90 Hz rhythmic circuits differently on correct and incorrect trials . For the principal cells coherent to beta15-35 Hz during Correct Trials Only , the preferred phase during correct trials was anti-correlated to the average phase during incorrect trials ( Rbeta-correct = -0 . 35 , p=0 . 0027 ) , indicating that the preferred phase of entrainment during correct trials in this population was often opposite ( i . e . separated by 180 degrees ) to the average phase during incorrect trials . In contrast , the principal cells coherent to low gamma35-55 Hz during Correct Trials Only exhibited similar phases across correct and incorrect trials ( Rlow gamma-correct = 0 . 35 , p=0 . 0006 ) , indicating that there is some similarity in how the spike timing of these cells relates to the low gamma35-55 Hz oscillation during both trial types despite a lack of significant coherence during incorrect trials . This suggests that the low gamma35-55 Hz rhythmic circuit tends to engage its principal cells at a specific phase irrespective of trial outcome , while the magnitude of this engagement is a stronger predictor of correct performance . For the relatively few principal cells that were coherent to theta4-12 Hz , beta15-35 Hz , low gamma35-55 Hz , and high gamma65-90 Hz during All Trials , the preferred phase angles during correct trials were not correlated to the preferred phase angles during incorrect trials ( Figure 4—figure supplement 1d; Rtheta-all = -0 . 12 , p=0 . 3693; Rbeta-all = -0 . 89 , p=0 . 1747; Rlow gamma-all = 0 . 07 , p=0 . 8175; Rhigh gamma-all = -0 . 05 , p=0 . 8751 ) , indicating that these principal cells are engaged by rhythmic circuits differently on correct and incorrect trials . These results further demonstrate that principal cells are distinctly engaged in rhythmic circuits when the rat is effectively processing information in order to correctly respond in the task . We then examined the extent to which interneurons and principal cells were exclusively coherent to one rhythm or flexibly engaged in many . In our previous analyses , cells were identified as coherent to one rhythm without regard for its coherence to the other three frequency ranges examined . It is possible that many of the cells coherent to beta15-35 Hz , for example , are also coherent to theta4-12 Hz , low gamma35-55 Hz , and high gamma65-90H . To examine the extent to which cells categorized as coherent to a specific rhythm are actually engaged in many , we first identified the number of interneurons and principal cells coherent to theta4-12 Hz , beta15-35 Hz , low gamma35-55 Hz , or high gamma65-90 Hz that were either coherent to only one rhythm or multiple rhythms ( Figure 2e ) . As the strength of coherence for both interneurons and principal cells is notably lower during incorrect trials , this analysis was restricted to Correct Trials Only . Tests for differences between the interneuron and principal cell populations revealed that interneurons were significantly more engaged in multiple rhythms than the principal cell population in each rhythmic category ( χ2theta ( P v I ) ( 1 , N=300 ) = 49 . 65 , p<0 . 00001; χ2beta ( P v I ) ( 1 , N=122 ) = 33 . 40 , p<0 . 00001; χ2low gamma ( P v I ) ( 1 , N=145 ) = 26 . 56 , p<0 . 00001; χ2high gamma ( P v I ) ( 1 , N=178 ) = 31 . 17 , p<0 . 00001; for the interneurons and principal cells engaged in multiple rhythms , see the specific combination of rhythms in Figure 2—source data 1 and 2 , respectively ) . Thus , interneurons often participate in multiple types of rhythmic circuits , whereas principal cells often exhibit engagement limited to a single rhythm . These results suggest that different mechanisms may be involved in engaging interneurons and principal cells in rhythmic circuits , leading to the capacity for interneurons to readily participate in multiple rhythmic processing states . Since the majority of interneurons are coherent to multiple rhythms during correct trials , it is possible that engagement in specific combinations of rhythms is important for successful performance of the task . To investigate this question , we determined the number of interneurons coherent to every combination of the four rhythmic categories during correct and incorrect trials ( Figure 5a ) . Interneurons were not equally distributed across the fifteen combinations of the four rhythms during correct trials ( χ2correct ( 14 , N=134 ) = 326 . 97 , p<0 . 00001 ) , with the largest number of the interneurons coherent to all four rhythms . In contrast , during incorrect trials , the interneurons were unequally distributed such that they were most often coherent to theta4-12 Hz only ( χ2incorrect ( 14 , N=100 ) = 203 . 30 , p<0 . 00001 ) . The distribution of interneurons across the fifteen rhythmic categories was significantly different during correct and incorrect trial types ( χ2correct v incorrect ( 14 , N=234 ) = 52 . 46 , p<0 . 00001 ) . The most striking differences were observed in the proportion of interneurons coherent to theta only and the proportion of interneurons coherent to all four rhythms ( χ2theta only ( 1 , N=234 ) = 18 . 99 , p<0 . 00001; χ2all rhythms ( 1 , N=234 ) = 21 . 67 , p<0 . 00001 ) . These results indicate that interneuron engagement in all four rhythms is strongly related to successful performance of the task , and suggests that interneuron engagement solely in a theta rhythmic circuit is a marker of a processing state that is maladaptive for good performance in our task . 10 . 7554/eLife . 09849 . 012Figure 5 . Profiles of interneuron and pyramidal cell recruitment into rhythmic circuits during correct and incorrect performance . ( a ) The proportions of interneurons coherent during correct trials ( top ) or incorrect trials ( bottom ) that were coherent to each combination of the four rhythms examined in this study ( theta4-12 Hz , beta15-35 Hz , low gamma35-55 Hz , and high gamma65-90 Hz ) . While a large proportion of interneurons demonstrate coherence to all four frequency ranges during correct trials , interneurons are often engaged in a single rhythmic circuit during incorrect trials . ( b ) Same as in a , for principal cells . The majority of principal cells exhibited coherence to a single rhythmic circuit during both correct and incorrect trial types . To view the number of neurons coherent during correct trials as a ratio of all the neurons coherent in each rhythmic category see Figure 5—figure supplement 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 01210 . 7554/eLife . 09849 . 013Figure 5—figure supplement 1 . The ratio of neurons coherent during correct trials to all coherent neurons in each rhythmic category . ( a ) The number of interneurons coherent during correct trials as a proportion of the numbers coherent during correct trials or incorrect trials for each rhythmic category . ( b ) Same as in a , for principal cells . ( c ) The proportions of interneurons coherent during correct trials ( top ) or incorrect trials ( bottom ) that were coherent to each combination of the four rhythms examined in this study , from Figure 5 for reference . ( d ) Same as in c , for principal cells . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 01310 . 7554/eLife . 09849 . 014Figure 5—figure supplement 2 . Correlations of interneuron coherence for different pairs of rhythms . Correlations between the magnitude of coherence to one frequency and the magnitude of coherence to another frequency across correct trials for interneurons that exhibited coherence to all rhythms . Each point represents the correlation value for a single interneuron . To better visualize the distribution of points , normalized 2D histograms surround the data points for every pair of frequencies . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 01410 . 7554/eLife . 09849 . 015Figure 5—figure supplement 3 . Amplitude correlations for different pairs of rhythms . ( a ) Mean correlations between the amplitudes for every pair of frequencies , averaged across sessions . Theta4-12 Hz and low gamma35-55 Hz are often anti-correlated , while beta15-35 Hz is often strongly correlated with theta4-12 Hz and low gamma35-55 Hz . ( b ) The proportion of sessions exhibiting a given correlation value on the x-axis for every pair of frequencies . These correlation values were used to derive the averages in a . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 015 Principal cells were also not equally distributed across the fifteen combinations of the four rhythms during correct trials ( Figure 5b , χ2correct ( 14 , N=453 ) = 1162 . 70 , p<0 . 00001 ) , with the largest numbers of principal cells coherent to single rhythms and more specifically to theta4-12 Hz only . A similar bias in the distribution was observed during incorrect trials ( χ2correct ( 14 , N=131 ) = 518 . 35 , p<0 . 00001 ) , with no significant differences across the fifteen rhythmic categories between correct and incorrect trial types . Thus , whereas larger numbers of principal cells are engaged in rhythmic activity during correct trials ( Figure 2b ) , co-participation in multiple rhythms is not related to task performance ( χ2correct v incorrect ( 14 , N=584 ) = 12 . 13 , p=0 . 5958 ) . While the interneurons engaged in all four rhythms over the course of the session may have been independently engaged by each rhythmic circuit , it is possible that some rhythmic circuits operate cooperatively . If the latter possibility were the case , we would expect the coherence of the interneurons to be similarly modulated on each trial by our rhythms of interest . To examine whether interneurons were engaged in combinations of rhythms at similar times during the nose poke , we performed multi-taper spike-phase coherence analysis to acquire the magnitude of coherence at four frequencies ( 7 Hz , 20 Hz , 45 Hz , and 75 Hz ) within each of the four bands used in this study for every correct trial . We then asked whether the magnitude of coherence to one frequency was correlated with the magnitude of coherence to another frequency across trials by performing correlations on the coherence values across trials for every pair of frequencies ( Figure 5—figure supplement 2 ) . There were no significant differences in the correlations values obtained for any pair of rhythms for the subpopulation of interneurons that were coherent to all four rhythms ( one-way repeated measures ANOVA , N=53 , d . f . = 5 , F= 1 . 59 , p=0 . 1648 ) . Although we are open to the possibility that there is coordinated ( e . g . correlated or anti-correlated ) coherence to multiple rhythms , there do not appear to be any obvious trends in our data . We also attempted to examine whether the amplitudes of the four rhythms were co-modulated during correct trial nose pokes . To do this , we examined amplitude dynamics in the local field potential alone , independently of the spiking activity . For every session , we calculated the instantaneous amplitude of each of the four frequency ranges during correct trials and performed correlations on the amplitude values for every pair of frequencies ( Figure 5—figure supplement 3a , see Materials and methods ) . We observed significant differences in the correlations between different pairs of frequencies ( one-way ANOVA , d . f . = 5 , F= 42 . 06 , p<0 . 00001 ) . Post hoc pairwise comparisons revealed that some pairs of frequencies appear to be more correlated or anti-correlated than others ( for the results from all pairwise comparisons , see Figure 5—figure supplement 3a ) . Specifically , theta4-12 Hz and low gamma35-55 Hz amplitudes are more anti-correlated than all other pairs of frequencies ( Tukey’s Honest Significant Difference test , p<0 . 00001 , for all post hoc pairwise comparisons of theta4-12 Hz and low gamma35-55 Hz amplitude correlations against all other pairs of frequencies ) . In contrast , theta4-12 Hz and beta15-35 Hz amplitudes , as well as beta15-35H and low gamma35-55 Hz amplitudes were more correlated than most other pairs of frequencies ( Tukey’s Honest Significant Difference test , ptheta-beta v theta-low gamma < 0 . 00001 , ptheta-beta v theta-high gamma < 0 . 00001 , ptheta-beta v beta-low gamma = 0 . 248 , ptheta-beta v beta-high gamma = 0 . 020 , pbeta-low gamma v beta-high gamma = 0 . 676; pbeta-low gamma v theta-low gamma < 0 . 00001 , pbeta-low gamma v theta-high gamma < 0 . 00001 , pbeta-low gamma v beta-high gamma < 0 . 00001 , pbeta-low gamma v low gamma-high gamma = 0 . 003 ) . Correlations between high gamma65-90 Hz amplitude and other frequencies ( theta4-12 Hz , beta15-35 Hz , and low gamma35-55 Hz ) were lower than beta15-35 Hz and low gamma35-55 Hz correlations , indicating that co-modulation of frequencies with high gamma65-90 Hz is generally not as strong . These findings suggest that while theta4-12 Hz and low gamma35-55 Hz amplitudes tend to undergo opposing changes , they both are often co-modulated with beta15-35H amplitude at some point during correct nose pokes . Importantly , many pairs of frequencies appear correlated in some sessions and anti-correlated in others ( Figure 5—figure supplement 3b ) . Thus , while these results reveal that there are trends towards co-modulation of some frequencies , we did not observe strikingly robust co-modulation or mutual exclusivity for any pair of frequencies in this study . Consequently , it seems that each rhythmic circuit can operate independently , and therefore the engagement of interneurons into each rhythmic network also likely occurs independently . Principal cells in the hippocampus have been shown to exhibit activity that is highly selective to spatial positions , sensory stimuli , and the co-occurrence of these features during associative memory tasks ( Figure 6a , b , Figure 6—figure supplement 1–4 ) ( Komorowski et al . , 2009; O'Keefe and Dostrovsky , 1971; Eichenbaum et al . , 1999 ) . Consequently , we characterized the degree to which rhythmic populations of principal cells encode specific task dimensions . We calculated the information for odors , positions , or odor-position conjunctions contained in the spiking activity of theta4-12 Hz , beta15-35 Hz , low gamma35-55 Hz , or high gamma65-90 Hz coherent principal cells during three non-overlapping time intervals during correct trials: a baseline 500 ms interval prior to the nose poke ( before ) , a 500 ms directly after odor delivery ( odor ) , and 500 ms at the end of the nose poke when the rat committed to a decision ( end; Figure 6c–e , for information analysis see Materials and methods ) ( Markus , et al . , 1994 ) . Position information describes the extent to which the spiking activity of a neuron differentiates between the four possible odor port positions . Odor information describes the ability of the neuron to discriminate between odors ( A , B , C , and D ) . Odor-position information describes the extent to which neural activity was selective for the co-occurrence of a specific odor in a particular odor port location . A neuron was considered to have significant information for a task dimension if the information score exceeded the 95% confidence interval of 1000 scores calculated from trial-shuffled conditions . 10 . 7554/eLife . 09849 . 016Figure 6 . Principal cells exhibiting strong spike-phase coherence during the odor sampling contain information for task-relevant features . ( a ) Representative activity of a single CA1 principal cell during the odor sampling intervals of Correct Trials Only . Each row of tick marks represents spiking during a single trial . This cell demonstrates activity selective for position 2 . ( b ) Same as in a , showing the activity of a principal cell that is selective for the co-occurrence of odor A in position 2 . For additional examples of theta4-12 Hz , beta15-35 Hz , low gamma35-55 Hz , and high gamma65-90 Hz coherent principal cells , see Figure 6—figure supplement 1–4 . ( c ) Median information ( bits/spike ) in theta4-12 Hz , beta15-35 Hz , low gamma35-55 Hz , and high gamma65-90 Hz coherent populations for odors during a 500 ms interval prior to nose poke ( before ) , 500 ms directly after odor delivery ( odor ) , and 500 ms prior to the end of the nose poke when the rat committed to a decision ( end ) . Vertical gray bars indicate the inter-quartile range . The top vertical line indicates q3 + 1 . 5 x ( q3 – q1 ) and the bottom vertical line indicates q1 – 1 . 5 x ( q3 – q1 ) , where q1 and q3 are the 25th and 75th percentiles , respectively . Asterisks ( * ) indicate a significant pair-wise comparison using a Tukey’s Honest Significant Difference test , p<0 . 05 . ( d ) Same as in c , for position information . ( e ) Same as in c , for odor-position information . ( f ) Cartoon diagram indicating potential mechanisms for the generation of the beta15-35 Hz rhythm in CA1: 1 ) beta15-35 Hz rhythmic input is received from an upstream structure , 2 ) beta15-35 Hz is an internally generated rhythm , or 3 ) long-range communication across multiple interacting networks is facilitated by coordination in beta15-35 Hz by a third-party structure . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 01610 . 7554/eLife . 09849 . 017Figure 6—figure supplement 1 . Theta ( 4–12 Hz ) coherent principal cell . ( a ) Relative waveform shape and amplitude across tetrode wires and clustering of spike features . ( b ) Circular histogram representing the phase of theta4-12 Hz at the time of each spike with the direction of the mean resultant length vector ( R ) shown by the arrow in black . Histograms were calculated from spiking activity during odor sampling intervals , during correct trials only . ( c ) Spiking activity of the cell during odor sampling of correct odors . Pairs of odors are differentially rewarded depending upon the context in which they are presented . Each row of tick marks represents spiking during a single trial . This cell exhibits activity that is selective for Position 2 , for both odors rewarded in that position . The spiking activity of this cell contains significant information ( see Materials and methods ) for odors , positions , and odor-position combinations . Information scores are indicated in the top right . ( d ) Magnitude of the mean resultant length vector over the time course of nose pokes leading to a correct choice . The strength of theta4-12 Hz coherence is greatest at 0 . 75 s after nose poke onset . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 01710 . 7554/eLife . 09849 . 018Figure 6—figure supplement 2 . Beta ( 15–35 Hz ) coherent principal cell . ( a ) Relative waveform shape and amplitude across tetrode wires and clustering of spike features . ( b ) Circular histogram representing the phase of beta15-35 Hz at the time of each spike with the direction of the mean resultant length vector ( R ) shown by the arrow in black . Histograms were calculated from spiking activity during odor sampling intervals , during correct trials only . ( c ) Spiking activity of the cell during odor sampling of correct odors . Pairs of odors are differentially rewarded depending upon the context in which they are presented . Each row of tick marks represents spiking during a single trial . This cell exhibits activity that is selective for Position 4 , for both odors rewarded in that position . The spiking activity of this cell contains significant information ( see Materials and methods ) for odors , positions , and odor-position combinations . Information scores are indicated in the top right . ( d ) Magnitude of the mean resultant length vector over the time course of nose pokes leading to a correct choice . The strength of beta15-35 Hz coherence is greatest at 0 . 75 s after nose poke onset . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 01810 . 7554/eLife . 09849 . 019Figure 6—figure supplement 3 . Low Gamma ( 35–55 Hz ) coherent principal cell . ( a ) Relative waveform shape and amplitude across tetrode wires and clustering of spike features . ( b ) Circular histogram representing the phase of low gamma35-55 Hz at the time of each spike with the direction of the mean resultant length vector ( R ) shown by the arrow in black . Histograms were calculated from spiking activity during odor sampling intervals , during correct trials only . ( c ) Spiking activity of the cell during odor sampling of correct odors . Pairs of odors are differentially rewarded depending upon the context in which they are presented . Each row of tick marks represents spiking during a single trial . This cell exhibits activity that is selective for Position 3 and Position 4 , for both odors rewarded in that position . The spiking activity of this cell contains significant information ( see Materials and methods ) for odors , positions , and odor-position combinations . Information scores are indicated in the top right . ( d ) Magnitude of the mean resultant length vector over the time course of nose pokes leading to a correct choice . The strength of low gamma35-55 Hz coherence is greatest prior to odor sampling , and at 1 . 05 s after nose poke onset . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 01910 . 7554/eLife . 09849 . 020Figure 6—figure supplement 4 . High Gamma ( 65–90 Hz ) coherent principal cell . ( a ) Relative waveform shape and amplitude across tetrode wires and clustering of spike features . ( b ) Circular histogram representing the phase of high gamma65-90 Hz at the time of each spike with the direction of the mean resultant length vector ( R ) shown by the arrow in black . Histograms were calculated from spiking activity during odor sampling intervals , during correct trials only . ( c ) Spiking activity of the cell during odor sampling of correct odors . Pairs of odors are differentially rewarded depending upon the context in which they are presented . Each row of tick marks represents spiking during a single trial . This cell exhibits activity that is selective for Position 3 , for both odors rewarded in that position . The spiking activity of this cell contains significant information ( see Materials and methods ) for odors , positions , and odor-position combinations . Information scores are indicated in the top right . ( d ) Magnitude of the mean resultant length vector over the time course of nose pokes leading to a correct choice . The strength of high gamma65-90 Hz coherence is greatest at 0 . 25 s after nose poke onset and peaks again at 1 . 15 s after nose poke onset . DOI: http://dx . doi . org/10 . 7554/eLife . 09849 . 020 We observed principal cells coherent to each of the four rhythms that also exhibited significant information for task dimensions ( Figure 6—figure supplement 1–4c ) . We then tested whether the information for a specific task dimension changed over the course of a trial . For principal cells coherent to each rhythm that also exhibited significant information for a specific task dimension , we compared the information content during the before , odor , and end intervals described above . Theta4-12 Hz coherent cells exhibited increases in information for each task dimension across the three intervals examined ( Figure 6c–e , position: Friedman’s testtheta: d . f . = 2 , χ2 = 47 . 19 , p<0 . 00001; odor: Friedman’s testtheta: d . f . = 2 , χ2 = 15 . 89 , p=0 . 0004; odor-position: Friedman’s testtheta: d . f . = 2 , χ2 = 31 . 54 , p<0 . 00001 ) . Post hoc comparisons revealed that these increases occurred after the odor delivery , and lasted until the end of the nose poke ( Tukey’s Honest Significant Difference test , p<0 . 05 for comparisons of the before interval to odor and end intervals ) . Theta4-12 Hz coherent cells thus contained greater information during intervals after odor onset than during approach . Low gamma35-55 Hz and high gamma65-90 Hz coherent cells also exhibited increases in information across the three intervals examined ( low gamma: position: Friedman’s testlow gamma: d . f . = 2 , χ2 = 33 . 33 , p<0 . 00001; odor: Friedman’s testlow gamma: d . f . = 2 , χ2 = 5 . 93 , p=0 . 05; odor-position: Friedman’s testlow gamma: d . f . = 2 , χ2 = 33 . 73 , p<0 . 00001; high gamma: position: Friedman’s testhigh gamma: d . f . = 2 , χ2 = 20 . 56 , p<0 . 00001; odor: Friedman’s testhigh gamma: d . f . = 2 , χ2 = 7 . 93 , p=0 . 0189; odor-position: Friedman’s testhigh gamma: d . f . = 2 , χ2 = 26 . 7 , p<0 . 00001 ) . Post hoc comparisons revealed that the increases for position and odor-position information were sustained during both intervals after the odor onset , while information for odors increased only at the end of the poke ( Tukey’s Honest Significant Difference test , p<0 . 05 ) . In contrast , beta15-35 Hz coherent cells exhibited significant increases in information only for odor-position conjunctions , and only at the end of the poke , when the rat had committed to a decision ( position: Friedman’s testbeta: d . f . = 2 , χ2 = 5 . 78 , p=0 . 0556; odor: Friedman’s testbeta: d . f . = 2 , χ2 = 2 . 96 , p=0 . 2275; odor-position: Friedman’s testbeta: d . f . = 2 , χ2 = 6 . 56 , p=0 . 0376 ) . Together , these results suggest that rhythmic circuits differentially contribute task-relevant information , and that these contributions occur during different key intervals of the task . Moreover , beta15-35 Hz rhythmic circuits may be particularly selective for processing odor-position information , which is critical for successful task performance . Neural oscillations provide insight into organized interactions between cells at both local circuit and cross-regional scales . Understanding the flexible engagement of neurons into rhythmic circuits allows us to uncover the mechanisms through which single cells contribute to systems level processes . Our results indicate that the hippocampus can support multiple distinct , rhythmically identifiable processing states that are tightly linked to behavior in a context-guided odor-reward association task . During odor-sampling intervals , we observed transient amplitude dynamics in theta ( 4–12 Hz ) , beta ( 15–35 Hz ) , low gamma ( 35–55 Hz ) , and high gamma ( 65–90 Hz ) oscillations , indicating that there is a shift in processing state within the hippocampal circuit . We suggest that this shift serves to coordinate local neural activity for the processing of task-relevant information . We found that task-related processing coincided with the engagement of interneurons into local rhythmic circuits . We quantified this engagement by determining which interneurons demonstrated significant spike-phase coherence to the ongoing rhythms during odor sampling , as well as the magnitude and phase of their coherence . These measures demonstrated that engagement arises from a reorganization of spike timing with respect to each rhythm during task-related processing ( Figure 2a , Figure 3a and b ) , and not from enhanced participation through differences in overall firing rate between correct and incorrect trials ( Figure 3c ) . Importantly , the relationship between task performance and interneuron engagement differed across the four frequency ranges examined , suggesting that coordination within each rhythm may have a differential contribution to task-related processing . Notably , the largest proportion of interneurons exhibiting significant spike-phase coherence to theta4-12 Hz did so irrespective of whether the trial was correct or incorrect ( Figure 2a ) , suggesting that recruitment of these interneurons into theta4-12 Hz rhythmic circuits is not related to successful performance of the task . In contrast , interneurons that exhibited significant spike-phase coherence to beta15-35 Hz were coherent almost exclusively during correct trials , suggesting that task-related processing consistently recruits interneurons into beta15-35 Hz rhythmic circuits . Thus coordination of interneuron activity within each rhythm corresponds to task-related processing to a different extent . We further characterized task-related coordination in our analysis of phase preferences during correct and incorrect trials ( Figure 3—figure supplement 1 ) . We found that the phase preference between trial types was grossly different for the class of interneurons coherent during Correct Trials Only , but not for the interneurons that were coherent during All Trials . This indicates that task-related engagement manifests in both magnitude and preferred phase , suggesting that interneuron spike timing relative to the ongoing rhythms reflects its participation in task-related processing states . As the proportions of cells coherent during Correct Trials Only and All Trials differ across the four rhythms , we can surmise that each rhythm contributes uniquely to the orchestration of spike timing in the service of task-related processing . The task-related engagement of the principal cells into rhythmic circuits was strikingly different than the engagement of interneurons . Notably , principal cell engagement in each rhythm occurred almost exclusively during correct trials ( Figure 2b , Figure 4a and b ) , despite similar firing rates between correct and incorrect trials ( Figure 4c ) . This could be observed through the large proportions of principal cells with significant spike-phase coherence to each rhythm during Correct Trials Only ( Figure 3a ) . Selective engagement of the principal cell population during correct trials was not equal across the four frequencies examined . Beta15-35 Hz , low gamma35-55 Hz , and high gamma65-90 Hz coherent principal cells exhibited this preferential coherence during Correct Trials Only more often than theta4-12 Hz coherent principal cells , which consisted of a larger proportion of cells that were coherent irrespective of trial outcome . Though this difference is less pronounced in theta rhythmic principal cells than theta rhythmic interneurons , it nonetheless suggests that task-related processing more consistently recruits principal cell activity in beta15-35 Hz , low gamma35-55 Hz , and high gamma65-90 Hz rhythmic circuits than in theta4-12 Hz rhythmic circuits . The strength of principal cell coherence to each rhythm was also greater during correct trials than during incorrect trials ( Figure 4a and b ) , suggesting that there is an organization of principal cell spike timing that is unique to correct trials . The analysis of phase preference between correct and incorrect trials also revealed unique spike timing properties across the rhythms . The principal cells coherent to theta4-12 Hz , beta15-35 Hz , and high gamma65-90 Hz during Correct Trials Only exhibited inconsistent phases of entrainment across correct and incorrect trials , providing further evidence that these principal cells are engaged in their rhythmic circuits differently on correct and incorrect trials . In contrast , the principal cells coherent to low gamma35-55 Hz during Correct Trials Only exhibited similar phases across correct and incorrect trials , suggesting that there are features of engagement in low gamma35-55 Hz rhythmic circuits that are independent of trial outcome . Despite these finer differences in the nature of principal cell coordination within each rhythmic circuit , principal cells demonstrate a selective reorganization of spike timing during task-related processing . The extent of engagement in multiple rhythms also differed between the two cell types . Interneurons flexibly interacted in multiple rhythmic circuits , which was most apparent in the large proportion of interneurons that exhibited coherence to all four rhythms during correct trials , though potentially not all at the same time ( Figure 5a ) . Indeed , interneuron spike-phase coherence to multiple rhythms might be a signature of correct performance , whereas interneuron engagement in a single rhythm is more often observed during incorrect trials . This result indicates that the flexible engagement of the interneuron population in multiple rhythmic circuits is important for task-related processing . In contrast , larger numbers of principal cells were preferentially coherent to only one rhythm ( Figure 5b ) , providing evidence that theta4-12 Hz , beta15-35 Hz , low gamma35-55 Hz and high gamma65-90 Hz reflect functionally distinct processing states . This result suggests that principal cells can often participate in segregated rhythmic circuits with separable functions . Moreover , as many principal cells were engaged in single rhythmic circuits during correct and incorrect trials , engagement in multiple rhythmic circuits appears more related to task performance in the interneuron population . These dramatically different profiles of engagement between interneurons and principal cells indicate that each group interacts with the surrounding rhythmic circuits in markedly different ways . The engagement of interneurons in multiple rhythmic circuits could indicate co-modulation of interneuron activity in more than one rhythmic circuit at the same time . Although the strength of interneuron coherence did not appear to be co-modulated for any pair of rhythms ( Figure 5—figure supplement 2 ) , correlations in amplitude between pairs of rhythms can indicate potential for co-modulation in the network . The instantaneous amplitudes of certain pairs of rhythms were correlated during the task , although not consistently across all sessions . Specifically , beta15-35 Hz amplitude was often correlated with theta4-12 Hz and low gamma35-55 Hz amplitude , revealing a potential for beta15-35 Hz coherent interneurons to be co-modulated by theta4-12 Hz and low gamma35-55 Hz rhythmic circuits . Interestingly , theta15-35 Hz amplitude was often anti-correlated with low gamma35-55 Hz amplitude , suggesting that interneuron engagement in theta4-12 Hz and low gamma35-55 Hz rhythmic circuits may occur at different time points during the nose poke interval . These coordinated dynamics among specific pairs of rhythms suggest that there is some consistency in the organization of interneuron activity within each of the four rhythms that can include both co-modulation within some rhythmic circuits and temporally segregated entrainment in others . The engagement of principal cells in single rhythmic circuits may provide a mechanism for the hippocampus to simultaneously engage subpopulations of cells in distinct task-relevant processes . It is possible that subpopulations of principal cells become differentially engaged in their surrounding rhythmic circuits as a consequence of subtle differences in innervation from heterogeneous afferents or differences in the narrow-band intrinsic resonance of CA1 principal cells ( Freund and Buzsáki , 1996; Stark et al . , 2013 . ) . Task-related increases in coherence could then facilitate the effective communication of subpopulations of CA1 neurons with specific downstream targets that exhibit similar resonance . The ability of a single network to process information at multiple different frequencies thus creates an opportunity for multiplexing through frequency division and the selective readout of signals that are segregated into different frequency bands ( Akam and Kullmann , 2014 . ) . This feature could be particularly useful in the hippocampus for combining input from multiple different afferents while maintaining the ability to transmit individual signals . We demonstrate that the principal cells coherent to each rhythm can contribute task-relevant information at different time intervals during the odor-sampling epoch ( Figure 6c–e ) , providing further evidence that each rhythm reflects a distinct circuit process . Theta4-12 Hz , low gamma35-55 Hz and high gamma65-90 Hz coherent principal cells exhibited significant increases in information for position and odor-position after odor delivery that lasted until the end of the nose poke . Information for odors increased earlier in theta4-12 Hz coherent principal cells than in low gamma35-55 Hz and high gamma65-90 Hz coherent principal cells , indicating that theta4-12 Hz coherent principal cells have a dissociable contribution to the processing of odor information . Notably , beta15-35 Hz coherent principal cells only exhibited increases in information for odor-position conjunctions , which are a hallmark of associative memory ( Komorowski et al . , 2009 ) . Coupled with the fact that interneurons are coherent to beta15-35 Hz primarily during correct trials ( Figure 2a ) , our results strongly suggest that beta15-35 Hz rhythmic circuits might be uniquely processing information that is critical for successful utilization of associative memory . Multiple mechanisms could generate beta15-25 Hz rhythmicity within the hippocampal network . Identifying the mechanisms that give rise to the beta15-35 Hz rhythm can provide insight into how it contributes to associative memory processes . The selective entrainment of the interneuron population to beta15-25 Hz during correct trials ( Figure 2a ) could reflect the receipt of rhythmic input from an upstream structure that engages neurons in the hippocampus ( Figure 4d , mechanism 1 ) . This hypothesis is supported by recent studies suggesting that coherence between the CA1 and the lateral entorhinal cortex ( LEC ) parallels the onset of learning in an olfactory discrimination task and the development of task-selective ensemble activity ( Igarashi et al . , 2014; Rangel and Eichenbaum , 2014 ) . As the LEC has been shown to be important for the processing of multi-modal object information ( Young et al . , 1997; Deshmukh and Knierim , 2011 ) , communication between LEC and hippocampus could be critical for the association of odors with their reward contingencies . In addition , a recent study that used current source density analysis to reveal the location of current sources and sinks corresponding to beta oscillations in the dentate gyrus found that beta oscillations in this structure are likely driven by perforant path input ( Rangel et al . , 2015 ) . It is thus possible that beta oscillatory dynamics in the CA1 are the product of entrainment by an upstream cortical afferent . As an alternative to inheriting beta15-25 Hz rhythmicity from upstream structures , it is also possible that the hippocampus locally generates a beta15-25 Hz rhythm when engaged by specific afferents ( Figure 4d , mechanism 2 ) . Under this hypothesis , a change in communication between local neurons produces a functional circuit that resonates at beta15-25 Hz , which optimally facilitates associative memory processing . As a third possibility , the beta15-25 Hz rhythm could reflect the broad coordination of activity across disparate neural networks ( Kopell et al . , 2000; Bibbig et al . , 2002; Pinto et al . , 2003 ) . The hippocampus is just one of many structures that engage in beta rhythmic processing during the presentation of meaningful cues ( Igarashi et al . , 2014; Kay and Freeman , 1998; Rangel et al . , 2015; Quinn et al . , 2010; Buschman et al . , 2012; Leventhal , 2012; Tingley et al . , 2015 ) . Together , these structures could constitute an integrated network of circuits that span the brain . Rhythmic synchronization of this distributed network through a central rhythm generator could then enable information processing as a coordinated unit ( Figure 4d , mechanism 3 ) . Our study provides insight into the flexible coordination and engagement of distinct cell types within hippocampal circuits . We found two major differences in the rhythmic organization of CA1 interneuron and principal cell activity during a memory task . First , interneuron and principal cell coherence have distinct relationships to task performance . Second , we show that interneurons , unlike principal cells , are often flexibly engaged in multiple rhythms . These differences may shed light upon the distinct roles of excitation and inhibition in processing within rhythmically identifiable hippocampal circuits . Taken together , our results suggest that different rhythms make unique contributions to information processing within the hippocampus , and changes in the rhythmic profile reflect dynamic coordination of its cell activity . Further characterization of these rhythmic circuits will be critical for understanding how cell activity within the hippocampus is flexibly coordinated in the service of memory . All animal procedures were performed in accordance with NIH and Boston University Institutional Animal Care and Use Committee guidelines . Subjects were six male Long-Evans rats ( Charles River Laboratories ) housed individually and maintained on a 12-hr light/dark cycle . All neural recordings were performed during the light cycle . Rats were food and water restricted , and maintained at a weight of at least 85–90% of ad libitum body weight . Weights ranged from 450-–600g . All rats were handled daily for at least two weeks before beginning the experiment . Upon entering the study , rats were first exposed to the testing apparatus: a two-arm apparatus constructed of black plastic ( Figure 1—figure supplement 1 ) . Two 45-cm arms extended from opposite ends of a 30-cm central chamber . The central chamber consisted of 20-cm high walls and two pairs of doors , one black and one clear , that opened onto either arm . All doors could be independently raised and lowered by electric actuators . The end of each arm contained a widened area with two circular odor ports ( Figure 1—figure supplement 1 , inset ) , into which an odorant could be released by opening an air solenoid . Odorants were delivered by air flowing over vials of oil-based scents . A total 12 odors were used over the course of training and experimental sessions . Some of the odors were natural scents ( maple , cedar , spearmint , strawberry , sweet orange , mango , lemon ) while the others were chemical odorants ( 2-phenylpropionaldehyde , allyl-a-ionone , cis-3-hexen-1-ol , guaiacol , isoamyl acetate ) . Each port also contained a vacuum , which removed the odorant after release to prevent cross-contamination of odors between trials . Below each odor port was a tray with a well , into which water could be released to reward successful performance . Throughout the apparatus there were LED sensors to verify rat movement: two along the length of each arm , one in each odor port , one in each water well , and two in the central chamber . LED sensors in each odor port acted to record nose poke onset . All apparatus functions were controlled via a computer by customized MATLAB programs ( MathWorks , Natick , MA ) . Each rat underwent a behavioral shaping process to first poke its snout into an odor port and then increase the length of each nose poke . Rats received training sessions in which initially 100 ms nose pokes elicited a water reward . The nose poke criterion increased by 100 ms for each poke of sufficient length and dropped by 100 ms for every two successive unsuccessful nose pokes until rats learned to consistently poke for 1 . 5 s . During this process , rats only had access to one port at a time and received equal exposure to all four ports on the apparatus . Following nose poke training , each rat was taught to discriminate between two odors of an odor pair . During early discrimination sessions , two different context overlays made of distinct materials were placed over each side of the apparatus , and the rat was given access to one arm of the apparatus at a time . Rats alternated between arms in blocks of 20 trials during initial training , followed by blocks of 10 trials upon improved performance . Ultimately , arms switched in a pseudorandom , counterbalanced fashion in all later sessions . During each trial , each of the odors would be presented on one side of the apparatus , one odor in each of the two ports . Rats could initiate the delivery of an odor by poking their snouts into an odor port . Odors were released within the port after a 250 ms delay . One odor of the pair was designated as a “correct” odor , and sustaining a nose poke of 1 . 5 s in the odor port containing this odor resulted in a water reward . The other odor was not rewarded , and a white noise buzz would occur if the rat sustained a nose poke for 1 . 5 s in that port . After each trial , the rat returned to the central chamber , the doors surrounding the chamber were elevated , and the next trial would begin . We analyzed neural activity during trials when the rat maintained a nose poke for 1 . 5 s while sampling a rewarded odor ( correct trials ) and during trials when the rat maintained a nose poke for 1 . 5 s while sampling the non-rewarded odor ( incorrect trials ) . The same odor was always correct for a given context overlay and side of the apparatus . The location of the correct odor could switch between the left and right ports each trial , and trials were counterbalanced and pseudo-randomized before each session . The reward contingencies of the odors were reversed for each arm: the incorrect odor from the first arm was the correct odor on the second arm and vice versa . Each rat underwent 80 trials a day until reaching a criterion of 75% accuracy . With this final paradigm , rats were trained to perform this task with four pairs of odors ( eight odors total ) in a 96-trial session . Each odor pair was presented in a discrete block of 24 trials . For each recording session , distinct context overlays were placed on the apparatus for every two consecutive odor pairs . That rat was removed from the apparatus when the context overlays were replaced . Each half of the session was analyzed separately . All conditions were counterbalanced and pseudo-randomized within each 24-trial block before each session . For data analysis , only sessions in which the rat performed at 75% accuracy were used . Following training , each rat was surgically implanted with a hyperdrive containing 24 microdrives , each with an independently drivable tetrode . Each tetrode was composed of four strands of 0 . 0005” ( 12 µm ) Nickel-Chrome wire ( Sandvik , Stockholm , Sweden ) , gold-plated to reduce impedance to 200–250 kOhms at 1000 Hz . The implant site was located over the right dorsal hippocampus ( A/P = -4 . 0 mm; M/L = 2 . 2 mm ) , and tetrodes were turned down an initial 1 . 6 mm into the brain immediately following surgery . After rats received a two-week recovery period , tetrodes were progressively lowered over the training period ( 6–7 weeks ) to the principal cell layer of CA1 ( D/V = ~1 . 9 mm ) . Signals were amplified by a preamplifier 20x and amplified again to 4 , 000–6 , 000x ( Plexon , Dallas , TX ) , with a band-pass filter of 400–8 , 000 Hz to digitally isolate spikes ( OmniPlex , Plexon ) . Local field potentials ( LFPs ) were digitally isolated with a band-pass filter from 1–400 Hz . LFP and spike channels were globally referenced to a wire above the cerebellum , and spike channels were also locally referenced to a wire with low activity . Throughout the session , the rat’s location was recorded via digital video and tracking software ( CinePlex , Plexon , Dallas , TX ) that monitored the motion of two LEDs mounted at the top of the rat hyperdrive . Tracking data was time stamped and synchronized with neural recording data , all of which was stored offline for later analysis . Only sessions in which the rats performed at greater than 75% accuracy were used in analysis . Electrophysiological features such as the presence of theta ( 4–12Hz ) oscillations , sharp-wave ripples , and theta-modulated complex spiking activity were used to estimate tetrode locations . To confirm tetrode locations , rats were anesthetized with 2 . 5% isofluorane and small lesions were made by passing 40 µA of direct current through each wire . Final tetrode locations were visualized via a Nissl stain in 40 µm coronal sections . Single units were isolated in OfflineSorter ( Plexon ) by comparing waveform features across tetrode wires including peak and valley voltage amplitudes , total peak-to-valley distance , and principal component analysis . Principal cells and interneurons were classified according to both firing rate and waveform characteristics . Interneurons clustered according to mean firing rate , mean width at half the maximum amplitude of the waveform , and mean temporal offset from peak to trough . Interneurons exhibited mean firing rates of at least 5 Hz , a mean width at half-max less than 150 μs , and a mean peak to trough waveform width less than 350 μs , while principal cells exhibited mean firing rates of less than 3 Hz with wider waveforms ( Stark et al . , 2013; Csicsvari et al . , 2003; Bartho , 2004 ) . Approximately 5–10 recording sessions were performed over the course of 3–5 weeks from each rat , with rats resting on days between sessions . For populations of principal cells coherent to theta4-12 Hz , beta15-25 Hz , low gamma35-55 Hz , and high gamma65-90 Hz during correct trials , we calculated the information for positions , odors , and odor-position conjunctions expressed in their spiking activity . All information scores were calculated using the spiking activity from correct trials only . Position information describes the extent to which the spiking activity of a neuron differentiates between the four possible odor port positions . Odor information describes the ability of the neuron to discriminate between four odors ( A , B , C , or D ) . Since odors are only rewarded on one side of the maze and we only considered correct trials , odor identity is nested within the identity of a given side of the maze . Odor-position information describes the extent to which neural activity was selective for the co-occurrence of a specific odor in a particular odor port location . The position information score ( Markus et al . , 1994 ) was calculated as follows: I = ∑ Pi FiF log2 FiF Where i is the odor port position number ( four possible positions ) , Pi is the probability of occupancy in position i , Fi is the mean firing rate for position i , and F is the overall mean firing rate of the cell . For the calculation of odor information , a similar formula was used where Pi is the probability of experiencing an odor , and i is the number of the odor ( four possible odors ) . For the calculation of odor position information , Pi is the probability of experiencing an odor in a given position , and i is the number of the odor-position combination ( eight possible odor-position combinations for correct responses in two consecutive blocks of odor pairs ) . To hold reward value constant , odor A was compared at each position with odor C , and B with D . In order to determine whether calculated scores could be acquired by chance from the spiking behavior of a given principal cell , task conditions were randomly shuffled 1000 times and the observed information was considered significant if greater than the 95% confidence interval of the condition-shuffled scores . Principal cells could exhibit information for more than one task dimension . Information scores were then calculated for three different 500 ms intervals: 750 ms-–250 ms prior to nose poke onset ( before ) , the 500 ms after odor onset ( odor ) , and the last 500 ms of the nose poke ( end ) . This latter analysis was performed on all cells that exhibited significant information in their spiking behavior for a given task dimension during the entire duration of correct nose pokes . Differences in the median information across the three time intervals were assessed using a Friedman’s test , with post-hoc pairwise comparisons performed using a Tukey’s Honest Significant Difference test . Average spectrograms for the odor-sampling intervals of correct and incorrect trials were calculated using a multi-taper method from the Chronux open source MATLAB toolbox ( available at: http://chronux . org/ ) ( Mitra and Bokil , 2008 ) . Spectrograms were calculated for intervals beginning 0 . 5 s prior to the nose poke and lasting until 1 . 5 s after its initiation . The results from multiple correct trials in a single session were averaged and then divided by the mean amplitude observed during 2 . 0 s baseline inter-trial intervals in the center chamber of the behavioral apparatus . The log of these values was then taken before averaging across all sessions from all rats . For comparison between correct and incorrect trials , the results from multiple correct trials in a single session were averaged and then divided by the amplitude during incorrect trials . The log of these values was then taken before averaging across all sessions from all rats . A 3rd-order Butterworth filter was first used to bandpass filter the LFP for theta ( 4–12 Hz ) , beta ( 15–35 Hz ) , or low gamma ( 35-–55 Hz ) , or high gamma ( 65–90 Hz ) ( Rubino et al . , 2006 ) . These frequency ranges were chosen based upon the observed frequencies present in the average spectrogram during odor sampling ( Figure 1 ) . The instantaneous amplitude for the entire session was then calculated by taking the magnitude of the complex Hilbert transform of the filtered signal . The mean amplitudes during correct and incorrect trials were then calculated by averaging the amplitudes observed during the 1 . 5 s nose poke intervals leading to a correct or incorrect response for a single session . To examine increases or decreases in amplitude over the course of a nose poke , the mean amplitudes for each session were averaged within six 250 ms time bins spanning the 1 . 5 s nose poke interval . A two-factor repeated measures ANOVA was used to determine whether amplitudes changed over the course of the six time bins or differed according to behavioral outcome ( correct or incorrect ) . To examine whether amplitude was co-modulated across rhythms during correct trials of a session , we calculated the instantaneous amplitude of each of the four frequency ranges during correct trials and performed correlations on the amplitude values for every pair of frequencies . For each session , a random temporal jitter was applied to instantaneous amplitude of each rhythm 1000 times to create a null distribution of correlations that would be expected by chance . All correlations in the data were significantly above the 95% confidence interval of distributions derived in this manner . To determine whether there were any differences in the correlations between pairs of frequencies , we performed a one-way ANOVA with post-hoc pairwise comparisons performed using Tukey’s Honest Significant Difference test . As mentioned above , a 3rd-order Butterworth filter was first used to bandpass filter the LFP for theta ( 4–12 Hz ) , beta ( 15–35 Hz ) , low gamma ( 35–55 Hz ) , or high gamma ( 65–90 Hz ) ( Rubino , et al . , 2006 ) . The instantaneous phase was then calculated by taking the arctangent of the complex Hilbert transform of the filtered signal . Single cell spike-phase relationships to the filtered LFP during the odor sampling intervals were assessed using a Rayleigh statistic , and categorized as significantly phase coherent if exhibiting a p<0 . 05 . Spike-phase relationships were calculated for correct trials only or incorrect trials only for the 1 . 5 s prior to a reward or white noise buzz outcome , respectively . The magnitude and phase of coherency for cell spiking activity with respect to each rhythm was calculated using a multi-taper method from the Chronux open source MATLAB toolbox ( available at: http://chronux . org/ ) ( Mitra and Bokil , 2008 ) , and adjusted for differences in firing rate between correct and incorrect trials by estimating a correction factor that is conceptually equivalent to spike thinning procedures ( Aoi et al . , 2015 ) . The number of tapers ranged from 3–15 , to maximize the number of tapers that could be used while avoiding contamination from neighboring frequencies outside the boundaries of a given frequency range . The magnitude and phase of coherence at approximately 7 Hz , 20 Hz , 45 Hz , and 75 Hz were reported for the theta4-12 Hz , beta15-35 Hz , low gamma35-55 Hz , or high gamma65-90 Hz frequency ranges , respectively . To test whether there was any consistency in the preferred phase angles during correct and incorrect trials , we performed circular correlations on the phases observed across a given population of interneurons or pyramidal cells during each trial type . For the interneuron population that exhibited significant spike-phase coherence to all four rhythms at some point during correct trials , we tested whether the magnitude of coherence to each rhythm was correlated across trials . For each session , we first determined the magnitude of coherence at 7 Hz , 20 Hz , 45 Hz , and 75 Hz for every correct trial . We then determined whether the magnitude of coherence to one frequency was correlated with the magnitude of coherence to another frequency across trials by performing correlations on the coherence values across trials for every pair of frequencies . To test for significant differences in the correlations between pairs of frequencies , we performed a one-way repeated measures ANOVA .	Electrodes placed on the surface of the scalp can reveal rhythmic patterns of electrical activity within the brain . These rhythms reflect the coordinated firing of large numbers of neurons that are connected together within a network in order to process information . A single network can show rhythms with various different frequencies depending on its local connections and the pattern of input that it receives at any given time . One region that exhibits striking changes in these rhythmic patterns is the hippocampus: a brain area that plays a key role in memory . The hippocampus contains many cell types , including interneurons ( which form connections with nearby cells ) and principal cells ( which connect with cells outside of this region ) . Though both participate in rhythmic circuits , little is known about the different extents to which these distinct cell types are engaged in rhythmic processing , or how rhythmic processing might support memory . Rangel , Rueckemann , Rivière et al . have now addressed these questions by using electrodes to record from the hippocampus as rats learned to associate specific odors in different environments with a reward . As the rats sniffed the odors , their brains showed four different hippocampal rhythms: from a low frequency called “theta” , through “beta” and “low gamma” up to “high gamma” frequencies . Each of these hippocampal rhythms varied in strength over time , indicating that rhythmic processing is dynamic during the task . Rangel , Rueckemann , Rivière et al . found that neurons fired rhythmically during trials in which the rat chose the correct odor-environment combination . In these correct trials , individual principal cells were more likely to fire in synchrony with only one of the rhythms . In contrast , interneurons were more likely to fire in synchrony to each of the four rhythms at some point during a correct choice . Among the four rhythms , coordinated principal cell and interneuron firing with respect to the beta rhythm was most tightly linked with a correct choice . These findings reveal that investigation of rhythmic dynamics in the hippocampus can provide insight into how the timing of cell activity is coordinated to support memory .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2016	Rhythmic coordination of hippocampal neurons during associative memory processing
Molecular and cellular processes in neurons are critical for sensing and responding to energy deficit states , such as during weight-loss . Agouti related protein ( AGRP ) -expressing neurons are a key hypothalamic population that is activated during energy deficit and increases appetite and weight-gain . Cell type-specific transcriptomics can be used to identify pathways that counteract weight-loss , and here we report high-quality gene expression profiles of AGRP neurons from well-fed and food-deprived young adult mice . For comparison , we also analyzed Proopiomelanocortin ( POMC ) -expressing neurons , an intermingled population that suppresses appetite and body weight . We find that AGRP neurons are considerably more sensitive to energy deficit than POMC neurons . Furthermore , we identify cell type-specific pathways involving endoplasmic reticulum-stress , circadian signaling , ion channels , neuropeptides , and receptors . Combined with methods to validate and manipulate these pathways , this resource greatly expands molecular insight into neuronal regulation of body weight , and may be useful for devising therapeutic strategies for obesity and eating disorders . Neurons that express Agouti related protein ( Agrp ) and Proopiomelanocortin ( Pomc ) comprise two intermingled molecularly defined populations in the hypothalamic arcuate nucleus ( ARC ) that mediate whole-body energy homeostasis in conjunction with other cell types . AGRP and POMC neurons positively and negatively regulate body weight , respectively ( Aponte et al . , 2011; Krashes et al . , 2011 ) . AGRP neurons transduce circulating signals of energy deficit into increased output of the neuropeptides AGRP and Neuropeptide Y ( NPY ) as well as the neurotransmitter γ-aminobutyric acid ( GABA ) , each of which contribute to increased appetite and body weight . Correspondingly , weight loss is accompanied by increased Agrp and Npy gene co-expression in AGRP neurons ( Hahn et al . , 1998 ) , as well as increased electrical activity ( Takahashi and Cone , 2005 ) and synaptic plasticity ( Yang et al . , 2011; Liu et al . , 2012 ) . In contrast , during energy deficit , POMC neurons decrease electrical activity due to inhibitory synaptic input from AGRP neurons ( Takahashi and Cone , 2005; Atasoy et al . , 2012 ) , and Pomc neuropeptide gene expression is reduced ( Schwartz et al . , 1997 ) . AGRP and POMC neurons are thus both associated with sensing and counteracting energy deficit states . Because these neurons play major reciprocal roles in energy homeostasis , investigations of the molecular response pathways for AGRP and POMC neurons to weight-loss are critical for identifying key control points associated with regulation of body weight . AGRP and POMC neurons sense energy deficit , in part , through responding to the metabolic hormones ghrelin , leptin , and insulin . Signaling pathways downstream of the receptors for these hormones have been elucidated ( Banks et al . , 2000; Kitamura et al . , 2006 ) , but most of the other molecular processes involved in the cellular response to systemic metabolic challenge in AGRP and POMC neurons remain unexplored . In light of this , a transcriptome-wide view of gene expression changes can provide a foundation for investigating the neuronal cell biology of these energy homeostasis sensing neurons during a state of energy deficit . The transcriptional response to food-deprivation has been reported previously using tissue samples from the entire hypothalamus ( Guarnieri et al . , 2012 ) or ARC ( Li et al . , 2005; Jovanovic et al . , 2010 ) , but these studies lacked cell type-specificity necessary to understand the molecular response properties of individual neural circuit nodes . Recent approaches employing immunoprecipitation of messenger RNA ( mRNA ) in molecularly defined and even projection-specific populations ( Heiman et al . , 2008; Dalal et al . , 2013; Ekstrand et al . , 2014; Allison et al . , 2015 ) require large numbers of cells , and therefore have been challenging to perform for neurons with small population sizes , such as AGRP and POMC neurons . A transcriptional profile of AGRP neurons has been obtained previously from dissociated tissue in which fluorescently labeled AGRP neurons were sorted and pooled from ∼40 neonatal mice and compared to a similar number of neonatal AGRP neuron-specific Foxo1 knockout mice ( Ren et al . , 2012 ) . In neonatal mice , cells are readily dissociated , but AGRP neurons are not necessary for early neonatal life and their axons are not developed ( Bouret et al . , 2004; Luquet et al . , 2005 ) , thus , the relevance of neonatal gene expression patterns to those in adult mice is uncertain . Moreover , comparing only one sample from two conditions prevents statistical analysis of differentially expressed genes ( DEG ) . Recent technical improvements in cell sorting and transcriptional profiling methods have enabled the generation of high quality gene expression profiles from small numbers of fluorescently labeled neurons ( typically 40–250 neurons ) from single adult mouse brains ( Sugino et al . , 2006; Okaty et al . , 2011 ) . Importantly , this permits use of individual animals as replicates for comparing gene expression profiles under different conditions , which is the approach that we used here . We performed RNA sequencing ( RNA-Seq ) using AGRP and POMC neurons from ad libitum fed young adult mice as well as from mice after 24-hr food deprivation . We confirmed a small number of previously reported changes in gene expression , and also identified hundreds of additional DEG . These changes in gene expression allowed identification of coordinated signaling pathways that are involved in the response to food deprivation , and we focus here on neuropeptides , G-protein coupled receptors ( GPCRs ) , as well as pathways associated with neuron electrical activity , circadian regulation , and endoplasmic reticulum ( ER ) -stress signaling . This resource , which includes gene expression profiles for AGRP and POMC neurons as well as methods for validating and evaluating the functional significance of these changes , provides a foundation for in depth analysis of the molecular control points for neuron populations involved in energy homeostasis . In addition to changes in expression of individual genes and gene families such as kinases , phosphatases , and transcription factors that are evident from RNA-Seq data ( Figure 1—figure supplement 2C–E ) , the coordinate regulation of multiple genes allows predictions to be made about pathways involved in the transition from energy replete to energy deficient physiological states . Gene annotation enrichment analysis highlighted several pathways that were significantly affected by food-deprivation ( Table 1 ) . One such previously established pathway in AGRP neurons is leptin receptor signaling . Leptin levels fall in energy deficit and this is associated with a concomitant rise in Lepr ( +5 . 3-fold , q = 1 . 3e−10 ) in AGRP neurons , as was previously reported for whole-hypothalamus tissue samples ( Baskin et al . , 1999 ) , but Lepr was not significantly changed in POMC neurons ( −1 . 6-fold , q = 0 . 39 ) . Consistent with low circulating leptin , which leads to reduced Lepr signaling , Jak2 , the Lepr-associated kinase , and Socs3 , a downstream target of Lepr signaling , were reduced in AGRP neurons ( Figure 1—figure supplement 2F , G ) . Conversely , Foxo1 , a transcription factor that is negatively regulated by leptin receptor signaling , was selectively increased in AGRP neurons with food deprivation . 10 . 7554/eLife . 09800 . 006Table 1 . Gene annotation enrichment analysis of differentially expressed genesDOI: http://dx . doi . org/10 . 7554/eLife . 09800 . 006Pathway−log ( p-value ) Agouti related protein Leptin signaling4 . 9 Glutamate signaling/Axonal guidance/Ephrin/Rho GTPase3 . 5 Endoplasmic reticulum stress/Oxidative stress3 . 2 G-protein coupled receptor signaling3 . 0 Circadian rhythm signaling2 . 7 Sperm motility1 . 8Proopiomelanocortin Gαi signaling6 . 3 Tetrahydrobiopterin biosynthesis ( Gch1 ) 2 . 1 Zymostrerol biosynthesis ( Msmo1 ) 1 . 8 Opposite to leptin , levels of the hormone ghrelin are elevated with food-deprivation . We found that Ghsr was upregulated in AGRP neurons after food-deprivation ( +3 . 4-fold , q = 4 . 9e−9 ) , but was nearly absent in POMC neurons ( Figure 1—figure supplement 2F , G ) . Signaling components downstream of Ghsr were also upregulated , such as Prkca , a protein kinase C isoform . Therefore , our cell type-specific transcriptomic data recapitulates known leptin receptor and ghrelin receptor signaling pathways in AGRP and POMC neurons . More importantly , though , this comprehensive transcriptomic resource can be used to identify pathways that have not been previously implicated in the physiological response to energy deficit in these cell types . We focused on investigating pathways that had not been examined in AGRP neurons , such as systems for ER-stress , circadian regulation , and synaptic function , as well as genes encoding ion channels , GPCRs , and secreted proteins . Gene annotation enrichment analysis ( Table 1 ) highlighted , selectively in AGRP neurons , a transcriptional response for genes associated with ER-stress: the unfolded protein response ( UPR ) and ER-associated degradation ( ERAD ) of misfolded proteins . This pathway has not been examined previously in AGRP neurons; instead , based on whole hypothalamus analysis , UPR has been primarily associated with overnutrition states and leptin resistance ( Ozcan et al . , 2009 ) , as opposed to the energy-deficit condition examined here . ER-stress responses occur during high levels of protein translation where unfolded proteins elicit a program of downstream transcriptional responses to increase protein folding and processing in the ER , and this pathway showed the most prominent effect on gene expression that we identified in AGRP neurons . This pattern of differential gene expression was not observed in POMC neurons , indicating that it is not an artifact of the neuronal isolation procedure . Consequently we examined multiple aspects of ER-stress signaling by analysis of our RNA-Seq data , coupled with preliminary evaluation of ER-stress signaling using immunohistochemistry , a cell type-specific UPR pathway reporter , and smFISH . ER-localized proteins were significantly overrepresented in the group of food-deprivation-regulated genes in AGRP neurons ( p = 8 . 3e−16 , hypergeometric-test ) , and 95% of these DEG were upregulated ( Figure 2A ) . A key UPR marker , Hspa5 , which encodes the canonical ER-localized unfolded protein-sensing chaperone BiP , was selectively upregulated in AGRP neurons during food deprivation ( AGRP: +3 . 3-fold , q = 2 . 5e−10; POMC: −1 . 2-fold , q = 0 . 52 ) . Correspondingly , BiP-immunoreactivity was significantly increased by food-deprivation in AGRP neurons ( p = 2 . 5e−42 , ks-test , Figure 2B , D ) , indicating UPR activation in AGRP neurons . In contrast , we also found that BiP-immunoreactivity was significantly reduced in POMC neurons during food-deprivation ( p = 2 . 5e−19 , ks-test , Figure 2C , D ) , suggestive of reduced protein translational-load in this population . 10 . 7554/eLife . 09800 . 007Figure 2 . Food deprivation induces unfolded protein response in AGRP neurons . ( A ) Log2 ( fold-change ) [log2 ( fc ) ] and q-values for genes associated with endoplasmic reticulum ( ER ) localization ( KEGG pathway: mmu04141 ) that are affected by food deprivation in AGRP ( left columns ) or POMC ( right columns ) neurons . ( B , C ) Representative images showing BiP-immunofluorescence from NpyhrGFP or PomctopazFP mice . Arrows: examples of fluorescently labeled ( B ) AGRP and ( C ) POMC neurons used for BiP quantification . Scale: 10 μm . ( D ) Population counts of BiP somatic intensity in AGRP or POMC neurons . AGRP . fed , n = 209; AGRP . FD , n = 283; POMC . fed , n = 121; POMC . FD , n = 92; 3 mice per condition . Bars: mean values . Rank-sum test . **p < 0 . 001 . ( E ) Fraction of spliced to total Xbp1 transcript isoforms in AGRP neurons . Unpaired one-tailed t-test . p < 0 . 05 . ( F , G ) Representative images ( F ) and cumulative probability distribution ( G ) of TDP43-immunoreactive-granules ( red ) in GFP-expressing AGRP neurons ( p = 0 . 76 , ks-test ) . Fed , n = 276; FD , n = 174; 2 mice per condition . Scale , 10 μm . ( H–J ) Representative images ( H , I ) of GFP:ATF6 expression AgrpCre;ai9 ( tdtomato ) mice . Cumulative probability distribution ( J ) of the ratio of nuclear to cytoplasmic GFP fluorescence in AGRP neurons: fed vs FD ( p = 0 . 44 , ks-test ) Fed , n = 92; FD , n = 63; 4 mice per condition; or AGRP neurons: DMSO vs tunicamycin ( p = 2 . 9e−16 , ks-test ) DMSO , n = 42; Tunicamycin , n = 38; 1 mouse per condition . ( K ) Differentially expressed cell survival genes . Left , mean expression level [log2 ( TPM ) ] of each transcript of each experimental group . Right , log2 ( fold-change ) and q-values for differential expression between FD and fed states separately for AGRP and POMC neurons . ( L ) Schematic for DEG in ER stress-associated pathways in AGRP neurons after food-deprivation . Red: upregulated expression , Blue: downregulated expression . DOI: http://dx . doi . org/10 . 7554/eLife . 09800 . 007 UPR involves a program of gene expression that is regulated by Ern1/Ire1 , which was selectively increased in AGRP neurons by food deprivation ( +2 . 7-fold , q = 0 . 001 ) . In response to the accumulation of unfolded proteins in the ER lumen , Ern1/Ire1 undergoes a conformational change , which activates endoribonuclease activity for the short-lived transcript Xbp1 and splices it into a more stable mRNA , Xbp1s . These RNA-Seq data reveal increased abundance for the Xbp1s spliced mRNA in AGRP neurons after food-deprivation ( Xbp1s/Xbp1 , fed: 7 . 4 ± 1 . 0% , FD: 15 . 6 ± 3 . 8% , p = 0 . 043 , one-tailed t-test; Figure 2E ) , consistent with activation of Ern1/Ire1 endonuclease activity in AGRP neurons after 24-hr food deprivation . Xbp1s is a transcription factor that regulates UPR-related gene expression . Two well-established Xbp1s-dependent gene targets ( Lee et al . , 2003 ) were selectively upregulated in AGRP neurons after food deprivation: Dnajc3/p58IPK ( +1 . 9-fold , q = 0 . 0007 ) and Dnajb9 ( +2 . 2-fold , q = 0 . 0003 ) . Many other genes regulated by Xbp1s were also increased in AGRP neurons during food-deprivation , including protein folding chaperones of the Hsp40 ( Dnaj ) , Hsp70 ( Hspa ) , and Hsp90 families , Calr and Canx , as well as protein disulfide isomerases ( Pdia3-6 ) , which aid in protein folding by catalyzing the formation of disulfide bonds ( Figure 2A ) . Atm ( −3 . 1-fold , q = 7e−5 ) , a DNA damage sensing-enzyme that , when reduced leads to increased Xbp1 splicing ( He et al . , 2009 ) , was downregulated with food deprivation . Taken together , Xbp1-splicing and the patterns of downstream gene expression indicate a previously unreported role for Ern1/Ire1 → Xbp1s signaling in the adaptive response of AGRP neurons to energy deficit . We also investigated the involvement of two other UPR signaling pathways regulated by ER-bound transmembrane proteins , Eif2ak3 ( also called PERK ) and Atf6 . The Eif2ak3/PERK ( AGRP: −2 . 1-fold , q = 0 . 23 ) arm of the UPR pathway is the first to be engaged during ER stress and suppresses translation of most mRNA . We expected that prolonged translational arrest in AGRP neurons was unlikely because it is inconsistent with elevated neuropeptide production in AGRP neurons during energy deficit . Activation of Eif2ak3/PERK leads to mRNAs sequestration into messenger ribonucleoprotein particles that aggregate into stress granules containing the RNA binding protein , TDP43 ( Colombrita et al . , 2009 ) . To assess stress granule formation , we used anti-TDP43 immunostaining . Elevated stress granule formation was not detected in AGRP neurons from FD mice ( p = 0 . 76 , ks-test , Figure 2F , G ) , which provides preliminary evidence that Eif2ak3/PERK-mediated translational arrest may not be engaged at this 24-hr food deprivation time-point . Moreover , transcripts for ER protein translocation ( Srp68 , Srp72 , Sec61a1 , Sec61b1 , Sec63 , Serp1/Ramp4 ) and Golgi trafficking ( Sec14l1 , Sec22b , Sec24d ) were upregulated in AGRP neurons , possibly indicating increased protein translation and folding capacity during energy deficit and consistent with the requirement for increased peptidergic neurotransmission for AGRP neuron function . An additional signaling arm of the UPR is mediated through the ER-bound protein Atf6 ( AGRP: +2 . 0-fold , q = 0 . 0016 ) . In response to ER stress , Atf6 is subject to proteolytic cleavage and releases an N-terminal transcription factor domain that translocates to the nucleus and initiates gene expression to increase protein folding . To examine the extent of Atf6 nuclear translocation in the response of AGRP neurons to food-deprivation , we developed a Cre recombinase-dependent N-terminal green fluorescent protein ( GFP ) :Atf6 fusion ( Samali et al . , 2010 ) reporter of Atf6 cleavage for use in the brain ( Figure 2H ) . After expression in AGRP neurons , we measured the nucleus:cytoplasm ratio of GFP , which was very low and was not significantly increased in FD vs ad libitum fed mice; whereas the ER-stress-inducer , tunicamycin , strongly increased nuclear localization of GFP in AGRP neurons ( fed vs FD: p = 0 . 44 , dimethyl sulfoxide ( DMSO ) vs tunicamycin: p = 2 . 9e−16 , ks-test , Figure 2H–J ) . Low nuclear localization indicates minimal activation of the Atf6-signaling arm of UPR in AGRP neurons after 24-hr food deprivation . Consistent with this , a transcript selectively regulated by Atf6 , Herpud1 ( Lee et al . , 2003 ) , was not significantly ( q = 0 . 22 ) elevated in AGRP neurons from FD mice . However , because many Atf6-regulated genes are also regulated by Xbp1s , a robust ER-stress response can be maintained without Atf6 signaling ( Yamamoto et al . , 2007 ) . Oxidative stress pathways associated with elevated protein production are also increased in AGRP neurons during food deprivation . The transcription factor Nfe2l2 ( also called Nrf2 ) regulates oxidative stress and is normally rapidly degraded in the cytoplasm , in part through association with Keap1 ( Kobayashi et al . , 2004 ) , which shows reduced expression with food-deprivation ( −2 . 2-fold , q = 0 . 05 ) . Nfe2l2/Nrf2 upregulates expression of the key glutathione biosynthetic enzymes Gclc ( +2 . 3-fold , q = 0 . 0002 ) and Gclm ( +1 . 8-fold , q = 0 . 002 ) , which are elevated in AGRP neurons with food-deprivation . Nfe2l2/Nrf2 increases other transcripts associated with oxidative stress , such as Mt1 ( +34-fold , q = 0 . 004 ) , Mt2 ( 79-fold , q = 0 . 004 ) , and Srxn1 ( +17-fold , q = 1 . 2e−6 ) , which are among the most strongly upregulated transcripts in AGRP neurons during food-deprivation . Other oxidative stress transcripts were also upregulated , such as peroxiredoxin 2 , Prdx2 ( +1 . 9-fold , p = 0 . 0004 ) , the transcription factor Hif1a ( +1 . 8-fold , p = 0 . 005 ) and the oxidative regulatory enzyme prolyl hydroxylase ( P4hb , +2 . 7-fold , p = 8e−7 ) . Together , this group of cell type-selectively upregulated genes is consistent with an adaptive response to increased oxidative stress in AGRP neurons during energy deficit . UPR also typically leads to ERAD of unfolded proteins . E2 ubiquitin conjugating enzyme subunits ( Ube2j1 , Ube2g2 , Ube2q1 , Ube2z , Ube4b ) and the E3 ligase Park2 were upregulated in AGRP neurons during food-deprivation ( Figure 2A ) . Os9 , a lectin that senses misfolded proteins , and Edem3 and Stt3b ( Figure 2A ) , which are enzymes that mark unfolded proteins for ERAD ( Sato et al . , 2012 ) , were also increased . In addition , Vcp , a component for retrotranslocation of polyubiquitinylated unfolded proteins for degradation , was elevated in AGRP neurons during energy deficit . Collectively , this pattern of gene expression indicates engagement of ERAD during food-deprivation . UPR is protective to cells for short periods of elevated ER-stress , but prolonged activation can result in apoptosis . The pro-apoptotic Ddit3 ( Chop ) is typically upregulated during UPR , however Ddit3/Chop was unchanged in AGRP neurons after 24-hr food deprivation ( Figure 2K ) . Instead , a variety of anti-apoptotic transcripts Bcl2l ( Bcl-xl ) , Manf , Mcl1 , and the caspase inhibitor Cflar were upregulated ( Figure 2K ) . Mcl1 upregulation is consistent with a previously established pathway Creb3l2 → Atf5 → Mcl1 ( Izumi et al . , 2012 ) , each member of which is upregulated in food deprivation ( Figure 2K ) . Irak2 ( −4 . 7-fold , q = 3 . 7e−5 ) was strongly suppressed , consistent with reports that its reduction is protective against apoptosis ( Benosman et al . , 2013 ) . In addition , the cellular stress-induced transcript Atf3 ( +15 . 5-fold , q = 0 . 005 ) also promotes neuron survival ( Francis et al . , 2004 ) and is increased with food deprivation . This pattern of gene expression provides preliminary evidence of a potential role for anti-apoptotic pathways in AGRP neurons during activation in response to energy deficit . Taken together , cell type-specific transcriptional profiling , immunohistochemistry , and a cell type-specific UPR reporter construct indicate engagement of ER-stress pathways selectively in AGRP neurons during energy deficit . Increased neuron activity and elevated neuropeptide production associated with food-deprivation is expected to increase the translational-load in AGRP but not POMC neurons , and UPR may serve to cope with elevated neuropeptide and synaptic output . This is due , in part , to Xbp1s signaling and also results in induction of gene expression associated with oxidative stress responses , ERAD , and protection of cells from apoptotic pathways that might otherwise be associated with prolonged ER-stress ( Figure 2L ) . Expression of genes associated with circadian regulation was strongly altered by food-deprivation . We examined the expression levels of 19 core circadian reference genes ( Yan et al . , 2008; Rey et al . , 2011 ) and found that after food-deprivation 9/19 of these genes were downregulated in AGRP neurons ( Bhlhe40 [Dec1] , Bhlhe41 [Dec2] , Nr1d1 [Rev-erbα] , Nr1d2 [Rev-erbβ] , Dbp , Hlf , Tef , Per2 , and Per3 ) , and only 3/19 were upregulated ( Rorb , Nfil3 , Per1 ) ( Figure 3A ) . For POMC neurons , only one transcript was significantly changed ( Nr1d2/Rev-erbβ ) ( Figure 3A ) . The nine circadian genes downregulated in AGRP neurons during food-deprivation are established targets of two important circadian transcription factors , Clock and Arntl ( also called , Bmal ) , which heterodimerize to regulate a large number of downstream genes through E-box transcriptional response elements ( Rey et al . , 2011 ) . We analyzed a collection of E-box regulated genes that have been previously determined by Arntl ( Bmal ) chromatin binding , bioinformatic analysis , and transcriptional profiling ( Rey et al . , 2011 ) . E-box-containing genes were over-represented among differentially expressed transcripts in AGRP neurons but not in POMC neurons ( p = 0 . 00075 and 0 . 31 , respectively; hypergeometric-test ) . For the E-box-containing genes that were significantly differentially expressed more than twofold in AGRP neurons after food-deprivation , 90% ( 18/20 ) were downregulated ( Figure 3B ) . Conversely , in POMC neurons , Nr1d2/Rev-erbβ was the only E-box-containing gene that was significantly differentially expressed more than twofold , and it showed increased expression ( Figure 3B ) . This indicates that differentially expressed E-box-containing genes during energy deficit are selectively reduced in AGRP neurons . 10 . 7554/eLife . 09800 . 008Figure 3 . Changes in the expression of circadian and synapse-associated genes after food-deprivation . ( A ) Gene expression changes for core circadian genes . Left , mean expression level [log2 ( TPM ) ] of each transcript of each experimental group . Right , log2 ( fold-change ) and q-values for differential expression between FD and fed states separately for AGRP and POMC neurons . ( B ) E-box genes differentially expressed during food-deprivation . ( C ) DEG with synapse-localized functions ( Gene Ontology: 0045202 ) . Top , AGRP neurons: FD vs fed; bottom , POMC neurons FD vs fed . ( D , E ) Gene expression changes for granin genes ( D ) and synaptotagmins ( E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09800 . 00810 . 7554/eLife . 09800 . 009Figure 3—figure supplement 1 . Arntl/Bmal expression in AGRP neurons . ( A ) Representative images for Arntl/Bmal-immunofluorescence in AGRP neurons from fed and FD mice . Scale , 10 µm . ( B ) Arntl/Bmal immunofluoresence intensity ( bars , mean values ) in AGRP neurons from fed and FD mice . Fed , n = 159; FD , n = 196; 3 mice per condition . DOI: http://dx . doi . org/10 . 7554/eLife . 09800 . 009 Many genes regulated by E-box transcriptional response elements show reduced expression in Arntl−/− mice ( Rey et al . , 2011 ) , but Clock and Arntl expression levels in AGRP neurons were not significantly altered by food-deprivation . Immunohistochemistry for Arntl protein expression in AGRP neurons from fed and FD mice also showed similar levels ( fed: 115 ± 4 a . u . , n = 158 neurons; FD: 127 ± 4 a . u . , n = 195 neurons; p = 0 . 058 , rank sum test , Figure 3—figure supplement 1 ) . An alternative pathway that has been shown to regulate E-box genes is the transcriptional splicing factor Sfpq ( Psf ) ( Duong et al . , 2011 ) , which shows selectively increased expression in AGRP neurons after food-deprivation ( AGRP: +2 . 0-fold , q = 0 . 00064; POMC: 1 . 0-fold , q = 1 . 0 ) . However , detailed examination of pathways that regulate circadian E-box containing genes in AGRP neurons during energy deficit states is required . Excitatory synaptic plasticity occurs in AGRP neurons during energy deficit ( Yang et al . , 2011; Liu et al . , 2012 ) . Gene annotation enrichment analysis revealed significant changes in expression of genes associated with glutamate signaling , synaptic plasticity , and presynaptic function ( Table 1 ) . Moreover , genes encoding proteins that are localized to the synapse were over-represented in the DEG ( q < 0 . 05 ) from AGRP neurons ( p = 1 . 2e−6 , hypergeometric-test; Figure 3C ) . For example , α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid ( AMPA ) and kainate glutamate receptors , which mediate excitatory synaptic transmission , were upregulated in AGRP neurons ( Gria3: +3 . 4-fold , q = 2e−5 , Grik1: +3 . 0-fold , q = 0 . 03; Grik3: +3 . 2-fold , q = 0 . 005 ) , but not in POMC neurons ( see Ion Channels , below ) . Food-deprivation also induced upregulation of excitatory synaptogenic genes in AGRP neurons ( Syndig1: +8 . 3-fold , q = 0 . 008; Syndig1l: +13 . 2-fold , q = 0 . 0001 ) ( Kalashnikova et al . , 2010; Lovero et al . , 2013 ) as well as kinases ( Figure 1—figure supplement 2C ) that regulate activity-dependent changes in spine morphology and excitatory synaptic plasticity , such as p21-associated kinase 3 ( Pak3 , +2 . 0-fold , q = 4e−7 ) , Ptk2b ( +31 . 3-fold , q = 0 . 0002 ) , and Plk2 ( +8-fold , q = 5 . 2e−6 ) ( Boda et al . , 2004; Seeburg et al . , 2008; Bartos et al . , 2010 ) . Therefore , a number of upregulated genes are associated with elevated excitatory synaptic input , potentially contributing to synaptic plasticity and increased AGRP neuron activity previously reported for FD mice ( Takahashi and Cone , 2005; Yang et al . , 2011; Liu et al . , 2012 ) . During food deprivation , elevated AGRP neuron activity results in increased neurotransmitter and neuropeptide release ( Atasoy et al . , 2012 ) . In line with correspondingly high Agrp and Npy expression ( Figure 1C and Figure 1—figure supplement 1D ) , transcripts for neuropeptide secretory vesicle-associated granin-family molecules were selectively upregulated in AGRP neurons ( Figure 3D ) . Moreover , a range of synaptotagmin-family transcripts , which mediate different aspects of calcium-dependent vesicle release , are selectively regulated in AGRP neurons by food deprivation ( Figure 3E ) . For example , Syt5 , Syt9 , and Syt10 are upregulated in AGRP neurons from FD mice , and these gene products localize to peptidergic vesicles and regulate activity-dependent peptide release ( Cao et al . , 2011 ) . GABA signaling is also important in AGRP neurons , and synaptic vesicle glycoprotein 2C ( Sv2c , +5 . 4-fold , q = 5e−10 ) , which regulates the readily releasable pool ( Xu and Bajjalieh , 2001 ) is increased , as is Snap25 ( +2 . 1-fold , p = 1 . 5e−6 ) , a key SNARE complex component responsible for vesicle fusion . Collectively , these changes in gene expression show some of the molecular underpinnings for processes that mediate increased AGRP neuron output in energy deficit due to alterations of excitatory synaptic inputs , synaptic plasticity , as well as elevated GABA and neuropeptide release . AGRP and POMC neuron electrical properties are determined by expression of distinct groups of ion channels . Despite the importance of neuron electrical activity for influencing appetite and body weight , only a few ion channel subunits have been determined in these cell types and little is known about their regulation by energy deficit state . Previous work has shown transient receptor potential C ( TRPC ) ion channel subunit expression in POMC neurons ( Qiu et al . , 2010 ) , which is confirmed by our RNA-Seq data , and we find that an overlapping set of TRPC channels ( Trpc1 , Trpc3-Trpc7 ) is also expressed in AGRP neurons ( Figure 4A ) . Moreover , Kcnq3 is downregulated in AGRP neurons with food-deprivation ( Figure 4A ) , as previously reported ( Roepke et al . , 2011 ) . Thus , these RNA-Seq data are consistent with established changes in ion channel expression . 10 . 7554/eLife . 09800 . 010Figure 4 . Ion channel gene expression in AGRP and POMC neurons . ( A–C ) Gene expression for voltage-gated ion channels ( A ) , ligand-gated ion channels ( B ) , and other ion channels ( C ) . For each colormap , left , mean expression level [log2 ( TPM ) ] of each transcript of each experimental group . Right , log2 ( fold-change ) and q-values for differential expression between FD and fed states separately for AGRP and POMC neurons . ( D ) Tail currents elicited in AGRP neurons from fed mice by a voltage step from −30 mV to −60 mV in the absence ( n = 5 ) and presence of apamin ( n = 5 ) . Unpaired t-test . p < 0 . 05 . Lines show mean , shading shows s . e . m . ( E ) Firing rate from current injection to AGRP neurons from fed mice in the absence ( n = 8 ) and presence ( n = 13 ) of apamin . Unpaired t-test . ( F ) Example of action potential firing in AGRP neurons from fed mice in response to −10 , 0 , +10 pA in the absence and presence of apamin . ( G ) Tail currents in AGRP neurons from FD mice ( −apamin , n = 10; +apamin , n = 5 ) . Unpaired t-test . n . s . , p > 0 . 05 . ( H ) Firing rate from AGRP neurons from FD mice ( −apamin , n = 4; +apamin , n = 4 ) . Unpaired t-test . ( I ) Example of action potential firing in AGRP neurons from FD mice ( current injections: −10 , 0 , +10 pA ) . DOI: http://dx . doi . org/10 . 7554/eLife . 09800 . 01010 . 7554/eLife . 09800 . 011Figure 4—figure supplement 1 . Burst firing in AGRP neurons with apamin or food deprivation . ( A ) Burst firing and plateau potential from current injection ( +45 pA ) in AGRP neurons from a well-fed mouse after apamin block of SK channels . ( B ) Burst firing and plateau potential from current injection ( +10 pA ) in an AGRP neuron from a FD mouse . DOI: http://dx . doi . org/10 . 7554/eLife . 09800 . 011 This transcriptional profiling resource also identifies other potentially important ion channel genes that cell type-selectively change expression with energetic state and have not been previously investigated in AGRP and POMC neurons ( Figure 4A–C ) . One of the most striking changes in ion channel gene expression in AGRP neurons during food-deprivation was a sharp reduction of Kcnn3 ( also called Sk3 , −5 . 4-fold , q = 0 . 0006 ) , which is a small conductance calcium-activated potassium channel ( SK ) that attenuates action potential firing rate during elevated activity . Other SK channels were not appreciably expressed in AGRP neurons ( Figure 4A ) . Electrophysiological characterization of AGRP neurons showed the presence of SK-mediated tail currents , confirmed by blockade with the highly selective SK channel antagonist , apamin ( Figure 4D ) . Notably , the apamin-sensitive SK conductance was absent in FD mice ( Figure 4G ) , consistent with reduced Kcnn3 expression . Apamin blockade of SK channels in AGRP neurons resulted in higher firing rates ( Figure 4E , F ) , which often elicited bursts and plateau potentials ( Figure 4—figure supplement 1A ) . Similarly , AGRP neurons from FD mice showed a comparable increase in excitability , but in this state the neurons were not sensitive to apamin , consistent with greatly reduced SK-channel expression ( Figure 4H , I and Figure 4—figure supplement 1B ) . Thus , by starting from cell type-specific RNA-Seq analysis , we found that Kcnn3 regulates firing rate as well as burst firing in AGRP neurons , and its reduction with food-deprivation plays an important role in increasing the excitability of AGRP neurons . In addition to the examples highlighted here , this resource shows many other ion channels and regulatory subunits that are altered by energy deficit in AGRP neurons , including ionotropic glutamate receptors , GABA receptors , the ionotropic ATP receptor P2rx4 , sodium channels , calcium channels , additional potassium channels and others ( Figure 4A–C ) . This resource provides a list of ion channels expressed in AGRP and POMC neurons , and it is a foundation for a concrete understanding of the electrical properties of AGRP and POMC neurons in basal and energy deprived states . GPCRs are critical molecular control points for neuronal function . AGRP and POMC neurons express multiple GPCRs , many of which respond to circulating hormones , neuropeptides , or neurotransmitters . RNA-Seq provides a cell type-specific taxonomy of GPCRs expressed under different conditions . For each cell type , many GPCRs were detected at expression levels greater than 20 transcripts per million ( TPM ) under either fed or FD conditions ( AGRP: 59 , POMC: 61 , AGRP or POMC: 80 ) ( Figure 5A ) , which is more than previous estimates of GPCRs for hypothalamic regulation of energy homeostasis ( Schioth , 2006 ) . Also , several GPCRs show high differential expression between AGRP and POMC neurons: 8 GPCRs were >10-fold differentially expressed in AGRP neurons and 13 GPCRs were >10-fold overrepresented in POMC neurons ( Figure 5A ) . Therefore , differential GPCR expression can separately regulate the function of AGRP and POMC neurons . 10 . 7554/eLife . 09800 . 012Figure 5 . G-protein coupled receptors regulated by food-deprivation . ( A , B ) All GPCR genes expressed ( TPM > 20 ) in at least one group , sorted by log2 ( fold-change ) between AGRP and POMC neurons ( A ) or AGRP or POMC neurons FD/fed ( B ) . Bars indicate genes with >10-fold change ( A ) or >twofold change ( B ) . ( C ) Double smFISH for Agrp and Gpr6 . Scale , 10 μm . ( D , E ) Population counts ( bars: mean values ) ( D ) and cumulative probability distributions ( E ) of Gpr6 puncta per cell volume in AGRP neurons ( p = 4 . 8e−10 , ks-test ) . Fed , n = 115; FD , n = 122; 3 mice per condition . ( F ) Cre-dependent viral vector for cell type-specific Gpr6 overexpression in AGRP neurons . hSyn: synapsin promoter . Black and white triangles denote heterotypic loxP sites for stable inversion of Gpr6-IRES-GFP . ( G ) Schematic for viral transduction and cell-type specific overexpression of Gpr6 in AgrpCre mice . ( H ) Representative image showing Gpr6-IRES-GFP-transduced AGRP neurons . Scale , 100 µm . ( I ) Body weight change from pre-injection weight ( starting age: 8 weeks ) in AgrpCre mice expressing Gpr6-IRES-GFP or BFP ( 2-way ANOVA , one factor repeated measures , transgene: F1 , 65 = 19 . 6 , p < 0 . 001; time: F5 , 65 = 30 . 1 , p < 0 . 001; interaction: F5 , 65 = 4 . 2 , p = 0 . 002 ) . Holm-Sidak correction for multiple comparisons . AGRPGpr6 n = 9 mice , AGRPBFP n = 7 mice . Data is mean ± s . e . m . p < 0 . 05 , p < 0 . 01 , p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 09800 . 01210 . 7554/eLife . 09800 . 013Figure 5—figure supplement 1 . Opposite differential expression of Hrh3 in AGRP and POMC neurons after food deprivation . ( A–F ) Representative images of double smFISH for Hrh3 and Agrp ( A ) or Pomc ( B ) . Scale , 10 μm . Population counts ( bars , mean values ) ( B , D ) and cumulative probability distributions ( C , E ) of Hrh3 puncta per cell volume in AGRP and POMC neurons ( p = 3 . 4e−34 , p = 0 . 0037 , respectively , ks-test ) . *p < 0 . 05 . AGRP fed , n = 159 cells; AGRP FD , n = 143 cells; POMC fed , n = 36 cells; POMC FD , n = 44 cells; 3 mice per condition . DOI: http://dx . doi . org/10 . 7554/eLife . 09800 . 013 GPCR expression was significantly overrepresented in genes that were differentially expressed after food deprivation in AGRP and POMC neurons ( AGRP: p = 0 . 0002 , POMC: p = 3 . 3e−5 , hypergeometric-test ) . For AGRP neurons , 13 genes were >twofold upregulated and 11 genes were >twofold downregulated by food deprivation ( Figure 5B ) . All significantly upregulated GPCRs were either Gq-protein coupled ( e . g . , Ghsr , Nmur2 , Gpr83 , Htr2a , Nmbr , Hcrtr2 ) or Gs-protein coupled receptors , several of which show high ligand-independent basal Gs-protein activity ( Gpr3 , Gpr6 , Gpr64 ) . The most strongly downregulated genes were Gi-protein coupled receptors ( Npy2r , Hrh1 , Hrh3 , Htr1a ) . Conversely , POMC neurons showed the opposite pattern with six differentially regulated GPCR transcripts where all upregulated genes were Gi-protein coupled ( 4/6 ) and the downregulated genes were Gq- or Gs-protein coupled ( 2/6 ) . For example , two GPCRs were reciprocally regulated in AGRP and POMC neurons with food-deprivation , most strikingly Gi-protein-coupled Hrh3 ( AGRP: −9 . 5-fold , q = 4 . 7e−8; POMC: +5 . 3-fold , q = 0 . 003; Figure 5—figure supplement 1 ) . Taken together , these changes reveal an opposite response pattern during food deprivation for GPCR expression levels in AGRP and POMC neurons , highlighting the importance of cell type-specific transcriptional profiling for assessing gene expression regulation in the brain . Because Gq-protein signaling increases AGRP and POMC neuron activity and Gi-protein signaling reduces activity in these cell types ( Krashes et al . , 2011; Atasoy et al . , 2012 ) , the pattern of GPCR expression is consistent with elevated AGRP neuron activity and reduced POMC neuron activity during food deprivation . In addition to Gq- and Gi-protein coupled signaling , elevated Gs-protein coupled signaling was prominent in AGRP neurons , especially upregulation of Gpr3 , Gpr6 , and Gpr64 , which are constitutively active Gs-coupled receptors ( Uhlenbrock et al . , 2002 ) . Moreover , the Gs subunit Gnas was upregulated selectively in AGRP neurons ( AGRP: +1 . 8-fold , q = 0 . 0002 , POMC: −1 . 1-fold , q = 0 . 73 ) . However , the physiological consequences of Gs-protein coupled signaling in AGRP neurons are not well understood . We extended our RNA-Seq observations by examining Gpr6 expression using quantitative smFISH , which showed a significant increase in the number of Gpr6 transcripts in AGRP neurons after food deprivation ( p = 4 . 8e−10 , ks-test , Figure 5C–E ) . To test the functional consequence of elevated Gpr6 expression in AGRP neurons , we transduced AgrpCre mice with a Cre-dependent virus co-expressing Gpr6 and a fluorescent protein ( AGRPGpr6 mice; Figure 5F–H ) . AGRPGpr6 mice showed significantly elevated body weight compared to mice expressing a fluorescent protein alone ( Figure 5I ) . These experiments reveal a potential role for Gpr6 and Gs-coupled signaling in AGRP neurons for positive regulation of body weight . Neuropeptide expression is typically used to discriminate AGRP and POMC neurons . Transcriptional profiling shows that these two cell populations are distinguished by several additional secreted proteins ( Figure 6A ) . Neuropeptides were also among the most highly regulated genes in response to food-deprivation in AGRP neurons ( Figure 6B ) . Several neuropeptide transcripts with increased expression in our dataset have been previously shown to increase appetite , for example , Agrp , Npy , Vgf , Pdyn . We also observed increased expression of peptides that are associated with reduced appetite , based on pharmacological experiments: Nmb ( +8 . 0-fold , q = 6 . 0e−5 ) , Nts ( +107-fold , q = 0 . 0001 ) , Nucb2 ( +2 . 3-fold , q = 0 . 0002 ) . However , these neuropeptides may have a local signaling role that promotes appetite . For example , Nmbr ( +8 . 8-fold , q = 0 . 0008 ) and Ntsr1 ( +41-fold , q = 0 . 005 ) were also strongly upregulated in AGRP neurons , are Gq-coupled receptors , and Nmb has been previously demonstrated to increase AGRP neuron electrical activity ( van den Pol et al . , 2009 ) . 10 . 7554/eLife . 09800 . 014Figure 6 . Secreted proteins regulated by food-deprivation . ( A , B ) Secreted protein genes differentially expressed between AGRP and POMC neurons ( A ) or AGRP or POMC neurons FD/fed ( B ) . Genes mentioned in the text are labeled with an asterisk . ( C ) Double smFISH for Agrp and Ccl17 . Scale , 10 μm . ( D , E ) Population counts ( bars: mean values ) ( D ) and cumulative probability distributions ( E ) of Ccl17 puncta per cell volume in AGRP neurons ( p = 1 . 5e−20 , ks-test ) . Fed , n = 144; FD , n = 230; 3 mice per condition . ( F ) Food intake at start of light period 1-hr and 2-hr after intracerebroventricular injection of either saline or recombinant CCL17 ( 500 ng ) . Rank-sum test . ( G ) Cre-dependent viral vector for cell type-specific Ccl17 overexpression in AGRP neurons . hSyn: synapsin promoter . Black and white triangles denote heterotypic loxP sites for stable inversion of Ccl17-IRES-GFP . ( H ) Schematic for viral transduction and cell-type specific overexpression of Ccl17 in the brains of AgrpCre mice . ( I ) Representative image showing Ccl17-IRES-GFP-transduced AGRP neurons . Scale , 100 µm . ( J ) Body weight change from pre-injection weight ( starting age: 8 weeks ) in AgrpCre mice expressing Ccl17-IRES-GFP or BFP ( 2-way ANOVA , one factor repeated measures , transgene: F1 , 65 = 12 . 0 , p = 0 . 004; time: F5 , 65 = 14 . 8 , p < 0 . 001; interaction: F5 , 65 = 14 . 8 , p < 0 . 001 ) . BFP data is same as Figure 5I . Holm-Sidak correction for multiple comparisons . AGRPCcl17 n = 9 mice , AGRPBFP n = 7 mice . Data is mean ± s . e . m . n . s . , p > 0 . 05 , p < 0 . 05 , p < 0 . 01 , p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 09800 . 014 Other genes were upregulated by energy deficit that have not been previously associated with appetite regulation , for example Ccl17 ( +21 . 6-fold , q = 0 . 0012 ) and Thbs1 ( +12 . 4-fold , q = 0 . 0002 ) . We further investigated the chemokine CCL17 , which was highly upregulated in AGRP neurons . CCL17 is a 103 aa member of the CC chemokine group with no previously reported functional characterization in the central nervous system . Assessment of Ccl17 expression using double smFISH in AGRP neurons confirmed food deprivation-induced upregulation ( p = 1 . 4e−20 , ks-test , Figure 6C–E ) . Microinjection of recombinant CCL17 ( 500 ng ) into the lateral cerebral ventricle elicited a small increase of food intake during the light period ( 1-hr: p = 0 . 02 , 2-hr: p > 0 . 05; rank sum-test , Figure 6F ) . More strikingly , chronic viral overexpression of Ccl17 and GFP in adult mice by selective expression in AGRP neurons ( AGRPCcl17 mice ) resulted in a progressive elevation of weight gain compared to mice expressing a fluorescent protein alone ( Figure 6J ) , providing preliminary evidence that CCL17 may play a role in regulating body weight . Conversely , we considered the possibility that genes coding for secreted proteins with anorexigenic properties were selectively repressed in AGRP neurons after food-deprivation . For example , Fgf1 , which has been shown to reduce appetite ( Sasaki et al . , 1991 ) , was strongly downregulated ( −4 . 1-fold , q = 0 . 0004 ) . To explore this idea , we tested a number of secreted proteins that were downregulated in AGRP neurons after food deprivation and previously had not been examined for regulation of food intake . Based on these criteria , we selected four peptides for further analysis: Pleiotrophin ( Ptn , −2 . 3-fold , q = 0 . 005 ) , a heparin-binding cytokine , which inhibits receptor protein tyrosine phosphatase β/ζ; autotaxin ( Enpp2 , −2 . 2-fold , q = 0 . 00013 ) , a secreted enzyme that converts lysophosphatidylcholine into the lipid second messenger lysophosphatidic acid; cerebellin 4 ( Cbln4 , −7 . 6-fold , q = 2 . 9e−5 ) , a transneuronal regulator of synaptic function; and bone morphogenic protein 3 ( Bmp3 , −17 . 7-fold , q = 2 . 0e−5 ) , a member of the TGFβ superfamily ( Figure 6B ) . Based on smFISH , we confirmed that Cbln4 and Bmp3 were downregulated in AGRP neurons after food-deprivation ( Cbln4: p = 6 . 8e−17 , Bmp3: p = e −12 , ks-test; Figure 7A–F ) . To test whether these proteins influence food intake , we delivered them by intracerebroventricular injection . All four proteins significantly reduced food intake over a 24-hr period ( Figure 7G ) , consistent with reduction of their expression levels during food deprivation . Therefore , cell type-specific RNA-Seq is also effective for identifying new secreted proteins that regulate appetite . 10 . 7554/eLife . 09800 . 015Figure 7 . Secreted proteins that are downregulated in AGRP neurons with food-deprivation reduce food intake . ( A–F ) Double smFISH for Agrp and ( A ) Bmp3 or ( D ) Cbln4 . Scale , 10 μm . Population counts ( B , E ) and cumulative probability distributions ( C , F ) for Bmp3 and Cbln4 ( p = 9e−12 and p = 6 . 8e−17 , ks-test ) . Bmp3 fed , n = 142; Bmp3 FD , n = 157 cells; Cbln4 fed , n = 145; Cbln4 FD , n = 132; 3 mice per condition . ( G ) Mean food intake ( 24 hr ) after intracerebroventricular injection of either saline or recombinant BMP3 ( ANOVA , F2 , 24 = 7 . 5 , p = 0 . 003 ) , CBLN4 ( ANOVA , F2 , 24 = 8 . 0 , p = 0 . 002 ) , ENPP2 ( ANOVA , F2 , 27 = 7 . 3 , p = 0 . 003 ) , or Pleiotrophin ( unpaired t-test ) . Holm-Sidak correction for multiple comparisons . Data is mean ± s . e . m . n . s . , p > 0 . 05 , p < 0 . 05 , p < 0 . 01 , p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 09800 . 015 The molecular processes governing the function of AGRP and POMC neurons have been a central focus for understanding energy homeostasis and for identifying new approaches to influence appetite and body weight in humans . This cell type-specific RNA-Seq resource from AGRP and POMC neurons reveals an extensive program of gene expression changes selectively in AGRP neurons during energy deficit associated with increased protein translation and folding , circadian gene expression , increased neuronal activity and synaptic release of neurotransmitter and neuropeptides , and alterations in secreted protein and cell surface receptor expression . POMC neurons show a much smaller number of changes , and some of these are associated with reduced activity , as would be expected from suppression of POMC neuron function under energy deficit conditions . Therefore , AGRP neurons are much more sensitive to energy deficit states than POMC neurons . Differential gene expression in AGRP neurons after food deprivation is also considerably greater than has been observed in other cell types after various perturbations of brain state . Miller et al . ( 2011 ) compared gene expression in fast-spiking GABAergic interneurons from 48-hr muscimol-treated and saline-treated sides of motor cortex , which led to dramatic alteration in firing properties of these neurons but only 13 DEG ( q < 0 . 05 ) . We observed >160-fold more DEG in AGRP neurons after food-deprivation using the analysis criteria as Miller et al . ( 2011 ) . Another strong neuronal perturbation , genetic ablation of the important transcriptional repressor Mecp2 , only resulted in 10–50% of the number of DEG in various cell types ( Sugino et al . , 2014 ) compared to AGRP neurons from FD mice using the same analysis criteria . Finally , RNA-Seq analysis of motor neurons in an amyotrophic lateral sclerosis mouse model showed only 62 genes differentially expressed ( q < 0 . 05 ) relative to wildtype motor neurons ( Bandyopadhyay et al . , 2013 ) , which is 34-fold fewer differentially expressed transcripts than from AGRP neurons under food deprivation using the same analysis criteria . The high number of DEG in our dataset is influenced by our study's greater statistical power , use of RNA-Seq ( instead of microarrays ) , as well as high purity and more narrowly defined cell types . However , the relatively small number of DEG measured in POMC under similar experimental conditions indicates that the magnitude of changes in gene expression in AGRP neurons is also based on a selectively tuned response in these neurons to the energy deficit physiological state . Therefore , our results show that neuronal gene expression in vivo can have much greater dynamic changes from alteration of genetic , physiological , or behavioral state than previously reported . Many DEG were consistent with ER-stress responses associated with the transition to elevated protein translation in AGRP neurons during periods of increased neuropeptide release . Indeed , secreted proteins are among the most abundant transcripts in AGRP neurons and their abundance increases further during energy deficit . We observed increased splicing of the UPR regulator Xbp1s , as well as multiple protein folding chaperones , transcripts encoding protein degradation machinery , and oxidative stress signaling molecules during food deprivation . This adaptive cellular response to food deprivation in AGRP neurons is different to what has been found using whole hypothalamus analysis , where UPR signaling has been primarily associated with overnutrition states ( Ozcan et al . , 2009 ) . Interestingly , overexpression of Xbp1s in POMC neurons has been shown to facilitate the function of those neurons ( Williams et al . , 2014 ) , and a similar effect may be operating in AGRP neurons . In addition , we found that anti-apoptotic pathways were also prominently increased in AGRP neurons during food-deprivation , which is also indicative of considerable cellular stress and the importance of protecting against cell death for this critical energy homeostasis neuron population . We also found reduced expression of E-box-regulated circadian genes during energy deficit in AGRP neurons . AGRP neurons contribute to increased locomotor activity during scheduled feeding ( Tan et al . , 2014 ) , which is associated with the food-entrained circadian rhythm . Moreover , E-box genes are expressed primarily during the light period ( Rey et al . , 2011 ) , when mice normally eat little , and suppression of these genes is similar to the expression level of these genes at night , a time when mice consume the most food . The cause and downstream consequences of this ‘night-like’ pattern of circadian gene expression are an important area for further investigation into the circadian control of AGRP neurons , but transcriptional repression involving Sfpq ( Psf ) ( Duong et al . , 2011 ) is a candidate molecule for this pathway . GPCR gene expression was also substantially altered by food-deprivation in both AGRP and POMC neurons . For AGRP neurons , this was largely associated with Gq- and Gs-coupled GPCR upregulation as well as reduced Gi-coupled GPCRs . This is in line with activation of AGRP neurons by Gq-protein-coupled signaling pathways through the Gq-coupled DREADD hM3Dq ( Krashes et al . , 2011 ) . In addition , overexpression of the constitutively active Gs-protein-coupled Gpr6 in AGRP neurons led to a significant increase in body weight . A similar role may extend to other constitutively active Gs-protein-coupled receptors that are upregulated by energy deficit , such as Gpr3 and Gpr64 , and this effect is likely enhanced by the concomitant fall in Gi-protein coupled receptor expression . Interestingly , we noted that Hrh1 and Hrh3 are sharply reduced in AGRP neurons during energy deficit , indicating reduced responsiveness to histamine . Because GPCRs are of considerable interest as targets for drug development , this resource of AGRP and POMC neuron-expressed GPCRs provides multiple possibilities for modulating the function of these key energy homeostasis-regulating neuron populations . Neuropeptide genes were strongly differentially expressed in AGRP neurons . Ccl17 has not been functionally investigated in the brain , but it was highly increased by food-deprivation as confirmed by RNA-Seq and smFISH , and overexpression in AGRP neurons increased body weight , consistent with a role in energy homeostasis . Notably , several neuropeptide genes that were strongly downregulated reduced appetite when injected in the lateral cerebral ventricle . For example , the neuropeptide CBLN4 strongly reduced food intake , and Cbln4 has been implicated in regulating inhibitory synapse function and is thought to play a neuroprotective role ( Chacon et al . , 2015 ) . In addition , PTN is an inhibitor of receptor tyrosine phosphatase β/ζ , and its reduction might reduce cytokine signaling through tyrosine kinase receptors , which include leptin and insulin receptors . Another important application of cell type-specific RNA-Seq data is to investigate the electrical properties of neurons . A list of expressed ion channels and how they are altered with food-deprivation provides a molecular framework for electrical activity in these neurons . Ex vivo and in vivo recordings of AGRP neuron activity have shown elevated activity and indicated the importance of burst firing in AGRP neurons during energy deficit ( Betley et al . , 2015 ) . Here , we show that an important contributor to AGRP neuron excitability and burst firing in FD mice is reduced expression of the SK3 channel transcript Kcnn3 . In fed mice , this calcium-activated potassium channel reduces electrical activity during elevated firing , and we find that SK-channel blockade in AGRP neurons promotes burst firing in brain slices . Taken together , the data described in this resource provide a framework for extensive analysis of energy homeostasis circuits as well as other cell type-specific circuit nodes in the brain . Transcriptomic profiling was performed only on male mice , and future work should examine gene expression in female mice . However , these data allow insights into the cell biology of energy deficit-responsive neurons and highlight the remarkable specificity of gene expression responses of AGRP and POMC neurons . Moreover , in light of the intensive focus on new therapies for obesity as well as undereating disorders ( Gautron et al . , 2015 ) , this detailed resource of molecular components in key energy homeostasis neurons will be valuable to researchers aiming to devise small molecule and peptide modulators of these circuits . In addition , together with the rapid increase in known genetic variants associated with obesity ( van der Klaauw and Farooqi , 2015 ) , cell type-specific RNA-Seq data from these and numerous other populations essential for energy homeostasis may strengthen understanding of the cell type basis of obesity found in the human population . Mice were housed on a 06:00–18:00 light cycle with water and mouse chow ad libitum ( PicoLab Rodent Diet 20 , 5053 tablet , TestDiet , St . Louis , MO , United States ) unless otherwise noted . The following mouse lines were used: AgrpCre ( Jackson Labs Stock 012899 , Agrptm1 ( cre ) Lowl/J ) , Ai9 ( ROSA-loxPStoploxP-tdTomato , Jackson Labs Stock 007909 ) , PomctopazFP ( Jackson Labs Stock 008322 ) , and NpyhrGFP ( Jackson Labs Stock 006417 ) . Young adult male mice ( >6 . 5 weeks old ) were used for experiments . For tdtomato expression in AGRP neurons , AgrpCre mice were crossed to Ai9 mice . AGRP and POMC neurons were obtained by sorting fluorescent neurons form the ARC of NpyhrGFP mice and PomctopazFP mice , respectively . Male mice ( age 6 . 5–8 weeks ) were used for the experiments . Both control and experimental mice were separated into individual fresh cages a day before experiment around noon . For mice in the 24-hr food deprivation condition , only water was provided . On the next day , between 11:00 and 13:00 , mice were sacrificed and labeled neurons were manually sorted as described previously ( Hempel et al . , 2007 ) . Briefly , a horizontal hypothalamic slice ( 300 μm thick ) containing the ARC was obtained using Leica Vibratome VT1200S . After digesting with 1 mg/ml Pronase ( P5147 , Sigma–Aldrich , St . Louis , MO , United States ) for 1 hr at room temperature , the ARC was dissected from the slices and triturated using three Pasteur pipettes with decreasing tip sizes in artificial cerebrospinal fluid ( ACSF ) with 1% FBS ( 1 ml ) . Triturated cells were diluted ( 25 ml ) and poured into a 100 mm Petri dish . After cells settled ( 5–10 min ) , labeled neurons were picked using a glass pipette ( 30–50 μm tip size ) and transferred to a clean 35 mm dish containing ACSF with 1% FBS ( 2 ml ) . This manual sorting process was repeated two additional times and the final sorted neurons were transferred to a PCR tube containing extraction buffer XB ( 47 μl ) from PicoPure Kit ( KIT0204 , Life Technologies , Carlsbad , CA , United States ) , incubated ( 42°C , 30 min ) , and the mixture was stored at −80°C until library preparation . The whole process was complete 3–3 . 5 hr after sacrifice . RNA was extracted according to PicoPure manufacturer's instructions . Either 1 μl of 10−5 dilution of External RNA Controls Consortium ( ERCC ) spike-in control ( Life Technologies , #4456740 ) or ( number of sorted cells/50 ) × ( 1 μl of 10−5 dilution of ERCC ) was added to the purified RNA and speed-vacced down to 5 μl and immediately processed with reverse transcription by NuGEN Ovation RNA-Seq System V2 ( #7102 , NuGEN , San Carlos , CA , United States ) which yielded 4–8 μg of amplified DNA . This amplified DNA was fragmented to average size of ∼200 bp using Covaris E220 . Then , NuGEN Encore NGS Multiplex System I kit ( NuGEN , 0314 ) was used to prepare for sequencing with Illumina HiSeq2500 . Libraries were sequenced with either fourfold or eightfold multiplexing in one or two lanes . In total six lanes and four sequencing runs were used for the data collection , and lanes were mixed with samples from fed and FD mice as well as from AGRP and POMC neurons . On average 53 million of 100 bp single end reads were obtained per sample ( range: 26–97 million reads ) . Of these , on average 68% mapped uniquely to University of California , Santa Cruz ( UCSC ) mm10 genome , 2 . 2% mapped non-uniquely , 4 . 6% were unmappable and 25% mapped to abundant sequences such as ribosomal RNA , mitochondrial or phiX sequences . RNA-Seq data is available at Gene Expression Omnibus ( GEO ) ( accession number GSE68177 ) . Adaptor sequences ( AGATCGGAAGAGCACACGTCTGAACTCCAGTCAC for Illumina sequencing and CTTTGTGTTTGA for NuGEN SPIA ) were removed from de-multiplexed FASTQ format data from Illumina HiSeq2500 using cutadapt v1 . 7 . 1 ( http://dx . doi . org/10 . 14806/ej . 17 . 1 . 200 ) with parameters ‘--overlap = 7 --minimum-length = 30’ . Then abundant sequences ( ribosomal RNA , mitochondrial , Illumina phiX and low complexity sequences ) were detected using bowtie2 ( Langmead and Salzberg , 2012 ) v2 . 1 . 0 with default parameters . The remaining reads were mapped to mm10 genome using STAR ( Dobin et al . , 2013 ) v2 . 4 . 0i with parameters ‘--chimSegmentMin 15 --outFilterMismatchNmax 3’ . Uniquely mapped reads were then assigned to genes using HTSeq ( Anders et al . , 2015 ) v0 . 6 . 1p1 with parameters ‘-s no -m intersection-nonempty’ . For genome annotation , we used GencodeVM4Basic downloaded from UCSC genome browser , which included 35 , 266 genes . Fragments per kilobase of transcript per million mapped reads ( FPKM ) for a gene was calculated by dividing counts of reads assigned to the gene by the sum of the length ( in kb ) of all the exons belonging to the gene and then normalized by library size ( in millions ) . FPKM was then transformed to TPM by dividing by the sum of all FPKM values and multiplying with 1e6 . Expression levels in figures are TPM . ERCC analysis ( similar to Zeisel et al . , 2015 ) indicated we had 50% detection rate at 22 copy × kb of ERCC spike-ins in a tube . Because we had on average 102 cells in a tube ( minimum 44 ) , this suggests we had at least 50% detection rate of all the transcripts larger than 1 kb , even if there was only 1 transcript/cell . We also estimated that 1 TPM corresponded to 3 . 2 ± 1 . 9 copies of transcripts/cell from the ERCC data . This was done using a linear fit between log ( ERCC TPM ) and log ( ERCC copy number ) , ratio of total reads between ERCC and all the genes and number of cells used . To detect DEG , we used limma-voom packages ( Law et al . , 2014; Ritchie et al . , 2015 ) . Raw read counts for genes with counts per million ( CPM ) value larger than 1 in at least three samples ( 16 , 513 genes ) were used as inputs to limma-voom package . Trimmed mean of M-values ( TMM ) normalization method ( Robinson and Oshlack , 2010 ) was used for normalization and the Benjamini and Hochberg method was used for adjusting for multiple tests . Adjusted p-values ( q-values ) are reported throughout the paper . The log-ratio ( coefficients ) outputs from limma package was used as adjusted log2 ( fold-change ) [log2 ( fc ) ] and reported instead of simple log2 ( fold-change ) calculated from raw TPM values ( differences between coefficients and raw TPM fold-change values are due to weights assignment to each observation calculated by the voom package ) . Reported DEG were obtained by requiring q-value <0 . 05 , abs[log2 ( fc ) > 1] and mean CPM > 20 in at least 1 cell type/condition . Post-hoc power analysis using the ‘RNASeqPower’ package for R ( Hart et al . , 2013 ) indicates that for our sample size ( 5–6 mice per group ) , 80% of genes can be detected that have at least a twofold change in expression . To evaluate this pipeline of DEG calculation , we randomly permuted assignment of four conditions ( Agrp . FD , Agrp . fed , Pomc . FD , Pomc . fed ) on 21 samples and recalculated DEG . With 1000 permutations , there were 1 . 7 ± 25 ( mean ± sd ) DEG for AGRP . fed vs AGRP . FD and 0 . 77 ± 16 for POMC . fed vs POMC . FD and 0 . 64 ± 14 for AGRP . fed vs POMC . fed , yielding empirical false discovery rate ( FDR ) of 0 . 2% , 1 . 6% and 0 . 1% respectively . The majority of permutations resulted in zero DEG with only biased permutations yielding any DEG . To assess contamination of non-neuronal cell types , we used RNA-Seq data by Zhang et al . ( 2014a ) . Raw RNA-Seq data was downloaded from GEO ( accession number GSE52564 ) and processed as for the current data . Four specifically and highly expressed genes from each of the previously described non-neuronal cell types ( Zhang et al . , 2014a ) were used to assess potential contamination by these cell types . All samples were processed together with apt-probeset-summarize in Affymetrix Power Tools version 1 . 17 . 0 with ‘–a rma-sketch’ option . Using ‘MoEx-1_0-st-v1 . r2 . dt1 . mm9 . core . mps’ provided by Affymetrix as meta-probesets ( i . e . , gene expression value summaries were obtained ) . MDS was used to visualize gene expression differences between samples . Pseudo-distance , 1 minus correlation coefficient , was used as an input for MDS . Only genes with CPM > 20 in at least 1 cell type/condition were used for this calculation ( 8198 genes met this criterion ) . The dataset with ( adjusted ) log-fold-change ( lfc ) and q-values were uploaded to the Ingenuity Pathway Analysis ( IPA ) server and genes with q < 0 . 05 and abs ( lfc ) > 1 ( AGRP: 1276 , POMC: 53 genes ) were used to evaluate pathway enrichment using IPA core analysis option . Core circadian reference genes and E-box target genes shown in Figure 3 were described previously ( Rey et al . , 2011 ) . E-box targets were selected with criteria: conservation score ≥0 . 9 , cycling score ( sum of ZT2-ZT22 score ) ≥160 and number of associated E-boxes ( either E1 or E1–E2 ) ≥1 . Out of 16 , 513 genes that had read counts in AGRP neurons , 113 were associated with E-boxes , 1346 genes satisfied qval < 0 . 05 and abs[log2 ( fc ) ] > 1 and of these 20 were associated with E-box . From these numbers , overrepresentation probability of food deprivation affected E-box genes was calculated using the hypergeometric test . Even when we did not threshold E-box genes with conservation score and cycling score , E-box overrepresentation in AGRP food deprivation affected genes ( qval < 0 . 05 , abs[log2 ( fc ) ] > 1 ) was significant ( p = 0 . 0006 ) . If genes with smaller fold-changes ( q < 0 . 05 , abs[log2 ( fc ) ] > 0 ) were included , statistical significance was further improved ( p = 1 . 5e−6 ) . IUPHAR database ( Pawson et al . , 2014 ) was used to obtain a list of GPCRs and ion channels . Ion channel classification is based on IUPHAR assignment . Secreted Protein Database ( SPD ) ( Chen et al . , 2005 ) and Gene ontology ( GO ) annotation were used to create a list of secreted proteins . Only genes with confidence level ≤2 from SPD that also intersected with genes which had GO annotation of ‘extracellular region’ ( GO: 0005576 ) were used . Genes associated with GO: 0005576 were obtained using the QuickGO website . KEGG PATHWAY mmu04141 was used to obtain a list of ER-associated genes . QuickGO was used to obtain genes that encode proteins localized to ‘Synapse ( GO: 0045202 ) ’ and proteins with ‘Sequence-specific DNA binding transcription factor activity ( GO: 0003700 ) ’ . Previously reported lists of mouse kinases ( Caenepeel et al . , 2004 ) and phosphatases ( Sacco et al . , 2012 ) were used . To assess intracellular localization of the UPR-related protein ATF6 , the DNA encoding enhanced green fluorescent protein ( EGFP ) :ATF6 ( Addgene , Plasmid #32955 ) was inserted into a rAAV2-hSynapsin-FLEX ( FLEX: flip-excision ) vector in the inverted orientation to create the Cre-dependent viral expression vector ( serotype 1 ) rAAV2/1-hSyn-FLEX-rev-EGFP:Atf6 ( 6 . 4e12 Genomic Copies ( GC ) /ml ) . For experiments involving Cre-dependent overexpression of Ccl17 or Gpr6 in AGRP neurons , an Origene synthesized fragment containing either Ccl17 or Gpr6 cDNA from mouse was cloned into a rAAV2-hSynapsin-FLEX vector in an inverted orientation , yielding the Cre-dependent viral expression vectors rAAV2/1-hSyn-FLEX-rev-Ccl17-IRES-EGFP ( 3 . 3e12 GC/ml ) , and rAAV2/1-hSyn-FLEX-rev-Gpr6-IRES-EGFP ( 7 . 1e12 GC/ml ) . Cre-dependent expression of blue fluorescent protein ( BFP ) used rAAV2/1-hSyn-FLEX-rev-BFP ( 1 . 9e13 GC/ml ) . FLEX , Cre-dependent flip-excision switch ( Atasoy et al . , 2008 ) . Viral vectors were produced by the Janelia Farm Molecular Biology Core Facility . For transgene expression in AGRP neurons , male mice 8 weeks of age or older were anaesthetized with isoflurane , and placed into a stereotaxic apparatus ( David Kopf Instruments ) . After introducing a small incision to expose the skull surface , small holes were drilled in skull for immediate viral injections and/or cannula implantation . rAAV was delivered via a pulled glass pipette with diameter between 20 and 40 μm . For targeted rAAV delivery to the ARC , bilateral injections were made at two depths using the coordinates: bregma −1 . 3 mm; midline ±0 . 25 mm; dorsal brain surface −6 . 0 mm and −5 . 90 mm in AgrpCre mice . For experiments involving long term ( 6 weeks ) monitoring of body weight small volume injections ( 50 nl total injected at each of the two sites ) were performed to avoid potential long-term effects on body weight associated with large volume viral injections in this brain region . Mice injected with rAAV2/1-hSyn-FLEX-rev-EGFP:Atf6 were used for experiments 7 days post-infection; transgene expression for significantly longer periods of time resulted in aberrant localization of the EGFP signal to the nucleus under baseline conditions . For experiments involving acute administration of recombinant peptides into the brain , a craniotomy was drilled over the right lateral ventricle and a cannula was implanted at the coordinates: bregma −0 . 58 mm , midline +1 . 25 mm , skull surface −2 . 0 mm; Grip cement ( DENTSPLY ) was used to anchor the cannula to the skull . For all animal surgeries , postoperative analgesia was provided . Buprenorphine was administered intraperitoneally ( 0 . 1 mg/kg ) along with ketoprofen administered subcutaneously ( 5 mg/kg ) . Rabbit anti-GRP78 ( also called BiP; 1:4000 , Novus Biologicals , San Diego , CA , United States ) , monoclonal mouse anti-TDP43 ( 1:32 , 000 , Abcam , San Francisco , CA , United States ) , guinea pig anti-BMAL1 ( 1:15 , 000 , Millipore , Billerica , MA , United States ) , and sheep anti-GFP ( 1:3 , 000 , AbD Serotec , Raleigh , NC , United States ) were used . Species appropriate , fluorophore-conjugated , minimally cross reactive secondary antibodies were obtained from Jackson Immuno ( West Grove , PA , United States ) and used at a concentration of 1:500 . Specificity of anti-BiP antibody was previously verified via siRNA knockdown ( Kitahara et al . , 2011; Maddalo et al . , 2012 ) . Specificity of anti-TDP43 antibody was previously verified via shRNA knockdown ( Lee et al . , 2015 ) . Specificity of anti-Bmal1 antibody was verified in Bmal−/− tissue ( LeSauter et al . , 2012 ) . Mice were transcardially perfused with 4% paraformaldehyde ( PFA ) in 0 . 1 M phosphate buffer fixative ( pH 7 . 4 ) . Brains were postfixed in this solution ( 3–4 hr ) and washed overnight in phosphate buffered saline ( PBS ) ( pH 7 . 4 ) . Brain sections ( 50 μm thick ) were incubated ( 24–48 hr , 4°C ) with primary antibodies diluted in PBS , supplemented with 1% bovine serum albumin ( BSA ) and 0 . 1% Triton X-100 . Slices were washed three times and incubated with species appropriate secondary antibodies ( 2 hr , room temperature ) and mounted using VECTASHIELD ( Vector Laboratories , Burlingame , CA , United States ) hard set mounting medium with DAPI ( 4′ , 6-diamidino-2-phenylindole ) . Images were collected by confocal microscopy ( Zeiss 510 , Zeiss 710 , and Nikon A1R ) , using identical imaging conditions for each experimental group . Fed and FD groups used same conditions as for RNA-Seq data . Analysis of BiP antibody staining intensity in somatic regions of AGRP and POMC neurons was performed using the software program FIJI on maximal intensity z-projected ( 10 μm ) image stacks . Immunofluorescence intensity was calculated from the average pixel intensity value contained within a 80 pixel × 80 pixel circular region of interest placed over the somatic region of each cell . TDP43-associated granule abundance in individual AGRP neurons ( Figure 2F ) was measured on 5 μm projections from confocal images acquired with a 63× objective , using the StarSeach analysis tool ( http://rajlab . seas . upenn . edu/StarSearch/launch . html ) with a threshold setting of 50 . Quantification of nuclear to cytoplasmic ratio of the GFP:ATF6 signal ( Figure 2H ) was performed using FIJI . Nuclear GFP fluorescence intensity was quantified on a per cell basis from a single optical section ( as defined by a binarized mask created from the corresponding DAPI signal ) , then subtracted from the total somatic integrated intensity for each cell ( as defined by a binarized mask created from the corresponding tdtomato signal ) to yield the uniquely cytoplasmic intensity signal . Nuclear and cytoplasmic values were normalized by area before being used to calculate the relative ratio of nuclear to cytoplasmic fluorescence intensity ratio for each infected cell . To verify efficacy of reporter , rAAV2/1-hSyn-FLEX-rev-EGFP:Atf6 injected mice were implanted with a cannula over the right lateral ventricle and allowed 1 week of post-surgical recovery time before intracerebroventricular injection with either DMSO or the UPR-inducing agent tunicamycin ( 40 mg/ml , 1 μl total volume ) . Mice were perfused 24-hr post injection to assess potential alterations in nucleus:cytoplasm GFP ratio . Initial attempts to quantify somatic Arntl/Bmal immunofluorescence in z-projected confocal stacks or single optical sections produced considerable variability under baseline conditions that was likely due to differences in fixation quality . To ameliorate this issue , we acutely sliced brain sections of 250 µm thickness ( using standard brain slicing procedures [Atasoy et al . , 2012] for electrophysiological recordings ) , and these were fixed via submersion in 4% PFA in 0 . 1 M phosphate buffer fixative ( pH 7 . 4 ) for 1-hr at room temperature . These slices were then washed and subjected to immunofluorescence staining procedures similar to those described earlier . We restricted analysis to the top 20 µm of the slice . Analysis of Arntl/Bmal immunofluorescence intensity in somatic regions of AGRP neurons ( identified from NpyhrGFP mice ) , was performed using custom MATLAB scripts ( Source code 1 ) to automatically detect the optical section of maximal intensity in the EGFP fluorescence channel for a given neuron , then to return the average pixel value of the Arntl/Bmal immunofluorescence channel within a 60 pixel × 60 pixel ROI placed over the somatic region of the each cell . Acute coronal slices ( 200 μm ) were prepared from the ARC of fed and fasted male NpyhrGFP mice ( 8 weeks old ) and incubated at 37°C for 1 hr before being kept at room temperature prior to experiments , in ACSF containing ( in mM ) : NaCl2 125 , KCl 2 . 5 , NaHCO3 26 , NaH2PO4 1 . 25 , glucose 25 , CaCl2 2 , MgCl2 1 ( pH 7 . 3 when bubbled with 95% O2 and 5% CO2 ) . Fluorescent cells were visualized on an upright Slicescope ( Scientifica , UK ) using a 60× objective , and whole-cell patch clamp recordings ( Rseries < 30 MΩ ) were performed at 35–37°C using a HEKA 800 Amplifier ( HEKA , Germany ) and borosilicate glass micropipettes with a 3–6 MΩ resistance ( Harvard Apparatus , UK ) filled with ( in mM ) : K-Gluconate 130 , KCl 10 , HEPES 10 , EGTA 1 , Na2ATP 2 , Mg2ATP 2 , Na2GTP 0 . 3 . Apamin-sensitive tail currents were recorded in the presence of Kynurenic Acid ( 2 mM , Sigma ) , Picrotoxin ( 50 μM , Sigma ) , TTX ( 1 μM , Tocris , Minneapolis , MN , United States ) , ±Apamin ( 100 nM , Tocris ) . Current clamp recordings did not contain TTX . Slices were perfused in blockers at least 10 min prior to obtaining recordings , and all comparisons are between populations of cell recorded in the different conditions . Data was analyzed in Python 2 . 7 using custom written routines . To assess the effect of intracerebroventricular peptide injection on feeding behavior , adult ( 8–9 weeks ) male mice were implanted with a cannula over the right lateral ventricle , and were allowed to recover for 1 week prior to further manipulation . All animals were singly housed for at least 5 days following surgery . Assessment of food intake over 24-hr was performed in home cages with ad libitum access to standard mouse chow . Saline or peptide-containing solution ( 1 μl total volume ) was delivered via a micromanipulator ( Narishige ) at a speed at 30 nl/min under isoflurane anesthesia . Injections were performed at 17:00 . The following commercially available peptides were used for assessment of food intake alterations during the dark cycle: recombinant mouse PTN ( R&D Systems , #6580-PL-050 , Minneapolis , MN , United States ) , recombinant mouse ENPP-2/Autotaxin ( R&D Systems , #6187-EN-010 ) , recombinant human BMP-3 ( R&D Systems , #113-BP-100/CF ) , and recombinant human CBLN4 ( Abnova , #H00140689-PO1 ) . Cohorts of experimental animals were randomly assigned into saline or peptide groups . Mice were given at least 1 day post injection before performing additional manipulations . Experiments involving acute alterations in food intake were in the early light period and used a similar procedure for dark cycle experiments . Singly housed , cannulated adult male mice ( 8–9 weeks of age ) were injected intracerebroventricularly with either saline or recombinant mouse ( R&D Systems , #529-TR-025/CF ) at 09:00 and food intake was monitored in the home cage 1 and 2 hr post-injection . Mice received 1 day between experimental sessions before the subsequent injection . To measure the effects of cell-type specific overexpression of Ccl17 , Gpr6 and Bfp , adult male mice ( 8–9 weeks of age at start of experiment ) were injected with a Cre-dependent rAAV as described above . Body weight was assessed between 11:00 and 13:00 for 6 weeks . Two-color smFISH was performed on hypothalamus containing fixed frozen sections from male AgrpCre mice ( 8–9 weeks old ) , using the proprietary probes and methods of Advanced Cell Diagnostics ( Hayward , CA , United States ) ( ACD Technical notes #320535 for tissue prep , and #320293 for Multiplex labeling , http://www . acdbio . com/technical-support/downloads ) . Fed and FD groups were same conditions as for RNA-Seq data . Briefly animals were anesthetized and sequentially perfused with RNase free solutions of PBS and 4% PFA in PBS . The brains were removed and post-fixed ( 24 hr , 4°C ) in 4% PFA in PBS , incubated in 30% sucrose ( 12 hr ) , and the blocked brain was mounted in cryo-embedding media ( OCT , Ted Pella , Redding , CA , United States ) on a cryostat for sectioning . Frozen sections ( 15 μm ) were mounted on slides , which were air dried ( 20 min at −20°C ) or stored at −80°C for later use . The OCT was washed off with PBS before pretreatment with ACD proprietary reagents PT2 and PT4 . After boiling for 5 min in PT2 , sections were rinsed in distilled water then ethanol , air dried , and then incubated with ACD proprietary reagent PT4 ( 30 min , 40°C ) in a HybEZ sealed humidified incubator ( ACD ) . We performed dual probe labeling , using probes for Bmp3 ( Mm-Bmp3-C1 , #428461-C1 ) , Cbln4 ( Mm-Cbln4-C1 , #428471-C1 ) , Ptn ( Mm-Ptn , custom order ) , Hrh3 ( Mm-Hrh3-C1 , #428481-C1 ) , Ccl17 ( Mm-Ccl17-C1 , #428491-C1 ) , and Gpr6 ( Mm-Gpr6-C2 , #318251-C2 ) in one channel and either Agrp ( Mm-Agrp-C1 , #400711-C1; Mm-Agrp-C2 , #400711-C2 ) or Pomc ( Mm-Pomc-C1 , # 314081-C1; Mm-Pomc-C2 , #314081-C2 ) in the complementary channel . Probes were mixed at a 1:50 ratio of Channel 2 and Channel 1 probes . Wax-outlined tissue sections were immersed in Probe mix and incubated ( 2 hr , 40°C ) in the HybEZ humidified incubator , rinsed in ACD Wash Buffer ( 2 × 2′ ) then sequentially incubated in ACD proprietary reagents alternating AMP1-FL and AMP3-FL ( 30 min ) with AMP2-FL and AMP4-FL ( 15 min ) with two washes ( 2 min ) between each step . Brain sections were then labeled with DAPI and coverslips were applied . Slides were stored at 4°C before image acquisition at 63× using a Zeiss 710 confocal on an Axio Examiner Z1 upright microscope . Quantification of mRNA particles was performed on cell volumes obtained by maximal intensity projection of 5 mm of tissue acquired with a 63× objective , using the StarSeach Java applet ( http://rajlab . seas . upenn . edu/StarSearch/launch . html ) with a threshold setting of 50 . For statistical analyses of RNA-Seq data see above . For other experimental data , comparisons were calculated by unpaired or paired two-tail Student's t-test , rank sum test , or analysis of variance ( ANOVA ) . Post hoc multiple comparisons used Holm-Sidak correction . Statistical analyses were performed using Origin , Matlab , or SigmaPlot . Values are means ± s . e . m , unless otherwise noted . n . s . , p > 0 . 05 , p < 0 . 05 , p < 0 . 01 , *p < 0 . 001 . RNA-Seq data is available at GEO , accession number GSE68177 .	Humans and other animals must get adequate nutrition in order to survive . As a result , the body has several systems that work side by side to maintain a healthy body weight and ensure that enough food gets eaten to provide the energy that the body needs . Problems with these systems can contribute towards obesity and other eating disorders . Certain types of cells in the brain play important roles in controlling weight and appetite , although the genes and cellular mechanisms that underlie these abilities are not well understood . When an animal is deprived of food , so-called AGRP neurons produce molecules that increase appetite and make it easier to gain weight . These neurons also go through structural changes and increase their electrical activity during weight loss . Another group of cells , called the POMC neurons , becomes less active when an animal is deprived of energy . Using a technique called cell type-specific transcriptomics , Henry , Sugino et al . have now revealed that the expression of hundreds of genes in AGRP and POMC neurons changes depending on whether mice are well fed or food deprived . Food deprivation also affects more genes in AGRP neurons than has been seen in other types of brain cell , and the AGRP neurons are also more sensitive to a change in food intake than POMC neurons . In the future , this gene expression data and knowledge of the pathways affected by the genes could help researchers to develop new treatments for obesity and other disorders that affect appetite . Henry , Sugino et al . then mapped how these changes in gene expression trigger molecular “pathways” in the neurons that alter how the cells work . These affect many parts of the cells , including ion channels , transcription factors , receptors , and secreted proteins . In addition , food deprivation activated pathways in AGRP neurons that protect the cells from damage and death caused by elevated neuron activity and also triggered signaling pathways that increase body weight . In the future , this gene expression data and knowledge of the pathways affected by the genes could help researchers to develop new treatments for obesity and other disorders that affect appetite .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "tools", "and", "resources", "neuroscience" ]	2015	Cell type-specific transcriptomics of hypothalamic energy-sensing neuron responses to weight-loss
The inner ear is a fluid-filled closed-epithelial structure whose function requires maintenance of an internal hydrostatic pressure and fluid composition . The endolymphatic sac ( ES ) is a dead-end epithelial tube connected to the inner ear whose function is unclear . ES defects can cause distended ear tissue , a pathology often seen in hearing and balance disorders . Using live imaging of zebrafish larvae , we reveal that the ES undergoes cycles of slow pressure-driven inflation followed by rapid deflation . Absence of these cycles in lmx1bb mutants leads to distended ear tissue . Using serial-section electron microscopy and adaptive optics lattice light-sheet microscopy , we find a pressure relief valve in the ES comprised of partially separated apical junctions and dynamic overlapping basal lamellae that separate under pressure to release fluid . We propose that this lmx1-dependent pressure relief valve is required to maintain fluid homeostasis in the inner ear and other fluid-filled cavities . Understanding the mechanisms by which organs use water-filled cavities to compartmentalize biochemical and biophysical environments is a fundamental problem . Because water is nearly incompressible , several organs harness and cope with water as an object that transmits force . Hydrostatic pressure inflates the eye during development ( Coulombre , 1956 ) . Later , unstable ocular pressure from reduced production of aqueous humor or reduced drainage can lead to blindness , as occurs in hypertonic maculopathy or glaucoma ( Costa and Arcieri , 2007; Leske , 1983 ) . Hydrostatic pressure appears to drive expansion of brain ventricles during development ( Desmond , 1985; Desmond and Levitan , 2002; Lowery and Sive , 2005 ) . Later , unstable hydrostatic pressure in brain ventricles is correlated with hydrocephaly and mental disorders ( Hardan et al . , 2001; Kurokawa et al . , 2000 ) . Hydrostatic pressure inflates and controls the size of the developing ear ( Abbas and Whitfield , 2009; Hoijman et al . , 2015; Mosaliganti et al . , unpublished ) . Later , unstable pressure in the ear can cause deafness and balance disorders like Pendred syndrome and Ménière’s disease ( Belal and Antunez , 1980; Schuknecht and Gulya , 1983 ) . This theme of harnessing hydrostatic pressure for normal development and controlling pressure for healthy physiology raises the question of how tissues regulate pressure . Tissue structures identified as important for pressure control include Schlemm’s canal in the eye , arachnoid granules and the choroid plexus in the brain , and the endolymphatic duct and sac in the ear ( Johnstone et al . , 2011; Kimura and Schuknecht , 1965; Naito , 1950; Orešković et al . , 2017; Pollay , 2010; Symss and Oi , 2013; Tripathi , 1974 ) . A cavity’s pressure could be managed via mechanisms involving molecular pores , molecular transporters , and the physical behavior of the tissue . Observing the mechanisms by which these tissue barriers control pressure has been limited by a range of obstacles such as optical accessibility and uncertainty in both the time- and length-scale on which they function . As such , not much is known about how these tissues manage fluctuating pressures because they have not been observed in vivo . The inner ear is a prominent example of an organ whose tissue form determines its physiology . It is filled with endolymph , the composition of which differs from other fluids such as plasma , perilymph , and cerebral spinal fluid in that it contains high potassium , low sodium , and high electric potential ( Lang et al . , 2007 ) . This endolymph composition is necessary to drive ion currents into hair cells to convert fluid movement , caused by either the head’s acceleration or by sound , into biochemical signals that underlie balance and hearing . The ear’s endolymphatic duct connects the endolymph in the semicircular canals , cochlea , and other chambers to the endolymphatic sac ( ES , Figure 1A , B ) . An ES-like structure is present in basal vertebrates , including lamprey and hagfish ( Hammond and Whitfield , 2006 ) , suggesting it plays an ancient role in inner ear function . The epithelium of the ES , as well as that of the rest of the ear , has an apical surface facing the internal endolymph and a basal surface facing the external perilymph ( Figure 1A ) . Most of the adult ear is enclosed in a bony labyrinth within the temporal bone and perilymph is located in the space between the epithelium and bone ( Figure 1A ) . In adults , the endolymphatic duct and sac are partially encased within the temporal bone such that the distal end of the endolymphatic sac protrudes out of the temporal bone into the cranial cavity ( Figure 1A ) . Excessive hydrostatic pressure can tear the inner ear’s epithelium , disrupting its electric potential ( a pathology called endolymphatic hydrops ) . In contrast , low potassium or reduced endolymph production can lower hydrostatic pressure within the ear to the point where the ear chambers may collapse ( Lang et al . , 2007 ) . Early work showed that ES ablation causes hydrostatic pressure to rise and the epithelium to tear ( Kimura and Schuknecht , 1965; Naito , 1950 ) , suggesting that the ES may maintain endolymph homeostasis . Recent work indicated that an explanted mouse ES gradually loses endolymph ( Honda et al . , 2017 ) . While many molecular pores and transporters are expressed in the ES , it is unclear why it is organized as a dead-end tube and how it might reduce endolymph volume . The ES has not been studied much in zebrafish because it has not always been clear that zebrafish possess this structure or at what time it forms during development ( Haddon and Lewis , 1996; Hammond and Whitfield , 2006 ) . In situ hybridization images at 52 hr post fertilization ( hpf ) showing expression of ES markers , foxi1 and bmp4 , restricted to an ES-like structure positioned on the dorsal side of the otic vesicle are the strongest and developmentally earliest evidence for zebrafish having an ES ( Figure 1B’ ) ( Abbas and Whitfield , 2009; Geng et al . , 2013 ) . However , its formation and physiological function remain unknown . Here , we present a detailed characterization of the ES in zebrafish performed using confocal microscopy , serial-section electron microscopy , adaptive optics lattice light-sheet microscopy , and a genetic mutant that together demonstrate that the ES contains a physical relief valve for regulating inner ear fluid pressure through regulated transepithelial fluid release . We first established an imaging system to identify the developmental origins of the zebrafish ES . Extended time-lapse imaging required long-term immobilization with α-bungarotoxin that permits normal development without the reduction in ear growth caused by long-term treatment with tricaine , bright fluorescent transgenic fish with contrast from a membrane-localized fluorescent protein to minimize bleaching and photoxicity , and mounting in a submerged agarose canyon that permits positioning of the zebrafish ear close to the coverslip ( 400 μm wide , walls of canyon secure the yolk and head , Figure 1C ) ( Swinburne et al . , 2015 ) . The zebrafish ear develops from a collection of epithelial cells that form a closed fluid-filled ellipsoidal structure called the otic vesicle ( Whitfield , 2015 ) . We saw that ES morphogenesis begins at 36 hpf as an evagination in the dorsal epithelial wall of the otic vesicle ( Figure 1—figure supplement 1A , Video 1 , Figure 1A–B , and ES identity confirmed by expression of foxi1 , Figure 1B’ ) . Between 36 and 60 hpf , the ES grows and elongates . These findings established that the zebrafish ES is optically accessible during embryonic and larval stages and that ES morphogenesis begins at 36 hpf . The zebrafish ear starts to sense body acceleration for the nervous system between 60 and 72 hpf , as demonstrated by the onset of the vestibulo-ocular reflex ( Mo et al . , 2010 ) . We found that the ES begins to exhibit a physical behavior during the same time window . We observed that the lumen of the ES remains closed until 60 hpf , but between 60 and 65 hpf it begins cycles of slowly inflating and rapidly deflating ( sagittal slices , Figure 1D , lumen volumes , Figure 1E , Figure 1—figure supplement 1 , and Videos 2–3 ) . Three-dimensional ( 3D ) measurements showed that the ES lumen volume changes 5–20-fold through the course of each cycle ( Figure 1E , F , and Figure 1—figure supplement 1B–C ) . The period between peak ES volumes exhibited a broad distribution ( 0 . 3–4 . 5 hr ) with an average of 1 . 6 ± 0 . 8 hr ( mean ± SD , histogram compiled from eight time courses and 54 peaks , Figure 1G ) . These initial observations suggested several potential causes of the inflation-deflation cycles of the ES including: a response to organ-wide increases in fluid pressure within the otic vesicle , a local tissue behavior wherein the ES inflates with fluid from the perilymph , or a local tissue behavior wherein cells in the ES periodically coordinate their movements to expand the ES volume by sucking fluid from the otic vesicle . To assess whether ES inflations occur by transmission of endolymph pressure between the otic vesicle and the ES , we performed a single injection of a small volume of solution containing 3 kDa fluorescent dextran into the otic vesicle and followed its movement during inflation and deflation . We found that the duct is open and that endolymph flows from the otic vesicle into the ES during inflation ( Figure 2A–D , Video 4 ) . Additionally , upon deflation , endolymph rapidly leaked out of the ES and into the perilymphatic space ( 75:22 panel , Figure 2D ) . To determine whether pressure in the otic vesicle is transmitted to the ES for its inflation , we laser-ablated 2–3 cells within the wall of the otic vesicle distant from the ES at 64 hpf ( Figure 2E ) . Shortly after ablation , the ES deflated and completely collapsed within 20 min ( Figure 2F ) . These results indicate that fluid pressure in the otic vesicle inflates the ES volume , the ES tissue has elastic material properties , and a loss of epithelial integrity is sufficient for ES deflation . A core function of epithelial sheets is to act as a barrier that can prevent passive movement of molecules between an organ’s interior and exterior . Deflations could be driven by local breaks in the epithelial barrier combined with elastic tissue collapse or by an alternative mechanism such as cellular reabsorption of endolymph . To distinguish between these mechanisms , we used dye injections into the heart to label the perilymph that surrounds the otic vesicle and ES ( Figure 1D ) . Quantification of fluorescence within the lumen of the ES revealed that perilymph dye begins to leak into the ES lumen at the onset of each deflation ( Figure 1F , visible leakage in Figure 1D , Videos 2–3 ) . In the example shown deflation coincided with an influx of fluorescence from the perilymph into the lumen of the ES in each of the nine major inflation-deflation cycles . This coincident timing suggests that both deflation and dye leakage were due to transient breaks in the epithelial barrier , an interpretation that is consistent with the rapid release of endolymph prior to deflation ( Figure 2D , Video 4 ) . The observation that perilymph enters the deflating ES may seem counterintuitive , since the contents of a punctured high-pressure elastic vessel might be expected to primarily flow outwards . Estimation of the rates of advection and diffusion supports the presence of upstream movement of perilymph by diffusion . Due to limited spatial resolution and the absence of contrast for trafficking vesicles , we could not dismiss alternative mechanisms such as contributions from rapid transcytosis or cellular absorption . To determine whether endolymph efflux occurs at the same time as perilymph influx , we simultaneously labeled the two fluids with different colored dyes and imaged their localization with higher temporal resolution ( Figure 2G , Video 5 ) . We found a fluttering behavior underlying deflation in which endolymph leaked out ( 62:41:20 ) , perilymph leaked in ( 62:55:20 ) , leaked-in perilymph flushed out with more endolymph efflux ( 62:55:40 ) , followed by complete deflation ( 63:12:00 ) . It is difficult to explain the simultaneous inward and outward movement of fluid with the alternative interpretation that rapid transcytosis or cellular absorption processes deflate the ES . A plausible mechanism for this observed behavior involves a structure that normally resists flow driven by a pressure differential that rapidly opens during deflation to allow for fast efflux by advection followed by influx by diffusion before re-closing . The uniqueness of the ES inflation and deflation cycles suggests specific genes might be involved . To identify pathways that contribute to the emergence of this physiology and to reveal ways in which it can malfunction , we examined a mutant caused by a premature stop codon in the transcription factor lmx1bb ( Obholzer et al . , 2012 ) that exhibited an enlarged ES . We found that the ES in lmx1bb mutants became greatly enlarged ( >4 times the inflated wild-type ES volume ) making it readily visible at 80 hpf by bright-field microscopy ( Figure 3A ) . To determine if lmx1bb is expressed at an appropriate place and time for a mutation to be causing an ES defect , we imaged a transgenic reporter line , Tg ( lmx1bb:eGFP ) mw10 , driven by the lmx1bb promoter McMahon et al . , 2009 . This reporter was expressed in ES cells beginning at 52–58 hpf , just before the first ES inflation ( cyan , Figure 3B ) , consistent with lmx1bb expression being instrumental for development of the ability of the ES to release pressure . Earlier in the development of the otic vesicle , lmx1bb is expressed in portions of the nascent semicircular canals and sensory patches , regions of the inner ear that also exhibit abnormal development in the mutant ( Obholzer et al . , 2012; Schibler and Malicki , 2007 ) . There is no precedent for ES development being dependent on those portions of the otic vesicles and there are many mutants with similar SCC or sensory defects that do not have ES phenotypes ( Fekete , 1999; Malicki et al . , 1996; Whitfield et al . , 1996 ) . Live imaging and perilymph tracking in lmx1bbjj410/jj410 mutant embryos revealed that the ES lumen over-inflates ( Figure 3C–D , Figure 3—figure supplement 1 , and Videos 6–7 ) . As in the wild-type analysis , we quantified the presence of perilymph leaking into the ES lumen ( secondary axes of Figure 3D and Figure 3—figure supplement 1A ) . In the mutant , however , we never observed perilymph entering the ES . Additionally , we imaged mutants where the endolymph was labeled and did not observe leakage out of the distended ES epithelium ( Figure 3E ) . These findings suggest that the epithelial diffusion barrier remains intact in the mutant ES . The absence of ES deflation in the lmx1bb mutant suggests that a structural deficiency may be present . Indeed , by comparing the dorsal ES tissue between wild-type and lmx1bb mutants using confocal microscopy and membrane-localized fluorescent proteins , we observed that the ES tissue was much thinner in wild-type embryos during peak inflation than in the corresponding region in mutants inflated to a similar volume ( thin region indicated by asterisk , Figure 3F , G , data in G compiled from 65 to 80 hpf inflation events ) . In wild-type embryos , this region often appeared as a thin sheet ( 1 . 0 ± 0 . 3 µm , mean ±SD ) rather than thick as seen in the un-inflated wild-type ES or the lmx1bb mutants in which two distinct surfaces ( apical and basal ) were observed ( 6 . 2 ± 3 . 2 µm , the mutant , n = 9 , is significantly thicker than wild-type , n = 14 , with a Mann-Whitney-Wilcoxon one tailed p-value of 4 × 10−5 ) . Imaging the uneven signal of the Tg ( lmx1bb:eGFP ) reporter revealed thin protrusions that extend along the basal side of neighboring ES cells ( white arrow , Figure 3H ) . Sparse mosaic labeling of cells in the wild-type further confirmed the presence of thin basal protrusions in the ES , as well as diversity in their organization ( Figure 3I ) . Similar labeling in the mutant revealed the absence of thin basal protrusions in the ES ( Figure 3J ) . To determine the gross organization of the ES tissue , we stained for basal proteins collagen ( Figure 3K–L ) and laminin ( not shown ) . We found that the basal surface of the ES tissue faces the perilymph outside surrounding the otic vesicle ( Figure 3L ) . We could not resolve a clear difference in the staining pattern of collagen between wild-type and the lmx1bb410/410 mutant . Prior work suggested that Col1a2 may be a direct target of LMX1B in the limb ( Haro et al . , 2017 ) and additional studies will be necessary to determine whether collagen or other extracellular matrix components are abnormally expressed in the ES of lmx1bb410/410 mutants . We also examined localization of the apical marker ZO-1 and found that it stained the ES tissue that faces the endolymph-filled lumen ( Figure 3K–L ) . This apical-basal organization was also present in the lmx1bb410/410 mutant ( Figure 3K–L ) . The absence of local breaks in the mutant ES , the absence of regional thinning in the mutant inflated ES , and the absence of thin basal protrusions suggests that these thin areas may be structurally relevant to ES pressure relief . To investigate this possibility , we examined the ultrastructure of these thin areas using serial-section electron microscopy ( EM ) at a resolution of 4 . 0 × 4 . 0 × 60 . 0 nm per voxel . We found the dorsal tissue of the inflated wild-type sac consists of an extremely thin shell of broadly overlapping lamellae extending from the basal sides of multiple adjacent cells ( Figure 4; Video 8 ) . These basal extensions appear to be the distal barrier in the ES because the endolymph they contain is continuous with the lumen of the ES and endolymphatic duct ( extending ventrally from outlined ES lumen in third panel of Figure 4A ) . We term these ‘lamellar barriers’ because of their thin , plate-like structure and apparent function as a barrier that holds the elevated hydrostatic pressure of the endolymph . We observed lamellar sheets that extend for distances as long as a typical cell body but are as thin as 40 nm ( Figure 4A , F; lamella from cell segmented in purple extends 7 . 5 μm in the x-y plane and 6 . 6 μm along the z-axis ) . In contrast to junctions between thin aveolar cells , which kiss at their tips with tight junctions , the lamellae from ES cells formed large zones of overlap and sometimes bifurcated to form a tongue-and-groove structure ( Figure 4B , inset ) or interwoven structures ( Figure 4C , F , black mesh highlights area of overlap , dotted perimeter indicates full spread ) . These structures lacked electron-dense signal , indicating that they were unlikely to contain tight junctions between lamellae . We also identified an ear where the lamellar barriers were separated as if forced into an open configuration ( Figure 4D ) . This may be an ES in the act of deflating via bursting or sliding of the lamellae . The full serial-section EM dataset also confirmed that the endolymphatic duct connects the lumen of the otic vesicle to the tip of the ES , where basal lamellae form a complete barrier ( available at http://zebrafish . link/hildebrand16 , ES-specific links presented in methods ) ( Hildebrand et al . , 2017 ) . Basal lamellae were present along the length of the endolymphatic duct . However , cell bodies in the duct remain tightly packed with apical junctions ( Figure 4A , E ) unlike the apical and lateral membrane separations exposing lamellae in the ES . We identified electron-dense tight junctions between cells at the tip of the ES . These same cells also had lamellar projections ( yellow arrows indicate tight junctions , Figures 4G and 3 serial EM sections , 1 . 2 μm apart , the diameter of the apical opening is ~1 . 2 μm ) . The electron-dense signal formed a ring sealing the apical side of these ES cells that was continuous except for an opening that connected fluid from the duct and sac to the exposed lamellae ( magenta arrow directed from duct to sac , Figure 4G ) . Apical junctions were also present in the mutant ES of lmx1bbjj410/jj410 larvae ( yellow arrows , Figure 4H , H’ ) . Unlike the wild-type ES , however , the mutant ES appeared to completely lack openings in its apical junctions , as we were unable to identify any in mutants imaged by serial-section scanning EM or transmission EM ( four total mutant ears ) . Consistent with the absence of thin protrusions in the sparsely labeled mutants ( Figure 3J ) , we could not find long lamellar projections on the basal side of the mutant ES ( Figure 4I ) . We were unable to identify apical openings in the wild-type ZO-1 immunostains ( Figure 3L ) . This could be due to difficulty resolving small openings ( ~1 μm diameter ) or to ZO-1 remaining present at the openings such that the apical barrier could reform after ES deflation to hold hydrostatic pressure . These data suggest that lmx1bb-dependent activity is necessary for openings to form in the apical junctions of ES tissue and for ES cells to extend thin basal protrusions . The stress from increased pressure in the otic vesicle likely causes an increased ES volume by inducing cell stretching or expansion of the lamellar barrier . To distinguish how the tissue behaves through cycles of ES inflation and deflation , we sought better resolution during live imaging using lattice light-sheet microscopy ( LLSM ) , which generates thin light-sheets using Bessel beams to enhance axial resolution while minimizing phototoxicity and bleaching ( Chen et al . , 2014; Gao et al . , 2014 ) . When using LLSM to image the ES , emitted light passes through brain tissue that scatters and refracts light in a complex , spatially uneven manner . Recent advances in the application of adaptive optics ( AO ) to microscopy can compensate for these aberrations ( Wang et al . , 2014 ) . A microscope combining live-cell lattice light-sheets and adaptive optics was built ( Liu et al . , 2018 ) and used here to image ES cycles with significantly improved spatial and temporal resolution and reduced photo-damage owing to 2–4 times less laser power being distributed over a large volume ( the imaging plane is volumetrically ~105 times larger than the confocal point ) . LLSM requires the illumination plane and optical axis to be perpendicular to one another , so we developed a mold for an agarose mount shaped like a volcano that positions the embryo in the desired orientation ( Figure 5A ) . The LLSM produced slightly blurred images with low signal-to-noise ( SNR , Figure 5B ) without adaptive optics correction . Adding the correction with an AO-LLSM system adequately compensated for tissue aberrations and produced images with higher SNR and better contrast . The resulting high-quality images were then analyzed using our software , ACME , to reconstruct membrane signals ( Figure 5C ) , segment cells from the lumen of the ES ( Figure 5D , E ) , and track cells and lumen accurately through time ( Mosaliganti et al . , 2012 ) . As with confocal live imaging , we observed cycles of the ES lumen inflating and deflating ( Figure 5F , G; Videos 9–10 ) . Cell bodies within the ES stretch in response to increasing pressure . While the increased resolution of AO-LLSM did not allow us to resolve overlapping basal lamellae ( thickness ~40 nm ) , it did enable segmentation of individual cell bodies ( lengths and thicknesses ~1–10 μm ) . We found that some cells stretched and thinned when the ES lumen was inflated and thickened upon deflation , likely a result of their elastic properties ( Figure 5H , I ) . However , there were instances when the cells at the tip of the ES did not thin during inflation , or while they thinned during some events , they did not thin during others ( Figure 5J , K ) . To quantify these correlations and obtain an overview of the range of behaviors , we determined the Spearman correlation coefficient between trajectories of lumen volume and cell thickness for intervals that spanned individual inflation and deflation events ( Figure 5L , bracketed examples in Figure 5H , J ) . For 43 of 99 tested trajectories , either cell thinning significantly correlated ( p-value less than 10−3 ) with inflation or cell thickening significantly correlated with deflation ( green region , Figure 5L , example of significantly correlated interval , bracket with green arrowhead , Figure 5H , same data point indicated with green arrowhead in Figure 5L ) . For 42 of 99 tested trajectories , cell behavior did not significantly correlate with inflation or deflation ( grey region , Figure 5L , example of uncorrelated inflation interval , bracket with grey arrowhead , Figure 5J , same data point indicated with grey arrowhead in Figure 5L ) . Unexpectedly , for 14 of 99 tested trajectories there was significant correlation between inflation and cell thickening or deflation and cell thinning ( magenta region , Figure 5L , example of correlated interval , bracket with magenta arrowhead , Figure 5J , same data point indicated with magenta arrowhead in Figure 5L ) . While each cell’s behavior is varied , these data suggest that stretching of a subset of cells contributes to regulation of increased pressure by allowing the ES lumen to expand . Additionally , some cells appear to be pushed away from the ES lumen during inflation and pulled back into the ES epithelium during deflation . The complexity of the response is likely the result of features we did not resolve such as the dynamics of each cell’s basal lamellae , each cell’s residual apical and lateral adhesion , and each cell’s basal interface with the extracellular matrix . The low temporal resolution of our original confocal time courses made it unclear whether basal lamellae are static structures like floodgates , only opening to release endolymph , or dynamic . 3D rendering of the AO-LLSM signal shows that some lamellae are rapidly extending and retracting like they are crawling over neighboring cells or neighboring lamellae ( Video 11 ) . The basal lamellae are composed of two juxtaposed membranes and can overlay abutting membranes of adjacent cells or additional lamellae ( Figure 4 ) . While we observed stacked membranes by EM , they are still unresolved by AO-LLSM . However , one might expect patches of increased membrane signal several times the brightness of that of a lone cell membrane . In 3D renderings of the time course we see dynamic patches of increased intensity that can shift rapidly ( Figure 6A , Video 11 ) . By merging 3D renderings of 3 consecutive time points , 30 s apart , in red , blue , and then green , we observed the relative displacement of these lamellae that can crawl a few micrometers per minute ( Figure 6B , C ) . While the resolution of AO-LLSM is improved , it does not resolve the double membrane structure of lamellae or enable identification of the cell from which a lamella extends . Thus , in our surface-rendered segmentations , we give the cell membranes of each cell body a unique pseudo-color while the lamellae that are just exposed to the lumen and not adjacent to another cell body are collectively colored magenta ( Figure 5E , G , corresponds to black dotted outline in Figure 4F ) . By visualizing the segmented objects in 3D we can identify when lamellar barriers are exposed based on when and where the magenta appears ( Figures 5G and 6D , Video 10 ) . Inflations of the ES lumen coincided with slow increases in the surface area of exposed lamellae ( 30 min expansion , Figure 6D ) . The expansion of thin membrane signal was also observable in the raw data ( bottom portion of panel , Figure 6D ) . Some secondary loci included smaller lamellae that engaged and inflated much more quickly ( 1–2 min expansion , Figure 6E ) . We can visualize their sudden expansion by again merging consecutive time points , 30 s apart , with red , blue , and then green ( Figure 6F ) . This presentation reveals that the same locus can rapidly expand multiple times ( one minute rapid expansions an hour apart , Figure 6F , G ) . In summary , multiple sets of basal lamellae can simultaneously engage at different loci , either through slow spreading or rapid expansion . Cells and tissues modulate tension for crawling , retracting , bending , constricting , rounding-up , and resisting stretch by organizing an extensive variety of complexes of actin filaments , actin binding proteins , and myosin motors ( Grill , 2011; Munjal and Lecuit , 2014; Rafelski and Theriot , 2004; Stewart et al . , 2011 ) . Myosin distribution correlates strongly with contractile forces ( Fournier et al . , 2010 ) . Therefore , to begin to determine how myosin and actin might contribute to tension during ES inflation and deflation , we imaged myosin and actin using non-muscle myosin light chain fused to eGFP and utrophin , an actin binding protein , fused to mCherry ( confocal microscopy , Figure 6H , Video 12 ) . Myosin localized to the apical domains of ES cells as well as to dynamic puncta at the basal membrane throughout both inflation and deflation of the ES . During ES inflation , contraction of apical myosin likely counteracts the stress of hydrostatic pressure and maintains the integrity of apical junctions ( Rauzi et al . , 2010 ) . Tension from apical myosin , strain from hydrostatic pressure , and the adhesion strength between junctional protein complexes likely determines the elastic limit of apical junctions . The crossing of this elastic limit may cause the small apical openings observed in Figure 4G , which ultimately communicates the otic vesicle’s hydrostatic pressure to the lamellar protrusions . The dynamic spots of basal myosin could indicate two possible activities: contraction to retract lamellar protrusions as observed during cell migration or contraction at focal adhesions . Utrophin localized strongest at apical domains of ES cells ( Figure 6H ) , but unfortunately bleached too quickly to generate any meaningful insights into the potential of actin localization dynamics . It had been shown that utrophin based actin highlighters do not localize to actin within lamellae ( Belin et al . , 2014 ) , which explains the relatively low amounts of basal utrophin signal . In contrast , fluorescent phalloidin labels actin filament enrichment at both apical and basal domains of ES cells ( not shown ) . Our observations of consistent apical localization of myosin throughout ES cycling suggests that the same contractility both counteracts hydrostatic pressure during inflation and drives deflation when hydrostatic pressure drops as the relief valve is opened . We did not observe any sudden changes in myosin localization or behavior that would indicate regulation for either inflation or deflation . Analysis of myosin localization in lmx1bb crispants ( CRISPR-Cas9 knockout embryos ) revealed similar apical localization and dynamic basal puncta ( Figure 6I , Video 12 , representative of 2 time-lapses ) . To determine whether the mutant ES tissue has elastic material properties , we punctured the otic vesicle with a tungsten needle and found that the distended ES rapidly collapsed ( 5/5 mutant puncture experiments caused ES collapse , Figure 6K ) . Global inhibition of actin or myosin with cytochalasin D or ( S ) -nitro-blebbistatin correlated with ES deflation . However , it is currently impossible to determine whether this is owed to hydrostatic pressure not being maintained by the otic vesicle epithelium . Given the finding that ES lamellae are dynamic with myosin puncta and phalloidin stain , we asked whether Rac1 , known to promote actin polymerization in lamellipodial protrusions , was important for ES valve function ( Waterman-Storer et al . , 1999 ) . Heat-shock induction , throughout the embryo , of a dominant negative Rac1 at 56 hpf resulted in otic vesicles that became leaky and subsequently collapsed between 60 and 70 hpf ( Video 13 , 19/34 leaky and collapsed ears versus 1/12 in a heat-shock gfp control ) ( Kardash et al . , 2010 ) . In contrast , heat shock induction of the dominant negative Rac1 at 32 hpf , prior to ES formation , did not result in leaking or ear collapse during the subsequent 15 hr ( while 0/20 leaky or collapsed ears versus and 0/14 in a heat-shock gfp control ) . These results are consistent with an important role for the basal lamellipodia of the ES for maintaining barrier integrity during inflation and deflation cycles . Precise analysis of the roles of actin and myosin networks during ES inflation cycles will require either the identification of regulators specific to the ES or actomyosin perturbations specific to the ES . While we had evidence that lamellae can separate in our EM images and evidence that the epithelial barrier breaks during deflation , as seen in the leak in fluorescence from labeled perilymph , it remained uncertain whether separation of lamellae precipitated the deflation events . To observe evidence of the lamellar valve opening , we looked for instances when the membrane signal of lamellar barriers was disrupted prior to deflation events using AO-LLSM . We were able to witness instances when membrane signal from inflated lamellae was disrupted prior to deflation ( yellow arrows , Figure 7A–D , lumen magenta , adjacent cells colored in at first at last time-points ) . These events can be accompanied by a thin lamellar protrusion sticking out into the perilymph ( arrow , Figure 7D ) . Examination of the raw and rendered segmentations prior to and following breaks in membrane continuity showed that it preceded a deflation event ( reduction in local volume of ES lumen , magenta , Figure 7A , C ) . To confirm that the small openings cause the rapid expulsion of endolymph during deflation events , we more closely examined time-lapse movies with labeled endolymph ( Figure 7E , yellow endolymph volume before and after two rapid openings ) . Close examination of the path of endolymph release revealed a small opening through which endolymph travels out into the periotic space ( x-y and y-z planes at relief valve openings , yellow arrows , Figure 7E ) . Narrow paths of pressure release were observed in all high time-resolution ( 20 s ) time-lapse movies of endolymph dynamics ( four fish in Video 14 exhibit 16 opening events ) . We report identification of the ES as a pressure relief valve based on six lines of evidence . First , the lumen of the ES slowly inflates and then rapidly deflates every ~0 . 3–4 . 4 hr . Second , deflation of the ES coincides with a breach in its epithelial diffusion barrier . Third , laser ablation of the otic epithelium is sufficient to induce rapid deflation of the inflating ES . Fourth , the ultrastructure of the closed and opened relief valves in the ES shows that lamellar barriers can pull apart . Fifth , high temporal and spatial resolution time course imaging reveal lamellae that behave like lamellipodia: constantly crawling over one another before they separate to relieve pressure and excess volume . Sixth , perturbation of the valve’s development with a genetic mutation causes distension of the ES as found in common ear disorders such as Ménière’s disease , Pendred syndrome , and Enlarged Vestibular Aqueduct syndrome ( Levenson et al . , 1989; Jackler and De La Cruz , 1989; Belal and Antunez , 1980; Schuknecht and Gulya , 1983 ) . Upon first observing the inflation and deflation cycles of the ES , we postulated that the underlying mechanism could either be a response to organ-wide hydrostatic pressure within the otic vesicle , a local tissue behavior where the ES inflates with fluid from the perilymph , or a local tissue behavior where cells in the ES periodically coordinate their movements to expand the ES volume . Our ablation experiments indicated that pressure is transmitted to the ES from the OV and that the tissue is elastic because it collapses quickly when the stress from hydrostatic pressure is removed ( Figure 2E–F ) . The second and third models lack support: dye tracing shows that during inflation the ES is filled with endolymph from the otic vesicle , while perilymph only enters the ES during deflation when the epithelium is open . We did not observe coordinated cell movements within the ES other than stretching . Additionally , myosin appears to act consistently throughout the ES cycles likely providing a constant apical tension that both resists pressure-induced stretch during inflation and drives collapse during deflation by contributing to tissue elasticity . This constant activity precludes the need for sensors or signals to tell the valve when to inflate , when to open , and when to collapse . Furthermore , dynamic basal puncta of myosin may provide integrity to adhesion between basal lamellae with added tension ( Ingber , 1997 ) . Throughout our imaging experiments , whether the contrast was produced at plasma membranes , endolymph , or perilymph , we observed what appear to be small pockets of lumen that pinch off from the rest of the ES lumen ( see Figures 2D , G , 5G , 6D–E and 7E , and Videos 4 , 5 , 9 , 10 , 11 and 14 ) . These pockets were not present in the lmx1bb mutant , ( Figure 3E ) . The simplest explanation is that fluid gets trapped in the lateral intracellular space as the apical junctions open but before the basal lamellae open or when the basal lamellae close after a deflation event . This is consistent with how the inclusions are often resolved in the next inflation cycle . Incorporating all of our observations , we surmise that increased pressure is managed through a combination of strain that stretches viscoelastic cells in the ES and adhesion distributed across the surface area of dynamic basal lamellae . Excess pressure and volume is then released when small openings form amongst compliant apical junctions of ES cells that transmits the stress of the ear’s hydrostatic pressure to the basal lamellae that hold until they separate to release excess pressure and volume . The elastic properties of the tissue then drive its deflation , basal lamellae reunite , and the homeostatic cycle begins again ( Figure 7F ) . A physical relief valve in the ES , composed of overlapping basal lamellae , had not been previously identified because of the lack of sufficient temporal and spatial resolution as well as the lack of a system with an optically accessible ES . Thin overlapping basal junctions are seen at the marginal folds of lymph and blood capillaries ( Baluk et al . , 2007 ) . Capillaries flank the developing and adult ES and the similarity between the cross-sectional ultrastructure of lamellar barriers and capillaries , as well as vacuoles , could have led to inaccurate annotation of lamellar barriers ( Kronenberg and Leventon , 1986 ) . EM micrographs of the rat , guinea pig , tree frog , and human ES showed instances of thin cytoplasmic extensions enclosing large lumens , which were annotated as capillaries or large vacuoles ( Bagger-SjobackSjöbäck et al . , 1986; Bagger-Sjöbäck and Rask-Andersen , 1986; Dahlmann and von Düring , 1995; Kawamata et al . , 1987; Møller et al . , 2013; Qvortrup et al . , 1999 ) . More recent EM studies of the human ES , which used more refined techniques to prepare samples from adult cadavers , recognized similar structures as belonging to the distal portion of the ES and described them as interconnected lumen resembling a network of cisterns ( Møller et al . , 2013 ) . While appearing more elaborate in humans , the structural unit of the relief valve seems to be conserved . However , because of insufficient spatial and temporal resolution , it remains to be determined whether these cells have overlapping lamellae and whether they function as a physical relief valve . Primarily , pressure relief prevents endolymphatic hydrops—the buildup of pressure and potential tearing of the ear’s epithelium—that is associated with many inner ear pathologies . Secondarily , low-frequency deflations of the ES may drive intermittent longitudinal flow of endolymph from the cochlea and semicircular canals into the endolymphatic duct and sac . The presence of longitudinal flow had been proposed to explain the accumulation of debris and tracers within the ES and , while never directly observed in an unperturbed inner ear , longitudinal flow can be induced and observed through manipulations of the inner ear ( Salt , 2001; Salt and DeMott , 1998; Salt and Plontke , 2010; Salt et al . , 1986 ) . The localized inflation and patterned formation of valve cells leads to a localized break at the ES lamellar barriers . The local break could create a transient pressure gradient that leads to flow from other parts of the inner ear towards the endolymphatic duct and sac . The intermittent and slow nature of this flow could explain conflicting interpretations of classic studies using tracer injections where tracer accumulated in the ES after long periods of time but significant flow was absent when observed on short time scales ( Guild , 1927; Manni and Kuijpers , 1987; Salt , 2001; Salt et al . , 1986 ) . We observed longitudinal flow in action with the expulsion of leaked-in perilymph ( Figure 2G , Video 5 ) Lmx1bb is a LIM homeobox transcription factor . Mutations in LMX1B cause Nail Patella syndrome in humans and a third of Nail Patella syndrome patients develop glaucoma , a phenotype also observed in mouse models ( Cross et al . , 2014; Liu and Johnson , 2010; Mimiwati et al . , 2006 ) . The mechanism by which glaucoma arises in the disease and in mouse models is not clear; however , the trabecular meshwork and Schlemm’s canal appear abnormal in these cases . Giant vacuoles and pores in Schlemm’s canal may be used to relieve intraocular pressure , and their ultrastructure resembles the lamellar barriers of the ES ( Gong et al . , 2002 ) . Like the ear , pulsatile pressure relief occurs in the eye , although at the higher frequency of the heart rate , and Schlemm’s canal is a major site of the pressure relief ( Ascher , 1962; Johnstone et al . , 2011 ) . Additionally , up to half of Nail Patella syndrome patients suffer from kidney disease ( Bennett et al . , 1973 ) . During podocyte maturation apical junctions are lost , thereby enabling development of the basal slit diaphragm that filters blood ( Pavenstädt et al . , 2003 ) . In mice lacking LMX1B activity , podocytes fail to lose their apical junctions ( Miner et al . , 2002 ) . Removal of apical junctions may thus be a specialized mechanism used in organs where elaborations of junction architectures allow controlled release of fluids from one compartment into another . Furthermore , the mouse Lmx1a mutant exhibits defects ( Koo et al . , 2009 ) like the fish lmx1bb mutant in that they exhibit similarly enlarged ears . Similarity in its protein sequence and expression suggests that mouse Lmx1a and fish lmx1bb likely have comparable activities . We speculate that LMX1 transcription factors may have a common set of targets in the ES , eye , and kidney involved in apical junction remodeling to regulate fluid flow across epithelia . We find that apical junctions and basal lamellae in the endolymphatic sac behave like a pressure relief valve to regulate fluctuating volume and pressure that arises from excess endolymph in the inner ear ( Figure 7F ) . The high temporal and spatial resolution of AO-LLSM , combined with image processing tools for segmenting , tracking , and quantifying cell geometries , was necessary to reveal the dynamic cell behaviors underlying the pulsing of the ES . While cells of the ES are immobile epithelial cells , the cell extensions resemble lamellipodia of crawling cells in thickness ( 40 nm ) , speed ( as fast as ~1 μm per minute ) , and regulatory mechanism ( Rac1 ) ( Abercrombie , 1980 ) . Because of limitations of imaging , it remains unclear how each cell’s basal lamellae , residual apical and lateral adhesion , and basal interface with the extracellular matrix contribute to the physical behavior of the ES . Of related interest is how the crawling lamellae adhere to their substrate in the ES to generate a barrier without gaps during the inflation phase . The force of this adhesion , in combination with the viscoelastic properties of the ES , likely determines the relief valve’s set point for volume and pressure homeostasis . For the inner ears of adult mammals , this set point is likely around 100–400 Pascals , although it is unknown how much the inner ear’s pressure and volume fluctuate ( Park et al . , 2012 ) . In our mutant over-inflation videos we observed an instance of an incomplete deflation ( Video 6 ) that contrasts with wild-type deflation cycles and mutant ES collapse after the otic vesicle was punctured ( Figure 6J ) . This difference raises the questions of how pathological tearing of otic epithelium from endolymphatic hydrops might lead to hearing and balance symptoms and whether the organization of the endolymphatic duct and sac prevents an influx of low potassium perilymph from coming into direct contact with the stereocilia of sensory hair cells . Determination of the molecular basis of lamellar barrier organization and apical junction dynamics will be necessary to make progress on how the physical relief valve works and how it might fail to cause disease . Zebrafish were maintained at 28 . 5°C using standard protocols ( Westerfield , 1993 ) . The Harvard Medical Area Standing Committee on Animals approved zebrafish work under protocol number 04487 . Adult zebrafish , 3 months to 2 years of age , were mated to produce embryos and larvae . Adults were housed in a main fish facility containing a 19-rack Aquaneering system ( San Diego , CA ) and a 5-rack quarantine facility for strains entering the system from outside labs or stock centers . The systems’ lights turn on at 9am and go off at 11pm . Fish matings were set up the night before with males separated from females with a divider in false-bottom crossing cages in a pair wise , or 2 × 2 fashion to maximize embryo yield . The divider was pulled the following morning , shortly after the lights turned on . Egg production was monitored to establish the time of fertilization . Manipulations and observations of baby zebrafish were performed between fertilization and 5 . 5 days post fertilization . These studies were performed using the AB wild-type strain , the lmx1bbjj410/jj410 mutant and the following transgenic lines: Tg ( actb2:mem-citrine-citrine ) hm30 , Tg ( actb2:mem-citrine ) / ( actb2:Hsa . H2b-tdTomato ) hm32 , Tg ( actb2:mem-citrine ) / ( actb2:Hsa . H2b-tdTomato ) hm33 ( these three alleles were combined for maximal membrane signal , hm32 and hm33 are two separate alleles of the same construct that is composed of two divergent beta-actin promoters , one driving membrane citrine and the other driving histone tdTomato , actb2:Hsa . H2b-tdTomato of these divergent constructs tends to be silenced in transgenic fish and is not useful ) , Tg ( −5 . 0lmx1bb:d2eEGFP ) mw10 , Tg ( actb2:mem-mcherry2 ) hm29 , Tg ( hsp70:rac1_T17N-p2a-mem-cherry2 ) hm35 , Tg ( elavl3:GCaMP5G ) a4598 , Tg ( actb2:myl12 . 1-EGFP ) e2212 , and Tg ( actb2:mCherry-Hsa . UTRN ) e119 ( Aguet et al . , 2016; Ahrens et al . , 2013; Compagnon et al . , 2014; McMahon et al . , 2009; Obholzer et al . , 2012; Schibler and Malicki , 2007; Xiong et al . , 2014; Xiong et al . , 2013 ) . The foxi1 in situs that confirm the identity of the ES tissue ( Figure 1B’ ) is representative of 25 stained embryos . A similar ES tissue labeling was obtained when staining for bmp4 ( 12 embryos at 55hpf ) . In situs were performed as previously described ( Thisse and Thisse , 2014 ) . The foxi1 probe was made using T7 RNA polymerase ( Sigma ) off of a template made from cDNA using these PCR primers: ctccATGTTTCTGGAGGGAGAG and TAATACGACTCACTATAGGGagaGATCCGTCCCGGTTGTATATGAG . The bmp4 probe was made using T7 RNA polymerase ( Sigma ) off of a template made from cDNA using these PCR primers: GTCGAGACATCATGATTCCTGG and TAATACGACTCACTATAGGG aga GAGTCTCCGTTTAGCGGCAGC . Whole-mount fluorescent immunostains were performed using standard protocols and a mouse monoclonal antibody against ZO-1 ( ZO1-1A12 , Thermo Fisher Scientific , Waltham , MA ) , a rabbit polyclonal antibody against collagen II ( ab209865 , Abcam ) , and a rabbit polyclonal antibody against laminin ( L9393 , Sigma ) . In Figure 3J , the 50 hpf wt collagen stain is representative of 16 stained and imaged embryos , the 50 hpf wt ZO-1 stain is representative of 22 stained and imaged embryos , the 80 hpf wt collagen stain is representative of 11 stained and imaged embryos , the 80 hpf wt ZO-1 stain is representative of 32 stained and imaged embryos , the 80 hpf mutant collagen stain is representative of 7 stained and imaged embryos , and the 80 hpf mutant ZO-1 stain is representative of 6 stained and imaged embryos . In all stained embryos , the membrane contrast comes from the membrane citrine transgenic alleles . Crispants for lmx1bb in the Tg ( actb2:myl12 . 1-EGFP ) e2212 background were made using previously described approaches ( Gagnon et al . , 2014 ) . sgRNA’s were designed and synthesized to target exon 2 ( GTTGCTTGTCCCGACCGCAGCGG ) and exon 3 ( ACGGAGTGCCATCACCAGGCGG ) of lmx1bb . These guides were pooled and combined with purified Cas9 protein and injected into 1 cell stage embryos . Analyses were performed on F0 crispants that exhibited the full mutant phenotype . Embryos were immobilized with either 50 pg of α-bungarotoxin mRNA injected into the 1 cell embryo or 2 . 8 ng of α-bungarotoxin protein injected into the heart at 59 hpf ( Swinburne et al . , 2015 ) . α-bungarotoxin mRNA was synthesized from a linearized plasmid using the mMessage mMachine T7 ULTRA kit ( Thermo Fisher Scientific ) . Subsequently , mRNA was purified using RNAeasy Mini Kit ( Qiagen , Hilden , Germany ) . α-bungarotoxin protein was obtained from Tocris Bioscience ( Bristol , United Kingdom ) . 2 . 3 nL injections were performed using Nanoject II ( Drummond Scientific , Broomall , PA ) . For tracking perilymph , ~10 ng of 3 kDa dextran-Texas red neutral ( Thermo Fisher Scientific ) was injected into the zebrafish heart at 59 hpf . For tracking endolymph , ~1 ng of 3 kDa dextran-Texas red neutral was injected into the otic vesicle using a FemtoJet 4x ( Eppendorf , Hamburg , Germany ) to minimize the volume injected to ~0 . 2 nl . When labeling both endolymph and perilymph , 3 kDa dextran-Texas red neutral was used for endolymph and 10 kDa dextran- Alexa Fluor 488 ( Thermo Fisher Scientific ) was used for perilymph . Immobilization with α-bungarotoxin permits the ear to grow and develop normally ( Swinburne et al . , 2015 ) . Paralyzed embryos were cultured in Danieau buffer and mounted in an immersed 1 . 5% w/v agarose canyon mount or the volcano mount for AO-LLSM , that were generated from custom-made molds . The canyons were 0 . 4 mm wide and 1 . 5 mm deep . The volcano mold for AO-LLSM was printed by shapeways . com ( New York City , NY ) . The inner width of the volcano was also 0 . 4 mm and was printed using their ‘Frosted Extreme Detail’ material . The embryo was placed dorso-laterally in the canyon or volcano so that dorsal portion of the left otic vesicle was either flush with a #1 coverslip placed over the canyon ( embryos were tilted approximately 30 degrees laterally from their dorsal axes ) or protruding from the mouth of the volcano mold . No coverslip was placed above embryos for AO-LLSM . Homemade hair loops were used to position embryos . Mounted embryos were imaged on an upright LSM 710 ( Carl ZEISS , Göttingen , Germany ) with a C-apochromat 40x/NA 1 . 2 objective ( ZEISS ) . The objective’s corrective ring was adjusted to account for the use of #1 coverslips ( 0 . 13–0 . 16 mm thick ) . Imaging took place within a homemade foam-core incubator maintained at 28 . 5°C . Lasers of wavelength 405 , 488 , 514 , and 594 nm were used to image GFP , citrine , Texas red , and mCherry2 . A typical imaging session used the following set up: citrine excited with 514 nm laser 5% ( Ch1 filters 519–584 nm , gain 885 ) , Texas red excited with 594 nm laser 15% ( Ch2 filters 599–690 nm , gain 864 ) , 1 . 27 μs pixel dwell , one line average , 137 μm pinhole , 1 . 2 zoom , 458/514/594 beam splitter , and 0 . 173 × 0 . 173×1 . 135 μm voxel scaling . Around 20–30 long-term time courses were attempted to establish the method . The data presented are representative of 8 wild-type time courses and four mutant time courses . Additionally , hundreds of ES distensions in mutants were observed during the cloning of the mutant and phenotype characterization at 72 and 96 hpf . Laser ablations of epithelial cells in the otic vesicle were performed using a similar imaging set-up as for time-lapse confocal imaging . For the cell ablations a Mai-Tai HP 2-photon laser ( Spectra-Physics , Santa Clara , CA ) was used . After a target was chosen the 2-photon laser was tuned to 800 nm at 50% power , the pinhole was opened completely , the 690 + beam splitter was selected , and a spot scan was performed for 10 , 000 cycles . 2 or three spots were targeted in adjacent cells to ensure the wounds disrupted the epithelial barrier . After ablations , time-lapse microscopy was performed with 1-photon microscopy , as described above . The data presented are representative of 4 such experiments as well as seven ablations that were evaluated without time courses , 1 hr after ablation . For puncturing with a tungsten needle , 78 hpf mutant ears were imaged on the confocal microscope , taken from the microscope to a dissecting microscope where the otic vesicle was rapidly punctured , remounted , and then reimaged on the confocal microscope within 2 min . 0 . 25 mm tungsten wire ( Sigma ) was sharpened by electrolysis to make tungsten needles . For imaging with the adaptive optics lattice light-sheet microscope , zebrafish immobilized with α-bungarotoxin were mounted in a 1 . 5% low melt agarose volcano mold on a 5 mm coverslip and imaged starting at 68–72 hpf . We submerged the excitation and detection objectives along with the 5 mm coverslip in ~8 ml of 1X Danieau buffer at room temperature ( 22 ± 2°C ) . The zebrafish tissues expressing membrane citrine marker and 3 kDa dextran Texas Red fluid phase marker were excited using 488 nm and 560 nm lasers ( 488 nm operating at ~3 mW and 560 nm operating at 5 mW corresponding to ~16 µW and ~31 µW at the back aperture of the excitation objective ) both sequentially exposed for 15 msec . The tissues were excited with 488 nm and 560 nm sequentially by dithering a multi-Bessel light-sheet arranged in a square lattice configuration ( corresponding to an excitation inner/outer numerical aperture of 0 . 517/0 . 55 , respectively ) . The optical sections were collected by scanning the detection objective with 300 nm steps . Each imaging volume consisted of 131 optical planes capturing 1024 × 1024 pixels , thereby capturing a volume of ~97 µm x 97 µm x 39 µm every 15 . 5 or 29 . 9 s using a dual Hamamatsu ORCA-Flash 4 . 0 sCMOS camera setup ( Hamamatsu Photonics , Hamamatsu City , Japan ) . Prior to the acquisition of the time series data consisting of 300–530 time points , the imaged volume was corrected for optical aberrations ( manuscript in preparation ) . The imaged volumes were deconvolved with experimentally measured point-spread functions measured with 0 . 17 µm tetraspec beads ( Thermo Fisher ) excited with 488 nm and 560 nm lasers in MATLAB using Lucy-Richardson algorithm on HHMI Janelia Research Campus’ computing cluster and locally with a 3 . 3 GHz 32-Core Intel Xeon E5 with 512 GB memory . The LLSM was operated using a custom LabVIEW software ( National Instruments , Woburn , MA ) on a 3 . 47 GHz Intel Xeon X5690 workstation with 96 GB memory running Microsoft Windows seven operating system . The images presented are from one AO-LLSM time course that is representative of three other similar acquisitions also obtained using AO-LLSM . ‘Wild-type’ Tg ( elavl3:GCaMP5G ) a4598 and ‘mutant’ lmx1bbjj410/jj410 larvae at 5 . 5 days post-fertilization ( or 4 . 5 days for the mutant that begins to suffer from pleiotropic toxicity ) were briefly anesthetized with 0 . 02% tricaine methanesulfonate ( Sigma-Aldrich , St . Louis , MO ) , and subsequently preserved by dissection and immersion into an aldehyde fixative solution ( 2% paraformaldehyde and 2 . 5% glutaraldehyde in 0 . 08M Sorenson's phosphate buffer , Electron Microscopy Sciences , Hatfield , PA ) . Each specimen was then post-fixed and stained en bloc with reduced osmium solution ( 1% osmium tetroxide ( Electron Microscopy Sciences ) and 1 . 5% potassium ferricyanide ( Sigma-Aldrich , St . Louis , MO ) ) followed by uranyl acetate ( 1% uranyl acetate ( Electron Microscopy Sciences ) in 0 . 05 M maleate buffer ) . Finally , samples were dehydrated with serial dilutions of acetonitrile in distilled water , infiltrated with serial dilutions of epoxy resin in acetonitrile ( Electron Microscopy Sciences ) , and embedded in low-viscosity resin ( Koehler and Bullivant , 1973 ) . The epoxy resin for infiltrations and embedding was composed of 63% nonenyl succinic anhydride ( Electron Microscopy Sciences ) , 35 . 5% 1 , 2 , 7 , 8-diepoxyoctane ( 97% , Sigma ) , and 1 . 5% 2 , 4 , 6-tri ( dimethylaminomethyl ) phenol ( DMP-30 , Electron Microscopy Sciences ) , v/v . Serial sections through the endolymphatic sac were continuously cut with a nominal thickness of 60 nm using a 45° diamond knife ( Diatome , Biel , Switzerland ) affixed to an ultramicrotome ( Leica EM UC6 , Leica Microsystems , Wetzlar , Germany ) and collected with an automated tape-collecting ultramicrotome ( Hayworth et al . , 2014 ) . Field-emission scanning EM of back-scattered electrons was conducted either on a Zeiss Merlin ( ZEISS ) or a FEI Magellan XHR 400L ( Thermo Fisher Scientific ) with an accelerating voltage of 5 . 0 kV and beam current of 1 . 6–7 nA ( Hildebrand et al . , 2017 ) . Image registration was performed with Fiji TrakEM2 alignment plug-ins ( Saalfeld et al . , 2010; Saalfeld et al . , 2012; Schindelin et al . , 2012 ) . Manual image segmentation was performed with ITK-SNAP ( Yushkevich et al . , 2006 ) . The data presented are from a single wild-type and single mutant sample because of the time and effort required technology development as well as in sample preparation , acquisition , and processing . Additional wild-type and lmx1bbjj410/j4410 mutant specimens were analyzed by transmission electron microscopy at the Harvard Medical School EM facility . Samples were fixed in 2 . 5% paraformaldehyde , 2 . 5% glutaraldehyde , 0 . 03% picric acid in 0 . 1M sodium cacodylate buffer , pH 7 . 4 . Samples were stained for three hours in 1% osmium tetroxide and 1 . 5% potassium ferrocyanide , washed , and then stained with 1% uranyl acetate in maleate buffer , pH 5 . 2 for one hour . Samples were then embedded in Taab 812 Resin ( Marivac Ltd . , Nova Scotia , Canada ) . Blocks were sectioned with a Leica ultracut microtome to a thickness of 80 nm . Sections were collected on carbon coated , formvar coated , copper slot grids . Sections were further stained with with 0 . 2% lead citrate prior to viewing on a JEOL 1200EX ( JEOL USA , Peabody , MA ) using an AMT 2 k CCD camera ( Advanced Microscopy Techniques , Corp . , Woburn , MA ) . Time-lapse confocal data sets were converted from Zeiss’s LSM format to a series of image files ( * . png ) with a header file containing information on the imaging set up ( * . meg ) . This was performed with a script called lsmtomegacapture available in our open source ‘in toto image analysis tool-kit’ ( ITIAT , https://wiki . med . harvard . edu/SysBio/Megason/GoFigureImageAnalysis ) . The image data sets were then loaded into GoFigure2 , open-source software developed in the Megason lab for image analyses with a database ( http://www . gofigure2 . org; Xiong et al . , 2013 ) . In GoFigure2 the ES was manually contoured and these contours were exported as XML-files in the GFX format . Exported contours were modeled in 3D and volumes were quantified using a script called GoFigure2ContoursToMeshes , which uses the power crust reconstruction algorithm ( Amenta et al . , 2001 ) . 3D meshes were generated in the VTK format and viewed using ParaView ( KitWare ) ( Henderson , 2007 ) . Leak in fluorescence was measured by placing a sphere with radius 1 μm into the lumen of the ES and GoFigure2 recorded the integrated fluorescence intensity . Automated image analysis of the AO-LLSM data was performed using modified versions of the ACME software , python shell scripts , and ITIAT scripts ( Mosaliganti et al . , 2012 ) . Starting with 3D tif files , the ACME pipeline includes the following steps , parameters used in ( bold ) : Correcting of segmentations and tracking were done using iterative implementations of the ITIAT scripts membraneSegmentationWithMarkersImageFilter and MorpholigcalErosionLabelImageFilter with manual error corrections done in ITK-SNAP ( Yushkevich et al . , 2006 ) . 3D rendering of segmented labels and overlays of segmented labels and raw membrane signal were performed using the software ITK-SNAP ( www . itksnap . org , [Yushkevich et al . , 2006] ) . 3D visualization of raw membrane data was performed using the software FluoRender ( www . sci . utah . edu/software/fluorender . html , [Wan et al . , 2012] ) . Videos were compiled and encoded using a combination of ImageJ and HandBrake ( [Schindelin et al . , 2012] , www . handbrake . fr ) . For tracking cell thickness , we used the ITIAT scripts CellSegmentationStatistics to map the coordinates of each cell’s centroid , sizeThresh to extract label image files of just the ES lumen , and DistanceFromMask to calculate the physical distance from cell centroids to the surface of the ES lumen labe , which was then multiplied by two to estimate the height of each cell . Cell and lumen volumes were also determined using CellSegmentationStatistics . The outputs from DistanceFromMask and CellSegmentationStatistics were loaded into MATLAB for both graphing and further analysis . For determining the Spearman correlation coefficient between lumen volumes we first parsed the data matrix into time windows when lumen volume exhibited approximately monotonic increasing or decreasing values . For these time windows we used MATLAB’s statistical toolbox to determine the Spearman correlation coefficient and its associate p-value . The syntax for this analysis was: [RHO , PVAL]=corr ( LumenVolume , CellThickness , 'type' , 'Spearman' ) . This test was performed on 99 data windows . MATLAB was also used to perform a Mann-Whitney-Wilcoxon test on the significance cell thickness differences between wild-type and mutant images ( Figure 3 ) . The image processing scripts described above are available through Github at the following repositories: https://github . com/krm15/ACME/tree/MultithreadLookup ( Mosaliganti , 2016; copy archived at https://github . com/elifesciences-publications/ACME ) , https://github . com/krm15/AO-LLSM ( Mosaliganti , 2017a; copy archived at https://github . com/elifesciences-publications/AO-LLSM ) and https://github . com/krm15/GF2Exchange ( Mosaliganti , 2017b; copy archived at https://github . com/elifesciences-publications/GF2Exchange ) The full data-set of serial-section EM images are publicly available at resolutions of 4 . 0 × 4 . 0×60 nm3 per voxel for the closed valve and 18 . 8 × 18 . 8×60 nm3 per voxel for the open valve . To examine the ES anatomy viewers should navigate to http://zebrafish . link/hildebrand16/data/es_closed ( closed valve ) or http://zebrafish . link/hildebrand16/data/es_open ( open valve ) . This set of multi-resolution serial-section EM image volumes was co-registered ( Wetzel et al . , 2016 ) to allow one to navigate the lumen of the endolymphatic duct from the lumen of the otic vesicle to the tip of the ES .	The most internal part of the human ear , the inner ear , is essential for us to hear and have a sense of balance . It is formed by a complex series of connected cavities filled by a liquid . When sound waves and changes in the position of the body make this liquid move , specialized ‘hair’ cells can detect these subtle movements; neurons then relay this information to the brain where it is decoded and interpreted . For the inner ear to work properly , the body needs to finely regulate the pressure created by the liquid inside the cavities . For example , people with unstable pressure in their ears can experience deafness or problems with balance . A structure known as the endolymphatic sac , which is a balloon-like chamber connected to the rest of the inner ear by a thin tube , helps with this regulation . However , scientists are still unsure about how exactly the sac performs its role . One problem is that the inner ear is difficult to study because it is encased in one of the densest bones in the body . Many other animals also have inner ears , from fish to birds and mammals . Here , Swinburne et al . examine the inner ear of zebrafish embryos because , in this fish , the ear starts working before the bones around it form; the structure is therefore accessible for injections and microscopy . Experiments show that when the pressure in the inner ear rises , the endolymphatic sac slowly fills up with the ear liquid , and then it rapidly deflates . Fish with mutations that stop the sac from deflating have overinflated sacs , which is a symptom also found in certain patients with hearing and balance disorders . Looking into the details of these inflation-deflation cycles , Swinburne et al . found that the cells that form the sac have gaps between them , unlike a normal sheet of cells . A flap covers these gaps to keep the liquid in , but under pressure , the flap opens and the liquid can escape . These results show that the endolymphatic sac works as a pressure relief valve for the inner ear . Ultimately , understanding how pressure is regulated in the ear could help patients with inner ear disorders . It could also serve as a template to investigate how eyes , kidneys and the brain , which all have liquid-filled cavities , control their internal pressure .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2018	Lamellar projections in the endolymphatic sac act as a relief valve to regulate inner ear pressure
Horizontal gene transfer ( HGT ) plays a major role in the spread of antibiotic resistance . Of particular concern are Acinetobacter baumannii bacteria , which recently emerged as global pathogens , with nosocomial mortality rates reaching 19–54% ( Centers for Disease Control and Prevention , 2013; Joly Guillou , 2005; Talbot et al . , 2006 ) . Acinetobacter gains antibiotic resistance remarkably rapidly ( Antunes et al . , 2014; Joly Guillou , 2005 ) , with multi drug-resistance ( MDR ) rates exceeding 60% ( Antunes et al . , 2014; Centers for Disease Control and Prevention , 2013 ) . Despite growing concern ( Centers for Disease Control and Prevention , 2013; Talbot et al . , 2006 ) , the mechanisms underlying this extensive HGT remain poorly understood ( Adams et al . , 2008; Fournier et al . , 2006; Imperi et al . , 2011; Ramirez et al . , 2010; Wilharm et al . , 2013 ) . Here , we show bacterial predation by Acinetobacter baylyi increases cross-species HGT by orders of magnitude , and we observe predator cells functionally acquiring adaptive resistance genes from adjacent prey . We then develop a population-dynamic model quantifying killing and HGT on solid surfaces . We show DNA released via cell lysis is readily available for HGT and may be partially protected from the environment , describe the effects of cell density , and evaluate potential environmental inhibitors . These findings establish a framework for understanding , quantifying , and combating HGT within the microbiome and the emergence of MDR super-bugs . The spread of antibiotic resistance among pathogenic microbes is a major and growing threat to public health . Gram-negative Acinetobacter spp . are a particularly worrisome example - these bacteria thrive in hospital settings , causing around 9% of nosocomial infections , particularly in the respiratory tract ( Joly-Guillou , 2005 ) . This prevalence , combined with high levels of antibiotic resistance ( Antunes et al . , 2014 ) , has led the Infectious Diseases Society of America to designate Acinetobacter baumanii one of six particularly problematic multidrug-resistant ( MDR ) pathogens ( Talbot et al . , 2006 ) and the US Centers for Disease Control to assign it threat level ‘Serious’ ( Centers for Disease Control and Prevention ( CDC ) , 2013 ) . Much of the threat posed by Acinetobacter stems from its ability to acquire drug resistance via horizontal gene transfer ( HGT ) ( Peleg et al . , 2012 ) . Acinetobacter has a remarkably high rate of HGT ( Touchon et al . , 2014 ) , and many antibiotic resistance genes in clinical isolates appear to have been recently acquired from other human pathogens ( Adams et al . , 2008; Fournier et al . , 2006; Imperi et al . , 2011 ) . Although A . baumanii can be naturally competent in vitro under some conditions ( Ramirez et al . , 2010; Wilharm et al . , 2013 ) , it is not understood at the population level how Acinetobacter acquires foreign genes at such high rates in real-world conditions , particularly considering that extracellular DNA is rapidly degraded or sequestered ( Nielsen et al . , 2007 ) . For example , DNA lost its transforming ability with a half life of around 1 hr in soil ( Nielsen et al . , 2000 ) and 1 min in saliva ( Mercer et al . , 1999 ) , and ingested DNA was unable to transform even highly competent A . baylyi within the mouse gut ( Nordgård et al . , 2007 ) . Many Streptococcus species also actively exchange genetic material in natural environments . Some of these species use quorum sensing to enhance HGT by co-regulating competence ( the ability to take up extracellular DNA ) with secretion of diffusible bacteriocins ( small protein toxins ) ( Steinmoen et al . , 2002 ) . These toxins cause the release of potential donor DNA by lysing nearby sister cells ( fratricide ) or closely related species ( sobrinocide ) ( Johnsborg et al . , 2008; Kreth et al . , 2005; Wei and Håvarstein , 2012 ) , thereby making their genes available for uptake . However , fratricide-enhanced HGT has not been observed outside Streptococcus , and the narrow target range of bacteriocins limits potential DNA donors . Another killing mechanism , the contact-dependent type-VI secretion system ( T6SS ) , is conserved across a much broader range of bacteria ( Schwarz et al . , 2010a ) . Recently , Vibrio cholera was shown to similarly co-regulate T6SS expression with competence , also enhancing HGT within biofilms ( Borgeaud et al . , 2015 ) . In this work , we extend the observation of killing-enhanced HGT to Acinetobacter , and we develop an experimental and modeling framework to quantify the population dynamics of this phenomenon within microbial communities . As a model system , we used Acinetobacter baylyi , which is closely related to A . baumannii ( Touchon et al . , 2014 ) , because it is genetically tractable , fully sequenced , and well-studied ( Elliott and Neidle , 2011 ) . The two also share the key features of T6SS-mediated killing ( Weber et al . , 2013; Carruthers et al . , 2013 ) and natural competence ( Ramirez et al . , 2010; Wilharm et al . , 2013 ) , and they are similar enough that standard phenotypic assays used in the clinic often fail to distinguish between them ( Chen et al . , 2007 ) . We show that A . baylyi uses its T6SS to lyse and acquire genes from neighboring E . coli cells , and that this process is frequent enough to observe functional and adaptive HGT from prey to predator cells in real time . Prey cell DNA released by predator cells is also available for uptake by other nearby , non-killing cells , highlighting the importance of polymicrobial population dynamics in HGT and the emergence of MDR . To understand the population dynamics of killing-enhanced HGT , we develop a model quantifying both neighbor killing and natural transformation . HGT parameters fit using naked DNA predict transfer of genes released via neighbor killing surprisingly well , suggesting DNA released from cell lysis in situ is readily available for uptake . Using this model , we characterize the impact of neighbor killing on HGT in a wide range of environmental conditions and evaluate potential inhibition strategies , and we experimentally confirm the key predictions . Contact-dependent neighbor killing may be a widespread contributor to HGT among gram-negative bacteria , and for Acinetobacter in particular , killing-enhanced HGT may play a key role in the emergence of clinically pervasive MDR ‘super-bug’ strains . In this context , our dynamic model should be useful for predicting and combating HGT of antibiotic resistance . To study the population dynamics generated by contact-dependent cell killing , we designed a custom microfluidic device ( Figure 1—figure supplement 1 ) enabling visualization of spatially structured communities at single-cell resolution in a monolayer . We seeded the device with a simplified community consisting of predator A . baylyi strain ADP1 ( Elliott and Neidle , 2011 ) ( the predator ) that kill directly adjacent , GFP-expressing Escherichia coli ( the prey ) ( Basler et al . , 2013 ) . As expected , we observed spontaneous lysis of individual E . coli adjacent to Acinetobacter cells . Serendipitously , a few Acinetobacter cells spontaneously began expressing GFP , suggesting that they had acquired the GFP gene via HGT from DNA released by lysed E . coli . In our initial experiments , the GFP-expressing plasmid in E . coli had a ColE1 origin of replication that does not stably replicate in Acinetobacter ( Palmen et al . , 1993 ) . To better observe HGT , we constructed the broad-host plasmid pBAV1k-GFP ( Bryksin and Matsumura , 2010 ) , which can replicate in both Acinetobacter and E . coli . Expression of the GFP gene from this plasmid is repressed in E . coli by LacI , which is absent in Acinetobacter . We transformed E . coli with this plasmid and then seeded them in our microfluidic device alongside mCherry-expressing Acinetobacter ( Figure 1a–d , Videos 1–6 ) . After 3 hr , Acinetobacter had lysed a large number of E . coli cells , and multiple independent HGT events were visible within the device’s approximately 104 um2 traps ( Figure 1b–d ) . Importantly , horizontally acquired GFP was stably maintained during cell division , giving rise to clumps of cells that were both red and green . In the movies , note that although GFP expression in E . coli was repressed by LacI , some E . coli appeared green due to incomplete repression ( see also Supplementary Note 1 ) . HGT of antibiotic resistance allows pathogenic bacteria to survive treatment with antibiotics that would kill the parental strain . In our experiments , the transferring plasmid , pBAV1k-GFP , contained a kanamycin resistance gene in addition to GFP . Therefore , expression of GFP should indicate a newly kanamycin-resistant strain of Acinetobacter , whose de novo appearance would demonstrate the potential clinical relevance of HGT within a microbial community . To test whether this directly observed HGT would be enough to provide a population-level selective advantage , we used our microfluidic chip to visualize functional HGT of antibiotic resistance in real time . We seeded the microfluidic device with mCherry-expressing Acinetobacter alongside E . coli carrying pBAV1k-GFP , grew them for just under 12 hr to allow E . coli lysis and HGT to occur , and then added kanamycin ( Videos 7 , S8 ) . Within about 7 hr of kanamycin addition , multiple HGT events could be visibly identified by the emergence of GFP-expressing Acinetobacter ( arrows in Figure 1e–g ) . Only those Acinetobacter that were expressing GFP , indicating horizontal transfer of kanamycin resistance , continued to grow ( Figure 1h , Videos 7 , 8 ) , while the parental GFP-negative Acinetobacter cells became smaller and stopped dividing . The red and green , dual-labeled Acinetobacter quickly dominated the device , lysing neighboring E . coli ( dark ) and pushing the non-dividing , GFP-negative Acinetobacter cells out of the trap ( Figure 1i–k ) . To our knowledge , these experiments represent the first real-time observation of adaptive , cross-species HGT via natural competence that rapidly enables invading cells to thrive in a new niche . The type VI secretion system ( T6SS ) of Vibrio cholera was recently shown to enhance HGT by promoting the release of DNA from lysed prey cells , thereby making it available for uptake ( Borgeaud et al . , 2015 ) . We hypothesized that T6SS-mediated killing was similarly contributing to the high rate of HGT we observed with Acinetobacter . To test this , we quantified HGT from E . coli to Acinetobacter using bulk experiments , in which we dried small spots of liquid culture onto agar plates . These experiments involve surface-attached cells that are developing spatially-structured communities . This is a qualitatively different condition from growth in shaking culture tubes and more similar to biofilm dynamics , although we did not study true biofilms , which take longer to mature . In these bulk experiments , we co-cultured Acinetobacter carrying genomic spectinomycin ( spect ) resistance and E . coli carrying both a genomic chloramphenicol ( cm ) resistance marker and the ‘bait’ kanamycin ( kan ) resistance plasmid pBAV1k . We placed the cm resistance gene in a region of the E . coli genome with minimal homology to Acinetobacter to preclude its horizontal transfer , enabling us to quantify the remaining E . coli and thereby measure killing efficiency . We co-cultured Acinetobacter ( spect ) and E . coli ( cm+kan ) , seeded together at roughly equal concentrations , on agar plates at 30∘C overnight to allow for T6SS-mediated killing and HGT . We also cultured each species alone as controls . We quantified the efficiency of E . coli lysis and HGT of kanamycin resistance by resuspending the initial spots , spotting serial dilutions of that resuspension onto selective agar plates ( spect , cm+kan , and spect+kan ) , and counting colony-forming units ( CFUs ) . As expected , Acinetobacter dramatically reduced E . coli numbers during co-culture ( Figure 2a , cm+kan ) . Simultaneously , a novel spect+kan double antibiotic-resistant phenotype appeared . We picked three clones with this novel spect+kan phenotype and confirmed them to be Acinetobacter by microscopy and 16S sequencing . Transfer of the genomic cm E . coli marker to Acinetobacter was below our limit of detection ( 2500 CFUs in this case ) , as we did not recover any spect+cm resistant cells . Next , we confirmed the role of T6SS-mediated killing in HGT to Acinetobacter by replacing the T6SS structural gene hcp ( Carruthers et al . , 2013; Basler et al . , 2013 ) with tetracycline resistance . To compare these non-killing Acinetobacter Δhcp with the wild type ( WT ) , we incubated E . coli carrying pBAV1k on agar plates at 37°C for 150 min either alone or mixed with WT , Δhcp , or both strains of Acinetobacter , with Acinetobacter at optical density ( OD ) five and E . coli at OD 1 ( Figure 2b , c ) . No E . coli were detected after co-incubation with WT Acinetobacter ( Figure 2b , second bar ) . The limit of detection was 10 CFUs , which was significantly different from all other conditions at p=0 . 011 ( see Materials and methods for statistical tests ) . In contrast , survival of E . coli mixed with Acinetobacter Δhcp was not different from cultures containing E . coli alone ( p=0 . 063 ) , confirming a dramatic reduction of killing efficiency for Acinetobacter Δhcp ( Figure 2b , third bar ) . At the same time , elimination of killing impaired HGT by about 100-fold ( Figure 2c , dark bars , p<0 . 001 ) . Impaired HGT was not due to a competence defect , since WT and Δhcp Acinetobacter had equal competence for purified plasmid DNA ( Figure 2d , p=0 . 57 ) . Furthermore , in a time course experiment with each Acinetobacter strain co-cultured at 37°C with approximately equal concentrations of E . coli ( Figure 2e–g ) , we detected the spect+kan dual drug-resistant phenotype ( indicating HGT ) 2 hr earlier for WT cells ( Figure 2e ) than for non-killing Δhcp cells ( Figure 2f , also compare solid to dashed lines in Figure 2g for fraction of transformed Acinetobacter cells ) . Presumably , the residual HGT to Acinetobacter Δhcp results from slow , but continued , spontaneous DNA release from E . coli , possibly from spontaneous cell lysis . Together , these results and those reported for Vibrio cholera ( Borgeaud et al . , 2015 ) support a model in which the T6SS of Acinetobacter promotes HGT by killing neighboring E . coli , which then releases its DNA for subsequent uptake by Acinetobacter . If this model is correct , then WT Acinetobacter would complement HGT to the killing-defective Δhcp Acinetobacter strain in trans by lysing neighboring E . coli , thereby releasing their DNA into an extracellular pool accessible to all competent cells . To test this , we mixed the tetracycline ( tet ) -resistant Acinetobacter Δhcp with the spect-resistant WT Acinetobacter and co-cultured them in a ‘3-strain’ spot along with E . coli , seeding Acinetobacter at total OD = 5 and E . coli at OD = 1 . Survival of E . coli in the three-strain culture was reduced by about 104-fold ( Figure 2b fourth bar , p<0 . 001 ) , although their survival was improved relative to E . coli mixed with only WT Acinetobacter , where E . coli were reduced below the 10 CFUs limit of detection ( p=0 . 011 ) , presumably due to the lower fraction of killing cells in the three-strain culture . Consistent with the model , killing of E . coli by WT Acinetobacter in three-strain cultures was sufficient to increase HGT to co-cultured Acinetobacter Δhcp by about 100-fold ( Figure 2c , compare two- to three-strain communities in right bar group , p<0 . 001 ) , while HGT to the WT remained unchanged relative to a two-strain community ( Figure 2c , left bar group , p=0 . 68 ) . We found similar results for a time course experiment comparing HGT in two-strain ( Figure 2h ) and three-strain ( Figure 2i ) communities: co-culture with WT Acinetobacter dramatically increased HGT from E . coli to non-killing Acinetobaceter Δhcp . Interestingly , even in three-strain communities , HGT to Acinetobacter Δhcp was slightly less than HGT to the WT ( Figure 2c light bars , p<0 . 014 , and Figure 2i ) . This likely indicates a spatial effect - actively killing predator cells will be physically closest to the DNA released from their lysed prey . Together with experiments below showing HGT inhibition by extracellular DNase ( Figure 7e ) , these results support a model in which the T6SS releases DNA into the extracellular environment via lysis of neighboring cells , thereby making it available for uptake by any nearby cell , but the T6SS is not directly involved in DNA uptake ( Figure 2j ) . To further characterize killing-enhanced HGT in Acinetobacter , we analyzed the effect of genetic context . Previous experiments have shown that A . baylyi is competent to take up and replicate plasmids with a limited set of origins of replication , they require genomic homology for efficient recombination into their genome , and both genomic and plasmid DNA can serve as a source for recombination ( Palmen et al . , 1993 ) . However , those experiments were done using chemically purified , exogenously added DNA . Therefore , we experimentally confirmed them for the specific case of killing-enhanced HGT . We co-cultured Acinetobacter on agar plates with E . coli containing each of three different bait plasmids conferring kan resistance: pBAV1k that replicates in both species , pRC03 that cannot replicate in Acinetobacter , and pRC03H - a derivative of pRC03 containing 3 kb of Acinetobacter genomic homology to promote genomic integration ( de Vries and Wackernagel , 2002 ) . Acinetobacter acquired the kan resistance gene from either pBAV1k or pRC03H , but not from pRC03 ( Figure 2—figure supplement 1a ) . We also tested whether contact-dependent killing enables Acinetobacter to efficiently acquire genes from the E . coli genome . To do so , we created a bait plasmid containing kan resistance adjacent to 22 kb of Acinetobacter genomic homology . We transformed this plasmid into E . coli to create a plasmid bait strain , and then we integrated it into the E . coli genome to create a genomic bait strain ( see Materials and methods ) . Acinetobacter acquired the kan resistance gene regardless of its location ( Figure 2—figure supplement 1b ) , but HGT was approximately 100-fold lower when the kan gene was in the E . coli genome . Considering that the bait plasmid had a pBR322 origin of replication , with 10–20 plasmids per cell , the per-copy difference was only about 5- to 10-fold . Finally , we picked three double-resistant Acinetobacter clones each that had acquired either the replicating or homology plasmid for further analysis , isolated both plasmid and genomic DNA , and confirmed that the replicating plasmid pBAV1k had transferred as an episomal unit , whereas the homology plasmid pRC03H had integrated into the genome ( Figure 2—figure supplement 2 ) . These results do not rule out additional lower frequency mechanisms , but they indicate that previous characterizations of competence using purified DNA are equally applicable to in situ HGT within spatially structured microbial communities . While killing-enhanced HGT has now been observed in multiple genera , little is known about the population dynamics of this process , its efficiency , or how that efficiency is influenced by environmental conditions . This lack of understanding is largely due to the absence of a quantitative method to measure mechanistic HGT parameters . Currently , HGT is generally quantified using the fraction of transformed CFUs or transformed CFUs per ng of DNA . This is problematic , because results depend on multiple parameters extrinsic to inherent competence , including cell concentration , DNA concentration , and incubation time . This makes results difficult to compare across experiments , conditions , strains , or even labs . A more useful approach would be to quantify transformation using variables intrinsic to the cells and the DNA , such as the DNA uptake rate and transformation efficiency per molecule of DNA . To this end , we developed a population dynamic model of spatially structured microbial communities that couples transformation via natural competence and contact-dependent killing ( summarized in Box 1 and explained in more detail in the Materials and methods ) . The model parameters pertaining to growth , transformation , and killing can each be measured sequentially in simplified conditions ( Figure 3 , Figure 3—figure supplements 1–3 , Table 1 ) . For all parameter fitting , we incubated cells in spots on agar plates , the same condition used to measure HGT in dual-species communities ( see Materials and methods ) . First , we measured growth parameters for each species separately ( Figure 3—figure supplements 1–2 , and see Materials and methods ) . Second , we fit the natural transformation parameters using data obtained by mixing Acinetobacter with known concentrations of the genomically integrating kan resistance plasmid pRC03H-2S , derived from pRC03H used earlier , with higher efficiency integration via fully homolgous recombination ( HR ) ( Palmen et al . , 1993 ) , rather than homology-facilitated illegitimate recombination ( de Vries and Wackernagel , 2002 ) as in Figure 2 , see Materials and methods . We incubated the cell-DNA mixtures on agar plates , and then we counted the number of cells that had acquired kan resistance . To obtain values for the HGT parameters , we simultaneously fit data from time courses with either limiting ( Figure 3a ) or saturating ( Figure 3b ) DNA , and from DNA dilution series with cells harvested at different time points ( Figure 3c ) . Third , we fit parameters for T6SS-mediated killing by mixing different concentrations of E . coli and Acinetobacter , incubating them together on agar plates , and counting the cells of each type that were present upon harvesting . As with DNA uptake , we fit killing parameters simultaneously for several time courses ( Figure 3d–i and Figure 3—figure supplement 3 ) and dilution series ( Figure 3j ) . Finally , to determine the ‘leak’ rate of DNA from E . coli , we measured HGT to non-killing Acinetobacter Δhcp and found the best fit for rl⁢e⁢a⁢k , with the killing rate rk⁢i⁢l⁢l set to 0 in the model ( Figure 3k , l ) . The leak rate may be due to spontaneous E . coli cell death , but we do not include spontaneous death in the differential equation for E . coli because it is inherently included when measuring the bulk growth rate γE . Using this experimentally parameterized model , we asked how well the transformation parameters that we measured using purified DNA would predict killing-enhanced HGT in dual-species microbial communities . The model closely matched experimental results , suggesting that DNA released from cells in situ is equivalently available for uptake as DNA purified in vitro ( Figure 4 ) . Note that counting CFUs in a spot on an agar plate is a destructive measurement , as it requires harvesting and resuspending the entire spot , so successive data points in a time course are actually from different individual spots that were seeded at the same time from the same cell mixture . Experimental measurement of HGT is time-consuming and labor-intensive , limiting the study of how it is affected by environmental conditions . However , our experimentally parameterized , mechanistic model allows us to simulate a much wider variety of conditions . This allows us to predict the most conducive conditions for HGT , the conditions in which contact-dependent killing plays an important role , and what strategies are most likely to inhibit HGT , and thus the spread of MDR . First , to determine the effect of bacterial seeding density on the fraction of transformed Acinetobacter , we simulated surfaces seeded with varying densities of the two species and incubated for two hours . Wild-type Acinetobacter had the highest transformation frequency when both species were seeded at higher densities , allowing maximal contact ( Figure 5a ) . This can be understood by considering that the killing rate , and thus DNA release , depends on the product of the cell counts of both species ( Box 1 ) , so the most DNA is released when both species are at high density . In contrast , the transformation frequency of killing-deficient Acinetobacter was mainly dependent on only the E . coli seeding density ( Figure 5b ) , because DNA release from spontaneous lysis depends only on the number of E . coli . From these results , we calculated how the relative importance of killing for HGT depends on seeding density . We defined the degree to which killing of E . coli increases HGT to Acinetobacter - that is , the enhancement factor - as the ratio of HGT to WT Acinetobacter divided by HGT to the killing mutant Acinetobacter in the same condition . This killing enhancement was greatest when Acinetobacter was seeded at high density and E . coli at low density , in which case killing enhanced HGT by nearly 1000-fold ( Figure 5c ) . Next , we used our model to explore how killing-enhanced HGT interacts with incubation time . We simulated a surface seeded with both Acinetobacter and E . coli at 10−3 of their respective carrying capacities and calculated HGT to either killing or non-killing Acinetobacter over time . HGT increased with time for both the WT and the killing mutant Acinetobacter ( Figure 5d ) , but the enhancement of HGT provided by killing was greatest within the first few hours ( Figure 5e ) . Varying the initial seeding density ( Figure 5f ) revealed that at higher densities , the enhancement factor was greatest immediately after seeding , whereas at lower seeding densities , the enhancement factor did not peak until up to 4 hr after seeding . Killing still enhanced HGT by more than 10-fold for a wide range of seeding densities even after 10 hr ( Figure 5g ) , which is long enough to reach the carrying capacity . Overall , these results show that the degree to which contact-dependent killing increases HGT to predatory bacteria is influenced by the total initial cell density , the ratio of the two species , and the length of time that the community has to grow . In the case of transferring antibiotic resistance genes , it would be desirable to inhibit HGT . Therefore , we used our model to explore how well killing-enhanced HGT could be blocked by two potential environmental perturbations: either degradation by DNase or competitive inhibition with added DNA . First , we determined the DNA degradation rate and amount of competing DNA that would be required to inhibit HGT . We simulated surfaces seeded with both species at 10−3 of their respective carrying capacities and grown for 2 hr . At this seeding density , DNase was effective when it reduced DNA half life to less than about 2–3 min ( Figure 6a ) , and competing DNA was effective at about 109 kb or greater ( Figure 6e ) . We then explored how well each condition would inhibit HGT in microbial communities seeded at different cell densities and species ratios . We simulated surfaces seeded with varying numbers of E . coli mixed with either killing or non-killing Acinetobacter , either with DNA half life fixed to 1 min ( Figure 6b–d ) or with 1011 kb of competing DNA added at time 0 ( Figure 6f–h ) . To quantify the efficacy of each inhibition strategy , we calculated the HGT reduction factor , defined as the ratio of HGT without inhibition to HGT with inhibition . The HGT reduction factor depended on seeding density to varying degrees for both WT and non-killing ( Δhcp ) Acinetobacter for both strategies - degradation with DNAse ( Figure 6b , c ) and competitive inhibition ( Figure 6f , g ) . Our model had previously predicted contact-dependent killing-enhanced HGT to WT Acinetobacter to occur most frequently when both species begin at high density ( see Figure 5a ) , so that may be the most important condition in which to inhibit HGT . Importantly , our model predicted DNAse to remain effective even at high initial cell density ( Figure 6b ) , whereas competing DNA was predicted to dramatically lose efficacy in that condition ( Figure 6f ) . This suggests that DNase may better inhibit HGT in dense communities such as biofilms than competitive inhibition with exogenous DNA . Regardless of the inhibition strategy , killing consistently increased HGT to the wild type relative to the killing mutant , even in the presence of the inhibitors ( Figure 6d , h ) . Finally , we experimentally tested the predictions from our model , focusing on cases with relatively high cell numbers where HGT events are detectable . With respect to seeding density , the model predicted experimental results quite well ( see Figure 5a–c ) . For WT Acinetobacter mixed with E . coli carrying pRC03H-2S , HGT was greatest when both species were at high initial density , and it decreased along with seeding counts of either species ( Figure 7a ) . In contrast , for non-killing Acinetobacter Δhcp , the HGT frequency was mainly dependent on the initial number of E . coli ( Figure 7b ) . The HGT enhancement factor was greatest when the predator was seeded at high density , while the prey was at low density ( Figure 7c ) . Also consistent with the model , the HGT enhancement decreased for all seeding densities after overnight growth ( compare Figure 7c to Figure 7d ) . Units of DNase are not easily converted to DNA half life on an agar substrate , so we tested a four-fold DNase dilution series ( Figure 7e ) with the two species seeded at approximately equal numbers . As predicted , DNase effectively inhibited HGT even to the wild type at high seeding density ( approximately 3 × 106 CFUs of each species ) . When cells were seeded at 100-fold lower density , or Acinetobacter was killing-deficient , DNase reduced HGT to an even greater extent . All tested levels of DNase reduced HGT below detection ( 5 CFUs ) for Acinetobacter Δhcp seeded at the lower density . In contrast , 1011 kb of competing DNA was ineffective against HGT to WT Acinetobacter at high seeding density ( Figure 7f ) . At 100-fold lower seeding density , or when Acinetobacter was unable to kill E . coli , competing DNA did inhibit HGT . The standard error for these HGT reductions appears relatively large because calculating HGT reduction requires four measurements - total CFUs and transformed CFUs for each of two conditions - and the Poisson-distributed measurement error compounds with each additional measurement ( see Materials and methods ) . Nevertheless , the results were repeatable over 3 separate days , and the figure shows their average . While our model predicted the key qualitative features of cell seeding density , extracellular DNase , and competing DNA , there were some discrepancies . In particular , HGT inhibition by DNase leveled off above 0 . 1 units per spot , suggesting a sub-population of Acinetobacter with privileged access to released DNA , and inhibition by competing DNA was less than predicted in all conditions . This may be a result of the fact that our model does not capture spatial heterogeneity in DNA concentrations . Lysed E . coli release DNA at a locally high concentration , which may provide privileged access for directly adjacent Acinetobacter . The discrepancies may also reflect the technical difficulty of delivering molecules evenly to real-world communities , which would be exacerbated for mature biofilms . Humanity’s dwindling arsenal of antibiotics is a significant and growing concern . This threat stems from pathogens such as Acinetobacter that are able to rapidly accumulate multiple resistance genes , which can make them nearly impossible to treat . High-throughput sequencing has revealed evidence for widespread horizontal gene transfer , but unlike conjugation and phage transduction , we still know relatively little about the microbial dynamics underlying HGT via natural competence ( Mao and Lu , 2016; Johnsborg et al . , 2007 ) . It has been observed qualitatively that killing of nearby cells - via fratricide ( Kreth et al . , 2005; Wei and Håvarstein , 2012 ) , sobrinocide ( Johnsborg et al . , 2008 ) , or contact-dependent killing ( Borgeaud et al . , 2015 ) - can enhance HGT , but a lack of quantitative methods has precluded a fuller understanding of the importance of microbial combat in the horizontal spread of genes . In this paper , we showed that contact-dependent killing by Acinetobacter can increase HGT rates from E . coli by up to 3 orders of magnitude , making it frequent enough to observe multiple events in real time within microfluidic chips . By subsequently adding kanamycin to our chips , we observed functionally adaptive emergence of newly drug-resistant bacteria in situ . Significantly , we also showed that killing by one strain in a spatially-structured community makes DNA available for uptake by nearby , non-killing cells . This highlights the role that polymicrobial interactions can play in facilitating HGT . We then developed population dynamic models for both natural transformation and contact-dependent killing , and we fit them to experimental data to obtain biologically relevant parameters . Interestingly , DNA uptake and transformation parameters fit using purified plasmid DNA accurately predicted HGT by DNA released in situ . This was not obvious a priori , particularly given that previous experiments have shown extracellular DNA to rapidly lose its transforming ability in real-world conditions ( Mercer et al . , 1999; Nielsen et al . , 2000; Nielsen et al . , 2007; Nordgård et al . , 2007 ) . For contact-dependent killing , predation can occur only at the perimeter of micro-colonies , where there is contact between predator and prey cells ( Schwarz et al . , 2010a; Borenstein et al . , 2015; Hood et al . , 2010; Schwarz et al . , 2010b; MacIntyre et al . , 2010; LeRoux et al . , 2012 ) ( see also Supplemental Videos 9–12 ) . In our model , we derived an approximation for the number of cells at the perimeter of their respective colonies , N* ( see Box 1 and Materials and methods ) . This allowed us to approximately account for the difference between interior and exterior prey cells , while maintaining the simplicity of a system of ordinary differential equations . This use of N* to represent perimeter cells yielded a slightly better optimal fit to our data than assuming all cells are equally vulnerable , but the difference was not dramatic . Using all cells N rather than perimeter cells N* , the optimal fit was rk⁢i⁢l⁢l=17 hr−1 , Kk⁢i⁢l⁢l=1 . 3 x 106 , KE⁢_⁢k⁢i⁢l⁢l=21 ( compare to values in Table 1 ) , and the best fit sum of squares of residuals increased from 87 . 0 to 93 . 9 ( calculated using log10-transformed data ) . See also Figure 3—figure supplement 4 for a comparison of simulations using the same parameters , fit using the restriction of killing to perimeter cells and shown in Table 1 , simulated with and without that restriction . The difference between the total number of cells and the number of perimeter cells only becomes significant when the population has grown much greater than the initial cell number ( compare solid blue to dashed orange lines in Figure 3—figure supplement 5 ) , so we would expect it to be more important in longer time courses . The model presented here can be modified for the case when killing is mediated by diffusible molecules , where there is no distinction between interior and exterior cells in a micro-colony , by simply replacing N* with N . Using our experimentally parameterized model , we characterized contact-dependent killing-enhanced HGT in a wide range of simulated conditions . Our model revealed that killing is most important for HGT when the prey is at low density , the predator is at high density , and the interaction time is short ( Figure 5c , f ) , which was confirmed by experimental data ( Figure 7a–d ) . Killing is less important for HGT when the prey outnumbers the predator , because enough prey DNA is released by spontaneous lysis that the additional DNA released by killing no longer provides much advantage . Similarly , killing provides less HGT enhancement when the interaction time is longer , because killing Acinetobacter deplete their donor DNA source by killing neighboring prey E . coli , whereas the non-killing mutants have a slower , but exponentially growing , source of DNA from spontaneously lysing E . coli . Interestingly , this seeding ratio at which contact-dependent killing most enhances HGT - when the predator outnumbers the prey - is the same condition in which it provides the strongest competitive advantage in terms of cell growth ( Borenstein et al . , 2015 ) . High predator cell density is also the condition that induces fratricide-mediated HGT in Streptococci ( Steinmoen et al . , 2002 ) and simultaneously induces both T6SS expression and competence in V . cholera ( Borgeaud et al . , 2015 ) . However , T6SS regulation does not always depend on high cell density , and it can be quite complex and varied , even within strains of the same species , likely reflecting the wide range of ecological contexts and functions performed by the T6SS ( Miyata et al . , 2013; Bernard et al . , 2010; Weber et al . , 2013; Weber et al . , 2015 ) . Given that the only demonstrated advantages are at high cell density , it remains an open question what , if any , selective advantages the T6SS may provide at lower cell density . While T6SS regulation in A . baylyi has not been extensively characterized with respect to cell density ( Weber et al . , 2016 ) , it is active and functional in standard in vitro conditions ( Weber et al . , 2013; Basler et al . , 2013 ) , and we observed T6SS-mediated killing of E . coli at all times in the Supplemental Videos 1–6 . We also used our model to evaluate the effect of two conditions that may inhibit HGT: addition of either competing DNA or DNAse . Although our model predicted neither condition would eliminated the HGT enhancement provided by killing in our model , both strategies were predicted to reduce HGT when at realistic levels . DNA degradation began to reduce HGT when DNA half life was less than 2–3 min ( Figure 6a ) , which is just longer than the half life of free DNA in saliva , about 1 min ( Mercer et al . , 1999 ) . Competing DNA began to reduce HGT at about 109 kb or greater ( Figure 6e ) , and given that our agar spots could support 107–108 cells , that would be 10–100 kb of extracellular DNA per cell at carrying capacity . Importantly , DNase was predicted to be more effective than competing DNA at high cell seeding densities , which is the condition where contact-dependent killing-enhanced HGT is most efficient . Our experimental results supported this , showing that DNase can reduce HGT to wild type Acinetobacter by nearly 104-fold even at high seeding densities , whereas 1011 kb of competing DNA was ineffective at high cell density . The reduced efficacy of competing DNA in this condition is likely because the high initial killing rate releases a large amount of prey DNA at once , overcoming competitive inhibition . Interestingly , competing DNA was nearly 10-fold less effective than predicted , which may reflect its physical exclusion from dense microbial communities , or the fact that our model does not reflect the spatial heterogeneity of DNA released from lysing E . coli . The potential importance of non-isotropic DNA concentration is also suggested by the observation that there appears to be a small sub-population of Acinetobacter with access to prey cell DNA protected from DNase ( Figure 7e ) , and the observation that even in three-strain communities , HGT to Acinetobacter Δhcp remains slightly lower than HGT to WT cells ( Figure 2c , i ) . Spatial dependence of DNA concentration would be a valuable limitation to address in future work; nevertheless , the qualitative trends predicted by our model show its usefulness in guiding experiments and intuition . Additionally , the observation that contact-dependent killing appears to provide a sub-population of cells with privileged access to DNA that is protected from DNase may help explain how HGT can occur so frequently in real-world environments where DNA is quickly degraded . While we used A . baylyi as a model for the more clinically threatening A . baumannii , the two are closely enough related that they are difficult to distinguish with standard phenotypic tests used in the clinic ( Chen et al . , 2007 ) . Indeed , careful re-examination of clinical isolates recently revealed that A . baylyi can also opportunistically infect humans , including case clusters in hospitals ( Chen et al . , 2008 ) . It was a clinical A . baylyi isolate that contained the first blaSIM-1 carbapenemase reported in China ( Zhou et al . , 2011 ) , and the difficulty of distinguishing between the two suggests that A . baylyi may in fact be causing additional infections that are either misidentified as A . baumannii or identified nonspecifically as Acinetobacter ( Chen et al . , 2007 , 2008 ) . Perhaps more importantly , A . baylyi may serve as an easily sampled reservoir of genetic diversity for A . baumannii , due to the extremely high transformability of A . baylyi coupled to the ease with which Acinetobacter species exchange DNA . For example , another carbapenemase gene , blaOXA23 , has been found in multiple Acinetobacter species isolated from both humans and animals , including both A . baumannii and A . baylyi ( Smet et al . , 2012; Nigro and Hall , 2016 ) . In addition to helping to explain , predict , and combat the rapid spread of multi drug-resistance , particularly among Acinetobacter , the population dynamics revealed here are likely important in the microbiome and microbial evolution more broadly . Both the T6SS ( Schwarz et al . , 2010a ) and natural competence ( Johnsborg et al . , 2007 ) are widespread among gram-negative bacteria , and both competence and other neighbor killing mechanisms are found in both gram-negatives and gram-positives . Therefore , our results are likely broadly generalizable and may contribute to understanding the known increase of HGT within biofilms ( Madsen et al . , 2012 ) . The method presented here for quantifying natural competence with standardized parameters may also help researchers address other outstanding questions about microbial evolution ( Vos et al . , 2015 ) . In the light of a growing threat from antibiotic resistance , the quantitative methods presented here should greatly aid study of the mechanisms by which bacteria swap genes , shortcut evolution , and outsmart our drugs . We obtained Acinetobacter baylyi sp . ADP1 from the ATCC and used a lab strain of E . coli MG1655 . We inserted spectinomycin resistance and mCherry into a putative prophage region of the Acinetobacter genome , as described previously ( Murin et al . , 2012 ) . To give E . coli genomic chloramphenicol resistance , we first inserted a short ‘landing pad’ from pTKS/CS , including tetracycline resistance , into a neutral site on the genome ( between atpI and gidB ) , using the recombineering helper plasmid pTKRED ( Kuhlman and Cox , 2010 ) . We then replaced this landing pad with the chloramphenicol marker from donor plasmid pTKIP-cat , as described previously ( Kuhlman and Cox , 2010 ) . We cured the cells of pTKRED by growing them overnight with saturating IPTG and arabinose and then screening for spectinomycin sensitivity . For a replicating bait plasmid , we used the broad-host plasmid pBAV1k ( Bryksin and Matsumura , 2010 ) , to which we introduced a GFP gene . We derived our genomically integrating plasmids from and the ColE1 origin plasmid pRC03 , which replicates in E . coli but not in Acinetobacter . To add Acinetobacter homology to pRC03 , we inserted 7 kb of Acinetobacter genomic sequence ( covering genes ACIAD3424 to ACIAD3429 ) adjacent to the kanamycin marker , yielding pRC03H . To make a higher efficiency integrating plasmid with genomic homology on both sides of the marker ( pRC03H-2S ) , we moved the kanamycin marker into the middle of the Acinetobacter sequence . To compare acquisition of plasmid vs . genomic resistance genes , we inserted 22 kb of Acinetobacter genomic sequence ( covering genes ACIAD2681 to ACIAD2697 ) into the donor plasmid pTKIP-neo and transformed this into the landing pad-containing strain described above . We then used the helper plasmid pTKRED to integrate this sequence in place of the landing pad and finally cured the integrants of pTKRED ( Kuhlman and Cox , 2010 ) . We constructed the the non-killing Acinetobacter Δhcp by first fusing the tetracycline resistance marker from pTKS/CS to approximately 400 bp homology arms amplified from either side of hcp ( ACIAD2689 ) in the A . baylyi genome . For microfluidic experiments , we seeded a custom microfluidic device ( Figure 1—figure supplement 1 ) with both Acinetobacter and E . coli and imaged it on a Nikon TI microscope with an incubated stage set to 37°C and a gravity-driven nutrient flow ( LB broth ) ( Ferry et al . , 2011 ) . We imaged the experiment shown in Figure 1a–d with a 40x objective and that shown in Figure 1e–k with a 60x objective . Note that microfluidics with A . baylyi are extremely challenging technically , due to its propensity to adhere to and clog the feeding channels ( see the horizontal channels at the tops and bottoms of Videos 1-S6 ) . We grew spatially structured communities on LB agar plates ( 1 . 5% ) in an incubator at 30∘C overnight or 37°C for the indicated time . We seeded the communities with 2 ul spots of cell culture , being careful not to introduce bubbles , which can spray aerosolized cells across the plates when they burst . Before seeding , we grew the cells overnight , resuspended them at 1:50 into fresh media , grew them again for 2–3 hr , washed them , and then resuspended again in fresh media . For time-course experiments , the initial cell density can be seen from the CFUs at time 0 . We harvested spots by cutting out spots with a razor blade and resuspending them in 500 ul of PBS buffer . To count CFUs , we made serial ten-fold dilutions of the resuspended cells , spotted 2 ul of each dilution onto selective plates , and counted colonies after incubation overnight . The limit of detetion by this method is 5002=250 CFUs . To lower our limit of detection ( Figure 2b bars 2 and 4 , Figure 2c bar 3 ) , we also spread 50 ul of resuspended cells across selective plates , achieving a theoretical limit of detection of 10 CFUs . To characterize the genomic result of HGT , we picked three clones each of Acinetobacter that had acquired kan resistance via either pBAV1k or pRC03H . We isolated any plasmid DNA from these clones and from the bait E . coli carrying pBAV1k using a standard miniprep kit ( Qiagen , Hilden , Germany ) , digested with SspI , which should cut twice , and ran the result on a gel ( Figure 2—figure supplement 2a ) . To verify genomic integration of the kan gene from pRC03H , we isolated genomic DNA and performed inverse PCR from the kan gene . In particular , we digested the genomic DNA with SalI , diluted to 2 ng/ul , and used T4 DNA ligase to circularize the resulting fragments . Next , we performed PCR using primers pointing outward from the kan gene to amplify the circularized , surrounding genomic region . All three clones had the same PCR pattern ( Figure 2—figure supplement 2b ) , and sequencing confirmed that the surrounding region was Acinetobacter genome . All enzymes used for cloning were from New England Biolabs ( Ipswitch , Massachusetts ) . When DNA or DNase was added to the cells , we added it to the cell mixture on ice just before spotting . We used PureLink DNase , resuspended in water at the recommended concentration ( 2 . 7 units/ul ) , frozen in aliquots , and diluted to the desired concentration when needed . For competitive inhibition , we used a high-copy , ampicillin resistance plasmid with a ColE1 origin of replication , which cannot replicate in Acinetobacter and does not provide resistance to any of the antibiotics in our experiments ( pBest , Promega , Madison , Wisconsin ) . We grew this in E . coli , extracted it with a maxiprep kit , and concentrated it using ethanol precipitation . Error bars in Figure 2b–d indicate the standard deviations of measurements pooled across two culture replicates , each with three measurement replicates . Culture replicates refer to distinct spots seeded with the same cell mixture at the same time . Each resuspended culture was then serially diluted , and these 10-fold dilutions were themselves spotted onto selective plates to count CFUs ( measurement spots ) , as described above . However , the CFUs in each of these measurement spots are expected to be Poisson-distributed , so we spotted each dilution three times , to obtain measurement replicates . Error bars in Figure 2a are for three measurement replicates of one culture each . Where data was less than 250 CFUs , only one ( larger volume ) measurement replicate was counted for each community , as described above . To calculate the pooled variance of HGT frequency across both types of replicates , we followed the following procedure . To calculate statistical significance , we performed analysis of variance ( ANOVA ) . In particular , we calculated p-values with the MATLAB function multcompare , using the condition means and variances calculated above . Spotting serial dilutions to count CFUs measures data on a log base 10 scale , so to compare data separated by more than one order of magnitude , we first calculated the log base 10 value of each data point ( Figure 2b , survival of E . coli with both strains of Acinetobacter , Figure 2c , HGT to Acinetobacter Δhcp in 2-strain communities ) . To test for significance where data was below the limit of detection ( Figure 2b , survival of E . coli with WT Acinetobacter ) , we used one-tailed t-tests to determine whether each of the other conditions were significantly greater than that limit of detection ( not log-transformed ) , reporting the largest p-value . In our model ( Box 1 , Figure 2j , Table 1 ) , both E . coli and Acinetobacter grow according to the logistic growth equation d⁢Nd⁢t=γN⁢N⁢ ( 1-NKN ) , where N is the number of cells of one species , γN is the growth rate , and KN is the carrying capacity for 2 ul spots , and we multiplied the two saturation terms ( 1-NKN ) to couple the growth saturation of the two species . Acinetobacter takes up DNA with maximal rate c and saturation constant KD⁢N⁢A , this DNA transforms Acinetobacter from species A1 to A2 with efficiency ϵ , Acinetobacter kills E . coli , thereby releasing DNA , and E . coli leaks DNA at a basal level rl⁢e⁢a⁢k . All variables are absolute numbers , not concentrations . We modeled the DNA uptake and E . coli killing terms using Michaelis-Menton kinetics . Since we did all experiments on equal-sized agar spots ( resulting from evaporation of 2 ul droplets ) , we assumed constant volume and combined V with the Michaelis constant KMD⁢N⁢A to obtain a new parameter KD⁢N⁢A , with units kb , as demonstrated below for the DNA uptake term: ( 1 ) d⁢Dd⁢t=-c⁢A⁢DVKMD⁢N⁢A+DV ( 2 ) d⁢Dd⁢t=-c⁢A⁢DKD⁢N⁢A+D;KD⁢N⁢A≡KMD⁢N⁢A⁢V Analogously , the killing term -rk⁢i⁢l⁢l⁢A⁢EVKMk⁢i⁢l⁢l+AV+KEk⁢i⁢l⁢l⁢EV becomes -rk⁢i⁢l⁢l⁢A⁢EKk⁢i⁢l⁢l+A+KEk⁢i⁢l⁢l⁢E , where Kk⁢i⁢l⁢l≡KMk⁢i⁢l⁢l⁢V is unitless . For contact-dependent killing , predation can occur only at the perimeter of micro-colonies where there is contact between predator and prey cells ( Schwarz et al . , 2010a; Borenstein et al . , 2015; Hood et al . , 2010; Schwarz et al . , 2010b; MacIntyre et al . , 2010; LeRoux et al . , 2012 ) ( see also Supplemental Videos 9–12 ) , and that perimeter grows proportionally to the square root of the total cell count . Therefore , the relevant quantities in the killing terms are the numbers of perimeter cells , A* and E* , not the total numbers of cells , A and E . Consider a large micro-colony i with Ni cells of radius r and total colony radius Ri , where Ni≫1 . The number of cells in the micro-colony is the area of the colony divided by the area of a single cell , Ni=π⁢Ri2π⁢r2 , so the colony radius is Ri=r⁢Ni . The number of perimeter cells is Ni=2⁢π⁢ ( Ri-r ) 2⁢r , and substituting for Ri , we find Ni=π⁢ ( Ni-1 ) . Each micro-colony grows from a single cell , so assuming equal distribution of cells across micro-colonies and initial number of cells N0 , we have Ni=NN0 , so Ni=π⁢ ( NN0-1 ) . Since there are N0 micro-colonies , the total number of perimeter cells is: ( 3 ) N=π⁢N0⁢ ( NN0-1 ) Equation 3 is valid for large micro-colonies ( Ni≫1 ) , but when colonies consist of only one cell each; that is , N≤N0 , all cells are perimeter cells , so ( 4 ) N=N Therefore , we define a function for the number of perimeter cells that transitions smoothly between the two limits of Equation 3 and Equation 4 at a threshold micro-colony size of 5 ( Figure 3—figure supplement 5 ) , which we used in the killing term for both A and E: ( 5 ) N=N⁢ ( 11+ ( N5⁢N0 ) 2 ) +π⁢N0⁢ ( NN0-1 ) ⁢ ( ( N5⁢N0 ) 21+ ( N5⁢N0 ) 2 ) A . baylyi natural competence is largely proportional to its growth rate ( Palmen et al . , 1993 ) , so we assumed the DNA uptake slows with the same saturation factor as cell growth . Over a longer time scale , communities on agar plates will continue to grow slowly as fresh nutrients diffuse in and to replace dying cells , but this model focuses on the early dynamics of a growing community in the very early stages of biofilm development . Transformation of Acinetobacter by homologous recombination ( HR ) is more efficient than via replicating plasmids ( Palmen et al . , 1993 ) , so it allows measurement of HGT with higher sensitivity . Therefore , we measured HGT parameters using pRC03H-2S , which cannot replicate in Acinetobacter but contains homology to the Acinetobacter genome , allowing Acinetobacter to integrate the kan gene via HR . All parameter fitting ( Table 1 ) was performed using the least squares nonlinear optimization function lsqnonlin in MATLAB . We fit each component of the model sequentially using simplified experiments . To determine the growth parameters , we measured growth curves for monoculture cell spots on agar and fit them using the logistic growth equation N⁢ ( t ) =KN⁢N0⁢eγN⁢tKN+N0⁢ ( eγN⁢t-1 ) , where N is the number of cells , γN is the growth rate , and KN is the saturation level for 2 ul spots ( Figure 3—figure supplements 1 and 2 ) . Using these growth parameters , we then fit the killing rate by measuring CFUs in co-cultured communities over time and using that data to fit the model as integrated using ode23 in MATLAB ( Figure 3d–j ) . For simultaneous fitting of data from multiple experiments ( Figure 3 , Figure 3—figure supplement 3 ) , we defined custom objective functions to calculate the prediction error for all data points , and then minimized the total sum of squares with lsqnonlin . For DNA uptake rate c , we used the previously measured 60 bp/s ( Palmen et al . , 1993 ) . To measure the other key HGT parameters KD⁢N⁢A and transformation efficiency ϵ , we seeded monoculture Acinetobacter spots on agar with known numbers of plasmids , and measured HGT over time . We simultaneously fit the HGT curves for time course and serial dilution data to the model as integrated using ode23 ( Figure 3a–c ) . We measured plasmids per E . coli cell by growing a shaking tube of cells to exponential phase , counting cell density by spotting serial dilutions , isolating plasmid DNA with a standard miniprep kit , measuring obtained DNA density with a spectrophotometer , and converting ng of DNA to plasmids per cell . While plasmid copy number may vary as conditions change ( e . g . growth phase and liquid vs . solid media ) , most of our results present log-scale effects , so this should not affect the qualitative conclusions . For genomic DNA per cell G , we divided the E . coli genome ( 3 . 6 Mb ) by the plasmid length ( 9093 bp ) to obtain the number of plasmid equivalents of genomic DNA . Full simulations of contact-dependent killing and HGT in microbial communities were performed by integrating the population dynamic model with the MATLAB function ode23 , using parameters measured as described above for pRC03H-2S . For DNA , we defined D to be the number of plasmids , so we converted the units of c and KD using the conversion factor 9093 bp/plasmid . In Videos 1–6 and 9–12 , GFP expression in E . coli is poorly repressed early in the movies , and mCherry expression in Acinetobacter becomes bleached later in the movies . We believe both of these phenomena are likely due to nutrient restriction , caused by Acinetobacter adhesion , growth , and eventual clogging in the channels meant to supply nutrients ( see the top and bottom strips in the Supplemental Movies ) . Indeed , in our experience , the stickiness of Acinetobacter makes microfluidic experiments much more challenging than with only lab E . coli strains . LacI repression in E . coli is less complete when nutrients are limiting ( Grossman et al . , 1998 ) . The E . coli in Videos 1–6 are likely nutrient-deprived because Acinetobacter are coating the channels , both consuming nutrients and reducing flow rate . In Videos 7–8 , on the other hand , the media had been switched to include kanamycin , which inhibited all Acinetobacter that had not acquired resistance via HGT from E . coli . This is likely why the GFP repression in E . coli is less complete in Videos 1–6 than in Videos 7–8 . Conversely , Acinetobacter appear to reduce expression of mCherry as they become more nutrient-limited . We speculate that mCherry fluorescence fades while GFP remains visible in Acinetobacter because ( i ) the mCherry gene is single-copy on the genome while the GFP plasmid is multi-copy , ( ii ) mCherry bleaches more readily and has lower intrinsic brightness than GFP ( Shaner et al . , 2005 ) , and ( iii ) there may be promoter , ribosome-binding site , and codon usage effects affecting expression in A . baylyi , which is not nearly as well-characterized as E . coli in these respects . Regardless , in the Supplemental Videos 1–6 , it is clear when and where all E . coli are killed , because the GFP disappears suddenly within a single time-frame , indicating lysis as opposed to gradual bleaching . For GFP to re-emerge in an area where all E . coli have been killed ( white circles in Videos 1–6 ) , it must be expressed by the remaining Acinetobacter .	Every year , antibiotics save millions of lives , but this may not last forever . The bacteria that cause infections are getting smarter and continuously evolve genes to become resistant to antibiotics , which makes it harder to kill them . In many cases , using stronger drugs can bypass this problem , but some 'super-bugs' are developing resistance to every drug we have . For example , the bacterium Acinetobacter baumannii has recently been classified as a global threat that kills thousands of people every year , which placed it on a top six 'most wanted' list for multi drug-resistant bacteria . Worryingly , this drug resistance seems to develop faster than 'standard' evolution would allow , making it difficult to keep up with developing new effective drugs . One way bacteria can shortcut the evolution of resistance is through a process called horizontal gene transfer , in which they collect resistance genes from other bacteria . Some bacteria can speed up this gene transfer by actively killing their neighbors to extract their DNA . However , until now , this process has not been observed directly , and it was not fully understood where and when killing neighbors becomes important for gene transfer . Now , Cooper , Tsimring and Hasty have studied a relative of A . baumannii called A . baylyi . Together with another type of bacteria that contained green fluorescence genes , A . baylyi was placed onto a surface that allowed both species to grow . As the two types of bacteria grew together , A . baylyi started to kill the other one and stole their genes . This happened so often that some started to become fluorescent , which could be observed in real time under a microscope . A . baylyi also stole genes for antibiotic resistance , and when an antibiotic was added , the bacteria with the stolen resistance genes kept growing and dividing , while the others were killed . Cooper et al . then developed a mathematical model to quantify and simulate this killing-enhanced horizontal gene transfer . The results showed that killing other bacteria made gene transfer more effective when the number of A . baylyi was high and the number of 'victims' was low – and also when they were together for a shorter period . This work may help to explain how Acinetobacter and similar bacteria develop drug resistance so quickly . A next step will be to measure and compare gene transfer parameters in different types of bacteria . A better understanding of how , where , and when gene transfer happens , may in the future help to guide strategies to fight resistance .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "computational", "and", "systems", "biology", "microbiology", "and", "infectious", "disease" ]	2017	Inter-species population dynamics enhance microbial horizontal gene transfer and spread of antibiotic resistance
Dystrophin forms an essential link between sarcolemma and cytoskeleton , perturbation of which causes muscular dystrophy . We analysed Dystrophin binding dynamics in vivo for the first time . Within maturing fibres of host zebrafish embryos , our analysis reveals a pool of diffusible Dystrophin and complexes bound at the fibre membrane . Combining modelling , an improved FRAP methodology and direct semi-quantitative analysis of bleaching suggests the existence of two membrane-bound Dystrophin populations with widely differing bound lifetimes: a stable , tightly bound pool , and a dynamic bound pool with high turnover rate that exchanges with the cytoplasmic pool . The three populations were found consistently in human and zebrafish Dystrophins overexpressed in wild-type or dmdta222a/ta222a zebrafish embryos , which lack Dystrophin , and in Gt ( dmd-Citrine ) ct90a that express endogenously-driven tagged zebrafish Dystrophin . These results lead to a new model for Dystrophin membrane association in developing muscle , and highlight our methodology as a valuable strategy for in vivo analysis of complex protein dynamics . Muscle Dystrophin establishes a link between Dystroglycan complexes at the cell membrane and actin in the cortical cytoskeleton ( Ibraghimov-Beskrovnaya et al . , 1992; Levine et al . , 1992; Ervasti and Campbell , 1993; Rybakova et al . , 1996 , 2000 ) . Mutations in the Dystrophin gene often lead to a non-functional protein and Duchenne muscular dystrophy ( DMD ) , characterised by severe muscle degeneration from early childhood . In-frame deletions within the Dystrophin sequence can result in a shortened but partially functional protein that causes Becker muscular dystrophy ( BMD ) ( Koenig et al . , 1989 ) . A major international effort aims to develop gene therapy for DMD . Yet , there are still big gaps on our understanding of how Dystrophin works within cells . It is important to understand the dynamics of Dystrophin in vivo and how this could vary within cellular context , influencing the phenotype of BMD and gene therapy planning for patients with DMD . For example , many current approaches for gene therapy in DMD aim to restore ‘short’ Dystrophins , known to be partially functional from studies of patients with BMD and murine transgenic models ( Konieczny et al . , 2013 ) . How the dynamics of these proteins compare with those of full-length Dystrophin has not been addressed due to the lack of a suitable method . However , if some short Dystrophin forms bind more efficiently and stably than others this will have an impact on the relative amount of protein necessary to recover function . The knowledge of Dystrophin dynamics and a methodology to perform comparative studies is therefore needed . Dystrophin is well studied in zebrafish and its homology with the human Dystrophin is well documented ( Guyon et al , 2003; Jin et al . , 2007; Berger et al . , 2011; Lai et al . , 2012 ) . Several mutant and transgenic lines have been used as model for Duchenne muscular dystrophy and testing potential therapeutic targets ( Kunkel et al . , 2006; Johnson et al . , 2013; Kawahara and Kunkel , 2013; Waugh et al . , 2014; Wood and Currie , 2014 ) . The loss of Dystrophin is lethal to both people and zebrafish , primarily due to striated muscle defects ( Bassett et al . , 2003; Berger et al . , 2010 ) . Both species show developmental progression towards the adult localisation of Dystrophin . In human embryos , Dystrophin first appears in the cytoplasm , at the tips of myotubes , then becomes widespread throughout the myofibres in foetal stages ( Wessels et al . , 1991; Clerk et al . , 1992; Chevron et al . , 1994; Mora et al . , 1996; Torelli et al . , 1999 ) . In embryonic zebrafish muscle , Dystrophin transcripts are reported to accumulate initially in the cytoplasm , and from 24 hr post fertilization ( hpf ) until early larval stages , Dystrophin protein and transcripts are primarily located at muscle fibre tips ( Bassett et al . , 2003; Guyon et al . , 2003; Jin et al . , 2007; Böhm et al . , 2008; Ruf-Zamojski et al . , 2015 ) . In both species , Dystrophin becomes localised under the sarcolemma in maturing and adult muscle fibres where it concentrates at costameres , neuromuscular and myotendinous junctions ( Samitt and Bonilla , 1990; Miyatake et al . , 1991; Chambers et al . , 2001; Guyon et al . , 2003 ) . Dystrophin half-life is believed to be very long ( Tennyson et al . , 1996; Verhaart et al . , 2014 ) . Therefore , to study Dystrophin binding dynamics , it may be advantageous to look at the moment where binding complexes are actively forming , during muscle development . Study of protein dynamics in living tissue faces many technical hurdles that no available method can tackle satisfactorily . Fluorescence correlation spectroscopy ( FCS ) requires stable confocal imaging of submicron volumes and is thus sensitive to drift in living tissue . Moreover , FCS is only applicable over a limited range of fluorophore concentrations and is greatly impeded by the presence of significant quantities of immobile fluorophores . Fluorescence recovery after photobleaching ( FRAP ) avoids these problems . However , imaging in a living organism is challenging due to low signal-to-noise ratio that worsens as tissue thickness increases and protein abundance decreases . In addition , cells are located at variable optical depths and have varying shapes and protein levels , all of which introduces variability . This hampers identification of real variation in protein dynamics and prevents the common procedure of pooling data from multiple cells to reduce noise . In this study , we assess human Dystrophin dynamics in muscle cells of host zebrafish embryos , using a new approach to perform and analyse FRAP in the context of the living muscle fibre that specifically deals with the challenges of in vivo protein analysis . We thoroughly characterize the expression of the exogenous human Dystrophin within zebrafish host muscle cells . Overexpression often results in an excess of cytoplasmic Dystrophin , which is taken into account on the analysis of Dystrophin binding dynamics . We demonstrate that Dystrophin diffuses freely in the zebrafish muscle fibre cytoplasm and determine its diffusion constant . At the binding sites localised at the muscle cell tips , we found the existence of two membrane-bound pools with distinct binding constants: an immobile pool bound stably during our imaging timescale and a mobile-bound pool with a highly dynamic turnover . We test several potential factors that could potentially interfere with the binding dynamics of Dystrophin , or with its analysis , and result in wrong identification of a labile-bound pool: lateral diffusion of bound Dystrophin , transient dark state of fluorescent proteins , artificial increase of the cytoplasmic pool , competition with endogenous zebrafish Dystrophin , or weak interaction between inter-species proteins . Our data allowed us to dismiss all these hypotheses , supporting the real existence of two bound forms of Dystrophin in maturing fibres of the zebrafish embryo . Taken together , these results suggest a model for Dystrophin association with the membrane and provide a baseline and a validated methodology to analyse how modifications in Dystrophin structure may alter its dynamics . We set out to analyse human Dystrophin protein dynamics in vivo in the physiological environment of the muscle fibres of zebrafish embryos ( Figure 1 ) . We engineered expression constructs based on the full-length 427 kd human cDNA sequence ( huDys; Figure 1A; ‘Materials and methods’ ) . Expression of huDys or GFP control in zebrafish embryos was achieved through the injection of the DNA constructs into newly fertilized embryos at the early 1 cell stage , aiming to obtain mosaic expression to facilitate single cell analysis ( Figure 1B , C ) . From 24 hpf onwards , huDys ( Figure 1C , green ) accumulated progressively at both ends of transgenic fibres ( hereafter referred to as ‘tips’ ) , as observed for endogenous zebrafish Dystrophin ( Figure 1C , red ) . GFP control showed no tip accumulation ( Figure 1D ) . In addition , huDys was often detected accumulating at putative neuromuscular junctions ( NMJ ) , like endogenous Dystrophin ( arrows in Figure 1E , F ) . We conclude that human Dystrophin localises in zebrafish skeletal muscle like zebrafish Dystrophin , making it likely , in a first approach , that the zebrafish embryo could be a suitable host to study human Dystrophin in vivo . 10 . 7554/eLife . 06541 . 003Figure 1 . Human Dystrophin expression in the zebrafish embryo . ( A ) Main features of the human Dystrophin expression constructs engineered for this study . ( B ) Schematic illustrating 2 dpf zebrafish embryo . Slow muscle fibres within the chevron-shaped somite , one magnified and highlighted in blue , are typically aligned anterior-posteriorly with their tips ( dark blue ) attaching at vertical somite borders . ( C ) Immunofluorescent detection of exogenous huDys ( green , arrows ) at fibre tips , co-localizing with endogenous zebrafish Dystrophin ( zfDys , red ) that accumulates at the tips of every muscle fibre , marking the somite border . ( D ) In vivo expression of control GFP shows accumulation in muscle fibre cytoplasm without enrichment at the fibre tips . ( E ) Immunodetection with antibody specifically recognizing human Dystrophin on whole mount 2 dpf embryo shows punctate accumulation of exogenous huDys ( arrow ) suggestive of localization at the NMJ , in addition to fibre tips ( arrowheads ) . ( F ) Immunodetection on longitudinal cryostat sections of 2 dpf somitic muscle shows enrichment of endogenous zebrafish Dystrophin ( zfDys ) at NMJ ( arrows ) . Note concentration of most zfDys at fibre tips ( arrowheads ) . ( G ) Maximum intensity projection of a confocal stack showing accumulation of huDysGFP in a muscle fibre in vivo . Strong enrichment is noticeable at the tips ( arrows ) , membrane protrusions ( yellow arrowheads ) , and NMJ ( red arrowheads ) . ( H ) Double immunofluorescent detection of GFP in a huDysGFP-expressing embryo ( huDysGFP , green ) and α-bungarotoxin ( BTX , red ) confirms co-localization at the NMJ ( insert ) . ( I , J ) huDysGFP mRNA detected by in situ hybridization ( arrows in I , Nomarski ) localises at fibre tips like GFP fluorescence detected while in vivo ( arrows in J; confocal maximum projection ) . Scale bars = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 003 To allow the in vivo study of huDys dynamics , the expression construct was modified to produce huDys tagged with GFP at its C-terminus ( huDysGFP; Figure 1A; ‘Materials and methods’ ) . This produces a bright fluorescent signal easily detectable at fibre tips ( Figure 1G , arrows ) . Occasionally , some cells showed accumulations at membrane protrusions ( Figure 1G , yellow arrowheads ) and NMJs ( Figure 1G , red arrowhead ) . The latter was confirmed by double staining with α-bungarotoxin ( Figure 1H , inset ) . Compared to GFP alone , huDysGFP was generally less bright ( Figure 2A ) but was , nevertheless , more readily detected in muscle than non-muscle tissue ( Figure 2B ) , suggesting that binding and stabilization at the membrane differ between tissues . 10 . 7554/eLife . 06541 . 004Figure 2 . Comparison of huDysGFP and GFP expression in 2 dpf zebrafish embryos . ( A ) Total cellular GFP signal ( sum of pixel values ) of sum projections made from confocal optical sections of individual muscle fibres expressing GFP or huDysGFP in vivo . NGFP = 10 fibres , NhuDysGFP = 32 fibres; p < 0 . 0001 . ( B ) Fraction of muscle fibres among positive cells in embryos expressing huDysGFP or GFP in vivo . NGFP = 1593 cells in 27 embryos , NhuDysGFP = 472 cells in 28 embryos; p = 0 . 0032 . Error bars show S . E . M . DOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 004 To determine whether human Dystrophin mRNA becomes localised in zebrafish muscle like the endogenous transcripts , in situ mRNA hybridization with a human Dystrophin-specific probe was performed on injected embryos . In most cases , localisation of human Dystrophin mRNA was observed at fibre tips ( Figure 1I , J ) . Thus , GFP-tagged Dystrophin localises similarly to its untagged counterpart , and to the endogenous Dystrophin mRNA and protein , and it is suitable for in vivo imaging . Both endogenous Dystrophin and huDysGFP accumulate at the fibre tips , yet the endogenous form is not readily detected in cytoplasm in immunofluorescence assays , that is , it is not clear that the fluorescence detected is higher than background ( Figure 1F ) , in contrast most fibres expressing huDysGFP show weak but detectable fluorescence in the cytoplasm ( Figure 1G , H ) . We investigated this difference . As intensity around 3 units above the background is easily detected under our imaging conditions ( ‘Materials and methods’ ) , we can distinguish huDysGFP in a cytoplasmic voxel ( a three dimensional pixel of 0 . 024 µm3 ) down to a number per voxel around 60 times lower than in the brightest fibre tip voxel ( avoiding saturation of the detector by setting it to under 255 on 8-bit grayscale ) . As less than 1% of the entire cell volume is in the tip region , it is possible that even in cells with cytoplasmic huDysGFP below detectable levels there could be as much huDysGFP in the cytoplasm as in the tip region . This could equally be the case for endogenous Dystrophin . Therefore , the observed difference may be partially due to lower sensitivity of the antibody detection of cytoplasmic endogenous Dystrophin compared to the higher sensitivity of GFP detection . However , higher levels of cytoplasmic accumulation are likely an artefact of the overexpression of exogenous Dystrophin . Therefore , to confidently analyse Dystrophin binding dynamics , the presence of this cytoplasmic pool has to be taken into account and a deeper characterisation is required . We analysed in more detail how each pool , tips , and cytoplasm , distribute . As predicted , in the majority of the muscle fibres , most huDysGFP is in the cytoplasm , with only a minority at the tips ( Figure 3A ) , even though the higher concentration at the tips might have suggested otherwise ( Figure 1G , H ) . Even in cells with cytoplasmic levels close to the detection limit , there is at least as much huDysGFP dispersed in the whole cytoplasm as that concentrated at the tips ( Figure 3A ) . Across a population of fibres , more huDysGFP fluorescence was detected in the cytoplasm of fibres with higher total huDysGFP levels ( blue triangles in Figure 3A ) . In contrast , the fluorescence at the tips does not increase with the total fluorescence of the fibre ( green circles in Figure 3A ) , indicating that tip binding is limited by the presence of a limited number of binding sites that easily saturate . Thus , the accumulation of huDysGFP in the fibre cytoplasm does not appear to affect the binding at the tips . Also , fibre tips generally had greater fluorescence intensity than fibre cytoplasm ( Figure 3B , C ) . High intensities at the tips can be achieved even with low cytoplasmic huDysGFP concentrations ( Figure 3B ) . Moreover , at low overall fibre intensities , there is clear preference for accumulation at the tips ( Figure 3C ) . All these data indicate that human Dystrophin is preferentially bound at the zebrafish fibre tips regardless of the amount of cytoplasmic Dystrophin . 10 . 7554/eLife . 06541 . 005Figure 3 . Comparison of tip and cytoplasm huDysGFP . ( A ) Variation of the total fluorescence ( as raw integrated density or sum of pixel values ) at tips ( green circles ) and cytoplasm ( blue triangles ) over a population of 32 fibres expressing huDysGFP . ( B ) Mean voxel intensity at tips versus cytoplasm in sum projection of confocal z-stacks . The mean voxel intensity is calculated as the integrated density per pixel in the sum projections , therefore taking into account the fibre and tip size . ( C ) Ratio tip/cytoplasm voxel intensities shows an inverse correlation with the total fibre voxel intensity in sum projections of confocal z-stacks . p value = 0 . 0004 , R2 = 0 . 3443 , n = 32 . DOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 005 To study Dystrophin dynamics in our system , we still have to take into account the presence of a cytoplasmic pool . Dystrophin is a high molecular mass protein with multiple actin binding sites . We asked whether huDysGFP can diffuse freely in muscle fibre cytoplasm or whether it may be bound to cytoplasmic structures such as actin fibres . We developed a modified FRAP approach to analyse protein dynamics in vivo . Even in the best imaging conditions , one faces the challenge of low signal-to-noise ratio that worsens as tissue thickness increases and protein abundance decreases . Although we are able to detect cytoplasmic huDysGFP , the signal is weak and the signal-to-noise ratio in single pixels or even small volumes is low ( Figure 4A ) . To address these issues , we increase the signal by increasing laser power and studying large areas ( ‘Materials and methods’ ) . Under these conditions , we acquire a consistent signal both for GFP and huDysGFP . However , using a high laser power to image over large areas results in significant photobleaching during imaging , and , even at 100% laser power , bleaching is so slow that significant diffusion occurs during bleaching ( Weiss , 2004 ) . We compensate for both using a mathematical model applied to the FRAP experimental data ( ‘Materials and methods’ ) . This was integrated in a user-friendly application written to allow easy data analysis of multiple experiments ( see ‘Materials and methods’ ) . 10 . 7554/eLife . 06541 . 006Figure 4 . Bleaching areas: size optimization , orientation , and definition of Cartesian coordinates . ( A ) Noise reduction in GFP FRAP curve for increasing areas: 1 ( blue ) , 25 ( red ) and 256 ( black ) pixels . ( B ) To determine D along the X-axis ( arrowheads ) of individual muscle cells in the embryo , two large regions of different widths ( narrow and wide; see Table 1 ) are bleached sequentially and separated by >1 min to ensure full recovery . ( C ) XYZ view from Volocity of a typical muscle fibre expressing huDysGFP in vivo . The cut planes shown correspond to panels D–F . ( D ) Muscle fibre imaged in the XY plane from a lateral position of the zebrafish embryo as embedded for FRAP . When referring to the fibre tips , we use a different set of axes: the long axis of the tip is the XT axis and the shorter is the YT axis . ( E , F ) YZ and XZ sections , respectively , through the muscle fibre , showing lower Z resolution . Scale bars = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 006 To validate our experimental conditions and FRAP analysis method in the muscle cells of zebrafish embryos , we first analysed GFP diffusion within the cytoplasm . We studied diffusion along the long axis of the muscle cell ( Figure 4B ) . The selected fibre is oriented such that the image X-axis aligns to the fibre long axis ( i . e . , roughly anterior-posterior in the animal ) and the Y-axis is dorso-ventral ( Figure 4C–F ) . In each cell , one or two large rectangles of different sizes ( narrow and wide ) were bleached , avoiding nuclei ( Figure 4B ) . The bleached rectangles are shorter along the X-axis than in Y and cross the entire cell transversely . This makes recovery almost entirely due to mobility along the X-axis , simplifying modelling and fitting . We analyse the profile along the X-axis taken immediately after bleaching GFP-expressing fibres ( Figure 5A ) . This profile is not a top-hat but a Gaussian , much wider than the region actually bleached , consistent with diffusion during the bleaching phase . This diffusion is taken into account in the modelling as neglecting it would lead to significant error ( Castle et al . , 2011; González-Pérez et al . , 2011; Müller et al . , 2012 ) ( ‘Materials and methods’ ) . 10 . 7554/eLife . 06541 . 007Figure 5 . Analysis of cytoplasmic diffusion . ( A–D ) FRAP experimental data and fitting curves . Normalized intensity profile along the X-axis of GFP ( A ) or huDysGFP ( C ) at the first time point after bleaching ( blue crosses ) and Gaussian fits ( red curves ) . Recovery curves for GFP ( B ) and huDysGFP ( D ) along X-axis in which the cyan crosses show normalized fluorescence intensity in the bleached region . Curves are fits of the diffusion model to ∼7 s ( red ) or ∼40 s ( two-parameters , solid black; one-parameter , dashed ) post-bleach . ( E ) DGFP and DhuDysGFP obtained from two-parameter fits to ∼40 s of FRAP experimental data ( see Table 1 ) . Graph shows median , quartile , range , and n . Comparison was by two tailed t-test after test for normality . ( F ) Scatter plot of D values obtained for pairs of bleaching experiments performed in the same cell . For huDysGFP ( triangles ) , the two D values measured in the same cell , in areas of different widths ( narrow/wide ) , show a good correlation ( triangles; Pearson R = 0 . 98 ) . The triangles are not far from falling on a straight line of slope one . The small cell-to-cell variation in huDysGFP , relative to the variation between cells suggests that the mobility of huDysGFP genuinely varies from one fibre to another . For GFP , there is no definite trend visible , just scatter , presumably due to lower signal-to-noise with the more rapid diffusion of GFP . It is , therefore , not clear whether the mobility of GFP varies significantly from one fibre to another . DOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 007 We fit the recovery for each individual case to obtain the diffusion constant ( D ) and a bleaching-due-to-imaging parameter ( β ) ( Figure 5B ) . We tested modelling of the recovery at short and long times post-bleaching . When analysing long ( ≥10 s ) timescales , it is essential to take β into account , while for short times bleaching due to imaging is insignificant ( Figure 5B , Table 1 ) . The Gaussian profile and the ability to fit a simple diffusion model to the recovery are strong evidence for GFP diffusion . DGFP best-fit values ( fitted to data over ∼40 s ) averaged 13 . 2 μm2 s−1 , ranging from 8 . 6 to 20 . 8 μm2 s−1 ( Table 1 ) . This matches the range 7 . 6–15 . 8 μm2 s−1 previously reported for muscle cells ( Arrio-Dupont et al . , 2000; Kinsey et al . , 2011 ) , which indicates that our approach for in vivo FRAP analysis is able to achieve similar results to those previously reported for isolated cells in culture . 10 . 7554/eLife . 06541 . 008Table 1 . Diffusion constants , D , for GFP and Dystrophin in the cytoplasm obtained from fitting to FRAP experimental dataDOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 008Data set number ( cell number ) Bleach width ( pixels ) Final fitted time pointFibre length ( µm ) Bleach position ( µm ) D ( µm/s2 ) βGFP , standard fibre length set 1 ( cell 1 ) 1050090 . 045 . 09 . 30 . 001753 set 2 ( cell 1 ) 2050090 . 045 . 08 . 20 . 001572 set 3 ( cell 2 ) 1050090 . 045 . 09 . 60 . 001502 set 4 ( cell 2 ) 2050090 . 045 . 09 . 60 . 001483 set 5 ( cell 3 ) 1050090 . 045 . 06 . 30 . 002618 set 6 ( cell 3 ) 2050090 . 045 . 07 . 30 . 002503 set 7 ( cell 4 ) 1050090 . 045 . 010 . 20 . 001980 set 8 ( cell 4 ) 2050090 . 045 . 012 . 80 . 001181 set 9 ( cell 5 ) 1050090 . 045 . 012 . 40 . 002358 set 10 ( cell 5 ) 2050090 . 045 . 010 . 20 . 000961 set 1 ( cell 1 ) 1020090 . 045 . 011 . 00 . 002259 set 2 ( cell 1 ) 2020090 . 045 . 011 . 10 . 002861 set 3 ( cell 2 ) 1020090 . 045 . 010 . 60 . 001721 set 4 ( cell 2 ) 2020090 . 045 . 09 . 50 . 001418 set 5 ( cell 3 ) 1020090 . 045 . 08 . 60 . 003959 set 6 ( cell 3 ) 2020090 . 045 . 011 . 50 . 004743 set 7 ( cell 4 ) 1020090 . 045 . 09 . 80 . 001765 set 8 ( cell 4 ) 2020090 . 045 . 015 . 00 . 002000 set 9 ( cell 5 ) 1020090 . 045 . 014 . 40 . 002998 set 10 ( cell 5 ) 2020090 . 045 . 011 . 90 . 001895 set 11 ( cell 6 ) 1020090 . 045 . 016 . 00 . 001196 set 12 ( cell 6 ) 2020090 . 045 . 018 . 40 . 002364 set 1 ( cell 1 ) 105090 . 045 . 010 . 10 . 000000 set 2 ( cell 1 ) 205090 . 045 . 09 . 80 . 000000 set 3 ( cell 2 ) 105090 . 045 . 09 . 40 . 000000 set 4 ( cell 2 ) 205090 . 045 . 08 . 30 . 000000 set 5 ( cell 3 ) 105090 . 045 . 07 . 60 . 000000 set 6 ( cell 3 ) 205090 . 045 . 09 . 90 . 000000 set 7 ( cell 4 ) 105090 . 045 . 09 . 00 . 000000 set 8 ( cell 4 ) 205090 . 045 . 013 . 80 . 000000 set 9 ( cell 5 ) 105090 . 045 . 011 . 50 . 000000 set 10 ( cell 5 ) 205090 . 045 . 011 . 70 . 000000 set 11 ( cell 6 ) 105090 . 045 . 015 . 20 . 000000 set 12 ( cell 6 ) 205090 . 045 . 014 . 50 . 000000GFP , comparing measured and standard fibre length set 13 ( cell 7 ) 820082 . 041 . 013 . 30 . 001050 set 14 ( cell 7 ) 3220082 . 041 . 08 . 90 . 001208 set 15 ( cell 8 ) 820079 . 040 . 018 . 50 . 001700 set 16 ( cell 8 ) 3220079 . 040 . 010 . 80 . 002270 set 17 ( cell 9 ) 820083 . 030 . 019 . 80 . 000616 set 18 ( cell 10 ) 8200102 . 044 . 017 . 10 . 001177 set 13 ( cell 7 ) 820090 . 045 . 013 . 50 . 001084 set 14 ( cell 7 ) 3220090 . 045 . 09 . 00 . 001231 set 15 ( cell 8 ) 820090 . 045 . 019 . 30 . 001828 set 16 ( cell 8 ) 3220090 . 045 . 011 . 10 . 002377 set 17 ( cell 9 ) 820090 . 045 . 020 . 80 . 000828 set 18 ( cell 10 ) 820090 . 045 . 016 . 90 . 001144 set 13 ( cell 7 ) 85090 . 045 . 010 . 90 . 000000 set 14 ( cell 7 ) 325090 . 045 . 08 . 00 . 000000 set 15 ( cell 8 ) 85090 . 045 . 012 . 90 . 000000 set 16 ( cell 8 ) 325090 . 045 . 09 . 00 . 000000 set 17 ( cell 9 ) 85090 . 045 . 018 . 50 . 000000 set 18 ( cell 10 ) 85090 . 045 . 015 . 20 . 000000huDysGFP , standard fibre length set 19 ( cell 11 ) 820090 . 045 . 010 . 10 . 005649 set 20 ( cell 11 ) 1620090 . 045 . 06 . 50 . 005012 set 21 ( cell 12 ) 820090 . 045 . 07 . 20 . 002094 set 22 ( cell 12 ) 1620090 . 045 . 05 . 20 . 001708 set 23 ( cell 13 ) 820090 . 045 . 01 . 70 . 003047 set 24 ( cell 13 ) 1620090 . 045 . 01 . 80 . 001818 set 25 ( cell 14 ) 820090 . 045 . 09 . 40 . 008203 set 26 ( cell 15 ) 820090 . 045 . 06 . 00 . 004196 set 27 ( cell 16 ) 420090 . 045 . 01 . 40 . 001003 set 28 ( cell 16 ) 1020090 . 045 . 01 . 90 . 000966 set 19 ( cell 11 ) 85090 . 045 . 08 . 70 . 000000 set 20 ( cell 11 ) 165090 . 045 . 05 . 20 . 000000 set 21 ( cell 12 ) 85090 . 045 . 06 . 40 . 000000 set 22 ( cell 12 ) 165090 . 045 . 04 . 90 . 000000 set 23 ( cell 13 ) 85090 . 045 . 01 . 50 . 000000 set 24 ( cell 13 ) 165090 . 045 . 01 . 70 . 000000 set 25 ( cell 14 ) 85090 . 045 . 07 . 70 . 000000 set 26 ( cell 15 ) 85090 . 045 . 05 . 10 . 000000 set 27 ( cell 16 ) 45090 . 045 . 01 . 30 . 000000 set 28 ( cell 16 ) 105090 . 045 . 01 . 80 . 000000huDysGFP , comparing measured and standard fibre length set 29 ( cell 17 ) 4200130 . 062 . 01 . 70 . 001319 set 30 ( cell 17 ) 10200130 . 062 . 01 . 60 . 001140 set 31 ( cell 18 ) 4200124 . 049 . 04 . 00 . 000757 set 32 ( cell 18 ) 10200124 . 049 . 03 . 90 . 001221 set 33 ( cell 19 ) 4200106 . 051 . 05 . 30 . 001332 set 34 ( cell 19 ) 10200106 . 051 . 03 . 60 . 001408 set 35 ( cell 20 ) 10200112 . 031 . 03 . 40 . 004444 set 29 ( cell 17 ) 420090 . 045 . 01 . 70 . 001319 set 30 ( cell 17 ) 1020090 . 045 . 01 . 60 . 001140 set 31 ( cell 18 ) 420090 . 045 . 04 . 00 . 000757 set 32 ( cell 18 ) 1020090 . 045 . 03 . 90 . 001221 set 33 ( cell 19 ) 420090 . 045 . 05 . 30 . 001332 set 34 ( cell 19 ) 1020090 . 045 . 03 . 60 . 001408 set 35 ( cell 20 ) 1020090 . 045 . 03 . 40 . 004447 set 29 ( cell 17 ) 45090 . 045 . 01 . 40 . 000000 set 30 ( cell 17 ) 105090 . 045 . 01 . 60 . 000000 set 31 ( cell 18 ) 45090 . 045 . 03 . 60 . 000000 set 32 ( cell 18 ) 105090 . 045 . 03 . 50 . 000000 set 33 ( cell 19 ) 45090 . 045 . 04 . 00 . 000000 set 34 ( cell 19 ) 105090 . 045 . 03 . 50 . 000000 set 35 ( cell 20 ) 105090 . 045 . 03 . 10 . 000000Diffusion is measured along the X ( long ) axis of the fibre . For most cells , two different size regions were bleached per fibre . The width of the bleached region in pixels is indicated for each data set ( one pixel is 0 . 147 μm wide ) . Intentional bleaching was performed between time points 20 and 21 . Two-parameter , D and β , fits were performed to long acquisition times , either to time points 21 to 200 ( ∼40 s ) or time points 21 to 500 ( ∼110 s ) . One-parameter fits were also performed to only the initial recovery curve ( points 21 to 50 , or ∼7 s ) . For the latter fits , bleaching during imaging is too small to fit β , so we fix β = 0 . Results of fits to FRAP curves for GFP and huDysGFP are presented for a model using either the actual fibre length and bleach position or with the standard length of 90 μm and a bleach position at 45 μm . Note that in most GFP cases , the difference between the fitted D values for standard and actual lengths and bleach positions is less than 1 μm2/s . This is smaller than our estimate of the uncertainties in these D values , which is several μm2/s . For huDysGFP , there was no difference within two significant figures . Due to the smaller diffusion constants of huDysGFP , varying the cell length within these limits makes no difference to the values of D . During a ∼30-s experiment , the bleached profile is not affected by a fibre tip ∼45 μm away , for values of D typical of huDysGFP . Values of D in the main text are obtained using all D values obtained for two-parameter fits to data to point 200 , using model cells with the standard cell length and bleach position . We used the FRAP method described above to analyse cytoplasmic huDysGFP dynamics . The Dystrophin profile immediately after bleaching , similar to that of GFP , is not a top-hat but a Gaussian , wider than the region actually bleached , and consistent with diffusion during the bleaching phase ( Figure 5C ) . However , the Gaussian's amplitude is narrower and the depth greater for huDysGFP than GFP , consistent with slower diffusion ( compare Figure 5A , C ) . Indeed , DhuDysGFP best-fit values ( fitted to data over ∼40 s ) ranged from 1 . 4 to 10 . 1 μm2 s−1 , with a mean of 4 . 4 μm2 s−1 ( Figure 5D; Table 1 ) . Again , the Gaussian profile and our ability to fit a simple diffusion model to the recovery are strong evidence for huDysGFP diffusion , and rule out large scale ( above μm ) directed motion along the long axis of the cell , or significant binding of huDysGFP to immobile structures . huDysGFP has significantly lower D than GFP , reflecting the different protein sizes of 454 kD and 27 kD , respectively ( Figure 5E , p < 0 . 0001 ) . By comparing narrow and wide bleaches in the same cell across our population of muscle fibres , we found that DhuDysGFP appears to vary between cells , that is , there is real variability in huDysGFP cytoplasmic dynamics from one cell to another . The difference between DhuDysGFP values for a pair of experiments in the same cell is significantly smaller than between values in different cells ( Figure 5E , F; Table 1 ) . Pairs of D values obtained from the same cell are well correlated ( RhuDysGFP = 0 . 9813; Figure 5F ) . This indicates that Dystrophin's diffusion in the cytoplasm shows important variations from cell to cell but that it is consistent within one cell . Further , this finding highlights the limitations of reducing noise by ‘pooling’ FRAP results from more than one cell . Therefore , each FRAP curve was analysed independently throughout this study and pooling was deliberately avoided . The original data and relevant analysis files for each case are available online ( Bajanca et al . , 2015 ) to complement representative examples shown in figures and main data summarised in tables . We conclude that , although there is fibre-specific variation of D , there is no evidence that cytoplasmic Dystrophin either binds cytoskeletal elements or is actively transported towards fibre tips . The fibre tip region presumably contains huDysGFP bound to the Dystroglycan complex at the membrane ( Guyon et al . , 2003; Böhm et al . , 2008 ) and a portion diffusing in the adjacent cytoplasm . FRAP can test whether huDysGFP is able to bind stably at the tips and reveal binding dynamics , but this is a challenging task . Diffusion of cytoplasmic protein into the bleached area , combined with bleaching-due-to-imaging , masks the real binding dynamics . We tested diffusion-plus-binding models to analyse the tip FRAP curves but these require too many variables to be reliably fitted , reflecting the complexity of the binding dynamics , as will be shown . The relatively featureless , and noisy , recovery curves of in vivo FRAP within a complex tissue cannot adequately constrain this number of variables ( Sadegh Zadeh et al . , 2006 ) . Instead , to understand human Dystrophin dynamics at the tip , we use direct semi-quantitative analysis of bleaching and recovery . A FRAP protocol that bleached only part of the tip allowed comparison of bleached and unbleached regions of the tip ( Figure 6A ) . We combined this with analysis of the cytoplasm intensity near the tip in each cell to measure how much recovery to expect due to diffusion . In addition , recovery is followed at an initial fast acquisition rate to detect fast recovery , and then at a slower pace over longer time periods with reduced bleaching-due-to-imaging ( Figure 6A ) . 10 . 7554/eLife . 06541 . 009Figure 6 . Analysis of bound Dystrophin . ( A ) FRAP protocol for tips and schematic FRAP curves showing effect of bleaching-due-to-imaging at high and low imaging frequency . Colour-coding of regions analysed in huDysGFP-expressing cell tips , correspond to traces in A–E . ( B ) Scatter plot of fractional recovery in tip pixels as a function of the cytoplasmic intensity . ( C–E ) Examples of normalized fluorescence curves , for tip FRAP recovery of three examples ( indicated in B ) of different tip recovery levels ( C > E > D ) and cytoplasmic intensity ( E > C > D ) . Intensities in intentionally bleached region ( green circles ) are lower than in the unbleached tip region ( red squares ) , yielding a difference ( black crosses = red − green ) . Cytoplasm recovers almost fully ( blue triangles ) . See Table 2 ( C = tip7 , D = tip27 , E = tip32 ) . ( F ) Examples of unnormalized fluorescence curves for unbleached tip ( squares ) and cytoplasm ( triangles ) , in tips shown in C ( open symbols ) , D ( closed symbols ) , and E ( hatched symbols ) . Note that absolute intensity recovery in tip is larger than in the cytoplasm , but higher cytoplasmic intensity does not result in higher tip recovery ( C vs E ) . DOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 009 On most tips , we identified three signatures of a population of Dystrophin that is bound at the tip and immobile ( not turning over ) on the timescale of our experiment . First , there is only partial recovery in the bleached region of the tip regardless of the huDysGFP expression levels ( Figure 6B–F ) . Second , the normalized intensity difference between the unbleached and bleached regions in most tips ( black crosses in Figure 6C–E ) reaches a plateau above zero ( Table 2 ) . At the end of the experiment , both halves of the tip have received the same amount of light ( and thus bleaching ) for ∼250 s after the intentional bleaching phase , so the difference observed is created by the huDysGFP bleached earlier that did not turn over . Third , in the unbleached half of the tip , which receives the same bleaching-due-to-imaging as the cytoplasm , both the normalized and the absolute drops in intensity from start to end of the experiment are much larger than in the cytoplasm ( compare drops in the tips with those in the cytoplasms in Figure 6F ) . Thus , our FRAP data consistently confirm that a population of huDysGFP is effectively bound and immobile at zebrafish muscle fibre tips . 10 . 7554/eLife . 06541 . 010Table 2 . Analysis of FRAP data on the tips of huDysGFP-expressing cells in wild-type embryosDOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 010Tip number ( cell number ) Cytoplasm intensityFractional recoveryFinal normalized unbleached-bleached intensitiesUnbleached-cytoplasm 50% recovery1 ( cell 1 ) 5 . 800 . 240 . 14<10 s2 ( cell 1 ) 4 . 580 . 340 . 23<10 s3 ( cell 2 ) 21 . 580 . 160 . 25no recovery4 ( cell 2 ) 21 . 900 . 46−0 . 06<20 s5 ( cell 3 ) 8 . 680 . 280 . 22<10 s6 ( cell 3 ) 4 . 040 . 310 . 14<10 s7 ( cell 4 ) 5 . 940 . 590 . 06<10 s8 ( cell 4 ) 15 . 030 . 640 . 09<10 s9 ( cell 5 ) 0 . 800 . 080 . 31no recovery10 ( cell 5 ) 1 . 810 . 170 . 36<10 s11 ( cell 6 ) 0 . 470 . 11−0 . 03<10 s12 ( cell 7 ) −0 . 080 . 070 . 38<10 s13 ( cell 7 ) 0 . 020 . 080 . 17<10 s14 ( cell 8 ) 0 . 500 . 080 . 13<10 s15 ( cell 8 ) 2 . 020 . 09−0 . 03<10 s16 ( cell 9 ) 2 . 22−0 . 020 . 10<10 s17 ( cell 10 ) 0 . 730 . 260 . 24<30 s18 ( cell 11 ) 1 . 260 . 100 . 31<10 s19 ( cell 12 ) 2 . 310 . 350 . 07<30 s20 ( cell 12 ) 1 . 630 . 080 . 20<10 s21 ( cell 13 ) 3 . 180 . 410 . 06<10 s22 ( cell 14 ) 0 . 070 . 020 . 24<10 s23 ( cell 15 ) 0 . 930 . 150 . 15<10 s24 ( cell 16 ) 0 . 15−0 . 090 . 16<30 s25 ( cell 17 ) 11 . 290 . 340 . 26>30 s26 ( cell 17 ) 7 . 880 . 32−0 . 03<20 s27 ( cell 18 ) 1 . 06−0 . 010 . 26<10 s28 ( cell 19 ) 2 . 050 . 090 . 22<20 s29 ( cell 20 ) 0 . 990 . 170 . 15<10 s30 ( cell 21 ) 0 . 150 . 020 . 45<10 s31 ( cell 22 ) 1 . 520 . 120 . 28<20 s32 ( cell 22 ) 11 . 860 . 350 . 15<10 s33 ( cell 23 ) 3 . 260 . 280 . 12<10 sCytoplasm intensity is the background-subtracted intensity on an 8-bit scale . It is calculated in a rectangular region of a few hundred pixels inside the cell but away from the tip and is averaged over images 4 to 20 ( the last one before bleaching ) . Fractional recovery is the ratio RT/IT , where , RT is the intensity recovery in the tip , and IT is the average pre-bleach intensity in the bleached region . RT is the average intensity in the bleached region in the final time point ( 200 ) , minus that in the first point after bleaching ( 21 ) , and IT is averaged over images 4 to 20 . Final normalised unbleached minus bleached intensities is the difference between the average normalized intensities in the unbleached and bleached regions , derived from the average of the final 20 time points ( 181–200 ) . Unbleached minus cytoplasm 50% recovery evaluates if at least 50% of the final recovery of unbleached minus cytoplasm curves was rapidly attained , at the first ( <10 s ) , second ( <20 s ) , or later time points after switching from fast to slow acquisition rates . Note that both tips of some fibres were analysed . We analysed the dynamics of exchange of cytoplasmic freely diffusing huDysGFP with bound huDysGFP . Surprisingly , the FRAP curves revealed complex binding dynamics of Dystrophin at the tips that vary between fibres . We analysed the recovery pattern in the bleached region when switching to a slow acquisition regimen ( transition from step 3 to 4 in Figure 6A ) . Despite the very slow turnover that characterizes the immobile-bound pool at the tips , for many fibres there is an almost immediate partial recovery ( see Figure 6C–E ) . A shift of the dynamic equilibrium between dark-state and excitable GFP could result in apparent recovery following a switch to slow acquisition speed ( Mueller et al . , 2012 ) . However , we verified that the extent to which this phenomenon occurs in huDysGFP could only justify a very small fraction of recovery ( <1% , Figure 7A ) . We next evaluated the contribution of free diffusion of cytoplasmic huDysGFP into the bleached region , occurring as demonstrated with a D around 4 μm2 s−1 . As the bleached region contains bound and unbound huDysGFP and both are subjected to bleaching , it is not straightforward to analyse the recovery against a control cytoplasmic region , which is not subjected to intentional bleaching . Therefore , we compared unbleached tip and cytoplasm traces , which were submitted to comparable experimental conditions , without intentional bleaching but with strong bleaching-due-to-imaging during the fast acquisition phase ( Figure 6F ) . On switching to slow acquisition , any recovery on the unbleached tip would be expected to parallel the recovery on the cytoplasm . Nevertheless , the immediate recovery is in most cases larger than that expected from diffusion of cytoplasmic huDysGFP despite a steady state is reached soon after , characteristic of the large immobile pool ( Figure 6F , compare open or hatched squares [unbleached tip] with the respective open or hatched triangles [cytoplasm] ) . In most cases , at least 50% of the total recovery of unnormalised unbleached minus cytoplasm curves occur within 10 s ( Table 2 ) . This indicates that there is an unaccounted third pool of huDysGFP , in addition to the immobile-bound pool and the cytoplasmic-free diffusible pool: a rapidly turning-over dynamic bound huDysGFP pool . 10 . 7554/eLife . 06541 . 011Figure 7 . Effect of dark state and lateral mobility on huDysGFP intensity recovery . ( A ) To evaluate the recovery fraction due to dark state , huDysGFP was bleached in entire muscle fibres in vivo . Images shown were taken at t = 0 ( top panel ) and right after bleaching ( middle panel ) ; the red line defines the bleached region . Lower panel: plot of normalised fluorescence shows very low recovery after photobleaching ( 0 . 6% ) , presumably due to a shift from dark state to excitable huDysGFP . ( B ) FRAP tests for lateral mobility of bound huDysGFP . Top panel: initial image acquired from muscle fibre tip 7 showing the area to be bleached in green and the unbleached tip region in red . Middle and bottom panels: one-dimensional profiles along XT for an example of substantial recovery ( middle panel , tip7 ) and little recovery ( bottom panel , tip27 ) . Profiles are shown at three time points: before deliberate bleaching ( t = 0 . 93 s , black squares ) , just after bleaching ( t = 5 . 3 s , orange triangles ) , and at the end of the experiment ( t = 230 s , turquoise crosses ) . The intensity at each point is the average ( background corrected ) unnormalized intensity over the 20 pixel strip along the YT axis and averaged over 3 images , at the given time , plus the previous and next images . DOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 011 Recovery that is presumably due to the bound mobile pool is seen in fibres with no detectable or very low levels of cytoplasmic unbound huDysGFP ( Table 2 ) . However , the extent of the recovery is still dependent on the amount of cytoplasmic huDysGFP in a fibre ( Figure 6B , F ) . Note that in the scatter plot of fractional recovery against cytoplasmic huDysGFP intensity , fibre tips with negligible cytoplasmic huDysGFP show very low recovery ( Figure 6B ) . This suggests that cytoplasmic protein is required for the dynamic recovery . Interestingly , tips with high cytoplasmic huDysGFP intensity near the tip do not necessarily recover more ( compare Figure 6C , E , F ) indicating real tip-to-tip variation in the dynamic bound pool . The two tips of the same fibre behave similarly in some but not all cases ( Table 2 ) . Most of the recovery that is detected occurs within a few seconds in most cases ( Table 2 ) . Given that DhuDysGFP is a few μm2 s−1 , diffusion times across the tip region will be of order a second . Thus , recovery of the dynamic bound pool is fast enough that it may be limited by diffusion . If there is a characteristic binding lifetime , it is at most a few seconds . In summary , our data indicate that most fibres have two pools of huDysGFP bound at their tips , a tightly bound pool stable for at least minutes and another with a sub-second to a few seconds turnover time . To confirm that the mobility of bound Dystrophin can only be due to exchange with the free-diffusing pool , we tested for lateral mobility of bound huDysGFP , that is , we searched for evidence that huDysGFP can move along the membrane without unbinding and becoming part of the cytoplasmic pool . We examined intensity profiles along the XT axis , parallel to the tip membrane ( Figure 7B and Figure 4D ) . After deliberately bleaching part of the tip , we looked for a gradient in fluorescence recovery along XT that could arise from mobility of bound Dystrophin . If the bound Dystrophin is mobile at the tip in the plane of the membrane , bleached bound huDysGFP would leave the deliberately bleached half of the tip and be replaced by fluorescent huDysGFP from the unbleached half . This would help drive fluorescence recovery and would be especially pronounced at the boundary between the bleached and unbleached halves of the tip , generating a gradient of fluorescence along the XT axis . We see no evidence of such gradients ( Figure 7B ) . In particular , in the middle panel of Figure 7B , although there is substantial recovery ( compare the turquoise crosses with the immediately post-bleach orange triangles ) , there is no overall tendency in the intensity in the unbleached part of the tip to decrease from right to left as the bleached tip area is approached . The same applies to cases of little recovery ( Figure 7C ) , where only substantial bleaching-due-to-imaging is observed on the unbleached tip half . The lack of evidence of lateral mobility of bound huDysGFP argues in favour of the existence of a bound-mobile pool with a fast turnover rate responsible for the fast fluorescence recovery observed in most fibres . The experiments above were performed in wild-type zebrafish embryos . We asked whether the mobile-bound Dystrophin pool observed may result from competition of the exogenous human Dystrophin with endogenous zebrafish Dystrophin . To address this question , huDysGFP was expressed in Dystrophin-null zebrafish embryos ( dmdta222a/ta222a ) . We first evaluated the ability of human Dystrophin to rescue the zebrafish dystrophic fibres . Typically , in the absence of Dystrophin , the zebrafish muscle fibres detach upon contraction . At 3 days post fertilisation , nearly all dmdta222a/ta222a embryos show signs of dystrophy ( Figure 8A ) . To quantify rescue efficiency , we first evaluated the percentage of cells that detach by mosaically expressing GFP in dmdta222a/ta222a embryos . Injecting a GFP control plasmid did not affect muscle fibres in siblings ( N = 39 ) , but about 23% of the GFP positive fibres in dmdta222a/ta222a embryos detached ( Figure 8B ) . In marked contrast , cells expressing huDysGFP looked healthy and no detachment was found , suggesting full rescue of the dystrophic phenotype in dmdta222a/ta222a muscle fibres ( Figure 8B; p = 0 . 000126 in Chi-square test for significance between GFP and huDysGFP ) . 10 . 7554/eLife . 06541 . 012Figure 8 . huDysGFP rescuing and binding dynamics in dmdta222a/ta222a embryos . ( A ) 3 dpf dmdta222a/ta222a zebrafish embryo with typical dystrophic muscles as shown by actc1b:mCherry reporter ( red ) in vivo , with several healthy fibres expressing huDysGFP ( green ) . ( B ) Control GFP mosaically expressed in dmdta222a/ta222a embryos is found in both healthy ( 77% ) and dystrophic ( 23% ) fibres ( N = 56 ) . Expression of huDysGFP fully rescues the dystrophic phenotype in dmdta222a/ta222a muscle fibres , as no cells expressing huDysGFP were found detached or unhealthy in any visible aspect ( N = 56 ) . p = 0 . 000126 in Chi-square test for significance between GFP and huDysGFP . GFP and huDysGFP-positive cells in regions of very dystrophic muscles in dmdta222a/ta222a zebrafish embryos are shown . The actc1b:mCherry reporter filling up the cytoplasm and huDysGFP expression at the tip suggest that the fibre structure is kept intact even without support from neighbouring cells , unlike the GFP-positive cell . ( C ) huDysGFP ratio tip intensity/cytoplasm intensity in wild-type ( mean = 3 . 7 ± 1 . 1 s . e . m . ; n = 33 ) and dmdta222a/ta222a ( mean = 9 . 6 ± 3 . 2 s . e . m . ; n = 13 ) zebrafish embryos . In the wild-type background , where huDysGFP competes with endogenous Dystrophin for available binding sites , the average ratio is 2 . 5 times lower than in the mutant background ( p = 0 . 03 ) , where huDysGFP can occupy all available sites . ( D ) Scatter plot of fractional recovery in bleached tip pixels as a function of the cytoplasmic intensity . ( E ) Comparative scatter plots , with mean and SD , of huDysGFP fractional recovery in bleached tip pixels in wild-type ( wt ) , dmdta222a/ta222a ( dmd ) and their siblings ( sibs ) . There were no statistically significant differences between groups as determined by one-way ANOVA [F ( 2 , 67 ) = 0 . 8628 , ns] . ( F ) Comparative scatter plots , with mean and SD , of huDysGFP final unbleached tip minus bleached tip intensities in wild-type ( wt ) , dmdta222a/ta222a ( dmd ) and their siblings ( sibs ) . There were no statistically significant differences between groups as determined by one-way ANOVA [F ( 2 , 67 ) = 2 . 845 , ns] . ( G ) huDysGFP fraction of cases showing no recovery , or 50% recovery at the first ( <10 s ) , second ( <20 s ) , or later ( >20 s ) time points , calculated from unbleached tip minus bleached tip intensities , in wild-type ( wt ) , dmdta222a/ta222a ( dmd ) and their siblings ( sibs ) . There were no statistically significant differences between groups as determined by one-way ANOVA [F ( 2 , 67 ) = 0 . 1521 , ns] . Scale bars = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 012 Next , huDysGFP intensity in the whole fibre and locally at the fibre tips was measured in wild-type embryos , where it coexists with the endogenous Dystrophin and in dmdta222a/ta222a dystrophic embryos ( Figure 8C ) . In the absence of competition with the endogenous protein , huDysGFP appears to occupy more of the available binding sites , showing a 2 . 5-fold increase in the intensity ratio tips:cytoplasm compared to that found in wild-type background . This indicates that fibres expressing the same intensity in the cytoplasm can accumulate more at the fibre tips in the mutant background , consistent with the view that huDysGFP overexpression in the wild-type background does not displace all endogenous zebrafish Dystrophin . Finally , tip FRAP curves of huDysGFP in dmdta222a/ta222a and siblings were analysed . Regardless of the genotype , most tips show the three typical signatures of an immobile-bound population described above: only partial recovery in the tip bleached region independently of the cytoplasmic intensity ( Figure 8D , E; Table 3 ) , the normalized intensity difference between the unbleached and bleached regions in most tips reaches a plateau above zero ( Figure 8F; Table 3 ) , and unbleached tip half intensity shows higher drop than the cytoplasm in the fast acquisition phase ( see original curves in Bajanca et al . , 2015 ) . Also regardless of the genotype , the tip intensity registers a rapid recovery that is uncharacteristic of an immobile fraction while higher than the estimated contribution of the cytoplasmic pool . Evidence is clear when comparing events on switching between fast and slow acquisition in the unbleached tip half and cytoplasm unnormalised intensity curves . At least 50% of the total recovery occurs within 10 s in most fibre tips , too fast for the immobile-bound pool ( Figure 8G; Table 3 ) . Overall , our analysis found no significant differences between huDysGFP binding dynamics in wild-type , dmdta222a/ta222a or their siblings . These results suggest that the presence of two bound populations of huDysGFP with different turnover times is not due to competition with endogenous zebrafish Dystrophin . 10 . 7554/eLife . 06541 . 013Table 3 . Analysis of FRAP data on the tips of huDysGFP-expressing cells in dmdta222a/ta222a embryos and siblingsDOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 013Tip numberEmbryo genotypeCytoplasm intensityFractional recoveryFinal normalized unbleached-bleached intensitiesUnbleached-cytoplasm 50% recovery1dmdta222a/ta222a51 . 560 . 120 . 18<10 s2dmdta222a/ta222a23 . 120 . 100 . 31<10 s3dmdta222a/ta222a38 . 040 . 070 . 25no recovery4dmdta222a/ta222a83 . 990 . 250 . 02<10 s5dmdta222a/ta222a4 . 460 . 060 . 30<10 s6dmdta222a/ta222a0 . 34−0 . 010 . 21<10 s7dmdta222a/ta222a3 . 160 . 210 . 17<20 s8dmdta222a/ta222a16 . 490 . 160 . 24<10 s9dmdta222a/ta222a18 . 670 . 060 . 07<10 s10dmdta222a/ta222a8 . 730 . 200 . 25<10 s11dmdta222a/ta222a7 . 120 . 470 . 12<20 s12dmdta222a/ta222a11 . 060 . 170 . 15<20 s13dmdta222a/ta222a10 . 650 . 090 . 23<10 s14dmdta222a/ta222a28 . 410 . 020 . 13<20 s15dmdta222a/ta222a8 . 910 . 170 . 14<10 s16dmdta222a/ta222a41 . 160 . 240 . 16<10 s17dmdta222a/ta222a17 . 710 . 360 . 50<20 s18dmdta222a/ta222a20 . 430 . 250 . 36<10 s19sibling30 . 85−0 . 020 . 20<10 s20sibling45 . 810 . 120 . 17<30 s21sibling13 . 890 . 110 . 23<10 s22sibling7 . 050 . 100 . 33<10 s23sibling12 . 780 . 010 . 32<10 s24sibling6 . 480 . 090 . 26<10 s25sibling10 . 950 . 070 . 28<10 s26sibling8 . 350 . 200 . 22<10 s27sibling10 . 140 . 150 . 15<10 s28sibling7 . 150 . 110 . 50<10 s29sibling7 . 770 . 060 . 57<10 s30sibling19 . 640 . 160 . 18<10 s31sibling16 . 080 . 320 . 36<10 s32sibling8 . 110 . 130 . 37<10 s33sibling7 . 300 . 200 . 36<10 s34sibling14 . 990 . 400 . 06<20 s35sibling29 . 820 . 280 . 06<20 s36sibling9 . 950 . 340 . 18<10 s37sibling32 . 10−0 . 020 . 12<10 sSee Table 2 for detailed information . We asked whether the labile membrane-bound pool may be a consequence of weaker binding between the human Dystrophin protein and the zebrafish endogenous protein complexes . To address this question , we analysed the dynamics of overexpressed zebrafish Dystrophin: GFP-tagged zebrafish Dystrophin ( zfDysGFP ) . Similar to huDysGFP , the overexpression of zfDysGFP results in mosaic expression and variable levels of accumulation both at the muscle fibre tips and cytoplasm ( Figure 9A; Table 4 ) . The cytoplasmic zfDysGFP diffusion dynamics was analysed by FRAP . DzfDysGFP best-fit values ( fitted to data over ∼40 s ) ranged from 0 . 6 to 6 . 7 μm2 s−1 , with a mean of 2 . 9 μm2 s−1 ( Figure 9B ) . zfDysGFP has statistically significantly lower D than GFP ( p < 0 . 001 ) but shows no difference with huDysGFP ( Figure 9B ) . 10 . 7554/eLife . 06541 . 014Figure 9 . zfDysGFP dynamics in wild-type and dmdta222a/ta222a embryos . ( A ) zfDysGFP variable intensity of expression in muscle fibres of 2 dpf wild-type embryo . Arrows point to low and arrowheads to high expressing tips . ( B ) Comparative scatter plots of DGFP , DhuDysGFP and DzfDysGFP . One-way ANOVA revealed a statistically significant difference between groups [F ( 2 , 44 ) = 57 . 08 , p < 0 . 0001] . Tukey post-hoc test revealed that DhuDysGFP ( 4 . 4 ± 2 . 7 μms2s−1 ) and DzfDysGFP ( 2 . 9 ± 1 . 7 μms2s−1 ) are not statistically significant but are statistically significantly lower ( p < 0 . 001 ) than DGFP ( 13 . 2 ± 3 . 7 μms2s−1 ) . ( C ) A rescued zfDysGFP ( green ) fibre within a 2 dpf dmdta222a/ta222a zebrafish embryo with otherwise typical dystrophic muscles as shown by actc1b:mCherry ( red , note extensive gaps in muscle ) reporter in vivo . ( D ) Scatter plot of fractional recovery in bleached tip pixels as a function of the cytoplasmic intensity . ( E ) Comparative scatter plots , with mean and SD , of the fractional recovery in bleached tip pixels of huDysGFP in wild-type ( wt ) embryos , and zfDysGFP in wild-type and dmdta222a/ta222a ( dmd ) embryos . There were no statistically significant differences between groups as determined by one-way ANOVA [F ( 2 , 70 ) = 3 . 019 , ns] . ( F ) Comparative scatter plots , with mean and SD , of final unbleached tip minus bleached tip intensities of huDysGFP in wild-type embryos , and zfDysGFP in wild-type and dmdta222a/ta222a embryos . One-way ANOVA revealed a statistically significant difference between groups [F ( 2 , 70 ) = 6 . 818 , p = 0 . 002] . Tukey post-hoc test revealed that zfDysGFP in wild-type ( 0 . 3 ± 0 . 16 , n = 22 ) and dmdta222a/ta222a embryos ( 0 . 3 ± 0 . 1 , n = 18 ) are not statistically significant but huDysGFP ( 0 . 17 ± 0 . 12 , n = 33 ) is statistically significantly lower than zfDysGFP in wild-type ( p < 0 . 01 ) and in dmdta222a/ta222a embryos ( p < 0 . 05 ) . ( G ) Fraction of cases showing no recovery , or 50% recovery at the first ( <10 s ) , second ( <20 s ) , or later ( >20 s ) time points , calculated from unbleached tip minus bleached tip intensities , in huDysGFP in wild-type embryos , and zfDysGFP in wild-type and dmdta222a/ta222a embryos . There were no statistically significant differences between groups as determined by one-way ANOVA [F ( 2 , 70 ) = 1 . 405 , ns] . Scale bars = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 01410 . 7554/eLife . 06541 . 015Table 4 . Analysis of FRAP data on the tips of zfDysGFP-expressing cells in wild-type and dmdta222a/ta222a embryosDOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 015Tip number ( cell number ) Embryo genotypeCytoplasm intensityFractional recoveryFinal normalized unbleached-bleached intensitiesUnbleached-cytoplasm 50% recovery1 ( cell 1 ) wild type14 . 590 . 080 . 46<10 s2 ( cell 2 ) wild type5 . 060 . 070 . 24<10 s3 ( cell 3 ) wild type9 . 440 . 170 . 12<20 s4 ( cell 4 ) wild type6 . 510 . 170 . 38<10 s5 ( cell 4 ) wild type8 . 020 . 300 . 05<10 s6 ( cell 5 ) wild type79 . 280 . 040 . 22<10 s7 ( cell 6 ) wild type1 . 780 . 090 . 51<20 s8 ( cell 6 ) wild type1 . 010 . 120 . 38<10 s9 ( cell 7 ) wild type0 . 280 . 210 . 13<10 s10 ( cell 8 ) wild type27 . 460 . 050 . 16<10 s11 ( cell 9 ) wild type12 . 130 . 080 . 17<10 s12 ( cell 9 ) wild type17 . 340 . 120 . 40<10 s13 ( cell 10 ) wild type5 . 090 . 130 . 23<10 s14 ( cell 10 ) wild type2 . 140 . 030 . 34<10 s15 ( cell 11 ) wild type−0 . 130 . 130 . 80<10 s16 ( cell 12 ) wild type2 . 090 . 060 . 17<10 s17 ( cell 12 ) wild type1 . 940 . 040 . 31<10 s18 ( cell 13 ) wild type5 . 800 . 030 . 29<10 s19 ( cell 13 ) wild type6 . 65−0 . 040 . 32<10 s20 ( cell 14 ) wild type6 . 420 . 270 . 31<10 s21 ( cell 14 ) wild type6 . 87−0 . 020 . 35<10 s22 ( cell 15 ) wild type8 . 960 . 250 . 23<10 s23 ( cell 16 ) dmdta222a/ta222a11 . 400 . 420 . 25<10 s24 ( cell 17 ) dmdta222a/ta222a4 . 700 . 050 . 35<10 s25 ( cell 17 ) dmdta222a/ta222a3 . 810 . 010 . 40<10 s26 ( cell 18 ) dmdta222a/ta222a6 . 770 . 330 . 12<10 s27 ( cell 19 ) dmdta222a/ta222a−0 . 38−0 . 020 . 36<10 s28 ( cell 19 ) dmdta222a/ta222a−0 . 600 . 110 . 32<10 s29 ( cell 20 ) dmdta222a/ta222a4 . 860 . 120 . 38<10 s30 ( cell 21 ) dmdta222a/ta222a3 . 450 . 470 . 24<10 s31 ( cell 21 ) dmdta222a/ta222a3 . 600 . 470 . 25<10 s32 ( cell 22 ) dmdta222a/ta222a7 . 600 . 390 . 10<10 s33 ( cell 23 ) dmdta222a/ta222a26 . 070 . 190 . 18<10 s34 ( cell 24 ) dmdta222a/ta222a6 . 080 . 150 . 45<10 s35 ( cell 24 ) dmdta222a/ta222a8 . 030 . 070 . 38no recovery36 ( cell 25 ) dmdta222a/ta222a17 . 930 . 280 . 22<10 s37 ( cell 26 ) dmdta222a/ta222a6 . 210 . 120 . 23<10 s38 ( cell 27 ) dmdta222a/ta222a5 . 620 . 070 . 13<20 s39 ( cell 28 ) dmdta222a/ta222a6 . 880 . 230 . 22<10 s40 ( cell 29 ) dmdta222a/ta222a23 . 050 . 170 . 29<20 sSee Table 2 for detailed information . Next , we analysed zfDysGFP dynamics at the fibre tips , in a wild-type context , where it competes for binding with endogenous Dystrophin , or in dmdta222a/ta222a mutants ( Figure 9A , C ) . Exogenous zfDysGFP is able to rescue the dystrophic phenotype in dmdta222a/ta222a mutants ( 32/32 , p = 0 . 003; Figure 9C ) . Like huDysGFP , most zfDysGFP tips show the typical signatures of an immobile-bound population regardless of the genotype ( Figure 9D–F; Table 4 ) . There are no statistically significant differences between the fractional recoveries of Dystrophin of the different species and in the different genetic backgrounds ( Figure 9E ) . However , the final unbleached minus bleached intensity of huDysGFP is statistically significantly lower than zfDysGFP ( Figure 9F ) . Using a two-way ANOVA to test the effect of genotypic background ( wild-type or dmdta222a/ta222a ) and Dystrophin origin ( human or zebrafish ) on the immobile fraction , we observed a significant effect of Dystrophin origin ( human vs zebrafish ) [F ( 1 , 87 ) = 11 . 21 , p = 0 . 0012] but not of the host genotype [F ( 1 , 87 ) = 0 . 02326 , p = 0 . 8791] , and there was no significant interaction between origin and genotype [F ( 1 , 87 ) = 1 . 367 , p = 0 . 2455] . While a final unbleached minus bleached intensity above zero indicates the existence of an immobile-bound fraction , its value is not a direct measurement of the amount of immobile-bound Dystrophin . However , we cannot exclude the possibility that huDysGFP forms less , or less stable , immobile-bound links than zfDysGFP , in the zebrafish environment . The background genotype appears not to influence this trait , suggesting that if there is a difference in the immobile fraction that is independent from competition with endogenous Dystrophin . Like huDysGFP , most zfDysGFP-expressing fibres show a fast recovery phase that is higher than the estimated contribution of the cytoplasmic pool . Typically , at least 50% of the total recovery of unbleached tip half minus cytoplasm intensity curves occurs within 10 s in the majority of the cases , regardless of species and genotype ( Figure 9G; Table 4 ) . This recovery is very fast and the characteristic immobile pool plateau is soon reached , suggesting the existence of an additional bound pool of zfDysGFP with rapid turnover . These results show that , like huDysGFP , zfDysGFP can be found in three populations , one diffusible and two bound with different binding lifetimes . Therefore , regardless of any differences between huDysGFP and zfDysGFP , our results confirm that the mobile-bound pool previously found for the human Dystrophin is not caused by weaker interactions with a different species environment . Analysis of FRAP curves of both huDysGFP and zfDysGFP indicated that when a low level of diffusible Dystrophin is detected in cytoplasm , there can be a significant amount of bound-mobile pool . However , there is the possibility that the labile-bound pool is caused by an excess of cytoplasmic Dystrophin resulting from overexpression . We questioned whether a mobile-bound pool can also be found in endogenous zebrafish Dystrophin , where a cytoplasmic pool of the protein is not known . We have analysed Gt ( dmd-Citrine ) ct90a zebrafish embryos , in which Citrine was inserted by gene-trap into the endogenous Dystrophin locus , creating fluorescently tagged zfDys ( Trinh et al . , 2011; Ruf-Zamojski et al . , 2015 ) . We first searched for signs of cytoplasmic Dystrophin . As every muscle fibre expresses Citrine-tagged Dystrophin ( zfDysCitrine; Figure 10A ) , and different tissues may have different background fluorescence , it is not easy to evaluate with confidence whether there are low levels of cytoplasmic zfDysCitrine above the background . We hypothesized that if there is zfDysCitrine in the cytoplasm , we should register a recovery after photobleaching . Control siblings lacking zfDysCitrine show a flat and noisy FRAP curve that reflects the background fluorescence ( Figure 10B , plot a ) . In contrast , the very low zfDysCitrine intensity detected in the cytoplasm could be bleached to even lower levels , and a recovery curve is registered , indicating the presence of diffusible zfDysCitrine in the cytoplasm ( Figure 10B , plot b ) . The switch from a dark state of Citrine was evaluated and contributes little to the recovery ( Figure 10B , plot c ) . Another signature of the presence of diffusible zfDysCitrine was the inverted bell-shaped , instead of top-hat , Gaussian curve that results from diffusion during intentional bleaching ( Figure 10C ) . Finally , zfDysCitrine recovery curves can be fit using a diffusion model ( Figure 10D ) . DzfDysCitrine best-fit values ranged from 0 . 9 to 4 . 3 μm2 s−1 and mean 2 . 2 μm2 s−1 ( Figure 10E ) . There is no significant statistical difference between DzfDysCitrine and DzfDysGFP . 10 . 7554/eLife . 06541 . 016Figure 10 . Endogenously driven zfDysCitrine dynamics . ( A ) In vivo zfDysCitrine ( green ) expression in muscle fibres of 2 dpf Gt ( dmd-citrine ) ct90a embryo contrasted with transmitted light . ( B ) Schematics of zebrafish muscle fibres showing origin of scatter plots a , b , and c . Bleaching a region in the cytoplasm of Citrine-negative siblings ( a ) results in a flat curve of background fluorescence intensity , while the same experiment in Citrine-positive embryos ( b ) results in a significant drop in fluorescence followed by recovery . Bleaching a large region to include the entire fibre abolishes recovery ( c ) , indicating that recovery from a citrine dark state makes a negligible contribution to recovery in ( b ) . ( C , D ) Normalized FRAP experimental data and fitting curves of a zfDysCitrine fibre cytoplasm . ( C ) Normalized intensity profile along the X-axis at the first time point after bleaching ( dots ) and Gaussian fit ( red line ) . ( D ) Recovery curves along X-axis ( dots ) and fit of the diffusion model to the post-bleach ( red line ) . ( E ) Comparative scatter plots of DzfDysGFP and DzfDysCitrine . t test shows no statistically significant differences . ( F ) Scatter plot of fractional recovery in bleached tip pixels as a function of the cytoplasmic intensity , for zfDysCitrine embryos of different developmental stages . HS = heat-shocked embryos . ( G ) Comparative scatter plots , with mean and SD , of the fractional recovery in bleached tip pixels of zfDysGFP and zfDysCitrine . t test shows no statistically significant difference . ( H ) Comparative scatter plots , with mean and SD , of final unbleached tip minus bleached tip intensities of zfDysGFP and zfDysCitrine . t test shows no statistically significant differences . ( I ) Fraction of cases showing no recovery , or 50% recovery at the first ( <10 s ) , second ( <20 s ) , or later ( >20 s ) time points , calculated from unbleached tip minus bleached tip intensities , in zfDysGFP and zfDysCitrine . t test shows a statistically significant difference ( p = 0 . 0016 ) . Scale bar = 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 016 To be able to analyse single fibres , some embryos were subjected to heat shock , which causes mosaic disruption of the somites ( Ruf-Zamojski et al . , 2015 ) . Embryos not subjected to heat shock , at different developmental stages ( 30 hpf , 40 hpf and 48 hpf ) were also analysed . However , no trend was obvious when analysing embryos of different developmental stages ( Figure 10F; Table 5 ) . Most tips showed the typical signatures of an immobile-bound pool , as expected , and no statistically significant differences were found between zfDysGFP and zfDysCitrine ( Figure 10F–H; Table 5 ) . There is fractional recovery in most bleached tips that tends to be low ( Figure 10F , Table 5 ) , which in accordance with the results obtained for huDysGFP and zfDysGFP cases when the cytoplasm intensity is just above detectable ( Figures 6B , 8D , 9D; Tables 2–4 ) . Therefore , we found no indications that endogenous zfDysCitrine behave differently from exogenous zfDysGFP regarding their immobile pool signatures . 10 . 7554/eLife . 06541 . 017Table 5 . Analysis of FRAP data on the tips of ZfDysCitrine expressing cells in Gt ( dmd-Citrine ) ct90a embryosDOI: http://dx . doi . org/10 . 7554/eLife . 06541 . 017Tip numberSetCytoplasm intensityFractional recoveryFinal normalized unbleached-bleached intensitiesUnbleached-cytoplasm 50% recovery1heat shock1 . 71−0 . 040 . 37<20 s2heat shock1 . 850 . 150 . 14<20 s3heat shock0 . 370 . 100 . 39<20 s4heat shock3 . 520 . 090 . 14<30 s5heat shock0 . 980 . 130 . 17<20 s6heat shock0 . 870 . 070 . 21<20 s7heat shock1 . 300 . 130 . 17<10 s848 hpf4 . 300 . 050 . 36<10 s948 hpf3 . 110 . 040 . 37<10 s1048 hpf1 . 720 . 120 . 53<20 s1148 hpf2 . 090 . 040 . 39<20 s1248 hpf2 . 010 . 070 . 42<10 s1348 hpf6 . 21−0 . 020 . 41<10 s1440 hpf3 . 260 . 100 . 27<20 s1540 hpf0 . 720 . 140 . 25<30 s1640 hpf3 . 570 . 030 . 26<20 s1740 hpf4 . 950 . 080 . 22<20 s1840 hpf0 . 790 . 010 . 25<10 s1940 hpf0 . 090 . 030 . 28<20 s2030 hpf1 . 530 . 070 . 38<10 s2130 hpf1 . 600 . 060 . 48<60 s2230 hpf5 . 020 . 090 . 50<20 s2330 hpf2 . 660 . 090 . 64<40 s2430 hpf3 . 040 . 110 . 44<10 s2530 hpf3 . 190 . 070 . 37<10 s2630 hpf2 . 780 . 030 . 28<10 sIn embryos subjected to heat shock ( tips 1–7 ) , individual fibres could be selected for FRAP , at 48 hpf , and background levels are taken into account , contrary to the remaining cases ( tips 8–26 ) . See Table 2 for detailed information . Despite the low cytoplasmic intensity , many zfDysCitrine tips still show a very fast recovery . At least 50% of the total recovery of the unbleached tip half minus the cytoplasm happens in almost half of the cases within 10 s , and still very fast at less than 20 s in the majority of the remaining cases ( Figure 10I; Table 5 ) . This suggests the presence of a mobile-bound pool , regardless of the statistically significant difference in the time it takes for zfDysCitrine and zfDysGFP to recover ( p = 0 . 0016 ) . The tendency to take longer to recover is likely to reflect the dependency of the mobile-bound pool on cytoplasmic availability of free Dystrophin . This is in accordance with our previous observations that recovery of the mobile-bound pool is fast enough that it may be limited by diffusion . The data therefore indicate that endogenous Dystrophin with an inserted Citrine , like exogenous zfDysGFP and huDysGFP , can be found in two states , immobile and mobile , regardless of a very low , but detectable , cytoplasmic pool . In various fields , transgenic and humanized animal models are a valuable resource where non-invasive methods to study human biology are lacking ( Boverhof et al . , 2011; Attfield and Dendrou , 2012; Akkina , 2013 ) . Here , we show that exogenous zebrafish and human Dystrophin have subcellular localizations , at both mRNA and protein levels , equivalent to that of endogenous Dystrophin ( Ruf-Zamojski et al . , 2015 ) . Furthermore , exogenous zebrafish and human Dystrophin have diffusion and binding dynamics similar to those of endogenous zebrafish Dystrophin , in spite of the artificially raised cytoplasmic levels caused by over-expression . Furthermore , we found that Dystrophin dynamics is not affected by the position of the fluorescent tag ( internally close to the actin binding site [Citrine] or in C-terminal position [GFP] ) . Importantly , both zebrafish and human Dystrophins were able to rescue the dystrophic phenotype of dmdta222a/ta222a embryos . Taken together , these data indicate that the zebrafish embryo is a good model system to study the dynamics of human Dystrophin in live muscle cells in vivo using fluorescently tagged versions of the protein . It is important to keep in mind that human Dystrophin may behave differently in zebrafish than in human muscle . The FRAP analysis methodology developed in this study could be applied for studies on human primary muscle cell cultures , or even pluripotent human stem cells differentiated into muscle fibres ( Chal et al . , 2015 ) . However , until a suitable 3D ex-vivo physiologically relevant human muscle system is readily available for routine experimentation , our methodology and findings provide a baseline for future comparative studies . For instance , the strategy presented here can be used to study the effects of shortening the protein on Dystrophin dynamics , as occurs in patients with BMD and planned exon-skipping gene therapies ( Koenig et al . , 1989; Cirak et al . , 2011; Konieczny et al . , 2013; Verhaart et al . , 2014 ) . We show that , in all three experimental conditions used ( exogenous zfDysGFP and huDysGFP or endogenously-driven zfDysCitrine ) , part of Dystrophin is found in a cytoplasmic freely diffusing pool . Despite considerable apparent variation in measured diffusion constant ( D ) from fibre to fibre ( 1 . 4–10 . 1 μm2 s−1 for DhuDysGFP , 0 . 6 to 6 . 7 μm2 s−1 for DzfDysGFP , 0 . 9 to 4 . 3 μm2 s−1 for DzfDysCitrine , mean 4 μm2 s−1 , 3 μm2 s—1 , and 2 μm2 s—1 , respectively ) , it is clear that the mobility of Dystrophin is on average around a fourth that of GFP , at approximately 13 μm2 s−1 . However , whereas GFP is a small ( 3 × 3 × 4 nm ) globular protein , Dystrophin is thought to be a rod perhaps 100 nm long ( Pons et al . , 1990; Arrio-Dupont et al . , 2000; Bhasin et al . , 2005; Kameta et al . , 2010; Kinsey et al . , 2011 ) , so a diffusion constant only a quarter that of GFP is surprising . As Dystrophin appears to diffuse , perhaps so-called ‘active diffusion’ due to energy-using cellular processes ( e . g . , molecular motors ) enhances its apparent mobility in a non-directed manner ( Brangwynne et al . , 2008 , 2009; Weber et al . , 2012 ) . As whole Dystrophin structure has not been reported , one can imagine that a compact rapidly diffusing Dystrophin conformation may account for Dystrophin dynamics in vivo . For now , it is not known whether cytoplasmic Dystrophin also exists in low amounts in adult human skeletal muscle cells . However , many studies have shown that , in human embryos and foetuses , Dystrophin first appears in the cytoplasm ( Wessels et al . , 1991; Clerk et al . , 1992; Chevron et al . , 1994; Mora et al . , 1996; Torelli et al . , 1999 ) . Interestingly , a cytoplasmic Dystrophin pool was also found in the adult heart ( Peri et al . , 1994 ) and in both regenerating fibres ( Kääriäinen et al . , 2000 ) and differentiating human primary muscle cultures ( Miranda et al . , 1988 ) , where Dystrophin localisation was suggested to recapitulate the embryonic process of Dystrophin deposition , accumulating initially in the cytoplasm and muscle–tendon junctions at the fibre ends , prior to maturing towards costameric localisation . Therefore , it would be particularly interesting to analyse Dystrophin dynamics in the cytoplasm of human dystrophic muscle cells , which undergo repeated cycles of regeneration . Our direct semi-quantitative FRAP analysis of bleaching and recovery allowed identification of two populations of immobile- and mobile-bound Dystrophin . The dynamic mobile-bound pool is found in fibres with undetectably low levels of cytoplasmic huDysGFP , zfDysGFP , or endogenously-driven zfDysCitrine , indicating that the mobile-bound state occurs independently of over-expression or high level cytoplasmic accumulation . A proportion of the Dystrophin located at muscle fibre tips is in free exchange with the cytoplasmic diffusible pool , whereas a further portion binds stably at muscle fibre tips . Whether this immobile pool is generated by stabilisation of the mobile-bound pool , or by another route , remains to be determined ( Figure 11 ) . It is not clear whether the dynamics of early tip-localised Dystrophin is different from the later costameric Dystrophin . The tip region of zebrafish muscle fibres corresponds to myotendinous junction at the ends of human muscle fibres , where Dystrophin is also found enriched in humans ( Zhao et al . , 1992 ) and in several other mammals , namely mouse ( Samitt and Bonilla , 1990 ) , rat ( Kääriäinen et al . , 2000 ) , and guinea pig ( Masuda et al . , 1992 ) . The immobile Dystrophin pool at the cell tips is bound tightly enough to transmit significant ( 10 pN/molecule ) forces for significant ( 1 s ) times , enough to unfold Dystrophin's spectrin domains ( Bhasin et al . , 2005 ) . Dystrophin in the bound mobile pool presumably cannot transmit significant force . However , weak binding would allow response to weak short-lived ( sub-pN/molecule , sub-second ) forces or may fulfil another function , such as structure assembly , sensing , or signalling . Interestingly , studies of mutants of ezrin , which , like Dystrophin , binds both β-dystroglycan and actin , revealed immobile and mobile membrane bound forms ( Coscoy et al . , 2002; Spence et al . , 2004 ) . The dynamic membrane-bound ezrin was suggested to be an intermediate conformation state leading to actin anchoring and full complex assembly . Similarly , it is possible that Dystrophin binding to β-dystroglycan facilitates a conformation change to promote actin binding and stabilization of the complex ( Friedel et al . , 2006 ) ( Figure 11 ) . Membrane localization , turnover , and clustering of other adhesion molecules such as cadherins is known to be influenced by the tension experienced by the cells ( Delanoë-Ayari et al . , 2004; Yonemura et al . , 2010; De Beco et al . , 2015 ) . Thus , it would be interesting to investigate whether muscle contraction might favour the conversion of some of the mobile-bound or cytoplasmic Dystrophin into immobile Dystrophin . Our expression vectors contain a CMV promoter that drives human and zebrafish Dystrophin expression throughout the fish . Nonetheless , we observed that muscle cells accumulate Dystrophin protein much more frequently than other cell types , suggesting that human Dystrophin stabilization is tissue dependent . The availability of suitable docking sites or specific partner proteins may play a role in stabilization ( Le Rumeur et al . , 1804 ) . In addition , sub-cellular accumulation of the mRNA itself may contribute to Dystrophin positioning at the tip . Like zebrafish Dystrophin mRNA , RNA encoding human Dystrophin localized near the tips of fibres . The expression constructs engineered in the present study do not contain the Dystrophin 5′- or 3′-UTR or introns . Instead , the coding sequence is preceded by a standard chimaeric intron . While previous studies showed a role for 5′ and 3′ UTR regions into controlling tissue-specific expression and transcriptional regulation of Dystrophin ( see Larsen and Howard , 2014 and references therein ) , our results show that UTR regions are dispensable for accumulation of the mRNA at the tip . It is therefore unlikely that Dystrophin RNA accumulation at the tips is due to specific RNA transport since that would most likely require the presence of the untranslated regions ( for review see Kloc et al . , 2002; Holt and Bullock , 2009 ) . Thus , the signals controlling the correct localization of the mRNA remain unclear . One possibility is anchorage by numerous nascent protein chains ( Figure 11 ) . This might then facilitate transition of Dystrophin into the strongly bound form . The new analysis methods developed here broaden the applications of FRAP . Specifically , our modelling overcomes low signal-to-noise ratios and accounts for diffusion during intentional bleaching and unintentional bleaching during imaging . Bleaching due to imaging is here used to analyse binding dynamics . Indeed , at the high laser powers needed in vivo a specific bleaching step may not be necessary , or even optimal . These advances may be generally valuable for studies of embryos and thick tissues or biomaterials by allowing higher laser powers . Also , our results show that pooling results from different cells in a complex in vivo environment may lead to error and mask individual cell-to-cell variability . Finally , our model for one-dimensional diffusion may be especially useful in elongated cells such as muscle cells or neurons . The model was implemented in open-source software that fits diffusion coefficients to data without programming ( ‘Materials and methods’ ) . Our methodology may have general application for analysis of protein dynamics in vivo . In conclusion , the present study reveals for the first time the complex dynamics of Dystrophin in maturing muscle cells within the intact animal . It reveals important cell-to-cell variations that most likely reflect fibre or tip maturation but could have another origin . Both developmental state and genetic background are thus expected to influence the stability of Dystrophin , which could prove important in the clinic . Unstable binding or overall shortage of a specific pool may affect Dystrophin turnover and muscle performance . In future , the methodology developed here can be used to test for the comparative performance of short Dystrophin forms in use in gene therapy trials , with the aim of focusing on stable versions that may favour a more successful clinical outcome . Full-length 427-kd human Dystrophin ( huDys ) was produced by generating a human Dystrophin cDNA using long-range PCR ( primers F1: SpeI_GACTAGTGTGTTCTTCATATGTATATCCTTCC; R1: MluI_CGACGCGTCATTGTGTCCTCTCTCATTG ) , digested with SpeI and MluI and cloned into pCI plasmid ( Promega , Madison , WI , United States ) downstream of a CMV promoter at the NheI and MluI restriction sites . Insertion of GFP tag: ( 1 ) primers F2 ( TCACCTCGAGAAAGTCAAGGCACTTCGAGGAGAAATTG , matching the 3′ huDystrophin cDNA plus a 5′ XhoI site ) and R2 ( CCTCGCCCTTGCTCACCATGGTTGTGGCCATTGTGTCCTCTCTCATTGGCTTTCCAGGGGTATTTCTTC , designed to remove the Dystrophin stop codon and harbouring first 30 nucleotides of eGFP cDNA ) were used on huDys; ( 2 ) eGFP cDNA was amplified with F3 ( GAAGAAATACCCCTGGAAAGCCAATGAGAGAGGACACAATGGCCACAACCATGGTGAGCAAGGGCGAGG , containing a 5′ free tail encoding the Dystrophin cDNA end ) and R3 ( GGTACCACGCGTTTACTTGTACAGCTCGTCCATGCC , plus a MluI site ) ; ( 3 ) finally , the two products were mixed , amplified with F2 and R3 , digested with XhoI and MluI and inserted into pre-digested huDys to generate huDysGFP . GFP was expressed from pCMV-GFP ( Addgene 11153 ) . Full-length zebrafish Dystrophin ( Lai et al . , 2012 ) GFP tagged was synthesized by GenBrick and subcloned into pCI-Neo at the MluI-SalI site ( GenScript USA Inc . , Piscataway Township , NJ , United States ) . All constructs were fully sequenced . Fish used were King's wild-type Danio rerio , dmdta222a/+ , Tg ( actc1b:mCherry ) pc4 ( Cole et al . , 2011 ) , and Gt ( dmd-Citrine ) ct90a ( Trinh et al . , 2011; Ruf-Zamojski et al . , 2015 ) and were staged and reared as described ( Westerfield , 1995 ) . Plasmids were injected into 1-cell stage embryos at 20–40 pg/embryo . Phenolthiourea ( 0 . 003% ) was added to inhibit pigmentation . Heat shock was performed at 6 s and embryos analysed at 48 hpf ( Ruf-Zamojski et al . , 2015 ) . To image , 48 hpf dechorionated embryos were anaesthetized with tricaine ( 0 . 2 mg/ml ) and embedded in 1 . 5% low melting point agarose diluted in fish water . Standard protocols were used . Embryos were fixed in cold methanol for Dystrophin staining , or otherwise in paraformaldehyde 4% . Antibodies were mouse anti-Dystrophin MANDRA1 ( 1:100; Novocastra , Roche , Basel , Switzerland ) , mouse anti-human Dystrophin Dy8 ( 1:100; Novocastra , Roche ) , rabbit anti-GFP ( 1:500; Roche ) , goat anti-mouse Alexa-543 , and goat anti-rabbit Alexa-488 . NMJ were detected with conjugated bungarotoxin-594 ( 1:1000 , Invitrogen , ThermoFisher Scientific , Waltham , MA , United States ) . In situ hybridization was performed as previously described ( Hinits and Hughes , 2007 ) , with a specific probe against human Dystrophin spectrin repeats 20–22 . An upright Zeiss Exciter laser scanning microscope ( LSM ) with a 40×/1 . 1 W Corr LD C-Apochromat objective , and an inverted Zeiss 710 LSM with a 20×/1 . 0 W Plan-Apochromat and a 40×/1 . 3 Plan-Apochromat objective were used for FRAP and Z-stacks . Acquisition and maximum intensity projections were made with ZEN 2009/2010 ( Zeiss , Jena , Germany ) . Volocity version 6 . 0 . 1 ( PerkinElmer , Waltham , MA , United States ) was used for XYZ projection , to which a fine Gaussian filter was applied and brightness corrected for visualization purposes only . Images were uniformly contrasted with Adobe Photoshop CS4 . Illustrations were made in Adobe Illustrator CS3 . GraphPad Prism 6 was used for statistical analysis and graph plotting . This parameter is a measure of the amount of fluorescence signal resulting from the expression of huDysGFP in a cell or defined sub-cellular region . Confocal Z-stacks of muscle fibres were acquired as 8-bit greyscale images with a voxel size of 0 . 147 µm × 0 . 147 µm × 1 µm ( x , y , z ) . ImageJ v1 . 45a was used for the next steps . The fibre tips or cytoplasm regions were manually delimited on sum projections of the pixel intensities over z-stacks . The raw integrated density ( sum of the values of the pixels ) in the tips and cytoplasm areas was measured and corrected for the average background . An approximate best correction for the contribution of cytoplasm signal at the tip region , typically at up a 45° angle to the field of view , was made by subtracting the corresponding cytoplasmic signal to half the tip area . Bleaching was performed at 100% intensity of an argon laser at 488 nm for GFP and 514 nm for Citrine . The acquisition region was a 300 × 60 pixel rectangle ( 44 µm × 8 . 8 µm ) , the interval between scanning rounds 0 . 2 s at a pixel dwell of 1 . 6 µs . For cytoplasmic studies , an open pinhole was used to optimize capture of the dim Dystrophin cytoplasmic signal , whereas a 1 Airy pinhole was generally used for GFP ( except cells 6 and 7 , Table 1 ) . For studying the tips , the pinhole was set at 1 Airy to improve imaging resolution of the tip region and bleaching was performed on a 60 × 20 pixels rectangle with a single scan at pixel dwell of 12 . 8 µs , thus minimizing bleaching time . Conditions for zfDysCitrine FRAP were adapted to avoid depleting too much of the very low cytoplasmic pool . Also see Figures 4 , 6A and Table 1 . The model considers diffusion in one dimension , X , of a single species in a spatially uniform background and includes bleaching by each imaging scan . The differential equation for the concentration of fluorescent protein Cf ( X , t ) is∂Cf ( X , t ) ∂t=D∂2Cf ( X , t ) ∂X2−βΘFV ( X ) ( ∑iδ ( ti ) Cf ( X , t ) ) , where D is the diffusion constant . The first two terms are the normal diffusion partial differential equation . The third term accounts for bleaching at each image acquisition . For each acquired image a fraction β of the fluorescent protein in the imaged area is assumed bleached . The sum is over the image acquisition times ti . The indicator function ΘFV ( X ) is one within the scanned field of view and zero outside of it . The delta function at time ti of the acquisition of the i'th image ( δ ( ti ) ) , assumes that bleaching occurs in the entire imaged region instantaneously . The fitting procedure is a least-squares two-parameter fit , D and β , to the FRAP curve . The boundary conditions for this equation are that the first image after bleaching is given by a Gaussian fit to the profile in this first image , and zero-flux boundary conditions at the two tips of the model cell . The equation is solved within a box approximately as long as the cell and much larger than the imaged region . The entire cell is modelled as the protein diffuses in and out of the imaged region . The effect of varying cell length and position of the bleached region in the model was quantified and deemed minimal ( Table 1 ) . To compare the experimental data with the model results , background subtracted pixel values are averaged along the YT-axis to obtain profiles of intensity as a function of XT . Background is defined as the average intensity signal from a rectangular region outside the cell over pre-bleach images 4–20 . A normalized profile for the first post-bleach point is constructed by dividing the first post-bleach profile by the average of the profiles in pre-bleach time points 4–20 . A Gaussian A0 − C0exp[− ( X − X0 ) 2/ ( 2σ2 ) ] is fit to this normalized post-bleach profile , providing an initial profile for the FRAP-curve simulation . The parameters of the profile at the end of bleaching , A0 , C0 , X0 , and σ are , respectively , the normalized intensity far from the region bleached , the maximum bleaching depth , the centre of the bleached region ( along the X-axis ) , and the bleaching width . A0 is set to A0 = 1 , the remaining three parameters are fit . The normalized FRAP curve experimental points are compared with the average value in the computed profile in the bleached region . Fitting is done by minimizing the sum-of-the-squares of the difference between the two , to obtain the best-fit values of D and β . For fits to the initial recovery only , bleaching is small and so a one-parameter fit to D is done in this case . Fibre tips are divided into two 60 × 20 pixels areas , each covering approximately half of a typical tip , together with some cytoplasm and some pixels outside the cell . ( Figure 6A ) . One box is intentionally bleached ( bleached tip ) , while the other is imaged in an identical way but not intentionally bleached ( unbleached tip ) . A region of similar size in the cytoplasm at a distance from the bright tip region is also analysed . For all regions , the background is subtracted and time points 4 to 20 are used to generate an initial pre-bleach average image for normalization , as for the cytoplasmic studies described above . The background-subtracted intensity is assumed to be proportional to Dystrophin concentration . To understand Dystrophin dynamics at the tip , and to distinguish between different bound populations , direct semi-quantitative analysis of bleaching and recovery is used . FRAP data are analysed in three ways:Direct analysis of the FRAP curve for the bleached half of the tip . Lack of recovery is strong evidence of immobility on the timescale of the experiment . Rapid , but partial , recovery , are indicative signatures of a mobile pool and an immobile pool . Comparative analysis of unnormalized intensities in the unbleached tip region and cytoplasm . Photobleaching due to imaging lowers the final intensity; this effect and a small immobile pool may be indistinguishable . To counter this problem , the identical photobleaching-due-to-imaging received by the unbleached tip region and cytoplasm is used to probe the dynamics . Unnormalized intensity plots ( background-subtracted ) allow direct comparison of the amount that is unintentionally bleached and then recovers in the tip and the cytoplasm . Presumably , the size of the cytoplasmic component in a tip pixel is at most equal to that in a cytoplasmic pixel , and its FRAP dynamics are similar . Therefore , a permanent drop in tip intensity that is much larger than the drop in the cytoplasmic intensity indicates a large immobile pool . In addition , a dip for the tip signal that is much larger than that in the cytoplasm , and that rapidly recovers , indicates that there is a dynamic bound pool at that tip . Analysis of unnormalized minus cytoplasm intensity curves evaluates if at least 50% of the final recovery occurred at the first or second time points after switching from fast to slow ( every 10 s ) acquisition rates . Analysis of the difference between bleached and unbleached tip regions . Both regions of the tip receive the same bleaching-due-to-imaging for ∼250 s . If the final difference is large , presumably there is a bound population with a bound lifetime of at least hundreds of seconds . However , if the difference is close to zero , then any bound species is dynamic on this timescale . Tests were made to evaluate whether bleaching may cause a significant portion of huDysGFP to enter a transient dark state , which then contributes to the fractional recovery after photobleaching . Typically , huDysGFP was bleached in an entire muscle fibre in vivo . Cells expressing high levels ( including in the cytoplasm ) were chosen to allow better detection of potential low levels of a dark-state pool . Bleaching was performed using the Argon laser at 100% , at which intensity we measured a direct laser power of 0 . 39–0 . 5 mW . To image the whole cell , a 20×/1 . 0 W Plan-Apochromat ( Zeiss ) objective was used . Due to the size of a muscle cell , similar tests bleaching the entire cell are not possible to perform using exactly the same conditions as in our FRAP experiments , where a 40×/0 . 8 Achroplan ( Zeiss ) objective was used . However , the laser power per area is higher in the conditions we use both in cytoplasmic and tips FRAP experiments , which is demonstrated by Mueller et al . ( 2012 ) to further decrease the effect of dark state . After whole-cell bleaching very low recovery after photobleaching is detected ( <1% ) , presumably due to a shift from dark state to excitable huDysGFP . A user-friendly application to analyse cytoplasmic diffusion along the long axis of a cell is freely available ( see instructions in Bajanca et al . , 2015 ) . The original datasets and main individual FRAP analysis files are deposited in the Dryad Digital Repository ( Bajanca et al . , 2015 ) .	A protein called Dystrophin plays a key role in maintaining the structural integrity of muscle cells as they contract and relax . Mutations in the gene that encodes Dystrophin can cause several different types of muscular dystrophy , a group of diseases in which muscle progressively weakens . Some mutations in Dystrophin can lead to mild symptoms that may affect the quality of life but are not life threatening . However , in more serious cases , patients lose the ability to walk in childhood and have shortened life expectancy . There is no cure for these diseases , and there are still big gaps in our understanding of how Dystrophin works , which makes it more difficult to develop efficient therapies . The zebrafish is often used as a model to study muscular dystrophies . In this study , Bajanca et al . introduced human Dystrophin into zebrafish muscle cells and analysed its behaviour using a combination of mathematical modelling and a method known as ‘fluorescence recovery after photobleaching’ . In these experiments , the human Dystrophin was attached to a tag that fluoresces green under a microscope , which allowed it to be easily seen and be followed in real time inside the cells of live animals . Bajanca et al . observed that Dystrophin could either remain firmly associated with the membrane that surrounds the cell over long periods of time or interact briefly with the membrane . Bajanca et al . carried out further experiments with the Dystrophin protein naturally found in zebrafish and observed that it behaved in a similar manner to the human protein , suggesting this behaviour is likely to be important for the ability of the protein to work . Bajanca et al . 's findings reveal that Dystrophin displays complex behaviour in living muscle cells . The fact that some Dystrophin molecules are firmly attached to the membrane support previous findings that this protein provides mechanical stability to the cells . However , the discovery that there is a group of more mobile Dystrophin molecules within muscle cells suggests that this protein may also play other roles . Therefore , these findings open a new avenue for research that may contribute to the development of new therapy approaches in future .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2015	In vivo dynamics of skeletal muscle Dystrophin in zebrafish embryos revealed by improved FRAP analysis
In meiosis , DNA double-strand break ( DSB ) formation by Spo11 initiates recombination and enables chromosome segregation . Numerous factors are required for Spo11 activity , and couple the DSB machinery to the development of a meiosis-specific ‘axis-tethered loop’ chromosome organisation . Through in vitro reconstitution and budding yeast genetics , we here provide architectural insight into the DSB machinery by focussing on a foundational DSB factor , Mer2 . We characterise the interaction of Mer2 with the histone reader Spp1 , and show that Mer2 directly associates with nucleosomes , likely highlighting a contribution of Mer2 to tethering DSB factors to chromatin . We reveal the biochemical basis of Mer2 association with Hop1 , a HORMA domain-containing chromosomal axis factor . Finally , we identify a conserved region within Mer2 crucial for DSB activity , and show that this region of Mer2 interacts with the DSB factor Mre11 . In combination with previous work , we establish Mer2 as a keystone of the DSB machinery by bridging key protein complexes involved in the initiation of meiotic recombination . Meiotic recombination is one of the defining features of eukaryotic sexual reproduction . In addition to creating the genetic diversity that fuels speciation and evolution , meiotic recombination fulfils a direct mechanistic role in establishing connections between initially unpaired homologous chromosomes . Meiotic recombination is initiated by programmed DNA double-strand break ( DSB ) formation by the transesterase Spo11 ( Keeney et al . , 1997 ) . Meiotic DSBs are preferentially repaired via recombination from the homologous chromosome which , depending on how recombination intermediates are processed , can yield crossovers ( reviewed in Hunter , 2015 ) . Together with sister chromatid cohesion , crossovers provide the physical linkage between homologous chromosomes which is necessary to ensure meiotic faithful chromosome segregation . In most organisms , crossover formation is associated with , and influenced by synapsis between homologs , established by the assembly of the SC . The formation of meiotic DSBs by Spo11 needs to be carefully orchestrated and controlled . In addition to Spo11 , at least 10 additional factors are required for Spo11-dependent DSB activity , and collectively these factors are referred to as the meiotic DSB machinery . Functional and biochemical analysis has begun to reveal the logic of the assembly of the DSB machinery . A picture is emerging in which several distinct subcomplexes are co-recruited into Spo11 activity proficient chromosomal foci . In addition to the core DSB machinery , several other factors promote meiotic DSB activity . For example , DSB formation occurs in the context of a distinctive chromatin loop axis architecture which is formed concomitantly with the entry of cells into the meiotic program ( Figure 1A ) . In budding yeast , this proteinaceous axis is made up of a meiosis-specific cohesin complex ( Rec8 cohesin ) in combination with the coiled-coil scaffolding protein Red1 and the HORMA domain protein Hop1 ( Smith and Roeder , 1997 ) . In cells lacking meiotic axis components , Spo11-dependent DSB formation is severely impaired ( but not completely abolished , as in DSB machinery mutants ) , and efficient recruitment of meiotic DSB factors depends on axis establishment . In addition , DSB placement and formation is influenced by histone modifications ( specifically histone H3-K4 methylation , which in budding yeast cells directs Spo11 to gene promotor regions ) . These nucleosomal interactions of the DSB machinery are proposed to occur within genomic regions which are located in the chromatin loops that emanate away from the chromosome axis ( to which the DSB machinery is tethered ) . A key component of the meiotic DSB machinery is Mer2 . This protein ( also known as Rec107 ) was originally identified as a high-copy number suppressor of the mer1 phenotype ( Mer1 was later shown to be a cofactor for the splicing of various meiotic mRNAs , including Mer2; Engebrecht et al . , 1991 ) , after which it was shown to be essential for meiosis ( Engebrecht et al . , 1990 ) . Mer2 is central to the temporal control of Spo11-dependent DSB formation , being the target of S-Cdk and DDK ( Cdc7-Dbf4 ) phosphorylation that presumably allows the binding of the Spo11-associated factors Rec114 and Mei4 to Mer2 ( Matos et al . , 2008; Murakami and Keeney , 2014; Wan et al . , 2008 ) . This regulation plays a crucial role in the spatiotemporal assembly of the DSB machinery . In addition to forming a complex with Rec114-Mei4 , Mer2 interacts directly with the PHD domain-containing protein Spp1 ( Acquaviva et al . , 2013; Sommermeyer et al . , 2013 ) . Spp1 binds to nucleosomes that are tri- ( or di- ) methylated on H3K4 ( i . e . H3K4me3 nucleosomes ) ( He et al . , 2019; Miller et al . , 2001 ) , and this association is important for the association of Spo11-dependent DSB formation with gene promoter regions . Spp1 is canonically part of the COMPASS ( aka Set1 complex ) , but during meiosis , Spp1 forms an independent , and mutually exclusive , interaction with Mer2 . The reciprocal interaction domains between Spp1 and Mer2 have been previously identified ( Acquaviva et al . , 2013; Sommermeyer et al . , 2013 ) . The C-terminal region of Spp1 interacts with a central , predicted coiled-coil , region of Mer2 ( Acquaviva et al . , 2013; Sommermeyer et al . , 2013; Figure 1B ) . A single amino acid substitution in Mer2 ( V195D ) is sufficient to disrupt the interaction with Spp1 ( as judged by yeast-2-hybrid [Y2H] analysis ) ( Adam et al . , 2018 ) . Through its interaction with Spp1 , a key role for Mer2 appears to link the Spo11 machinery directly to Spp1-mediated nucleosome interactions . Interestingly , Spp1 associated with Mer2 has a longer residence time on nucleosomes when compared with Spp1 when part of COMPASS ( Karányi et al . , 2018 ) , suggesting additional functions for Mer2 in mediating nucleosome tethering . In line with the central position for Mer2 in DSB machinery assembly is the observation that – in contrast to deletion of Set1 or Spp1 which severely reduces , but not eliminates DSB formation – mer2Δ cells completely fail to form meiotic DSBs ( Rockmill et al . , 1995 ) . Mer2 likely establishes additional biochemical interactions that enable a functional Spo11 assembly . For example , homologs of Mer2 in fission yeast ( Kariyazono et al . , 2019 ) and mouse ( Stanzione et al . , 2016 ) interact with meiotic chromosome axis-associated HORMA proteins , suggesting that Mer2 can mediate a link between the chromosome axis ( via HORMA protein interaction ) and chromatin loops ( through Spp1 association ) . Despite hints to the central position of Mer2 in assembly of DSB machinery , a more comprehensive biochemical understanding of these interactions is critically needed . Here , we use a combination of in vitro biochemical reconstitution with yeast genetics to investigate several distinct protein-protein interactions of Mer2 . We examine the interaction of Mer2 with Spp1 , nucleosomes , with proteins of the meiotic axis , and with additional members of the DSB machinery . Our results report a more complete picture of Mer2 as a foundational component of the meiotic DSB machinery , including novel functions , and provide mechanistic explanations for a number of previously observed phenomena revolving around the regulation of meiotic DSB formation . We first focussed on the described interaction between Mer2 and Spp1 . In order to probe the various possible functions of Mer2 , we made use of four principal expression constructs , the full-length protein ( residues 1–314 from hereon abbreviated to ‘Mer2FL’ ) , Mer2 amino acids 1–256 , lacking the C-terminal 58 residues ( from hereon ‘Mer2∆C’ ) , Mer2 residues 140–314 ( i . e . lacking the N-terminal 139 residues; from hereon ‘Mer2∆N’ ) and Mer2 containing residues 140–256 ( from hereon referred to as ‘Mer2core’ ) and a putative coiled-coil domain ( Figure 1B ) . Previous work has identified the Spp1 interaction region to be contained within Mer2core ( specifically Mer2 residues 165–232; Acquaviva et al . , 2013 ) . Using our in-house expression system , ‘InteBac’ ( Altmannova et al . , 2021 ) , we produced full-length Spp1 and all Mer2 proteins in Escherichia coli with N-terminal MBP tags to facilitate protein solubility . We could successfully remove the MBP tag using the 3C protease from both Spp1 and Mer2 , though in the case of Mer2∆N and Mer2Core the 3C cleavable MBP tag could not be removed , presumably due to steric hindrance by the MBP tag that precludes efficient cleavage . In co-lysis experiments , we found that we could purify a complex of Mer2 and Spp1 to homogeneity , and free of nucleic acid contamination ( Mer2FL with Spp1 shown as an example in Figure 1C , note the apparent A260 to A280 ratio as evidence of a lack of nucleic acid contamination ) , thus indicating that the interaction between Mer2 and Spp1 does not require any PTMs or additional cofactors and that the interaction was robust enough to survive extensive co-purification in >300 mM NaCl . Using microscale thermophoresis ( MST ) we measured the binding affinity of Spp1 to Mer2FL ( Figure 1D orange trace , squares ) and Spp1 to Mer2Core ( Figure 1D green trace , diamonds ) . Spp1 bound Mer2FL with a KD of 24 nM ( ±2 ) , and to Mer2Core with a KD of 137 nM ( ±4 ) . Mer2 constructs lacking the ‘core’ showed comparatively weak binding ( Figure 1—figure supplement 1 ) . Thus , we confirm that the majority of the Spp1 binding interface is indeed within the core of Mer2 , as reported earlier ( Acquaviva et al . , 2013; Adam et al . , 2018 ) , although there does appear to be some contribution to Spp1 binding provided by the N- and C-terminal regions of Mer2 . We next tested the reciprocal interaction using a C-terminally 2xStrep-II-tagged Mer2FL against full-length Spp1 , and two additional Spp1 constructs , Spp1∆C , containing amino acids 1–170 , and Spp1∆PHD containing amino acids 169–353 . We found that in a Strep-Mer2FL pulldown on Streptactin beads Spp1FL and Spp1∆PHD interacted with Mer2 , but Spp1∆C did not , consistent with previous studies ( Figure 1—figure supplement 2 ) . We measured the molecular mass of Mer2 by size exclusion chromatography coupled to multi-angle light scattering ( SEC-MALS ) and concluded that Mer2core is the tetramerisation region ( Figure 1E , orange trace and Figure 1—figure supplement 3A-D ) , consistent with recent observations of Mer2 ( Claeys Bouuaert et al . , 2021 ) . Interestingly , while Mer2∆C∆core is monomeric ( Figure 1—figure supplement 3D ) , Mer2∆N∆core is dimeric ( Figure 1—figure supplement 3C ) indicating the presence of a dimerisation region in the C-terminal region of Mer2 between residues 255 and 314 , which presumably aids in the stability of a full coiled-coil tetramer . Given that the tetramerisation region of Mer2 is also the principal Spp1 binding region ( Figure 1D ) , we determined the stoichiometry of the Mer2-Spp1 complex . First , we determined that full-length Spp1 alone is a monomer ( Figure 1E yellow trace , Figure 1—figure supplement 3E-F ) . We next analysed the stoichiometry of Mer2:Spp1 complexes . We measured the size of a complex of MBP-Mer2core with Spp1 ( Figure 1E green trace ) and determined its mass to be 290 kDa . The theoretical mass of a 4:2 ( MBP-Mer2core:Spp1 ) complex is 310 kDa , whereas a 4:1 ( MBP-Mer2core:Spp1 ) complex is 268 kDa . Given the possible ambiguity in determination of stoichiometry , we also measured complexes of MBP-Mer2FL and untagged Mer2FL together with Spp1 ( Figure 1—figure supplement 3G and H ) which gave complex sizes best fitting a 4:2 ( Mer2:Spp1 ) stoichiometry . Taken together we conclude that the Mer2 tetramer binds two copies of Spp1 , establishing a complex in a 4:2 stoichiometry . Thus we show a novel function for Mer2 in not simply binding Spp1 , but importantly , in mediating the dimerisation of Spp1 . In light of the inherent 2-fold symmetry of nucleosomes , we suggest that this 4:2 constellation might aid in the recognition of modified nucleosome tails by Spp1 . Next we probed the structural organisation of Mer2-Spp1 further using cross-linking coupled to mass spectrometry ( XL-MS ) using the 11 Å spacer crosslinker disuccinimidyl dibutyric urea ( DSBU ) ( Figure 1F ) . XL-MS of Mer2-Spp1 revealed that , while the ‘core’ of Mer2 showed many cross-links with Spp1 , these were , unexpectedly , not with the previously described C-terminal interaction domain ‘Mer2-ID’ of Spp1 ( Acquaviva et al . , 2013 ) . Instead the Mer2 core showed numerous cross-links with a region of Spp1 immediately C-terminal to the PHD domain . Furthermore there were additional cross-links between Spp1 and the N- and C-terminal regions of Mer2 , consistent with the residual binding affinity we observed in MST ( Figure 1—figure supplement 1 ) . The intramolecular cross-linked pattern of Mer2 alone ( Figure 1—figure supplement 4A ) was very similar in the presence and absence of Spp1 . As such we can possibly exclude a significant structural rearrangement of Mer2 upon association with Spp1 . We also compared the cross-linking pattern observed previously for Mer2 alone ( Claeys Bouuaert et al . , 2021; Figure 1—figure supplement 4B ) . This revealed that the pattern was broadly similar with a mixture of long- and short-distance cross-links . One striking difference is the extensive cross-links emanating from the N-term of our Mer2 . The most likely explanation is that the overhang remaining on our Mer2 preparation after removal of the N-terminal fusion protein is four amino acids longer , and thus more flexible . In order to study the role of the Mer2-Spp1 complex binding to H3K4me3 nucleosomes , we created synthetic H3K4me3 mononucleosomes . Briefly , we mutated K4 of histone H3 to cysteine whereas the single naturally occurring cysteine of the natural H3 sequence was mutated to alanine ( C110A ) . H3C4 was converted to H3K4me3 using a trimethylysine analogue as previously described ( Simon et al . , 2007 ) . We then reconstituted H3K4me3 into octamers , and subsequently into mononucleosomes using 167 bp Widom sequences ( see Materials and methods for further details ) . Due to the dimeric nature of nucleosomes , and because our reconstitutions showed a 4:2 Mer2:Spp1 complex stoichiometry , we hypothesised that dimerisation of Spp1 might lead to more tight binding to H3K4me3 nucleosomes , and set out to test this idea . We compared the apparent nucleosome binding affinity of Spp1 , GST-tagged Spp1 ( which mediates dimerisation ) , and the Mer2-Spp1 complex using both electrophoretic mobility shift assays ( EMSAs ) ( Figure 2A ) and pulldowns on biotinylated nucleosomes ( Figure 2B ) . We observed that the Mer2-Spp1 assembly ( in which two copies of Spp1 are present ) bound more tightly to nucleosomes , as compared to monomeric Spp1 alone ( which bound relatively weakly to nucleosomes , consistent with the reported ~1 µM affinity of the PHD domain with H3K4me3 peptide; He et al . , 2019 ) . Interestingly , when we compared the observed binding of Spp1:Mer2 with GST-Spp1 , we found that GST-Spp1 exhibited an intermediate apparent binding affinity . A potential corollary of this observation is that Mer2 in addition to triggering Spp1 ‘dimerisation’ might directly contribute to nucleosome binding . We established that the Mer2-Spp1 complex was capable of forming a stable complex with H3K4me3 nucleosomes in solution using analytical SEC ( Figure 2C ) . Note that we do not observe a complete shift of nucleosomes; we suspect that this is due to not having an optimal buffer condition ( in this experiment we have tried to balance the buffer conditions required for Mer2 [high salt] , Spp1 [Zn2+ ions for the PHD domain] , and nucleosomes [EDTA] ) . As our complex is currently not suitable for high-resolution structural studies , we made use again of XL-MS to determine a topological architecture of the Mer2-Spp1-H3K4me3 mononucleosome complex ( Figure 2D ) . We detected many more internal Spp1 cross-links than observed in the Mer2-Spp1 complex alone ( Figure 1F ) , suggesting that either the binding to nucleosomes brings the two Spp1 moieties ( in the 2:4 complex ) closer to one another , or that there is an internal rearrangement of domains of Spp1 . Most strikingly however , the cross-linking revealed that most of the cross-links between the Mer2-Spp1 complex and the nucleosomes are via regions of Mer2 , in the N-term , core , and C-term regions . This observation strengthened the idea that Mer2 might directly contribute to nucleosome binding . We modelled the location of the Mer2-Spp1 cross-links onto the previously determined structure of a mononucleosome ( Davey et al . , 2002; Figure 2E ) . We observe that the cross-links cluster around histone H3 , but more generally around the DNA entry/exit site on the nucleosomes . These observations suggest a large Mer2:nucleosome interface , and a considerably smaller Spp1:nucleosome interface . Our pulldown data , SEC experiments , and XL-MS data all suggested that Mer2 might bind to mononucleosomes directly , perhaps providing additional affinity to the Spp1:nucleosome interaction . If true , this would be a previously unreported function of Mer2 . We tested whether Mer2FL could bind to unmodified mononucleosomes in an SEC experiment , and surprisingly found that it formed a stable complex ( Figure 3A ) . Given that Mer2 is a tetramer that binds two copies of Spp1 , and given that Mer2 has been previously shown to form large assemblies on DNA ( Claeys Bouuaert et al . , 2021 ) , we asked what the stoichiometry of a Mer2-mononucleosome complex was . To do this , we first used mass photometry ( MP ) , a technique that determines molecular mass in solution at low concentrations based on the intensity of scattered light on a solid surface ( Young et al . , 2018 ) . In MP , we observe a mix of three species , free Mer2 tetramer ( measured at 127 kDa; theoretical mass 142 kDa ) , free mononucleosomes ( measured at 187 kDa; theoretical mass 202 kDa ) , and a complex at 303 kDa , which most likely corresponds to a 4:1 complex of Mer2:nucleosomes ( theoretical mass 344 kDa ) ( Figure 3B ) . The MP experiment was carried out at a protein concentration of 60 nM , which suggests that the dissociation constant ( KD ) is somewhat less than 60 nM ( at KD under equilibrium one would observe 50% complex formation , we observe less than 50% in Figure 3B ) . We next asked whether Mer2 might form larger assemblies on nucleosomes as higher concentrations . Using SEC-MALS ( Figure 3C ) , we observed a complex of 341 . 0 kDa which matches a theoretical complex consisting of one Mer2 tetramer plus one mononucleosome ( 340 kDa , summarised in Figure 3D ) . Additionally , we observe a small fraction of a very high molecular weight assembly ( though not an aggregate ) of 10 . 96 MDa . This could be an oligomer of Mer2 , of Mer2 on nucleosomes , or Mer2 on free DNA ( it has recently been reported that Mer2 binds directly to DNA; Claeys Bouuaert et al . , 2021 ) . We also observe a shoulder on the Mer2-nucleosome peak , and find that this is a mixture of molecular masses ( Figure 3—figure supplement 1A ) . Given that we observe Mer2 cross-links at the nucleosome DNA entry/exit site , we therefore tested whether Mer2 might simply be binding the free DNA ends on mononucleosomes . Using analytical EMSAs we found that Mer2 binds with a 6-fold higher apparent affinity to nucleosomes ( 5 nM vs . 30 nM ) than to the same 167 bp DNA used to reconstitute the nucleosomes ( Figure 3E ) . The discrepancy between KD determined by EMSA and the apparent KD from MP is presumably because EMSAs are non-equilibrium experiments , carried out by necessity at very low salt ( Fried and Bromberg , 1997 ) . Next we asked what effect using a smaller length of DNA to reconstitute nucleosomes might have . We reconstituted histone octamers on 147 bp DNA from now on referred to as nucleosome core particles ( NCP; Lohr and Van Holde , 1975; Sollner-Webb et al . , 1976 ) . We found that with no free DNA ends Mer2 did bind with a lower affinity to NCPs , but nonetheless still with an apparent KD of ~40 nM ( Figure 3—figure supplement 2A ) . We then asked whether mutating the common binding site , the ‘acidic patch’ on H2A ( E56T-E61T-E64T-D90S-E91T-E92T ) ( Kalashnikova et al . , 2013 ) might have an effect on Mer2 binding . In order to enhance potential differences in binding , this was also done on NCPs . Mer2 bound to NCP acidic patch mutants essentially as well as wildtype NCPs ( Figure 3—figure supplement 2A ) . We then asked whether Mer2 might be recognising the histone tails . We therefore prepared ‘tailless’ NCPs ( see Materials and methods ) . Surprisingly Mer2 bound very tightly to tailless NCPs with an apparent KD of ~5 nM ( Figure 3—figure supplement 2A ) . We suggest that this might indicate that under the conditions of an EMSA , the histone tails are shielding either the histone cores or the NCP DNA and interfering with binding by Mer2 . We next asked which region of Mer2 might be involved in binding nucleosomes ( reconstituted with 167 bp DNA ) . Initially we carried out EMSAs with Mer2core , Mer2∆N , and Mer2∆C . Mer2∆C and Mer2∆N appeared to bind slightly weaker than Mer2FL ( KD ~12 . 5 and ~30 nM , respectively ) , whereas Mer2core showed no binding at all ( Figure 3—figure supplement 2B ) . Apparently there was equal contribution to nucleosome binding from both termini . In order to refine this further , we carried streptavidin pulldown using mononucleosomes reconstituted with biotinylated DNA against different Mer2 constructs . This approach had the advantage of being able to use a more physiological , and as such more stringent , buffer . We confirmed that the Mer2core did not bind nucleosomes , but neither did Mer2∆C , strongly suggesting that the main nucleosome interaction region of Mer2 lies in the C-terminal 58 amino acids ( Figure 3F ) , with some additional contribution from the N-terminus of Mer2 ( summarised in Figure 3G ) . We propose that Mer2 , in addition to enabling the ‘dimerisation’ of Spp1 via its central tetramerisation domain , provides a direct binding interface with nucleosomes . In addition to Spp1 , several lines of evidence point to an association between Mer2 and meiotic HORMA domain-containing factors . In fission yeast and mouse , the functional homologs of Mer2 ( named Rec15 and IHO1 , respectively ) have been shown to interact with Hop1/HORMAD1 ( Kariyazono et al . , 2019; Stanzione et al . , 2016 ) . Likewise , in budding yeast Mer2 exhibits a chromatin-association pattern which is very similar to Hop1 ( Panizza et al . , 2011 ) . Meiotic HORMA proteins are integral members of the meiotic chromosome axis and are needed to recruit Mer2 to chromosomes ( Panizza et al . , 2011; Stanzione et al . , 2016 ) . Hop1 ( like most other known HORMA domains ) can exist in two topological states ( ‘open/unbuckled’ [O/U] or ‘closed’ [C] ) , in which the closed state can embrace a binding partner via a closed HORMA-closure motif ‘safety belt’ binding architecture ( West et al . , 2018 ) . The closure motif ( also referred to as CM ) is a loosely conserved peptide sequence encoded in HORMA binding partners . The meiotic HORMA proteins are unique among HORMA proteins in the fact that they contain a CM at the end of their own C-terminus ( Hop1 residues 585–605; West et al . , 2018 ) , endowing these factors with the ability to form an intramolecular ( closed ) HORMA-CM configuration . The association of Hop1 with chromosomes is mediated by an interaction with Red1 which depends upon a similar CM:HORMA-based interaction . The CM of Red1 ( located 340–362; West et al . , 2018 ) binds to Hop1 with a higher affinity than the CM of Hop1 ( West et al . , 2018 ) . There is mounting evidence in budding yeast and Arabidopsis , that , in addition to a chromosomal pool , a significant pool of Hop1 is non-chromosomal ( in the nucleoplasm or cytoplasm ) ( Herruzo et al . , 2021; Raina and Vader , 2020; Yang et al . , 2020 ) . Once bound to its own CM , Hop1 should not be able to interact with its chromosomal axis binding partner Red1 . Due to the high local concentration of the intramolecular CM , free Hop1 is expected to rapidly transition from the ‘open/unbucked’ into the intramolecular ‘closed’ state ( West et al . , 2018 ) . The closed state ( whether it is intramolecular or , for example , with the CM of Red1 ) can be reversed by the action of the AAA+ ATPase Pch2/TRIP13 ( Rosenberg and Corbett , 2015; Vader , 2015 ) . As such , within the nucleoplasm/cytoplasm , Pch2/TRIP13 activity serves to generate enough ‘open/unbuckled’ Hop1 that is proficient for incorporation into the chromosomal axis , via a CM-based interaction with Red1 . On the other hand , when recruited to chromosomes , that same Pch2/TRIP13 activity is expected to dismantle Hop1-Red1 assemblies , as such leading to removal of Hop1 from chromosomes ( Deshong et al . , 2014; Subramanian et al . , 2016 ) . We tested the ability of Mer2 to interact with the proteins of the meiotic axis ( i . e . Hop1 and Red1 ) , with a focus on Hop1 . Initially , we purified Hop1 using an N-terminal 2x-Strep-II tag and used it to pull down Mer2 . We observed a faint band corresponding to Mer2 in the Hop1 pulldown , indicative of a weak interaction ( Figure 4—figure supplement 1 , lane 1 , control in lane 2 ) . Note that based on the high relative concentration of the CM in Hop1 , this Hop1 is expected to largely consist of ( intramolecular ) closed Hop1 . We then co-expressed Red1-MBP containing a I743R ( from hereon referred to as Red1I743R-MBP ) mutation with Hop1 in insect cells . The I743R mutation should prevent Red1 from forming filaments , but still allow Red1 to form tetramers ( West et al . , 2019 ) . Given that we have an excess of Hop1 in our Hop1-Red1 purification , we carried out a pulldown on the MBP tag of Red1 ( using amylose beads ) with Mer2 as prey . In this case , we observed considerably more Mer2 binding , when measured relative to Hop1 , its putative direct binding partner ( Figure 4—figure supplement 1 , lane 3 , control in lane 4 ) . We quantitated the Mer2 intensity relative to the Hop1 band in both pulldowns , and from three independent experiments , and observed an ~6-fold increase in Mer2 binding ( Figure 4A ) . We reasoned that this difference could either be due to Red1 interacting directly with Mer2 , or that Red1 induces a conformational change in Hop1 that facilitates Mer2 binding . To test the former idea we purified Red1I743R-MBP in the presence or absence of Strep-Hop1 . Using amylose affinity beads to capture the MBP moiety of Red1 we tested the capture of Mer2 both in the presence and in the absence of Hop1 ( Figure 4B ) . We detected no Mer2 interaction when pulling on Red1I743R-MBP in the absence of Hop1 . As such , we conclude that Mer2 does not have significant affinity for Red1 . These data argue that Hop1 , when bound to Red1 , is acting as an efficient recruiter of Mer2 to this complex . How could this increased affinity of Hop1 for Mer2 be influenced by association with Red1 ? Based on the known biochemical basis of the Hop1-Red1 interaction , we can imagine that the interaction of Red1 with Hop1 ‘releases’ the C-terminus of Hop1 which could create a ‘chain’ of Hop1 moieties ( akin to what has been observed in Caenorhabditis elegans; Kim et al . , 2014 ) or could simply liberate the C-terminus of Hop1 for binding to non-self partners . We propose that such configurations would create or expose Mer2-specific binding interfaces ( which can conceivably be located on either the HORMA domain or within the C-terminal non-HORMA domain ) that are otherwise shielded when Hop1 is bound intramolecularly to its own C-terminal CM . This predicts that impairing intramolecular CM binding to the HORMA domain should ‘unlock’ the binding ability of Hop1 with Mer2 , regardless of Red1 presence . To test this prediction and to delineate the binding interface between Mer2-Hop1 , we created two additional Hop1 constructs , one where the N-terminal HORMA domain is missing ( Hop1∆HORMA ) and another where the conserved lysine in the CM has been mutated to alanine ( Hop1K593A ) disrupting the ability of the CM of Hop1 to interact with its HORMA domain thus forcing Hop1K593A into an ‘unlocked’ state ( West et al . , 2018 ) . Both Hop1∆HORMA and Hop1K593A were purified with an N-terminal 2xStrep-II tag ( as for Hop1WT ) and their ability to bind to Mer2 was tested . In order to prevent background binding , stringent binding conditions were used . We saw that both Hop1WT and Hop1K593A bind to Mer2FL , whereas Hop1∆HORMA does not ( Figure 4C ) . Quantification of the pulldown analyses showed that Hop1K593A has ~2-fold stronger binding to Mer2 than does Hop1WT ( Figure 4D ) . Based on these data , and taking into consideration the effect of Red1 on the Hop1-Mer2 interaction , we suggest the following model: Mer2 binding to Hop1 is driven largely by HORMA domain-mediated interactions , and promoted under conditions where the HORMA is not bound to its intramolecular CM . The most parsimonious molecular explanation of this behavior is that the C-terminal region of Hop1 , when kept in close proximity of the HORMA domain ( by virtue of CM-HORMA binding ) , can sterically interfere with the establishment of a Mer2-Hop1 association . Furthermore , the robust Mer2 binding we observe for the Hop1-Red1 complex ( Figure 4B ) suggests that there is additional contribution from Red1 , likely through promoting the closed conformation of the HORMA domain through Red1-CM-HORMA domain binding . We next asked which region of Mer2 is necessary for Hop1 interaction . Using N-terminally 2xStrepII tagged Hop1K593A as bait , we queried a variety of Mer2 constructs using Streptactin beads ( Figure 4E ) . Due to the weak staining of Mer2 under these conditions , we also carried out an anti-Mer2 Western blot ( Figure 4E lower panel ) . We determined that Hop1 was capable of interacting , apparently equally well , with all Mer2 constructs , except for Mer2∆C . Why would the Mer2core be capable of binding to Hop1 , but the ∆C not ? The answer may lie in the complex arrangement of the Mer2 coiled-coils and thus N- and C-termini of Mer2 relative to one another and to the core of Mer2 . As such one could imagine that part of the interaction region for Hop1 lies in the core of Mer2 , which might be shielded by the N-terminus when the C-terminus of Mer2 is not present . These results hint at a complex regulation , on the Mer2 side , underlying Mer2-Hop1 interaction , and we believe that high-resolution structural studies of the Mer2-Hop1 interface should eventually be able to provide deeper insight into this interaction . Taken together we propose a new model for Mer2 recruitment to the meiotic axis in which a ‘locked’ Hop1 cannot bind to Mer2 but once incorporated into Red1 , Hop1 recruits Mer2 to the meiotic axis . We sought to further characterise the interaction between Mer2 and Hop1 by again making use of XL-MS and the DSBU cross-linker ( Figure 4F ) . We observed the most cross-links between the core of Mer2 ( consistent with the binding of Mer2core to Hop1 , Figure 4E ) and both the HORMA domain of Hop1 and the C-terminal part of the protein ( few cross-links were detected within the putative PHD/wHTH like region of Hop1 , located in the non-HORMA domain C-terminal part of Hop1; Ur and Corbett , 2021 ) . We also observed additional cross-links with the C-terminal half of Mer2 , again consistent with the pulldown data ( Figure 4E ) . Using a model generated using open-Mad2 as a template ( Luo et al . , 2000 ) , we mapped Mer2-Hop1 cross-links onto the HORMA/SB domain of Hop1 ( Figure 4G ) . This revealed that multiple Mer2 cross-links were concentrated on one ‘face’ of the HORMA domain , whereas the other ‘face’ of the HORMA domain is essentially free of Mer2-Hop1 cross-links . We observed several cross-links in close proximity to the safety belt of the HORMA domain of Hop1 ( Figure 4G ) . While we assume that the Hop1K593A used in the cross-linking is mostly ‘unlocked’ , we acknowledge that the safety belt likely has a somewhat different orientation in ‘unlocked’ Hop1 vs . ‘open’ Mad2 ( West et al . , 2018 ) . Taken together these data reveal additional molecular details of the Mer2-Hop1 binding mode . The concentration of Mer2 cross-links on one face of Hop1 hints at a potential role for Red1 in enforcing a particular Hop1 orientation that is compatible with efficient recruitment of Mer2 . Due to the potential difficulties in assigning defined interaction regions within Mer2 when using truncation constructs , as so far described , we instead aimed to obtain separation of function alleles by introducing selected point mutants . Making use of sequence alignments from ( evolutionary closely and more distantly related ) Mer2 orthologs , we identified a conserved region , in the N-terminal domain between amino acids 52 and 71 ( Figure 5A , Figure 5—figure supplement 1 ) , that has also been previously annotated as Mer2 SSM1 ( Tessé et al . , 2017 ) . This particular stretch of amino acids stands out in the protein sequence of Mer2 ( and homologs ) since , in addition to the central coiled-coil region , this region is one of the few regions which shows sequence similarities across evolutionary distant species ( such as yeast and human ) . To probe a potential function of this conserved region , we created two different alleles containing mutations in this region , which we here refer to as mer2-3a and mer2-4a . In mer2-3a , we mutated three conserved residues W58 , K61 , and L64 to alanine ( Figure 5A , red stars ) . The significance here is both in the high level of conservation , and , in the case of W58 , in the presence of a bulky hydrophobic residue that might be involved in protein-protein interactions . Finally , the periodicity of the conserved residues combined with the secondary structure prediction hinted that these residues might be on the same face of an alpha-helical element ( see Figure 5—figure supplement 1 ) . mer2-4a contains the following mutations within the same domain: D52A , E68A , R70A , and E71A ( Figure 4A , yellow stars ) . We note that the residues in mer2-4a are less well conserved than those in the mer2-3a allele , but we wondered whether they might function together or in separate pathways . We integrated plasmids carrying C-terminally 3HA-tagged versions of wildtype MER2 , mer2-3a , or mer2-4a ( note that all these constructs are driven by pMER2 ) in mer2∆ strains . All three constructs lead to comparable expression levels of Mer2 during meiotic prophase , although we note different mobility of the Mer23A-3HA or Mer24A-3HA as compared to Mer2-3HA , which might indicate altered post-translational modifications , such as phosphorylation or SUMOylation , since Mer2 was recently shown to be heavily SUMOylated including sites in the region covered by the 3A and 4A mutations ( Mer2 K53 and K61; Bhagwat et al . , 2021; Figure 5B ) . Alternatively , it might reflect an inherent effect of the introduced mutations on electrophoretic behavior ( compare for example also the migrating patterns of wildtype Mer2 with Mer23A on SDS-PAGE from our in vitro preparations , where meiosis-specific post-translational modifications are most likely absent; see Figure 6B ) . The role of Mer2 in meiotic DSB activity is key in enabling faithful meiotic chromosome segregation and viable spore formation . Consequently , mer2Δ strains exhibit very strong spore viability defects , as reported previously ( Rockmill et al . , 1995 Figure 5C ) . As a first test of functionality , we investigated the effect of our MER2 alleles on spore viability . Expression of Mer2-3HA in mer2Δ overcame the spore viability defect seen in cells lacking Mer2 ( Figure 5C , blue bar ) demonstrating functionality of our MER2 expressing constructs . Strikingly , expression of Mer23A-3HA or Mer24A-3HA failed to restore spore viability in mer2Δ; these strains essentially behaved indistinguishably from the mer2Δ strain . This suggests that the designed mutants disrupt a functionality of Mer2 that is key to its role during meiotic prophase . When we probed for the activation of Mek1 kinase , which is regulated by DSB-dependent Mec1/Tel1 activation ( Chuang et al . , 2012; Niu et al . , 2005 ) , we noticed that expression of the Mer23A/4A mutants ( in contrast to wildtype Mer2 ) prevented the appearance of the phosphorylated version of histone H3-threonine 11 , a bona-fide Mek1 substrate ( Kniewel et al . , 2017; Figure 5B ) . Together with the fact that expression of Mer23A/4A mutants did not support the phosphorylation of Hop1 ( mediated by Mec1/Tel1 as a response to DSBs ) ( Carballo et al . , 2008; Figure 5B; as seen by an apparent electrophoretic shift in Hop1 ) , these data show that the expression of our Mer2 mutants leads to defective Mec1/Tel1-Mek1 signaling . This hints that DSB formation might be negatively affected by the mutations that we introduced in Mer2 . Since the investigated region of Mer2 is located within the N-terminal region of the protein , we expected that these mutants would not disrupt the interactions with Spp1 , Hop1 , and nucleosomes described above . In line with this idea , we observed that in vitro , recombinant Mer23A and Mer24A are able to interact with nucleosomes ( Figure 3F ) and Hop1 ( Figure 4E ) , albeit with slightly lower affinity for nucleosomes as determined by EMSA ( Figure 3—figure supplement 2C ) . To attempt to trace the defect of these MER2 alleles , we first investigated the association of Mer2 on spread meiotic chromosomes during meiotic prophase . We found that the recruitment of Mer2 to defined chromosomal foci was not disrupted in Mer23A and Mer24A expressing situations ( Figure 5D ) . In contrast , quantification of the number of Mer2 foci that were observed during meiotic prophase revealed an increase in cells expressing these mutant proteins ( Figure 5E ) . We currently do not understand the reason underlying this apparent increase in chromosome-associated Mer2 foci , but it could conceivably be related to known feedback regulation that ensures sustained DSB activity in conditions where chromosome synapsis fails or is delayed ( Thacker et al . , 2014 ) . In any case , recruitment of Mer2 to chromatin-associated foci was not disrupted , suggesting that Mer23A and Mer24A are proficient to interact with upstream factors that lead to chromatin recruitment . In order to identify a potential protein-protein interaction that is disrupted in the Mer2 mutants , we carried out a quantitative mass spectrometry analysis of samples immunoprecipitated from meiotic mer2∆ SK1 yeast expressing Mer24A-3HA or Mer2WT-3HA . Wildtype ( carrying an untagged MER2 allele ) SK1 cells were used as a control . In order to prevent possible confounding effects due to potential differences in DSB forming potential ( based on the effect on spore viability and the decrease in phospho-histone H3-T11 signal drop we observed in our mutants , Figure 5B and C ) , we performed the analysis in a DSB-deficient condition ( by using the catalytic-dead spo11-Y135F allele ) in both Mer24A-3HA and Mer2WT-3HA . The summary of the differences between Mer2WT and Mer24A are shown in the volcano plot in Figure 5F . In summary , the biggest change we see in the Mer24A mutant vs . Mer2WT is a reduced binding to Hht1 ( histone H3 ) . We also saw a reduced binding to the other histones Htb1 and Htb2 ( histone H2B ) . Interaction with Hop1 was largely unaffected ( Figure 5G ) – in agreement with our in vitro analysis ( Figure 4E ) , as was association with Spp1 and Red1; the latter presumably associated via association with Hop1 ( Figure 5—figure supplement 2A and B ) . Taken together , it appears from these analyses that Mer24A exhibits reduced direct nucleosome association . Nonetheless , the effect is likely mild , since we see in vitro association is at most reduced 5-fold in EMSAs for Mer23A ( Figure 3—figure supplement 2C ) , and both Mer23A and Mer24A robustly interact with nucleosomes in a pulldown ( Figure 3F ) , and Mer2 still localises to chromatin ( Figure 5D and E ) . What might be the cause of the penetrant phenotype seen in Mer23A and Mer24A ? Since HORMA proteins are suggested to drive recruitment of Mer2 to the chromosome axis ( Kariyazono et al . , 2019; Panizza et al . , 2011; Stanzione et al . , 2016; and see our earlier observations ) , these results argue that the defect that is triggered by mer2-3a and mer2-4a is independent of its interaction with Hop1 . Furthermore , the observation that spore viability was severely impaired in cells expressing mer2-3a and mer2-4a is in contrast with the relatively mild phenotype caused by the specific disruption of the Mer2-Spp1 interaction ( Adam et al . , 2018 ) . Finally , our Western blots analysis revealed a loss of Mek1-dependent histone H3-pT11 phosphorylation ( a proxy for meiotic DSB formation ) in the Mer23A/4A mutants ( Figure 5B , and see above ) . Together , these observations , in combination with the strong evolutionary conservation of the affected region , argue that the observed defects are caused by the disruption of yet another key Mer2 interaction , potentially with another essential DSB factor , leading to possible defects in DSB formation . Strikingly , we detected no Spo11-associated machinery in the Mer2WT IP ( Figure 5—figure supplement 2C and Supplementary mass spectrometry data ) , despite a large total number of enriched proteins . We therefore postulated that we might be missing key Mer2 interactions – which are potentially difficult to detect using in vivo biochemical purification ( ostensibly due to transient/low affinity interactions ) . The disruption of those interactions , as a result of the Mer2 mutants , might explain observed in vivo phenotypes . As such , we carried out a directed Y2H analysis against previously identified Mer2 interactors involved in DSB formation ( Xrs2 , Mre11 , Ski8 , Spo11 , Rec104 ) ( Arora et al . , 2004 ) and Spp1 ( as a positive control ) . Analysis of some of these interactions was complicated by the presence of apparent self-activation in our Y2H system . Nonetheless , we observed a clear difference in binding between the Mer2 and Mre11 when comparing MER2 to the mer2-3a or mer2-4a mutants ( Figure 6A , Figure 6—figure supplement 1 ) . Importantly , these mutants were proficient to interact with Spp1 , similar to what was indicated by our in vitro binding assays and in accordance with the mass spectrometry data ( Figure 5—figure supplement 2A ) . We aimed to confirm the interaction between Mer2 and Mre11 by other means , and sought to query whether the interaction with other DSB factors was not perturbed in Mer23A/4A mutants . We initially used in vivo co-immunoprecipitation ( co-IP ) to test for Mer2 interaction with Mre11 and Rec114 ( Figure 6—figure supplement 2A and B ) . Using this setup , we were not able to detect Mre11 nor Rec114 , although we reliably detected Hop1 in Mer2WT , Mer23A , and Mer24A co-IPs , indicating efficient purification of a relevant assembly . The apparent discrepancy between our Y2H and co-IP experiments might reflect an inherently transient or infrequently occurring interaction between Mer2 and Mre11 ( and Mer2 and Rec114 ) , precluding detection using our current in vivo methodology . We note that the results obtained with our co-IP experiments ( i . e . no detected interaction with Mer2 or Rec114 , but robust interaction with Hop1 ) were similar to the results we obtained using our IP-MS approach comparing Mer2WT and Mer24A ( Figure 5F and G ) . Therefore , we sought to corroborate an interaction between Mer2 and Mre11 using purified components . To this end , we expressed and purified recombinant Mre11 carrying a C-terminal 2xStrep-II tag using baculovirus-based expression . We then used this as a bait to pull down Mer2WT , Mer23A and Mer24A . We detected a robust interaction between Mer2WT and Mre11 , and in line with our earlier result , we observed a reduced interaction between Mre11 and Mer2 when Mer23A and Mer24A were used in the pulldown assay ( Figure 6B ) . We quantified the amount of Mer2 in the pulldown as a factor of Mre11 ( Figure 6C ) : Mer23A leads to a 2-fold reduction in binding , and Mer24A a reduction closer to 10-fold . So , despite the fact that the effects we observed in pulldown between recombinant Mre11 and Mer23A and Mer24A were not as striking the ones we detected in our Y2H analysis , we were able to confirm that Mer2 directly associates with Mre11 , and that our mutant Mer2 proteins exhibit reduced Mre11 association . DSB factors are recruited to chromosomes during meiotic prophase ( Panizza et al . , 2011 ) , and we were interested in addressing whether the introduction of Mer2 mutants affected the chromosomal association of Mre11 and Rec114 . We carried out immunofluorescence on meiotic chromosome spreads and analysed the signal corresponding to Rec114 or Mre11 ( both carrying a COOH-terminal 13XMYC tag ) in cells expressing the different MER2 alleles ( Figure 6—figure supplement 3A and B ) . We did not detect any obvious differences in the levels of chromosomally associated Rec114 or Mre11 in cells expressing wildtype or mutant versions of Mer2 . As such , while the interaction between Mer2 and Mre11 might be disrupted in the mutants , the ability of Mre11 ( and Rec114 ) to localise to chromosomes during meiosis appeared unaffected . We next were interested to shed more light on the interaction between Mer2 and Mre11 , and to this end , we carried out XL-MS analysis on a complex of Mer2 and Mre11 . These data revealed extensive cross-links across the length of Mre11 and Mer2 , with the exception of the C-terminal region of Mer2 ( Figure 6D ) . We paid particular attention to those cross-links that emanate from the residues that are in proximity of the Mer23A/4A mutant region ( amino acids 52–72 ) ( Figure 6D , pale orange ) and noted that these cross-links are predominantly formed with a region at the very C-terminus of Mre11 . This region is predicted to be unstructured , but we noted that this region harbors a high-confidence SUMO interaction motif ( SIM; residues 633–637 ) based on the SUMO-GPS prediction algorithm ( Zhao et al . , 2014 ) . Intriguingly , a recent study on the role SUMOylation in meiosis ( Bhagwat et al . , 2021 ) reported Mer2 as a SUMO-target , and found that Mer2K61 ( which is one of the residues mutated in Mer23A ) is directly SUMOylated . By comparing the location of the SIM ( Figure 6D , green bar ) with the Mer2-Mre11 cross-links , we noted that the SIM sits immediately adjacent to the most cross-linked region of Mre11 . Taken together , we consider it a distinct possibility that the interaction between Mer2 and Mre11 might be influenced ( and if so , likely enhanced ) by SUMOylation of Mer2 during meiotic prophase . Our earlier analysis of the phenotypes exhibited by cells expressing our Mer2 mutants , together with the fact that the MRX complex ( including Mre11 specifically; Johzuka and Ogawa , 1995 ) is generally required for meiotic DSB formation ( reviewed in Borde , 2007 ) , prompted us to investigate meiotic DSB formation in mer2-3a and mer2-4a mutants . We used Southern blotting to track meiotic DSB formation at YCR047C , a confirmed DSB hotspot . ( Note that we utilised the sae2Δ resection and repair-deficient background in order to enable DSB detection . ) We detected a strong impairment of meiotic DSB activity specifically in mer2-3a or mer2-4a , to an extent that was comparable to what was observed in mer2Δ strains ( Figure 6E ) . Thus , Mer23A and Mer24A disrupt a key functionality of Mer2 that is required for meiotic DSB formation . Based on this and our earlier data , we propose that , in addition to its centrally positioned role in mediating chromosome axis and chromatin loop-tethering of the DSB machinery , a third key contribution that Mer2 ( and its conserved N-terminal region ) makes to enable DSB activity is establishing an interaction Mre11 . While we cannot comprehensively rule out an alternative mechanism , our data suggest that not only the presence of Mre11 is required , but its specific interaction with Mer2 is essential for the initiation of DSB formation during meiotic prophase . We have biochemically dissected the function of Mer2 in vitro to reveal novel features that have the potential to explain several in vivo characteristics of Spo11-dependent DSB formation . Firstly , we shed light on the interaction between Spp1 and Mer2 . The interaction between Spp1 and Mer2 is tight ( ~25 nM ) and not dependent on any post-translational modification or additional cofactors . As such , the interaction between Spp1 and Mer2 is likely constitutive in vivo . Importantly , we also find that Mer2 serves as a dimerisation platform for Spp1 , effectively increasing its affinity for H3K4me3 nucleosomes . Presumably this is essential given that the interaction between COMPASS bound Spp1 and H3K4me3 is transient , whereas the association of the DSB forming machinery is an apparently more stable event ( Karányi et al . , 2018 ) . Given that the Spp1 interaction domain of Mer2 is the same as the tetramerisation domain , we speculate that the antiparallel arrangement of coiled-coils of Mer2core ( Claeys Bouuaert et al . , 2021 ) form two oppositely oriented binding sites for Spp1 , with V195 positioned at the centre ( Adam et al . , 2018; Claeys Bouuaert et al . , 2021 ) . We find that Mer2 itself is a bona-fide nucleosome binder . This interaction occurs at high affinity , though we assume that the true affinity is somewhat less than what we determine from EMSA titrations . Given that Mer2 has been previously shown to bind DNA ( Claeys Bouuaert et al . , 2021 ) , yet still binds to an NCP ( with 147 bp DNA and no DNA overhangs ) , and that neither the loss of histone tails nor the acidic patch mutant disrupted the interaction , we suggest that Mer2 binds to nucleosomal DNA . This idea is supported by the XL-MS data on the Mer2-Spp1 complex bound to nucleosomes , which places Mer2 proximal to the DNA entry/exit site ( Figure 2D and E ) . The tight and specific nucleosome binding ability of Mer2 provides a molecular basis for the observation that neither Spp1- nor Set1-mediated H3K4me3 are absolutely required to make meiotic DSBs ( Acquaviva et al . , 2013; Sommermeyer et al . , 2013 ) , unlike Mer2 itself ( Rockmill et al . , 1995 ) . As such we speculate that in the absence of H3K4me3 marks , or its reader Spp1 , Mer2 binds stochastically to nucleosomes that are positioned in chromatin loops . Such a model speculates that some of these binding events present a nucleosome-depleted loop region to Spo11 , but many do not , also explaining why meiotic DSBs are severely reduced in number in an spp1∆ or set1∆ background ( Acquaviva et al . , 2013; Sommermeyer et al . , 2013 ) . On the other hand , if Mer2 preferentially bound free DNA , as opposed to nucleosomal DNA , one might reasonably expect a less severe DSB phenotype in spp1∆ or set1∆ cells . The association between nucleosomes , Spp1 and Mer2 , is important to establish the connection between chromosome axis-associated DSB factors and DSB sites that are localised in the loop ( Acquaviva et al . , 2013; Sommermeyer et al . , 2013 ) . We speculate that the combination of ‘generic’ nucleosome binding ( via Mer2-nucleosomal DNA ) and specific histone tail recognition ( Spp1-PHD domain-H3 tail ) endows the DSB machinery with the required binding strength and specificity . Our observation and characterisation of the direct interaction between Hop1 and Mer2 offers a tantalising glimpse into the mechanism by which axial proteins may be recruiting DSB factors – through Mer2 – in order to regulate Spo11 activity . A key observation is that we only observed binding between Mer2 and Hop1WT in the presence of Red1 ( Figure 4A and B ) . We could exclude any significant direct binding to Red1 ( Figure 4B ) , which is consistent with the observations in mammals and fission yeast , where Hop1 orthologs have been implicated as the direct interactor of Mer2 orthologs ( Kariyazono et al . , 2019; Stanzione et al . , 2016 ) – though we here , for the first time , demonstrate a direct interaction in vitro between purified components . We found that the use of a constitutively ‘unlocked’ Hop1 ( Hop1K593A ) leads to an increase in Mer2 binding , whereas the isolated C-terminus ( Hop1∆HORMA ) does not show Mer2 interaction . These data clearly indicate a role for the HORMA domain in Mer2 association , but by which mechanism ? Given that Hop1WT is in equilibrium between ‘locked’ and ‘unlocked’ as it binds and unbinds its own CM , one explanation could simply be that if Mer2 is incubated with Hop1WT for longer time periods in vitro then eventually a complex will form , with presumably unlocked Hop1 . In vivo unlocking is likely promoted by the interaction of the Red1 CM with the HORMA of Hop1 . However , our in vitro pulldown experiments point towards an additional effect of Red1 on the Hop1-Mer2 binding . How might Red1 influence this binding ? We did not detect direct binding to Red1 in a pulldown experiment . We therefore hypothesise several possibilities by which Red1 contributes to Mer2 binding to Hop1 . The most simple explanation is that the CM of Red1 both unlocks Hop1 , but also catalyses the closed conformation of the HORMA domain ( West et al . , 2018 ) . Additionally , the oligomerisation effect of Red1 on Hop1 ( Red1 is a tetramer; West et al . , 2019 ) could promote Mer2 interaction by avidity . Alternatively , Red1 could ( allosterically ) influence the position of the HORMA domain . Our observation that the binding of Mer2 only apparently occurs with one ‘face’ of the HORMA domain suggests that the other ‘face’ might be directly oriented towards Red1 . As such , Red1 might act as a conformational and stoichiometric enhancer of Mer2 binding , recruiting multiple Hop1 HORMAs and orienting them in three-dimensional constellation so that they are positioned favourably for interaction with Mer2 . Intriguingly , in fission yeast , the zinc finger of Hop1 ( located in region corresponding with residues 348–364 within the C-terminal region of Saccharomyces cerevisiae Hop1 ) has been suggested to be required for Mer2 binding ( Kariyazono et al . , 2019 ) . While we show that this region alone ( which is now thought to be a PHD domain; Ur and Corbett , 2021 ) is not sufficient for Mer2 binding , it could nonetheless be involved in enhancing binding to Mer2 . Regardless , our observations provide mechanistic insights into the specific recruitment of Mer2 to meiotic chromosomes . For example , our model might explain how Mer2 can specifically be recruited to meiotic chromosomes , and not form spurious interactions with non-chromosomal Hop1: non-chromosomal , monomeric Hop1 is thought to be largely present in the intramolecular ‘closed’ form ( unless it is converted into the ‘open/unbuckled’ state by Pch2/TRIP13 , which is thought to promote rapid chromosomal incorporation of Hop1 ) ( Cardoso da Silva and Vader , 2021; Raina and Vader , 2020 ) . Conversely , it might explain how DSB activity is negatively regulated by chromosome synapsis . Synapsis leads to recruitment of Pch2/TRIP13 and removal of Hop1 from the chromosomal axis ( presumably by opening up Hop1 bound to Red1 ) . This would be associated with immediate co-removal of Mer2 , and once released Hop1 would transition into intramolecular Hop1 , leading to disruption of the Mer2-Hop1 complex ( Chen et al . , 2014; Vader , 2015; West et al . , 2018; Raina and Vader , 2020; Subramanian et al . , 2016; Yang et al . , 2020 ) . Finally , we attempted to create separation of function mutants for Mer2 by mutating a previously undescribed conserved region in the N-terminus of the protein . Both mutants in this region ( Mer23A and Mer24A ) were penetrant: they exhibited a loss of spore viability similar to cells lacking the Mer2 protein entirely ( Figure 5C ) . Nonetheless , these mutants maintained an interaction with Hop1 in vitro and in vivo , and – in agreement with the proposed role of the Mer2-Hop1 interaction – exhibited normal localisation to the meiotic axis in vivo . In addition , although quantitative EMSAs showed that there was a reduction in affinity of Mer23A for nucleosomes ( Figure 3—figure supplement 2C ) , under physiological salt conditions ( i . e . in the pulldown in Figure 3F ) both Mer2 mutants could still form a complex with nucleosomes . We discovered that our mutants interfere with an interaction between Mer2 and Mre11 . We note that the observed severity of the effect of the Mer23A and Mer24A on the Mer2-Mre11 interaction was somewhat variable: Y2H analysis indicated a strong disruption of the interaction when using Mer23A or Mer24A compared to Mer2 , and while we also observed interaction between Mer2 and Mre11 using in vitro purified proteins , and that the mutants reduced the interaction between Mer2 and Mre11 , the difference between the mutants and wildtype was less pronounced . We infer that a factor/condition that strengthens the interaction between Mer2 and Mre11 is present in vegetatively growing yeast cells ( i . e . the condition of Y2H analysis ) , but is lacking in our in vitro pulldown . This could be an additional protein factor , perhaps one of the other components of the Mre11-Rad50-Xrs2 complex , or a post-translational modification . For example , Mer2 has been previously shown to be phosphorylated by DDK and S-Cdk ( Henderson et al . , 2006; Murakami and Keeney , 2014 ) , and Mer23A or Mer24A could potentially affect these phosphorylation events . Moreover , Mre11 has been previously shown to contain two SIMs . Since many meiotic proteins have been recently shown to be SUMOylated in yeast , among them Mer2 , SUMOylation might serve as an important regulator ( Bhagwat et al . , 2021 ) . Importantly , the residues that are mutated in Mer23A are essentially universally conserved throughout the eukaryotic kingdom ( Figure 5—figure supplement 1 ) . As such , we surmise that the direct interaction between Mer2 and Mre11 , via these residues , is a universal feature of the meiotic program . Intriguingly , the Mer23A and Mer24A mutants do not disrupt the chromosomal localisation of Mre11 during meiosis ( Figure 6—figure supplement 3B ) . As such this suggests that rather than a simple disruption of localisation , the Mer2 mutants might be disrupting an allosteric effect of Mer2 on Mre11 . Alternatively , the efficient co-localisation of Mre11 and Mer2 into DSB foci might be disrupted ( as might the co-localisation of additional cofactors ) . We note that several other DSB factors have been shown via Y2H to interact with Mer2 ( Arora et al . , 2004 ) . An important future research goal should be to comprehensively biochemically anaylse the interactions between Mer2 and the DSB machinery , also in light of our findings here . Taken together our data show that Mer2 forms the keystone of meiotic recombination , binding directly to the axis – via Hop1- , and to the loop – via nucleosomes . Presumably , once assembled on the loop axis , Mer2 establishes interactions with Rec114 and Mei4 in phospho-dependent manner ( likely via the PH domains of Rec114; Boekhout et al . , 2019; Li et al . , 2006; Panizza et al . , 2011 ) , an association that may also be somehow further controlled by liquid-liquid phase separation ( Claeys Bouuaert et al . , 2021 ) . Simultaneously , our data indicate that the N-terminus of Mer2 recruits the MRX complex via Mre11 , a step that we suggest is critical for the formation of meiotic DSBs and the successful completion of meiosis . In organisms with a synaptonemal complex ( SC ) , downregulation of meiotic DSB formation is associated with chromosome synapsis . It has been shown that synapsis results in the Pch2TRIP13-mediated displacement of Hop1 from the axis ( Börner et al . , 2008; Chen et al . , 2014; Joshi et al . , 2009; Lambing et al . , 2015; San-Segundo and Roeder , 1999; Subramanian et al . , 2016; Wojtasz et al . , 2009 ) . Based on our work , we posit that this action would also result in the displacement of the crucial DSB factor Mer2 from the axis ( as removed Hop1 would no longer be bound to Red1 , thus weakening its affinity for Mer2 ) . Additionally , once released , Hop1 would be expected to re-bind its own CM , thus ‘snapping shut’ and preventing re-binding to Red1 , and also presumably preventing re-recruitment of Mer2 . Together , these effects provide a potential molecular rationale for the functional connection between chromosome synapsis and reduced DSB activity . Our findings highlight the power of biochemical reconstitution in dissecting the function of complex biological systems . Mer2 emerges as the keystone of meiotic recombination , establishing interactions with the chromosome axis , nucleosomes , and the DSB machinery ( Figure 6—figure supplement 4 ) . Larger and more ambitious reconstitutions will enable us to probe the role of additional protein cofactors and post-translational modifications in meiotic regulation . Sequences of S . cerevisiae SPP1 , HOP1 , RED1 , and MRE11 were derived from SK1 strain genomic DNA . Due to the presence of an intron in MER2 , this was amplified as two separate fragments and Gibson assembled together . All Mer2 constructs were expressed as an 3C HRV cleavable N-terminal MBP fusion in chemically competent C41 E . coli cells . Protein expression was induced by addition of 250 µM IPTG and the expression continued at 18°C overnight . Cells were washed with 1× PBS and resuspended in lysis buffer ( 50 mM HEPES pH 7 . 5 , 300 mM NaCl , 5% glycerol , 0 . 1% Triton-X 100 , 1 mM MgCl2 , 5 mM β-mercaptoethanol ) . Resuspended cells were lysed using an EmulsiFlex C3 ( Avestin ) in the presence of DNAse ( 10 μg/mL ) and AEBSF ( 25 μg/mL ) before clearance at 20 , 000 g at 4°C for 30 min . Cleared lysate was applied on a 5 mL MBP-trap column ( GE Healthcare ) and extensively washed with lysis buffer . Mer2 constructs were eluted with a lysis buffer containing 1 mM maltose and passed through a 6 mL ResourceQ column ( GE Healthcare ) equilibrated in 50 mM HEPES pH 7 . 5 , 100 mM NaCl , 5% glycerol , 5 mM β-mercaptoethanol . The proteins were eluted by increasing salt gradient to 1 M NaCl . Protein containing elution fractions were concentrated on Amicon concentrator ( 100 kDa MWCO ) and loaded a Superose 6 16/600 ( GE Healthcare ) pre-equilibrated in SEC buffer ( 50 mM HEPES pH 7 . 5 , 300 mM NaCl , 10% glycerol , 1 mM TCEP ) . Untagged Mer2FL was prepared likewise until concentration of protein eluted from ResourceQ . The concentrated eluent was mixed with 3C HRV protease in a molar ratio of 50:1 and incubated at 4°C for 6 hr . Afterwards , the cleaved protein was loaded on a Superose 6 16/600 pre-equilibrated in SEC buffer for cleaved Mer2 ( 20 mM HEPES pH 7 . 5 , 500 mM NaCl , 10% glycerol , 1 mM TCEP , 1 mM EDTA , AEBSF ) . Spp1 constructs were produced as an 3C HRV cleavable N-terminal MBP or GST fusion in a similar manner as MBP-Mer2 . To purify GST-Spp1 , cleared lysate was applied on GST-Trap ( GE Healthcare ) before extensive washing with lysis buffer . The protein was eluted with a lysis buffer with 40 mM reduced glutathione and passed through ResourceQ . Both GST and MBP could be cleaved by adding 3C HRV protease to concentrated protein ( using an Amicon concentrator with 30 kDa cutoff ) in 1:50 molar ratio . After an ~6 hr incubation at 4°C , the cleaved protein was loaded on Superdex 200 16/600 pre-equilibrated in SEC buffer ( 50 mM HEPES pH 7 . 5 , 300 mM NaCl , 10% glycerol , 1 mM TCEP ) . Hop1 constructs were produced as 3C HRV cleavable N-terminal Twin-StrepII tag in BL21 STAR E . coli cells . The expression was induced by addition of 250 µM IPTG and the expression continued at 18°C for 16 hr . Cleared lysate was applied on Strep-Tactin Superflow Cartridge ( Qiagen ) before extensive washing in lysis buffer . The bound protein was eluted with a lysis buffer containing 2 . 5 mM desthiobiotin and loaded on HiTrap Heparin HP column ( GE Healthcare ) and subsequently eluted with increasing salt gradient to 1 M NaCl . Eluted Strep-Hop1 constructs were concentrated on a 30 kDa MWCO Amicon concentrator and loaded on Superdex 200 16/600 pre-equilibrated in SEC buffer . Red1 was produced in insect cells as a C-terminal MBP-fusion either alone or co-expressed with Strep-Hop1 . Bacmids were in both cases produced in EmBacY cells and subsequently used to transfect Sf9 cells to produce baculovirus . Amplified baculovirus was used to infect Sf9 cells in 1:100 dilution prior to 72 hr cultivation and harvest . Cells were extensively washed and resuspended in Red1 lysis buffer ( 50 mM HEPES pH 7 . 5 , 300 mM NaCl , 10% glycerol , 1 mM MgCl2 , 5 mM β-mercaptoethanol , 0 . 1% Triton-100 ) . Resuspended cells were lysed by sonication in the presence of Benzonase and a protease inhibitor cocktail ( Serva ) before clearance at 40 , 000 g at 4°C for 1 hr . Cleared lysate was loaded on Strep-Tactin Superflow Cartridge ( in case of Red1-Hop1 complex ) or MBP-trap column ( in case of Red1 alone ) . Proteins were eluted using a lysis buffer containing 2 . 5 mM desthiobiotin and 1 mM maltose , respectively . Partially purified proteins were further passed through HiTrap Heparin HP column and eluted with increasing salt gradient to 1 M NaCl . Purified proteins were subsequently concentrated using Pierce concentrator with 30 kDa cutoff in 50 mM HEPES pH 7 . 5 , 300 mM NaCl , 10% glycerol , 1 mM TCEP . Because of the small yield of the proteins , the SEC purification step was neglected and the purity of the proteins was checked using the Refeyn One mass photometer . Mre11 was produced as a C-terminal Twin-StrepII tag in insect cells using the same expression conditions as for Red1 protein . The cell pellet was resuspended in Mre11 lysis buffer ( 50 mM HEPES pH 7 . 5 , 300 mM NaCl , 5% glycerol , 0 . 01% NP40 , 5 mM β-mercaptoethanol , AEBSF , Serva inhibitors ) . Resuspended cells were lysed by sonication before clearance at 40 , 000 g at 4°C for 1 hr . Cleared lysate was loaded on a 5 mL Strep-Tactin XT Superflow Cartridge ( IBA ) followed by first wash using 25 mL of high-salt wash buffer ( 20 mM HEPES pH 7 . 5 , 500 mM NaCl , 5% glycerol , 0 . 01% NP40 , 1 mM β-mercaptoethanol ) and second wash step using 25 mL of low-salt wash buffer ( 20 mM HEPES pH 7 . 5 , 150 mM NaCl , 5% glycerol , 0 . 01% NP40 , 1 mM β-mercaptoethanol ) . The Mre11 protein was eluted with 50 mL of low-salt wash buffer containing 50 mM biotin . Partially purified protein was further loaded onto a 5 mL Heparin column ( GE Healthcare ) pre-equilibrated in a low-salt wash buffer and eluted with increasing salt gradient to 1 M NaCl . The fractions containing Mre11 protein were concentrated on a 50 kDa MWCO Amicon concentrator and applied onto a Superdex 200 10/300 column ( GE Healthcare ) pre-equilibrated in Mre11 SEC buffer ( 20 mM HEPES pH 7 . 5 , 300 mM NaCl , 5% glycerol , 1 mM β-mercaptoethanol , 1 mM TCEP ) . Fifty µL samples at 5–10 µM concentration were loaded onto a Superose 6 5/150 analytical size exclusion column ( GE Healthcare ) equilibrated in buffer containing 50 mM HEPES pH 7 . 5 , 1 mM TCEP , 300 mM NaCl ( for samples without nucleosomes ) or 150 mM NaCl ( for samples with nucleosomes ) attached to an 1260 Infinity II LC System ( Agilent ) . MALS was carried out using a Wyatt DAWN detector attached in line with the size exclusion column . Triplicates of MST analysis were performed in 50 mM HEPES pH 7 . 5 , 300 mM NaCl , 5% glycerol , 1 mM TCEP , 0 . 005% Tween-20 at 20°C . The final reaction included 20 μM RED-NHS labelled untagged Spp1 ( labelling was performed as in manufacturer’s protocol – Nanotemper ) and titration series of MBP-Mer2 constructs ( concentrations calculated based on oligomerisation stage of Mer2 ) . The final curves were automatically fitted in Nanotemper analysis software . Recombinant Xenopus laevis histones were purchased from ‘The Histone Source’ ( Colorado State ) with the exception of H3-C110A_K4C cloned into pET3 , which was kindly gifted by Francesca Matirolli . The trimethylated H3 in C110A background was prepared as previously described ( Simon et al . , 2007 ) . X . laevis histone expression , purification , octamer refolding , and mononucleosome reconstitution were performed as described ( Luger et al . , 1999 ) . Plasmids for the production of 601–147 ( pUC19 ) and 601–167 ( pUC18 ) DNA were kindly gifted by Francesca Matirolli ( Hubrecht Institute , Utrecht ) and Andrea Musacchio ( MPI Dortmund ) , respectively . DNA production was performed as previously described ( Luger et al . , 1999 ) . Reconstituted mononucleosomes were shifted to 20 mM Tris pH 7 . 5 , 150 mM NaCl , 1 mM EDTA , 1 mM TCEP with addition of 20% glycerol prior to freezing in –80°C . Quadruplicate EMSAs were carried out as previously described ( Weir et al . , 2016 ) , at a constant nucleosome/NCP/DNA concentration of 10 nM with the DNA being post-stained with SYBRGold ( Invitrogen ) . Gels were imaged using a ChemiDocMP ( Bio-Rad Inc ) . Nucleosome depletion in each lane was quantitated by ImageJ , using measurements of triplicate of the nucleosome alone for each individual gel as a baseline . Binding curves were fitted using Prism software and the following algorithm ( Y = Bmax*Xh/ ( KDh + Xh ) ) . It was necessary in each Mer2 case to add a Hill coefficient to obtain the best fit . Streptactin pulldowns were performed using pre-blocked streptavidin magnetic beads ( Pierce ) in a pulldown buffer ( 20 mM HEPES pH 7 . 5 , 300 mM NaCl , 5% glycerol , 1 mM TCEP ) . One µM Strep-Hop1 or Mre11-Strep as a bait was incubated with 3 µM Mer2 as a prey in 40 µL reaction for 2 hr on ice without beads and another 30 min after addition of 10 µL of beads pre-blocked with 1 mg/mL BSA for 2 hr . After incubation , the beads were washed twice with 200 µL of buffer before elution of the proteins with a 1× Laemmli buffer . Samples were loaded on 10% SDS-PAGE gel and afterwards stained with InstantBlue . Amylose pulldowns were performed using pre-blocked Amylose beads ( New England BioLabs ) in a pulldown buffer ( 20 mM HEPES pH 7 . 5 , 300 mM NaCl , 5% glycerol , 1 mM TCEP ) . One µM RedI743R-MBP or RedI743R-MBP/Strep-Hop1 as a bait was incubated with 3 µM Mer2 as a prey in 40 µL reaction for 2 hr on ice without beads and another 1 hr after addition of 10 µL of beads pre-blocked with 1 mg/mL BSA for 2 hr . After incubation , the beads were washed twice with 200 µL of buffer before elution of the proteins with a buffer containing 1 mM maltose . Samples were loaded on 10% SDS-PAGE gel and afterwards stained with InstantBlue . Biotinylated nucleosomes ( 0 . 5 μM ) or NCP ( 0 . 4 µM ) were incubated with prey proteins ( 1 . 5 μM ) for 30 min on ice in buffer containing 20 mM HEPES pH 7 . 5 , 150 mM NaCl , 5% glycerol , 1 mM EDTA , 0 . 05% Triton-X100 , 1 mM TCEP in a reaction volume of 40 µL . Ten µL of protein mix were taken as an input before adding 10 µL of pre-equilibrated magnetic Dynabeads M 270 streptavidin beads ( Thermo Fisher Scientific ) to the reaction . The samples with beads were incubated on ice for 2 min before applying magnet and removing the supernatant . The beads were washed twice with 200 µL of buffer . To release the streptavidin from the beads , Laemmli buffer ( 1× ) was added to the beads and incubated for 10 min . Samples were analysed on 10–20% SDS-PAGE gel and stained by InstantBlue . Analytical SEC was performed using Superose 6 5/150 GL column ( GE Healthcare ) in a buffer containing 20 mM HEPES pH 7 . 5 , 150 mM NaCl , 5% glycerol , 1 mM TCEP , 1 mM EDTA . All samples were eluted under isocratic elution at a flow rate of 0 . 15 mL/min . Protein elution was monitored at 280 nm . Fractions were subsequently analysed by SDS-PAGE and InstantBlue staining . To detect complex formation , proteins were mixed at 5 µM concentration in 50 µL and incubated on ice for 1 hr prior to SEC analysis . MP was performed in 20 mM HEPES pH 7 . 5 , 150 mM NaCl , 5% glycerol , 1 mM TCEP , 1 mM EDTA . Mer2 and mononucleosomes ( 600 nM ) were mixed and incubated for 1 hr on ice prior to analysis using the Refeyn One mass photometer . Immediately before analysis , the sample was diluted 1:10 with the aforementioned buffer . Molecular mass was determined in Analysis software provided by the manufacturer using a NativeMark- ( Invitrogen ) based standard curve created under the identical buffer composition . For XL-MS analysis proteins were dissolved in 200 μL of 30 mM HEPES pH 7 . 5 , 1 mM TCEP , 300 mM NaCl ( for samples without nucleosomes ) or 150 mM NaCl ( for samples with nucleosomes ) to final concentration of 3 μM , mixed with 3 μL of DSBU ( 200 mM ) and incubated for 1 hr in 25°C . The reaction was quenched by addition of 20 μL of Tris pH 8 . 0 ( 1 M ) and incubated for another 30 min in 25°C . The crosslinked sample was precipitated by addition of 4× volumes of 100% cold acetone ON in –20°C and subsequently analysed as previously described ( Pan et al . , 2018 ) . All strains , except those used for Y2H analysis , are of the SK1 background . See Supplementary material for a description of genotypes of strains used per experiment . For MER2 alleles , constructs containing pMER2 ( −1000–1 ) , the coding sequence of MER2 lacking its intron ( i . e . wildtype , or 3a/4a-containing sequences ) , C-terminal 3HA tag , and 500 base pairs of downstream sequence flanked by HindIII and NarI restriction enzymes were custom-synthesised by Genewiz Inc These constructs were recloned in a YIPlac128 plasmid carrying LEU2 using restriction cloning . These plasmids were integrated at pMER2 in front of mer2::HISMX6 following EcoRI linearisation . Correct single copy integration was confirmed by PCR . Yeast strains were patched onto YP-Glycerol plates and transferred to YP-4%Dextrose plates . After this , cells were grown overnight in liquid YPD culture ( room temperature ) followed by inoculation in pre-sporulation media ( BYTA; 50 mM Sodium Phthalate-buffered , 1% yeast extract , 2% tryptone , and 1% acetate ) at OD600 = 0 . 3 . Cells were grown for 18 hr in BYTA at 30°C , washed twice in water and resuspended in sporulation media ( 0 . 3% potassium acetate ) at OD600 = 1 . 9 to induce meiosis at 30°C . Cells were synchronously induced into meiosis and incubated for 24 hr . Of each strain , 200 cells were counted using a standard bright-field microscope , and monads , dyads , and tetrads were scored . For viability , the indicated number of tetrads was dissected using standard manipulation methods and grown on YPD plates . Spore viability was calculated as a percentage of the total number of viable spores . Two mL of meiotic cells ( t = 3 hr induction ) were collected at indicated time points , killed by addition of 1% sodium azide and processed . Cells were treated with 200 mM Tris pH 7 . 5 , 20 mM dithiothreitol for 2 min at room temperature followed by spheroplasting at 30°C in 1 M sorbitol , 2% potassium acetate , 0 . 13 µg/µL zymolyase ( 20 min ) . Spheroplasts were washed two times in 1 mL ice-cold MES-Sorbitol solution ( 1 M sorbitol , 0 . 1 M MES pH 6 . 4 , 1 mM EDTA , 0 . 5 mM MgCl2 ) and resuspended in 55 µL of MES-Sorbitol . Twenty µL of spheroplasts were placed on clean glass slides ( that were dipped in ethanol overnight and air-dried ) and 2× volumes of fixing solution ( 3% paraformaldehyde , 3 . 4% sucrose ) were added . This was followed by addition of four volumes of 1% Lipsol , and mixing through gentle rotation . After 1 min , 4× volumes of fixing solution were added . A glass rod was used to mechanically spread chromosomes , after which samples were dried overnight at room temperature , and stored at −20°C . Slides were treated with 0 . 4% Photoflo ( Kodak ) /PBS for 3 min , after which slides were dipped in PBS with gentle shaking ( 5 min ) . Samples were blocked by incubation with 5% BSA in PBS for 15 min at room temperature . Overnight incubation with desired primary antibodies was performed in a humidified chamber at 4°C , after which slides were subjected to 2× washes of 10 min in PBS with gentle shaking followed by incubation with fluorescent-conjugated secondary antibody for 3 hr at room temperature . The slides were washed 2× and mounted using 20 µL of Vectashield mounting solution containing 4’ , 6-diamidine-2’-phenylindole dihydrochloride ( DAPI ) ( Vector Laboratories ) . Chromosome surface spreads were immunostained with rat α-HA at 1:200 ( Roche ) , mouse α-MYC ( 9E10 ) at 1:200 , and rabbit α-Hop1 at 1:200 ( home made ) . Hop1 antibody production was performed at the antibody facility of the Max-Planck-Institute of Molecular Cell Biology and Genetics ( Dresden , Germany ) using affinity purified full length 6xHis-tagged Hop1 . Secondary antibodies were used at the following concentrations: Alexa 488-conjugated donkey α-rat at 1:500; Texas Red 594-conjugated donkey α-rabbit at 1:500 , and Cy3-conjugated donkey α-mouse at 1:500 . Image acquisition was done by obtaining serial z-stacks of 0 . 2 μm thickness at room temperature using 100 × 1 . 42 NA PlanApo-N objective ( Olympus ) on a DeltaVision imaging system ( GE Healthcare ) equipped with an sCMOS camera ( PCO Edge 5 . 5 ) . The z-stack images were deconvolved using SoftWoRx software . Quantifications for the number of foci of Mer2WT-3HA , Mer23A-3HA Mer24A-HA were done using the ‘Spots’ function of the Imaris software ( Bitplane ) . Prism 8 ( GraphPad ) was used to generate the Scatter plots . Statistical significance was assessed by performing Mann-Whitney U-test . For representative images , Fiji/ImageJ software was used to obtain maximum intensity projection images . One-hundred mL of meiotic cultures ( at 4 hr into a meiotic time course ) were harvested by spinning down at 3000 rpm for 3 min . Samples were washed with cold H2O and snap frozen . To the pelleted cells , the following was added: 300 μL of ice-cold IP buffer ( consisting of 50 mM Tris–HCl pH 7 . 5 , 150 mM NaCl , 1% Triton X-100 and 1 mM EDTA pH 8 . 0 , and a cocktail of protease inhibitors which was freshly added ) and acid-washed glass beads . A FastPrep-24 disruptor ( MP Biomedicals ) was used to break the cells ( MP Biomedicals ) ( setting: two 50 s cycles ( speed 6 ) ) . Lysates were spun 30 s at 500 rpm , after which the supernatant was transferred to a 15 mL falcon tube . The resulting lysate was sonicated for 25 cycles ( 30 s on/30 s off , high power range ) using a Bioruptor-Plus sonicator ( Diagenode ) , and subsequently spun down 30 min at maximum speed at 4°C . The supernatant was transferred into new microcentrifuge tubes , and 10% of the supernatant ( ~50 μL ) was collected as input . One μL of antibody ( ⍺-HA; BioLegend ) was added , and the lysate was rotated for 3 hr at 4°C; 30 μL of buffer-washed Dynabeads protein G ( Invitrogen , Thermo Fisher Scientific ) was subsequently added , and the lysate was rotated overnight at 4°C . The next day , Dynabeads were washed four times with 500 μL of ice-cold IP buffer . For the final wash , beads were transferred to a new microcentrifuge tube . After removal of the supernatant , the beads were resuspended in 40 μL IP buffer , and 20 μL of 3XSDS-loading buffer was added . Samples were incubated for 5 min at 95°C . For the inputs , the following protocol was used: Inputs were precipitated by the addition of 5 μL 100% trichloroacetic acid ( TCA ) ( 10% final concentration ) , and incubated on ice for 30 min . Precipitates were collected by centrifugation , for 1 min at maximum speed , and washed with ice-cold acetone . Precipitations were dried and resuspended in resuspension buffer ( 40 μL , 50 mM Tris-HCl 7 . 5 , and 6 M urea ) for 30 min ( on ice ) . Dissolution was encouraged by pipetting and vortexing . Ten μL of 5XSDS-loading buffer was added , and samples were incubated for 5 min at 95°C . For Western blot analysis , protein lysates from yeast meiotic cultures were prepared using TCA precipitation and run on 8% or 10% SDS gels , transferred for 90 min at 300 mA and blotted with the selected antibodies , as described ( Kuhl et al . , 2020 ) . Primary antibodies with respective dilutions were used: rabbit α-Hop1 ( made in-house; 1:10 , 000 ) , rabbit α-Mer240-271 ( made in-house; 1:10 , 000 ) ; mouse α-Pgk1 ( Thermo Fisher , 1:5000 ) ; rabbit α-phospho-Histone-H3-Thr11 ( Abcam , 1:1000 ) , mouse α-HA ( Biolegend , diluted 1:500 ) , α-Myc-9E11 ( Abcam , ab56 , 1:1000 ) . For Southern blot assay , DNA from meiotic samples was prepared as described ( Vader et al . , 2011 ) . DNA was digested with HindIII ( to detect DSBs at the control YCR047C hotspot ) followed by gel electrophoresis , blotting of the membranes and radioactive ( 32P ) hybridisation using a probe for YCR047C ( chromosome III; 209 , 361–201 , 030 ) ( Raina and Vader , 2020 ) . DSBs signals were monitored by exposure of an X-ray film which was analysed using a Typhoon Trio scanner ( GE Healthcare ) after 1 week . MER2 variants and their potential interactors were cloned into pGAD-C1 or pGBDU-C1 vectors , respectively . The resulting plasmids were co-transformed into the S . cerevisiae reporter strain ( yWL365 ) and plated onto the selective medium lacking leucine and uracil . For drop assay , 2 . 5 μL from 10-fold serial dilutions of cell cultures with the initial optical density ( OD600 ) of 0 . 5 were spotted onto -Leu/-Ura ( control ) and -Leu/-Ura/-His plates . Cells were grown at 30°C for up to 7 days and imaged at time points indicated in the figures . Cells ( 450 mL ) were harvested after 3 hr of synchronised meiosis . Three independent cultures were prepared for each strain . Pellets were lysed in cryo-mill and the resulting powder was resuspended in 25 mL of lysis buffer ( 50 mM HEPES pH 7 . 5 , 300 mM NaCl , 5% glycerol , 0 . 01% NP40 , 5 mM β-mercaptoethanol , AEBSF , Serva protease inhibitors , 1 mM NEM , 1× Complete Mini EDTA-Free protease inhibitors; Roche ) . The lysate was cleared by centrifugation at 10 , 000 rpm for 1 hr . The supernatant was incubated with 5 μL of α-HA antibody ( Sigma-Aldrich , H6908 ) for 3 hr at 4°C followed by the addition of 25 μL magnetic Dynabeads Protein G ( Invitrogen , 10 , 004D ) pre-equilibrated with wash buffer ( 20 mM HEPES pH 7 . 5 , 100 mM NaCl , 5% glycerol , 0 . 01% NP40 , 1 mM β-mercaptoethanol ) . The samples were incubated overnight at 4°C . Next day , the samples were centrifuged at 2500 rpm for 1 min at 4°C and the beads were washed six times with 150 μL of 20 mM ammonium bicarbonate . Proteins were eluted with 75 μL of 2× SDS Laemmli buffer and boiled for 5 min at 95°C . The samples were analysed by Proteomics Core Facility ( EMBL , Heidelberg ) .	Organisms are said to be diploid when they carry two copies of each chromosome in their cells , one from each of their biological parents . But in order for each parent to only pass on one copy of their own chromosomes , they need to make haploid cells , which only carry one copy of each chromosome . These cells form by a special kind of cell division called meiosis , in which the two chromosomes from each pair in the parent cells are first linked , and then pulled apart into the daughter cells . Accurate meiosis requires a type of DNA damage called double-stranded DNA breaks . These breaks cut through the chromosomes and can be dangerous to the cell if they are not repaired correctly . During meiosis , a set of proteins gather around the chromosomes to ensure the cuts happen in the right place and to repair the damage . One of these proteins is called Mer2 . Previous studies suggest that this protein plays a role in placing the DNA breaks and controlling when they happen . To find out more , Rousova et al . examined Mer2 and the proteins that interact with it in budding yeast cells . This involved taking the proteins out of the cell to get a closer look . The experiments showed that Mer2 sticks directly to the chromosomes and acts as a tether for other proteins . It collaborates with two partners , called Hop1 and Mre11 , to make sure that DNA breaks happen safely . These proteins detect the state of the chromosome and repair the damage . Stopping Mer2 from interacting with Mre11 prevented DNA breaks from forming in budding yeast cells . Although Rousova et al . used budding yeast to study the proteins involved in meiosis , similar proteins exist in plant and animal cells too . Understanding how they work could open new avenues of research into cell division . For example , studies on plant proteins could provide tools for creating new crop strains . Studies on human proteins could also provide insights into fertility problems and cancer .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression", "biochemistry", "and", "chemical", "biology" ]	2021	Novel mechanistic insights into the role of Mer2 as the keystone of meiotic DNA break formation
Beige/brite adipocytes are induced within white adipose tissues ( WAT ) and , when activated , consume glucose and fatty acids to produce heat . Classically , two stimuli have been used to trigger a beiging response: cold temperatures and β3-adrenergic receptor ( Adrb3 ) agonists . These two beiging triggers have been used interchangeably but whether these two stimuli may induce beiging differently at cellular and molecular levels remains unclear . Here , we found that cold-induced beige adipocyte formation requires Adrb1 , not Adrb3 , activation . Adrb1 activation stimulates WAT resident perivascular ( Acta2+ ) cells to form cold-induced beige adipocytes . In contrast , Adrb3 activation stimulates mature white adipocytes to convert into beige adipocytes . Necessity tests , using mature adipocyte-specific Prdm16 deletion strategies , demonstrated that adipocytes are required and are predominant source to generate Adrb3-induced , but not cold-induced , beige adipocytes . Collectively , we identify that cold temperatures and Adrb3 agonists activate distinct cellular populations that express different β-adrenergic receptors to induce beige adipogenesis . The ability to defend core body temperature in response to cold environments is , in part , mediated by non-shivering thermogenesis , through the recruitment and activation of brown and beige/brite ( brown-in-white ) adipose tissues ( Cannon and Nedergaard , 2004 ) . Beige adipocytes are activated in white adipose tissue ( WAT ) by cold exposure or by beta3-adrenergic receptor ( Adrb3 ) agonists via the sympathetic nervous system ( Young et al . , 1984 ) . Activation of beige adipocyte thermogenesis requires the induction of numerous thermogenic and mitochondrial genes such as uncoupling protein 1 ( Ucp1 ) ( Cannon and Nedergaard , 2004 ) . To perform thermogenic function , beige adipocytes , like classical brown adipocytes , take up substantial amounts of glucose and free fatty acids and convert these molecules into heat rather than ATP . This uncoupling ability is clinically desirable because it has the potential to lower body fat percentage , reduce blood sugars and increase metabolic rate ( Sidossis and Kajimura , 2015; Blondin et al . , 2014; Blondin et al . , 2017 ) . Using radiolabeled glucose uptake and positron electron tomography ( PET ) imaging , brown/beige fat has been identified in cold-exposed adult humans ( Cypess et al . , 2009; van Marken Lichtenbelt et al . , 2009 ) . These imaging studies have indicated that glucose uptake occurs often in the supraclavicular region . Furthermore , supraclavicular WAT biopsies from cold-exposed humans have shown the presence and activation of brown/beige fat , which appears to resemble rodent beige fat rather than rodent brown fat ( Shinoda et al . , 2015; Wu et al . , 2012 ) . In addition to cold exposure , Adrb3 agonists also stimulate and promote beige adipocyte formation and activation ( Himms-Hagen et al . , 1994 ) . However , clinically used non-selective Adrb3 agonists have led to off target side-effects such as tachycardia and hypertension , which precludes the use of these Adrb3 agonists as beiging agents in the clinic ( Arch , 2011 ) . Yet , recent efforts by Cypess and colleagues have identified a more selective Adrb3 agonist , mirabergon ( MB ) , which is clinically used to treat overactive bladder conditions , as a potential human ‘beiger’ ( Cypess et al . , 2015 ) . Numerous studies have proposed various origins of beige adipocytes . For example , the notion that white adipocytes can transdifferentiate into beige adipocytes has been a conventional notion ( Barbatelli et al . , 2010; Smorlesi et al . , 2012 ) . On the other hand , several recent perivascular genetic fate-mapping studies have indicated that cold-induced beige adipocytes are generated from blood vessel-derived progenitors within WAT . For instance , perivascular smooth muscle cells that express Myh11 , Pdgfrb and Acta2 have been shown to be a progenitor cell pool for cold-induced beige adipocytes ( Berry et al . , 2016; Long et al . , 2014; Shao et al . , 2016 ) . Also , platelet-derived growth factor receptor α ( Pdgfra ) positive fibroblasts can form some Adrb3-induced beige adipocytes within the perigonadal adipose tissue ( PGW ) but not within the subcutaneous inguinal adipose ( IGW ) WAT depot ( Lee et al . , 2012 ) . Yet , Pdgfra+ cells do not appear to serve as a cold-induced beige progenitor pool ( Lee et al . , 2015; Berry et al . , 2016 ) . Thus , it is unclear if smooth muscle cells are the major contributing cellular source for Adrb3-induced beige adipocytes . Mechanistically , studies have also suggested that both cold and Adrb3 stimulate brown adipocyte activation and beige adipocyte recruitment similarly through the activation of Adrb ( Ramseyer and Granneman , 2016 ) . Norepinephrine signaling , via Adrb3 , not only stimulates thermogenic action in mature brown adipocytes but also recruits and activates beige adipocytes within WAT ( Grujic et al . , 1997 ) . Further , norepinephrine signaling through Adrb3-cAMP activation upregulates thermogenic genes such as Ucp1 potentiating the beiging process ( Cao et al . , 2001 ) . However , cellular expression studies have indicated that Adrb3 expression is restricted to mature white and brown adipocytes and is not expressed within the stromal vascular fraction ( SVF ) , the proposed site of both white and beige progenitors ( Collins et al . , 1994; Berry et al . , 2016 ) . In contrast to Adrb3 , Adrb1 is expressed in the SVF , not in mature adipocytes , and is thought to mediate the proliferation and differentiation of classical brown adipocyte progenitors ( Bronnikov et al . , 1992 ) . Thermogenically , Adrb1 null mice are unable to defend their body temperature in response to the cold ( Ueta et al . , 2012 ) ; whereas , transgenic overexpression of Adrb1 showed robust cold-induced beiging potential within WAT ( Soloveva et al . , 1997 ) . Overall , several studies suggest mechanistic differences between cold- and Adrb3-induced beige adipocyte formation ( Vosselman et al . , 2012 ) . But whether these two receptors mediate or recruit beige adipocytes similarly remains unknown . Here , we show that smooth muscle/mural cells ( Acta2 and Myh11 ) do not fate-map into Adrb3-induced beige adipocytes rather most Adrb3-induced beige adipocytes emanate from pre-existing white adipocytes . White adipocyte beiging necessity tests demonstrate the requirement of white adipocytes to form a significant percentage of beige adipocytes . Moreover , we find that that cold-induced beiging requires Adrb1 activation but not Adrb3 signaling . These data suggest that several cellular sources exist for beige adipocyte formation , which could provide mechanistic insight and clinical utility into enhancing beige adipocyte formation , function and perdurance . To begin to explore possible differences between cold-induced and Adrb3-induced beiging , we evaluated the relative appearance of in vivo beige adipogenesis . C57BL/6J-inbred mice were randomized to room temperature ( 23°C ) , cold ( 6 . 5°C ) or CL316 , 243 ( CL , 1 mg/Kg ) , a Adrb3 selective agonist , for 1 , 3 , or 7 days . After 1 day of cold exposure , beige adipocyte appearance was minimal within the subcutaneous inguinal adipose depot ( IGW ) , the predominant beiging WAT . Conversely , 1 day of CL treatment produced some Ucp1+ multilocular beige adipocytes and after 3 days of CL many Ucp1+ beige adipocytes could be observed . In contrast , after 3 days of cold exposure , a small number of Ucp1+ beige adipocytes were present . By day 7 , both beiging agents produced many Ucp1+ beige adipocytes throughout the IGW depot ( Figure 1A and Figure 1—figure supplement 1 ) . These findings were also confirmed by quantitative real time-PCR analysis of beige and thermogenic gene expression ( Figure 1—figure supplement 1 ) . To continue to test differences between cold- and Adrb3-induced beiging , we aged ( six months ) C57BL/6J inbred male mice . Aged mice have been shown to be defective in cold-induced beige adipocyte formation attributed to a cellular aging senescence-like phenotype of beige progenitors ( Rogers et al . , 2012; Berry et al . , 2017 ) . Mice were then randomized to room temperature , cold or CL for seven days . Cold exposure was unable to induce beiging or a thermogenic program , as previously reported ( Figure 1B–F ) ( Berry et al . , 2017 ) . Strikingly , CL induced robust beiging , glucose consumption , and thermogenic gene expression compared to room temperature controls ( Figure 1B–G ) . To continue to explore potential difference between cold-induced and Adrb3-induced beige adipocytes , we attempted to probe whether perivascular mural cells that express Acta2 , which function as a major cellular source for cold-induced beige adipocytes , also served as a progenitor source for Adrb3 activation . Recent studies have used the Acta2Cre-ERT2; Rosa26RRFP as beige fate-mapping model ( Figure 2A ) ( Berry et al . , 2016; Wendling et al . , 2009 ) . Notably , neither the Acta2-driven reporter nor endogenous Acta2 is expressed in beige or brown adipocytes or white adipocytes ( Berry et al . , 2016 ) . To induce recombination , one dose of tamoxifen ( TM , 50 mg/Kg ) for two consecutive days was administered to two-month-old Acta2Cre-ERT2; Rosa26RRFP male mice ( Figure 2A ) . After a two-week TM washout period , mice were randomized to vehicle , CL , or mirabegron ( MB , 1 mg/Kg ) , a clinically used Adrb3 agonist , for seven days ( Figure 2A ) . Histological assessment revealed both CL and MB induced beige adipocyte formation , similarly , in the IGW and PGW WAT depots ( Figure 2—figure supplement 1 ) . Nevertheless , beige adipocytes induced by CL- or MB- did not originate from an Acta2-RFP perivascular source ( Figure 2B and Figure 2—figure supplement 2 ) . Acta2-RFP expression was restricted to the vasculature under both Adrb3 agonist treatments , similar to vehicle sections ( Figure 2B and Figure 2—figure supplement 2 ) . We next tested if chronic cold or CL treatment could provoke Acta2+ mural cells to form beige adipocytes . We cold exposed or CL treated TM-induced Acta2-RFP mice for 14 days . Fate mapping studies demonstrated that Acta2+ cells formed beige adipocytes after 14 days of cold exposure , comparable to seven days of cold treatment ( Figure 2—figure supplement 3 ) ( Berry et al . , 2016 ) , However , CL-induced Ucp1+ beige adipocytes were RFP negative after 14 days of treatment ( Figure 2—figure supplement 3 ) . We then examined if age impacts Acta2-RFP fate mapping into Adrb3-induced beige adipocytes . We administered TM to P30 , P90 and P180 Acta2Cre-ERT2; RFP mice and then randomized mice to vehicle or CL for seven days , 2 weeks post TM . Histological assessment demonstrated that Adrb3-induced Ucp1+ beige adipocytes were Acta2-RFP negative ( Figure 2—figure supplement 4 ) . Taken together , these data suggest that Acta2+ mural cells do not serve as a progenitor cell source for Adrb3-induced beige adipocytes . This negative tracing by Acta2-RFP into Adrb3-induced beige adipocytes may be , in part , due to undesired biological actions of TM on white adipose tissues thereby masking or altering fate mapping capabilities ( Ye et al . , 2015 ) . To overcome possible TM-tracing issues , we employed a doxycycline ( Dox , 0 . 5 mg/L ) inducible Acta2rtTA; TRE-Cre; Rosa26RRFP mouse model ( Figure 2C ) ( Berry et al . , 2016 ) . Acta2rtTA-RFP was restricted to the vasculature and did not generate CL- or MB-induced beige adipocytes , confirming our TM-induced Acta2Cre-ERT2; Rosa26RRFP fate-mapping studies ( Figure 2D and Figure 2—figure supplement 5 ) . To support our Adrb3 Acta2 beiging fate-mapping studies , we performed two necessity strategies: ( 1 ) a cell-killing strategy ( diphtheria toxin fragment A; DTA ) ( Ivanova et al . , 2005 ) and ( 2 ) a blockade of adipocyte differentiation ( Pparg deletion ) ( Figure 2—figure supplement 6 ) ( Berry et al . , 2016; Jiang et al . , 2014 ) . A potential concern of these methods is that these deletions will occur in other Acta2+ compartments which could alter beiging potential ( Wendling et al . , 2009 ) . To minimize off-target affects , we conducted these studies in a temporally regulated fashion and proximate to Adrb3 agonist administration ( Figure 2—figure supplement 6 ) . Both strategies ( DTA , Ppargfl/fl ) did not appear to alter Adrb3-induced beige adipocyte formation or the physiological responses to CL treatment ( Figure 2—figure supplement 6 ) , indicating that Acta2+ cells are not a primary source of Adrb3-induced beige adipocytes . Other perivascular sources such as Myh11+ cells have been shown to fate map into cold-induced beige adipocytes ( Long et al . , 2014 ) . So , we employed the Myh11Cre-ERT2 combined with RosaR26RRFP to generate Myh11Cre-ERT2; R26RRFP fate mapping mice . We randomized TM-induced mice to vehicle or CL for seven days and found that CL-induced Ucp1+ beige adipocytes that were RFP negative ( Figure 2—figure supplement 7 ) . To further test if Myh11+ cells could commit to the Adrb3-induced beige progenitor lineage , we administered TM to two-month-old Myh11Cre-ERT2; R26RRFP mice and administered vehicle or CL one month post TM for seven days . We found that no new Adrb3-induced beige adipocytes emanated from an Myh11+ smooth muscle cell under this longer time interval ( Figure 2—figure supplement 8 ) . Collectively , our fate mapping and necessity tests demonstrate that smooth muscle/mural cells , which are required for cold-induced beiging , are not a cellular source of Adrb3-induced beige adipocytes . Adrb3 agonists can stimulate perivascular Pdgfra+ fibroblasts within visceral PGW , but not IGW depots , to form beige adipocytes ( Lee et al . , 2012 ) . Yet , PdgfraCre-ERT2; RFP cells do not appear to generate cold-induced beige adipocytes ( Berry et al . , 2016; Lee et al . , 2015 ) . As a potential difference between cold- and Adrb3-induced beiging , we re-visited if Pdgfra+ cells are a source of Adrb3-induced beige adipocytes by employing the PdgfraCre-ERT2; RFP model ( Figure 3—figure supplement 1 ) . TM induced PdgfraCre-ERT2; Rosa26RRFP two-month-old male mice were randomized to vehicle , CL or MB ( Figure 3—figure supplement 1 ) . Fate-mapping revealed that roughly 2% of subcutaneous IGW multilocular Ucp1+ beige adipocytes were RFP positive ( Figure 3A ) . Conversely , ~10–15% of perigonadal multilocular Ucp1+ beige adipocytes were RFP positive ( Figure 3—figure supplement 1 ) . However , many Ucp1 antibody positive beige adipocytes were RFP negative , suggesting that Pdgfra+ cells are not a major source used to generate Adrb3-induced beige adipocytes . One potential caveat of inducible systems is recombination efficiency ( Feil et al . , 1997 ) . To examine recombination efficiency , we TM pulsed two-month-old PdgfraCre-ERT2; Rosa26RRFP mice . SV cells were isolated from subcutaneous adipose depots ( inguinal and periscapular ) and subjected to flow cytometric analysis , 24 hr after the last TM injection . We found that Pdgfra-RFP+ cells were 100% positive for Pdgfra antibody staining ( Figure 3B ) . Conversely , ~70% of Pdgfra antibody+ cells were RFP+ indicating a high correspondence between reporter and endogenous Pdgfra expression ( Figure 3C , D ) . We also performed quantitative real-time PCR analysis of isolated Pdgfra-RFP+ cells and found that these cells did not express a smooth muscle signature ( Figure 3E ) . Immunohistochemistry also demonstrated that Pdgfra-RFP+ cells did not overlap with Acta2+ cells , as previously reported ( Figure 3—figure supplement 1 ) ( Lee et al . , 2012 ) . To support our Pdgfra fate-mapping studies , we incorporated a Ppargfl/fl allele with the PdgfraCre-ERT2 mouse model ( denoted PRa-Pparg ) to block adipogenic action in Pdgfra+ cells . Two-month-old PRa-Pparg male mice were TM-induced two weeks prior to vehicle or CL treatment for seven days ( Figure 3F ) . We found that Adrb3-induced beige adipocyte formation was unaltered in response to blocking adipogenesis in Pdgfra+ cells ( Figure 3G–I and Figure 3—figure supplement 2 ) . However , in PGW depots , some beige genes were dampened ( Figure 3J ) . Together , it appears , under our conditions , that Pdgfra+ cells are a minor WAT depot-specific subset of progenitors for Adrb3-induced beiging . To globally examine if Adrb3-induced beige adipocytes were generated from a common white and beige adipose progenitor cell source , we employed the AdipoTrak system ( PpargtTA; TRE-H2BGFP ) , which marks the entire adipose lineage ( Jiang et al . , 2014; Tang et al . , 2008 ) . We Dox suppressed the AdipoTrak system from conception until postnatal day 60 ( P60 ) ( Figure 3—figure supplement 3 ) . Under these conditions , PpargtTA activity is suppressed and prevents nucleosome incorporation of H2BGFP . Dox removal , allows PpargtTA to become active and H2BGFP is incorporated into proliferating nucleosomes of the progenitor compartment but not into post-mitotic cells such as existing white and beige adipocytes . GFP will be detected in the nuclei of adipocytes if they differentiate from a proliferating GFP-marked progenitor source . Reactivation of the AdipoTrak system after Dox suppression occurs predominantly around the vasculature , and these GFP-labeled progenitors can then be traced into a fraction of cold-induced beige progenitors ( Berry et al . , 2016 ) . However , after CL administration , histological staining revealed that <2% of beige adipocytes were GFP+ ( Figure 3—figure supplement 3 ) . These data suggest that the proliferating perivascular Pparg+ adipose progenitors do not generate Adrb3-induced beige adipocytes . Transdifferentiation or interconversion of white adipocytes to beige adipocytes has long been considered the standard notion for the beiging phenomena ( Barbatelli et al . , 2010 ) ; however , recent studies have challenged this view ( Berry et al . , 2016; Wang et al . , 2013; Vishvanath et al . , 2016 ) . Yet , our data thus far have not identified a progenitor cell source for Adrb3-induced beiging . Therefore , we further probed if existing white adipocytes could generate Adrb3-induced beige adipocytes by generating AdiponectinCre-ERT2; Rosa26RRFP ( Adpn-RFP ) mice . Adpn-RFP marks 100% of pre-existing white and beige adipocytes but does not mark the adipose SV compartment or newly generating adipocytes ( Figure 4A and Figure 4—figure supplement 1 ) ( Berry et al . , 2016; Sassmann et al . , 2010 ) . TM-induced Adpn-RFP mice were randomized to vehicle , CL , or MB treatment for seven days , two-weeks post TM administration ( Figure 4B ) . Beige adipocytes generated by CL or MB were formed from an Adpn-RFP+ source ( Figure 4C , D ) . That is ~75% of all Adrb3-induced Ucp1+ beige adipocytes were RFP+ , suggesting that white adipocytes could interconvert to beige adipocytes ( Figure 4E ) . These data were confirmed by longer pulse-chase experiments in which we TM induced two-month-old AdiponectinCre-ERT2; Rosa26RRFP mice and then treated them with CL two months later . Fate mapping assessment demonstrated that ~70% of Ucp1+ beige adipocytes were RFP+ , suggesting that pre-existing white adipocytes are a major source of Adrb3-induced beige adipocytes ( Figure 4—figure supplement 1 ) . To continue to test the interconversion of white adipocytes to beige adipocytes in response to Adrb3 agonists , we turned to an in vitro system in which we could test the interconversion of white adipocytes to beige adipocytes . We isolated SV cells from un-induced Ucp1Cre-ERT2; RFP mice ( marks only mature beige adipocytes ) and were induced with white adipogenic media for seven days . Cell culture-induced white adipocytes were then treated with vehicle , CL , or MB for 4 hr and thermogenic genes were examined ( Figure 4—figure supplement 1 ) . Vehicle-treated white adipocytes did not express Ucp1 mRNA nor other thermogenic genes . However , CL- or MB-treated white adipocytes had elevated expression of Ucp1 and Pgc1a , as previously observed ( Figure 4—figure supplement 1 ) ( Cao et al . , 2001 ) . Next , we tested if Ucp1-RFP ( Ucp1Cre-ERT2; RFP ) expression could be turned on in cultured white adipocytes as a marker of beige adipocyte interconversion . Cultured white adipocytes were treated with three different conditions to monitor Ucp1-RFP expression: vehicle and TM , TM and CL , and MB and TM ( Figure 4F ) . Under vehicle conditions Ucp1-RFP was not expressed . However , treating the cells with CL or MB , and then administering TM , induced the expression of the Ucp1-RFP reporter ( Figure 4G ) . As a control , we administered TM first then added CL . Under this condition , we did not observe Ucp1-RFP reporter expression ( Figure 4G ) . Of note , we treated SV cells with CL or MB , and found that Ucp1 and other thermogenic genes remained undetectable ( not shown ) . Taken together , these data indicate that both pre-existing white adipocytes and cell culture-induced white adipocytes can induce a beige adipocyte morphology and thermogenic program . To test if white adipocytes are required to form Adrb3-induced beige adipocytes , we incorporated a Prdm16fl/fl allele with AdiponectinCre-ERT2; Rosa26RRFP mice ( Figure 5A ) . We used AdiponectinCre-ERT2 to specifically target pre-existing mature white and beige adipocytes to avoid potentially affecting newly evolving progenitors undergoing adipogenesis . Additionally , we are not testing the indisputable role of Prdm16 in beige adipocyte formation but are testing the requirement of white adipocytes to generate beige adipocytes under Adrb3 agonists ( Cohen et al . , 2014 ) . In these studies , if white adipocytes are not required then new Adrb3-induced beige adipocytes should be observed . We TM induced two-month-old AdiponectinCre-ERT2; Rosa26RRFP; Prdm16fl/fl ( Adpn-Prdm16 ) male mice; two weeks later mice were randomized to vehicle , CL , or cold exposed for seven days ( Figure 5B ) . We found that cold-exposed Adpn-Prdm16 mice showed normal beiging ( Figure 5—figure supplement 1 ) . In contrast , CL-induced beiging was diminished , but not completely absent , in Adpn-Prdm16 mice ( Figure 5C–G and Figure 5—figure supplement 1 ) . As a next step , we isolated SV cells from un-induced Adpn-Prdm16 mice . Cells were cultured in white adipogenic media and then administered TM at the end of differentiation to delete Prdm16 in mature white adipocytes ( Figure 5H ) . Adipocytes were treated with vehicle or CL for 24 hr then examined for triglyceride levels and thermogenic gene expression . We found that adipocytes deficient in Prdm16 were unable to lower triglyceride levels compared to control CL-treated adipocytes ( Figure 5I ) . Further , mRNA expression induction of beige and thermogenic genes was dampened in response to Prdm16 deletion ( Figure 5J ) . To substantiate our white adipocyte necessity results and to test the requirement of Prdm16 in cold-induced beiging process , we incorporated the Prdm16fl/fl allele with the Acta2Cre-ERT2 ( Acta2-Prdm16 ) . We TM induced two-month-old Acta2-control and Acta2-Prdm16 mice; two weeks later mice were cold exposed for one week ( Figure 5—figure supplement 2 ) . We found that control mice exhibited beiging whereas Acta2-Prdm16 mice had reduced beiging potential as assessed by rectal temperature , sera glucose , H&E staining , Ucp1 IHC , Acta2 fate mapping and beige and thermogenic gene expression ( Figure 5—figure supplement 2 ) . Collectively , these data suggest that white unilocular adipocytes are a major source and are required for beige adipocyte formation in response to Adrb3 activation . The above data suggested that Adrb3 agonists convert white adipocytes to beige adipocytes whereas cold provokes perivascular smooth muscle cells to form beige adipocytes . Previous reports have indicated that Adrb signaling mediates beige and brown adipocyte recruitment and thermogenic action in response to cold temperatures and Adrb3 agonists ( Cannon and Nedergaard , 2004; Lidell et al . , 2014 ) . Since cold and Adrb3 agonists trigger different cell types to form beige adipocytes , we tested if inhibiting Adrb3 activation altered cold-induced beiging . To test this , C57BL/6J-inbred mice were administered vehicle or SR59230A ( SR59 1 mg/Kg/day ) , a Adrb3 antagonist , for 5 days prior to room temperature , cold exposure or CL administration ( Figure 6A ) . SR59 had little to no effect on cold-induced beige adipocyte formation ( Figure 6B–F ) . In contrast , blocking Adrb3 by SR59 reduced the ability of CL to promote beige adipocyte formation ( Figure 6B–G ) . Of note , SR59 did not appear to alter brown adipose tissue , morphologically or genetically , in response to either CL or cold exposure ( Figure 6—figure supplement 1 ) , as observed in Adrb3 null genetic models ( de Jong et al . , 2017 ) . In contrast to Adrb3 expression , which is restricted to mature white and brown adipocytes ( Collins et al . , 1994 ) , other adrenergic receptors such as Adrb1 are expressed in the WAT SVF , the source of cold-induced beige progenitors ( Bronnikov et al . , 1992 ) . We too examined if Adrb1-3 expression changes in response to white adipogenic media . We found that under white adipogenic conditions , Adrb1 and Adrb2 were slightly upregulated , and Adrb3 was significantly upregulated ( Figure 7—figure supplement 1 ) . As a next step , we performed quantitative real-time PCR analysis on FACS isolated Acta2-RFP+ cells to examine if Acta2+ cells expressed any Adrbs . We found that Acta2-RFP+ cells were enriched for Adrb1 but not for Adrb2 , and Adrb3 was undetectable in both SV compartments ( Figure 7—figure supplement 1 ) . We also performed quantitative real-time PCR analysis of FACS isolated PRα-RFP+ fibroblast from WAT depots . We found that neither Adrb1 nor Adrb2 were enriched in PRα-RFP+ cells and Adrb3 was undetectable in both compartments ( Figure 7—figure supplement 1 ) . Since Acta2+ cells are enriched in Adrb1 this suggested that these cells might be engaged by Adrb1 activation to generate cold-induced beige adipocytes . To test this , we inhibited Adrb1 by treating mice with vehicle or talinolol ( 1 mg/Kg/day ) , a specific Adrb1 blocker , for 5 days prior to randomization to room temperature , cold or CL ( Figure 7A ) . Under room temperature conditions , talinolol had no effect on temperature , adipose tissue mass , adipose tissue histology and beige adipocyte gene expression ( Figure 7B–G and Figure 7—figure supplement 2 ) . Similarly , talinolol appeared to have no effect on beiging when combined with CL-treatment; that is temperature , adipose tissue mass , histology and beige gene expression were similar between CL alone and CL with talinolol ( Figure 7B–F and Figure 7—figure supplement 2 ) . Conversely , talinolol-treated cold-exposed mice demonstrated an inability to defend body temperature , reduce blood glucose and maintained elevated levels of adipose tissue mass ( Figure 7B , C and Figure 7—figure supplement 2 ) . Further , talinolol-treated cold-exposed mice had reduced beiging potential as assessed by histology and beige and thermogenic gene expression ( Figure 7D , E and G ) . Taken together , these data indicate that Adrb1 signaling is involved in cold-induced beige adipocyte formation but not Adrb3-induced beiging . Beige adipocytes have the potential to increase energy expenditure through the consumption of glucose and free fatty acids ( Kajimura et al . , 2015 ) . These potential cellular energy sinks could help address the worldwide epidemic of obesity and its associated metabolic disorders ( Yoneshiro et al . , 2013 ) . Several studies have indicated the presence and induction of metabolically relevant beige adipocytes in humans ( Cypess et al . , 2015; Saito , 2013; Saito et al . , 2009; van der Lans et al . , 2013; Wu et al . , 2012; Yoneshiro et al . , 2011b ) . For example , administration of mirabegron ( MB ) increased the resting metabolism of patients and induced the formation of beige adipocytes ( Cypess et al . , 2015 ) . Other studies using cold exposure have seen a similar phenomenon including the induction of beige adipocytes ( Cypess et al . , 2009; Yoneshiro et al . , 2011a ) . The full therapeutic potential of beige adipocytes has not been fully realized due , in part , to the lack of genetic tools to mark , track and manipulate the beige adipocyte progenitor cell source; however , the current study appears to overcome some of these barriers . The cellular sources for cold- and Adrb3-induced beige adipocytes have attracted much attention but whether these two beiging stimuli trigger the same or different progenitor sources has remained ambiguous . The same set of tools , used herein , allowed us to identify that Acta2+ perivascular progenitor cells could serve as an indispensable source of cold-induced beige adipocytes ( Berry et al . , 2016 ) . However , using two independent Acta2 genetic tools , we were unable to observe Acta2 beige adipocyte fate mapping in response to Adrb3 agonists . In addition , we were also unable to observe noticeable fate mapping of Myh11+ smooth muscle cells , which also serve as a cold-induced beige adipocyte source , into Adrb3-induced beige adipocytes . These data suggest that Adrb3 agonists do not engage smooth muscle-perivascular cells . These findings also correlated with our gene expression data that Acta2+ cells express Adrb1 not Adrb3 . Moreover , our pharmacological studies indicate that cold-induced beige adipocyte formation relies primarily on Adrb1 activation rather than ARDB3 stimulation , which agrees with both our fate mapping and gene expression studies . These findings echo with a recent genetic study that showed that cold-induced beiging is intact in Adrb3 null mice ( de Jong et al . , 2017 ) . However , other studies have demonstrated that Adrb3 null mice have less beiging potential ( Barbatelli et al . , 2010; Jimenez et al . , 2003 ) . The underlying differences between these studies may be , in part , due to genetic backgrounds of the null mice , further studies would be required . Our studies support the notion that Adrb3-induced beige adipocytes originate from white adipocytes . Using AdiponectinCre-ERT2; RFP fate mapping studies , we found that the majority of Adrb3-induced beige adipocytes emanate from these pre-existing adipocytes . However , cold relies on Acta2+ mural cells . WAT necessity tests using AdiponectinCre-ERT2; Prdm16 demonstrated that pre-existing white adipocytes are required for Adrb3-induced beige adipocyte formation . A previous study demonstrated that deleting Prdm16 within the WAT lineage sufficiently blocked beige adipocyte formation from both cold exposure and Adrb3 agonists ( Cohen et al . , 2014 ) . The major difference between the two studies is the use of the temporal AdiponectinCre-ERT2 genetic tool compared to the global AdiponectinCre genetic tool . In our studies using the AdiponectinCre-ERT2 , Prdm16 is specifically deleted in all pre-existing adipocytes ( beige and white ) and not in newly generated adipocytes . In contrast , using AdiponectinCre model , Prdm16 is continuously deleted in all pre-existing and newly developing adipocytes ( Jeffery et al . , 2014; Shao et al . , 2016; Wang et al . , 2013 ) , meaning that under all conditions , cold or Adrb3 agonist , beige adipogenesis is disrupted . Yet , fate mapping limitations aside , the AdiponectinCre tool identified the critical role of Prdm16 in beige adipogenesis . Classical biochemical and pharmacokinetic studies have indicated that Adrb1 activation is not coupled to thermogenesis in brown adipocytes . In agreement , our data imply that Adrb1 is critical for Acta2+ perivascular cells to leave their vascular niche and form beige adipocytes . However , using pharmacological agents such as talinolol , could have systemic affects that may alter beiging potential independent of Adrb1 blockade within WAT . Pharmacokinetically , talinolol is highly hydrophilic with a half-life of 6–7 hr and has difficulty crossing the blood-brain barrier ( Neil-Dwyer et al . , 1981; Sourgens et al . , 2003 ) . Of note , mice were exposed to the cold or CL , 24 hr post talinolol treatment . Further investigation into the downstream signaling of Adrb1 in Acta2+ cells in response to the cold could provide important insight into potential mechanisms and therapeutic targets to stimulate cold-induced beiging perhaps in the absence of cold treatment . The current study demonstrates that several cell types can form multilocular beige-like adipocytes within white adipose depots . This is consistent with other reports that Pdgfra+ perivascular cells as well other non-adipocyte sources generate beige adipocytes ( Wang et al . , 2013; Lee et al . , 2012 ) . Additionally , other studies have suggested that beige and white adipocytes are bi-potential; that is beige adipocytes become white adipocytes but can revert to beige adipocytes in response to a beiging stimulus ( Rosenwald et al . , 2013; Altshuler-Keylin et al . , 2016 ) . Our studies resonate with these previous findings; however , the data herein suggest that a majority of Adrb3-induced beige adipocytes are generated from mature white adipocytes . A potential limitation of these study is that we cannot decipher if these white adipocytes that converted to Adrb3-induced beige adipocytes were once beige . Further studies into genetic program and memory could help elucidate how these cells interconvert in response to sympathetic activation . These studies also raise another interesting question; do beige adipocytes that form in response to cold exposure or by Adrb3 agonists activate thermogenesis in an analogous manner ? That is; does cold temperatures invoke the same thermogenic program as Adrb3 agonist ? More studies directed at examining beige metabolic properties could provide pivotal insight into which beige stimuli would be more clinically germane regarding energy utilization . Also do these differently induced beige adipocytes have different genetic signatures ? Towards this end , a recent study showed that beige adipocytes generated by CL compared to beige adipocytes produced by roscovitine/rosiglitazone had unique genetic signatures suggesting that not all beige adipocytes are created equally ( Wang et al . , 2016 ) . Hence , further research into the genetic regulation and programs could yield answers to these questions . In summary , using several genetic models encompassing several proposed sources of Adrb3-induced beige adipocytes , we found that white adipocytes generate most Adrb3-induced beige adipocytes . Further , we found that cold-induced beiging relies on Adrb1 activation , and not Adrb3 , to engage and license smooth muscle cells to form beige adipocytes . These data highlight that several cellular sources exist for two different beiging stimuli and that not all beiging stimuli should be considered equal . All animals were maintained under the ethical guidelines of the UT Southwestern Medical Center Animal Care and Use Committee according to NIH guidelines under the protocol number 2016–101336 . Mice were housed in a 12:12 light:dark cycle and chow and water were provided ad libitum . Experiments were performed on two or six-month-old male mice unless otherwise stated . For all experiments , mice were randomized to their respective groups without restrictions . All mouse lines were backcrossed on a C57BL/6J- 129S1/SvImJ mixed background for at least nine generations . AdipoTrak mice were previously established in our lab ( Tang et al . , 2008 ) . C57BL/6J ( stock no: 000664 ) , AdiponectinCre-ERT2 ( stock no: 025124 ) , Myh11Cre-ERT2 ( stock no: 019079 ) , Ppargfl/fl ( stock no: 004584 ) , Prdm16fl/fl ( stock no: 024992 ) , Rosa26RDTA ( stock no: 006331 ) , and Rosa26RRFP ( stock no: 007908 ) mice were obtained from the Jackson Laboratory . Acta2Cre-ERT2 mice were generously provided by Dr . Pierre Chambon . Acta2rtTA mice were generously provided by Dr . Beverly Rothermel . Drs . Sean Morrison and Bill Richardson generously provided the PdgfraCre-ERT2 mice . Ucp1Cre-ERT2 were generously provided by Dr . Eric Olson . Cre recombination was induced by administering one dose of tamoxifen dissolved in sunflower oil ( Sigma , 50 mg/Kg interperitoneally injection ) for two consecutive days . rtTA activation was induced by Doxycycline ( 0 . 5 mg/ml in 1% sucrose ) provided in the drinking water and protected from light , and it was changed every 2–3 days . For cold experiments , mice were placed in a 6 . 5°C cold chamber or maintained at room temperature ( 23–25°C ) for seven days . CL316 , 243 was purchased from Tocris and dissolved in water . CL316 , 243 was administered at one dose ( 1 mg/Kg/day ) for seven consecutive days by interperitoneal injections ( IP ) . Mirabegron was purchased from Cayman Chemical and dissolved in water and was administered at one dose ( 1 mg/Kg/day ) for seven consecutive days by IP . Talinolol was purchased from Cayman Chemical and was dissolved in DMSO . Administration solution was dissolved further in water ( DMSO ~5% ) and was administered at one dose ( 1 mg/Kg/day ) for five consecutive days by IP . SR59230A was purchased from Sigma-Aldrich and was dissolved in DMSO . Administration solution was dissolved further in water ( DMSO ~5% ) and was administered at one dose ( 1 mg/Kg/day ) for five consecutive days by IP . No cell lines were used . Stromal-vascular ( SV ) cells were isolated from pooled subcutaneous ( inguinal , interscapular ) white adipose tissues for fractionation , unless indicated otherwise . After 2 hr of slow shaking in isolation buffer ( 100 mM HEPES pH7 . 4 , 120 mM NaCl , 50 mM KCl , 5 mM glucose , 1 mM CaCl2 , 1 . 5% BSA ) containing 1 mg/ml collagenase at 37°C . The suspension was then spun at 800 g for 10 min and the pellet contained a crude SV fraction . The SV pellet was then re-suspended in erythrocyte lysis buffer ( 0 . 83% NH4Cl in H2O ) for 5 min , spun at 800xg for 5 min . The pellet was washed once in 1X PBS , re-suspended and passed through 40 μm mesh . Isolated SV cells were cultured in DMEM supplemented with 10% FBS with 1% penicillin and streptomycin . White adipogenesis was induced by treating confluent cells with DMEM containing 10% FBS , insulin ( 0 . 5 µg/ml ) , dexamethasone ( 5 µM ) , and isobutylmethylxanthine ( 0 . 5 mM ) . To induce beige and thermogenic genes , cells were treated with 5 µM CL316 , 243 or 5 µM mirabegron for 4 or 24 hr . mRNA was harvested or cells were imaged ( Wu et al . , 2012 ) . Triglyceride accumulation was performed using a kit from ZenBio ( Berry and Noy , 2009; Berry et al . , 2010 ) and manufacturer's protocol was followed . SV cells were isolated as above and washed , centrifuged at 1200 g for 5 min , and analyzed with a FACScans analyzer or sorted with a BD FACS Aria operated by the UT Southwestern Flow Cytometry Core . Data analysis was performed using BD FACS Diva and FlowJo software . For RFP+ sorting , live SV cells from Acta2Cre-ERT2; R26RRFP , or PdgfraCre-ERT2; R26RRFP , or AdiponectinCre-ERT2; R26RRFP mice were sorted based on native fluorescence ( RFP ) . The SV cells from TM-induced Acta2Cre-ERT2 or PdgfraCre-ERT2 ( without RFP ) control mice were used to determine background fluorescence levels . SV cells were incubated with primary antibodies on ice for 30 min and then washed twice with the staining buffer and incubated with secondary antibody for another 30 min on ice before flow cytometry analysis . Primary antibodies include: rabbit-anti-Pdgfrα ( 1:100 , Santa Cruz Biotechnology ) . Total RNA was extracted using TRIzol ( Invitrogen: item no: 15596026 ) from either mouse tissues or cells . Mouse tissues ( n= 6 individual tissues ) and cells ( n= 4 individual wells ) were pooled and analyzed in technical quadruplicates . These experiments were performed on three independent cohorts . cDNA synthesis was performed using RNA to cDNA high capacity kit ( Invitrogen: item no: 4387406 ) . Gene expression was analyzed using Power SYBR Green PCR Master Mix with ABI 7500 Real-Time PCR System . qPCR values were normalized by 18 s rRNA expression . Primer sequences are available in Supplementary file 1 . Adipose tissues were fixed in 10% formalin , dehydrated , embedded in paraffin , and sectioned with a Microm HM 325 microtome at 5–15 μm thickness . For immunofluorescence staining , paraffin sections were preincubated with permeabilization buffer ( 0 . 3% Triton X-100 in PBS ) for 30 min at room temperature and then incubated sequentially with primary antibody ( 4°C , overnight ) and secondary antibody ( 2 hr at room temperature ) , all in blocking buffer ( 5% normal donkey serum in 1X PBS ) . Antibodies used for immunostaining are: rabbit-anti-Ucp1 ( 1:500 , Abcam ) , mouse-anti-RFP ( 1:200 , Clontech ) , mouse-anti-Acta2 ( 1:500 , Abcam ) , rabbit-anti-Pdgfrα ( 1:100 , Santa Cruz Biotechnology ) , and goat-anti-Perilipin ( 1:500 , Abcam ) . Secondary antibodies , Alexa Fluor 488 donkey anti-rabbit , cy3 donkey anti-mouse , and Alexa Fluor 647 donkey anti-goat , were from Jackson ImmunoResearch . All secondary antibodies were used at a 1:500 dilution . For immunohistochemistry staining , slides were deparaffinized and rehydrated before heat-induced antigen retrieval . Antigens were detected using primary antibody and in conjunction with an HRP/DAB ( ABC ) detection kit ( Abcam; ab64264 ) according to the manufacturer’s instructions ( R&D Systems ) . Immunostained images were collected on a Zeiss LSM500 confocal microscope , or a Lecia DMi6 . For quantification of images , two independent observers assessed three random fields in 10 random sections from at least three mice per cohort and used Image J to quantify co-localization . Temperature was monitored daily using a rectal probe ( Physitemp ) . The probe was lubricated with glycerol and was inserted 1 . 27 centimeters ( 1/2 inch ) and temperature was measured when stabilized . Glucose monitoring: tail blood glucose levels were measured immediately after cold exposure ( 9am CST ) with a Contour glucometer ( Bayer ) . Statistical significance was assessed by an unpaired Student’s t-test using Origin Labs 8 . 1 software , Excel or GraphPad Prism 6 . p<0 . 05 was considered statistically significant . Mouse experiments were performed in biological triplicate with at least six mice/group and results are expressed as means ± SEM . No power analysis was used to calculate samples size rather sample size was calculated on historical fate-mapping and metabolic studies from the ours and other’s research group . For fate-mapping experiments , a minimum of 6 mice/group were analyzed replicated thrice . For metabolic studies , a minimum of 8 mice/group were analyzed replicated thrice . Experiments were performed on male mice at denoted ages . No mice were excluded from the study unless visible fight wounds were observed . Cell culture experiments were collected from three or four independent cultures for each sample and pooled .	Excess accumulation of a type of fat called white fat is associated with obesity and metabolic problems . White fat cells store energy . White fat tissue also contains some beige fat cells , which burn fats and sugars to produce heat . Cold temperatures trigger the production and activity of beige fat cells , which allows the body to stay warm . People with obesity tend to have less beige fat and more white fat . This has led scientists to test whether treatments that increase the number of beige fat cells a person has could reduce fat mass and improve metabolism . To develop treatments that increase beige fat , scientists must first understand where it comes from and how cold and other factors stimulate its growth . Recent studies have shown that smooth muscle cells , which surround blood vessel walls , make cold-induced beige fat cells . A widely used drug that turns on the β3 adrenergic receptor , which is found in the cell membrane , also boosts the creation of beige fat cells . Yet , it was not clear exactly how cold or this drug triggers the production of beige fat . Now , Jiang et al . show that drugs that target β3 adrenergic receptors cause white fat cells in mice to change into beige fat cells . The experiments also showed that cold turns on a different receptor called the β1 adrenergic receptor on smooth muscle cells causing them to make beige fat cells . This shows that there is more than one source for beige fat cells in the body and that different strategies for increasing beige fat cell numbers do not work the same way . More studies are needed to learn whether beige fat cells produced after exposure to cold or drugs behave in the same way and have similar affects on metabolism . This could help scientists determine if one of these strategies could make a better treatment for obesity or other metabolic disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "developmental", "biology", "cell", "biology" ]	2017	Distinct cellular and molecular mechanisms for β3 adrenergic receptor-induced beige adipocyte formation
Osteocalcin ( OCN ) is an osteoblast-derived hormone with pleiotropic physiological functions . Like many peptide hormones , OCN is subjected to post-translational modifications ( PTMs ) which control its activity . Here , we uncover O-glycosylation as a novel PTM present on mouse OCN and occurring on a single serine ( S8 ) independently of its carboxylation and endoproteolysis , two other PTMs regulating this hormone . We also show that O-glycosylation increases OCN half-life in plasma ex vivo and in the circulation in vivo . Remarkably , in human OCN ( hOCN ) , the residue corresponding to S8 is a tyrosine ( Y12 ) , which is not O-glycosylated . Yet , the Y12S mutation is sufficient to O-glycosylate hOCN and to increase its half-life in plasma compared to wildtype hOCN . These findings reveal an important species difference in OCN regulation , which may explain why serum concentrations of OCN are higher in mouse than in human . Osteocalcin ( OCN ) is a peptide hormone secreted by osteoblasts , the bone forming cells ( Lee et al . , 2007 ) . It regulates glucose metabolism by promoting beta cell proliferation and insulin secretion , and by improving insulin sensitivity ( Pi et al . , 2011; Ferron et al . , 2012 ) . In addition to its role in the regulation of energy metabolism , OCN is also involved in male fertility by promoting testosterone synthesis by Leydig cells ( Oury et al . , 2011 ) , in muscle adaptation to exercise by improving glucose and fatty acid uptake in myocytes ( Mera et al . , 2016a ) , and in acute stress response through the inhibition of post-synaptic parasympathetic neurons ( Berger et al . , 2019 ) . Overall , OCN might acts as an ‘anti-geronic’ circulating factor preventing age-related cognitive decline and muscle wasting ( Khrimian et al . , 2017; Mera et al . , 2016b; Oury et al . , 2013b ) . The G protein coupled receptor family C group six member A ( GPRC6A ) mediates OCN function in beta cells , muscles and testis ( Mera et al . , 2016a; Pi et al . , 2011; Oury et al . , 2013a ) , while the G protein coupled receptor 158 ( Gpr158 ) mediates its function in the brain ( Khrimian et al . , 2017; Kosmidis et al . , 2018 ) . Within the bone tissue , OCN undergoes a series of post-translational modifications ( PTM ) that are critical for the regulation of its endocrine functions . Prior to its secretion , in the osteoblast endoplasmic reticulum , the OCN precursor ( pro-OCN ) is γ-carboxylated on three glutamic acid residues ( Glu ) by the vitamin K-dependent γ-glutamyl carboxylase ( Ferron et al . , 2015 ) . In the trans-Golgi network , pro-OCN is next cleaved by the proprotein convertase furin releasing mature carboxylated OCN ( Gla-OCN ) ( Al Rifai et al . , 2017 ) . The presence of the negatively charged Gla residues allows Gla-OCN to bind hydroxyapatite , the mineral component of the bone extracellular matrix ( ECM ) . It is during bone resorption that Gla-OCN is decarboxylated through a non-enzymatic process involving the acidic pH generated by the osteoclasts , ultimately leading to the release of bioactive uncarboxylated OCN ( ucOCN ) in the circulation ( Ferron et al . , 2010a; Lacombe et al . , 2013 ) . The conclusion that ucOCN represents the bioactive form of this protein in rodents is supported by cell-based assays , mouse genetics and in vivo studies [reviewed in Mera et al . , 2018] . The role of OCN in the regulation of glucose metabolism appears to be conserved in humans . Human ucOCN can bind and activate human GPRC6A ( De Toni et al . , 2016 ) and promotes beta cell proliferation and insulin synthesis in human islets ( Sabek et al . , 2015 ) , while mutations or polymorphisms in human GPRC6A are associated with insulin resistance ( Di Nisio et al . , 2017; Oury et al . , 2013a ) . Finally , several cross-sectional and observational studies have detected a negative association between OCN or ucOCN , and insulin resistance or the risk of developing type 2 diabetes in various human populations ( Lin et al . , 2018; Turcotte et al . , 2020; Lacombe et al . , 2020 ) . Yet , some important species divergences exist between mice and humans with regard to OCN biology . First , only 30 out of the 46 amino acids ( i . e . 65% ) composing mature mouse OCN are conserved in human OCN . This is in striking contrast with other peptide hormones involved in the control of energy metabolism such as leptin and insulin whose respective sequence display about 85% conservation between mouse and human . Second , the circulating concentrations of OCN , even though decreasing with age in both species , are five to ten times higher in mice than in humans throughout life span [ ( Mera et al . , 2016a ) see also Table 1] . Based on these observations , we hypothesized that the post-translational regulation of OCN may be different between these two species , resulting in increased mouse OCN half-life in circulation . Here , using proteomics and cell-based assays , we identified O-glycosylation as a novel PTM presents in mouse OCN , and showed that this modification increases mouse OCN half-life in plasma ex vivo and in vivo . In contrast , mature human OCN does not contain the O-glycosylation site found in the mouse protein and consequently is not normally glycosylated . Yet , a single point mutation in human OCN is sufficient to elicit its O-glycosylation and to increase its half-life in plasma . To better document the circulating level of OCN in humans and mice , we measured the serum concentration of OCN in wildtype mice at different ages ( 2 to 60 weeks ) and compared the values with the reported serum level of OCN at corresponding life periods in humans ( Table 1 ) . This analysis reveals that serum OCN level is five- to ten-time lower in humans than in mice throughout life . One potential explanation for this observation could be that a mouse specific PTM increases mouse OCN half-life in circulation . Since OCN Gla residues and pro-OCN cleavage site are conserved between mouse and human , we searched for additional PTMs present in mouse OCN and characterized their impact on OCN half-life . To that end , OCN was immunoprecipitated from the secretion medium of primary mouse osteoblast cultures or from mouse bone protein extracts using specific polyclonal goat antibodies recognizing OCN C-terminus sequence ( Ferron et al . , 2010b ) . OCN was then characterized without proteolysis by reverse-phase HPLC followed by mass spectrometry ( MS ) and tandem mass spectrometry ( MS/MS ) . This ‘top-down’ analysis revealed that the most abundant OCN forms have a monoisotopic mass ranging from 5767 . 6961 to 6441 . 7636 Da which exceeds the predicted mass of 5243 . 45 Da corresponding to the fully carboxylated ( 3 × Gla ) OCN ( Table 2 and Table 3 ) . According to the various monoisotopic masses observed , we predict that this difference could be mainly explained by the presence of a single O-linked glycan adduct composed of one N-acetylgalactosamine ( GalNAc ) , one galactose ( Gal ) and one or two N-acetylneuraminic acid ( NANA ) . In addition to O-glycosylation , minor additional monoisotopic mass change corresponding to oxidation , unidentified modifications and/or adduct ions were also detected . Uncarboxylated OCN does not accumulate in bone ECM ( Ferron et al . , 2015 ) and accordingly , only fully or partially carboxylated OCN was detected in bone extracts , both of which were found to be O-glycosylated . In contrast , significant amount of ucOCN could be detected in osteoblast culture supernatants , most likely due to the low level of vitamin K present in fetal bovine serum . Importantly , in osteoblast supernatant the O-linked glycan adduct could be detected on both carboxylated and uncarboxylated OCN ( Table 2 ) . Using multiple approaches , we next established that mouse OCN is indeed subjected to O-glycosylation in cells and in vivo . First , in SDS-PAGE analyses the apparent molecular weight of OCN is reduced when expressed in HEK293 cells where the O-glycosylation capacity has been engineered to truncate O-glycans by knockout of C1GALT1C1 which encodes COSMC ( core 1β3-Gal-T-specific molecular chaperone ) , a private chaperone required for the elongation of O-glycans ( Figure 1A; Steentoft et al . , 2011 ) . Second , when expressed in CHO-ldlD cells , which have defective UDP-Gal/UDP-GalNAc 4-epimerase and are hence deficient in O-glycosylation ( Kingsley et al . , 1986 ) , OCN apparent molecular weight is also reduced compared to the same protein expressed in the parental CHO cell line . Supplementation of Gal and GalNAc in the culture medium rescued the O-glycosylation defect of CHO-ldlD and restored the molecular shift in the secreted OCN ( Figure 1B ) . Third , treatment of primary osteoblasts with benzyl-N-acetyl-α-galactosaminide ( GalNAc-bn ) , an inhibitor of N-acetylgalactosaminyltransferases ( GalNAc-Ts ) , the enzymes responsible for initiating O-glycosylation , decreases the apparent molecular weight of OCN secreted in the medium ( Figure 1C ) . Finally , treatment of mouse bone extracts with neuraminidase and O-glycosidase , which removes respectively NANA , and core 1 and core 3 O-linked disaccharides , also decreases the apparent molecular weight of endogenous OCN ( Figure 1D ) . We next aimed at identifying which OCN residue ( s ) is ( are ) O-glycosylated . Mature mouse OCN contains three serine ( S ) and three threonine ( T ) residues ( Figure 1E ) , the two main types of amino acids on which O-glycosylation occurs ( Steentoft et al . , 2013 ) . As expected , mutating all serine and threonine residues into alanine abrogates OCN glycosylation in primary mouse osteoblasts as assessed by SDS-PAGE ( Figure 1F ) . Further mutagenesis studies revealed that the O-glycosylation site resides within the N-terminal part of the protein , that is on S5 , S8 or T15 ( Figure 1F ) . Single amino acid mutagenesis allowed the identification of S8 as the O-glycosylation site of OCN in osteoblasts ( Figure 1G ) , a result consistent with the MS/MS analysis of OCN isolated from bone which also suggested that this residue is the O-glycosylation site ( Figure 1H ) . Together , these results establish that mouse OCN is O-glycosylated on at least one serine residue in cell culture and in vivo . Protein O-glycosylation is initiated by the transfer of a GalNAc to a serine or threonine residue , a reaction taking place in the Golgi and catalyzed by GalNAc-Ts , a family of enzymes comprising 19 different members in mice ( Bennett et al . , 2012 ) . Quantitative PCR on mRNA isolated from undifferentiated and differentiated primary mouse osteoblasts revealed that several GalNAc-Ts are expressed in this cell type , with Galnt1 and Galnt2 being the most strongly expressed ones ( Figure 2A ) . We noticed that S8A mutation abrogates OCN O-glycosylation in HEK293 cells and in primary osteoblasts ( Figure 1G and data not shown ) . GalNAc-T3 and its paralogue GalNAc-T6 are known to be expressed in HEK293 ( Narimatsu et al . , 2019b ) and our data shows they are also expressed in primary mouse osteoblasts and induced during osteoblast differentiation . Although these observations suggest one or both of these enzymes may be involved in OCN O-glycosylation , the inactivation of GALNT3 and/or GALNT6 genes failed to alter OCN O-glycosylation in HEK293 ( Figure 2B ) . Since GALNT1 and GALNT2 are also highly expressed in osteoblasts and induced during osteoblast differentiation , we inactivated these two genes in combination with GALNT3 , and assess the impact on OCN O-glycosylation in HEK293 cells . This partially abolished OCN glycosylation ( Figure 2B ) , suggesting that these three GalNAc-Ts are the primary isoenzymes that redundantly initiate the O-glycosylation of OCN . Processing of pro-OCN by the proprotein convertase furin and its γ-carboxylation are two post-translational modifications regulating OCN endocrine function ( Ferron et al . , 2015; Al Rifai et al . , 2017 ) . We therefore next aimed at testing whether OCN O-glycosylation can interfere with its γ-carboxylation or processing , or inversely , if O-glycosylation is modulated by γ-carboxylation or processing of pro-OCN . Pharmacological inhibition of γ-carboxylation or furin , using warfarin or Dec-RVKR-CMK ( RVKR ) respectively , did not impact OCN O-glycosylation in primary osteoblasts and HEK293 cells ( Figure 2C and Figure 2—figure supplement 1 ) . Similarly , inhibition of OCN O-glycosylation through GalNAc-bn treatment or the S8A mutation did not significantly affect its processing or its γ-carboxylation ( Figure 2C–E and Figure 2—figure supplement 1 ) . We also tested whether OCN processing influences its O-glycosylation in vivo . As shown in Figure 2F , both mature OCN present in control bones and pro-OCN present in furin-deficient bones are de-glycosylated by neuraminidase and O-glycosidase , indicating that pro-OCN is normally O-glycosylated in absence of processing by furin in vivo . Altogether , these results support the notion that osteocalcin O-glycosylation is not affected by its carboxylation status or by its processing by furin . Moreover , blocking O-glycosylation does not prevent pro-OCN processing by furin or its carboxylation . The results presented above suggest that O-glycosylation is not regulating the processing of pro-OCN by furin or the secretion of mature OCN by osteoblasts . It was recently observed that O-glycosylation can also increase the stability of some peptide hormones in the circulation by preventing proteolytic degradation ( Hansen et al . , 2019; Madsen et al . , 2020 ) . We therefore aimed at testing the impact of O-glycosylation on OCN half-life in plasma . To that end , we produced and purified O-glycosylated ucOCN from HEK293 and first compared its purity and molecular weight to native non-glycosylated ucOCN produced in bacteria by LC-MS , LC-MS/MS and SDS-PAGE ( Figure 3A , B; Figure 3—figure supplement 1 ) . Importantly , we observed that >99% of the ucOCN purified from HEK293 is O-glycosylated , with a certain proportion ( ~30% ) containing two glycan adducts ( Figure 3A ) , suggesting that in this context O-glycosylation may occurs on more than one residue . Freshly isolated Bglap-/- plasma , which is depleted of endogenous OCN , was next used to assess ucOCN half-life ex vivo . In all the following experiments , concentrations of ucOCN were measured with an ELISA assay recognizing both non-glycosylated and glycosylated mouse ucOCN ( see Materials and methods and Figure 3—figure supplement 2 ) . Non-glycosylated ucOCN has a half-life of ~120 min when incubated in plasma at 37°C , while O-glycosylated ucOCN is stable for more than 5 hr in the same conditions ( Figure 3C ) . Non-glycosylated ucOCN was stable when incubated in a saline solution containing 3 . 5% BSA at 37°C for 2 hr ( Figure 3—figure supplement 3 ) , implying that ucOCN is not intrinsically unstable . In addition , stability of the non-glycosylated ucOCN was restored when incubated in plasma at 4°C or in heat-inactivated ( HI ) plasma at 37°C ( Figure 3D ) , suggesting that non-glycosylated ucOCN’s decline involves the action of a protease . Pepstatin A ( Pep A ) an aspartic proteases inhibitor , RVKR a proprotein convertases inhibitor and ethylenediaminetetraacetic acid ( EDTA ) which inhibits metalloproteases did not affect non-glycosylated ucOCN stability in plasma ex vivo ( Figure 3—figure supplement 3 ) . The OCN sequence surrounding S8 contains several proline residues ( Figure 1E ) and could therefore be recognized by prolyl endopeptidases , such as the fibroblast activation protein ( FAP ) which is present in the circulation ( Sánchez-Garrido et al . , 2016; Coutts et al . , 1996 ) . We thus also tested the effect of talabostat , an inhibitor of FAP and dipeptidyl peptidases , on ucOCN stability in plasma . Surprisingly , treatment of plasma with 10 mM talabostat did not inhibit , but rather increased non-glycosylated ucOCN degradation ( Figure 3E ) . A reduced stability of non-glycosylated OCN in plasma was also observed in the presence of PMSF ( Figure 3F ) at a concentration ( 10 mM ) that was also shown to inhibit dipeptidyl peptidases and prolyl endopeptidases ( Banbula et al . , 2000; Bermpohl et al . , 1998 ) . One function of FAP in vivo is to activate the α2-antiplasmin precursor releasing the active α2-antiplasmin , which in turn acts as an inhibitor of plasmin activity ( Lee et al . , 2011; Lee et al . , 2006 ) . Interestingly , mouse OCN contains arginine residues in its N- and C-terminus ( R16 and R40 , respectively ) , which could be potential cleavage sites for plasmin ( Rawlings et al . , 2008 ) . Together these observations led us to hypothesize that plasmin could be a protease responsible for ucOCN degradation in plasma . Supporting this notion , ucOCN is rapidly degraded when low concentration of recombinant plasmin is added to previously heat-inactivated plasma or in Tris buffered solution ( Figure 3G and data not shown ) . Moreover , O-glycosylation partially protect ucOCN from plasmin-mediated degradation in the same assay ( Figure 3G ) . Altogether , these data indicate that O-glycosylation protects ucOCN from plasmin mediated proteolysis , thereby increasing its half-life in plasma in vitro . We next examined the stability of glycosylated and non-glycosylated mouse ucOCN in vivo by injecting an equal dose of each of these proteins in Bglap-/- mice which are depleted of endogenous OCN . In fasted animals , following an injection of 40 ng/g of body weight of ucOCN , the level of glycosylated ucOCN remains higher compared to the non-glycosylated form for the following 90 min ( Figure 3H ) . Moreover , when expressed as a percentage of the maximum concentration reached at 30 min , the concentration of non-glycosylated OCN declines more rapidly than the one of glycosylated ucOCN ( Figure 3I ) . Based on these curves , we estimated the in vivo half-life of O-glycosylated and non-glycosylated ucOCN to ~182 and~108 min respectively . In fed mice , glycosylated ucOCN serum concentration also remains higher than the one of non-glycosylated ucOCN for up to 2 hr following an injection of 40 ng/g of body weight ( Figure 3—figure supplement 4 ) . Circulating level of glycosylated ucOCN after 2 hr was further increased when 80 ng/g of body weight of protein was injected , while non-glycosylated ucOCN was not significantly increased with this higher dose ( Figure 3—figure supplement 4 ) . These results establish that O-glycosylation increases the stability of mouse OCN protein in vivo . Using a cell-based assay , we next tested if O-glycosylation impacts the biological activity of OCN . INS-1 832/3 cells , a sub-clone of the INS-1 rat insulinoma cell line previously shown to express GPRC6A and to respond to OCN ( Ferron et al . , 2010a; Pi et al . , 2016 ) , were treated with vehicle or low doses of non-glycosylated and glycosylated ucOCN for 8 hr . At the end of the stimulation period , the expression of Ins1 , the gene encoding pro-insulin 1 , was assessed by quantitative PCR as a readout of OCN activity . As shown in Figure 4A , non-glycosylated ucOCN stimulation at 0 . 3 ng/ml could increase Ins1 expression by 1 . 5-fold . However , a lower dose of glycosylated ucOCN ( 0 . 1 ng/ml ) was sufficient to significantly increase the expression of Ins1 as compared to vehicle or to 0 . 3 ng/ml non-glycosylated ucOCN stimulation . These difference in biological activity could be explained at least in part by an increased stability of glycosylated ucOCN as compared to non-glycosylated ucOCN in INS-1 832/3 cell cultures ( Figure 4B ) . Overall , these results suggest that glycosylated ucOCN is biologically active , at least in the setting of this cell-based assay , and that O-glycosylation also increases the stability of ucOCN in cell culture . Sequence alignments revealed that the residue corresponding to S8 in the mouse protein is a tyrosine ( Y12 ) in human OCN ( Figure 5A ) . In addition , human OCN does not contain any serine or threonine residues and migrates at a lower molecular weight compared to mouse OCN when expressed and secreted by osteoblasts , HEK293 or CHO cells ( Figure 5B and data not shown ) . Since mouse and human ucOCN have a very similar predicted molecular weight , that is 5 . 1 and 5 . 8 kDa , respectively , these observations suggested that human OCN may not be O-glycosylated . Remarkably , introduction of a single serine residue ( Y12S mutation ) in the human protein is sufficient to induce its O-glycosylation in osteoblasts as visualized by western blot ( Figure 5C ) . In contrast , introducing a leucine at the same position ( Y12L ) did not alter human OCN apparent molecular weight , indicating that this tyrosine residue is not normally subjected to O-glycosylation . Since the apparent molecular weight of both native and Y12S human OCN are increased following treatment with RVKR , we concluded that O-glycosylation does not affect human OCN processing by furin ( Figure 5D ) . These results establish that mature human OCN is not normally subjected to O-glycosylation , but that a single amino acid change ( Y12S ) is sufficient to induce its O-glycosylation in osteoblasts . Because O-glycosylation impacts mouse ucOCN half-life in plasma , we next tested whether this PTM had a similar effect on human ucOCN . We produced and purified O-glycosylated human ucOCNY12S from HEK293 and compared its purity and molecular weight to native non-glycosylated human ucOCN by LC-MS , LC-MS/MS and SDS-PAGE ( Figure 5E and F , and Figure 5—figure supplement 1 ) . Confirming what was observed in osteoblasts , human ucOCN containing the Y12S mutation purified from HEK293 was found to be O-glycosylated ( Figure 5E ) . These proteins were then incubated in Bglap-/- mouse plasma at 37°C and the concentration of ucOCN monitored over time using a specific ELISA assay ( Lacombe et al . , 2020 ) , which can quantify both non-glycosylated and O-glycosylated human ucOCN ( see Figure 5—figure supplement 2 and Materials and method ) . As shown in Figure 5G , in the conditions of this assay , non-glycosylated human ucOCN level declines by 50% within 180 min , while the concentration of the O-glycosylated version remains stable over the course of the experiment ( i . e . 5 hr ) . As observed with the mouse protein , human ucOCN degradation was only inhibited when the plasma was heat-inactivated or incubated at 4°C ( Figure 5H and Figure 5—figure supplement 3 ) , suggesting that glycosylation protects mouse and human ucOCN from degradation through a similar mechanism . In this study we identified O-glycosylation as a novel PTM regulating mouse ucOCN half-life in the circulation . We also showed that O-glycosylation is not found on human mature OCN , but that O-glycosylation of human OCN by means of a single amino acid change can improve its half-life in plasma ex vivo . These findings reveal an important species difference in the regulation of OCN and may also have important implication for the future use of recombinant ucOCN as a therapeutic agent in humans . Numerous secreted proteins are subjected to mucin-type ( GalNAc-type ) O-glycosylation , a PTM which is initiated in the Golgi apparatus and involves multiple sequential glycosylation steps to produce diverse O-glycan structures ( Steentoft et al . , 2013; Bennett et al . , 2012 ) . The initiation of O-glycosylation is catalyzed by the GalNAc-Ts isoenzymes , however , the specific protein sequence ( s ) targeted by each of the GalNAc-Ts remain poorly characterized , although weak acceptor motifs for these have been identified by in vitro analyses ( Perrine et al . , 2009; Gerken et al . , 2006 ) . Here , we demonstrate that mouse OCN is O-glycosylated likely on serine 8 , which is located within the amino acid sequence SVPSPDP11 . Interestingly , this sequence strongly matches the consensus site previously defined for GalNAc-T1 and GalNAc-T2 ( Gerken et al . , 2006 ) . In particular , the presence of proline residues in position −one , +one and +three has been shown to be determinant in the recognition of peptide substrates by GalNAc-T1 and GalNAc-T2 in vitro . We used an isogenic cell library with combinatorial engineering of isoenzyme families to explore the regulation of osteocalcin O-glycosylation by GalNAc-Ts ( Narimatsu et al . , 2019a; Narimatsu et al . , 2019b ) . Combined knock out of GalNAc-T1 , 2 and 3 only partially abolishes mouse OCN O-glycosylation in HEK293 cells , suggesting that OCN may be a substrate for additional GalNAc-Ts . O-glycosylation was shown to interfere with the action of proprotein convertases on several pro-hormones and receptors ( Goth et al . , 2015; Kato et al . , 2006; May et al . , 2003; Schjoldager and Clausen , 2012; Schjoldager et al . , 2010; Goth et al . , 2017 ) . This appears not to be the case for OCN as its O-glycosylation does not interfere with its processing by furin in vitro and in vivo . In other proteins , such as leptin and erythropoietin , glycosylation adducts were shown to increase protein stability and half-life in circulation ( Elliott et al . , 2003; Creus et al . , 2001 ) . More recently , O-glycosylation was shown to protect atrial natriuretic peptide from proteolytic degradation by neprilysin or insulin-degrading enzyme in vitro ( Hansen et al . , 2019 ) . Additional studies showed that sialic acid residues present in the glycosylation adducts increase protein charge , thereby improving serum half-life and decreasing liver and renal clearance ( Morell et al . , 1971; Runkel et al . , 1998; Perlman et al . , 2003; Ziltener et al . , 1994 ) . Although our data suggest that protection from proteolytic cleavage by plasmin in the plasma might be one mechanism by which O-glycosylation extend OCN half-life , we cannot exclude that in vivo , O-glycosylation may also decrease liver and renal clearance of OCN . Mouse OCN contains arginine ( R ) residues at position 16 and 40 , corresponding to R20 and R44 in human OCN , which could be potential cleavage sites for plasmin ( Figure 5A ) . A putative plasmin cleavage site is present in mouse ( i . e . AYK↓R40 , where the arrow indicates the cleavage site ) and human ( i . e . AYR↓R44 ) OCN at the corresponding position ( Backes et al . , 2000 ) . Bovine OCN , like the human protein , does not possess the O-glycosylation site found in mouse , and was shown to be cleaved by plasmin in vitro between R43 and R44 ( Novak et al . , 1997 ) . Here we show that plasmin is sufficient to reduce the concentration non-glycosylated ucOCN , by more than 90% in two hours , confirming that mouse ucOCN is also a plasmin substrate . Supporting a role for plasmin in the cleavage of mouse OCN in vivo , OCN serum levels are decreased in mice deficient in plasmin activator inhibitor I ( PAI-I ) , which have higher plasmin activity ( Tamura et al . , 2013 ) . Interestingly , circulating plasmin activity increases with age in rats and humans ( Paczek et al . , 2009; Paczek et al . , 2008 ) , an observation which could in part explain the gradual reduction of serum OCN concentrations during aging ( Mera et al . , 2016a ) . Notably , a nonsynonymous rare variant ( rs34702397 ) in OCN resulting in the conversion of R43 in a glutamine ( Q ) exists in humans of African ancestry and was nominally associated with insulin sensitivity index and glucose disposal in a small cohort of African Americans ( Das et al . , 2010 ) . Since the R43Q variant eliminates the potential plasmin cleavage site in the C-terminal region , it will be interesting to further investigate if subjects carrying this variant have higher OCN circulating levels . We also show that O-glycosylation protects mouse ucOCN from degradation in INS-1 cell cultures and consequently increases its biological activity . Interestingly , non-glycosylated ucOCN is stable when incubated in the INS-1 cell culture media at 37°C without cells ( data not shown ) , suggesting that the proteolytic activity originates from the INS-1 cells and not from the culture media . The identity of a putative β-cell-derived protease responsible for ucOCN degradation remains unknown , since plasmin is expressed exclusively by the hepatocytes . Nevertheless , these results suggest that it may be more appropriate to use glycosylated ucOCN when studying the function of mouse OCN in cell culture . We found that human OCN is not subjected to O-glycosylation and that consequently it has a reduced half-life in plasma ex vivo . The O-glycosylation sequence ‘SVPSPDP11’ of mouse OCN is conserved in the human protein , except for the amino acids corresponding to S5 and S8 , which are replaced by a proline ( P9 ) and a tyrosine ( Y12 ) ( i . e . ‘PVPYPDP15’ ) . Remarkably , we could introduce O-glycosylation into human OCN by a single amino acid change ( Y12S ) . O-glycosylated human OCN is protected from degradation in plasma ex vivo similarly to the glycosylated mouse protein . Hence , this difference in the O-glycosylation status of OCN could potentially explain why circulating level of OCN in 1- to 6-month-old mice is 5–10 times higher than the level measured in young or adult human ( Table 1 ) . It remains unknown if the increased half-life of O-glycosylated ucOCN will result in improved biological activity in vivo , although it was shown to be the case for other proteins ( Baudys et al . , 1995; Runkel et al . , 1998; Elliott et al . , 2003 ) . Our cell-based assay does suggest that O-glycosylated ucOCN is active , at least on β-cells . We also showed that mouse OCN is endogenously O-glycosylated in vivo and that more than 80% of OCN , including the ucOCN fraction , is O-glycosylated in osteoblast supernatant and in serum . This suggests that the bioactive form of OCN is O-glycosylated in vivo in mice . In summary , this work identified O-glycosylation as a previously unrecognized OCN PTM regulating its half-life in circulation in mice . This modification is not conserved in human , yet introducing O-glycosylation in human ucOCN also increases its half-life in plasma . These findings reveal an important difference between mouse and human OCN biology and also provide an approach to increase recombinant human OCN half-life in vivo , which might be relevant for the future development of OCN-based therapies for human diseases . The Furinfl/fl;BGLAP-Cre mice were generated by breeding Furinfl/fl ( Furintm1Jwmc/Furintm1Jwmc ) mice with BGLAP-Cre ( Tg ( BGLAP-cre ) 1Clem ) transgenic mice that express Cre recombinase under the control of human OCN promoter as described previously ( Al Rifai et al . , 2017 ) . Bglap-/- mice ( Bglap/Bglap2tm1Kry/Bglap/Bglap2tm1Kry ) were generated using homologous recombination to replace OCN1 ( Bglap1 ) and OCN2 ( Bglap2 ) genes in the mouse Bglap cluster with a neomycin resistance cassette ( Ducy et al . , 1996 ) . All strains used in this study were backcrossed on a C57BL/6J genetic background more than 10 times and maintained under 12 hr dark/12 hr light cycles in a specific pathogen–free ( SPF ) animal facility at IRCM . Male mice were used in all experiments , and they were fed a normal chow diet . All animal use complied with the guidelines of the Canadian Committee for Animal Protection and was approved by IRCM Animal Care Committee . Mouse pro-OCN cDNA was cloned into the pIRES2-EGFP-V5 plasmid in EcoRI and AgeI cloning sites . S5A/S8/AT15A pro-OCN , S29A/T36A/T45A pro-OCN and S5S/S8A/T15A/S29A/T36A/T45A ( i . e . 6XST→6XA ) pro-OCN mutant were purchased from Thermo Fisher . Human pre-pro-OCN cDNA cloned into pcDNA3 was purchased from GenScript . Each construct was used as PCR template for amplification and to introduce EcoRI and AgeI cloning sites and cloned in pIRES2-EGFP-V5 plasmid . Point mutations in mouse pro-OCN ( S5A , S8A , T15A ) and Y12S in human pro-OCN were generated by site-directed mutagenesis using specific primer ( Appendix 1—table 1 ) . The cDNA coding of the Fc and hinge region of human immunoglobulin flanked with HindIII-BamHI restriction sites was amplified using standard PCR ( Appendix 1—table 1 ) and pTT5-Fc1_CTL vector as a template ( Saavedra et al . , 2013 ) . The PCR product was cloned in pcDNA3 . 1-myc-His B in HindIII-BamHI cloning site , generating the pcDNA3 . 1-Fc-hinge-myc-His vector . The cDNA coding for mature hOCN ( Y12S ) was generated using pIRES2-EGFP-hOCN ( Y12S ) -V5 as a template , to which a thrombin ( Thr ) cleavage site was added at the N-terminus and BglII and EcoRI restriction sites were introduced by standard PCR amplifications . The Thr-hOCN ( Y12S ) product was cloned in the pcDNA3 . 1-Fc-hinge-myc-His vector . The generated vector pcDNA3 . 1-Fc-hinge-Thr-hOCN ( Y12S ) is an expression vector of human OCN fusion protein composed of the Fc and hinge region of human IgG1 , thrombin cleavage site and human OCN ( Y12S ) . Mouse OCN fused to Fc were generated following the same procedure and using different primers ( Appendix 1—table 1 ) . All cell lines tested negative for mycoplasma . The identity of the cell lines was confirmed by STR when obtained from ATCC or Millipore . Otherwise , the identity was based on the phenotype reported by the providing investigator ( e . g . lack of glycosylation for CHO-ldld , lack of expression of the GALNTs for the knockout HEK293 cell lines ) . Primary osteoblasts were prepared following a previously described protocol ( Ferron et al . , 2015 ) . In brief , calvariae were collected from 3 days old mice and washed with 1 × PBS and digested 2 times for 10 min in digestion solution ( αMEM , 0 . 1 mg/ml collagenase type 2 [Worthington Biochemical Corporation] and 0 . 00075% trypsin ) that was discarded after incubation . Following two 30 min incubations , the digestion solutions were collected , centrifuged and cells recovered were cultured in αMEM supplemented with 10% FBS , penicillin and streptomycin ( PS ) , and L-glutamine . Culture media was supplemented with 5 mM β-glycerophosphate and 100 µg/ml L-ascorbic acid to induce osteoblasts differentiation and it was replaced every 2 days for 21 days . Primary osteoblasts were transfected using jetPRIME Reagent ( Polypus transfection ) . After an overnight incubation , media were changed to secretion media ( FBS-free αMEM plus 2 mM L-glutamine , PS ) . After 24 hr of secretion , media were collected , and cells were lysed in protein lysis buffer ( 20 mM Tris-HC pH 7 . 4 , 150 mM NaCl , 1 mM EDTA , 1 mM EGTA , 1% Triton , 1 mM PMSF , and 1 × protease inhibitor cocktail ) and analyzed by western blotting . In some experiments , osteoblasts were treated with the γ-carboxylation inhibitor warfarin ( 50 μM; Santa Cruz Biotechnology ) , the N-acetylgalactosaminyltransferase inhibitor GalNAc-bn ( 2 mM , Sigma ) or the proprotein convertase inhibitor Dec-RVKR-CMK ( 50 μM , Tocris ) , combined with 22 μM vitamin K1 ( Sigma ) . Chinese hamster ovary ( CHO ) cells , originally purchased from ATCC , and Chinese hamster ovary ldlD cells ( CHO-ldlD; originating from the M . Krieger laboratory Kingsley et al . , 1986 ) were cultured in DMEM-F12 containing PS and 5% FBS for CHO cells or 3% FBS for CHO-ldlD cells and transfected using Lipofectamine 2000 ( Thermo Fisher ) following standard protocol . Secretion was performed in DMEM-F12 media supplemented with PS and 22 µM VK1 . In some experiments , CHO-ldlD culture , transfection and secretion media was supplemented with 0 . 1 mM galactose and/or 1 mM N-acetylgalactosamine ( GalNAc ) to rescue the O-glycosylation defect as previously reported ( Kingsley et al . , 1986 ) . Human embryonic kidney cells HEK293 were originally purchased from ATCC . C1GALT1C1 knockout HEK293sc ( HEK293 simple cell or COSMC ) cells and GALNTs deficient HEK293 cells were generated using Zinc-finger nuclease ( ZFN ) gene editing as described previously ( Goth et al . , 2015; Schjoldager et al . , 2012; Steentoft et al . , 2013; Steentoft et al . , 2011; Goth et al . , 2017 ) . Cells were transfected using Lipofectamine 2000 reagent and secretion was performed over 24 hr in EMEM supplemented with PS and 22 µM VK1 . In some experiments , HEK293 cells were treated with warfarin , GalNAc-bn or Dec-RVKR-CMK combined with 22 μM vitamin K1 . For western blot analysis , proteins were resolved on 15% Tris-tricine gel and blotted overnight with indicated antibody . Antibody used in this study are: anti-V5 ( mouse , clone V5-10 , V8012; Sigma-Aldrich ) , anti–β-actin ( mouse , clone AC-15 , A5441; Sigma-Aldrich ) , anti-GFP ( mouse , clones 7 . 1 and 13 . 1 , 11814460001; Sigma ) , anti-Gla-OCN goat antibody which recognize amino acids 11–26 of carboxylated mature OCN and anti-Cterm OCN goat antibody which recognize amino acids 26–46 of mature mouse OCN ( Ferron et al . , 2010b ) . Flushed mouse femur and tibia from C57BL/6J were homogenized in lysis buffer ( 20 mM Tris-HCl pH 7 . 4 , 150 mM NaCl , 1 mM EDTA , 1 mM EGTA , 1% Triton , 1 mM PMSF , and 1 × protease inhibitors cocktail ) . Tissue homogenates were then centrifuged for 10 min at 4000 rpm to remove insoluble material . In-vitro de-glycosylation assay was performed on 10 μg of bone homogenate . Briefly , proteins were denatured in denaturing buffer ( 0 . 5% SDS , 40 mM DTT ) at 95°C for 5 min and incubated with 80000 units of O-glycosidase and 100 units of neuraminidases for 4 hr at 37 °C following the NEB kit protocol ( E0540S; NEB ) . Samples were resolved on 15% Tris-tricine SDS-PAGE gel and blotted using anti-Cterm OCN goat antibody . Differentiated osteoblast secretion medium was spun down at 1500 rpm for 5 min to remove cells debris . The supernatant was then incubated overnight at 4°C with anti-Cterm OCN antibody in the presence of 1 × protease inhibitors cocktail . For bone extract , flushed femur and tibia from wild type mice were homogenized in lysis buffer containing ( 20 mM Tris-HCl pH 7 . 4 , 150 mM NaCl , 1 mM EDTA , 1 mM EGTA , 1% Triton , 1 mM PMSF , and 1 × protease inhibitors cocktail ) . 100 μg of protein homogenate was diluted in 1 , 6 ml of 100 mM phosphate buffer pH 7 . 4 and incubated overnight at 4°C with anti-OCN antibody . After overnight incubation , samples were centrifuged at 10000 rpm for 10 min to remove precipitate and supernatant was incubated with protein-G agarose beads pre-washed with 1X PBS . After for 4 hr of rotation at 4°C , beads were spun down , washed twice with 1X PBS and three times with 50 mM Ammonium Bicarbonate pH 8 . 0 . OCN was then eluted with 100 μl of 0 . 5 M NH4OH , snap frozen in liquid nitrogen and evaporated under vacuum using speedvac concentrator ( Thermo scientific ) . Samples were diluted in 25% ACN 0 . 3%TFA and loaded onto a 50 × 4 . 6 mm PLRP-S 300A column ( Agilent Technologies ) connected to an Accela pump ( Thermo Scientific ) and a RTC autosampler ( Pal systems ) . The buffers used for chromatography were 0 . 1% formic acid ( buffer A ) and 100% acetonitrile/0 . 1% formic acid ( buffer B ) . Proteins and peptides were eluted with a two slopes gradient at a flowrate of 120 µL/min . Solvent B first increased from 12% to 50% in 4 . 5 min and then from 50% to 70% in 1 . 5 min . The HPLC system was coupled to a Q Exactive mass spectrometer or an Orbitrap Fusion ( Thermo Scientific ) through an Ion Max electrospray Ion Source equipped with a HESI-II probe and operated in positive ion mode . The spray and S-lens voltages were set to 3 . 6 kV and 60 V , respectively . Capillary temperature was set to 225°C . Full scan MS survey spectra ( m/z 600–2000 ) in profile mode were acquired in the Orbitrap with a resolution of 70 , 000 or 120 , 000 with a target value at 3e6 . The four most intense protein/peptide ions were fragmented in the HCD ( higher-energy collision dissociation ) collision cell upon collision with nitrogen gas and analyzed in the Orbitrap with a target value at 5e5 and a normalized collision energy at 33 . The data acquisition software were Xcalibur 3 . 1 and Q Exactive 2 . 8 SP1 for the Q Exactive instrument and Xcalibur 4 . 0 and Tune 2 . 0 for the Orbitrap Fusion instrument . Data processing protocol: the identification of the different forms of OCN was performed by manual denovo sequencing using Qual Browser ( Xcalibur 4 . 0 ) . The source files for the proteomics analyses can be downloaded through this link: https://doi . org/10 . 6084/m9 . figshare . 13259891 . v1 . RNA was extracted from non-differentiated and differentiated calvariae osteoblasts using Trizol reagent ( Thermo Fisher Scientific ) following standard protocol . RNA was treated with DNAse I and reverse transcribed using poly dT primers , random primers and MMLV reverse transcriptase ( Thermo Fisher Scientific ) . QPCR was performed on standards of diluted genomic DNA and cDNA products using specific primers ( Appendix 1—table 1 ) on a ViiA 7 Real-Time PCR system ( Thermo Fisher Scientific ) . Galnts gene copy numbers were calculated using the genomic DNA as a standard curve and variation between biological replicate was normalized using Actb expression level . To generate stable clonal cell lines expressing glycosylated human and mouse OCN , HEK293 were transfected with pcDNA3 . 1-Fc-hinge-Thr-hOCNY12S and pcDNA3 . 1-Fc-hinge-Thr-OCN respectively using Lipofectamine 2000 reagent . Following 48 hr of transfection , cells were trypsinized and resuspended in sorting buffer containing ( 1 × sterile PBS , 2% FBS and 1 mM EDTA ) . Cells were sorted at a concentration of 5–10 cells/well in 96 well plates containing the selection media EMEM , 10% FBS supplemented with G418 sulfate ( 500 μg/ml; Wisent ) . Following two weeks of selection , isolated colonies appeared and the expression of mouse and human OCN was assessed using ELISA assay described below . Clones expressing the highest levels of OCN were amplified and frozen . TM102F12 clone expressing IgFc-mOCN fusion protein and 22H5 clone expressing IgFc-hOCNY12S fusion protein were cultured in triple layer 175 cm2 flasks . After reaching 100% confluency , cells were kept in secretion media ( EMEM media supplemented with 1% FBS and 10 μM warfarin to block γ-carboxylation ) for 72 hr . Secretion media was collected , filtered with 0 . 45 μm filter , and media was pH buffered with 10 × binding buffer ( 0 . 2 M phosphate buffer , pH 7 ) . This cell supernatant was then loaded into protein A affinity column ( HiTrap protein A high performance , GE29-0485-76; GE Healthcare Life Sciences , ) using liquid chromatography system ( GE AKTA Prime Plus ) . Column was then washed with 20 ml of 1 × binding buffer ( 0 . 02 M phosphate buffer , pH 7 ) and 5 ml of filtered 1 × PBS . To release OCN from the column , OCN fusion protein was digested with thrombin ( 27-0846-01 , GE Healthcare Life Sciences ) and eluted with 1 × PBS . Thrombin was subsequently removed using benzamidine sepharose ( 17-5123-10 , GE healthcare ) . Mouse and human OCN purity were assessed using Coomassie staining and liquid chromatography-mass spectrometry ( LC-MS ) analysis compared to purified non-glycosylated mouse or human ucOCN . Mouse OCN was quantified using ELISA assay as described previously ( Ferron et al . , 2010b ) . Human ucOCN measurements were performed using a commercially available human ucOCN ELISA ( BioLegend , 446707 ) ( Lacombe et al . , 2020 ) , which recognizes equally glycosylated and non-glycosylated human OCN ( Figure 5—figure supplement 2 ) . The capture antibody in this assay is a mouse monoclonal antibody ( 8H4 ) specific to the C-terminal region of human OCN ( i . e . amino acids 30 to 49 ) . The detection antibody is a mouse monoclonal antibody ( 4B6 ) specific to the mid-region of human ucOCN ( i . e . amino acids 12 to 28 in ucOCN ) . The mouse OCN ex vivo half-life assays were performed with fresh plasma ( lithium heparin ) collected from four independent Bglap-/- mice . Glycosylated OCN produced in HEK293 cells and non-glycosylated OCN , produced in bacteria as previously described ( Ferron et al . , 2012 ) , were incubated at 100 ng/ml in plasma at 37°C and OCN level was measured at indicated time points using the total mouse OCN ELISA assay described previously ( Ferron et al . , 2010b ) . In this assay , the capture antibody is a goat polyclonal antibody directed against the central part ( amino acids 11 to 26 ) of mouse OCN ( anti-MID OCN ) and recognizing with equal affinity the uncarboxylated and carboxylated OCN proteins . The detection antibody is a goat polyclonal antibody directed against the C-terminal region ( amino acids 26 to 46 ) of mouse OCN ( anti-CTERM OCN ) . This assay detects with the same sensitivity glycosylated and non-glycosylated mouse OCN at concentration of 50 ng/ml or less ( Figure 3—figure supplement 2 ) . Therefore , all samples were diluted to be in this range before being measured by the ELISA . Human OCN half-life assay was performed ex vivo using Bglap-/- mice plasma and human OCN at 60 ng/ml . Human OCN level was measured at different time point using the human ucOCN ELISA described above . At the experiment start point ( T0 ) , an aliquot of plasma was diluted in ELISA assay buffer and kept on ice for OCN measurements . In some experiments , plasma was heat-inactivated for 30 min at 56°C , or treated with EDTA ( 10 mM , Wisent ) , phenylmethylsulfonyl fluoride ( PMSF , 10 mM , Sigma ) , pepstatin A ( Pep A , 10 μM , Sigma ) which inhibits aspartic proteases ( pepsin , cathepsin D , renin , chymosin ) , RVKR ( Dec-RVKR-CMK , 50 μM; Tocris ) or talabostat ( 10 mM; Tocris ) . Results are calculated in percentage relative to the initial concentration of non-glycosylated and glycosylated OCN respectively . The plasmin enzymatic assay was performed on plasma ex vivo in the presence of non-glycosylated OCN and glycosylated OCN as follows: plasma from Bglap-/- mice was heat-inactivated for 30 min at 56°C , then diluted two times in the enzyme assay buffer containing 100 mM Tris buffer pH 7 . 5 . Non-glycosylated OCN or glycosylated OCN was incubated at 50 ng/ml in the diluted plasma for two hours in the presence of different concentration of human plasmin ( Sigma ) ranging from 0 to 0 . 3 U/ml . After two hours , OCN level was measured using total mouse OCN ELISA . Results are represented as percentage of initial concentration of non-glycosylated and glycosylated OCN respectively . For in vivo half-life assay , Bglap-/- male mice were injected intraperitoneally with 40 or 80 ng/g of body weight of mouse O-glycosylated ucOCN or non-glycosylated ucOCN , serum OCN level was analyzed at indicated time points using total mouse OCN ELISA . In all ex vivo and in vivo studies , mouse or human proteins were prepared in saline solution ( 0 . 9% NaCl ) containing BSA ( 35 mg/ml ) as a carrier . The INS-1 832/3 rat insulinoma cell line ( Millipore-Sigma ) was cultured in RPMI-1640 with 11 . 1 mM D-glucose , supplemented with 2 mM L-glutamine , 1 mM sodium pyruvate , 10 mM HEPES , 0 . 05 mM β-mercaptoethanol and 10% FBS ( Hohmeier et al . , 2000 ) . On day 1 , cells were plated at 3 × 105 cells per well in 12-well plates and on day 5 , cells were washed with PBS and incubated with culture media containing 5 mM D-glucose . Twenty-four hours later , cells were washed with PBS and incubated for 4 hr in culture media containing 1% FBS , 5 mM D-glucose and 0 . 1% BSA . OCN was added for 8 hr before supernatants and cells were harvested . ucOCN and O-gly ucOCN concentration were measured in supernatants by ELISA and cells were lyzed and RNA extracted as described in the material and methods section . Gene expression was analyzed using rat specific primers for Ins1 and Gapdh . Statistical analyses were performed using GraphPad Prism software version 7 . 03 . Results are shown as the mean ± SEM . For single measurements , an unpaired , 2-tailed Student’s t test was used , while one-way or two-way ANOVA followed by Bonferroni’s post-test were used for comparison of more than two groups . For repeated measurements ( e . g . half-life study ex vivo and in vivo ) , a repeated-measurement two-way ANOVA followed by Bonferroni’s post-test was used . A P value of less than 0 . 05 was considered statistically significant . All experiments were repeated at least three times or performed on at least three independent animals or included at least three independent biological replicates . In the in vivo half-life experiments , the Bglap-/- mice were randomized before the experiment into experimental groups such that the average body weight of each group was similar .	Bones provide support and protection for organs in the body . However , over the last 15 years researchers have discovered that bones also release chemicals known as hormones , which can travel to other parts of the body and cause an effect . The cells responsible for making bone , known as osteoblasts , produce a hormone called osteocalcin which communicates with a number of different organs , including the pancreas and brain . When osteocalcin reaches the pancreas , it promotes the release of another hormone called insulin which helps regulate the levels of sugar in the blood . Osteocalcin also travels to other organs such as muscle , where it helps to degrade fats and sugars that can be converted into energy . It also has beneficial effects on the brain , and has been shown to aid memory and reduce depression . Osteocalcin has largely been studied in mice where levels are five to ten times higher than in humans . But it is unclear why this difference exists or how it alters the role of osteocalcin in humans . To answer this question , Al Rifai et al . used a range of experimental techniques to compare the structure and activity of osteocalcin in mice and humans . The experiments showed that mouse osteocalcin has a group of sugars attached to its protein structure , which prevent the hormone from being degraded by an enzyme in the blood . Human osteocalcin has a slightly different protein sequence and is therefore unable to bind to this sugar group . As a result , the osteocalcin molecules in humans are less stable and cannot last as long in the blood . Al Rifai et al . showed that when human osteocalcin was modified so the sugar group could attach , the hormone was able to stick around for much longer and reach higher levels when added to blood in the laboratory . These findings show how osteocalcin differs between human and mice . Understanding this difference is important as the effects of osteocalcin mean this hormone can be used to treat diabetes and brain disorders . Furthermore , the results reveal how the stability of osteocalcin could be improved in humans , which could potentially enhance its therapeutic effect .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "medicine" ]	2020	The half-life of the bone-derived hormone osteocalcin is regulated through O-glycosylation in mice, but not in humans
Rotavirus genome replication and assembly take place in cytoplasmic electron dense inclusions termed viroplasms ( VPs ) . Previous conventional optical microscopy studies observing the intracellular distribution of rotavirus proteins and their organization in VPs have lacked molecular-scale spatial resolution , due to inherent spatial resolution constraints . In this work we employed super-resolution microscopy to reveal the nanometric-scale organization of VPs formed during rotavirus infection , and quantitatively describe the structural organization of seven viral proteins within and around the VPs . The observed viral components are spatially organized as five concentric layers , in which NSP5 localizes at the center of the VPs , surrounded by a layer of NSP2 and NSP4 proteins , followed by an intermediate zone comprised of the VP1 , VP2 , VP6 . In the outermost zone , we observed a ring of VP4 and finally a layer of VP7 . These findings show that rotavirus VPs are highly organized organelles . Rotavirus is a non-enveloped virus composed of three concentric layers of proteins that enclose a genome constituted by eleven segments of double stranded RNA ( dsRNA ) that encode six structural proteins ( VP1 to VP4 , VP6 and VP7 ) and six non-structural proteins ( NSP1 to NSP6 ) . The inner layer is formed by dimers of VP2 that enclose the viral genome and small numbers of molecules of the viral RNA-dependent RNA polymerase ( RdRp ) , VP1 , and the capping enzyme , VP3 . This nucleoprotein complex constitutes the core of the virus , which is surrounded by an intermediate protein layer of trimers of VP6 , to form double-layered particles ( DLPs ) . The surface of the virion is occupied by two polypeptides , VP7 , a glycoprotein , and VP4 , which forms spikes that protrude from the VP7 shell ( Estes and Greenberg , 2013 ) . Replication of the rotavirus genome and assembly of DLPs take place in cytoplasmic electron dense inclusions termed viroplasms ( VPs ) ( Estes and Greenberg , 2013 ) . Once the double-shelled particles are assembled , they bud from the cytoplasmic VPs into the adjacent endoplasmic reticulum ( ER ) . During this process , which is mediated by the interaction of DLPs with the ER transmembrane viral protein NSP4 , the particles acquire a temporary lipid bilayer , modified by VP7 and NSP4 , which after being removed in the lumen of the ER by an unknown mechanism , yields the mature triple-layered virions ( Estes and Greenberg , 2013 ) . It has been reported that VP4 is located between the VP and the ER membrane and it is incorporated into triple-layered particles ( TLPs ) during the budding process and maturation of the virus particle inside the ER ( Estes and Greenberg , 2013; Navarro et al . , 2016 ) . The viral non-structural proteins NSP2 and NSP5 serve a nucleation role that is essential for the biogenesis of VPs ( Fabbretti et al . , 1999; Silvestri et al . , 2004; Vascotto et al . , 2004; Campagna et al . , 2005 ) . In addition to viral proteins and genomic dsRNA , cellular proteins such as ER chaperones ( Maruri-Avidal et al . , 2008 ) , proteins associated with lipid droplets ( Cheung et al . , 2010 ) , and ribonuclear proteins ( Dhillon et al . , 2018 ) , have been shown to colocalize with VPs . Several studies have characterized the intracellular distribution of the rotavirus proteins ( González et al . , 2000; Petrie et al . , 1982; Petrie et al . , 1984; Richardson et al . , 1986 ) . Immunofluorescence studies , based upon epifluorescence or confocal microscopy , have described the viral proteins that conform the VPs , however the images are inherently diffraction-limited to a spatial resolution in the range of hundreds of nanometer , precluding the identification of the nanoscopic molecular scale organization of VPs ( González et al . , 1998; González et al . , 2000; Eichwald et al . , 2004; López et al . , 2005b; Criglar et al . , 2014; Martin et al . , 2011; Contin et al . , 2010 ) . On the other hand , transmission electron microscopy ( TEM ) studies often provide images with nanometric resolution , nevertheless , immunoelectron microscopy is challenging when looking for the localization of more than a single protein ( Altenburg et al . , 1980; Petrie et al . , 1982; Petrie et al . , 1984 ) . Over the past 15 years , a variety of super-resolution microscopy ( SRM ) techniques have been developed to observe subcellular structures beneath the diffraction limit of optical microscopes , with resolutions in the tens of nanometers ( Schnitzbauer et al . , 2017; Deschout et al . , 2014; Cox et al . , 2011 ) . In this work , we determined the organization of rotaviral proteins within and around VPs through the ‘Bayesian Blinking and Bleaching’ ( 3B ) SRM technique . We developed a segmentation algorithm to automatically analyze and quantify the relative distribution of seven viral proteins , and propose a model that describes their relative spatial distribution . Also , we present a dependency model that explains the relationship between the viral proteins . This work establishes a structural framework for VP organization that future mechanistic and functional studies must take into account , and establishes key methodologies for future investigations on this subject . Rotavirus VPs are complex signaling hubs composed of viral and cellular proteins , packed together with viral RNAs . By TEM , they roughly resemble circular electrodense structures whose internal components lack an obvious degree of spatial organization ( Altenburg et al . , 1980; Eichwald et al . , 2012 ) . In this work , we determined the relative spatial distribution of VPs components by immunofluorescence and SRM in MA104 cells infected with the rhesus rotavirus strain RRV at 6 hr post-infection ( hpi ) , using protein-specific antibodies . Due to their important role as nucleating factors during VP biogenesis , we selected either NSP2 or NSP5 as spatial relative reference for the distribution of the VP1 , VP2 and VP6 proteins . VPs were optically sectioned through total internal reflection fluorescence microscopy ( TIRF ) , with an excitation depth of field restricted to 200 n⁢m from the coverslip . This approach avoids excitation of fluorophores marking structural components located away from this plane , that is towards the inner cellular milieu . Additionally , NSP2 was also co-immunostained with the viral outer layer protein VP4 as well as with the ER resident proteins NSP4 and VP7 , all of which have been reported to form separate ring-like structures that closely associate with VPs ( González et al . , 2000 ) . In order to gain more insight into the morphogenesis of rotavirus , we analyzed the distribution of both VP7 monomers ( VP7-Mon ) and trimers ( VP7-Tri ) since this protein is assembled into virus particles in the latter form ( Kabcenell et al . , 1988 ) . The nanoscale distribution of VPs was then analyzed through 3B-SRM , with improvements in the technique , developed in the present work , to solve nanoscopic structures ( ‘Stochastic model fitted for 3B super resolution microscopy’Appendix 1 ) . By different methods of analysis VPs exhibit roughly a circular shape ( Figure 1A–E ) . However , unlike the diffraction-limited image ( Figure 1B ) , in super-resolution microscopy structural details of VP are appreciated , like the different layer distributions of viral components with respect to NSP2 ( Figure 1C–E ) . In addition to VPs , by diffraction-limited TIRF microscopy we detected in the cytoplasm several small and dispersed puncta of fluorescence ( Figure 1B ) , and in these images it is also sometimes possible to differentiate the distribution of NSP2 from that of VP4 , a closely viroplasm-associated viral protein ( see also González et al . , 2000 ) ; in this case , VP4 is detected as a ring-like structure that surrounds the VP . Nevertheless , the small size of the VPs effectively precludes measurement of component distribution for the majority of its structural elements , as their separation is below the spatial resolution of typical optical microscopes . In contrast , images obtained by 3B-SRM do allow the study of the relative distribution of the VP components ( Figure 1C–E ) . In the case of SRM images of VP4 ( Figure 1C ) , we observed that this protein forms a ring-like structure that does not colocalize with NSP2 , and also ribbon-like projections that extend towards the cytoplasm , details that were not apparent in images captured with conventional fluorescence microscopy ( Figure 1B ) . Additionally , we observed that the small puncta of proteins detected in the cytoplasm were in fact ribbon-like structures composed of various viral proteins that may represent different organization forms of the viroplasmic proteins ( Figure 1C ) . In this regard , it is interesting to note that both NSP2 and VP4 have been reported to have at least two different intracellular distributions ( González et al . , 2000; Nejmeddine et al . , 2000; Criglar et al . , 2014 ) . An examination of 3B-SRM images of VPs ( Figure 1C–E ) revealed that the viral components form ring like structures within the VPs and are arrayed as rather discrete concentric layers . As seen in Figure 1C–E , we find that although the structural proteins VP1 , VP2 and VP6 partially overlap in position with NSP2 , the bulk of the proteins form separate and distinct layers . Also , the monomeric as well as the trimeric forms of VP7 are clearly distinguished from NSP2 , forming an outer ring . Of interest , the spatial distribution of NSP4 colocalized with that of NSP2 , an unexpected result since , as mentioned , NSP4 is an ER integral membrane protein ( see the Discussion section ) , and as such it was expected to colocalize with VP7 rather than with an internal viroplasmic protein ( Petrie et al . , 1984 ) . With regard to NSP5 , it was observed distributed inside the ring formed by NSP2 ( Figure 1E ) . A qualitative analysis of the distribution of the VP components through 3B-SRM suggested that these are arranged as concentric spherical shells; thus , we set out to quantitatively validate the circularity of the VP shape . For this , we developed a segmentation algorithm based on a least squares approach , which we called ‘Viroplasm Direct Least Squares Fitting Circumference’ ( VP-DLSFC ) ( see ‘Segmentation Algorithm’ in Appendix 1 ) , to measure the spatial distribution of the components within individual VPs by adjusting concentric circumferences . This method is automatic , deterministic , easy to implement , and has a linear computational complexity . The performance of VP-DLSFC was tested on approximately 40 , 000 ‘ground truth’ ( GT ) synthetic images , showing a high robustness to noise and partial occlusion scenarios . Additionally , we compared our method with two other alternative methods ( Gander et al . , 1994 ) , and our approach displayed an improved performance ( see ‘Algorithm Validation’ in Appendix 1 ) . Based on this new algorithm , we find that the mean radius of the NSP5 distribution was smaller than that of NSP2 , suggesting that NSP5 is located in the innermost section , as a component of the core of VPs ( Figure 2A ) . On the other hand , the distribution of the structural proteins VP1 , VP2 and VP6 exhibit slightly larger mean radii than that of NSP2 , and are thus primarily localized in a zone surrounding NSP2 . Continuing further towards the outer regions of the VP , we observed a region occupied by the spike protein VP4 . Finally , the two different forms of VP7 ( VP7-Mon and VP7-Tri ) were located together , close to the most external region of the VPs ( Figure 2A ) . The distribution of the glycoprotein NSP4 showed a similar mean radius to that of NSP2 ( around 0 . 4⁢μ⁢m ) suggesting , as described above , that these two proteins are located in the same structural layer of the VP ( Figure 2A ) . In order to confirm our preliminary observations and clarify the nanoscopic organization of the VPs , we evaluated the relative separation between NSP2 and each accompanying protein . Again , the results show a remarkable degree of organization in the structure of the VP ( Figure 2B ) . As predicted from Figure 2A , we found that NSP5 is located in the internal part of the VP , in close proximity ( ≈0 . 05⁢μ⁢m ) to the area occupied by proteins NSP2 and NSP4 , which themselves show the closest association . After the NSP2-NSP4 region , VP6 occupies a middle region at ≈0 . 05⁢μ⁢m from NSP2 , followed by the VP4 protein , which were located at a distance of ≈0 . 18⁢μ⁢m . Finally , the VP7-Mon and VP7-Tri were situated at ≈0 . 38⁢μ⁢m from NSP2 ( Figure 2B ) . A Mann-Whitney test showed that the distances of the various viral components in relation to NSP2 were significantly different ( Figure 2B ) , suggesting that they are situated in specific areas of the VPs . The two forms of VP7 were located at the same distance to NSP2 , suggesting that the formation of trimers of VP7 takes place at the ER membrane , where the VP7 monomers should also be located . Note that in Figure 2B the relative distance of VP1 and VP2 to NSP2 was not included , since the radii obtained for NSP2 in these two combinations were significantly smaller than those found when it was determined in combination with the other VP components ( see ‘Supplementary Exploratory Analysis’ ) . In addition to this , we found no significant differences between the distance of both VP1 and VP2 to NSP2 ( Figure 2C ) . Nonetheless , based on the inferential analysis , we could place these two proteins in the same layer as VP6 ( see below ) . Next , through a hierarchical cluster analysis , we studied the relationship between the components of the VP , taking into account multiple variables at the same time , like the mean distance to NSP2 [‘Mean ( Dist ) ”] , the standard deviation of the distance to NSP2 [‘Std ( Dist ) ”] , the mean radius of NSP2 [‘Mean ( NSP2 ) ”] , and the radii of the other proteins [‘Mean ( Other ) ”] ( Figure 2D ) . Note that the proteins within a cluster should be as similar as possible and proteins in one cluster should be as dissimilar as possible from proteins in another . Because our variables are related with the distance to NSP2 and the radii of the proteins , this is a no-parametric analysis that should provide evidence about the spatial distribution/order of the viral proteins into the VP . As we are considering the distance to NSP2 , VP1 and VP2 were not included in this analysis . The first level of the hierarchical agglomerative cluster ( Figure 2D , left ) partitioned the VPs and the surrounding proteins in five clusters , composed by NSP4 , NSP5 , VP6 , VP4 and {VP7-Mon , VP7-Tri} , which suggest that these five proteins compose different layers of the VP . The second agglomerative level merged into the same group the proteins NSP4 and NSP5 , meanwhile VP6 and VP4 continue as independent clusters , which indicate that NSP5 and NSP4 are closer to each other than to VP6 and VP4 in the VP . In the third level , VP6 and VP4 are clustered in the same group , and as consequence are more related between them than with the others viral proteins . The subsequent groups in the clustering analysis indicate that VP7 remains as an independent layer with respect to the other proteins . Based on this analysis , the viral proteins seem to be highly organized , with VP7 conforming the most external layer , while NSP5 , NSP4 , VP6 and VP4 are distributed very close but as independent layers . The clusters between NSP5-NSP4 and VP6-VP4 suggest that these two pairs of proteins ( in each cluster ) conform continuous layers in the VP . The scatterplot between the radius of the spatial distribution of NSP2 ( independent variable , x-axis ) and the radius of the distribution of other viral components ( response variable , y-axis ) showed a strong linear relationship ( Figure 3A ) . The distribution of NSP5 grows 0 . 87⁢μ⁢m for each 1⁢μ⁢m increase in the radius of NSP2 ( slope interpretation ) , whereas the radius of the distribution of NSP4 increases 0 . 99⁢μ⁢m ( Figure 3B ) . These findings indicate that NSP5 is distributed in a proportionally smaller region than NSP2 regardless of the absolute size of the VP , supporting our observation that NSP5 is a constituent of the core of the VP . Moreover , the fact that the increase in the radius of the fitted distribution of NSP4 is directly proportional to the same parameter measured for NSP2 supports the idea that these proteins are both constituents of a putative second layer . VP1 , VP2 and VP6 exhibit similar slopes which diverge between 0 . 03 and 0 . 05 μm ( Figure 3B and Appendix 1—table 6 ) ; thus , these results confirm that VP1 , VP2 and VP6 are components of the same layer in the VPs which , from the data in Figure 3 , is located just after the layer of NSP2 and NSP4 . Finally , as noted in our quantitative analysis , VP4 and VP7 form consecutive external layers with a slope of 1 . 39 and 1 . 94 μm , respectively ( Figure 3B and Appendix 1—table 6 ) . These findings indicate that the spatial distribution of the viral components in the VPs and in the surrounding areas is conserved regardless of their absolute size , and also form the basis of a predictive model , where , for a given radius of distribution of NSP2 , it is possible to predict the radii of the remaining VP components ( NSP5 , NSP4 , VP1 , VP2 , VP6 ) and of VP4 and VP7 proteins . This predictive model is available as a web app at https://yasel . shinyapps . io/Nanoscale_organization_of_rotavirus_replication_machineries/ . The mathematical details and the residual analysis that validate these linear models are available in the Appendix 1 , section ‘Linear dependency between the viral components’ , Appendix 1—table 6 and Appendix 1—figure 9 . In order to confirm the observed structural organization of VPs , we analyzed two more experimental conditions in which we chose a different reference protein for pairwise comparisons . The first was based on the distribution of NSP5 and its comparison with the relative localizations of VP6 and VP4 , and the second considered NSP4 as the reference protein to compare with the distribution of VP6 . We found that both analyses produced an identical structural organization for the VPs , with a comparative localization error of approximately 0 . 05⁢μ⁢m between models ( close to the effective resolution limit of the 3B algorithm; see ‘NSP5 and NSP4 as reference proteins’ in Appendix 1 ) . An extensive quantitative validation regarding the congruence between the NSP2 , NSP5 and NSP4 models is available in the Appendix 1 . Based on our extensive quantitative , descriptive and inferential statistical analyses , we propose that the VP and the surrounding viral proteins form an ordered biological structure composed of at least five concentric layers organized as depicted in Figure 4 . In this structure , NSP5 constitutes the innermost layer , followed by a {NSP2-NSP4} layer . Then , there is a layer composed by {VP1-VP2-VP6} and two consecutive external layers formed by VP4 and VP7 . The different layers of proteins are most likely highly porous to allow the entry of positive-sense single-stranded viral RNA ( +RNA ) during genome replication and also of the antibodies used for VP staining . VPs have been previously studied using electron and fluorescence microscopy , however , due to the limited resolution of classic fluorescence microscopy techniques , and the difficulty of analysis of immunoelectron microscopy , the existence of any complex structural organization of the viral elements inside the VPs has not been reported . In recent years , the development of SRM has facilitated research into the nanoscale organization of a diverse range of cellular structures ( Grant et al . , 2018; Reznikov et al . , 2018 ) , however , until now SRM had not been applied to study the replication cycle of rotavirus . In this work , thanks to the use of the 3B SRM algorithm , we visualized and determined quantitatively the location of several viroplasmic proteins , leading us to propose a detailed model of the VP that should be of great value for understanding virus morphogenesis . Other SRM algorithms had been used to study the organization of viral and cellular structures showing concentric arrangements , as those proposed by Laine et al . ( 2015 ) and Manetsberger et al . ( 2015 ) . The main similarities between those studies and our approach is the use of conics , such as circles or ellipses , to fitting structures showing concentric organization . The method provided by Manetsberger et al . ( 2015 ) could actually be implemented to analyze our data set , which as outcome will produce similar results . This method could also provide information about the degree of asymmetry within the VP , which may be valuable to establish functional relationships between the protein distribution belts that shape these intriguing structures . The selection of the 3B SRM algorithm over other super-resolution approaches was based on the fact that this method allows to deal with samples with high density of labeling , obtaining data with a reasonable resolution , although at the cost of higher computational effort . The quantification of the viral protein distribution within the VPs was possible thanks to a novel segmentation algorithm ( VPs-DLSFC ) that was proven to be robust and efficient in noisy and partial occlusion scenarios . The manual pre-segmentation step of this algorithm was necessary in our case because we did not want to introduce any bias in the isolation of the VPs through an automatic approach . Setting aside the manual pre-segmentation step , the VPs-DLSFC algorithm is automatic , deterministic , non-iterative and has a linear computational complexity . Previous reports have suggested that VPs have a spherical-like structure ( Eichwald et al . , 2004; Cabral-Romero and Padilla-Noriega , 2006; Campagna et al . , 2013 ) ; in this study we confirmed this suggestion by comparing the VPs-DLSFC approach with a similar approach based on an ellipse adjustment ( Garcés et al . , 2016 ) . The results showed no significant statistical differences between these two models , and as consequence we can confidently model the structure of the VPs as a circumference . We also ratified that the center displacement of the circumferences that adjust two paired proteins are not statistically different . Our study indicates that the viral components in the VPs , as well as VP4 and VP7 , are arranged as largely discrete concentric layers ( note that we are describing the structure of viroplasms , not of virus particles ) . This organization , however , does not preclude the interaction among the VP components proposed in this model as being located in separate layers since , for instance , NSP5 has been shown by different biochemical methods to interact with NSP2 ( Eichwald et al . , 2004; Poncet et al . , 1997; Afrikanova et al . , 1998; Jiang et al . , 2006 ) , VP1 ( Afrikanova et al . , 1998 ) and VP2 ( Berois et al . , 2003 ) . In this regard , based on the super resolution microscopy images , it seems clear that there is some overlapping between different protein layers , as is the case for NSP2 and NSP5 in Figure 1E , but also between NSP2 and VP1 , VP2 , VP6 and NSP4 in Figure 1 . These general overlapping zones between different proteins most likely are relevant for coordinating the genome replication and virion assembly , as suggested . Of interest , we observed the presence of protein projections ( ‘spike-like’ ) from different viral shells that could also contribute to the interaction of proteins mapped to different layers ( Appendix 1—figure 8 ) . Although our present analysis is limited to a general characterization of the spatial distribution of the viral proteins within VPs , and not to understand specific details about the interactions between proteins in different layers , it could be used as departure point to analyze these interactions . Taking as initial solution the result of the algorithm VPs-DLSFC and the SRM image , it is possible to employ other segmentation approaches , like deformable/active contours ( snakes ) ( Kass et al . , 1988 ) , level-set ( Osher and Fedkiw , 2003 ) , or region growing methods ( Mehnert and Jackway , 1997; Synthuja et al . , 2012 ) , to evolve the circular contour and fit precisely the spatial distribution of the viral proteins . Then , establishing a polar coordinate system in the VP , and considering the results of both segmentation algorithms , it would be possible to quantify the radial angle in which a spike from the central distribution of a viral protein that interacts with a different protein exists . It would be also possible to determine how strong these interactions are ( intersection between two segmentation curves ) , and to study whether the spikes are randomly distributed between layers or a specific pattern in the connection between different protein layers exists . In the latter case , this could allow us to explore whether these patterns influence the assembly of the virus-like particles or only provide a skeleton that maintain the structure of the VP . The results obtained could also be used to study topological changes that the VP might experience at different times post infection , and associate these changes with maturation of the subviral particles . In this regard , in preliminary experiments carried out at three hpi , the viral proteins in he VP have been found to have a similar ‘layered’ organization as shown for the mature VP at six hpi ( data not shown ) . This observation indicates that this organization is already present when the formation of viral particles has not yet taken place , suggesting that it might be important for the assembly of DLPs within VPs . In an additional application , SRM could also be used to observe the assembly of the virus particles and the interactions that may occur of these particles in different layers of the VP . Nevertheless , to develop this idea it would be important to establish an experimental protocol to observe the viral particles during the early stages of the assembly process , to distinguish simultaneously the layers of the viroplasm and the viral particles , and to collect the SRM images with a very short acquisition rate and a very high resolution ( 25–30 nm ) , which makes this experimental plan a challenge . Previous studies based on conventional microscopy techniques have reported that NSP5 and NSP2 colocalize ( González et al . , 2000; Eichwald et al . , 2004; Fabbretti et al . , 1999 ) ; in contrast , we found that although NSP5 and NSP2 are located in close proximity , their positions in the VP were separable . This difference is attributable to the increased spatial resolution in the final image created by the super-resolution techniques employed in our study . Here , NSP5 was found to represent the innermost layer of the VPs , suggesting that this protein might serve as the core scaffold upon which the subsequent viroplasmic proteins are assembled to form the VPs . This finding contrasts with a report by Eichwald et al . ( 2004 ) , who described that NSP5 locates to a region external to NSP2 . In addition to the superior spatial resolution obtainable through 3B-SRM , compared to the traditional confocal microscopy employed in the previous report , the difference might be due to the fact that in our study we characterized the endogenous structures produced during virus replication , while Eichwald et al . characterized VP-like structures formed by transiently expressed proteins fused to GFP . Immediately outside the NSP5 core , we observed a layer composed of NSP2 and NSP4 proteins . The finding that NSP4 is located in the inner part of VPs was unexpected , since it is known that NSP4 is an integral membrane protein of the ER and since it has been reported that functions as a receptor for the new DLPs located at the periphery of the VPs , during their budding towards the lumen of the ER ( Chasey , 1980; Petrie et al . , 1982; Petrie et al . , 1984; Au et al . , 1989 ) . Furthermore , it has been shown that NSP4 associates with VP4 and VP7 to form a hetero-oligomeric complex that could be involved in the last steps of rotavirus morphogenesis ( Maass and Atkinson , 1990 ) . Based on these findings , NSP4 was expected to locate close to VP4 and VP7 , in the surroundings of the VP . On the other hand , and in line with our observations , previous confocal microscopy studies have shown that a portion of NSP4 also shows a limited colocalization with NSP2 ( González et al . , 2000 ) . The dual location of NSP4 as an integral glycoprotein of the ER membrane and as internal to VPs , as our results indicate , is not easy to reconcile; however , in a previous work it was suggested that there are three pools of intracellular NSP4 molecules . The first pool is represented by NSP4 localized in the ER , a second minor pool localized in the ERGIC compartment , and the third pool distributed in cytoplasmic vesicular structures associated with the autosomal marker LC3 ( Berkova et al . , 2006 ) . Furthermore , in that work the authors suggested that NSP4 and autophagic marker LC3-positive vesicles may serve as a lipid membrane scaffold for the formation of large VPs by recruiting early VPs or VP-like structures formed by NSP2 and NSP5 ( Berkova et al . , 2006 ) . This observation is in line with our model that NSP4 lies in an internal protein shell within VPs . An additional , and very interesting possibility to explain the internal location of NSP4 in VPs is the hypothesis that VP morphogenesis occurs on the surface of lipid droplets ( LDs ) ( Cheung et al . , 2010 ) . In that work , it was proposed that LDs serve as a platform to which NSP2 and NSP5 proteins attach to form VP-like structures; NSP2 octamers , in turn , associate with the viral polymerase VP1 and rotavirus +RNAs . The assorted RNA complex containing NSP2 , VP1 , the capping enzyme VP3 and viral +ssRNA is predicted to nucleate VP2 core assembly . In this model , core assembly results in the displacement of +RNA-bound NSP2 octamers , while VP1 within new formed cores direct dsRNA synthesis , using +RNAs as templates ( Cheung et al . , 2010; Borodavka et al . , 2017; Borodavka et al . , 2018 ) . These events are followed by incorporation of the middle virus capsid protein VP6 to form DLPs . At some stage , these assemblies become VPs containing cores and DLPs and may lose some or all of their lipids ( Cheung et al . , 2010 ) . In this regard , it is important to have in mind that the currently accepted model for the LD biogenesis is that neutral lipids are synthesized between the leaflets of the ER membrane , and the mature LD is then thought to bud from the ER membrane to form an independent organelle that is contained within a limiting monolayer of phospholipids and associated LD proteins ( Walther and Farese , 2012 ) . Thus , during budding of the LDs from the ER membrane they could take along rotavirus NSP4 ( topologically oriented towards the cell's cytoplasm ) which could help as a scaffold on the surface of LDs for the assembly of other rotavirus viroplasmic proteins , localizing then to the interior of VPs . Further support for our model of localization of at least one pool of NSP4 molecules inside of the VPs is the observation that knocking-down the expression of NSP4 by RNA interference significantly reduces the number and size of VPs present in the cell , as well as the production of DLPs ( López et al . , 2005a ) . That study also showed that during RNAi inhibition of NSP4 expression the NSP2 and NSP5 proteins maintained an intracellular distribution restricted to VPs , while the VP2 , VP4 , VP6 and VP7 proteins failed to locate to VPs . Based on these observations , it is tempting to suggest that , in addition to the role NSP4 has on the budding of DLPs into the ER lumen , it may also play an important role as a regulator of VP assembly . After the NSP2/NSP4 layer , we observed a middle zone composed of the structural proteins VP1 , VP2 and VP6 . Their location in the same zone is expected given their close association in the assembled DLPs ( Estes and Greenberg , 2013 ) . Also , the fact that VP1 , VP2 and VP6 form a complex with NSP2 that has replicase activity ( Aponte et al . , 1996 ) , suggests that the production of new DLPs could take place in this zone of the VP . Finally , we found that VP4 and VP7 conform independent layers just external to the viroplasmic proteins . The position of these two proteins agrees with the proposed model of rotavirus morphogenesis in which VP4 is assembled first on DLPs , and subsequently VP7 binds the particles and locks VP4 in place ( Trask and Dormitzer , 2006 ) . Furthermore , the fact that VP7-Mon and VP7-Tri occupied the same layer in our model indicates that in the ER sites into which the DLPs bud , VP7 is already organized as trimers , which are subsequently assembled into the virus particles . Of interest , VP4 has been reported to exist in two different forms in infected cells . One of them is associated with microtubules ( Nejmeddine et al . , 2000 ) , while the other one has been reported to be found between the VP and the ER membrane ( González et al . , 2000 ) . In this regard , based on our findings , we suggest that the latter form of VP4 can be actually considered as an integral component of the VP . Since several studies have found the presence of different cellular proteins and lipids in association to VPs ( Maruri-Avidal et al . , 2008; Cheung et al . , 2010; Dhillon et al . , 2018 ) , it will be interesting to study the relative localization of this components using the methodologies described here . The rhesus monkey kidney epithelial cell line MA104 ( ATCC ) was grown in Dulbecco’s Modified Eagle Medium-Reduced Serum ( DMEM-RS ) ( Thermo-Scientific HyClone , Logan , UT ) supplemented with 5% heat-inactivated fetal bovine serum ( FBS ) ( Biowest , Kansas City , MO ) at 37°C in a 5% CO2 atmosphere . The cells were confirmed to be free of mycoplasm by testing with the INTRON Mycoplasma PCR Detection Kit ( #25234 ) . Rhesus rotavirus ( RRV ) was obtained from H . B . Greenberg ( Stanford University , Stanford , Calif . ) and propagated in MA104 cells as described previously ( Pando et al . , 2002 ) . Prior to infection , RRV was activated with trypsin ( 10 μg/ml; Gibco , Life Technologies , Carlsbad , CA ) for 30 min at 37°C . Monoclonal antibodies ( MAbs ) to VP2 ( 3A8 ) , VP4 ( 2G4 ) , VP6 ( 255/60 ) , VP7 ( 60 ) and VP7 ( 159 ) were kindly provided by H . B . Greenberg ( Stanford University , Stanford , CA ) ( Shaw et al . , 1986; Greenberg et al . , 1983 ) . The rabbit polyclonal sera to NSP2 , NSP4 and NSP5 , and the mouse polyclonal serum to NSP2 were produced in our laboratory ( González et al . , 1998 ) . The hyperimmune serum to NSP4 ( C-239 ) was generated in our laboratory by immunizing New Zealand white rabbits with a recombinant protein expressed in E . coli with a histidine-tail , representing the carboxy-terminal end ( amino acids 120 to 175 ) of the rhesus rotavirus RRV NSP4 protein; see also Maruri-Avidal et al . ( 2008 ) , in which this serum was used . The hyperimmune serum to VP1 was also generated in our laboratory by immunizing BALB/c mice with a recombinant protein expressed in E . coli with a histidine-tail , representing amino acids 227 to 539 of the rhesus rotavirus RRV VP1 protein . Goat anti-mouse Alexa-488- and Goat anti-rabbit Alexa-568-conjugated secondary antibodies were purchased from Molecular Probes ( Eugene , Oreg . ) . MA104 cells grown on glass coverslips were infected with rotavirus RRV at a multiplicity of infection ( MOI ) of 1 . Six hours post infection , the cells were fixed with and processed for immunofluorescence as described ( Silva-Ayala et al . , 2013 ) . Finally , the coverslips were mounted onto the center of glass slides with storm solution ( 1 . 5% glucose oxidase +100 mM β-mercaptoethanol ) to induce the blinking of the fluorophores ( Dempsey et al . , 2011; Heilemann et al . , 2009 ) . Cells grown in 75-c⁢m2 flasks were infected with rotavirus RRV at an MOI of 3 as described above . Six hours postinfection the cells were fixed in 2 . 5% glutaraldehyde-0 . 1 M cacodylate ( pH 7 . 2 ) , postfixed with 1% osmium tetroxide , and embedded in Epon 812 resin . The ultrathin sections obtained were stained with 2% uranyl acetate-1% lead citrate ( Reynolds mix ) . The grids were examined with a Zeiss EM-900 electron microscope at 80 kV . All super-resolution imaging measurements were performed on an Olympus IX-81 inverted microscope configured for total internal reflection fluorescence ( TIRF ) excitation ( Olympus , cellTIRFTM Illuminator ) . The critical angle was set up such that the evanescence field had a penetration depth of ~200 nm ( Xcellence software v1 . 2 , Olympus soft imaging solution GMBH ) . The samples were continuously illuminated using excitation sources depending on the fluorophore in use . Alexa Fluor 488 and Alexa Fluor 568 dyes were excited with a 488 nm or 568 nm diode-pumped solid-state laser , respectively . Beam selection and modulation of laser intensities were controlled via Xcellence software v . 1 . 2 . A full multiband laser cube set was used to discriminate the selected light sources ( LF 405/488/561/635 A-OMF , Bright Line; Semrock ) . Fluorescence was collected using an Olympus UApo N 100⁢x/1 . 49 numerical aperture , oil-immersion objective lens , with an extra 1 . 6x intermediate magnification lens . All movies were recorded onto a 128 × 128-pixel region of an electron-multiplying charge coupled device ( EMCCD ) camera ( iXon 897 , Model No: DU-897E-CS0-#BV; Andor ) at 100 nm per pixel , and within a 50 ms interval ( 300 images per fluorescent excitation ) . Sub-diffraction images were derived from the Bayesian analysis of the stochastic Blinking and Bleaching of Alexa Fluor 488 dye ( Cox et al . , 2011 ) . For each super-resolution reconstruction , 300 images were acquired at 20 frames per second with an exposure time of 50 ms at full laser power , spreading the bleaching of the sample over the length of the entire acquisition time . The maximum laser power coming out of the optical fiber measured at the back focal plane of the objective lens , for the 488 nm laser line , was 23 . 1 mW . The image sequences were analyzed with the 3B algorithm considering a pixel size of 100 nm and a full width half maximum of the point spread function of 270 nm ( for Alexa Fluor 488 ) , measured experimentally with 0 . 17 μm fluorescent beads ( PS-SpeckTM Microscope Point Source Kit , Molecular Probes , Inc ) . All other parameters were set up using default values . The 3B analysis was run over 200 iterations , as recommended by the authors in Cox et al . ( 2011 ) , and the final super-resolution reconstruction was created at a pixel size of 10 nm with the ImageJ plugin for 3B analysis ( Rosten et al . , 2013 ) , using parallel computing as described in Hernández et al . ( 2016 ) . The resolution increase observed in our imaging set up by 3B analysis was up to five times below the Abbe’s limit ( ~50 nm ) . The resolution provided by 3B was improved by computing the photo-physical properties of Alexa Fluor 488 , and Alexa Fluor 568 dyes , which were provided to 3B algorithm , as an input parameter which encompass the probability transition matrix between fluorophore’s states . The method was validated with 40 nm gattapaint nanorules ( PAINT 40RG , gattaquant , Inc ) labeled with ATTO 655/ATTO 542 dyes ( see ‘3B Algorithm’ in Appendix 1 ) . The segmentation algorithm ( VPs-DLSFC ) was developed in Matlab R2018a ( 9 . 4 . 0 . 813654 ) software . A detailed explanation of each the developed methods is available in Appendix 1 . Statistical analysis were performed using R version 3 . 4 . 4 ( 2018-03-15 ) software . All the codes are available at https://github . com/Yasel88/Nanoscale_organization_of_rotavirus_replication_machineries ( Garcés Suárez , 2019; copy archived at https://github . com/elifesciences-publications/Nanoscale_organization_of_rotavirus_replication_machineries ) .	Rotaviruses are small viruses that can infect cells in the intestine . They are responsible for most cases of severe infectious diarrhea , the most common cause of death among young children in developing countries . Controlling the spread of rotavirus infections is difficult , even with high levels of hygiene , so effective treatments are essential to curtail the virus’ infections . Understanding how new rotaviral particles are made in infected cells is one of the first steps toward developing new therapies . Once rotaviruses enter the cells , proteins from the virus and the cell aggregate into compact spheres called viroplasms to make new viral particles . Studying these viroplasms used to be difficult because they are too small to see with the resolution of standard microscopes . In recent years , advances in microscopy and mathematical methods have focused on breaking the existing resolution limits , leading to the development of super-resolution microscopy . This new technique has made it possible to study objects with sizes in the order of a billionth of a meter , known as nanoscopic structures , including viroplasms . Garcés et al . use super-resolution microscopy to determine how viral proteins are arranged in the viroplasm and gain a better understanding of how the viruses are assembled . The images revealed that , in infected monkey kidney cells , rotavirus proteins inside the viroplasm form highly organized concentric layers . This arrangement is reliably repeated in viroplasms of different sizes , indicating that the organization of the proteins is likely set up when the viroplasm starts to form . These findings make use of new microscopy , image analysis and statistical tools to study rotaviruses , providing a new framework to understand many aspects of rotaviral biology . Additionally , the result showing that proteins organize consistently in viroplasms is a first step towards understanding how the machinery that makes new rotaviruses works , which could lead to future treatments for severe infectious diarrhea .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "microbiology", "and", "infectious", "disease" ]	2019	Nanoscale organization of rotavirus replication machineries
Giant lipid vesicles are closed compartments consisting of semi-permeable shells , which isolate femto- to pico-liter quantities of aqueous core from the bulk . Although water permeates readily across vesicular walls , passive permeation of solutes is hindered . In this study , we show that , when subject to a hypotonic bath , giant vesicles consisting of phase separating lipid mixtures undergo osmotic relaxation exhibiting damped oscillations in phase behavior , which is synchronized with swell–burst lytic cycles: in the swelled state , osmotic pressure and elevated membrane tension due to the influx of water promote domain formation . During bursting , solute leakage through transient pores relaxes the pressure and tension , replacing the domain texture by a uniform one . This isothermal phase transition—resulting from a well-coordinated sequence of mechanochemical events—suggests a complex emergent behavior allowing synthetic vesicles produced from simple components , namely , water , osmolytes , and lipids to sense and regulate their micro-environment . Giant unilamellar vesicles ( GUVs ) are the simplest cell-like closed compartments consisting of semi-permeable flexible shells ( 4–6 nm thick , 5–50 µm diameter ) , isolating femto- to pico-liter quantities of aqueous core from the surrounding bulk ( Walde et al . , 2010 ) . Although water permeates readily across the vesicular walls ( 10−2–10−3 cm/s ) ( Fettiplace and Haydon , 1980 ) , passive permeation of solutes is significantly lower across the intact membrane ( Deamer and Bramhall , 1986 ) . As a result , osmotic differentials are readily established between the compartmentalized volume and the surrounding free bath . This in turn triggers a relaxation process , which acts to reduce the osmotic pressure difference across the closed semi-permeable membrane by influx ( or efflux ) of water depending on the sign of the pressure differential . Thus , for osmolyte-loaded vesicles in a hypotonic environment , water permeates and vesicle swells , until the internal Laplace pressure compensates the osmotic pressure , increasing its volume to surface area ratio ( Ertel et al . , 1993 ) . In this same vein , efflux of compartmentalized water from vesicles embedded in hypertonic media , conversely , decreases the volume to surface area ratio ( Boroske et al . , 1981 ) . Furthermore , because of their large area expansion moduli ( 102–103 mN m−1 ) and low bending rigidities ( 10−19 Nm ) , vesicular shells bend readily but tolerate only a limited area of expansion ( ∼5% ) ( Needham and Nunn , 1990; Hallett et al . , 1993; Seifert , 1997 ) . Consequently , GUVs experiencing solute concentration difference across their vesicular boundary adjust their volume , deforming in hypertonic media and swelling in hypotonic ones ( Boroske et al . , 1981; Ertel et al . , 1993 ) . A consequence of the osmotic influx of water in vesicles embedded in hypotonic media is the build-up of lateral membrane tension due to changes in the balance of forces within the bilayer producing high energy states ( compared to isotonic relaxed vesicles ) ( Needham and Nunn , 1990 ) . Beyond a threshold tension , rupture and pore formation become energetically favorable , lysing the GUVs at lateral tensions corresponding to ∼30–40 mNm−1 ( Needham and Nunn , 1990; Ertel et al . , 1993; Mui et al . , 1993 ) . The lytic process , however , is not catastrophic; rather it follows a step-wise sequence of events ( Ertel et al . , 1993; Peterlin et al . , 2012 ) . During each membrane rupture event , only a fraction of the intravesicular solute ( and water ) is released before the bilayer reseals leaving the vesicle hyperosmotic with a lower osmotic differential . This then prompts subsequent events of water influx , vesicle swelling , and rupture until sufficient intravesicular solute has been lost , and the membrane is able to withstand the residual sub-lytic osmotic pressure without collapsing ( Wood , 1999 ) . Thus , GUVs in hypotonic media exhibit oscillations in their sizes—characterized by alternating modulations of vesicular volume , tension , and solute efflux—prompted by repeated cycles of swelling and bursting ( Sandre et al . , 1999; Karatekin et al . , 2003b; Popescu and Popescu , 2008; Peterlin et al . , 2012 ) . In the work reported here , we show that the swell–burst cycles in hypertonic vesicles consisting of domain-forming lipid mixtures ( Baumgart et al . , 2003; Veatch and Keller , 2005 ) become coupled with the membrane's compositional degrees of freedom , producing a long-lived transient response characterized by damped oscillations in phase behavior at the membrane surface , cycling between the state characterized by large microscopic domains at the membrane surface and an optically uniform one . This oscillatory phase separation occurs isothermally , and it is driven by a sequence of elementary biophysical processes involving cyclical changes in osmotic pressure , membrane tension , and poration , which attend swell–burst cycles ( Koslov and Markin , 1984; Mui et al . , 1993; Popescu and Popescu , 2008 ) : in the swelled state , osmotic pressure and elevated membrane tension due to the influx of water promote the appearance of microscopic domains ( Akimov et al . , 2007; Ayuyan and Cohen , 2008; Hamada et al . , 2011 ) . During the burst phase , solute leakage through short-lived membrane poration ( Sandre et al . , 1999; Brochard-Wyart et al . , 2000; Karatekin et al . , 2003b ) relaxes the osmotic pressure and membrane tension , breaking up the domains producing an optically uniform membrane . This cyclical pattern does not persist indefinitely: a step-wise diminution of the osmotic pressure differential , because of the solute leakage during burst events , gradually dampens the oscillations ultimately equilibrating the GUV to the residual osmotic differential . The GUVs ( Morales-Penningston et al . , 2010 ) we investigated consist of ternary lipid mixtures composed of cholesterol ( Ch ) , sphingomyelin ( SM ) , and the unsaturated phospholipid , POPC ( 1-palmitoyl-2-oleoyl-sn-1-glycero-3-phosphocholine ) at room temperature ( 25°C ) ( ‘Materials and methods’ ) . Depending on precise composition and temperature , these mixtures are known to form a uniform single phase or exhibit microscopic phase separation ( Veatch and Keller , 2005 ) , including one characterized by two co-existing liquid phases: a dense phase enriched in SM and Ch designated as the Lo ( liquid-ordered ) phase and a second , less dense Ld ( liquid-disordered ) phase consisting primarily of POPC . To discriminate between the Lo and Ld phases by fluorescence microscopy , we doped our GUVs with a small concentration ( 0 . 5 mol% ) of a phase sensitive probe , N-lissamine rhodamine palmitoylphosphatidyl-ethanolamine ( Rho-DPPE ) ( Baumgart et al . , 2007 ) . For an equimolar lipid ( 1:1:1 ) composition , the phase diagram predicts the absence of large microscopic domain formation at optical length scales ( Veatch and Keller , 2005 ) . Consistent with this prediction , our GUVs encapsulating 200 mM sucrose appear optically homogeneous at room temperature in an osmotically balanced , isotonic medium also containing 200 mM sucrose ( Figure 1A , C ) . Moreover , they exhibit a flaccid , undulating surface topography ( Video 1 ) confirming bending-dominated shape fluctuations ( Seifert , 1997 ) . 10 . 7554/eLife . 03695 . 003Figure 1 . Subjecting giant unilamellar lipid vesicles to an osmotic differential . ( A–B ) , Schematic of a GUV immersed in an osmotically balanced isotonic bath ( A ) . Dilution of the extra-vesicular dispersion medium by water subjects the GUV to a hypotonic bath producing an osmotic differential ( B ) , which renders initially flaccid vesicles stiff and replaces the initially optically uniform membrane surface by one characterized by a domain pattern of co-existing Ld and Lo phases at microscopic length scales . Solute is rendered as white particles , membrane , pink , and domain pattern in pink and purple . ( C–D ) The process in ( A–B ) exemplified by wide-field fluorescence ( C ) and deconvolved ( D ) images of a solution of GUVs consisting of POPC:SM:Ch ( 1:1:1 ) labeled with 0 . 5 mol% Rho-DPPE at 25°C containing 200 mM sucrose concentration , osmotically balanced by 200 mM glucose in ( C ) , and under an osmotic differential of ∼200 mM in ( D ) at 25°C . Scale bars: 15 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 00310 . 7554/eLife . 03695 . 004Figure 1—figure supplement 1 . Undulating boundary of isotonic vesicles . Selected time-lapse fluorescence images from Video 1 revealing out-of-plane membrane fluctuations typical for non-tense GUVs in the absence of osmotic gradient . Panels correspond to ( A ) 0 s , ( B ) 3 s , ( C ) 5 s , ( D ) 11 s , ( E ) 16 s , ( F ) 27 s . The GUV is composed of POPC:SM:Ch ( 1:1:1 ) , labeled with 0 . 5% Rho-DPPE ( pseudo-colored magenta ) , and imaged at 25°C . Scale bar: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 00410 . 7554/eLife . 03695 . 005Video 1 . Thermally excited undulations of isotonic GUVs . A video assembled from time-lapse fluorescence images revealing out-of-plane membrane fluctuations typical for non-tense GUVs in the absence of an osmotic gradient ( 50 vol% glycerol inside and outside ) . The osmotically balanced GUV consists of POPC:SM:Ch ( 1:1:1 ) labeled with 0 . 5 mol% Rho-DPPE ( pseudo-colored magenta ) and is imaged at 25°C . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 005 Diluting the extra-vesicular dispersion medium with deionized water produces a hypotonic bath depleted in osmolytes , subjecting the GUVs to a trans-bilayer osmotic differential . A representative fluorescence image obtained shortly after imposing the concentration difference ( ∼200 mM , t < 60 s , reveals that the flaccid topography [Figure 1A , C , Figure 1—figure supplement 1] and uniform dye distribution of isotonic GUVs are abandoned , replaced by a swollen , spherical boundary and a heterogeneous fluorescence pattern characterized by microscopic , probe-enriched domains consistent with earlier reports ) ( Figure 1B , D ) ( Baumgart et al . , 2003; Hamada et al . , 2011; Oglecka et al . , 2012 ) . Because Rho-DPPE partitions preferentially into the Ld phase , the appearance of bright domains indicates microscopic Ld phase fluid domains in the Lo phase surroundings . A time-lapse video of a vesicular population subject to hypotonic conditions ( Figure 2A , Video 2 ) reveals that the phase separation is not static: optically homogeneous vesicles observed at a given instance break up into surface patterns consisting of large microscopic domains and conversely , those textured by domains adopt an optically homogeneous state over time , with each vesicle undergoing complete single cycles in roughly tens of seconds . Moreover , at any given instance , some vesicles appear homogeneous whereas others are microscopically phase-separated ( Figure 2B and Video 3 ) producing a heterogeneous landscape . The time-dependent process of the appearance and disappearance of large microscopic domains repeats itself multiple times ( n > 10 ) over several tens of minutes—ultimately producing a steady-state characterized by a fixed microstructure and a rounded boundary . The oscillatory phase separation behavior is fully reproducible for a variety of ( 1 ) neutral osmolytes ( e . g . , glycerol , glucose , lactose , galactose , dextran , sorbitol , and sucrose ) ; ( 2 ) GUV sizes ( ∼5–50 µm ) ; ( 3 ) initially imposed strengths of osmotic gradients ( 20–2000 mM ) ; and ( 4 ) lipid compositions within the phase co-existence window ( Veatch and Keller , 2005 ) . 10 . 7554/eLife . 03695 . 006Figure 2 . Oscillatory phase separation in hypertonic giant unilamellar vesicles subject to an osmotic imbalance . ( A ) , Selected frames from a video of time-lapse fluorescence images ( Video 2 ) illustrating stages of domain dynamics during two consecutive cycles of oscillatory phase separation ( t = 0 s , 9 s , 12 s , 15 s , 25 s , 27 s , 29 s , 188 s , 191 s , 193 s , 246 s , and 247 s ) . The GUVs imaged consist of POPC:SM:Ch ( 1:1:1 ) labeled with 0 . 5% Rho-DPPE , encapsulating 1 M sucrose , diluted in deionized water , at room temperature . Scale bar: 10 μm . ( B ) Selected images from time-lapse fluorescence images ( Video 3 ) showing asynchronous cycling in a population of GUVs ( t = 0 s , 98 s , 148 s , 294 s , and 448 s ) . The images are projections of Z-stacks of the lower hemispheres of GUVs consisting of POPC:SM:Ch ( 1:1:1 ) labeled with 0 . 5 mol% Rho-DPPE , encapsulating 200 mM sucrose , diluted in deionized water at 25°C ( n = 5 ) . Scale bar: 15 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 00610 . 7554/eLife . 03695 . 007Video 2 . Oscillatory domain dynamics in GUVs immersed in hypotonic bath . Time-lapse images of a bottom view of GUVs consisting of POPC:SM:Ch ( 1:1:1 ) labeled with 0 . 5% Rho-DPPE ( pseudo-colored magenta ) under a net osmotic differential . The GUVs encapsulate 1 M sucrose in their interior , and the external dispersion medium is MilliQ water . A striking temporal pattern of oscillatory phase separation revealing appearance , coalescence , and dispersion of optically resolved domains is evident ( see manuscript for details ) . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 00710 . 7554/eLife . 03695 . 008Video 3 . Domain dynamics of GUVs in hypotonic bath . Video assembled from time-lapse images of Z-stack projections of the bottom hemispheres of GUVs , consisting of POPC:SM:Ch ( 1:1:1 ) labeled with 0 . 5 mol% Rho-DPPE ( white ) . The vesicles encapsulated 200 mM sucrose and were diluted in MilliQ water at 25°C . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 008 It is known that continuously illuminating single vesicles deliberately using intense light can oxidize lipids ( Ayuyan and Cohen , 2006 ) or generate membrane tension by folding the excess membrane area within the localized regions of the enhanced electric field of the light thus suppressing undulations ( Barziv et al . , 1995; Sandre et al . , 1999 ) . To confirm that the unusual domain dynamics we witness do not result from these effects of optical illumination , we carried out additional experiments . In companion experiments where only occasional low-dose illumination ( as opposed to the rapid sequence of illumination used to capture detailed dynamics ) at arbitrary time intervals is used ( Figure 3 ) , we find that the oscillatory domain behavior is fully reproduced . Moreover , by preparing GUVs using gentle hydration ( Morales-Penningston et al . , 2010 ) , we further confirm that the observed behavior is not adversely affected by the electroformation method ( Video 4 , Figure 3—figure supplement 1 ) . 10 . 7554/eLife . 03695 . 009Figure 3 . Interrupted imaging of oscillatory phase separation . Z-stack projections of height-resolved fluorescence images of the lower hemisphere of a GUV consisting of POPC:SM:Ch ( 1:1:1 ) labeled with 0 . 5% Rho-DPPE ( pseudo-colored red ) . The GUV encapsulates 200 mM sucrose , and the external dispersion medium is diluted in MilliQ water . Images are acquired at 25°C at arbitrary time points; ( A ) 0 s , ( B ) 99 s , ( C ) 148 s , ( D ) 299 s , ( E ) 550 s , and ( F ) 692 s . The first image was taken ∼2 hr after imposing the osmotic gradient . Scale bar: 5 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 00910 . 7554/eLife . 03695 . 010Figure 3—figure supplement 1 . Oscillatory phase separation in complex GUVs prepared by hydration . Arbitrarily selected micrographs from a time-lapse sequence of fluorescence images from Video 4 for GUVs prepared by ‘gentle hydration’ . GUVs consist of POPC:SM:Ch ( 1:1:1 ) labeled with 0 . 5% Rho-DPPE ( pseudo-colored blue ) , encapsulate 200 mM sucrose , and are subject to hypotonic conditions by dilution in MilliQ water at 25°C . Panels correspond to ( A ) 0 s , ( B ) 3 s , ( C ) 6 s , ( D ) 11 s , ( E ) 14 s , ( F ) 22 s , starting about 2 min after establishing the osmotic differential . Daughter vesicles identified as 1 and 2 illustrate clear oscillatory phase separation behavior over the experimental time scale captured by the images displayed . Scale bar: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 01010 . 7554/eLife . 03695 . 011Video 4 . GUVs prepared by gentle hydration reproduce the oscillatory domain dynamics . GUVs prepared by ‘gentle hydration’ consisting of POPC:SM:Ch ( 1:1:1 ) labeled with 0 . 5 mol% Rho-DPPE ( pseudo-colored blue ) , encapsulating 200 mM sucrose . Hypotonic conditions are established by dilution in deionized water at 25°C . Vesicles exhibit osmotic swelling and oscillatory domain dynamics comparable to that seen for electroformed GUVs under the same conditions . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 011 This oscillatory pattern of phase separation appears to be a cyclical isothermal phase transition resulting from oscillations in osmotic pressure and membrane tension , which characterize osmotic relaxation in vesicular compartments subject to osmotic differentials . A synergistic interplay of well-understood fundamental biophysical mechanisms—including selective membrane permeability for water ( Deamer and Bramhall , 1986; Peterlin and Arrigler , 2008 ) , osmotically-generated membrane tension , tension- and pressure-dependent membrane phase behavior ( Hamada et al . , 2011; Portet et al . , 2012; Uline et al . , 2012; Givli et al . , 2012 ) , and poration ( Sandre et al . , 1999; Brochard-Wyart et al . , 2000; Karatekin et al . , 2003b ) —couple osmotic activity of water with spatial organization of membrane molecules ( i . e . , appearance of large , microscopic domains ) , such as those analyzed below . The existence of an osmolyte concentration difference across the vesicular boundary triggers an osmotic relaxation process , which acts to reduce the pressure difference across the semi-permeable membrane by an influx of water ( Mui et al . , 1993 ) . As water enters , the GUV swells ironing out the thermal undulations and rendering the vesicular boundary tense ( Figure 1A , C ) ( Haleva and Diamant , 2008 ) . At equilibrium , the lateral tension generated in the membrane compensates for the osmotic pressure , consistent with the law of Laplace , σ ( = ΔP r/2 , where ΔP is the osmotic pressure difference and r , the vesicle radius ) . A closer examination of the results above reveals that ( 1 ) the domains coarsen through collision and coalescence ( Figure 4A and Video 5 ) and ( 2 ) the appearance of phase-separated state invariably coincides with the swollen , tense state of the GUV during the cyclical swell–burst processes ( Figure 4B , C ) . Although both lateral tension and pressure difference influence membrane phase behavior in our osmotically driven case , it is instructive to consider how each of the two factors individually affects membrane phase behavior . A recent thermodynamic analysis and experiments examining the effects of mechanically generated tension reveal a lowering of miscibility phase transition temperature between the Lo and Ld phases with increase in tension ( dT/dσ , ∼−1 K [mNm−1] −1 ) ( Portet et al . , 2012; Uline et al . , 2012 ) . However , how this shift in transition temperature affects membrane phase behavior and domain morphology is not obvious: a recent experimental study suggests that even tension alone can stabilize complex domain morphologies ( Chen and Santore , 2014 ) . The current and earlier observations in which osmotic differentials induce phase separation ( Hamada et al . , 2011 ) are clearly at variance with these predictions . An alternate explanation involves separate theoretical arguments , which require pre-existing phase separated domains in the optically homogeneous state . It suggests that the lateral tension elevates line tension between co-existing phases ( Akimov et al . , 2007 ) . Therefore , although membrane tension disfavors nucleation of a new phase ( by raising the energy barrier that must be met for the formation of critical nuclei ) , it can promote coalescence of small pre-existing nanoscale domains driven by minimization of line tension between Lo and Ld phase . Additional experiments using ternary lipid mixtures ( DOPC , DPPC , and Chol ) , which have been thought not to produce nanodomains at temperatures above 20°C ( Hamada et al . , 2011 ) , also produces oscillatory phase behavior ( Video 6 ) . This then suggests that the osmotically generated tension alone might be insufficient to explain the observed osmotically induced isothermal phase transition , and that the non-ideality in mixing is likely a consequence of a combined effect of the pressure and tension . Indeed , a theoretical model by Givli and Bhattacharya ( Givli et al . , 2012 ) , explicitly introducing osmotic pressure contributions within the generalized Helfrich energy treatment , suggests that pressure can perturb isothermal phase diagram , driving domain formation primarily by affecting the interaction between geometry and composition . 10 . 7554/eLife . 03695 . 012Figure 4 . Mechanisms responsible for oscillatory phase separation in GUVs subject to osmotic differentials . ( A ) Domain coarsening . Selected frames from Video 4 illustrating collision and coalescence of domains during a swell segment of the GUV oscillations ( Ld phase , pseudo-colored magenta ) . Images are 1 s apart focused on a region of interest located at the bottom of a GUV . Scale bar: 5 μm . ( B–C ) Relationship between vesicle swelling and phase-separation . Fluorescence images revealing ( B ) that largest domains are observed under conditions of maximal swelling ( t = 0 s , 8 s , and 106 s ) . Scale bar: 10 μm . ( C ) Control experiment using single component POPC GUVs , labeled with 0 . 5% Rho-DPPE , encapsulating 200 mM sucrose , diluted in deionized water at 25°C , confirm that the GUV swelling does not require domain formation and/or reorganization . Scale bar: 10 μm . ( D–F ) Increase of cycle period during oscillatory domain dynamics . A bar chart showing successively increasing periods of domain growth/dispersion cycles in GUVs ( D ) 42 . 0 μm , ( E ) 26 . 3 μm , and ( F ) 10 . 7 μm in diameter . A cycle period is defined as the time elapsed between two consecutive instances of appearance of uniform fluorescence . Except for control in ( C ) , all data were collected using POPC:SM:Ch ( 1:1:1 ) GUVs , labeled with 0 . 5% Rho-DPPE , encapsulating 200 mM sucrose , diluted in deionized water at 25°C . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 01210 . 7554/eLife . 03695 . 013Video 5 . Evidence for domain merger by collision and coalescence . Time-lapse wide-field fluorescence images of the lower hemisphere of POPC:SM:Ch ( 1:1:1 ) GUVs labeled with 0 . 5 mol% Rho-DPPE ( pseudo-colored yellow ) , encapsulating 200 mM sucrose , diluted in MilliQ water at 23°C . Domain–domain coalescence , followed by line-tension driven shape transformations , drives domain growth . Frames are collected at 1 s intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 01310 . 7554/eLife . 03695 . 014Video 6 . Oscillatory domain dynamics in mixed ternary system known to exist in single liquid state in the absence of net osmotic differential . Time-lapse wide-field fluorescence images of DOPC:DPPC:Ch ( 5:2:3 ) GUVs labeled with 0 . 5 mol% Rho-DPPE ( pseudo-colored yellow ) , encapsulating 200 mM sucrose , diluted in MilliQ water at 23°C . Frames are collected at 1 s intervals . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 014 This tension and pressure-mediated appearance of the phase-separated state in osmotically swollen membranes , however , does not account for the oscillations in domain pattern: ( 1 ) why does the osmotically swollen vesicle characterized by large microscopic domains return to a homogeneous state , and ( 2 ) what prompts subsequent cycles of phase separation ? A closer look at the temporal dynamics reveals that the process does not persist indefinitely . The period of oscillation between optically uniform and phase-separated states increases with the passage of time ( Figure 4D–F ) . The cycle period—defined as the time elapsed between two consecutive instances of homogeneous fluorescence—increases three to 10-fold , before reaching a non-oscillating quiescent state , 60–120 min after the imposition of the osmotic differential . This ‘fatigue’ in the oscillatory phase separation process suggests that the driving force ( i . e . , the osmotic differential and accompanying tension ) must weaken with each cycle , which requires a separate mechanism for solute efflux . It is known that membrane lysis proceeds via cascades of pores during each cycle of the swell–burst sequence ( Karatekin et al . , 2003b ) . This strikingly regular , temporal cascade of pores is fully reproduced in our case ( Figure 5 , Figure 5—figure supplement 1 ) : during the swell segment of each oscillation cycle , a single microscopic pore , several micrometers across , becomes visible under conditions of maximum swelling and largest domain size , typically for a period not exceeding 1 . 0 s ( Video 7 ) . According to classical nucleation theory , the cost ( E ) of creating a pore in a tense membrane is determined by the competition between membrane tensional energy ( −πr2σ ) and the line tension energy ( +2πrγ ) at the edge of the pore . Thus , under conditions of sufficient membrane tension ( dE/dr > 0 ) , pores nucleate and grow , enabling solute efflux ( Sandre et al . , 1999; Peterlin and Arrigler , 2008 ) . Although domain formation is not required for pore-formation , the probability of pore-nucleation might be enhanced by surface defects , such as are present at the boundary between co-existing phases , since the energy required to open a pore ( >40 KBT ) is considerably higher than the thermal activation energy ( Karatekin et al . , 2003b ) . The long life spans ( ∼1 s ) of the pores are likely supported by two opposing processes , namely osmotic influx of water and the leakage rate of solute through the pore ( Koslov and Markin , 1984 ) . Subsequent healing of the pore is promoted by the reduction in the net membrane area and partial loss of the encapsulated solutes , both of which reduce membrane tension , σ ( Karatekin et al . , 2003b ) . Thus , during each membrane rupture event , only a fraction of the intravesicular solute is released before the bilayer reseals , leaving the vesicle hyperosmotic , albeit with a reduced osmotic differential . This then prompts subsequent cycles of water influx , vesicle swelling , and rupture until sufficient intravesicular solute is lost and the Laplace tension in the membrane is able to compensate for the residual osmotic pressure ( Ertel et al . , 1993 ) . 10 . 7554/eLife . 03695 . 015Figure 5 . Evidence for the formation of microscopic pores during each individual cycle of oscillatory phase separation . ( A–L ) Wide-field fluorescence images of microscopic pore formation ( ∼5–15 µm in diameter; indicated by arrows ) observed in three consecutive swell–burst cycles of a phase-separating GUV . A single pore appears during each phase separation cycle and reseals within 1 s . The POPC:SM:Ch ( 1:1:1 ) GUV labeled with 0 . 5% Rho-DPPE ( pseudo-colored yellow ) , encapsulates 200 mM sucrose , and is immersed in deionized water at 25°C . Images collected 20 min after imposition of the osmotic differential . Height-resolved ( increment , 0 . 5 µm ) images shown at arbitrary time intervals following the first frame . ( A–L ) 0 s , 0 . 3 s , 0 . 6 s , 0 . 9 s , 9 . 3 s , 9 . 6 s , 9 . 9 s , 10 . 1 s , 15 . 3 s , 15 . 6 s , 15 . 9 s , and 16 . 1 s . Scale bar: 15 µm . Cascades of pores have been observed more than five times . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 01510 . 7554/eLife . 03695 . 016Figure 5—figure supplement 1 . Evidence for pore-formation . Selected frames from Video 7 showing the equatorial view of a POPC:SM:Ch ( 1:1:1 ) GUV labeled with 0 . 5% Rho-DPPE ( pseudo-colored green ) , encapsulating 50 vol % glycerol , diluted in MilliQ water at 25°C . ( A ) The GUV is in a tense state exhibiting distinct domains . ( B ) Pore formation ( indicated by the arrow ) coincides with the onset of domain dispersion . ( C ) Further domain dispersion . ( D ) The GUV is returned to an optically homogenous state . Panels are taken with 0 . 5 s intervals . Scale bar: 5 μm . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 01610 . 7554/eLife . 03695 . 017Video 7 . Evidence for pore formation . Equatorial view of a POPC:SM:Ch ( 1:1:1 ) labeled with 0 . 5 mol% Rho-DPPE ( pseudo-colored green ) , encapsulating 50vol % glycerol , diluted in MilliQ water at 25°C . Pore formation can be clearly seen at 1 . 1 s , just prior to the disappearance of domains and size shrinkage of the GUV . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 017 The oscillatory phase separation above does not require isolated vesicles but becomes integrated with other known shape transformations in GUVs experiencing tension ( Seifert , 1997 ) . Using structurally complex GUVs , which hierarchically embed smaller ones with different osmolyte concentrations , we found that the domain dynamics becomes coordinated with the previously well-known process of expulsion of internal ‘organelle’ vesicles ( Video 8 and Figure 6 ) ( Moroz et al . , 1997; Oglecka et al . , 2012 ) . The observations above further show how local inhomogeneity in the distribution of solute , namely sucrose in the present case , in nested or hierarchical vesicular compartments in single solutions can produce localized oscillatory phase behavior in component vesicles . These observations also suggest that the oscillatory phase separation can be regarded as a type of amplified mechanosensor for solute concentration differences and osmotic differentials . 10 . 7554/eLife . 03695 . 018Video 8 . Oscillatory phase separation during expulsion of daughter GUVs . A video assembled from time-lapse fluorescence images of phase-separating GUVs containing internal ‘organelle’ vesicles . GUVs consisting of POPC:SM:Ch ( 2:2:1 ) labeled with 0 . 5 mol% Sphingomyelin-Atto647N ( SM-647N ) ( pseudo-colored green ) , encapsulate 1 M sucrose , and diluted in MilliQ water at 25°C . The video reveals shifting patterns of osmotic pressure and tension during expulsion of the internal vesicles after an osmotic differential had been established . Key steps include ( A ) a homogeneous , flaccid mother vesicle encapsulating tense daughter vesicles , at a time point prior to vesicle expulsion; ( B ) just after expulsion , the daughter GUV remains tense exhibiting oscillatory phase separation , while the mother GUV is left deflated and homogenous due to the sudden loss of volume; ( C ) The mother GUV subsequently becomes inflated by influx of water; and ( D ) the mother GUV begins to exhibit oscillatory phase separation . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 01810 . 7554/eLife . 03695 . 019Figure 6 . Osmotic gradients sensed by the membrane and visualized by oscillatory phase separation in nested vesosomes . Selected frames from Video 7 showing hierarchical membrane structures of POPC:SM:Ch ( 2:2:1 ) GUVs labeled with 0 . 5% SM-Atto647N ( pseudo-colored green ) , encapsulating 1 M sucrose , submerged in MilliQ water at 25°C . In panel ( A ) , we define the entrapping mother vesicle as M and daughter vesicle of interest as D . Both M and D initially exhibit homogenous fluorescence from their membranes , but store different amounts of tension ( M is flaccid , while D appears tense ) . ( B ) The homogeneous fluorescence from D is replaced by the appearance of optically resolved domains . In the meantime , M becomes more spherical . ( C ) The domains of D have increased in size , and M has now reached an almost spherical shape . ( D ) Expulsion of the tense D vesicle . This image acquired during a transient pore formation suggests that the intravesicular pressure and/or crowding is reduced via preferential expulsion of daughter . This event , we surmise , also delays the onset of domain formation by reducing the swelling of the M vesicle . ( E ) M is returned to a flaccid state , remaining homogenously fluorescent , consistent with the reduction in swelling and a reduction of osmotic pressure . At the same time , D experiencing a new hypotonic medium gets engaged in swell–burst cycles . ( F ) Further inflation of GUVs leads to M adopting a tense spherical configuration , while yet retaining homogenously fluorescent state , while D's domain sizes continue to grow . ( G ) The continued swelling of M finally leads to phase separation . ( H ) Domains in M disappear producing homogeneous state , consistent with the oscillatory phase separation under osmotically generated tension . Panels correspond to ( A–H ) 0 s , 6 s , 14 s , 18 s , 20 s , 62 s , 103 s , and 118 s . Scale bar: 10 µm . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 019 The emergence of oscillatory domain dynamics appears to be a well-coordinated membrane response to osmotic stress through an isothermal phase transition resulting from a highly coordinated interplay between elementary physical mechanisms ( Figure 7 ) . Specifically , the processes of ( 1 ) osmotically triggered water influx; ( 2 ) retention of osmotic pressure and build-up of membrane tension ironing out thermal undulations; ( 3 ) appearance of microscopic domains in the membrane subject to osmotic pressure and lateral tension; ( 4 ) coarsening of domains; ( 5 ) appearance of a short-lived transient pore , which enable partial solute efflux reducing osmotic pressure and membrane tension; and ( 6 ) consequent pore-closure resulting in closed GUVs with reduced osmotic differential—repeat until the sub-lytic osmotic pressure is reached . Although individual biophysical processes leading to the oscillatory domain dynamics during this osmotic relaxation process are well-appreciated , the observations reported here bring to focus several features of vesicle behavior , which are best poorly appreciated . First , the seemingly autonomous vesicle response—in which an external osmotic perturbation is managed by a coordinated and cyclical sequence of physical mechanisms allowing vesicles to sense ( by domain formation ) and regulate ( by solute efflux ) their local environment–suggests a primitive form of a quasi-homeostatic regulation in a synthetic material system ( He et al . , 2012 ) , that is , a simple microemulsion produced from simple components , namely , lipids , water , and osmolytes . Second , these observations illustrate how out-of-plane osmotic activity of water becomes coupled with membrane's in-plane compositional degrees of freedom producing an exquisite and complex response . It underscores the intrinsic coupling between membrane phase and mechanical tension . Third , by highlighting the complexity of lipid vesicles , these results offer an important caveat in implementing giant vesicles as experimental models in scenarios where osmotic imbalances can dominate vesicle response . 10 . 7554/eLife . 03695 . 020Figure 7 . Schematic representations of physical mechanisms and changes in membrane properties during vesicular osmoregulation . ( Left panel ) ( A ) GUV in isotonic medium exhibiting a flaccid morphology . ( B–C ) Immersion in a hypotonic bath initiates an osmotically triggered influx of water rendering the GUV tense . ( D–F ) The optically uniform vesicular surface breaks up into a pattern of microscopic domains , which grow by collision and coalescence . ( G ) Transient appearance of a microscopic pore ( ∼0 . 3–0 . 5 s lifetime ) , enabling solute efflux and tension relaxation , which drives pore closure , producing closed GUVs with a reduced osmotic differential and homogenous surface . Steps ( B–G ) repeat until the sub-lytic solute concentration differential is reached and the Laplace tension in the membrane is able to compensate for the residual osmotic pressure . ( Right panel ) Temporal cascades of osmotic pressure ( H ) and oscillations in membrane tension ( I ) during osmotic relaxation of giant vesicles subject to hypotonic bath . Note that the relative rates implied in the schematic are only best-guess estimates . DOI: http://dx . doi . org/10 . 7554/eLife . 03695 . 020 The findings reported here—illustrating a complex dynamic relationship between the membrane's compositional degrees of freedom with external osmotic imbalance—might be biologically relevant . While large microscopic domains of co-existing liquid phases are thought to not exist in many living cells , a recent study reveals such domain texture in yeast vacuole membranes ( Toulmay and Prinz , 2013 ) . In this study , under conditions of nutrient deprivation , pH changes , and changes in the growth medium have been shown to segregate into microscopic domains in what appears to be a sterol-dependent manner , reminiscent of synthetic giant vesicles . Since one of the functions of yeast vacuoles is to regulate osmotic pressure , it seems tempting to consider whether vacuolar domains also undergo large-scale domain reorganization under osmotic stimuli and contribute to the physiological function . The formation ( and dissolution ) of compositionally differentiated membrane domains ( e . g . , lipid rafts ) is often associated with alterations in the conformations ( and activity ) of many signaling proteins ( Simons and Toomre , 2000 ) , which partition within them . It appears plausible that the domain reorganization stimulated by the osmotic activity of water might provide the cell a generic mechanism to respond to the physical perturbation , such as by activating mechanosensitive ion-channels and serving as sensors for signaling and stress transmission ( DuFort et al . , 2011; Stamenovic and Wang , 2011; Wood , 2011 ) . Fourth , although fatty acid based prebiotic amphiphiles exhibit different mechanical properties ( e . g . , elastic properties and permeability characteristics ) compared to phospholipids , it appears likely that amphiphilic osmoregulation , such as we witness , might have given a thermodynamic advantage to early protein-free protocells ( Hanczyc et al . , 2003; Chen et al . , 2004; Oglecka et al . , 2012 ) to survive ( and even utilize ) drastic environmental osmotic shifts . Elementary physical mechanisms of membrane permeability ( Fettiplace and Haydon , 1980; Deamer and Bramhall , 1986; Rawicz et al . , 2008 ) , osmotic swelling ( Taupin et al . , 1975; Mui et al . , 1993; Haleva and Diamant , 2008; Peterlin and Arrigler , 2008; Peterlin et al . , 2012 ) , tension dependence of lateral phase separation ( Hamada et al . , 2011; Portet et al . , 2012; Uline et al . , 2012 ) , pressure-dependent membrane phase separation ( Givli et al . , 2012 ) , and membrane poration ( Needham and Hochmuth , 1989; Zhelev and Needham , 1993; Sandre et al . , 1999; Brochard-Wyart et al . , 2000; Karatekin et al . , 2003a; Karatekin et al . , 2003b; Levin and Idiart , 2004; Farago and Santangelo , 2005; Riske and Dimova , 2005; Evans and Smith , 2011 ) have all been extensively studied and well-documented in the existing literature . Below , we recapitulate the key aspects of these mechanisms ( and obtain rough estimates for key observables ) , which constitute key parts of the emergent behavior reported herein . Consider a case in which dilution of the external dispersion medium results in the creation of an initial osmotic differential of 200 mM ( sucrose ) . Applying van't Hoff's equation ( ΔPosm = RTΔc , where R is the gas constant and T the absolute temperature ) , this concentration gradient corresponds to an excess intravesicular osmotic pressure of 0 . 5 MPa . As a result of the pressure difference , an osmotic relaxation process sets in . Because of large differences in permeability of water ( ∼10−3–10−4 cm s−1 ) and sucrose ( ∼10−8 cm s−1 ) , however , the relaxation process is determined by the significant differences in time scales for permeation of water and the solute: rapid water permeation governs the response of the solute-encapsulating hypertonic GUV by rapidly adjusting its volume and reducing the effective pressure difference across the membrane ( Haleva and Diamant , 2008; Peterlin et al . , 2012 ) . The residual osmotic pressure then necessarily generates a lateral tension in the membrane , which compensates for the internal fluid force . Following the law of Laplace ( σ = ΔP r/2 ) , the imposed osmotic pressure of 0 . 5 MPa translates into the applied membrane tension , σ , of 2 . 5 N m−1 for a vesicle , 10 µm in radius—approximately three orders of magnitude larger than necessary for membrane lysis ( ∼3–5 mN m−1 ) ( Portet and Dimova , 2010 ) . It is instructive to note , however , that the actual tension that develops in the membrane is much lower . GUVs prepared by electroformation invariably display large variations in size , shape , and area to volume ratio . As a result , upon immersion in hypotonic solution , osmotic influx of water first transforms the initial non-spherical shapes into spherical ones ( Hamada et al . , 2011 ) . The drop in the osmolyte concentration during this transformation effectively reduces the osmotic pressure difference , which the vesicle experiences , compared to the applied one . Note also that this reduction in osmotic pressure during the initial shape transformation is different for different vesicles within single populations . Thus , although quantifying actual membrane tension in our experiments is difficult , the observations of tense membranes and subsequent poration indicate that the osmotic perturbation is sufficient for most GUVs to surpass their maximum volume limit , developing appreciable membrane tension and pressure differences . GUV membranes in isotonic media are flaccid ( Figure 1A , C ) . The actual membrane area , A , is higher than the projected area Ap . The excess area , ΔA ( =A − Ap ) , ensures that the flaccid GUV is essentially free of tension ( σ ) and exhibits thermally-excited undulations ( Seifert , 1997 ) . In a hypotonic bath , the GUV assumes a spherical shape and the membrane fluctuations become suppressed ( Figure 1B , D ) , dilating the vesicular volume by >15% . During this transformation , Ap also increases proportionately . In the low-tension regime , where the shape fluctuations are not completely ironed out , Ap increases logarithmically with σ∼ . This is followed by a high tension regime , in which Ap climbs linearly with σ‥ because of the stretching of the molecular areas . The area dilation ( α ) in the membrane is given by the superposition of an area increase due to reduction of membrane undulations and expansion in area per molecule ( Evans and Rawicz , 1990 ) . ( 1 ) α=kT8πκc[ln ( 1+cσκc ) +σE]where κ and E are the elastic moduli for bending and area expansion , respectively , and the coefficient c for tension is 1/24π . The cross-over between the two regimes occurs at the critical tension of E ( =kT/8πκc ) , which for a typical phospholipid membrane falls between 0 . 1–1 mN m−1 . Because of their large dimensions , GUVs develop appreciably greater membrane tensions ( >> 1 mN m−1 ) under even small osmotic gradients ( mM range ) , placing our experiments in the high-tension regime . This is also confirmed by the routine formation of pores evident in our data ( see above ) . ( Sandre et al . , 1999; Brochard-Wyart et al . , 2000; Karatekin et al . , 2003a; Karatekin et al . , 2003b ) An increase in the projected area ( Ap ) of the membrane in the presence of tension can be expressed in terms of a hypothetical radius , Ro , which the vesicle would adopt were its membrane tension absent ( σ = 0 ) . ( 2 ) 4πR02=4πR02[1+σ0E] After a pore opens , the tension in the membrane , σ0 , drops to σ . Thus , ( 3 ) 4πR02[1+σ0E]=4πR02[1+σE]+πr2which yields an estimate for the critical radius to which the pore must grow to relax the membrane tension completely . ( 4 ) rc=2R0 ( σ0E ) 1/2 Rearranging the Equation ( 3 ) in terms of critical radius , the stress equation can be written in terms of a vesicle's geometric parameters ( under conditions of no leakage ) . ( 5 ) σσ0=1−r2rc2−4 ( Ri2−R2 ) rc2 This equation describes two conditions for tension–relaxation following pore opening . First , as the pore grows , the first negative term in the equation above increases , reducing membrane tension consistent with the physical picture that pore opening causes lipids to distribute over a smaller area , thus reducing tension . Second , the efflux of the vesicular content following the opening of the pore reduces R , making the second negative term larger , reducing tension ( σ ) . Together , they set the stage for pore closure . Solute efflux through an open pore in an osmotically stretched membrane occurs under a complex hydrodynamic scenario . Pore radius , vesicle volume , and osmotic pressure differences all change with time , and persistent excess solute concentration maintains conditions for water influx , all of which influence shear stresses associated with the net outward flow . Below , we consider the effusion mechanism following Levin and Idiart ( Idiart and Levin , 2004; Levin and Idiart , 2004 ) . A simple diffusion analysis , such as summarized below , provides a comparison between the amount of solute released in each cycle for the experimental life-time of microscopic pores , which we witness . Diffusive current through a pore of size , r , is given by ( 6 ) j=πr2cDR , where c is the sucrose concentration , D is the solute diffusivity , and R represents the vesicle radius . Comparing the diffusive current with the rate of drop of sucrose concentration within the GUV , then yields , ( 7 ) 43πR3dcdt=−j Solving the differential Equation ( 7 ) above , we find that the concentration decay adopts an exponential profile , ( 8 ) c=c0e−t/τwhere the characteristic effusion time τ is ( 9 ) τ=4R23r2D Using D = 10−9 m2/s for sucrose in water and pore size , r = 5 μm , we find that for a GUV of radius 10 μm , the effusion time is ∼0 . 5 s , comparable to our experimental estimate for the lifetime of pores ( <1 s ) . This then suggests that pore lifetimes are sufficient to allow partial solute leakage ( fractional loss , ∼1/e ) required to relax membrane tension and promote pore closure through effusion alone per cycle . Although the actual dynamics of solvent and solute transport across osmotically imbalanced vesicles are likely to be much more complex , the model above provides approximate estimates for the expected values . Sphingomyelin ( chicken egg ) and cholesterol were purchased from Carbosynth , Berkshire , UK . POPC ( egg ) ( 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine ) and Rhodamine-DPPE ( lissamine rhodamine B 1 , 2-dihexadecanoyl-sn-glycero-3-phosphoethanolamine , triethylammonium salt ) , ( also abbreviated Rhodamine-DHPE ) were acquired from Avanti Polar Lipids , Alabama , USA . Sphingomyelin-Atto647N ( SM-647N ) was from Atto-Tec , Germany . Sucrose and glucose were from USB Corporation , Cleveland , OH , USA . Appropriate amounts of chloroform solutions of desired lipid mixtures doped with a small concentration of lipid-conjugated fluorescent dye were deposited onto clean ITO-coated glass surfaces within the area delimited by a small O-ring , and allowed to dry . Subsequently , the resulting dried lipid cake—containing ∼60 µg of lipids and 0 . 5 mol% lipid-conjugated dye—was hydrated with 300 µl sugar solution of choice , flooding the O-ring enclosed area to the rim . The hydrated sample was then carefully covered by placing a second ITO-coated glass slide , avoiding entrapment of air bubbles . Electroformation ( Angelova et al . , 1992 ) was carried out at 45°C , above the gel–fluid transition temperatures of the lipid mixtures , using a commercial Vesicle Prep Pro ( Nanion , Munich , Germany ) chamber . Application of an AC current at 5 Hz and 3 V for 120 min yielded high abundance of 5–50 µm sized GUVs with excellent reproducibility . Same as electroformation above , except that no electrical current was applied ( Morales-Penningston et al . , 2010 ) . A DeltaVision microscope ( Applied Precision , Inc . , Washington , USA ) , fitted with a PLAPON 60XO/1 . 42 NA oil-immersion objective from Olympus and DAPI , TRITC , FITC , and CY5 Semrock filters ( New York , USA ) , was used for imaging of GUVs in real-time using wide-field deconvolution fluorescence microscopy . Samples were imaged in 8-well chambers fitted with coverslip bottoms ( Nunc , Rochester , USA ) . The 8-well chamber was fitted inside a custom made housing attached to a heating/cooling system , which also was designed to regulate the temperature of the objective . This assured that temperature differences between the sample and the lens would be kept to a minimum , and thus potential convective water flow inside the sample avoided . The temperature was monitored using a thermostat that was submerged into the sample volume . Briefly , 5 µl of sugar-encapsulating GUVs were placed inside a well and gently diluted in 200 µl of deionized water ( 18 megaohm cm ) . The osmolyte-loaded GUVs subsequently settled to the bottom of the coverslip . Osmotic differentials were generated in all experiments by diluting the extra-vesicular bath with deionized water . Images were processed using ImageJ—a public-domain software obtained from http://rsbweb . nih . gov/ij/ .	All living cells are surrounded by a membrane that water can pass through . However , water often contains other molecules called solutes , and many of these cannot pass through the cell membrane . If the concentration of solutes outside the cell is , say , suddenly decreased , then water molecules will tend to move into the cell to lower the solute concentration there . This process , which is called osmosis , strives to equalize the solute concentrations inside and outside the cell . Osmosis can have dramatic consequences for cells . Animal cells need to be bathed in water to survive , but if the solute concentration outside a cell is higher than inside , the cell can lose a lot of water and die . And if the solute concentration outside is lower , then water enters the cell and it can burst . Single celled microbes use a variety of strategies to counter the movement of water by osmosis: strong cell walls prevent the cell from swelling too much , and channel proteins in the membrane can be opened to allow solutes to pass through . But it is not known how more primitive cells—cells that lived billions of years ago—might have responded to fluctuations in their environment . Oglęcka et al . have now used artificial membranes to make closed compartments called giant vesicles that mimic certain properties of cells . When giant vesicles are filled with a sugar solution and placed in water with a lower concentration of sugar , a series of events takes place that can lead to the sugar concentration inside and outside the vesicle becoming more equal . At first the vesicle expands as water enters . However , as the membrane stretches , a temporary hole opens up , which allows some of the excess solute molecules and water to escape , shrinking the vesicle . This sets up cycles of vesicle expansion and contraction that gradually lead to the solute concentrations on both sides of the membrane becoming more equal . This cyclical expansion and contraction of the vesicle also changes the membrane , decorating it with “domains” of specialized molecules , when expanded and uniform , when shrunk . It is possible that this process may have helped the first primitive cells to survive and , maybe , even benefit from changes in solute concentration in their environment .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "cell", "biology" ]	2014	Oscillatory phase separation in giant lipid vesicles induced by transmembrane osmotic differentials
Activating mutations in FLT3 confer poor prognosis for individuals with acute myeloid leukemia ( AML ) . Clinically active investigational FLT3 inhibitors can achieve complete remissions but their utility has been hampered by acquired resistance and myelosuppression attributed to a ‘synthetic lethal toxicity’ arising from simultaneous inhibition of FLT3 and KIT . We report a novel chemical strategy for selective FLT3 inhibition while avoiding KIT inhibition with the staurosporine analog , Star 27 . Star 27 maintains potency against FLT3 in proliferation assays of FLT3-transformed cells compared with KIT-transformed cells , shows no toxicity towards normal human hematopoiesis at concentrations that inhibit primary FLT3-mutant AML blast growth , and is active against mutations that confer resistance to clinical inhibitors . As a more complete understanding of kinase networks emerges , it may be possible to define anti-targets such as KIT in the case of AML to allow improved kinase inhibitor design of clinical agents with enhanced efficacy and reduced toxicity . Kinase inhibitors are among the fastest growing new class of therapeutics for treating cancer , with 25 new kinase inhibitors approved by the FDA in the last 14 years ( Mullard , 2014 ) . Since these agents almost exclusively target a kinase's highly conserved ATP binding pocket , achieving selectivity is problematic . A limiting feature of kinase inhibitors is that their ability to target multiple wild-type kinases in normal tissues limits the doses that can be used to target the mutant kinase in the tumor tissue . The key question , in light of the complex kinase networks in all cells , is which anti-targets should be avoided , in order to limit toxicity in normal tissues . Our work ( Kung et al . , 2006; Dar et al . , 2012 ) and that of others ( Ahmad and Eisen , 2004; Wilhelm et al . , 2006 ) have highlighted the potent effects of multi-targeted kinase inhibitors , revealing unexpected effects when several kinases are inhibited rather than each one individually ( synthetic lethal effects as well as positive epistatic effects ) . By understanding the synthetic lethal effects on normal cells and developing selective inhibitors which avoid even a small number of ‘off-target’ kinases , we believe that clinical agents with an improved therapeutic index can be developed . One disease where this anti-target ( Dar et al . , 2012 ) concept is particularly needed is AML , a rapidly fatal blood cancer comprising 2% of cancer deaths in the United States in 2013 and a disproportionate number of new cases ( ca . 19 , 000 , 30% ) and deaths ( ca . 10 , 500 , 44% ) relative to all leukemias ( Leukemia and Lymphoma Society , 2014 ) . FDA-approved chemotherapy has not advanced beyond general cytotoxic agents , and no highly active therapies have been approved in over 30 years . The cure-rate for AML remains approximately 25% . Fms-like tyrosine kinase 3 ( FLT3 ) is a receptor tyrosine kinase ( RTK ) regulating hematopoietic differentiation that is mutated in 30–35% of AML cases , making it the most frequently mutated gene in AML ( Mizuki et al . , 2000 ) . 25% of AML patients present with juxtamembrane ( JM ) domain duplications termed internal tandem duplications ( FLT3-ITD ) ( Kindler et al . , 2005 ) . The ITD mutation is thought to render FLT3 constitutively active by disrupting an autoinhibitory function of the JM domain . Patients with FLT3-ITD mutations have significantly increased relapse rates and shortened survival , thus illustrating a major unmet therapeutic need . FLT3-ITD was recently validated as a therapeutic target in AML ( Smith et al . , 2012 ) . Several targeted FLT3 tyrosine kinase inhibitors ( TKIs ) have been investigated; however , none have advanced beyond Phase III clinical trials ( Smith and Shah , 2013 ) . The FLT3 TKI AC220 ( Quizartinib ) ( Chao et al . , 2009 ) achieves substantial reduction in leukemic blasts initially in a high proportion of patients and is the most kinome-wide selective clinical candidate ( Zarrinkar et al . , 2009 ) . However , in spite of its promising selectivity , which is remarkably confined to inhibition of the Class III/PDGFR family of RTK's , AC220-induced myelosuppression represents a major dose-limiting toxicity ( Galanis et al . , 2012; Smith and Shah ) . As a result , while many patients achieve clearance of bone marrow ( BM ) blasts , most experience incomplete recovery of normal blood counts ( CRi ) and remain at risk of complications such as life-threatening infection or bleeding . The Class III family of RTKs ( comprising FLT3 , KIT , CSF1R , PDGFRα , and PDGFRβ ) are important regulators of normal hematopoiesis . In 1995 , Lemischka and coworkers showed that mice lacking function of either Flt3 or Kit maintained overall normal populations ( Mackarehtschian et al . , 1995 ) . However , mice lacking both Flt3 and Kit function had a dramatic reduction of hematopoietic cell numbers , ca . 15-fold white cell depletion , reduction of lymphoid progenitors , and postnatal lethality . We propose that KIT is an anti-target ( Dar et al . , 2012 ) in the context of pharmacologic inhibition of FLT3 . Thus , normal mature hematopoietic populations can be maintained in the context of either Flt3 or Kit inhibition alone but not dual Flt3/Kit inhibition ( Bershtein et al . , 2006 ) . This synthetic lethal toxicity relationship between FLT3 and KIT for maintaining normal hematopoietic populations may explain the adverse side effects of the current kinase targeted drugs in clinical development . In a recent single agent Phase II trial , PKC412 failed to achieve a single complete remission ( CR ) . When combined with cytotoxic agents PKC412 showed some promise , achieving a 25% CR rate , but responses were primarily incomplete recovery of peripheral blood counts ( CRi , 20% ) with over 90% of patients developing grade 3/4 myelosuppression ( Strati et al . , 2014 ) . While AC220 monotherapy impressively demonstrated a 50% CR rate in a Phase II trial , these consisted primarily of CRi ( 45% ) with few real CRs with complete recovery of blood counts ( Cortes et al . , 2013 ) , correlating with the similar potency of these agents for both FLT3 and KIT . A recent study showed increased selectivity of the clinical agent crenolanib for FLT3 over KIT and reinforced the correlation between target inhibition , and anti-target avoidance ( Dar et al . , 2012 ) , which lead to lowered toxicity towards normal hematopoiesis ( Galanis et al . , 2014 ) . However , the potency of crenolanib for KIT remains too high ( IC50 = 67 nM for p-KIT inhibition in TF-1 cells; 65% inhibition at 100 nM , in vitro ) ( Galanis et al . , 2014 ) . This is likely insufficient to fully minimize clinically relevant myelosuppression , as a recent interim analysis reported only a 17% ( 3/18 patients ) composite CR rate in AML patients , with 2/3 of these responders achieving only CRi ( Collins et al . , 2014 ) . These findings highlight the need for new clinical candidates that better minimize KIT and other Class III RTK inhibition . While avoiding inhibition of the presumed anti-target , KIT , is one chemical challenge toward inhibitor design , the emergence of on-target resistance is another clinical challenge . We ( Smith et al . , 2012 ) and others ( Wodicka et al . , 2010 ) have identified the acquisition of secondary FLT3 kinase domain ( KD ) mutations that cause drug resistance as another limitation of current clinically active FLT3 inhibitors . Mutations at the activation loop residue D835 are particularly clinically problematic . These mutations are proposed to bias the kinase toward the constitutively active conformation by disrupting a hydrogen bond from D835 to S838 , and thus limit the efficacy of Type II inhibitors such as AC220 . We have recently proposed that a Type I inhibitor , which binds to the active kinase conformation , would circumvent these mutations that confer resistance to AC220 ( Smith et al . , 2012 ) . New small molecule therapies have been reported to bypass these particular mutations , including crenolanib ( Galanis et al . , 2014 ) , a Type I inhibitor ( Lee et al . , 2014; Smith et al . , 2014 ) , but the CR rate of crenolanib remains modest ( Collins et al . , 2014 ) . Moreover , it is likely that a repertoire of drugs will be necessary to combat emerging resistance . We propose herein a solution to the FLT3/KIT selectivity problem designed to avoid myelosuppression and also retain potency against drug-resistant mutations . The staurosporine scaffold has been utilized pharmacologically for 30 years , and staurosporine analogs have been proven to be potent FLT3 inhibitors ( PKC412 , CEP701 ) ( Strati et al . , 2014 ) , though clinical activity of these compounds has been modest , perhaps caused by lack of potent FLT3 inhibition due to dose-limiting toxicity in vivo . The lactam ring C7 position remains virtually unexplored for modulating selectivity ( Wood et al . , 1999; Bishop et al . , 2000; Heidel et al . , 2005 ) . We recently reported that C7-substituted staurosporine analogs , we term ‘staralogs’ , are potent and selective inhibitors of engineered analog-sensitive ( AS ) kinases ( Lopez et al . , 2013 ) . For example , when C7 ( R1 ) equals isobutyl ( Star 12 ) , AS Src kinase is potently inhibited but WT kinases remain unaffected . However , we also observed that Star 12 , in a panel of 319 kinases , weakly inhibits only one WT kinase , FLT3 ( 57% inhibition at 1 µM; KIT , CSF1R , PDGFRα/β all inhibited <10% ) . Thus , the C7-alkyl group of Star 12 may allow for weak but selective inhibition of FLT3 over the anti-target KIT , which contributes to myelosuppression when FLT3 is also inhibited . Although substitution of an isobutyl group at C7 reduced potency , we hypothesized that combining the FLT3/KIT selectivity of Star 12 and the potency features of PKC412 would generate an optimal FLT3 inhibitor also capable of targeting emerging mutations ( Figure 1A ) . 10 . 7554/eLife . 03445 . 003Figure 1 . Purified kinase assays comparing SAR of staralogs . ( A ) PKC412 ( N-benzoylstaurosporine ) potently inhibits all Class III/PDGFR family hematopoietic stem cell kinases . Staurosporine analog ( staralog ) Star 12 , with large alkyl C7 ( R1 ) substituent is a weak inhibitor of FLT3 but suggested more selectivity for Class III RTKs . ( B ) Table shows purified kinase in vitro assays testing SAR showing optimal potency at R1 = methyl; rigid potency window found with Star 27 with n-propyl-NH2 at R2 but not R3 . Star ## ordered by increasing potency and numbering consistent with prior usage ( Lopez et al . , 2013 ) . Error ranges represent standard error of the mean ( SEM ) and are the result of at least three independent measurements , each in triplicate . ( C ) IC50 values of PKC412 and lead compounds Star 23 and 27 against the hematopoietically relevant five Class III RTKs showing the optimal selectivity achieved for Star 27 . Log10-scale heat map highlights IC50 ratios relative to FLT3 , indicating progression from PKC412 having lower selectivity towards Star 27 having high selectivity . ( D ) Dendrogram showing single point inhibition for 319 kinases for PKC412 . Each value represents the average of two experiments ± SEM ( performed by RBC ) . ( E ) Dendrogram showing single point inhibition for 319 kinases for Star 27 . ( F ) Graph of Tyr kinases . y-axis = potency by single point for Star 27 as a function of the corresponding potency with PKC412 . DOI: http://dx . doi . org/10 . 7554/eLife . 03445 . 003 In an effort to identify an optimal C7 substituent that retains selectivity away from KIT while enhancing potency for FLT3 , we synthesized and tested a panel of C7-substituted staralogs ( Figure 1B ) . The C7 substituent points toward the gatekeeper ( GK ) residue of the kinase ( Lopez et al . , 2013 ) , and FLT3 and KIT possess Phe and Thr gatekeepers , respectively . We chose an unbiased selection of groups easily derived from the chiral pool , and tested these derivatives in a radiolabeled purified kinase assay , finding that the alanine-derived methyl group provided the most potent inhibition . Synthesis of C7 analogs was further expedited by testing the fully aglyconic derivatives ( see Figure 1B , entries 1–8 ) . Replacement of the sugar moiety of PKC412 with a pendant amine has proven to be a means to reduce the chemical synthetic burden of carbohydrate synthesis while simultaneously maintaining potency . Attachment of n-propyl amine to mimic the N-methyl amide of PKC412 produced comparable values ( Figure 1B , Star 27: 1 . 5 nM; PKC412: 2 nM ) . Interestingly , both Star 27's one carbon homolog ( Star 23 ) and regioisomer ( Star 27-iso ) reduced activity , further indicating a well-defined structure activity relationship ( Figure 1B ) . We next evaluated our lead inhibitors to show the role of C7 modification on the ratio of KIT/FLT3 IC50s . While PKC412 exhibited potent inhibition of all five Class III RTKs ( Figure 1C , IC50 = <0 . 5–36 nM ) , Star 23 showed improved differentiation between FLT3 and other Class III RTKs ( IC50 = 9 – > 10 , 000 nM ) , and Star 27 proved to have the best selectivity ( IC50 = <0 . 6–9880 nM ) . This selectivity increases from ( a ) KIT/FLT3 ratio of >72 for PKC412 to >1700 for Star 27 and ( b ) CSF1R/FLT3 ratio of 8 for PKC412 to >16 , 500 for Star 27 ( Figure 1C ) . Correlation is seen with inhibition of kinases containing Phe GKs ( e . g . FLT3 ) and avoidance of those containing Thr GKs ( e . g . KIT ) for C7-Me containing staralogs . We then profiled Star 27 against 319 purified kinases using the same assay conditions used previously for PKC412 . We chose to screen for additional targets at 1 µM , or 1000-fold above the IC50 value for the desired target , which captures the maximum number of targets of each inhibitor , dendrograms showing in Figure 1D ( PKC412 ) and Figure 1E ( Star 27 ) . At this high concentration , the two drugs inhibit many of the same targets . Yet the two drugs show important differences in the tyrosine kinase family . The percent inhibition of 54 Tyr kinases by Star 27 with respect to their percent inhibition by PKC412 is shown in Figure 1F . This plot indicates a substantial shift towards inhibition by PKC412 but not Star 27 and correlates with a high number of Thr GK residues ( 42 Thr GKs for Tyr kinases vs 34 Thr GKs in non-Tyr kinases ) . This shift is particularly striking for the entire Src family ( Figure 1D , pink triangles ) , all members of which possess Thr GKs . Conversely , Tyr kinases that possess Phe GKs mostly display high and equipotent inhibition by both Star 27 and PKC412 ( including FLT3 , TRKA , TRKB , and TRKC ) . We compared PKC412 to Star 27 for the ability of each to inhibit cell proliferation of human cell lines addicted to FLT3 or KIT . We employed Molm14 and MV4;11 cells , harboring the FLT3-ITD mutation ( hetero- and homozygous , respectively ) ; HMC1 . 1 cells ( dependent on KIT V560G ) ; and K562 cells ( which express the BCR-ABL fusion kinase ) as a control for non-KIT-related toxicity . Sorafenib , AC220 , and ponatinib all displayed equipotent inhibition of FLT3 and KIT-driven cell lines ( see Figure 2A and reference values [de Jong and Zon , 2005; Guo et al . , 2007; Kampa-Shittenhelm et al . , 2013] ) . PKC412 increases this selectivity 22-fold ( HMC1 . 1/MV4;11 ) . Encouragingly , Star 27 inhibited proliferation of the FLT3-ITD+ cells while not affecting HMC1 . 1 proliferation leading to a selectivity window of >122-fold ( HMC1 . 1/MV4;11 ) . 10 . 7554/eLife . 03445 . 004Figure 2 . Cellular proliferation , apoptosis , and biochemical validation of Star 27’s potency against FLT3 mutants , but not in the KIT context . ( A ) Table of cellular IC50 values for leading experimental clinical therapies ( Sorafenib , AC220 , Ponatinib , and PKC412 ) and Star 27 against a panel of AML-relevant human-derived ( MV4;11 and molm14 ) cell lines and toxicity controls ( HMC1 . 1 and K562 ) ( adapted* from Guo et al . , 2007; Kampa-Shittenhelm et al . , 2013; Smith et al . , 2013 ) . Star 27 shows a similar order of magnitude potency against Molm14 and MV4;11 compared to PKC412 . Relevant therapeutic windows between PKC412 and Star 27: an over fivefold increase in selectivity for HMC1 . 1/MV4;11 cells and an 11-fold increase for HMC1 . 1/Molm14 is observed for Star 27 over PKC412 . Replicates shown are the result of at least three attempts , each in triplicate , and error ranges represent the standard error of the mean . ( B ) Normalized caspase-3 negative cells plotted against escalating drug dosage . Star 27 shows a similar degree of apoptosis to PKC412 in Molm14 cells . ( C ) In KIT-addicted HMC1 . 1 cells PKC412 exhibits potent toxicity while Star 27 shows very little up to 2 , 500 nM . Results performed in triplicate with two biological replicates . p < 0 . 01; p < 0 . 001 . ( D ) PKC412 and Star 27 have a similar degree of inhibition against p-FLT3 in Molm14 cells as well as a similar degree for p-STAT5 and p-S6 ( for p-ERK , p-AKT , and p-MEK data not shown ) . In the HMC1 . 1 cells , by contrast , PKC412 and Star 27 show a larger difference in p-KIT inhibition . DOI: http://dx . doi . org/10 . 7554/eLife . 03445 . 004 We next tested if this inhibition of proliferation was manifested in triggering apoptotic cell death . Analysis of caspase-3 activation revealed a similar induction of apoptosis in a dose–response manner between PKC412 and Star 27 in Molm14 cells ( Figure 2B ) . Conversely , PKC412-induced apoptosis in the KIT mutant HMC1 . 1 cells as predicted while minimal apoptosis was observed with Star 27 , further highlighting the selectivity of Star 27 for FLT3 over KIT ( Figure 2C ) . Given the similarity in potency between PKC412 and Star 27 towards FLT3 , we next tested both compounds for their ability to inhibit FLT3 autophosphorylation as well as downstream signaling in Molm14 cells . We observe a similar inhibition of p-FLT3 inhibition between the two drugs and downstream phosphorylation among three canonical signaling arms was uniform between PKC412 and Star 27 ( JAK/STAT , RAS/ERK , and mTOR/AKT/S6; Figure 2D ) . In contrast to similar phospho-signaling inhibitory profiles in Molm14 cells , PKC412 inhibited p-KIT and downstream signaling to a greater degree in HMC1 . 1 cells compared with Star 27 ( Figure 2D ) . We next asked if Star 27’s FLT3/KIT selectivity could maintain efficacy and reduce toxicity in primary patient-derived contexts . Colony-forming assays are useful for this application because they better represent the microenvironment of the BM niche than traditional cell proliferation of primary human cells ( Miller and Lai , 2005 ) . Testing both PKC412 and Star 27 demonstrated their ability to inhibit the growth of primary patient-derived FLT3-ITD+ blasts in colony-forming assays , with both demonstrating >80% inhibition of colony formation at 1000 nM ( Figure 3A , B ) . Because of the importance of maintaining potency against KD mutations in addition to those of the JM domain , we also tested the ability of both drugs to inhibit the colony-forming ability of primary patient blasts containing a FLT3-D835 mutation ( ITD−; point mutational status undetermined ) . Star 27 inhibited about 60% of colonies at 1000 nM while PKC inhibited about 80% ( Figure 3C ) . 10 . 7554/eLife . 03445 . 005Figure 3 . Colony forming assays comparing effect of Star 27 and PKC412 on normal donor-derived stimulated peripheral blood ( SPB ) , normal bone marrow ( BM ) growth , and malignant blast reduction . ( A , B , C ) Colony forming assays comparing effects of Star 27 and PKC412 on primary patient AML circulating blast growth in methylcellulose . Colonies consist primarily of leukemic blasts of various hematopoietic identities . BFU colonies are not shown . Colony numbers scored individually for each replicate 3 cm plate . ( A ) Images showing leukemic blast morphology in DMSO-treated plates and representative images in 1 , 000 nM-treated plates for both drugs as indicated . ( B ) Primary patient FLT3-ITD+ AML circulating blasts . Raw colony numbers range from ca . 20 in PKC412 or Star 27-treated cells at 1 , 000 nM to ca . 1 , 200 for CFUs in DMSO-treated plates . Concentrations shown were applied in triplicate to each concentration for two different cell densities , the larger ( 2 . 5 × 10^5 cells/ml ) showing adequate cell growth and colony formation for statistically significant counts , the averages of those trails then calculated for standard error of the mean . p < 0 . 05 . ( C ) Primary patient FLT3 D835Y AML circulating blasts . Raw colony numbers range from ca . 25 for PKC412 ( or 50 for Star 27 ) at 1 , 000 nM to ca . 120 for DMSO-treated plates . Conditions repeated in duplicate or higher replicates . Colonies scored individually for the highest cell density tested ( 2 . 5 × 10^5 cells/ml ) . Similar to primary ITD patient samples , colonies consist mostly of poly-hematopoietic leukemic blasts . p < 0 . 05 . ( D , E , F ) Normal BM and SPB colony data . ( D ) Images of colonies grown and derived from normal BM ( images of SPB not shown ) in methylcellulose . Raw colony numbers for BM range from zero in most PKC412-treated replicates at 1 , 000 nM to ca . 400 for BFUs and ca . 300 for CFUs in all Star 27-treated conditions . Non-magnified differences are particularly noticeable between the 1 , 000 nM and 250 nM dosages . ( E and F ) Concentrations shown were applied in triplicate to three normal SPB and one normal BM donors , the averages of those trials then calculated for standard error of the mean ( full data not shown ) . Colony counts for DMSO ranged from the low 100's to 400's . Graphs show Star 27 having no effect on hematopoiesis up to 1 , 000 nM while PKC412 eliminates most normal hematopoietic colony forming potential at 1 mM on colony forming units ( CFU ) and blood forming units ( BFU ) , respectively . CFU-GEMM colonies were counted as one CFU plus one BFU . p < 0 . 05; *p < 0 . 001 . DOI: http://dx . doi . org/10 . 7554/eLife . 03445 . 005 Alternatively , we tested PKC412 and Star 27 for their effects on hematopoietic colony formation of normal BM and stimulated peripheral blood ( SPB ) samples . One of the four donor samples is shown in Figure 3D . Figure 3E , F shows quantification of colony counts for both erythroid ( or burst ) -forming units ( BFUs ) and colony-forming units ( CFUs ) . PKC412 demonstrated dose-dependent inhibition of colony formation from normal hematopoetic progenitors while Star 27 demonstrated no significant effect up to a concentration of 1 , 000 nM ( Figure 3E , F ) . Having validated Star 27's biochemical potency and lack of myelosuppression , we next tested its effects in vivo . We chose the established model , Danio rerio ( zebrafish ) for studying hematopoiesis ( Davidson and Zon , 2004; de Jong and Zon , 2005 ) . In this model , embryos injected with FLT3-ITD and treated with AC220 have been shown to recapitulate knock-down of FLT3-ITD in clinical studies ( He et al . , 2014 ) . This model also provides for robust and sensitive measures of in vivo myelopoiesis ( He et al . , 2014 ) . We treated embryos with Star 27 up to 10 μM and studied the effects on WT morphology at 3 days post fertilization ( dpf ) , finding no change in heart and tail morphology ( tail length and lack of curvature , see Figure 4A–D ) . In contrast , PKC412 showed substantial morphological defects to the heart at 1 μM vs vehicle ( Figure 4E–I ) , tail length and tail curvature at 1 μM and 2 . 5 μM ( see Figure 4E , G , respectively ) . 10 . 7554/eLife . 03445 . 006Figure 4 . Effects of Star 27 on D . rario ( zebrafish ) WT morphology and myelopoiesis , and FLT3-ITD AML context . Effect of Star 27 on zebrafish normal myelopoiesis and the FLT3/ITD-induced myeloid cells expansion . ( A–D ) The effect of Star 27 on zebrafish embryonic development at 3 dpf , showing no noticeable morphological defects up to 10 μM . ( E–I ) The effect of PKC412 on zebrafish embryonic development at 3 dpf , showing substantial pericardial defects beginning at 500 nM ( data not shown ) , and tail curvature and length defects beginning at 1 μM and 2 . 5 μM , respectively vs . vehicle . ( J , K , M ) The effect of 10 μM Star 27 treatment on mpo+ myeloid cells development in the posterior blood island ( PBI ) at 30 hpf , showing no statistically significant change . ( J , L , M ) The effect of 10 μM PKC412 treatment on mpo+ myeloid cells development in the PBI at 30 hpf , showing a statistically significant granulogenesis/myelosuppression . ( N–P ) Three categories of mpo transcription ( N , normal; O , intermediate; P , severe ) were defined based on the WISH results ( three experiments ) . ( Q ) The rescue effect of 10 μM Star 27 treatment on FLT3/ITD-induced mpo+ myeloid cells expansion at 30 hpf , showing rescue of normal phenotype approaching that seen for AC220 ( He et al . , 2014 ) . Scale bar equals 500 μm . Blue arrows indicate the pericardial edema , green arrows indicate tail shortening and curving , and red arrows indicate the FLT3-ITD AML mpo+ myeloid cells expansion . PKC412 treatment on FLT3/ITD-induced mpo+ myeloid cells expansion at 30 hpf , showing an efficaciousness similar to Star 27 , consistent with Figures 1–3 . For all experiments: Zebrafish embryos were collected and kept in standard E3 medium at 28°C . Different concentrations of either drug were added to the E3 medium from 6 hr post fertilization ( hpf ) to 3 days post fertilization ( dpf ) . Embryos treated with DMSO or 10 μM drug from 6 to 30 hpf were collected for mpo whole mount in situ hybridization ( WISH ) analysis . 80 ng plasmid DNA containing FLT3/ITD sequence was microinjected into one-cell stage embryos , and the uninjected embryos were used as control . FLT3/ITD-injected embryos were treated with DMSO or 10 μM drug from 6 to 30 hpf . Embryos were collected at 30 hpf for mpo WISH analysis . p <0 . 01 . DOI: http://dx . doi . org/10 . 7554/eLife . 03445 . 006 We next examined the effect of Star 27 and PKC412 on WT granulopoiesis as a measure of KIT-related myelopoiesis/myelosuppression measured by myeloperoxidase ( mpo ) -stained in situ hybridization in the posterior blood island ( PBI ) . Representative cross sections for the PBI for DMSO , Star 27 , and PKC412 are shown ( Figure 4J , K , L , respectively ) . Quantification of mpo dots in Star 27-treated embryos shows a non-significant difference between 10 μM and DMSO treatments of 30 hr post fertilization ( hpf , Figure 4J , K , M ) . In contrast , PKC412 showed statistically significant suppression of myelopoiesis at 10 μM ( Figure 4M ) . Efficacy in the AML FLT3-ITD context is quantified by binning the FLT3-ITD blast phenotype into Normal , Intermediate , and Severe states ( Figure 4N–P , respectively ) ( He et al . , 2014 ) . Star 27 did have efficacy in preventing the spread of neutrophilic blasts in FLT3-ITD-transduced embryos ( Figure 4Q ) similar to that previously seen with AC220 ( He et al . , 2014 ) . Similar to Star 27 , PKC412 also showed an ability to limit the spread of FLT3-ITD-injected leukemic blasts ( Figure 4Q ) . The structural basis of selectivity of Star 27 for FLT3 over other Class III RTKs was computationally investigated , revealing two features of staralog binding that may explain Star 27's selectivity for FLT3 over KIT , vs PKC412's equipotency towards both kinases . ( a ) Electronics: we calculate that PKC412 maintains a partial positive charge at its lactam C7 position , matching the partial negative charge of KIT's Thr GK ( Figure 5A–C ) , while Star 27's C7 methyl renders its lactam relatively neutral ( Figure 5D–F ) , reducing this affinity interaction . ( b ) Sterics: staralogs with no C7 substitution maintain a flat lactam for binding to the hinge region , adjacent to the GK . Based on crystallographic data from KIT's active conformation ( Gajiwala et al . , 2009 ) , a simple steric argument may explain why Star 27's C7 methyl group prevents binding to KIT's restricted ATP binding pocket while PKC412 potently binds ( data not shown ) . Conversely , FLT3's active site is hypothetically large enough to accommodate the 17 . 4 Å3 van der Waal's volume of methyl ( vs . 1 . 17 Å3 for H ) . Taken together both arguments may explain the greater affinity of PKC412 for both kinases and of Star 27 for FLT3 but not KIT ( calculated using MOE ver2013 . 0801 ) . This model may account for the selectivity seen for Star 27 for kinases bearing Phe GKs ( FLT3 , TRK , etc ) and away from those bearing Thr GKs ( KIT , CSF1R , Src family , other Tyr kinases , see Figure 1D ) . 10 . 7554/eLife . 03445 . 007Figure 5 . Electronic model of selectivity . Star 27-des methyl ( PKC412-relevant ) and Star 27 are depicted in 3D with electron potential mapping ( EPM ) to highlight the difference of a single methyl group . ( A–C ) Star 27-des methyl maintains a partial positive charge at the C7 position which matches the partial negative charge on the Thr 670 gatekeeper of KIT , potentially explaining PKC412's relative affinity for KIT . ( D–F ) Star 27 maintains a neutral charged surface at C7 due to the presence of the methyl group , potentially explaining its lack of binding to KIT . DOI: http://dx . doi . org/10 . 7554/eLife . 03445 . 007 Mutations that confer drug resistance to existing Type II FLT3 inhibitors AC220 and sorafenib commonly prevent the kinase from efficiently adopting an inactive conformation required for drug binding . Type I FLT3 inhibitors , such as PKC412 and Star 27 , are predicted to be less vulnerable to such mutations . We tested both inhibitors against a panel of 17 cell lines containing different FLT3 TKI mutations with varying resistance to the leading clinical candidate therapies , sorafenib , ponatinib , and AC220 ( Figure 6A ) . IC50 values are presented as log10-fold resistance in a color-coded heat map to highlight the range of offset . Using published IC50 values for sorafenib , AC220 , and ponatinib for comparison ( Guo et al . , 2007; Kampa-Shittenhelm et al . , 2013; Smith et al . , 2013 ) , we observed equipotent values for PKC412 against most KD point mutants . Intriguingly , Star 27 maintained similar potency against most mutants compared to FLT3-ITD alone . 10 . 7554/eLife . 03445 . 008Figure 6 . Heatmap of cellular EC50 values for leading clinical therapies ( S = Sorafenib , A = AC220 , P = Ponatinib , PKC412 ) and Star 27 against a panel of AML-relevant human- and mouse-derived drug-resistant cell lines . ( A ) ( adapted from Guo et al . , 2007; Kampa-Shittenhelm et al . , 2013; Smith et al . , 2013 ) . Log10-scale fold resistance between Ba/F3 FLT3 ITD and drug-induced mutations ( ITD + KD double mutant ) shown via heat map . Point mutations corresponding to each drug-relevant resistance shown . Star 27 maintains potency between FLT3 ITD and resistance mutants comparable to PKC412 . Mutations derived independently from both saturating mutagenesis and patient-derived samples [Smith et al . , 2012] . Replicates shown are the result of at least two or three attempts , each in triplicate , and error ranges represent the standard error of the mean . ( B and C ) Crenolanib's effect on p-KIT inhibition and of colony growth in normal BM . ( B ) Crenolanib inhibits p-KIT in HMC1 . 1 cells at 12 nM IC50 . This level of inhibition translates to downstream kinases , with inhibition of p-AKT ( S473 ) and p-S6 ( S235/S236 ) . ( C ) Crenolanib potently inhibits normal BM colony formation at <63 nM and ca . 63 nM IC50's of CFU and BFU colonies , respectively . DOI: http://dx . doi . org/10 . 7554/eLife . 03445 . 008 Crenolanib has been reported to inhibit p-KIT in TF-1 cells ( Galanis et al . , 2014 ) and HMC1 . 2 cells ( Smith et al . , 2014 ) at 67 nM and <100 nM IC50's , respectively . In order to directly compare our results with PKC412 and Star 27 ( Figure 2D ) , we tested crenolanib's biochemical inhibition of p-KIT in HMC1 . 1 cells . Figure 6B shows that crenolanib inhibits p-KIT in these cells at an IC50 of 12 nM . This potency against KIT anti-target is propagated down to inhibition of p-AKT and p-S6 kinases . Crenolanib's potency against p-FLT3 has been reported to be 5 nM ( Galanis et al . , 2014 ) . Crenolanib has been reported to have only modest inhibition of normal BM colony formation of CFUs and no inhibition of BFUs up to 200 nM drug treatment ( Galanis et al . , 2014 ) . However , the lack of both higher drug dosing ( >200 nM ) and statistically comparable variation prompted us to retest crenolanib in the same colony forming assay up to 1 , 000 nM , to compare to PKC412 and Star 27 ( Figure 3D–F ) . We observe crenolanib to have a more substantial dampening of normal colony growth of both CFU and BFU subtypes in normal BM ( see Figure 6C ) . The aggressive nature of FLT3-driven AML and the lack of effective and well-tolerated targeted therapy represent a major unmet therapeutic need . Currently , the most promising clinical targeting agents cause myelosuppression in most patients regardless of remission status . As kinase inhibitor-based therapies evolve , refined selectivity against particular off-targets ( so-called anti-targets [Dar et al . , 2012] ) will drive chemical design ( Ciceri et al . , 2014 ) . Avoiding specific anti-targets will hopefully lead to a reduction in dose-limiting toxicities that will allow more complete inhibition of the oncogenic drivers of disease . A former targeted clinical candidate , tandutinib ( MLN518 ) , may further demonstrate directly the effects of the FLT3/KIT synthetic lethal toxicity model ( Griswold et al . , 2004 ) . MLN518 was indicated to have similar IC50s for FLT3 and KIT phosphorylation ( 220 and 170 nM , respectively ) . While MLN518 showed promise with reducing FLT3-ITD+ blasts in primary patient-derived samples , exposure of the drug to the BM of healthy individuals at a dose necessary for primary FLT3-ITD+ CFU reduction , led to severe depletion of healthy hematopoietic colony formation at 1 µM . The most direct evidence exists for FLT3/KIT as a synthetic lethal toxic pair , however , other Class III RTKs may exhibit synthetic lethal effects in terms of dose-limiting toxicity when treating FLT3-driven disease . Another Class III RTK , CSF1R , has been shown via the knockout ‘toothless’ rat model ( tl = csf1r−/− ) to be important to osteoclastogenesis ( Chen et al . , 2011 ) , colon development ( Huynh et al . , 2013 ) , and for normal macrophage and dendritic cell ( DC ) production and maintenance ( Stanley et al . , 1997 ) . Perhaps most importantly , tl rats have been reported with a 32% platelet loss , and this thrombocytopenia was not reversed by BM transplant ( Thiede et al . , 1996 ) . Star 27 exhibits a large selectivity window with a >16 , 500/1 CSF1R/FLT3 IC50 ratio ( Figure 1C ) , which may provide a clinical benefit . However , more work is needed to determine the synthetic lethal toxicity of CSF1R since other reports suggest an opposite and more complicated role for it in hematopoiesis ( Dai et al . , 2002 ) . Our focus on the staurasporine scaffold maintains the beneficial aspects of PKC412's potency against drug-resistant FLT3 mutations ( Figure 6A ) . Since both staralogs are Type I inhibitors , they are not expected to suffer from the Type II-resistance model at the GK ( F691L ) and especially activation loop ( D835 ) mutations observed in the clinic ( Smith et al . , 2012 ) . The reduced potency of Star 27 against FLT3 relative to PKC412 in Ba/F3 cells ( Figure 6A ) is mitigated by the more relevant comparison in human cells ( Figure 2A ) as well as the predicted increase in maximum tolerated dosing for clinical candidates avoiding the KIT anti-target . Robustness against emergent mutations is a desirable feature of PKC412 but its pan-Class III RTK inhibition ( especially KIT ) is likely a major limitation addressed here . Remarkably , the addition of a single methyl group ( CH3 ) likely caused the observed decrease in toxicity . The strategy reported here for gaining selectivity for FLT3 inhibition over KIT and other Class III RTKs may have importance in other hematopoietic pathology as well . FLT3 has been shown to be important for classical DC maintenance and necessary for inflammatory DCs ( Ramos et al . , 2014 ) . Consequently , FLT3 has been proposed to be a drug target for autoimmunity and inflammation , particularly atherosclerosis . The staralog , CEP701 , has been investigated in this context but it is expected that its promiscuity , particularly among KIT and CSF1R , will hamper clinical development ( Dai et al . , 2002 ) . Star 27 should enable a reduction in FLT3-associated inflammation and autoimmunity while allowing KIT and CSF1R to compensate with classical DC maintenance . Furthermore , the generality of our electrostatic Phe GK inhibition coupled with Thr GK avoidance model ( Figures 1D , 5 ) may guide future medicinal chemistry efforts , while our in vivo study shows any promiscuity seen in Figure 1E does not compromise Star 27's efficaciousness in a well-validated in vivo model of hematopoiesis ( Davidson and Zon , 2004; de Jong and Zon , 2005; He et al . , 2014 ) . Recent clinical experience with targeted inhibition of mutated kinases in cancer suggests that efficacy is driven by maximal pathway inhibition . In the case of vemurafenib , <80% pathway inhibition in BRAF ( V600E ) mutant melanoma afforded no tumor shrinkage , while >90% inhibition showed profound clinical benefit ( Bollag et al . , 2010 ) . The challenge is that inhibitors of the oncogene also exhibit off-target effects on closely related kinases that inevitably lead to dose limiting toxicities . The goal of advancing improved molecular-targeted therapies is to identify the key anti-targets that drive toxicity and develop agents that avoid these effects . Our work highlights the synthetic lethal toxicity of an anti-target ( KIT ) when the oncogenic kinase ( FLT3 ) is inhibited . As more complete understanding of kinase networks emerge it may be possible to delineate the anti-targets to avoid in numerous disease settings to allow improved kinase inhibitor design that can lead to greater ability to inhibit the disease causing pathway activation while avoiding systemic toxicities . PKC412 was purchased from Selleckchem ( Houston , TX ) . Staralogs ( Figure 1B ) were synthesized according to precedent ( Lopez et al . , 2013 ) . This protocol was adapted for synthesis of Star 27 . Materials obtained commercially were reagent grade and were used without further purification . 1H NMR and 13C NMR spectra were recorded on Varian 400 ( Palo Alto , CA ) or Brucker 500 ( Billerica , MA ) spectrometers at 400 and 125 MHz , respectively . Low resolution mass spectra ( LC/ESI-MS ) were recorded on a Waters Micromass ZQ equipped with a Waters 2695 Separations Module and a XTerra MS C18 3 . 5 mm column ( Waters ) . Reactions were monitored by thin layer chromatography ( TLC ) , using Merck silica gel 60 F254 glass plates ( 0 . 25 mm thick ) . Flash chromatography was conducted with Merck silica gel 60 ( 230–400 mesh ) . For Figure 1B purified FLT3 WT was diluted in kinase reaction buffer ( 10 mM HEPES [pH 7 . 6] , 10 mM MgCl2 , 0 . 2 mM DTT , 1 mg/ml BSA ) to a concentration of 2 nM and pre-incubated with 2 . 5% ( vol/vol ) DMSO , 100 µM peptide ( sequence EAIYAAPFKKK [Abltide] ) , and varying concentrations of inhibitor from 20 µM by fourths down to 1 . 2 nM for 10 min pre-incubation . Kinase reactions were initiated by the addition of 100 µM cold ATP supplemented with 2 . 5 µCi γ32P ATP per well and allowed to proceed at RT . At 15 min , 3 µL of the reactions were spotted onto phosphocellulose sheets ( P81 , Whatman ) and subsequently soaked in wash buffer ( 1 . 0% [vol/vol] phosphoric acid ) at least five times for 5 min each . The sheets were then dried , and transferred radioactivity was measured by phosphorimaging using a Typhoon scanner ( Molecular Dynamics ) . Radioactive counts were quantified using ImageQuant software ( GE Healthcare Biosciences , Pittsburgh , PA ) , and titration data were fit to a sigmoidal dose response to derive IC50 values using the Prism 4 . 0 software package . Experiments were performed 2–4 times , each in triplicate , with eight dosages , to derive standard error of the mean values . For Figure 1C , values obtained at Reaction Biology Corporation ( Malvern , PA ) . Briefly , specific kinase/substrate pairs were prepared in reaction buffer; 20 mM Hepes pH 7 . 5 , 10 mM MgCl2 , 1 mM EGTA , 0 . 02% Brij35 , 0 . 02 mg/ml BSA , 0 . 1 mM Na3VO4 , 2 mM DTT , 1% DMSO . Compounds were delivered into the reaction , followed 20 min later by addition of a mixture of ATP ( Sigma , St . Louis , MO ) and 33P ATP ( PerkinElmer ) to a final concentration approximating each kinase's Km-ATP ( FLT3: 50 μM; KIT: 150 μM; CSF1R: 150 μM; PDGFRa: 5 μM; PDGFRb: 50 μM ) . Reactions were carried out at 25°C for 120 min , followed by spotting of the reactions onto P81 ion exchange filter paper ( Whatman ) . Unbound phosphate was removed by extensive washing of filters in 0 . 75% phosphoric acid . After subtraction of background derived from control reactions containing inactive enzyme , kinase activity data were expressed as the percent remaining kinase activity in test samples compared to vehicle ( dimethyl sulfoxide ) reactions . IC50 values and curve fits were obtained using Prism ( GraphPad Software ) . Data obtained in singlicate with two biological replicates and IC50 values presented ±SEM in Figure 1C . 319 kinases were tested in a duplicate single dose ( 1 μM ) format using a 0P-labeled ATP activity assay performed by Reaction Biology Corp . Assays performed at identical conditions to those of PKC412 ( Anastassiadis et al . , 2011 ) . Stable Ba/F3 lines were generated by retroviral spinfection with the appropriate mutated plasmid as previously described ( Smith et al . , 2012 ) . Exponentially growing cells ( 5 × 103 cells per well ) were plated in each well of a 96-well plate with 0 . 1 ml of RPMI 1640 + 10% ( vol/vol ) FCS containing the appropriate concentration of drug in triplicate , and cell viability was assessed after 48 hr as previously described ( Smith et al . , 2012 ) . Log10-fold selectivity heat maps for Figures 1C , 6A: briefly , IC50 values were calculated as shown in figures , then ratios of RTK/FLT3 ( Figure 1C ) or point mutant/BaF3 FLT3 ITD ( Figure 6A ) were calculated . Log10-scale transformation of ratios , followed by conditional formatting using xcel spreadsheet software yielded the colored diagrams shown . Exponentially growing cells were plated in the presence of PKC412 or Star 27 in RPMI + 10% ( vol/vol ) FCS for 48 hr . Cells were fixed with 4% ( vol/vol ) paraformaldehyde ( Electron Microscopy Sciences ) and permeabilized with 100% ( vol/vol ) methanol ( Electron Microscopy Sciences , Hatfield , PA ) followed by staining with a FITC-conjugated antiactive caspase-3 antibody ( BD Pharmingen , San Jose , CA ) . Cells were run on a BD LSRFortessa cell analyzer , and data were analyzed using FlowJo ( Tree Star Inc . , Ashland , OR ) . Percentage of live cells was determined by negative staining for activated caspase-3 ( see Figure 2B , C ) . Exponentially growing Molm14 , HB119 , or Ba/F3 cells stably expressing mutant isoforms were plated in RPMI medium 1640 + 10% ( vol/vol ) FCS supplemented with PKC412 or Star 27 at the indicated concentration . HMC1 . 2 cells were cultured and treated in IMDM + 10% FCS . After a 90-min incubation , the cells were washed in PBS , lysed , and processed as previously described . Immunoblotting was performed using anti-phospho-FLT3 , anti-phospho-KIT , anti-phospho-STAT5 , anti-STAT5 , anti-phospho-ERK , anti-ERK , anti-phospho-S6 , anti-S6 , anti-KIT ( Cell Signaling , Beverley , MA ) and anti-FLT3 S18 antibody ( Santa Cruz Biotechnology , Santa Cruz , CA ) . Wild-type ( WT ) zebrafish were maintained under standard conditions , and embryos were staged as described ( He et al . , 2014 ) . The study was approved by the Committee of the Use of Live Animals for Teaching and Research in The University of Hong Kong . Primary AML blood samples and/or marrow aspirates were obtained on an IRB-approved protocol at the University of California , San Francisco . Informed consent was obtained in accordance with the Declaration of Helsinki . Mononuclear cells were purified by density centrifugation ( Ficoll–Paque Plus , GE Healthsciences ) before cryopreservation in 10% ( vol/vol ) DMSO or FCS ( primary ITD and D835-mutant assays , Figure 3A–C ) or immediate use in assays ( Normal BM and SPB , Figure 3D–F ) . Normal BM and stimulated peripheral blood ( SPB ) samples were collected from donors at the UCSF oncology/hematology division . Cells were then suspended in MethoCult methylcellulose ( Stem Cell Technologies , product no . H4435 Enriched , Canada ) in 15-ml Falcon tubes , vigorously vortexed , bubbles settled over 10 min . Triplicates then plated via blunt-nosed syringe ( 1 . 1 ml each ) into 3-cm dishes , avoiding bubble generation , and each dish placed in a larger 10-cm dish with water trough in an incubator ( 37°C , 5% CO2 ) for 13–14 days . Colonies were counted individually using traditional microscopy . CFU and BFU types reported separately ( CFU-GEMM colonies counted as one of each ) . Results shown for FLT3 ITD primary blasts ( Figure 3B ) are a combination of leukemic blast colonies and BFU colonies . Mutations isolated in the screen were engineered into pMSCVpuroFLT3–ITD by QuikChange mutagenesis ( Stratagene , La Jolla , CA ) as previously described ( Smith et al . , 2012 ) .	Major advances in cancer therapy have improved the treatment options for many patients . However , many cancer treatments are toxic or have severe side effects , making them difficult for patients to tolerate . One cause of these side effects is that many cancer therapies kill both normal cells and cancer cells . Developing cancer therapies that are more targeted is therefore a priority in cancer research . Acute myeloid leukemia is a type of blood cancer that has proven difficult to treat without causing serious side effects . This cancer is very aggressive and only about 1 in 4 patients are successfully cured of their cancer . At present , physicians treat acute myeloid leukemia with chemotherapy , which kills both the cancer cells and some of the patient's healthy cells . Many patients with acute myeloid leukemia have mutations in the gene encoding an enzyme called Fms-like tyrosine kinase 3 ( FLT3 ) . This mutation makes the enzyme permanently active , and patients with the mutation have a greater risk of their cancer recurring or death . Scientists have recently discovered that treatments that inhibit the FLT3 enzyme can be effective against cancer . However , the drugs investigated so far also interfere with the patient's ability to produce new blood cells , which can lead to infections or an inability to recover from bleeding . Therefore , no new drugs have yet been approved for general use . Warkentin et al . suspected the reason for the adverse effects of FLT3 inhibitors is that these drugs also inhibit another enzyme necessary for blood cell production . Previous work showed that inhibiting one or the other of the enzymes still allows blood cells to be produced as normal: it is only when both are inhibited that production problems arise . Warkentin et al . therefore looked for a chemical that inhibits only the FLT3 enzyme and found one called Star 27 . Tests revealed that this inhibits FLT3 and prevents the growth and spread of cancerous cells but does not impair blood cell production . Additionally , Star 27 continues to work even when mutations arise in the cancer cells that cause resistance to other FLT3 inhibitors . The findings demonstrate that when it comes to drug development , it is sometimes as important to avoid certain molecular targets as it is to hit others . Understanding the network of enzymes that FLT3 works with could therefore help researchers to develop more effective and safer cancer treatments .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "biochemistry", "and", "chemical", "biology", "cancer", "biology" ]	2014	Overcoming myelosuppression due to synthetic lethal toxicity for FLT3-targeted acute myeloid leukemia therapy
Drosophila melanogaster males perform a series of courtship behaviors that , when successful , result in copulation with a female . For over a century , mutations in the yellow gene , named for its effects on pigmentation , have been known to reduce male mating success . Prior work has suggested that yellow influences mating behavior through effects on wing extension , song , and/or courtship vigor . Here , we rule out these explanations , as well as effects on the nervous system more generally , and find instead that the effects of yellow on male mating success are mediated by its effects on pigmentation of male-specific leg structures called sex combs . Loss of yellow expression in these modified bristles reduces their melanization , which changes their structure and causes difficulty grasping females prior to copulation . These data illustrate why the mechanical properties of anatomy , not just neural circuitry , must be considered to fully understand the development and evolution of behavior . “The form of any behavior depends to a degree on the form of the morphology performing it . ” – West-Eberhard ( 2003 ) Over 100 years ago in Thomas Hunt Morgan’s fly room , Alfred Sturtevant described what is often regarded as the first example of a single gene mutation affecting behavior ( Sturtevant , 1915; reviewed in Drapeau et al . , 2003; Cobb , 2007; Greenspan , 2008 ) : he noted that yellow mutant males , named for their loss of black pigment that gives their body a more yellow appearance ( Figure 1A ) , mated successfully with wild-type females much less often than wild-type males . In 1956 , in what is often considered the first ethological study ( reviewed in Cobb , 2007; Greenspan , 2008 ) , Margaret Bastock compared courtship of yellow mutant and wild-type males and concluded that despite all courtship actions being present , loss of yellow function likely reduces courtship vigor or drive , leading to copulation inhibition ( Bastock , 1956 ) . Despite more recent data consistent with this hypothesis ( Drapeau et al . , 2003 ) , the precise mechanism by which the yellow gene affects male mating success in D . melanogaster has remained a mystery . Consequently , Bastock’s statement about yellow from her 1956 paper is equally true today: “It seemed worthwhile therefore to examine more closely one example of a gene mutation affecting behavior and to ask two questions , ( 1 ) how does it bring about its effect ? [and] , ( 2 ) what part might it play in evolution ? ” The D . melanogaster yellow gene encodes a protein hypothesized to act either structurally ( Geyer et al . , 1986 ) or enzymatically ( Wittkopp et al . , 2002 ) in the synthesis of dopamine melanin , and a Yellow homolog has been shown to bind dopamine and other biogenic amines in the sand fly Lutzomyia longipalpis ( Xu et al . , 2011 ) . The interaction between Yellow and dopamine might explain the protein’s effects on male mating success because dopamine acts as a modulator of male courtship drive in D . melanogaster ( Zhang et al . , 2016 ) . These effects of dopamine are mediated by neurons expressing the gene fruitless ( fru ) ( Zhang et al . , 2016 ) , which is a master regulator of sexually dimorphic behavior in D . melanogaster that can affect every component of courtship and copulation ( reviewed in Villella and Hall , 2008 ) . fru has also been shown to regulate expression of yellow in the central nervous system ( CNS ) of male D . melanogaster larvae ( Drapeau et al . , 2003 ) . These observations suggest that the pleiotropic effects of yellow on male mating success might result from effects of yellow in the adult CNS , particularly in fru-expressing neurons . Consistent with this hypothesis , functional links between the pigment synthesis pathway and behavior mediated by the nervous system have previously been reported for other pigmentation genes ( Hotta and Benzer , 1969; Heisenberg , 1971; Borycz et al . , 2002; Richardt et al . , 2002; True et al . , 2005; Suh and Jackson , 2007 ) . D . melanogaster males perform multiple behaviors , including tapping , chasing , singing , and genital licking , before attempting to copulate with females by curling their abdomen and grasping the female ( Figure 1B , Video 1 ) . In one-hour trials , we found that virgin males homozygous for a null allele of the yellow gene ( y1 ) successfully mated with wild-type virgin females only 3% of the time , whereas wild-type males mated with wild-type virgin females 93% of the time ( Fisher’s exact test , p=6×10−13; Figure 1C ) . Videos of mating trials ( e . g . , Videos 1 and 2 ) indicated that the difference in mating success between wild-type and yellow males did not come from differences in the amount of time spent courting ( courtship index , t-test , p=0 . 81; Figure 1D ) or the number of wing extensions during the trial period ( t-test , p=0 . 37; Figure 1E ) . Courtship song analysis also indicated similar amounts of pulse ( t-test , p=0 . 90; Figure 1F ) , sine song ( t-test , p=0 . 07; Figure 1G ) , and interpulse interval ( t-test , p=0 . 07; Figure 1H ) . Watching the courtship videos showed that copulation initiation was most strikingly different between the two genotypes , with copulation initiation reduced in yellow males compared to wild-type ( compare Videos 3 and 4 ) . To determine whether yellow activity in fru-expressing cells was responsible for this difference in mating success , we used the UAS-GAL4 system ( Brand and Perrimon , 1993 ) to drive expression of yellow-RNAi ( Dietzl et al . , 2007 ) with fruGAL4 ( Stockinger et al . , 2005 ) , knocking down native yellow expression in these cells . We also used fruGAL4 to drive yellow expression in y1 mutants . In both cases , when the experimental genotype was compared to the ( 1 ) GAL4 only and ( 2 ) UAS only control genotypes using a Fisher’s exact test ( FET ) with p-values adjusted ( Bonferroni ) for the ( n = 2 ) control comparisons , we found no significant effect on male mating success ( Figure 2A , p=1 for both tests; Figure 2B , p=0 . 07 and 0 . 2 ) , suggesting that expression of yellow in fru-expressing cells is neither necessary nor sufficient for yellow’s effect on male mating success . To continue searching for cells responsible for yellow’s effects on mating , we examined a 209 bp sequence 5’ of the yellow gene called the ‘mating-success regulatory sequence’ ( MRS ) in a prior study that reported it was required for male mating success ( Drapeau et al . , 2006 ) . We hypothesized that the MRS might contain an enhancer driving yellow expression and found that ChIP-seq data indicate the Doublesex ( Dsx ) transcription factor binds to this region in vivo ( Clough et al . , 2014 ) . Like fru , dsx expression is required to specify sex-specific behaviors in D . melanogaster ( Rideout et al . , 2010; Robinett et al . , 2010; reviewed in Villella and Hall , 2008; Yamamoto and Koganezawa , 2013 ) , suggesting that yellow expression regulated by Dsx through the MRS enhancer might be responsible for its effects on male mating behavior . We found that reducing yellow expression in dsx-expressing cells with either of two different dsxGAL4 drivers ( Robinett et al . , 2010; Rideout et al . , 2010 ) strongly reduced male mating success ( Figure 2C , FET , p=7×10−9 and 1 × 10−7; Figure 2—figure supplement 1A , FET , p=0 . 002 and 0 . 002 ) , whereas restoring yellow activity in cells expressing dsxGAL4 in y1 mutants significantly increased male mating success compared with y1 controls ( Figure 2D , FET , p=0 . 001 and 0 . 0004; Figure 2—figure supplement 1B , FET , p=5×10−10 and 5 × 10−10 ) . Video recordings of male flies with reduced yellow expression in dsx-expressing cells showed the same mating defect observed in y1 mutants: males seem to perform all courtship actions normally , but repeatedly failed to copulate ( Video 5 ) . We therefore conclude that yellow expression is required in dsx-expressing cells for normal male mating behavior . To determine whether the MRS sequence might be the enhancer mediating yellow expression in dsx-expressing cells that affects male mating success , we manipulated yellow expression with GAL4 driven by a 2 . 7 kb DNA region located 5’ of yellow that includes the wing , body , and putative MRS enhancers ( Gilbert et al . , 2006 , Figure 2—figure supplement 2A ) . Altering yellow expression with this GAL4 driver modified pigmentation as expected but did not affect male mating success ( Figure 2—figure supplement 2B–D ) , possibly because this GAL4 line did not show any detectable expression in the adult CNS ( Figure 2—figure supplement 2E ) . To test more directly whether the MRS was necessary for male mating success , we deleted 152 bp of the 209 bp MRS sequence using CRISPR/Cas9 gene editing ( Bassett et al . , 2013 ) ( Figure 2—figure supplement 2F , G ) . We found that this deletion had no significant effect on male mating success ( Figure 2—figure supplement 2H , FET , p=0 . 99 compared to wild-type ( CS ) ) , contradicting the previous deletion mapping data ( Drapeau et al . , 2006 ) . We conclude therefore that effects of yellow expression in dsx-expressing cells on mating behavior are likely mediated by other cis-regulatory sequences associated with the yellow gene . Although dsx is expressed broadly throughout the fly ( Robinett et al . , 2010; Rideout et al . , 2010 ) , we hypothesized that its expression in the nervous system would be responsible for yellow’s effects on mating because yellow has been reported to be expressed in the adult brain ( Hinaux et al . , 2018 ) and behavioral effects of other pigmentation genes are mediated by neurons ( Hotta and Benzer , 1969; Heisenberg , 1971; Borycz et al . , 2002; True et al . , 2005 ) . However , we found that suppressing yellow expression in the larval CNS , dopaminergic neurons , or serotonergic neurons ( Figure 2—figure supplement 3 , FET , P values ranging from 0 . 45 to 1 ) , or in all neurons ( Figure 2E , FET , p=1 in all cases ) or all glia ( Figure 2F , FET , p=1 ) , had no significant effect on male mating success . Specifically reducing yellow expression in either all dsx-expressing neurons ( Figure 2G , FET , p=1 and 0 . 45 ) or all dsx-expressing glutamatergic neurons that are required for genital coupling ( Pavlou et al . , 2016 ) ( Figure 2H , FET , p=1 and 0 . 68 ) also had no significant effect on male mating success . In addition , when we examined yellow expression in adult brains , we were only able to observe non-specific signal at the anterior of the adult brain in females ( Figure 2J , Figure 2—figure supplement 4 ) . Given this lack of evidence that yellow is required in neuronal cells for normal male mating behavior , we limited dsxGAL4 activation of yellow expression in y1 mutants to non-neuronal cells and found that these flies exhibited an increase in male mating success compared with y1 mutant males ( Figure 2I , FET , p=0 . 04 and 0 . 0002 ) , showing that yellow expression in non-neuronal dsx-expressing cells is required for normal male mating behavior . To identify which non-neuronal dsx-expressing cells require yellow expression for normal male mating success , we screened ten dsx-enhancer GAL4 lines that each contains a different ~ 3 kb region of dsx noncoding sequence ( Figure 2K; Pfeiffer et al . , 2008 ) . Two of these lines , 42D04-GAL4 and 40F03-GAL4 , significantly decreased male mating success when driving yellow-RNAi ( Figure 2L , FET , p=0 . 001 and 2 × 10−5 ) . These two GAL4 drivers contain overlapping sequences from intron 2 of dsx ( Figure 2K ) , suggesting that their similar effects result from reduction of yellow expression in the same cells . Line 42D04-GAL4 had stronger effects than 40 F03-GAL4 ( Figure 2M , FET , p=0 . 0009 for both controls for 42D04-GAL4 versus p=0 . 97 for both controls for 40 F03-GAL4 ) , so we performed all further analyses with 42D04-GAL4 . Males with yellow reduced by 42D04-GAL4 performed courtship behavior in a pattern similar to y1 mutant males: males performed all precopulatory courtship behaviors normally , but repeatedly failed to copulate , even after hours of attempts ( Video 6 ) . These data indicate that some or all cells in which 42D04-GAL4 drives expression require yellow expression for normal male mating behavior . 42D04-GAL4 drives expression in a sexually dimorphic pattern in multiple neurons of the adult male ( Figure 3A , B ) and female CNS ( Figure 3—figure supplement 1A , B ) , consistent with previously described dsxGAL4 expression in the posterior cluster , the abdominal cluster , and , in males , in the prothoracic TN1 neurons ( Robinett et al . , 2010 ) . 42D04-GAL4 also drives expression in male and female larval CNS and genital discs , with expression in the genital tissues persisting into the adult stage only in females ( Figure 3—figure supplement 1C–G ) . Finally , we observed 42D04-GAL4 expression at the base of the sex combs ( also observed by Robinett et al . , 2010 and Rice et al . , 2019 ) , which are modified bristles used during mating ( Cook , 1975; Ng and Kopp , 2008; Hurtado-Gonzales et al . , 2015 ) that are present only on the first tarsal segment of adult male forelegs ( Figure 3C–F ) . Yellow protein is expressed in sex combs ( Hinaux et al . , 2018 , Figure 3G , H ) , where it is presumably required for synthesis of black dopamine melanin in the sex comb ‘teeth’ . This expression of yellow in sex comb cells is driven by enhancer sequences in the yellow intron ( Figure 3—figure supplement 2 ) , potentially explaining why manipulating yellow expression using GAL4 driven by sequences 5’ of the yellow gene failed to affect mating ( Figure 2—figure supplement 2A–D ) . Driving expression of yellow-RNAi with 42D04-GAL4 eliminated expression of an mCherry tagged version of the native Yellow protein in sex combs and strongly reduced black melanin in the sex combs ( Figure 3I–L ) but not the abdomen ( Figure 3—figure supplement 1J ) . To test the impact of yellow expression in sex combs on male mating behavior , we used 42D04-GAL4 to drive yellow-RNAi , but inhibited the function of 42D04-GAL4 in the CNS with nysb-GAL80 ( courtesy of Julie Simpson ) . These flies showed no GAL4 activity in the CNS ( Figure 3M , N ) , but lost black melanin in the sex combs ( Figure 3O ) and had reduced male mating success ( Figure 3P , FET , p=0 . 002 and 0 . 08 ) . High-speed videos ( 1000 frames per second ) revealed that yellow mutant ( y1 ) males fail repeatedly to grasp the female abdomen with their sex combs when attempting to mount and copulate ( Video 7 ) , whereas wild-type males more readily grasp the female with their melanized sex combs and initiate copulation efficiently ( Video 8 ) . These observations suggest that yellow expression in sex combs affects their melanization , which in turn affects their function . To test whether sex comb melanization ( as opposed to some other unknown effect of losing yellow expression in sex combs ) is critical for male sexual behavior , we suppressed expression of Laccase2 ( Arakane et al . , 2005; Riedel et al . , 2011 ) in sex combs using 42D04-GAL4 and Laccase2-RNAi ( Dietzl et al . , 2007 ) . Laccase2 is required to oxidize dopamine into dopamine quinones and thus acts upstream of Yellow in the melanin synthesis pathway ( Figure 4A; Riedel et al . , 2011 ) . Males with Laccase2 suppressed in sex combs lacked both black and brown dopamine melanin , making these sex combs appear translucent ( Figure 4B ) . These males displayed strongly reduced mating success compared with wild-type males ( Figure 4C , FET , p=1×10−7 and 8 × 10−6 ) and behavioral defects similar to those observed for y1 mutants ( Videos 9 and 10 ) , including inefficient grasping of the female for mounting and copulation . We noticed , however , that flies with Laccase2-RNAi driven by 42D04-GAL4 also showed a loss of melanin in the aedeagus ( Figure 4—figure supplement 1A ) , which is the main part of the male genitalia used for copulation , despite no visible expression of 42D04-GAL4 in the adult male genitalia ( Figure 3—figure supplement 1G ) nor changes in aedeagus pigmentation in y1 mutants ( Figure 4—figure supplement 1A ) . We therefore used subsets of the 42D04 enhancer ( Figure 4—figure supplement 1B ) to drive expression of Laccase2-RNAi , separating the effects of expression in the sex combs from expression in the genitalia ( Figure 4—figure supplement 1C ) . Male mating success was reduced when Laccase2 suppression reduced melanization in the sex combs , but not the genitalia ( Figure 4—figure supplement 1D–G ) . How can sex comb melanization affect sex comb function ? In insects , melanization impacts not only the color of the adult cuticle but also its mechanical stiffness ( Xu et al . , 1997; Kerwin et al . , 1999; Vincent and Wegst , 2004; Andersen , 2005; Arakane et al . , 2005; Suderman et al . , 2006; Riedel et al . , 2011; Noh et al . , 2016 ) . For example , expressing Laccase2-RNAi in D . melanogaster wings softens the cuticle to such a degree that the wings collapse ( Riedel et al . , 2011 ) . Butterflies lacking dopamine melanin due to loss of yellow or another gene required for melanin synthesis , Dopa decarboxylase , also show changes in the fine structure of their wing scales ( Matsuoka and Monteiro , 2018 ) . Consistent with these observations , we observed structural changes in D . melanogaster sex comb teeth lacking yellow or Laccase2 expression using scanning electron microscopy ( SEM ) , with a crack appearing in one of the Laccase2-RNAi comb teeth ( Figure 4D ) . We thus conclude that these structural changes in sex combs are responsible for inhibiting the yellow mutant male’s ability to grasp a female for mounting and copulation ( Video 10 ) . In 1976 , Wilson et al . ( 1976 ) speculated about this very hypothesis based on their own observations of behavior in yellow mutant males . Data from other Drosophila species are also consistent with this structural hypothesis . Specifically , yellow mutants in D . subobscura , D . pseudoobscura , and D . gaucha , all of which have sex combs , show reduced male mating success with wild-type females ( Rendel , 1944; Tan , 1946; Frias and Lamborot , 1970; Pruzan-Hotchkiss et al . , 1992 ) whereas yellow mutants in Drosophila willistoni , a species that lacks sex combs ( Kopp , 2011; Atallah et al . , 2014 ) , do not ( da Silva et al . , 2005 ) . Sex comb morphology is highly diverse among species that have sex combs ( Kopp , 2011 ) , but these structures generally seem to be melanized ( Figure 4—figure supplement 2; Tanaka et al . , 2009 ) and used to grasp females ( Videos 11–15 ) . ( Our high-speed video recordings of mating in D . anannasae , D . bipectinata , D . kikkawai , D . malerkotiana , and D . takahashi show that differences in sex comb morphology ( Figure 4—figure supplement 2 ) correspond with differences in how ( where on the female and with which part of the male leg ) the male grasps the female prior to copulation ( Videos 11–15 ) . It remains unclear how D . willistoni males ( and males of other species without sex combs ) are able to efficiently grasp females prior to copulation ( Video 16 ) . However , differences in females might be part of the answer , as D . melanogaster y1 mutant males are able to mate with y1 mutant females at rates similar to wild-type males ( Bastock , 1956; Dow , 1976; Heisler , 1984; Liu et al . , 2019; Figure 4—figure supplement 3A , FET , p=1 ) . That said , removing all melanin from D . melanogaster sex combs by knocking down Laccase-2 reduced mating efficiency with y1 females ( Figure 4—figure supplement 3B , FET , p=0 . 02 and 0 . 0001 ) , suggesting that the brown melanin remaining in y1 sex-combs ( Figure 4B ) played a role in the mating success of y1 males with y1 females . Taken together , our data show that melanization of a secondary sexual structure affects mating in D . melanogaster . Specifically , we find that the reduced mating success of D . melanogaster yellow mutant males , which was perceived as a behavioral defect for decades , is caused by changes in the morphology of the structures used during mating . Other recent studies have also shown the importance of morphological structures for stickleback schooling ( Greenwood et al . , 2015 ) , water strider walking ( Santos et al . , 2017 ) , and cricket singing ( Pascoal et al . , 2014 ) behaviors . These observations all underscore that behavior cannot be understood by studying the nervous system alone; anatomy and behavior function and evolve as an interconnected system . The following lines were used for this work: y1 [which was backcrossed into a wild-type ( Canton-S ) line for six generations before starting our experiments; the y1 allele contains an A to C transversion in the ATG initiation and is considered a null allele ( Geyer et al . , 1990 ) ]; Canton-S as wild-type ( courtesy of Scott Pletcher ) ; UAS-yellow-RNAi obtained from the Vienna Drosophila Resource Centre ( VDRC ) ( Dietzl et al . , 2007 , KK106068 ) ; y1;UAS-y ( BDSC 3043 ) ; elav-GAL4 ( BDSC 49226 ) ; nsyb-GAL4 ( BDSC 39171 ) ; repo-GAL4 ( BDSC 7415 ) ; dsxGAL4 ( Robinett et al . , 2010 ) ( courtesy of Bruce Baker ) ; dsxGAL4 ( Rideout et al . , 2010 ) ( courtesy of Stephen Goodwin ) ; fruGAL4 ( Stockinger et al . , 2005 ) ( courtesy of Barry Dickson ) ; the following Janelia enhancer trap GAL4 lines ( Pfeiffer et al . , 2008 ) : 40A05-GAL4 ( BDSC 48138 ) , 41D01-GAL4 ( BDSC 50123 ) , 42D02-GAL4 ( BDSC 41250 ) , 41 F06-GAL4 ( BDSC 47584 ) , 41A01-GAL4 ( BDSC 39425 ) , 42D04-GAL4 ( BDSC 47588 ) , 40 F03-GAL4 ( BDSC 47355 ) , 39E06-GAL4 ( BDSC 50051 ) , 42 C06-GAL4 ( BDSC 50150 ) , 40 F04-GAL4 ( BDSC 50094 ) ; ymCherry ( courtesy of Nicolas Gompel ) ; nsyb-GAL80 ( courtesy of Julie Simpson ) ; UAS-Laccase2-RNAi obtained from the VDRC ( Dietzl et al . , 2007 , KK101687 ) ; dsxGAL4-DBD ( Pavlou et al . , 2016 ) ( courtesy of Stephen Goodwin ) ; vGlutdVP16-AD ( Gao et al . , 2008 ) ( courtesy of Stephen Goodwin ) ; BDSC 6993; BDSC 49365; BDSC 6927; BDSC 45175; BDSC 3740; BDSC 5820; BDSC 8848; BDSC 7010; TPH-GAL4 ( courtesy of Shinya Yamamoto ) ; wing-body-GAL4 ( BDSC 44373 ) ; D . melanogaster yellow 5’ up EGFP reporter ( Kalay and Wittkopp , 2010 ) ( courtesy of Gizem Kalay ) ; D . melanogaster yellow intron EGFP reporter ( Kalay and Wittkopp , 2010 ) ( courtesy of Gizem Kalay ) ; vasa-Cas9 ( BDSC 51324 ) ; UAS-cytGFP ( courtesy of Janelia Fly Core ) ; pJFRC12-10XUAS-IVS-myr::GFP ( courtesy of Janelia Fly Core ) . All flies were grown at 23°C with a 12 hr light-dark cycle with lights on at 8AM and off at 8PM on standard corn-meal fly medium . Supplementary file 2 is a Microsoft Excel file containing four worksheets with all of the data used for analysis . The worksheet labeled ‘Univar_Male_Mating_Success_Data’ contains a univariate description of each mating trial . The worksheet labeled ‘Summary of mating success data’ shows the number of successful and unsuccessful matings for each genotype tested ( grouped by figure panel including the data ) and was generated from the ‘Univar_Male_Mating_Success_Data’ worksheet using the Excel Pivot Table function . The worksheet labeled ‘Courtship_Data’ includes the data for courtship index and wing extensions shown in Figure 1D and E , respectively . The worksheet labeled ‘Song Data’ includes the measures of pulses per minute , sin per minute and interpulse interval ( labeled ‘ModeEndToStartIPI’ ) exported from the software described in Arthur et al . ( 2013 ) . R version 3 . 6 . 1 ( 2019-07-05 ) ( R Development Core Team , 2013 ) was used for all statistical analyses using the code included in Source code 1 . These analyses included t-tests comparing courtship index , number of wing extensions , pulses per minute , sine per minute , and interpulse interval that were run after exporting data in the Courtship Data and Song Data worksheets ( separately ) to tab delimited text files . Note that the default t-test parameters allowing for unequal variance between samples were used . Source code 1 also contains the R code for the Fisher’s Exact Tests , which were coded using data from the ‘Summary of mating success data’ worksheet . Supplementary file 3 contains a summary of all statistical tests . Whenever an experimental genotype was compared to two control genotypes , P-values were adjusted using a Bonferroni correction for N = 2 ( see Supplementary file 3 ) . We note that for N = 2 , alternative adjustments available with the p . adjust function in R ( ‘holm’ , ‘Hochberg’ , ‘hommel’ and ‘fdr’ ) give the same adjusted P-value .	More than 100 years ago , Nobel-prize winning geneticist Thomas Hunt Morgan and his colleagues discovered that some fruit flies inherited genetic mutations that caused their body color to change . The yellow flies had a mutation in one specific gene and these mutants did not only look different from normal flies , they behaved differently too . Specifically , yellow males were far less successful at mating than normal males , demonstrating for the first time that some behaviors had a genetic basis . Since then it has remained a mystery how the genetic mutations that cause yellow coloration in fruit flies lead to unsuccessful mating attempts . Geneticists have long suggested that mutations in insect pigment genes cause changes in the fly’s brain because these pigments are made from dopamine , a chemical messenger that acts in the brain . They proposed that yellow flies must have altered levels of dopamine in their brains which was causing them to fail at mating . To solve this mystery , Massey et al . used a series of genetic experiments and high speed-videos to assess how mutations in male yellow fruit flies affected their mating behavior . The experiments showed that yellow fruit flies mated poorly not because of changes in their brain but because of changes in specialized structures on their legs called sex combs . The yellow males lack melanin pigments in their sex combs , which changes their structure . As a result , the yellow males would court female flies but were then unable to grab and mount them . This explains why yellow flies often fail to mate and why fruit flies have sex combs in the first place . The study reveals the importance of scientists considering that genes that affect behavior may do so by changing anatomy rather than by altering the brain . The results also may benefit those working to control insect pests . For example , they could help insect pest managers to develop strategies that prevent reproduction in other insects that spread disease or destroy crops .	[ "Abstract", "Introduction", "Results", "and", "discussion", "Materials", "and", "methods" ]	[ "evolutionary", "biology", "genetics", "and", "genomics" ]	2019	The yellow gene influences Drosophila male mating success through sex comb melanization
The DUX4 transcription factor is encoded by a retrogene embedded in each unit of the D4Z4 macrosatellite repeat . DUX4 is normally expressed in the cleavage-stage embryo , whereas chromatin repression prevents DUX4 expression in most somatic tissues . Failure of this repression causes facioscapulohumeral muscular dystrophy ( FSHD ) due to mis-expression of DUX4 in skeletal muscle . In this study , we used CRISPR/Cas9 engineered chromatin immunoprecipitation ( enChIP ) locus-specific proteomics to characterize D4Z4-associated proteins . These and other approaches identified the Nucleosome Remodeling Deacetylase ( NuRD ) and Chromatin Assembly Factor 1 ( CAF-1 ) complexes as necessary for DUX4 repression in human skeletal muscle cells and induced pluripotent stem ( iPS ) cells . Furthermore , DUX4-induced expression of MBD3L proteins partly relieved this repression in FSHD muscle cells . Together , these findings identify NuRD and CAF-1 as mediators of DUX4 chromatin repression and suggest a mechanism for the amplification of DUX4 expression in FSHD muscle cells . Repetitive DNA sequences make up the majority of the human genome ( Birney et al . , 2007; de Koning et al . , 2011 ) , and these ubiquitous but understudied elements play a critical role in important biological processes such as embryogenesis and cellular reprogramming ( Chuong et al . , 2017; Elbarbary et al . , 2016; Gerdes et al . , 2016 ) . For example , each unit of the D4Z4 macrosatellite repeat array contains a copy of the double homeobox 4 ( DUX4 ) retrogene that is expressed in the germline and in four-cell human embryos where DUX4 activates a cleavage-specific transcriptional program ( De Iaco et al . , 2017; Hendrickson et al . , 2017; Snider et al . , 2010; Whiddon et al . , 2017 ) . This is in contrast to somatic tissues where DUX4 is silenced via repeat-mediated epigenetic repression of the D4Z4 arrays ( Das and Chadwick , 2016; Daxinger et al . , 2015; Snider et al . , 2010; van Overveld et al . , 2003; Zeng et al . , 2009 ) . To date , little is understood about how the epigenetic repression of DUX4 is relieved at specific times during germline and early embryo development , or what the mechanisms of establishing and maintaining epigenetic repression during later development and in somatic tissues are . Facioscapulohumeral muscular dystrophy ( FSHD ) is caused by the mis-expression of DUX4 in skeletal muscle ( Tawil et al . , 2014 ) and provides an experimentally tractable context in which to identify mechanisms that normally repress DUX4 in somatic cells as well as mechanisms that might regulate this repression during development . In individuals with FSHD , the epigenetic repression of DUX4 is incomplete as a consequence of having fewer than 11 D4Z4 repeats ( FSHD type 1 , FSHD1 ) or mutations in trans-acting chromatin repressors of D4Z4 ( FSHD type 2 , FSHD2 ) , either of which results in ectopic expression of DUX4 in skeletal muscle when combined with a permissive chromosome 4qA haplotype that provides a polyadenylation site for the DUX4 mRNA ( Lemmers et al . , 2012; Lemmers et al . , 2010; van den Boogaard et al . , 2016 ) . The mis-expression of DUX4 in skeletal muscle has many consequences that include induction of a cleavage-stage transcriptional program , suppression of the innate immune response and nonsense-mediated RNA decay ( NMD ) pathways , inhibition of myogenesis , and induction of cell death through mechanisms that involve the accumulation of aberrant and double-stranded RNAs ( Bosnakovski et al . , 2008; Feng et al . , 2015; Geng et al . , 2012; Kowaljow et al . , 2007; Rickard et al . , 2015; Shadle et al . , 2017; Snider et al . , 2009; Wallace et al . , 2011; Winokur et al . , 2003; Young et al . , 2013 ) . These cellular insults lead to progressive muscle weakness initiating in the face and upper body but eventually involving nearly all skeletal muscle groups ( Tawil et al . , 2014 ) . Previous studies investigating D4Z4 repeat-mediated epigenetic repression have shown the D4Z4 arrays to be silenced through multiple mechanisms , including DNA methylation and the repressive histone modifications di/trimethylation of histone H3 at lysine 9 ( H3K9me2/3 ) and trimethylation of histone H3 at lysine 27 ( H3K27me3 ) along with their binding proteins CBX3/HP1γ and EZH2 ( Cabianca et al . , 2012; Huichalaf et al . , 2014; van den Boogaard et al . , 2016; van Overveld et al . , 2003; Zeng et al . , 2009 ) . Other repressor proteins have been shown to be associated with the D4Z4 repeat , including DNMT3B , SMCHD1 , cohesin , CTCF , HDAC3 and a YY1/HMGB2/NCL complex ( Gabellini et al . , 2002; Huichalaf et al . , 2014; Lemmers et al . , 2012; Ottaviani et al . , 2009; van den Boogaard et al . , 2016; Zeng et al . , 2009 ) . In addition , DICER/AGO-dependent siRNA-directed silencing has also been demonstrated to play a role in repressing the D4Z4 array ( Lim et al . , 2015; Snider et al . , 2009 ) . The genetic lesions that cause FSHD disrupt these regulatory pathways resulting in D4Z4 DNA hypomethylation; reduced H3K9me2/3 and H3K27me3 levels; and loss of HP1γ , EZH2 , SMCHD1 and cohesin binding; which together culminate in ectopic DUX4 expression ( Cabianca et al . , 2012; Daxinger et al . , 2015; Jones et al . , 2014; Lemmers et al . , 2012; van den Boogaard et al . , 2016; van Overveld et al . , 2003; Zeng et al . , 2009 ) . Although each of the above-mentioned studies tested specific factors based on knowledge of their role in chromatin , to date no studies have taken an agnostic approach to identify how these individual components might be integrated into repressive complexes or to understand how these complexes might be regulated . Here , we report a locus-specific proteomics-based characterization of proteins that bind the D4Z4 array in human myoblasts and identify the NuRD and CAF-1 complexes as individually necessary to maintain DUX4 repression in skeletal muscle and induced pluripotent stem ( iPS ) cells . Further , we show that DUX4-mediated induction of the MBD3L family of factors relieves this repression and amplifies DUX4 expression . Together , these findings identify multiprotein complexes that regulate DUX4 expression and reveal a process for DUX4 amplification in FSHD muscle cells that provides a new candidate target for therapeutics . To identify regulators of the D4Z4 macrosatellite repeat , we carried out engineered DNA-binding molecule-mediated chromatin immunoprecipitation ( enChIP ) followed by mass spectrometry ( MS ) ( enChIP-MS ) ( Fujita and Fujii , 2013 ) ( Figure 1A ) . We transduced human MB135 control ( non-FSHD ) myoblasts with a lentiviral vector co-expressing FLAG-tagged , nuclease-deficient Cas9 ( FLAG-dCas9 ) and guide RNA ( gRNA ) targeting the 3’ end ( gD4Z4-1 ) , middle ( gD4Z4-2 ) or 5’ end ( gD4Z4-3 ) of the D4Z4 unit , or the MYOD1 distal regulatory region ( DRR ) ( gMYOD1 ) for comparison . After confirming the expression , subcellular localization , and specific chromatin occupancy of FLAG-dCas9 in each cell line ( Figure 1—figure supplement 1 ) , complexes containing FLAG-dCas9 were immunoprecipitated and subjected to liquid chromatography-tandem mass spectrometry for protein identification . A total of 261 proteins were identified ( Supplementary file 1 ) , including known D4Z4-associated factors SMCHD1 , CBX3/HP1γ and the cohesin complex components SMC1A , SMC3 , RAD21 and PDS5B ( Lemmers et al . , 2012; Zeng et al . , 2009 ) ( Table 1 ) . BRD3 and BRD4 were also identified ( Supplementary file 1 ) and BET inhibitor compounds have recently been shown to regulate D4Z4 repression ( Campbell et al . , 2017 ) . D4Z4-bound proteins were enriched in gene ontology categories that included telomere maintenance and chromatin silencing ( Supplementary file 2 ) , consistent with the subtelomeric localization and transcriptionally repressed state of the D4Z4 array . Strikingly , CHD4 , HDAC2 , MTA2 and RBBP4 , which comprise many of the components of the Nucleosome Remodeling Deacetylase ( NuRD ) complex ( Basta and Rauchman , 2015 ) , were among the isolated proteins ( Table 1 ) . While each of these factors was identified as associated with the D4Z4 repeat in more than one gD4Z4 sample , they were either absent or present in only a single replicate from the gMYOD1 pulldowns ( Supplementary file 1 ) . Occupancy of CHD4 , HDAC2 and MTA2 at the D4Z4 array was confirmed by chromatin immunoprecipitation ( ChIP ) in MB2401 myoblasts , an independent control muscle cell line ( Figure 1B–D ) . The NuRD complex can be recruited to methylated DNA by the MBD2 subunit ( Le Guezennec et al . , 2006; Zhang et al . , 1999 ) , and indeed , ChIP showed MBD2 enrichment at the D4Z4 region in MB2401 control myoblasts ( Figure 1E ) . Together , these data demonstrate that the D4Z4 macrosatellite repeat is bound by the MBD2/NuRD complex in control human muscle cells . The NuRD complex represses gene transcription via the concerted effort of the core subunits HDAC1 and HDAC2; CHD3 or CHD4; MBD2 or MBD3; MTA1 , MTA2 or MTA3; RBBP4 and RBBP7; and GATAD2A and GATAD2B ( Basta and Rauchman , 2015 ) ( Figure 2A ) . In MB2401 control myoblasts , small interfering RNA ( siRNA ) depletion of the lysine deacetylases HDAC1 or HDAC2 had no significant effect on DUX4 mRNA levels , whereas concurrent HDAC1/HDAC2 knockdown increased DUX4 mRNA 100-fold resulting in the activation of DUX4 target genes ZSCAN4 and TRIM43 ( Figure 2B and Figure 2—figure supplement 1A ) . In contrast , in MB073 FSHD1 and MB200 FSHD2 myoblasts , singular HDAC1 or HDAC2 depletion led to a ≥ 20-fold activation of DUX4 mRNA while dual HDAC1/HDAC2 knockdown increased DUX4 levels more than 140-fold , with comparable changes to DUX4 targets ( Figure 2C–D and Figure 2—figure supplement 1B–C ) . Pharmacological inhibition of HDAC1/HDAC2 activity by MS-275 ( Nebbioso et al . , 2009 ) also increased DUX4 and DUX4 target gene expression , and enhanced histone H4 acetylation at the D4Z4 array ( Figure 2—figure supplement 2 ) . Collectively , these results indicate that HDAC1 and HDAC2 are associated with , and function to transcriptionally repress , the D4Z4 array . These data also show that the D4Z4 repeat in control myoblasts is more resistant to de-repression than the D4Z4 repeat in FSHD cells , which are sensitized because of a shortened array ( FSHD1 ) or SMCHD1 mutation ( FSHD2 ) . We next evaluated the necessity of the ATP-dependent chromatin remodelers CHD3 and CHD4 for D4Z4 repeat repression . Depleting CHD4 from MB2401 control myoblasts had no effect on DUX4 expression ( Figure 2E and Figure 2—figure supplement 3A ) . However , CHD4 knockdown in MB073 FSHD1 or MB200 FSHD2 myoblasts increased DUX4 mRNA 20-fold and concomitantly activated DUX4 targets ( Figure 2F–G and Figure 2—figure supplement 3B–C ) . In contrast , CHD3 depletion did not lead to DUX4 de-repression in either control or FSHD cells ( Figure 2—figure supplement 4 ) , consistent with its absence from the gD4Z4 enChIP purifications and the mutually exclusive nature of CHD3 and CHD4 within the NuRD complex . Together , these results reveal that CHD4 binds the D4Z4 repeat and is necessary to silence DUX4 expression in FSHD cells , whereas control myoblasts have a more stably repressed D4Z4 array . Similar to CHD4 , depleting methyl-CpG-binding protein MBD2 from MB2401 control myoblasts had no effect on DUX4 mRNA levels ( Figure 2H and Figure 2—figure supplement 5A ) . However , depleting MBD2 from MB073 FSHD1 myoblasts moderately , but significantly , increased DUX4 expression , whereas DUX4 was not de-repressed when MBD2 was knocked down in MB200 FSHD2 myoblasts ( Figure 2I–J and Figure 2—figure supplement 5B–C ) . This difference suggests a possible D4Z4 context-dependent effect that was not observed for the single-copy NuRD complex-bound gene TMEM130 following MBD2 knockdown ( Figure 2—figure supplement 5 ) . We further observed that depletion of MBD3 , which can recruit the NuRD complex to unmethylated DNA ( Baubec et al . , 2013; Le Guezennec et al . , 2006; Saito and Ishikawa , 2002 ) , did not de-repress DUX4 in MB2401 control , MB073 FSHD1 or MB200 FSHD2 myoblasts ( Figure 2—figure supplement 6 ) . Together , these data show that MBD2 occupies the D4Z4 array and is necessary for DUX4 repression in at least some contexts , and suggest that factors in addition to MBD2 might recruit components shared by the NuRD complex to silence the D4Z4 macrosatellite repeat . The NuRD complex is known to cooperate with other complexes to carry out its cellular functions . For example , NuRD and the CAF-1 chromatin assembly complex work together in several molecular processes ( Helbling Chadwick et al . , 2009; Yang et al . , 2015 ) and share a core subunit , RBBP4 , which was identified as associated with the D4Z4 repeat by gD4Z4 enChIP purification ( Table 1 ) . CHAF1A and CHAF1B comprise the other core members of the CAF-1 complex ( Volk and Crispino , 2015 ) ( Figure 3A ) . Depleting CHAF1A or CHAF1B resulted in the activation of DUX4 and DUX4 target genes in FSHD myoblasts ( Figure 3B–D and Figure 3—figure supplement 1 ) , confirming a role for this complex in D4Z4 regulation . CAF-1 interacts with CpG-binding protein MBD1 , which binds both methylated and unmethylated DNA to inhibit transcription ( Jørgensen et al . , 2004; Reese et al . , 2003 ) . Knockdown of MBD1 led to DUX4 and DUX4 target gene activation in MB200 FSHD2 myoblasts but not in MB2401 control or MB073 FSHD1 myoblasts ( Figure 3E–G and Figure 3—figure supplement 2 ) , indicating a possible context-dependent relative necessity of MBD1 or MBD2 in different FSHD cells . Notably , although knockdown of CHAF1A or CHD4 alone did not induce DUX4 expression in MB2401 control myoblasts ( Figure 2E and Figure 3B ) , simultaneous depletion increased DUX4 mRNA levels over 150-fold ( Figure 3H and Figure 3—figure supplement 3A ) . An additive or greater impact was also observed with dual versus singular CHD4 and CHAF1A knockdown in MB073 FSHD1 and MB200 FSHD2 myoblasts ( Figure 3I–J and Figure 3—figure supplement 3B–C ) . Together , these results indicate that a combination of MBD1- and MBD2-mediated recruitment of the CAF-1 and NuRD repressive complexes , respectively , work together to silence the D4Z4 repeat in skeletal muscle cells . To extend these studies , we depleted CHD4 , CHAF1A , MBD2 , or MBD1 in five additional FSHD cell lines: one FSHD1 cell line ( 54-2 ) with three 4qA D4Z4 repeats ( compared to the 8 repeats of the MB073 line ) , and four FSHD2 lines ( 2305 , 2453 , 2338 , and 1881 ) with different SMCHD1 mutations and repeat sizes ranging from 11 to 15 D4Z4 units ( Supplementary file 3 ) . All five lines showed de-repression of DUX4 upon knockdown of MBD2 or CHAF1A , and all but one ( 2453 , an FSHD2 cell line ) showed increased DUX4 expression following CHD4 depletion , whereas de-repression following MBD1 knockdown was evident in the FSHD1 and two of the FSHD2 cell lines ( Figure 3—figure supplements 4–7 ) . Taken together , these data indicate the combined roles of the NuRD and CAF-1 complexes in repressing DUX4 , and that the relative necessity of specific components of each pathway might vary depending on the cellular context , or possibly the efficiency of each knockdown . To repress transcription , core members of the NuRD and CAF-1 complexes utilize a shared set of auxiliary factors , namely the tripartite motif-containing protein TRIM28 , the lysine methyltransferase SETDB1 , and the lysine demethylase KDM1A ( Ivanov et al . , 2007; Loyola et al . , 2009; Sarraf and Stancheva , 2004; Schultz et al . , 2001; Wang et al . , 2009; Yang et al . , 2015 ) . Knockdown of TRIM28 , SETDB1 or KDM1A de-repressed DUX4 in MB073 FSHD1 and MB200 FSHD2 myoblasts to varying degrees ranging from 3- to 130-fold ( Figure 4 and Figure 4—figure supplements 1–3 ) , implicating them in facilitating silencing of the D4Z4 array . Of these factors , only KDM1A knockdown de-repressed DUX4 mRNA in the MB2401 control myoblasts , indicating a necessary role for this demethylase in maintaining repression of both normal and pathological D4Z4 alleles in muscle cells . In support of these expression data , peptides for TRIM28 were present in gD4Z4 enChIP pulldowns , although they did not meet our filtering criteria to be included in the list of D4Z4-associated proteins . Similarly , SIN3A peptides were found in a gD4Z4 pulldown before our final filtering steps . The transcriptionally repressive SIN3 complex shares core proteins HDAC1 , HDAC2 , RBBP4 , and RBBP7 with the NuRD complex and is also composed of SDS3 , SAP18 , SAP30 and SIN3A or SIN3B subunits ( Grzenda et al . , 2009 ) ( Figure 4—figure supplement 4A ) . Therefore , we tested its role in D4Z4 repeat repression and found that SIN3A or SIN3B depletion led to the activation of DUX4 and DUX4 target genes in FSHD cells ( Figure 4—figure supplement 4B–G ) , supporting a role for the SIN3 complex in D4Z4 regulation . Taken together , these data indicate that D4Z4 array silencing is mediated by multiple chromatin regulatory factors that act together with core components of the NuRD complex and also depend on the CAF-1 chromatin assembly complex to achieve full epigenetic repression . We previously reported that DUX4 is expressed at very low levels in human iPS cell populations ( Snider et al . , 2010 ) and , similar to the expression pattern in FSHD myoblasts , this represents the occasional expression in a small number of cells ( JWL , unpublished data ) . We have more recently shown that DUX4 is present in four-cell human embryos and that when expressed in iPS cells or muscle cells it activates a cleavage-stage transcriptional program similar to the program expressed in a subset of ‘naïve’ iPS or embryonic stem ( ES ) cells ( Hendrickson et al . , 2017; Whiddon et al . , 2017 ) . To determine whether factors responsible for silencing the D4Z4 repeat in myoblasts have a similar function in a model of early development , we knocked down components of the NuRD and CAF-1 complexes in human eMHF2 iPS cells , which were derived from an unaffected ( non-FSHD ) individual , and assessed the impact on DUX4 expression . Similar to our myoblast results , depletion of HDAC1/HDAC2 , CHD4 , CHAF1A , SETDB1 or SIN3B de-repressed DUX4 in iPS cells; whereas , unlike in myoblasts , knockdown of KDM1A in iPS cells had a more minor effect on the levels of DUX4 mRNA ( Figure 5 and Figure 5—figure supplement 1 ) . To determine whether iPS cells have a greater necessity for NuRD and CAF-1 components to maintain DUX4 repression compared to somatic cells , we transduced a human foreskin fibroblast cell line ( HFF3 ) with the reprogramming factors Oct4 , Sox2 , Nanog , and Lin28 to generate isogenic iPS cell clones ( Figure 5—figure supplement 2 ) . Notably , depletion of NuRD and CAF-1 complex components did not lead to DUX4 de-repression in the parental HFF3 fibroblast line , whereas the HFF3 iPS lines responded similarly to the eMHF2 iPS line ( Figure 5—figure supplement 3 ) . These results indicate that the NuRD and CAF-1 complexes that silence the D4Z4 macrosatellite array in muscle cells also contribute to the regulation of this locus in human iPS cells , and that iPS cells have decreased D4Z4 repression compared to their somatic counterpart , similar to the decreased repression in FSHD myoblasts compared to control myoblasts . In prior studies of DUX4-induced gene expression , we identified the MBD3L family ( MBD3L2 , MBD3L3 , MBD3L4 , and MBD3L5 ) as a direct target of DUX4 that was expressed in FSHD , but not control , muscle cells and muscle biopsies , and activated by exogenous DUX4 in cultured human myoblasts ( Geng et al . , 2012; Yao et al . , 2014; Young et al . , 2013 ) . MBD3L family proteins can replace MBD2 or MBD3 in the NuRD complex but they lack the CpG-binding domain and antagonize NuRD-mediated transcriptional repression , possibly by preventing the complex from being recruited to its DNA targets ( Jiang et al . , 2002; Jin et al . , 2005 ) . To determine whether MBD3L proteins de-repress the NuRD complex-regulated D4Z4 array , we transduced control and FSHD myoblasts with a lentiviral vector delivering a doxycycline-inducible MBD3L2 transgene and , after selecting for transgene-expressing cells , analyzed DUX4 mRNA and protein after 48 hr of doxycycline treatment . Similar to the knockdown of NuRD complex members , expression of MBD3L2 induced DUX4 5–18-fold in MB073 FSHD1 and MB200 FSHD2 myoblasts and increased by 10-fold the number of myoblast nuclei expressing DUX4 protein , whereas DUX4 was not de-repressed in MB2401 control myoblasts ( Figure 6A–E and Figure 6—figure supplement 1A–C ) . When cultured in low mitogen differentiation media , myoblasts fuse to form multinucleated myotubes , and DUX4 expression increases in FSHD myotubes compared to myoblasts ( Balog et al . , 2015 ) . To determine whether the DUX4-induced MBD3L proteins might contribute to the increased DUX4 expression in myotubes , we expressed short hairpin RNA ( shRNA ) to inhibit MBD3L RNAs in MB073 FSHD1 and MB200 FSHD2 myotubes and found that these decreased DUX4 and DUX4 target gene expression by ~50% and~30% , respectively ( Figure 6F–G , Figure 6—figure supplement 1D–E and Figure 6—figure supplement 2 ) . Together , these data implicate MBD3L2 in the regulation of the D4Z4 array and demonstrate that endogenous DUX4-induced MBD3L proteins contribute to the amplification of DUX4 expression in FSHD myotubes . In this study , enChIP-MS identified factors that co-purified with the D4Z4 macrosatellite array in human myoblasts , and subsequent ChIP and knockdown studies revealed that the NuRD and CAF-1 complexes repress DUX4 expression from the D4Z4 repeat in skeletal muscle and iPS cells . To some extent , each complex appears to have a parallel , or redundant , function in DUX4 repression because knockdown of both pathways was necessary to induce DUX4 expression in MB2401 control myoblasts . The distinctive mutations causing FSHD , or other factors such as the distribution of DNA methylation on the D4Z4 , might preferentially weaken different specific components of each pathway , as evidenced by the relative necessity for CHD4 , MBD2 or MBD1 in different FSHD cell lines . However , the variable efficiencies of the individual knockdowns in each cell type and experiment might also contribute to these apparent differences . It is also important to note that CAF-1 is a chromatin assembly complex and that the knockdowns were performed in replicating myoblasts; therefore , CAF-1 knockdown might not have the same consequence in post-mitotic myotubes . Overall , despite the relative differences in the necessity of the specific protein knockdown of individual components of the NuRD or CAF-1 complexes in different FSHD cells , the data show that these complexes together are necessary to maintain D4Z4 repression . These two complexes also have shared auxiliary components , for example , TRIM28 , SETDB1 , and KDM1A , and knockdown of these factors also induced DUX4 expression in FSHD cells , with KDM1A knockdown being sufficient on its own to induce DUX4 in control myoblasts . Our en-ChIP pulldowns identified several D4Z4-associated proteins that are involved in epigenetic silencing of variegated gene expression in mice ( Blewitt et al . , 2005; Daxinger et al . , 2013 ) . One of this group of Modifier of murine metastable epiallele ( Momme ) genes , Smchd1 , was shown to directly repress DUX4 in human cells and to be a causative gene for FSHD2 ( Lemmers et al . , 2012 ) . In addition to finding SMCHD1 associated with the D4Z4 in our enChIP , we identified the Momme genes PBRM1 , RIF1 , SMARCA4 , SMARCA5 and UHRF1 as D4Z4-associated by enChIP-MS , and implicated the Momme genes HDAC1 , SETDB1 and TRIM28 in the regulation of DUX4 through knockdown experiments . The convergence and striking overlap of the results of these two complementary approaches to understanding variegated gene expression suggest that conserved machinery may be responsible for repressing this type of locus across species . The presence of chromatin remodelers and positive transcriptional regulators , such as SMARCA5 , BRD3 and BRD4 , at the D4Z4 locus in the control cells used for the enChIP also indicates a dynamic balance between activators and repressors , which is consistent with the identification of sense and anti-sense transcripts associated with the D4Z4 repeats in both control and FSHD cells ( Snider et al . , 2010 ) . Our findings also suggest that PBRM1 , RIF1 , SMARCA4 , SMARCA5 and UHRF1 are candidates for playing a role in DUX4 regulation and deserve additional attention in future studies . The necessity of the NuRD complex to maintain repression of DUX4 in FSHD cells suggests that the DUX4-mediated induction of the MBD3L family of factors might amplify DUX4 expression within a nucleus or facilitate the internuclear spreading of DUX4 in multinucleated myotubes . MBD3L factors replace MBD2 or MBD3 in the NuRD complex and antagonize its normal repressive function ( Jiang et al . , 2002; Jin et al . , 2005 ) . In this study , we showed that expression of MBD3L2 was sufficient to amplify DUX4 expression in FSHD cells and knockdown using an shRNA that targets the entire family showed that expression of the MBD3L family was necessary for the full induction of DUX4 expression in FSHD myotubes . The fact that DUX4 induces high expression of the clustered MBD3L genes reveals a positive feed-forward mechanism that might facilitate spreading of DUX4 expression between nuclei in myotubes . In FSHD myotubes , DUX4 expression apparently initiates in a single nucleus and the protein then spreads to adjacent nuclei in the syncytium . Similarly , MBD3L proteins are detected as spreading to adjacent nuclei ( AEC , unpublished data ) where they would facilitate DUX4 expression . In this manner , each DUX4 expressing nucleus would act to progressively amplify DUX4 expression in its neighbors , spreading DUX4 expression along the myofiber . This might be similar to , and additive with , the prior observation that the DUX4-mediated inhibition of NMD can amplify DUX4 expression by stabilizing the DUX4 mRNA , which is itself a target of NMD ( Feng et al . , 2015 ) . It is interesting to speculate that the internuclear amplification of DUX4 expression might contribute to the susceptibility of skeletal muscle to damage in FSHD . Together our data provide several complementary approaches to the challenge of creating an FSHD therapeutic . One strategy would be to enhance D4Z4 repression by designing drugs that increase NuRD complex-mediated repression . Although drugs that decrease epigenetic repression are in clinical use , including some that target members of the NuRD complex ( SAHA , targeting HDAC1/2; ORY-1001 , targeting KDM1A; GSK126 , EZH2 inhibitor ) drugs that enhance epigenetic repression have received less attention . This is partly due to concerns that they might also suppress important tumor suppressor genes , but the fact that mutations in SMCHD1 and DNMT3B that cause FSHD have limited genome-wide consequences suggests that some factors might be relatively specific for repressing repetitive regions of the genome . A second strategy would be to prevent the amplification of DUX4 after it stochastically ‘bursts’ on in a myotube nucleus . This might be accomplished by inhibiting the production of MBD3L proteins with small molecules or interfering RNAs . Alternatively , myoblast transplantation with cells containing larger D4Z4 repeat sizes or 4qB alleles might provide ‘decoy’ nuclei that would absorb MBD3L factors and not activate DUX4 , or , in a similar decoy approach , autologous transplants following deletion of the D4Z4 array and/or the MBD3L cluster . Although little is known about the regulation of DUX4 expression in cleavage-stage embryos and the testis luminal cells , it is evident from this study that the expression of DUX4 in a small percentage of iPS cells or ES cells shares mechanisms of molecular regulation with skeletal muscle cells . This also indicates similarities between the regulation of the human DUX4 retrogene and the mouse Dux retrogene that is also in a macrosatellite array , although thought to have arisen from a separate retrotransposition of the DUXC gene ( Clapp et al . , 2007; Leidenroth et al . , 2012; Leidenroth and Hewitt , 2010 ) . It was previously shown that CAF-1 depletion in mouse ES cells resulted in the expression of genes specific to two-cell embryos ( Ishiuchi et al . , 2015 ) , and later shown that induction of these genes was blocked by simultaneous knockdown of mouse Dux along with Chaf1a ( Hendrickson et al . , 2017 ) . Trim28 , Kdm1a , and HDAC inhibitors have been shown to regulate Zscan4 and the early cleavage program in mouse ES cells ( Macfarlan et al . , 2012 ) , and for Trim28 this activity was shown to be mediated through the induction of mouse Dux ( De Iaco et al . , 2017 ) . Similarly , the NuRD complex and MBD3 have been shown to inhibit cellular reprogramming in mouse ES cells , and , conversely , reprogramming to a naive stem cell state was facilitated by inhibition of these complexes ( Luo et al . , 2013; Rais et al . , 2013 ) . The fact that inhibiting NuRD or CAF-1 activity potentiates stem cell reprogramming in mouse ES/iPS cells and , as shown in this report , potentiates human DUX4 expression , suggests that DUX4 itself might facilitate reprogramming to the naive state and that mouse Dux and human DUX4 might be subject to similar regulation , a finding not entirely obvious given that these retrogenes are thought to have been generated by independent retrotranspositions of the parental DUXC gene , as noted above . In summary , we identified components of the NuRD and CAF-1 complexes as necessary to maintain repression of DUX4 expression from the D4Z4 repeat . In control myoblasts , either pathway was sufficient to maintain repression of DUX4 , whereas in FSHD cells inhibition of either pathway resulted in higher levels of DUX4 expression . These same mechanisms repress DUX4 expression in iPS cells . In addition , the DUX4 induction of the NuRD antagonist MBD3L family further de-repressed DUX4 in FSHD cells . Together , these findings provide the basis for therapies directed at repressing DUX4 in FSHD and reveal a mechanism for the regulation of DUX4 in stem cells . All reagents were obtained from Sigma-Aldrich ( St . Louis , MO ) unless otherwise specified . Human primary myoblast cell lines originated from the Fields Center for FSHD and Neuromuscular Research at the University of Rochester Medical Center ( https://www . urmc . rochester . edu/neurology/fields-center . aspx ) and were immortalized by retroviral transduction of CDK4 and hTERT ( Stadler et al . , 2011 ) . Myoblasts were maintained in Ham’s F-10 Nutrient Mix ( Gibco , Waltham , MA ) supplemented with 20% HyClone Fetal Bovine Serum ( GE Healthcare Life Sciences , Pittsburgh , PA ) , 100 U/100 μg penicillin/streptomycin ( Gibco ) , 10 ng/ml recombinant human basic fibroblast growth factor ( Promega Corporation , Madison , WI ) and 1 μM dexamethasone . Differentiation of myoblasts into myotubes was achieved by switching the fully confluent myoblast monolayer into Dulbecco’s Modified Eagle Medium ( DMEM , Gibco ) containing 1% horse serum ( Gibco ) , 100 U/100 μg penicillin/streptomycin , 10 μg/ml insulin and 10 μg/ml transferrin for 48–72 hr . Myoblasts harboring a transgene were additionally cultured in 2 μg/ml puromycin and transgene expression induced with 1 μg/ml doxycycline hyclate when required . Myoblast cell line identity was authenticated by monitoring fusion into myotubes , DUX4 expression , and the presence of a 4qA161 allele . Detailed characteristics of the myoblast lines used in this study are provided in Supplementary file 3 . Human control ( non-FSHD ) iPS cells were obtained from the University of Washington Institute for Stem Cell and Regenerative Medicine Tom and Sue Ellison Stem Cell Core ( eMHF2 ) ( Hendrickson et al . , 2017 ) or derived in-house from normal HFF3 foreskin fibroblasts reprogrammed via lentiviral transduction of Oct4 , Sox2 , Nanog and Lin28 ( Yu et al . , 2007 ) , and grown in DMEM:Nutrient Mixture F-12 ( 1:1 , Gibco ) with 100 U/100 μg penicillin/streptomycin , 10 mM MEM Non-Essential Amino Acids ( Gibco ) , 100 mM sodium pyruvate ( Thermo Fisher Scientific , Waltham , MA , USA ) , 20% KnockOut Serum Replacement ( Gibco ) , 1 mM 2-mercaptoethanol and 4 ng/ml recombinant human basic fibroblast growth factor under hypoxic ( 5% O2 ) conditions on 0 . 1% gelatin-coated plates pre-seeded with 1 . 3 × 104 cells/cm2 of irradiated mouse embryonic fibroblasts . While the full haplotypes are unknown , eMHF2 cells utilize DUX4 exon 3 , suggesting a 4qA161S allele , while HFF3 cells use DUX4 exon 3b , suggesting a 4qA161L allele ( Lemmers et al . , 2018 ) . HFF3 fibroblasts and 293T cells were maintained in DMEM supplemented with 10% HyClone Fetal Bovine Serum and 100 U/100 μg penicillin/streptomycin . Cell lines are tested periodically for Mycoplasma contamination by the Fred Hutchinson Cancer Research Center Specimen Processing/Research Cell Bank and have not shown evidence of Mycoplasma . To construct FLAG-dCas9-gRNA plasmids , the lentiCRISPRv2 vector ( a gift from Feng Zhang , Addgene plasmid #52961 ) ( Sanjana et al . , 2014 ) was digested with AgeI and BamHI , PCR was used to amplify AgeI-3xFLAG-EcoRI from a synthesized template and EcoRI-dCas9-BamHI from pHR-SFFV-KRAB-dCas9-P2A-mCherry ( a gift from Jonathan Weissman , Addgene plasmid #60954 ) ( Gilbert et al . , 2014 ) , the three fragments were ligated together to create a 3xFLAG-dCas9-HA-2xNLS vector , and then D4Z4 or MYOD1 gRNA were inserted by digesting 3xFLAG-dCas9-HA-2xNLS with BsmBI and ligating it to annealed gRNA oligos . To construct the doxycycline-inducible MBD3L2 plasmid , the MBD3L2 coding region was subcloned into the NheI and SalI sites of the pCW57 . 1 vector ( a gift from David Root , Addgene plasmid #41393 ) . The pGIPZ-shControl and -shMBD3L vectors were obtained from the Fred Hutchinson Cancer Research Center Genomics Shared Resource . Lentiviral particles were produced in 293T cells by co-transfecting the appropriate lentiviral vector with pMD2 . G ( a gift from Didier Trono , Addgene plasmid #12259 ) and psPAX2 ( a gift from Didier Trono , Addgene plasmid #12260 ) using Lipofectamine 2000 ( Invitrogen , Carlsbad , CA ) following the manufacturer’s instructions . To generate polyclonal transgenic cell lines , myoblasts were transduced with lentivirus in the presence of 8 μg/ml polybrene and selected using 2 μg/ml puromycin . Monoclonal transgenic lines were generated by transducing at a low cell density using a low multiplicity of infection ( MOI <1 ) and allowing cells that survived selection to form colonies before individual clones were isolated using cloning cylinders . Total protein extracts were generated by lysing cells in SDS sample buffer ( 500 mM Tris-HCl pH 6 . 8 , 8% SDS , 20% 2-mercaptoethanol , 0 . 004% bromophenol blue , 30% glycerol ) followed by sonication and boiling with 50 mM DTT . Samples were run on NuPage 4–12% precast polyacrylamide gels ( Invitrogen ) and transferred to nitrocellulose membrane ( Invitrogen ) . Membranes were blocked in PBS containing 0 . 1% Tween-20% and 5% non-fat dry milk for 1 hr at room temperature before overnight incubation at 4°C with primary antibodies in block solution . Membranes were then incubated for 1 hr at room temperature with horseradish peroxidase-conjugated secondary antibodies in block solution and chemiluminescent substrate ( Thermo Fisher Scientific ) used for detection on film . Cells were fixed in PBS containing 2% paraformaldehyde ( Electron Microscopy Sciences , Hatfield , PA ) for 7 min at room temperature and permeabilized for 10 min in PBS with 0 . 5% Triton X-100 . Samples were then incubated overnight at 4°C with primary antibodies , followed by incubation with appropriate FITC- or TRITC-conjugated secondary antibodies for 1 hr at room temperature prior to DAPI counterstaining and imaging with a Zeiss Axiophot fluorescent microscope , AxioCam MRc digital camera and AxioVision 4 . 6 software ( Carl Zeiss Microscopy , Thornwood , NY ) . Image J software ( Schneider et al . , 2012 ) was used for image analysis and quantification . FLAG-dCas9 chromatin occupancy was analyzed as previously described ( Fujita and Fujii , 2013 ) using chromatin extraction and fragmentation methods from ( Forsberg et al . , 2000 ) and the following minor modifications . Five million trypsinized myoblasts were crosslinked with 1% formaldehyde ( Thermo Fisher Scientific ) for 10 min at room temperature . Chromatin was diluted to 0 . 5% SDS with IP Dilution Buffer ( 20 mM Tris pH 8 . 0 , 2 mM EDTA , 150 mM NaCl , 1% Triton X-100 , 0 . 01% SDS , cOmplete EDTA-free Protease Inhibitor Cocktail , 100 mM PMSF ) and fragmented to an average length of 500 bp using a Fisher Scientific Model 500 Sonic Dismembrator probe tip sonicator . Soluble chromatin was diluted to 0 . 2% SDS with IP Dilution Buffer before pre-clearing with 5 μg of mouse IgG conjugated to 20 μl of Dynabeads-Protein G ( Thermo Fisher Scientific ) followed by immunoprecipitation with 5 μg of anti-FLAG M2 antibody conjugated to 50 μl of Dynabeads-Protein G . Quantitative PCR was carried out on a QuantStudio 7 Flex ( Applied Biosystems , Waltham , MA ) using locus-specific primers and iTaq SYBR Green Supermix ( Bio-Rad Laboratories , Hercules , CA ) . Primer sequences are listed in Supplementary file 4 . The enChIP-MS procedure was performed as described previously ( Fujita and Fujii , 2013 ) using chromatin extraction and fragmentation methods from ( Forsberg et al . , 2000 ) and the following minor modifications . Forty million myoblasts were harvested by trypsinization and lysed in Cell Lysis Buffer ( 10 mM Tris pH 8 . 0 , 10 mM NaCl , 0 . 2% IGEPAL-CA630 , cOmplete EDTA-free Protease Inhibitor Cocktail , 100 mM PMSF ) . The isolated nuclei were crosslinked with 1–2% formaldehyde at room temperature for 10–20 min and then lysed in Nuclei Lysis Buffer ( 50 mM Tris pH 8 . 0 , 10 mM EDTA , 1% SDS , cOmplete EDTA-free Protease Inhibitor Cocktail , 100 mM PMSF ) . Chromatin was diluted to 0 . 5% SDS with IP Dilution Buffer and fragmented using a Fisher Scientific Model 500 Sonic Dismembrator probe tip sonicator to an average length of 3 kb . Sonicated chromatin was diluted to 0 . 2% SDS with IP Dilution Buffer , pre-cleared with 25 μg of mouse IgG conjugated to 100 μl of Dynabeads-Protein G and immunoprecipitated with 70 μg of anti-FLAG M2 antibody conjugated to 180 μl of Dynabeads-Protein G . An additional two Dynabead washes in Low Salt Wash Buffer replaced the high-salt washes . Eluted and precipitated samples were resuspended in SDS sample buffer , boiled and subjected to SDS-PAGE . Entire gel lanes were excised and proteins analyzed using an OrbiTrap Elite mass spectrometer ( Thermo Fisher Scientific ) coupled to an Easy-nLC II ( Thermo Fisher Scientific ) at the Fred Hutchinson Cancer Research Center Proteomics Shared Resource . The raw spectra were searched against a UniProt human protein database that also included common contaminants as defined in Mellacheruvu et al . ( 2013 ) using Proteome Discoverer 1 . 4 software ( Thermo Fisher Scientific ) to generate peptide-spectrum matches . The number of peptides that mapped to each protein was summarized to generate a ‘pseudoquant’ metric . Proteins with at least one peptide-spectrum match in two experimental replicates were carried forward for further analysis , after filtering out common contaminants . Finally , the UniProt annotations for Function and Subcellular location were used to restrict the analysis to only the nuclear proteins to enrich for biologically relevant , nuclear interactions . The R code used for the proteomics data analysis can be accessed via github ( https://github . com/sjaganna/2017-campbell_et_al ) ( Jagannathan , 2017 ) . The gRNA sequences are listed in Supplementary file 4 . GO analysis was carried out with the PANTHER classification system ( Mi et al . , 2016 ) using the statistical overrepresentation test against all human genes and the complete GO Biological process annotation . p-Values were corrected for multiple hypothesis testing using the Bonferroni correction . The occupancy of NuRD complex components and acetyl-Histone H4 was determined using crosslinked ChIP coupled with micrococcal nuclease digestion as described previously ( Skene and Henikoff , 2015 ) . For acetyl-Histone H4 samples , the Lysis Buffer and IP Dilution Buffer were supplemented with 10 mM sodium butyrate . Quantitative PCR was carried out on a QuantStudio 7 Flex using locus-specific primers and iTaq SYBR Green Supermix . Primer sequences are listed in Supplementary file 4 . Flexitube and ON-TARGETplus duplex siRNAs were obtained from Qiagen ( Hilden , Germany ) or GE Dharmacon ( Lafayette , CO ) , respectively . Transfections of siRNAs into myoblasts and iPS cells were carried out using Lipofectamine RNAiMAX ( Invitrogen ) according to the manufacturer’s instructions . A double transfection protocol was followed in myoblasts to ensure efficient depletion of pre-existing proteins . Briefly , cells were seeded at ~30% confluence in six-well plates and transfected ~20 hr later with 6 μl Lipofectamine RNAiMAX and 25 pmol of either gene-specific siRNA ( s ) or a scrambled non-silencing control siRNA diluted in 125 μl Opti-MEM Reduced Serum Medium ( Thermo Fisher Scientific ) . Forty-eight hours following this , myoblasts were transfected a second time and harvested for RNA analysis 48–72 hr later . In iPS cells , the same procedure was followed except cells were treated with 10 μM Y-27632 ROCK inhibitor ( Miltenyi Biotec , Auburn , CA ) for 24 hr before being trypsinized and seeded in mTeSR1 medium ( STEMCELL Technologies , Vancouver , BC ) at 1 × 105 cells/well on Matrigel ( Corning Life Science , Tewksbury , MA ) -coated six-well plates , and were harvested 48 hr after a single transfection . The sequences of siRNAs are listed in Supplementary file 4 . Total RNA was extracted from whole cells using the RNeasy Mini Kit ( Qiagen ) according to the manufacturer’s instructions . The isolated RNA was treated with DNase I ( Thermo Fisher Scientific ) , heat inactivated , and reverse transcribed into cDNA using Superscript III ( Thermo Fisher Scientific ) and oligo ( dT ) primers ( Invitrogen ) following the manufacturer’s protocol . Quantitative PCR was carried out on a QuantStudio 7 Flex using primers specific for each mRNA and iTaq SYBR Green Supermix . The relative expression levels of target genes were normalized to that of the reference genes RPL27 , RPL13A or GAPDH by using the delta-delta-Ct method ( Livak and Schmittgen , 2001 ) after confirming equivalent amplification efficiencies of reference and target molecules . Primer sequences are listed in Supplementary file 4 . The following antibodies were used: α-Tubulin ( T9026 ) ; Acetyl-Histone H4 ( 06–866 lot#2554112 , EMD Millipore ( Billerica , MA ) ) ; CHD4 ( A301-081A , Bethyl Laboratories ( Montgomery , TX ) ) ; FITC anti-mouse ( 715-095-150 lot#115855 , Jackson ImmunoResearch ( West Grove , PA ) ) ; FLAG M2 ( F1804 lot#SLBG5673V and lot#124K6106 ) ; FLAG M2 ( F3165 lot#SLBL1237V ) ; HDAC2 ( ab7029 , lot#GR88809-7 , Abcam ( Cambridge , UK ) ) ; HRP anti-mouse ( 115-035-146 , Jackson ImmunoResearch ) ; MBD2 ( A301-632A , Bethyl ) ; mouse IgG ( 315-005-003 lot#120058 , Jackson ImmunoResearch ) ; MTA2 ( ab8106 lot#GR185489-3 , Abcam ) ; TRITC anti-rabbit ( 711-025-152 lot#114768 , Jackson ImmunoResearch ) ; rabbit monoclonal antibodies against DUX4 ( E5-5 and E14-3 ) were produced in collaboration with Epitomics and are described elsewhere ( Geng et al . , 2012 ) . All collected data were included in the analyses . Statistical significance was determined using Mann-Whitney U or Wilcoxon signed-rank tests , as indicated in the corresponding figure legends . As is convention , at least three biological replicates per condition were used for ChIP-qPCR and RT-qPCR , as indicated . Here a biological replicate is defined as an independent culture of cells that was separately manipulated and subsequently analyzed . The enChIP-MS studies were multiple singleton experiments performed using several different gRNA that targeted the same genomic locus , as described . No statistical methods were used to predetermine sample size . Masking was not used during group allocation , data collection or data analysis . The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium ( http://proteomecentral . proteomexchange . org ) via the PRIDE partner repository ( Vizcaíno et al . , 2016 ) with the dataset identifier PXD006839 .	The DNA sequences of humans and other mammals contain many repetitive regions . This repetition makes these regions difficult to study with conventional approaches , and so the exact role of repetitive DNA is not fully understood . A particular sequence of repetitive DNA that plays an important role in human health contains a gene called DUX4 in each repeat . DUX4 is normally active in stem cells and in early-stage embryos . This gene is then switched off or ‘silenced’ during later stages of development and in most cells of the body . However , in some individuals the DUX4 gene inappropriately activates in muscle cells . This causes a disease known as facioscapulohumeral muscular dystrophy ( FSHD ) , in which muscle weakness begins in the face and upper body and eventually spreads to other muscles . Currently , there is no cure for FSHD . Proteins that bind to DNA can control the activity of nearby genes . Little is known about which proteins silence DUX4 at the appropriate time and in the right cells , so Campbell et al . set out to identify the proteins that attach to the repetitive DNA sequences containing DUX4 . Further investigation showed that several of these proteins play an important role in keeping DUX4 turned off , including two protein complexes called NuRD and CAF-1 . These complexes are necessary to silence DUX4 in human muscle cells and stem cells . Campbell et al . also identified a protein that can increase the activity of the DUX4 gene in FSHD muscle cells by overcoming the silencing activity of the NuRD complex . Overall , the results presented by Campbell et al . provide the groundwork for developing new treatments for FSHD . The next step will be to discover ways of enhancing the ability of NuRD and CAF-1 to silence the DUX4 gene .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "chromosomes", "and", "gene", "expression" ]	2018	NuRD and CAF-1-mediated silencing of the D4Z4 array is modulated by DUX4-induced MBD3L proteins
Many types of adult stem cells exist in a state of cell-cycle quiescence , yet it has remained unclear whether quiescence plays a role in maintaining the stem cell fate . Here we establish the adult germline of Caenorhabditis elegans as a model for facultative stem cell quiescence . We find that mitotically dividing germ cells—including germline stem cells—become quiescent in the absence of food . This quiescence is characterized by a slowing of S phase , a block to M-phase entry , and the ability to re-enter M phase rapidly in response to re-feeding . Further , we demonstrate that cell-cycle quiescence alters the genetic requirements for stem cell maintenance: The signaling pathway required for stem cell maintenance under fed conditions—GLP-1/Notch signaling—becomes dispensable under conditions of quiescence . Thus , cell-cycle quiescence can itself maintain stem cells , independent of the signaling pathway otherwise essential for such maintenance . Stem cells in adult tissues were once thought to exist primarily in a state of cell-cycle quiescence . Such quiescence was viewed as an inherent property of the stem cell fate and thus essential for a tissue’s long-term self-renewal ( Hall and Watt , 1989; Potten and Loeffler , 1990 ) . More recently , however , it has become clear that adult stem cells are not universally quiescent but instead cycle in accordance with the needs of the tissue: Some types of stem cells proliferate continuously , whereas others switch from quiescence to rapid proliferation in response to certain stimuli ( e . g . wounding or hormones ) ( Wabik and Jones , 2015 ) . In mammals , for example , hematopoietic and neural stem cells reversibly switch between quiescence and active proliferation in response to tissue injury ( Doetsch et al . , 1999; Harrison and Lerner , 1991; Lugert et al . , 2010 ) , and mammary stem cells expand transiently during pregnancy and the estrus cycle ( Asselin-Labat et al . , 2010; Joshi et al . , 2010 ) . Though periods of sustained stem cell proliferation enable rapid tissue growth or turnover , they challenge the view of quiescence as a prerequisite for the stem cell fate . Thus , a long-standing question has remained unanswered: Does cell-cycle quiescence play a role in stem cell maintenance ? Understanding the relationship between cell-cycle quiescence and stem cell maintenance has been difficult because tractable models of facultative stem cell quiescence have been lacking . Perturbations affecting the cell cycle can in some cases impact stem cell maintenance ( Orford and Scadden , 2008; Pietras et al . , 2011; Yilmaz et al . , 2012 ) , but whether quiescence can maintain stem cells independent of the signals otherwise required for their maintenance has been untested . Such a test requires a system in which cell-cycle quiescence can be readily induced , and in which the signals otherwise required for stem cell maintenance can be readily removed . In this study , we establish the adult germline of Caenorhabditis elegans as a model fitting these criteria . We describe a previously uncharacterized state of cell-cycle quiescence among adult germline stem cells , emerging under conditions of starvation . We then test whether this quiescence can maintain stem cells , independent of the signal required for their maintenance under conditions of active proliferation . The adult germline of C . elegans presents a tractable model for studying stem cell behavior because of its simple , linear organization ( Figure 1A ) . Mitotically dividing germ cells—including germline stem cells—reside in the distal region of the gonad ( the ‘progenitor zone’ ) . Differentiating germ cells , in meiotic prophase , are located more proximally . ( Here , we use the term ‘progenitor zone’ rather than the earlier term ‘mitotic zone’ or ‘proliferative zone’ to reflect the facultative nature of germ cell divisions . ) The progenitor zone has been studied under fed conditions and is composed of a distal pool of germline stem cells and a more proximal pool of cells that have begun to differentiate ( Cinquin et al . , 2010 ) . This proximal pool comprises cells dividing mitotically , as well as cells completing their final passage through interphase in preparation for entry into the meiotic cell cycle . We collectively refer to these cells as ‘transient progenitors’ , to reflect their continued mitotic divisions and transitional state ( Figure 1A ) . Under fed conditions , cells throughout the progenitor zone cycle asynchronously and continuously ( Crittenden et al . , 2006; Fox et al . , 2011; Jaramillo-Lambert et al . , 2007; Morgan et al . , 2010 ) , with transient progenitors undergoing one or two rounds of division as they pass through the proximal progenitor zone ( Fox and Schedl , 2015 ) . 10 . 7554/eLife . 10832 . 003Figure 1 . Fed versus starved adult hermaphrodite gonad of Caenorhabditis elegans . ( A ) Schematic of an adult hermaphrodite gonadal arm , with the progenitor zone at its distal end and maturing gametes at its proximal end . Germline stem cells and transient progenitors are located in the distal and proximal progenitor zone , respectively . Cells in both pools cycle asynchronously , although they are partially connected via a cytoplasmic core . Filled circles , germ cell nuclei in the progenitor zone . Open circles , germ cell nuclei in meiotic prophase , including developing oocytes . Gonads in males and larval hermaphrodites are organized similarly , although their proximal germ cells differentiate as sperm . This same gonad organization is also seen in starved animals of any stage or sex for time intervals examined in this work . ( B ) Images of distal gonads dissected from adult hermaphrodites and stained with DAPI to visualize DNA ( magenta ) and anti-phospho-histone H3 to visualize M-phase chromosomes ( green ) . M-phase cells are outlined and numbered . Left , fed early adult hermaphrodite . Right , hermaphrodite starved from early adult for 8 hr . ( See Materials and methods for definition of ‘early adult’ . ) Asterisks , distal gonad ends . Images are maximum-intensity z-projections . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 003 Prior to this work , germ cell proliferation in C . elegans adults had not been examined in detail under food-limited conditions . However , the effects of such conditions have been examined during larval development in C . elegans , as well as in adult Drosophila , and in both contexts , germ cells respond robustly to nutritional cues . In Drosophila , nutrient limitation or changes in nutrient-sensing pathways slow germ cell proliferation , reduce germline stem cell number , or both ( Armstrong et al . , 2014; Drummond-Barbosa and Spradling , 2001; Hsu et al . , 2008; LaFever et al . , 2010; McLeod et al . , 2010; Roth et al . , 2012; Sheng and Matunis , 2011 ) . These effects are mediated in part by changes in the somatic gonad ( Yang and Yamashita , 2015 ) , including changes in the size of the somatic niche supporting germline stem cells ( Bonfini et al . , 2015; Hsu and Drummond-Barbosa , 2011 ) . In C . elegans , primordial germ cells are born in the early embryo and arrest in the G2 phase of the cell cycle until newly hatched larvae begin to feed ( Butuci et al . , 2015; Fukuyama et al . , 2006; Fukuyama et al . , 2012 ) . This response to feeding has been hypothesized to involve food-related signals traveling through soma-to-germline gap junctions , which are required early in larval development for germ cell proliferation and survival ( Starich et al . , 2014 ) . Later in development , germ cells stop dividing if animals enter the non-feeding dauer larval stage ( Narbonne and Roy , 2006 ) . Even in non-dauer larvae , germ cells proliferate less when food is scarce , an effect mediated in part by communication between food-sensing neurons and the somatic gonad ( Dalfo et al . , 2012; Korta et al . , 2012 ) . In adult C . elegans , decreased food intake slows mitotic and meiotic progression and oogenesis ( Gerhold et al . , 2015; Lopez et al . , 2013; Salinas et al . , 2006; Seidel and Kimble , 2011 ) , and limited observations suggest that germ cell proliferation is also reduced ( Salinas et al . , 2006 ) . More strikingly , full starvation from the L4 larval stage causes dramatic germline shrinkage in adult hermaphrodites , and this shrinkage is reversible upon re-feeding ( Angelo and Van Gilst , 2009; Seidel and Kimble , 2011 ) . These observations motivated us to examine in greater detail how mitotically dividing germ cells in adult C . elegans respond to food removal . Here , we report that in the absence of food , mitotically dividing germ cells in adult C . elegans stop dividing and become quiescent . This quiescence is characterized by a dramatic slowing of S phase , cell-cycle arrest in G2 , and the ability to re-enter M phase rapidly in response to re-feeding . We investigate these cell-cycle responses in wildtype animals and in germline tumors , and we test whether this cell-cycle quiescence requires factors controlling larval or behavioral responses to food . We next investigate the control of stem cell maintenance under starved conditions . We uncover a major difference in the requirement for GLP-1/Notch signaling in the maintenance of actively proliferating versus quiescent germline stem cells . This work establishes the C . elegans germline as a model of facultative stem cell quiescence and demonstrates the utility of such a model in clarifying the role of quiescence in maintaining the stem cell state . To investigate how starvation affects germ cell division in adults , we removed food from early adult hermaphrodites and males and monitored the number of germ cells in M phase over the following 10 . 5 hr . Cells in M phase were identified by staining for phospho-histone H3 ( Figure 1B ) , a marker of M phase ( Hans and Dimitrov , 2001 ) . Food removal caused a drop in the number of M-phase cells ( Figure 2A ) , and this response was fast: In hermaphrodites , the number of M-phase cells per progenitor zone dropped from an average of 7 . 6 before food removal to 2 . 1 after 30 min without food ( n = 7 replicates of 220–551 gonadal arms per replicate per time point ) ( Figure 2A ) . The number of M-phase cells continued to decline thereafter , and after 3 . 5 hr without food , M-phase cells were virtually absent ( Figure 2A ) . This drop in M-phase cells did not occur in hermaphrodites fed continuously ( Figure 2—figure supplement 1B , D ) , nor in hermaphrodites exposed to a mock starvation procedure ( Figure 2—figure supplement 1C ) . In males , M-phase cells also disappeared rapidly in response to food removal , although the initial drop in M-phase cells was not monotonically decreasing ( Figure 2B ) . We conclude that in adults of both sexes , germ cells stop dividing quickly in the absence of food . 10 . 7554/eLife . 10832 . 004Figure 2 . Mitotic divisions in adult progenitor zones respond quickly to food removal and re-feeding . Time courses showing the number M-phase cells per progenitor zone after food removal or re-feeding . Time zero indicates the start of food removal or re-feeding . Animals in A , B and D , E were starved from early adult . Animals in C were starved from mid L4 . Animals in F were starved from mid L4 for 24 hr or from early L4 for 72 hr . Independent replicates are overplotted with transparency . For each replicate , lines connect means , and shaded areas show interquartile ranges . Sample sizes indicate numbers of gonadal arms . Source data are available in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 00410 . 7554/eLife . 10832 . 005Figure 2—source data 1 . Counts of M-phase cells for starvation and re-feeding time courses of wildtype animals . This comma-separated value file contains counts of M-phase cells per gonadal arm for starvation , re-feeding , and control time courses of wildtype animals ( Figure 2 and Figure 2—figure supplement 1 ) . Each row of the file represents a single gonadal arm . Descriptors include treatment group ( starvation , re-feeding , mock starvation , continued feeding ) , animal sex ( hermaphrodite , male ) , age at the start of treatment ( L4 , early adult , L4 + 24 hr ) , replicate , and time . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 00510 . 7554/eLife . 10832 . 006Figure 2—figure supplement 1 . Comparison of numbers of M-phase cells in fed , starved , and re-fed animals . ( A–D ) Time courses showing the number M-phase cells per progenitor zone in adult hermaphrodites exposed to food removal , mock food removal , or continuous feeding . For food removal or mock food removal , time zero indicates start of food removal or mock food removal . For continuous feeding , time zero indicates the time at which animals reached the early adult stage or the ‘24-hr post mid L4’ stage . ( E , F ) Normalized histograms showing the number of M-phase cells per progenitor zone for fed or re-fed adults . Data for fed animals are reproduced from the ‘before food removal’ time points of Figure 2A , C . Data for re-fed animals are reproduced from Figure 2B , D . Source data are available in Figure 2—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 006 We next investigated how germ cell division responds to re-feeding . We removed food from early adult hermaphrodites and males , allowed animals to remain in starvation for 12 hr , then re-fed animals , and monitored the number of M-phase cells , as above . In hermaphrodites , this treatment triggered a burst of M-phase cells 1 . 5 hr after the start of re-feeding ( Figure 2D ) . Males showed a similar response to re-feeding , but the burst of M-phase cells occurred 1 hr earlier ( Figure 2E ) . In both sexes , these bursts included some individual germlines having approximately twice as many M-phase cells as were observed among continuously fed animals ( Figure 2—figure supplement 1E , F ) . These results demonstrate that in both sexes , germ cells resume mitotic division rapidly in response to re-feeding . The faster response in males is consistent with germ cells in males having a faster cell-cycle under continuously fed conditions ( Morgan et al . , 2010 ) . Further , the higher maxima of M-phase cells in re-fed versus continuously fed animals is consistent with germ cells collecting at the G2-to-M transition during starvation and entering M phase semi-synchronously upon re-feeding . We next extended our results to adult hermaphrodites starved from the L4 larval stage . This extension was motivated by the need to perform certain later experiments in such animals , as starvation from L4 prolongs the amount of time that adult hermaphrodites can be maintained without food ( Angelo and Van Gilst , 2009; Seidel and Kimble , 2011 ) . We removed food from mid-L4 hermaphrodites and monitored the number of cells in M phase , as above . In starved L4s , M-phase cells persisted for ~4–5 hr after food removal , with the average number of M-phase cells only moderately reduced relative to fed animals ( Figure 2C ) . Thereafter , the number of M-phase cells declined rapidly , and after 10 . 5 hr without food , M-phase cells had virtually disappeared ( Figure 2C ) . The disappearance of M-phase cells coincided with the molt into adulthood ( ~5–8 hr after food removal ) , and the coincidence of these events persisted even under conditions where the timing of this molt was changed: Hermaphrodites starved from early L4 molted into adulthood ~12–20 hr after food removal , and gonads in these animals contained , on average , 1 . 6 M-phase cells per progenitor zone before the molt ( n = 57 , gonads collected 10 . 5 hr post food removal ) and 0 . 0 M-phase cells after the molt ( n = 63 , gonads collected 24 hr post food removal ) . These results demonstrate that germ cells in hermaphrodites starved from L4 do not immediately stop dividing in response to food removal , but germ cell division eventually ceases , at or near the molt into adulthood . This finding suggests that mitotically dividing germ cells in L4s are not equivalent to those in adults , a result consistent with previous studies ( Crittenden et al . , 2002; Dalfo et al . , 2012; Gerhold et al . , 2015; Michaelson et al . , 2010 ) . In other systems , re-entry into the mitotic cell cycle following a period of quiescence occurs more slowly after longer periods of quiescence ( Lum et al . , 2005; Soprano , 1994 ) . We therefore tested whether longer periods of starvation would delay mitotic re-entry upon re-feeding . We repeated the re-feeding time course in two types of animals having experienced longer starvation: Adult hermaphrodites starved from mid-L4 for 24 hr and adult hermaphrodites starved from early L4 for 72 hr . In both types of animals , re-feeding triggered a burst of M-phase cells 1 . 5 hr after re-feeding ( Figure 2F ) , similar to the re-feeding response in animals starved for only 12 hr ( compare Figure 2D versus F ) . We conclude that the timing of M-phase entry upon re-feeding is largely unaffected by the duration of preceding starvation , at least during the first 72 hr of starvation . We next examined how starvation affects progression of germ cells through G1 , S phase , and G2 . First , we monitored cell-cycle progression in fed animals . By labeling germlines with the thymidine analog 5-ethynyl-2′-deoxyuridine ( EdU ) and monitoring the fraction of EdU+ M-phase cells over time , we estimated a median cell-cycle length in fed early adult hermaphrodites of ~6 . 2 hr , with S phase lasting ~4 . 4 hr , G2 lasting ~1 . 3 hr , and G1 and M phase together lasting ~30 min ( Figure 3—figure supplement 1 ) . We also measured cell-cycle length in fed hermaphrodites aged 24-hr post mid-L4 ( ~12–16 hr past the early adult stage ) . For this age group , we estimated a median cell-cycle length of ~9 . 8 hr , with median G1 , S-phase , and G2 lengths of <30 min , ~6 . 8 hr , and ~2 . 3 hr , respectively , but with a long-tailed distribution of G2 lengths ( Figure 3—figure supplement 1 ) . Our estimates of median cell-cycle length are within the range reported by others ( Crittenden et al . , 2006; Fox et al . , 2011; Jaramillo-Lambert et al . , 2007; Morgan et al . , 2010 ) , and the long-tailed distribution of G2 lengths is consistent with estimates of maximal cell-cycle length being considerably longer than estimates of median cell-cycle length ( Crittenden et al . , 2006; Fox et al . , 2011; Jaramillo-Lambert et al . , 2007; Morgan et al . , 2010 ) . Second , we monitored cell-cycle progression during starvation . We removed food from animals of both sexes and , after varying amounts of time , calculated the fraction of progenitor-zone cells in G1 , S phase , and G2 . S-phase cells were identified by EdU labeling , and G1 versus G2 cells were distinguished by nuclear size ( see ‘Use of nuclear size to distinguish G1 versus G2 cells’ in Materials and methods and Figure 3—figure supplement 2 ) . The following time points were collected: Animals starved from early adult for 3 . 5 , 6 . 5 , and 10 . 5 hr; animals starved from mid-L4 for 10 . 5 and 24 hr; and animals starved from early L4 for 10 . 5 , 24 , 48 , and 72 hr . For all three age groups and for both sexes , G1 cells disappeared following food removal , and the timing of their disappearance coincided with the disappearance of M-phase cells ( Figure 3A ) . This result demonstrates that during starvation , G1 cells continued to initiate S phase without a pronounced delay . Our second observation was that for all age groups and sexes , the fraction of S-phase cells decreased over time during starvation , and the fraction of G2 cells increased ( Figure 3A ) . However , S-phase and G2 fractions changed more slowly during starvation than expected from measurements of S-phase length in fed animals ( Figure 3A ) . ( For example , in hermaphrodites starved from early adult , S-phase fractions changed very little during the 7 hr between time points 3 . 5 hr and 10 . 5 hr [Figure 3A] , indicating that progression through S phase during this time period was minimal . By contrast , cells in fed adult hermaphrodites complete S phase in a median of ~4 . 4–6 . 8 hr [Figure 3—figure supplement 1] . ) We conclude that during starvation , germ cells do not pause in G1 and instead continue through S phase and arrest in G2 . However , S-phase progression occurs much more slowly than under fed conditions . 10 . 7554/eLife . 10832 . 007Figure 3 . Starvation slows S-phase and causes germ cells to arrest in G2 . ( A ) Time courses showing the proportion of progenitor-zone cells in G1 , S-phase , or G2 , for animals starved from early adult or from mid or early L4 . Animals starved from mid L4 were adults at the 10 . 5- and 24-hr time points . Animals starved from early L4 were adults at the 24- , 48- , and 72-hr time points . n = 19–40 gonadal arms per time point . ( B ) Time courses showing the number of cells in the progenitor zone , for the same gonads used in A . Within each plot , lines connect means , and shaded areas show interquartile ranges . ( C ) Schematic summarizing the effect of starvation on the mitotic cell cycle of germ cells . Cell-cycle length under fed conditions was measured in adult hermaphrodites ( Figure 3—figure supplement 1 ) . Source data are available in Figure 3—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 00710 . 7554/eLife . 10832 . 008Figure 3—source data 1 . Counts of cells in each phase of the cell cycle for starvation time courses of wildtype animals . This comma-separated value file contains counts of progenitor-zone cells in each phase of the cell cycle for starvation time courses of wildtype animals ( Figure 3 ) . Each row of the file represents a single cell . Descriptors include cell-cycle phase ( G1 , S-phase , G2 , M-phase ) , animal sex ( hermaphrodite , male ) , age at the start of starvation ( early L4 , mid L4 , early adult ) , time , and a gonad identifier . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 00810 . 7554/eLife . 10832 . 009Figure 3—source data 2 . Counts of cells in each phase of the cell cycle for fed adult wildtype hermaphrodites . This comma-separated value file contains counts of progenitor-zone cells in each phase of the cell cycle for fed early adult hermaphrodites and fed hermaphrodites aged 24 hr post mid L4 ( Figure 3—figure supplement 1A–B ) . Each set of four rows of the file represents a single progenitor zone . Descriptors include cell-cycle phase ( G1 , S-phase , G2 , M-phase ) , age ( early adult , L4 + 24h ) , and a gonad identifier . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 00910 . 7554/eLife . 10832 . 010Figure 3—source data 3 . EdU labeling in fed adult wildtype hermaphrodites . This comma-separated value file contains counts of EdU-positive and EdU-negative M-phase cells in hermaphrodites exposed to EdU from the early adult stage or from 24 hr post mid L4 ( Figure 3—figure supplement 1C ) . Each rows of the file represent a single time point . Descriptors indicate age at the start of EdU labeling ( early adult , L4 + 24h ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 01010 . 7554/eLife . 10832 . 011Figure 3—source data 4 . Propidium iodide intensities versus nuclear volume in the progenitor zone . This comma-separated value file contains propidium iodide intensities and nuclear volumes of interphase progenitor-zone nuclei and distal tip cell nuclei of six EdU-labeled gonadal arms ( Figure 3—figure supplement 2 ) . Gonadal arms are from fed early adult hermaphrodites or hermaphrodites starved from early adult for 3 . 5 hr . Each row of the file represents a single cell . Descriptors include treatment group ( fed , starved ) , cell-cycle phase ( G1 , S-phase , G2 , distal tip cell ) , and a gonad identifier . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 01110 . 7554/eLife . 10832 . 012Figure 3—figure supplement 1 . Measurements of cell-cycle length in fed animals . ( A ) Proportions of progenitor-zone cells in each phase of the cell cycle in fed early adult hermaphrodites and fed adult hermaphrodites aged 24-hr post mid L4 ( ‘L4 + 24 h’ ) . Data for early adults derives from the ‘before food removal’ time point of the time course in Figure 3A . Data for L4 + 24 animals derived from 0 . 5-hr time point of the time course in C . Sample sizes indicate number of gonadal arms . ( B ) Number of progenitor-zone cells for the same gonads used in C . ( C ) EdU-labeling time course to measure G2 length . Fed early adult hermaphrodites and hermaphrodites aged 24-hr post mid L4 were exposed to continuous EdU labeling . The proportion of M-phase cells that were EdU+ was measured at the time points shown . Open green circle , sample contained some gonadal arms in which all progenitor-zone cells were EdU+ . Filled green circle , all progenitor-zone cells in all gonadal arms were EdU+ cells . ( D ) Median indices for each cell-cycle phase , calculated from the data in A . ( E ) Time at which 50% , 95% , or 100% of M-phase cells were EdU+ , from the time course in C . ( F ) Median length of each cell-cycle phase and the total cell cycle , calculated from the data in D , E . ( G ) Estimates of maximum total cell-cycle length , calculated from the data in E , F . Source data for A , B are available in Figure 3—source data 2 . Source data for C are available in Figure 3—source data 3 . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 01210 . 7554/eLife . 10832 . 013Figure 3—figure supplement 2 . G1 versus G2 cells can be distinguished by nuclear size in EdU-labeled gonads . ( A ) Propidium iodide intensities ( a measure of DNA content ) versus nuclear volume for interphase progenitor-zone nuclei and the distal tip cell nucleus in six gonadal arms . Propidium iodide intensities are scaled relative to the distal tip cell and the mean of G2 cells , to account for gonad-to-gonad differences in staining . Left , fed early adult hermaphrodites . Right , hermaphrodites starved from early adult for 3 . 5 hr . Cell-cycle phases were assigned by EdU uptake ( S-phase cells ) and nuclear size ( G1 versus G2 ) . AU , arbitrary units . ( B ) Images of EdU-labeled distal gonads ( 15 min EdU pulse ) dissected from fed early adult hermaphrodites and stained with DAPI to visualize DNA ( magenta ) . Asterisks , distal gonad ends . Dashed lines , progenitor zone boundaries . Yellow and blue circles , examples of G1 and G2 nuclei , respectively . Caret , paired occurrence of G1 cells . ( C ) Images of progenitor-zone nuclei in each stage of the cell cycle . Nuclei derive from gonads described as in B . n/a , not applicable . Source data are available in Figure 3—source data 4 DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 013 We next examined the effect of re-feeding on cell-cycle progression through S phase and G2 . We re-fed starved animals and determined the length of time until all germ cells entered M phase . This analysis is complementary to our initial monitoring of M-phase cells ( Figure 2 ) because our initial monitoring allowed us to detect the earliest germ cells to enter M-phase upon re-feeding , whereas this analysis allowed us to detect the slowest such cells . To detect M-phase entry of all cells , we blocked M-phase exit using a temperature-sensitive mutation in emb-30 , which encodes a component of the anaphase-promoting complex ( Furuta et al . , 2000 ) . emb-30 ( tn377ts ) hermaphrodites were starved at the permissive temperature ( 15°C ) , then shifted to the restrictive temperature ( 25°C ) , and re-fed . At the time of re-feeding , germlines contained a mixture of S-phase and G2 cells ( data not shown ) . In response to re-feeding , germ cells entered M phase , as expected , but were unable to complete the metaphase-to-anaphase transition , thus enabling us to quantify the accumulation of M-phase cells: Two hours after re-feeding , 55% of cells , on average , had entered M phase ( n = 60 gonadal arms ) ; after 4 hr , this number reached 99% ( n = 66 gonadal arms ) ( Figure 4B ) . ( This analysis is restricted to the distal-most 50 germ cells , because germ cells located more proximally sometimes directly entered the meiotic cell cycle upon re-feeding . See Figure 4 . ) Therefore , virtually all mitotically dividing germ cells completed the remainder of S phase and G2 within 4 hr of re-feeding , a time shorter than the time required to complete all of S phase and G2 in continuously fed adult hermaphrodites ( ~4 . 4–6 . 8 hr for S-phase , plus ~1 . 3–2 . 3 hr for G2 ) . Thus , not only does re-feeding trigger germ cells arrested in G2 to enter M phase , but re-feeding also restores the rate of progression through S phase and earlier stages of G2 . 10 . 7554/eLife . 10832 . 014Figure 4 . Re-feeding restores the rate of progression through S phase and G2 , as well as the meiotic entry of transient progenitors . ( A ) Images of distal gonads dissected from emb-30 ( tn377ts ) adult hermaphrodites and stained with DAPI to visualize DNA ( magenta ) and anti-GLD-1 ( green ) . Animals were starved at 15°C , shifted to 25°C , and then either re-fed or maintained in starvation . Time = 0 hr indicates the time at which re-feeding was started or starvation continued . Top , starved adult hermaphrodite , Time = 0 hr . Middle , re-fed adult hermaphrodite , Time = 6 hr . Bottom , adult hermaphrodite maintained in starvation , Time = 6 hr . Dashed lines , ‘crescent’/‘non-crescent’ boundaries ( defined by the second distal-most ‘crescent’ cell—see Materials and methods section ) . Arrowhead , an example of a metaphase-arrested cell having a high level of GLD-1 . Asterisks , distal gonad ends . Images are maximum-intensity z-projections . ( B ) Time course showing the proportion of cells in M-phase ( among the distal-most 50 germ cells ) for emb-30 ( tn377ts ) adult hermaphrodites treated as in A . n = 31–66 gonadal arms per time point . ( C ) Number of cells distal to the ‘crescent’/‘non-crescent’ boundary for the same gonads used in B . ( D ) Cell-cycle behavior , among cells distal to the ‘crescent’/‘non-crescent’ boundary , for the same gonads used in B . Cells from individual gonadal arms are plotted along lines , with color indicating cell-cycle phase ( interphase or M-phase ) and location determined by each cell’s position along the distal-to-proximal axis of distal gonad . For each time point , this axis is scaled relative to the mean number of cells distal to the ‘crescent’/‘non-crescent’ boundary . ( E ) Summary of the effects of re-feeding and continued starvation on germ cells in emb-30 ( tn377ts ) hermaphrodites in the distal versus proximal progenitor zone . Source data are available in Figure 4—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 01410 . 7554/eLife . 10832 . 015Figure 4—source data 1 . Temperature-shift experiments of emb-30 ( tn377ts ) hermaphrodites . This comma-separated value file contains counts of interphase and M-phase progenitor-zone cells for emb-30 ( tn377ts ) hermaphrodites re-fed at the restrictive temperature or maintained in continued starvation ( Figure 4 ) . Each row of the file represents a single cell . Descriptors include treatment group ( re-fed , starved ) , cell-cycle phase ( interphase , M-phase ) , ranked proximity to the distal tip cell ( numeric ) , time , and a gonad identifier . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 015 Mitotically dividing germ cells in adults are confined to the distal gonad , where they contact a single somatic cell—the distal tip cell ( or pair of distal tip cells , in males ) ( Figure 1A ) . The distal tip cell forms the niche for germline stem cells ( Kimble and Seidel , 2013 ) and influences how germ cells respond to physiological cues ( Dalfo et al . , 2012 ) . We therefore tested the effect of proximity to the distal tip cell on cell-cycle responses to food removal and re-feeding . We monitored M-phase cells , as above , in two mutant backgrounds in which mitotically dividing germ cells fill the gonad ( i . e . germline tumors ) : ( i ) glp-1 ( oz112gf ) /Notch gain-of-function mutants , in which constitutive GLP-1/Notch signaling maintains all germ cells in the mitotic cell cycle ( Berry et al . , 1997 ) and ( ii ) gld-3 ( q730 ) nos-3 ( q650 ) loss-of-function mutants , in which meiotic entry is inhibited irrespective of GLP-1/Notch signaling ( Byrd et al . , 2014; Eckmann et al . , 2004 ) . Germ cells in both genotypes of germline tumors responded normally to starvation and re-feeding: Outside the region normally corresponding to the progenitor zone ( i . e . outside the distal-most 20 rows of germ cells ) , M-phase cells disappeared quickly in response to food removal and re-bounded 1 to 2 hr after re-feeding ( Figure 5 ) . These results demonstrate that proximity to the distal tip cell—the germline stem cell niche—is not required for a normal starvation and re-feeding response . Additionally , these results refine our understanding of the control of germ cell fate: Constitutive GLP-1/Notch signaling or combined loss of gld-3 and nos-3 does not promote germ cell proliferation per se , but rather promotes an undifferentiated fate in which cells divide only in the presence of food . 10 . 7554/eLife . 10832 . 016Figure 5 . Ectopic mitotic divisions outside the progenitor zone respond normally to food removal and re-feeding . Time courses showing the number of M-phase cells—outside the distal-most 20 rows of germ cells—after food removal or re-feeding in adult glp-1 ( oz112gf ) hermaphrodites or adult gld-3 ( q730 ) nos-3 ( q650 ) hermaphrodites . Time zero indicates the start of food removal or re-feeding . Lines connect means , and shaded areas show interquartile ranges . n = 50–249 gonadal arms per time point . Source data are available in Figure 5—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 01610 . 7554/eLife . 10832 . 017Figure 5—source data 1 . Counts of M-phase cells for starvation and re-feeding time courses of glp-1 ( oz112gf ) and gld-3 ( q730 ) nos-3 ( q650 ) hermaphrodites . This comma-separated value file contains counts of M-phase cells per gonadal arm for glp-1 ( oz112gf ) and gld-3 ( q730 ) nos-3 ( q650 ) starvation and re-feeding time courses ( Figure 5 ) . Each row of the file represents a single gonadal arm . Descriptors include treatment group ( starvation , re-feeding ) , genotype ( glp-1 ( oz112gf ) , gld-3 ( q730 ) nos-3 ( q650 ) ) , and time . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 017 In multiple types of stem and progenitor cells , cell-cycle quiescence correlates with the capacity for long-term self-renewal ( Cheung and Rando , 2013; Orford and Scadden , 2008 ) . We therefore investigated how quiescence affects the maintenance of C . elegans germline stem cells . Under fed conditions , maintenance of these stem cells requires GLP-1/Notch signaling ( Austin and Kimble , 1987 ) . The glp-1 gene , which encodes one of two Notch receptors in C . elegans , is expressed in the progenitor zone , and the receptor is activated by ligands expressed in the adjacent distal tip cell ( Henderson et al . , 1994; Kimble and Crittenden , 2007; Nadarajan et al . , 2009 ) . We used the temperature-sensitive glp-1 allele q224ts to test whether GLP-1/Notch signaling is similarly required for maintenance of germline stem cells under starved conditions . The q224ts allele is the strongest of all known temperature-sensitive glp-1 alleles and behaves like a null at the restrictive temperature ( Austin and Kimble , 1987; Kodoyianni et al . , 1992 ) . In fed animals , removal of GLP-1/Notch signaling causes germline stem cells to be lost: Germline stem cells fail to self-renew , and instead all germ cells enter the meiotic cell cycle ( Austin and Kimble , 1987; Cinquin et al . , 2010; Fox and Schedl , 2015 ) . We tested whether the loss of GLP-1/Notch signaling produces these same effects in starved animals . We removed food from glp-1 ( q224ts ) hermaphrodites at the permissive temperature ( 15°C ) , shifted starved animals to the restrictive temperature ( 25°C ) for 8 hr , and then evaluated germ cell fate . Cell fate was assessed by staining for the meiosis-associated protein GLD-1 ( Jones et al . , 1996 ) and by scoring cells for the ‘crescent’ chromosome morphology indicative of meiotic chromosome pairing ( Dernburg et al . , 1998 ) . This morphology is readily distinguishable from the ‘non-crescent’ morphology found in mitotic interphase . In fed controls , incubation at the restrictive temperature caused all germ cells to enter the meiotic cell cycle: Chromosomes adopted the ‘crescent’ shape ( Figure 6A ) , and GLD-1 levels in the distal-most germ cells rose ( Figure 6I ) . In starved animals , by contrast , meiotic entry did not occur: 99% ( n = 214 ) of progenitor zones retained germ cells with a interphase chromosome morphology ( Figure 6C ) , and GLD-1 levels in the distal-most germ cells remained low ( n = 47 of 47 gonadal arms ) ( Figure 6I ) . Importantly , germ cells in starved animals retained the capacity for mitotic cell division , because when starved animals were re-fed at the restrictive temperature , their germ cells re-entered M phase ( Figure 6G ) . Germ cells in starved animals also retained the capacity for long-term self-renewal , because when starved animals were instead returned to the permissive temperature and re-fed for 2–3 days , 91% ( n = 148 ) of progenitor zones retained germ cells in the mitotic cell cycle ( Figure 6C ) . Similar results were observed when incubation at the restrictive temperature was extended to 16 hr or 24 hr ( Figure 6D ) . These results demonstrate that starved animals maintain germline stem cells independent of GLP-1/Notch . In other words , starvation inhibits the meiotic entry of germline stem cells , even in the absence of GLP-1/Notch . Similar to cell-cycle quiescence , this inhibition of meiotic entry was reversible upon re-feeding , because when starved glp-1 ( q224ts ) animals were re-fed at the restrictive temperature , all germ cells eventually entered the meiotic cell cycle ( Figure 6H ) . 10 . 7554/eLife . 10832 . 018Figure 6 . Quiescence induced by three different conditions maintains germline stem cells independent of GLP-1/Notch . ( A–F ) Distance between the gonad distal tip and the first ‘crescent’ germ cell for adult hermaphrodites of the genotypes and treatments shown . Animals were grown at 15°C , shifted to 25°C for 8 , 16 , or 24 hr , and then returned to 15°C for 2–3 days . Gonads were collected prior to the temperature shift ( 15°C ) , immediately following the temperature shift ( 25°C for 8 hr , 25°C for 16 hr , or 25°C for 24 hr ) , and following the 15°C recovery period ( … + 15°C for 2–3 days ) . Data are plotted as vertical histograms , with black circles denoting means . n = 55–488 gonadal arms per time point . ( G ) Time course showing the number of M-phase cells per progenitor zone in glp-1 ( q224ts ) hermaphrodites starved from early adult , shifted to 25°C for 8 hr , and then re-fed at 25°C . Lines connect means , and shaded areas show interquartile ranges . ( H ) Time courses showing distance between the gonad distal tip and the first ‘crescent’ germ cell . Top , hermaphrodites starved from early adult , shifted to 25°C for 8 hr , and then re-fed at 25°C . Bottom , hermaphrodites fed continuously and shifted to 25°C at the early adult stage . Data are plotted as vertical histograms , with lines connecting means . Sample sizes in G , H indicate number of gonadal arms . ( I ) Images of distal gonads dissected from adult hermaphrodites and stained with DAPI to visualize DNA ( magenta ) and anti-GLD-1 ( green ) . Gonads were dissected before and after 8 hr at 25°C . Left-hand panels , fed adult hermaphrodites . Right-hand panels , hermaphrodites starved from early adult . Asterisks , distal gonad ends . ( J ) Schematic summarizing the effect of cell-cycle quiescence on germline stem cell maintenance . Under conditions of active proliferation , GLP-1/Notch is required for germline stem cell maintenance . Under quiescent conditions , GLP-1/Notch is dispensable . Source data for A–F are available in Figure 6—source data 1 . Source data for G , H are available in Figure 6—source data 2 . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 01810 . 7554/eLife . 10832 . 019Figure 6—source data 1 . Temperature-shift experiments of glp-1 ( q224 ) hermaphrodites . This comma-separated value file contains measurements of the distance between the gonad distal tip and the first ‘crescent’ germ cell in temperature-shifted glp-1 ( q224 ) hermaphrodites ( Figure 6A–F ) . Distances are measured in germ cell diameters . Each row of the file represents a single gonadal arm . Descriptors include growth conditions ( fed , cdk-1 RNAi , starved from early adult , starved from mid-L4 , 300 mM NaCl , no sperm ) and temperature ( 15°C; 25°C for 8 hr; 25°C for 16 hr; 25°C for 24 hr; 25°C for 8 hr + 15°C for 2–3 days; 25°C for 16 hr + 15°C for 2–3 days; 25°C for 24 hr + 15°C for 2–3 days ) . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 01910 . 7554/eLife . 10832 . 020Figure 6—source data 2 . Time courses of temperature-shifted glp-1 ( q224 ) hermaphrodites . This comma-separated value file contains counts of M-phase cells and distances between the gonad distal tip and the first ‘crescent’ germ cell for two groups of animals: ( i ) glp-1 ( q224ts ) hermaphrodites starved from early adult , shifted to 25°C for 8 hr , and then re-fed at 25°C , and ( ii ) glp-1 ( q224ts ) hermaphrodites fed continuously and shifted to 25°C at the early adult stage ( Figure 6G–H ) . Distances are measured in germ cell diameters . Each row of the file represents a single gonadal arm . Descriptors include treatment group ( fed , starved ) and time . NA , not determined . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 02010 . 7554/eLife . 10832 . 021Figure 6—figure supplement 1 . Cell-cycle arrest caused by cdk-1 RNAi does not maintain germline stem cells in the absence of GLP-1/Notch . Images of distal gonads dissected from glp-1 ( q224ts ) adult hermaphrodites exposed to cdk-1 RNAi ( lower panels ) or the empty RNAi vector , L440 ( upper panels ) . Gonads were dissected before and after 8 hr at 25°C and stained with DAPI to visualize DNA ( magenta ) and anti-HIM-3 ( green ) to visualize formation of the synaptonymal complex , indicating meiotic entry ( Zetka et al . , 1999 ) . Asterisks , distal gonad ends . Arrowheads , examples of nuclei in the distal ends in which HIM-3 has been loaded onto chromosomes . Dashed lines , progenitor zone boundaries . Images are maximum-intensity z-projections . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 021 To further investigate the relationship between cell-cycle quiescence and stem cell maintenance , we asked , apart from starvation , do other conditions that inhibit germ cell division also maintain germline stem cells independent of GLP-1/Notch ? To answer this question , we performed temperature-shift experiments in glp-1 ( q224ts ) hermaphrodites under two additional conditions: High NaCl and absence of sperm , each of which causes a twofold drop in the germ cell mitotic index ( Morgan et al . , 2010; Salinas et al . , 2006 ) . Animals exposed to high NaCl ( 300 mM ) or lacking sperm ( via loss-of-function mutation in fog-1 ) were grown at the permissive temperature , shifted to the restrictive temperature for 8 hr , and then returned to the permissive temperature for 2–3 days . Following this treatment , germ cell fate ( mitotic versus meiotic ) was evaluated by scoring germ cells for the ‘crescent’ chromosome morphology indicative of meiotic prophase ( described above ) . For both high NaCl and absence of sperm , 48–79% ( n = 129–231 ) of gonadal arms retained germ cells in the mitotic cell cycle ( Figure 6E–F ) . To control for possible pleiotropic effects of the mutation used to eliminate sperm ( q785 ) , sperm was introduced to fog-1 ( q785 ) ; glp-1 ( q224ts ) animals by mating with wildtype males , and these animals were examined in parallel . In such animals , incubation at the restrictive temperature caused all germ cells to enter the meiotic cell cycle ( data not shown; n = 37 gonadal arms , scored immediately following incubation at the restrictive temperature ) . Thus , three stress conditions that inhibit or reduce germ cell division ( starvation , high NaCl , and absence of sperm ) also permitted maintenance of germline stem cells independent of GLP-1/Notch . This commonality suggests that cell-cycle quiescence itself is an effector of stem cell maintenance ( Figure 6J ) . Nonetheless , such maintenance is not simply a function of inhibiting passage through M-phase , because stem cell maintenance in the absence of GLP-1/Notch was not permitted by cell-cycle arrest caused by RNAi knockdown of cyclin-dependent kinase 1 ( cdk-1 ) ( Figure 6B ) , a result we confirmed by visualizing formation of the synaptonemal complex ( Figure 6—figure supplement 1 ) . The results above demonstrate that starvation inhibits the meiotic entry of germline stem cells , even in the absence of GLP-1/Notch . We therefore hypothesized that starvation might also control the meiotic entry of transient progenitors , located in the proximal progenitor zone ( Figure 1A ) . No markers currently exist to distinguish transient progenitors from germline stem cells , and the boundary between these pools of cells is not clear , but the two pools can be distinguished under fed conditions by restricting movement of cells out of the progenitor zone ( Cinquin et al . , 2010 ) . Under such conditions , transient progenitors enter the meiotic cell cycle , whereas germline stem cells do not ( Cinquin et al . , 2010 ) . To compare meiotic entry of transient progenitors under fed versus starved conditions , we re-fed starved animals and restricted cell movement during re-feeding; we then compared meiotic entry after re-feeding to meiotic entry after continued starvation . Meiotic entry was assessed by staining for GLD-1 and by scoring cells for the ‘crescent’ chromosome morphology indicative of meiotic prophase ( described above ) . Cell movement was restricted by performing this experiment in emb-30 ( tn377ts ) hermaphrodites at the restrictive temperature ( 25°C ) , a condition that induces metaphase arrest ( also described above , under ‘Re-feeding restores the rate of progression through S phase and G2’ ) ( Furuta et al . , 2000 ) . emb-30 ( tn377ts ) animals were starved at the permissive temperature ( 15°C ) , shifted to the restrictive temperature ( 25°C ) , and then re-fed for 6 hr or maintained in starvation . Our primary result from this experiment was that re-feeding and continued starvation affected meiotic entry differently . After 6 hr of re-feeding , cells in the proximal ( but not distal ) progenitor zone entered the meiotic cell cycle: GLD-1 levels in the proximal half of the progenitor zone rose ( Figure 4A ) , and the boundary between ‘crescent’ and ‘non-crescent’ germ cells moved distally , such that the number of cells distal to this boundary was reduced by about half ( Figure 4C ) . After 6 hr of continued starvation , by contrast , very little meiotic entry was observed: GLD-1 levels in the proximal progenitor zone remained largely unchanged ( Figure 4A ) , and the number of cells distal to the ‘crescent’/’non-crescent’ boundary was only slightly reduced ( Figure 4C ) . Our inference from these results is that the proximal half of the progenitor zone was composed of transient progenitors at the start of re-feeding , and that 6 hr of re-feeding—but not 6 hr of continued starvation—allowed for their timely progression into the meiotic cell cycle . We conclude that starvation slows or blocks the meiotic entry of transient progenitors , similar to its effect on the meiotic entry of germline stem cells . As a first step towards understanding the regulation of cell-cycle quiescence induced by starvation , we compared this quiescence to cell-cycle arrest caused by DNA damage , another perturbation causing G2 arrest ( Gartner et al . , 2000; Kuntz and O'Connell , 2009 ) . Following three criteria , starvation-induced quiescence was distinct from cell-cycle arrest caused by DNA damage . First , DNA damage strongly up-regulates inhibitory phosphorylation of CDK-1 ( Figure 7A; Figure 7—figure supplement 1; Craig et al . , 2012 ) . By contrast , starvation-induced quiescence did not up-regulate this phosphorylation ( Figure 7A ) . Second , DNA damage causes germ cell nuclei to enlarge ( Figure 7A; Gartner et al . , 2000 ) , a phenotype replicated by RNAi knockdown of cdk-1 ( Jeong et al . , 2011 ) . By contrast , starvation-induced quiescence did not cause nuclei to enlarge ( Figure 7A ) . Third , cell-cycle arrest in response to DNA damage requires the p53 homolog cep-1 ( Derry et al . , 2007 ) . By contrast , starvation-induced quiescence did not require cep-1 ( Figure 7B ) . We conclude that starvation and DNA damage induce cell-cycle arrest differently . 10 . 7554/eLife . 10832 . 022Figure 7 . Starvation-induced quiescence is distinct from the DNA damage response and does not require factors involved in larval or behavioral responses to food . ( A ) Images of distal gonads dissected from adult hermaphrodites and stained with DAPI to visualize DNA ( magenta ) and anti-phosopho-CDK-1 ( green , Santa Cruz #sc-28435-R ) to visualize inhibitory phosphorylation of CDK-1 . The phospho-specificity of anti-phosopho-CDK-1 is shown in Figure 7—figure supplement 1 . Top row , fed adult hermaphrodite . Center row , adult hermaphrodite starved from early adult for 8 hr . Bottom row , fed adult hermaphrodite treated with UV light to induce DNA damage . Asterisks , distal gonad ends . Arrowheads , example of an enlarged nucleus having elevated phospho-CDK-1 . Similar results were observed using a different phospho-CDK-1 antibody ( Calbiochem #219440 , data not shown ) . ( B ) Time courses showing the mean number of M-phase cells per progenitor zone after food removal for hermaphrodites of the genotypes shown . Animals were starved from early adult . Grey curves represent seven replicates of wild type , reproduced from Figure 2A . Time zero indicates the start of food removal . Lower right corner , wildtype hermaphrodites exposed to 20 mM serotonin at the onset of food removal ( 0 hr ) or 3 hr later . n = 46–452 gonadal arms per time point . Source data are available in Figure 7—source data 1 . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 02210 . 7554/eLife . 10832 . 023Figure 7—source data 1 . Starvation time courses of mutants and of wildtype hermaphrodites exposed to exogenous serotonin . This comma-separated value file contains counts of M-phase cells per gonadal arm for the starvation time courses shown in Figure 7B . Each row of the file represents a single gonadal arm . Descriptors include genotype ( cep-1 ( gk138 ) , aak-2 ( gt33 ) , etc . ) and time . For serotonin exposure , genotype is listed as ‘serotonin 0 hr’ or ‘serotonin 3 hr’ . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 02310 . 7554/eLife . 10832 . 024Figure 7—figure supplement 1 . Validation of phospho-specificity of anti-phosopho-CDK-1 . Images of whole gonadal arms dissected from fed early adult hermaphrodites . Gonads were treated with or without lambda protein phosphatase and stained with anti-phosopho-CDK-1 ( Santa Cruz #sc-28435-R ) and DAPI to visualize DNA . Images are maximum-intensity z-projections . DOI: http://dx . doi . org/10 . 7554/eLife . 10832 . 024 As a second step towards understanding the regulation of starvation-induced quiescence , we tested whether this quiescence requires factors influencing the larval germline’s response to food ( Dalfo et al . , 2012; Michaelson et al . , 2010 ) , including factors controlling germ cell quiescence during the two larval diapause states—L1 diapause ( Fukuyama et al . , 2006; Fukuyama et al . , 2012 ) and the mid-larval dauer diapause ( Narbonne and Roy , 2006 ) . Food was removed from early adult hermaphrodites homozygous for mutations in the transforming growth factor beta ( TGF-β ) pathway , the insulin/insulin-like growth factor 1 ( IGF-1 ) pathway , or the AMP-activated protein kinase ( AMPK ) pathway . M-phase cells were then monitored after food removal . Additionally , to test for a requirement for factors affecting behavioral responses to food , this same experiment was performed in animals defective for neuropeptide processing , neuropeptide secretion , or chemosensation , as well as in animals exposed to exogenous serotonin , which in some contexts acts as a food signal ( Luedtke et al . , 2010 ) . In all experiments , M-phase cells disappeared quickly in response to food removal ( Figure 7B ) . We conclude that each of the genes tested is not individually required for starvation-induced quiescence in adult germ cells; likewise , quiescence is not affected by exogenous serotonin . These findings suggest that quiescence in adults is controlled differently than cell-cycle responses in the larval germline and is largely independent of behavioral responses to food . These results are consistent with adult versus larval germ cells responding to food removal differently ( Figure 2A versus C ) and with reduced insulin/IGF-1 signaling or TGF-β signaling not affecting the germ cell mitotic index in fed adult hermaphrodites ( Dalfo et al . , 2012; Michaelson et al . , 2010 ) . Similarly , these results are consistent with no requirement for daf-16/FOXO in reducing the proliferation of Drosophila germline stem cells in response to poor diet ( Hsu et al . , 2008 ) . Cell-cycle quiescence was once thought to be a near universal feature of stem cells in adult tissues ( Hall and Watt , 1989; Potten and Loeffler , 1990 ) . This view arose from the theory that biological systems ought to protect stem cells from the risks of DNA replication and led to the notion of quiescence as an inherent property of the stem cell fate . According to this model , tissues were maintained by a hierarchy of relatively quiescent master stem cells and their faster cycling but short-lived daughters . In recent years , however , work in several mammalian and invertebrate tissues has shown that quiescence is not a prerequisite for the stem cell fate ( Barker et al . , 2010a; Crittenden et al . , 2006; Doupe and Jones , 2013; Fuchs , 2009; Maciejowski et al . , 2006; Simons and Clevers , 2011 ) . Some types of stem cells do not exhibit quiescence under conditions assayed ( e . g . Barker et al . , 2010b; de Navascues et al . , 2012; Snippert et al . , 2010 ) , and still others vary their cell-cycle length in accordance with the physiological circumstances surrounding them ( e . g . Harrison and Lerner , 1991; Hartman et al . , 2013; Lugert et al . , 2010; Qiao et al . , 2007 ) . Nevertheless , genetic or environmental perturbations that impact the cell cycle can also affect stemness ( Orford and Scadden , 2008; Pietras et al . , 2011; Yilmaz et al . , 2012 ) . Thus , a long-standing question in the field of stem cell biology has remained: Does cell-cycle quiescence play a role in maintaining the stem cell fate ? Our results answer this question by showing that quiescence itself alters the genetic requirements for stemness: Actively dividing germline stem cells in C . elegans require GLP-1/Notch signaling for their maintenance ( Austin and Kimble , 1987 ) ; we find that cell-cycle quiescence—induced by starvation , high NaCl , or absence of sperm—relieves this requirement ( Figure 6J ) . Thus , cell-cycle quiescence maintains germline stem cells independent of the signal required for such maintenance under conditions of active proliferation . The molecular mechanisms maintaining stem cells during periods of cell-cycle quiescence remain to be determined . Acting downstream of GLP-1/Notch signaling to maintain germline stem cells are the PUF-family translational repressors FBF-1 and FBF-2 ( Crittenden et al . , 2002 ) and the proteins of unknown molecular function LST-1 and SYGL-1 ( Kershner et al . , 2014 ) . Quiescence might stabilize these stem cell regulators—for example , by inhibiting the protein degradation machinery linked to the cell-cycle . Alternatively , quiescence might substitute for the repressive effects of FBF-1 and FBF-2 by repressing translation on a global level . Global repression of translation is a conserved stress response ( Spriggs et al . , 2010 ) , and the stress of starvation represses translation of at least a few genes in C . elegans ( Lascarez-Lagunas et al . , 2014 ) . Another possibility is that stem cell maintenance might be regulated by a metabolite or metabolic process whose levels change during quiescence . Such connections between metabolism and developmental processes have been observed in a variety of vertebrate and invertebrate cell types ( Agathocleous and Harris , 2013 ) . We find that adult germ cells in C . elegans do not arrest in G1 during starvation but instead progress slowly through S phase and arrest in G2 . Most eukaryotic cells can transiently pause in G2 in response to DNA damage or microtubule disassembly ( Rieder , 2011 ) , but the G2-to-M transition has not been viewed as a point of cell-cycle control in response to growth factors or nutrients , largely because early studies in mammalian tissue culture showed that cell-cycle events in fibroblasts become independent of extracellular cues after entry into S phase ( Pardee , 1989 ) . Despite this view , the G2-to-M transition is emerging as the primary point of cell-cycle control in C . elegans germ cells , and the G2-to-M transition also responds to growth factors or nutrients in other systems . C . elegans germ cells arrest in G2 during embryonic development ( Fukuyama et al . , 2006 ) and two larval diapause states—L1 diapause ( Fukuyama et al . , 2006 ) and dauer ( Narbonne and Roy , 2006 ) . Our results show that G2 arrest also occurs in starved adults ( Figure 3C ) , and that the length of G2 can vary more than eightfold even under well-fed conditions ( Figure 3—figure supplement 1 ) . At the same time , G1 is very short ( Fox et al . , 2011 , Figure 3—figure supplement 1 ) , and cyclin E/cyclin-dependent kinase 2 , which drives the G1-to-S transition ( Orford and Scadden , 2008 ) , is active throughout the cell cycle ( Fox et al . , 2011 ) . These observations support a model of cell-cycle control in C . elegans germ cells in which regulation in response to extracellular cues relies on the G2-to-M transition . G2 arrest in response to nutrient limitation has also been observed for fission yeast ( Costello et al . , 1986 ) , budding yeast ( Laporte et al . , 2011 ) , and Tetrahymena ( Cameron and Bols , 1975 ) . Moreover , the G2-to-M transition is a point of cell-cycle control in response to poor nutrient conditions in germline stem cells of Drosophila ( Hsu et al . , 2008; LaFever et al . , 2010; Roth et al . , 2012 ) . Even under replete conditions , germline stem cells in Drosophila are thought to be paused in G2 ( Morris and Spradling , 2011 ) , as are a substantial fraction of Drosophila intestinal stem cells ( Zielke et al . , 2016 ) . In mammals , cells can pause in G2 for prolonged periods of time before dividing in response to various stimuli ( e . g . wounding , hormones ) ( reviewed in Gelfant , 1977 ) . Such pausing in mammals has not been the subject of recent investigation , although the G2-to-M transition is known to be regulated by the growth factor IGF-1—for example , in mammalian uterine cells ( Adesanya et al . , 1999 ) , oligodendrocyte progenitors ( Frederick and Wood , 2004; Min et al . , 2012 ) , spermatogonial stem cells ( Wang et al . , 2015 ) , and multiple myeloma cells ( Stromberg et al . , 2006 ) . Proper timing of the G2-to-M transition is also essential during development ( reviewed in Bouldin and Kimelman , 2014 ) , with some populations of cells naturally held in G2 in both Drosophila and zebrafish ( e . g . Bouldin et al . , 2014; Usai and Kimura , 1992 ) . Thus , the control of the G2-to-M transition by growth factors or nutrients may be a conserved feature of the eukaryotic cell cycle . Tissues in adult organisms can be remarkably plastic in their ability to shrink and re-grow in response to changing physiological demands ( e . g . Bergtold , 1926; Secor and Diamond , 1998 ) . Such plasticity requires broad flexibility in a range of cellular behaviors , yet our understanding of tissue plasticity on a cellular level is limited , primarily because tractable models of tissue plasticity are only now being developed ( e . g . O'Brien et al . , 2011 ) . The C . elegans germline presents such a model . This tissue undergoes dramatic shrinkage in adult hermaphrodites starved from the L4 larval stage , and such shrinkage is reversible upon re-feeding ( Angelo and Van Gilst , 2009; Seidel and Kimble , 2011 ) . Previous studies showed that germline shrinkage occurs in part through programmed cell death and oogenesis ( Angelo and Van Gilst , 2009; Seidel and Kimble , 2011 ) . This work establishes facultative stem cell quiescence as a third major force: Because germ cells stop dividing during starvation , cells lost to cell death and oogenesis are not replaced , thus causing the germline tissue to shrink . Quiescence also contributes to re-growth during re-feeding by ensuring that germ cells are able to re-enter the cell cycle rapidly in response to food . Rapid exit from quiescence is a characteristic shared by analogous re-feeding responses in other animals , at least in the handful of examples where such responses have been examined at short time scales—for example , in the ovary of protein-limited Drosophila ( Hartman et al . , 2013; Jouandin et al . , 2014 ) , in the gut of fasted rats , squirrels , and chicks ( Aldewachi et al . , 1975; Cameron and Cleffmann , 1964; Hagemann and Stragand , 1977; Kruman et al . , 1988 ) , and in the retina of yolk-deprived frog embryos ( Love et al . , 2014 ) . These observations stand in contrast to the comparatively longer times required for exit from quiescence in mammalian tissue culture ( Lum et al . , 2005; Pardee , 1974; Soprano , 1994; Zetterberg and Larsson , 1985 ) and suggest that in vivo models of quiescence may uncover new mechanisms of cell-cycle control . N2 , CB4856 ( Hodgkin and Doniach , 1997 ) , TJ1 cep-1 ( gk138 ) I ( Consortium , 2012 ) , TG38 aak-2 ( gt33 ) X ( Lee et al . , 2008 ) , RB754 aak-2 ( ok524 ) X ( Narbonne and Roy , 2006 ) , JK5399 aak-1 ( tm1944 ) III; aak-2 ( ok524 ) X ( Fukuyama et al . , 2012 ) , MR507 aak-2 ( rr48 ) X ( Narbonne and Roy , 2006 ) , JK326 par-4 ( it57ts ) V ( Watts et al . , 2000 ) , GR1310 akt-1 ( mg144gf ) V ( Paradis and Ruvkun , 1998 ) , CF1038 daf-16 ( mu86 ) I ( Lin et al . , 1997 ) , NS3227 daf-18 ( nr2037 ) IV ( Mihaylova et al . , 1999 ) , RB712 daf-18 ( ok480 ) IV ( Fukuyama et al . , 2006 ) , JK5011 daf-3 ( e1376 ) X ( Patterson et al . , 1997 ) , JK4971 daf-5 ( e1386 ) II ( da Graca et al . , 2004 ) , IK130 pkc-1 ( nj3 ) V ( Sieburth et al . , 2007 ) , CB169 unc-31 ( e169 ) IV ( Speese et al . , 2007 ) , KP2018 egl-21 ( n476 ) IV ( Husson et al . , 2007 ) , MT1241 egl-21 ( n611 ) IV ( Husson et al . , 2007 ) , VC461 egl-3 ( gk238 ) V ( Husson et al . , 2006 ) , VC671 egl-3 ( ok979 ) V ( Husson et al . , 2006 ) , JK4963 bbs-5 ( gk507 ) III ( Lee et al . , 2011 ) , JK4962 bbs-5 ( gk537 ) III ( Lee et al . , 2011 ) , JK4960 bbs-8 ( nx77 ) V ( Blacque et al . , 2004 ) , JK4964 bbs-9 ( gk471 ) III ( Chen et al . , 2006 ) , JK4970 tom-1 ( ok285 ) I ( Gracheva et al . , 2006 ) , CB1377 daf-6 ( e1377 ) X ( Perens and Shaham , 2005 ) , BS860 unc-32 ( e189 ) glp-1 ( oz112gf ) /dpy-19 ( e1259 ) glp-1 ( q172 ) III ( Berry et al . , 1997 ) , JK3182 gld-3 ( q730 ) nos-3 ( q650 ) /mIn1[mIs14 dpy-10 ( e128 ) ] II ( Eckmann et al . , 2004 ) , JK5098 fog-1 ( q785 ) I/hT2[qIs48] ( I;III ) ; glp-1 ( q224ts ) III/hT2[qIs48] ( I;III ) ( Morgan et al . , 2010 ) , JK5336 weSi2[Pmex-5::gfp::his-58::tbb-2 3' UTR; Cbr-unc-119 ( + ) ] II; emb-30 ( tn377ts ) III ( Furuta et al . , 2000 ) , JK4605 glp-1 ( q224ts ) III ( Kodoyianni et al . , 1992 ) Unless otherwise noted , worms were maintained on nematode growth media spotted with Escherichia coli OP50 at 20°C . Nematode growth media contained 3 g/L NaCl , 2 . 5 g/L peptone , 20 g/L agar , 25 ml/L 1 M potassium phosphate buffer ( 1 M K2HPO4 mixed with 1 M KH2PO4 to reach a pH of 6 . 0 ) , 1 mM CaCl2 , 1 mM MgSO4 , 5 µg/ml cholesterol , and 2 µg/ml uracil . Worms were synchronized by bleaching gravid hermaphrodites for 5–8 min in a 1:2:12 solution of 5 M NaOH: household bleach: M9 ( 3 g/L KH2PO4 , 6 g/L NaHPO4 , 5 g/L NaCl , and 1 mM MgS04 ) . Embryos were allowed to hatch overnight in M9 in an aerated flask , with shaking at ~170 rpm . L1 larvae were then plated onto 10-cm plates , at a density of ~1000 per plate , and grown to the appropriate developmental stage . We staged animals as ‘early adult’ when hermaphrodites had molted into adulthood and recently begun to ovulate , with most hermaphrodites containing one or two embryos in utero , but with some hermaphrodites containing zero embryos or up to four embryos . Animals with germline tumors ( and therefore no ovulation ) were staged according to the ovulation status of their non-tumorous siblings . At 20°C , early adulthood was reached ~48–52 hr after L1 feeding . For staging of L4 animals , we examined the extent to which the hermaphrodite gonad had migrated from the loop towards the vulva . We define populations as ‘early L4’ when the gonad had migrated ~1/4 of this distance and ‘mid-L4’ when the gonad had migrated ~1/2–3/4 of this distance . Males were staged according their hermaphrodite siblings . Food was removed by gently washing animals from plates with M9 , pelleting animals by spinning at 100–200g for ~1 min , and then washing animals 3–6 additional times with M9 , using 15 ml M9 per wash per 1000–3000 animals . Animals were then deposited onto unseeded 10-cm plates and gently spread across plates such that all liquid was absorbed into the plate within 5 min . Media for starvation plates contained 3 g/L NaCl , 25 g/L agar , 25 ml/L 1 M potassium phosphate buffer ( see above ) , 1 mM CACL2 , 1 mM MgSO4 , and 5 µg/ml cholesterol . This starvation procedure lasted ~10–15 min from start to finish , with time zero being the moment at which M9 was first added to the bacterially seeded plates . Mock food removal was performed in the same manner , except that animals were washed in M9 + ~0 . 5% OP50 and deposited onto bacterially seeded plates . In most experiments , animals were starved at densities of 500–1000 animals per 10-cm plate . Exceptions were starvations beginning from L4 , in which animals were starved at densities of 2000–3000 per 10-cm plate , and experiments requiring that animals be hand-picked , in which animals were starved at densities of <500 per 10-cm plate . Re-feeding was performed by washing animals from starvation plates with M9 + ~0 . 5% OP50 , spinning at 100–200g for ~1 min to pellet animals , and then depositing animals onto 10-cm nematode growth media plates seeded with OP50 . Animals were spread across the plate such that all liquid was absorbed into the plate within 5 min . glp-1 ( q224ts ) and fog-1 ( q785 ) ; glp-1 ( q224ts ) : Animals were grown at 15°C . All plates and M9 solutions used to handle animals were pre-equilibrated to 15°C . Animals were synchronized as described above , but the bleaching protocol was modified , as follows , because glp-1 ( q224 ) embryos are bleach-sensitive . Gravid hermaphrodites were incubated in bleaching solution for ~1 min to kill the adult hermaphrodites but allow their carcasses to remain intact . Embryo-containing carcasses were incubated at 15°C for 4–8 hr . Carcasses were then bleached again , for 3–4 min , to liberate embryos . Hatching of L1s in M9 and was performed at 15°C and extended to 36–40 hr , to account for longer embryonic development times at 15°C . At 15°C , animals reached the early adult stage ~90–96 hr after L1 feeding . par-4 ( it57ts ) : Animals were grown at 15°C until the L3 stage , and then shifted to 25°C until animals reached early adult . The par-4 ( it57ts ) starvation time course was performed at 25°C . aak-1 ( tm1944 ) ; aak-2 ( ok524 ) : Animals were grown at 15°C until the early L4 stage , and then transferred to 20°C until animals reached early adult . Growth at 15°C was used because aak-1 ( tm1944 ) ; aak-2 ( ok524 ) animals grown at 20°C showed high sterility . Additionally , an alternate synchronization protocol was used because hatching of aak-1 ( tm1944 ) ; aak-2 ( ok524 ) L1s in M9 causes sterility ( Fukuyama et al . , 2012 ) . Early adults were bleached according to the protocol described for glp-1 ( q224ts ) , and embryos were deposited directly onto food . emb-30 ( tn377ts ) : Animals were grown at 15˚C prior to temperature shifts . Gonads were dissected in M9 + 0 . 1% Tween-20 + 0 . 25 mM levamisole . For GLD-1 and phospho-histone H3 staining , gonads were fixed in PBS + 3% paraformaldehyde + 0 . 1% Tween-20 ( PBSTween ) for 30 min , followed by −20°C methanol for 15 min ( anti-GLD-1 and anti-HIM-3 ) or ≥15 min ( anti-phospho-histone H3 ) . For phospho-CDK-1 staining , gonads were fixed in PBSTween + 3 . 7% formaldehyde for 10 min , followed by −20°C methanol for 5 min . Gonads were blocked for 30 min at room temperature in PBSTween + 3–5% normal donkey serum ( anti-GLD-1 and anti-phospho-CDK-1 ) or 1–3% bovine serum albumin ( anti-HIM-3 and phospho-histone H3 ) . Incubations with primary antibodies were performed overnight at 4°C , with antibodies diluted in blocking solution . Dilutions were as follows: mouse anti-phospho-histone H3 ( Cell Signaling Technology , Danvers , MA , #9706 ) , 1/150; rabbit anti-GLD-1 ( Cinquin et al . , 2010 ) , 1/100; rabbit anti-HIM-3 ( Novus Biologicals , Littleton , CO , #53470002 ) , 1/200; rabbit anti-phospho-CDK-1 Thr14/Tyr15 ( Santa Cruz , Dallas , Texas , #sc-28435-R ) ( Rahman et al . , 2014 ) , 1/200; rabbit anti-phospho-CDK-1 Tyr15 ( Calbiochem , San Diego , CA , #219440 ) ( Hachet et al . , 2007 ) , 1/100 . Incubations with secondary antibodies were performed for 1–2 hr at room temperature , using Cy-3 donkey anti-mouse ( Jackson ImmunoResearch , Westgrove , PA , #715-165-151 ) or Cy-3 donkey anti-rabbit ( Jackson ImmunoResearch #711-165-152 ) , diluted 1/1000 . Gonads were mounted in Vectashield containing DAPI ( Vector Labs , Burlingame , CA , #H-1200 ) . For DAPI staining in the absence of antibody staining , gonads were fixed as for phospho-histone H3 staining , and then mounted in Vectashield containing DAPI . To test the phospho-specificity of anti-phospho-CDK-1 , gonads were dissected and fixed , as above . After fixation , gonads were treated with 20 U/µl Lambda protein phosphatase ( NEB , Ipswich , MA , #P0753S ) in protein metallophosphatase buffer ( 50 mM HEPES , pH 7 . 5 , 100 mM NaCl , 2 mM DTT , 0 . 01% Brij 35 , and 1 mM MnCl2 ) for 1 hr at 30°C Control gonads were treated the same , but Lambda protein phosphatase was omitted from the reaction . Gonads were blocked and stained with anti-phospho-CDK-1 as above . Unless otherwise noted , images were obtained on a Leica SP8 . In all experiments , identical imaging conditions and brightness adjustments were used across samples . Unless otherwise noted , M-phase cells were scored by examining gonads for phospho-histone H3+ cells at 63× magnification . In all experiments , a subset of gonads was examined via DAPI staining to confirm the correspondence between phospho-histone H3+ cells and mitotic figures . In germline tumors , M-phase cells occurring in the distal-most 20 rows of germ cells ( i . e . the region corresponding to the progenitor zone in wildtype gonads ) were excluded from analysis . Additionally , we excluded germline tumors with patches of differentiation ( as assessed by DAPI staining ) , which sometimes occurred in glp-1 ( oz112gf ) animals . EdU labeling was performed by soaking or by feeding . Soaking was used for starvation time courses ( Figure 3A ) and for testing whether G1 versus G2 cells could be distinguished by nuclear size ( Figure 3—figure supplement 2 ) . Feeding was used for measuring cell-cycle length in fed animals ( Figure 3—figure supplement 1 ) . For the soaking procedure , animals were incubated with rocking in M9 + 0 . 1% Tween-20 + 1 mM EdU for 15 min at room temperature . Gonads were dissected as for antibody staining and fixed in 3% paraformaldehyde in PBSTween for 30 min , followed by −20°C methanol for ≥15 min . Gonads were blocked in PBSTween + 3% bovine serum albumin for 30 min at room temperature . Click-iT reactions were performed using the Click-iT EdU Alexa Fluor 488 Imaging Kit ( Invitrogen , Carlsbad , CA , #C10337 ) , according to the manufacturer’s instructions , except that two back-to-back half reactions of 250 µl volume were performed . Gonads were mounted in Vectashield containing DAPI . EdU labeling by feeding was performed similar to previous studies ( Crittenden et al . , 2006; Fox et al . , 2011; Morgan et al . , 2010 ) . E . coli strain MG1693 was grown overnight at 37°C in M9 minimal media ( 3 g/L KH2PO4 , 6 g/L Na2HPO4 , 0 . 5 g/L NaCl , 1 g/L NH4Cl , 2 mM MgSO4 , 0 . 1 mM CaCl2 , 0 . 4% glucose , and 1 µg/ml thiamin ) supplemented with 5 µg/ml thymine . This culture was diluted 1:50 in M9 minimal media supplemented with 0 . 5 µM thymidine and 20 µM EdU and grown for 32 hr at 37°C . Cells were re-suspended in ~1/100th of their original volume in M9 , and then spread onto 6-cm plates , using 100 µl of E . coli solution per plate . Plate media was identical to standard nematode growth media except that 60 µg/ml carbenicillin was added , peptone was omitted , and agar was exchanged for 12 g/L agar + 6 g/L agarose . Plates were seeded 1 day prior to adding worms . Worms were transferred to plates for the required period of time , and then gonads were dissected and processed as above . EdU-labeled gonads were stained with propidium iodide to measure DNA content and to allow for simultaneous imaging of Alexa Fluor 488 and DNA . Propidium iodide staining was performed by adding two steps to the aforementioned EdU-labeling protocol . First , prior to the blocking step , gonads were incubated in PBSTween + 20 µg/ml RNase A for 1 hr at 37°C . Second , after the Click-iT reaction , gonads were incubated for 30 min at room temperature in PBSTween + 50 µg/ml propidium iodide . To quantify propidium iodide staining , gonads were imaged at 63× magnification on a Zeiss LSM510 laser-scanning confocal microscope , with a z-stack interval of 0 . 37–0 . 39 µm . Pixel intensity per nucleus was calculated as the summation of all pixel intensities within a best-fit cylinder whose height matched the height of the focal nucleus in the z-dimension and whose cross-sectional diameter matched the largest dimension of the focal nucleus in the x–y dimension . Cylinders were defined manually by drawing circles around nuclei in ImageJ . This quantification method is undeniably crude because cylinders often included portions of neighboring nuclei . Nevertheless , this method allowed us to distinguish non-S-phase cells as having a DNA content less than S-phase cells ( G1 ) or a DNA content greater than S-phase cells ( G2 ) ( Figure 3—figure supplement 2 ) . Cells were classified as G1 , S-phase , or G2 by a combination of EdU labeling ( to mark S-phase cells ) and nuclear size . This method has not been used previously—although others have noted a correlation between cell-cycle stage and nuclei size ( Chiang et al . , 2015; Fukuyama et al . , 2006; Lawrence et al . , 2015 ) —and we justify its use here . In pilot experiments involving EdU labeling and DNA quantification , we noticed a correlation between cell-cycle stage and nuclear size: G1 and early S-phase nuclei were small , G2 and late S-phase nuclei were large , and mid S-phase nuclei were intermediate in size ( Figure 3—figure supplement 2 ) . Additionally , G1 nuclei nearly always occurred in pairs , consistent with G1 being very short ( Fox et al . , 2011 ) . The size difference between G1 and G2 nuclei was large enough that in EdU-labeled gonads , G1 and G2 nuclei could be distinguished by eye ( n > 100 nuclei , from a total of seven progenitor zones ) . To test the accuracy of this method more thoroughly , we obtained z-stack images of EdU-labeled , propidium iodide-stained progenitor zones from three fed early adult hermaphrodites and three hermaphrodites starved from early adult for 3 . 5 hr . We first classified each cell as S-phase ( EdU+ ) , G1 ( EdU− and having a nuclear size equal to or smaller than the smallest EdU+ cells ) , or G2 ( EdU− and having a nuclear size equal to or larger than the largest EdU+ cells ) . We then quantified propidium iodide staining ( a measure of DNA content ) and compared our ‘by size’ classification to the classification given by propidium iodide staining ( Figure 3—figure supplement 2 ) . For all cells in all six progenitor zones , the two classification systems matched perfectly ( Figure 3—figure supplement 2 ) . We therefore used the ‘by size’ classification system for counting G1 , S-phase , and G2 cells throughout . EdU-labeled progenitor zones were imaged at 63× magnification with a z-stack interval of 1 µm . Cells were counted as belonging to the progenitor zone if they had an interphase or M-phase chromosome morphology and if their midpoint was located distal to a cross-sectional line drawn through the midpoint of the second most distal ‘crescent’ cell ( i . e . the second most distal meiotic prophase cell ) . Progenitor-zone cells were classified as G1 ( EdU− and nuclear size equal to or smaller than the smallest EdU+ cells ) , S-phase ( EdU+ ) , G2 ( EdU− and nuclear size equal to or larger than the largest EdU+ cells ) , or M-phase ( mitotic figures ) . Classifications were recorded using the Cell Counter plug-in for ImageJ ( http://rsb . info . nih . gov/ij/plugins/cell-counter . html ) , by marking each cell in all z-slices in which it was observed . A custom R script was then used to identify marks belonging to the same cell . Cell-cycle parameters in fed animals were determined by first measuring the length of G2 . G2 was measured by labeling animals with EdU ( via feeding ) and calculating the fraction of M-phase cells ( mitotic figures ) that were EdU+ over time ( equation 1 ) . Next , the length of G2 was combined with the G2 index to calculate the total length of the cell cycle ( equation 2 ) . The lengths of G1 , S-phase , and M-phase were calculated by multiplying the total length of the cell cycle by the G1 , S-phase , or M-phase indices ( equations 3–5 ) . Calculations of the maximum total length of the cell cycle depend on assumptions about covariance between the length of G2 and the length of M + G1 + S . If cells having a longer G2 are assumed to have a proportionally longer M + G1 + S , then the maximum length of the cell cycle for 95% or 100% of cells is given by equation 6 . If cells having a longer G2 are not assumed to have a proportionally longer M + G1 + S , then the maximum length of the cell cycle for 95% or 100% of cells is given by equation 7 . We performed both calculations . Cell-cycle length in fed animals was measured in early adult hermaphrodites and in adult hermaphrodites aged 24 hr post mid-L4 . For calculations involving early adults , indices for each cell-cycle phase were derived from the 0-hr time point of the time course in Figure 3 . For calculations involving animals aged 24 hr post mid-L4 , indices were derived from the 0 . 5-hr time point of the time course in Figure 3—figure supplement 1 , for consistency with Fox et al . ( 2011 ) . Gonads were stained with DAPI and examined at 63× magnification . In a central focal plane , one edge of the gonad was chosen at random , and the distal-most ‘crescent’ germ cell along that edge was identified . The number of germ cells ( along the gonad edge ) between this first ‘crescent’ cell and the distal tip of the gonad was counted . L4 hermaphrodites close to the adult molt were transferred to an unseeded plate and exposed to 100 J/m2 of 254 nm UV light in Spectrolinker XL-1000 UV Crosslinker . Animals were immediately returned to food and incubated for 8 hr at 20°C before dissection . Animals were adults at the time of dissection . Serotonin creatinine sulfate was dissolved in M9 to a concentration of 50 mg/ml . This solution was spread onto starvation plates to a final concentration of 20 mM serotonin . Plates were incubated for at least 1 hr before worms were added . To expose animals to serotonin at the onset of starvation , animals were deposited directly onto serotonin plates after food removal . To expose animals to serotonin after 3 hr of starvation , animals were starved for 3 hr on standard starvation plates and then were transferred to serotonin plates via washing in M9 + 0 . 01% Tween-20 . RNAi was performed by feeding . A cdk-1 RNAi clone and the empty RNAi vector ( L440 ) were obtained from the Ahringer library ( Kamath and Ahringer , 2003 ) and grown overnight at 37°C in liquid Luria Broth + 60 µg/ml carbenicillin + 10 µg/ml tetracycline . Cells were concentrated fivefold , and then spotted onto plates containing nematode growth media supplemented with 60 µg/ml carbenicillin , 10 µg/ml tetracycline , and 1 mM IPTG . Plates were spotted 1 day before adding worms . Temperature shifts were performed by transferring plates of worms from 15°C to 25°C or the reverse . To expedite equilibration of plates to the new temperature , plates were buried in a single layer within a 5 × 7 × 13 inch box full of unseeded plates pre-equilibrated at the new temperature . For temperature shifts of fed glp-1 ( q224ts ) animals , fed early adult hermaphrodites were shifted to 25°C . Gonads were dissected every 30–60 min after the shift ( for Figure 6H ) or after 8 hr ( for Figure 6A , I ) . Alternately , animals were maintained at 25°C for 8 hr , then returned to 15°C for 2–3 days ( Figure 6A ) . For temperature shifts of cdk-1 RNAi-treated animals ( Figure 6B and Figure 6—figure supplement 1 ) , glp-1 ( q224ts ) animals were grown at 15°C ( on OP50 ) to mid L4 , and then transferred to cdk-1 RNAi plates or L440 plates for 42 hr hours . Plates were then shifted to 25°C for 8 hr , and then returned to 15°C for 2–3 days . At the beginning of the temperature shift , animals were adults , and germ cells in the progenitor zone had uniformly arrested in interphase , as evidenced by the absence of mitotic figures . Some nuclei in the progenitor zone had slightly enlarged , characteristic of cdk-1 RNAi-induced cell-cycle arrest ( Jeong et al . , 2011 ) . Nuclear morphology was otherwise normal ( Figure 6—figure supplement 1 ) . For temperature shifts of starved glp-1 ( q224ts ) animals , hermaphrodites were starved at 15°C from early adult for 2–4 hr or from mid L4 for 24 hr . For Figure 6C–D , starved animals were shifted to 25°C for 8 , 16 , or 24 hr , returned to 15°C for 8–12 hr , then re-fed at 15°C for 2–3 days . At the beginning of the temperature shift , animals starved from L4 were adults . For Figure 6G–H , animals were shifted to 25°C for 8 hr , and then re-fed at 25°C . For Figure 6I , animals were shifted to 25°C for 8 hr . For temperature shifts of glp-1 ( q224ts ) animals exposed to high NaCl ( Figure 6E ) , early adult hermaphrodites were transferred to plates containing high NaCl media and incubated at 15°C for 2 hr . Animals were then shifted to 25°C for 8 hr , returned to 15°C for 12 hr , and then transferred to plates containing standard media for 2–3 days . High NaCl media were identical to standard media , except that it contained a total of 17 . 4 g/L ( 300 mM ) NaCl . For temperature shifts of ( unmated ) fog-1 ( q785 ) ;glp-1 ( q224ts ) animals ( Figure 6F ) , hermaphrodites aged 40 hr post mid L4 were shifted to 25°C for 8 hr , and then returned to 15°C for 2–3 days . Animals aged 40-hr post mid L4 were used because the reduced mitotic index caused by the absence of sperm is not fully evident at the early adult stage ( 40-hr post mid L4 at 15°C is roughly equivalent to 24-hr post mid-L4 at 20°C ) . For temperature shifts of mated fog-1 ( q785 ) ;glp-1 ( q224ts ) animals , fog-1 ( q785 ) ;glp-1 ( q224ts ) mid L4 hermaphrodites were transferred to plates with CB4856 males at a ratio of 1:2 . After 40 hr of mating , hermaphrodites wearing copulatory plugs were transferred to fresh plates at 15°C , and then shifted to 25°C for 8 hr . Temperature shifts of emb-30 ( tn377ts ) animals are described below . emb-30 ( tn377ts ) animals were grown at 15°C to mid L4 , and then starved at 15°C for 24 hr , by which time animals were adults . Animals were then shifted to 25°C for 14 hr and either re-fed or maintained in starvation . Gonads were dissected immediately before re-feeding and at 1-hr intervals 2–6 hr after re-feeding . Gonads were also dissected after 6 hr of continued starvation . Gonads were stained for phospho-histone H3 and GLD-1 and imaged at 63× magnification with a z-stack interval of 1 µm . Cells were identified using IRISES ( Vogel et al . , 2014 ) , followed by manual correction using the Cell Counter plug-in in Image J . Cells were classified as interphase , M-phase , or ‘crescent’ ( i . e . meiotic prophase ) according to chromosome morphology and phospho-histone H3 staining . Occasionally , at the later re-feeding time points , cells were observed bypassing the metaphase arrest; for the purposes of this experiment , such cells were classified as M-phase . Cells distal to the ‘crescent’/‘non-crescent’ boundary were defined as cells whose midpoints were located distal to a cross-sectional line drawn through the midpoint of the second most distal ‘crescent’ cell . Relative positions of cells along the distal-to-proximal axis were determined by collapsing cell positions along the z-axis and fitting these positions to a second-degree polynomial curve . Positions along this polynomial curve closest to each germ cell and closest to the distal tip cell were identified by solving , for each cell , the polynomial whose roots minimize this distance . Using this new set of positions , distances between the distal tip cell and each germ cell were calculated . Cells were then ranked according to these distances . In experiments requiring image acquisition , an attempt was made to examine at least 20 gonads . In other experiments , an attempt was made to examine at least 50 gonads . Plots were generated in part using the ggplot package for R ( ggplot2 . org ) .	Adult stem cells can divide to produce cells that can develop into one of many different specialist cell types in a tissue , and so are vitally important for tissue repair and maintenance . Some types of adult stem cells exist primarily in a non-dividing state known as quiescence , which for a long time was thought to be essential for maintaining the stem cell state . However , researchers have discovered some adult stem cells that are either not quiescent , or only enter this state rarely . Until now , biologists have lacked an experimental model in which the role of quiescence in maintaining stem cells can be easily investigated . Seidel and Kimble have now investigated the role of quiescence in the germline stem cells – which give rise to egg and sperm cells – of the roundworm Caenorhabditis elegans . The results of the study revealed that although the germline stem cells divide continuously when the worms are well fed , starving the worms causes these stem cells to become quiescent . Maintaining C . elegans germline stem cells in a stem cell state normally involves a process called Notch signaling , which cells use to communicate with each other . However , Seidel and Kimble found that the germline quiescence caused by starvation maintains the stem cell state even when Notch signaling is prevented . This suggests that , in the absence of food , quiescence alone can maintain germline stem cells , although how it does so remains a question for future work . One possibility is that quiescence stabilizes other molecules involved in the Notch signaling pathway or prevents the production of proteins that enable a stem cell to develop into a specialized cell .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "stem", "cells", "and", "regenerative", "medicine", "cell", "biology" ]	2015	Cell-cycle quiescence maintains Caenorhabditis elegans germline stem cells independent of GLP-1/Notch
Dysfunction of the basal ganglia produces severe deficits in the timing , initiation , and vigor of movement . These diverse impairments suggest a control system gone awry . In engineered systems , feedback is critical for control . By contrast , models of the basal ganglia highlight feedforward circuitry and ignore intrinsic feedback circuits . In this study , we show that feedback via axon collaterals of substantia nigra projection neurons control the gain of the basal ganglia output . Through a combination of physiology , optogenetics , anatomy , and circuit mapping , we elaborate a general circuit mechanism for gain control in a microcircuit lacking interneurons . Our data suggest that diverse tonic firing rates , weak unitary connections and a spatially diffuse collateral circuit with distinct topography and kinetics from feedforward input is sufficient to implement divisive feedback inhibition . The importance of feedback for engineered systems implies that the intranigral microcircuit , despite its absence from canonical models , could be essential to basal ganglia function . The basal ganglia are a collection of interconnected subcortical regions of the vertebrate brain ( DeLong , 2000 ) . Pathological disruptions of basal ganglia signaling produce profound deficits in the timing ( Buhusi and Meck , 2005 ) , vigor ( Turner and Desmurget , 2010 ) , and initiation ( Mink , 1996 ) of voluntary movements . While it is thus clear that the basal ganglia are critical for voluntary movement , the specific mechanisms by which movement is controlled by basal ganglia activity remain unclear ( Turner and Desmurget , 2010 ) . Voluntary control of movement can be explained in terms of optimal feedback control theory ( Diedrichsen et al . , 2010 ) . The basal ganglia circuit can be described as an extended loop that begins with projections from deep layer cortical neurons and ultimately returns to the cortex via projections from the basal ganglia to the ventral thalamus ( Haber , 2003 ) . However , the basal ganglia circuit also contains intrinsic feedback projections ( Gerfen , 2004 ) . In engineered control systems , feedback is critical for stable performance ( Astrom and Murray , 2008 ) . The substantia nigra ( SN ) pars reticulata ( SNr ) is the primary source of output from the sensorimotor basal ganglia in rodents ( Gerfen , 2004 ) . The vast majority of neurons in the SNr are projection neurons that synthesize and release the neurotransmitter ϒ-aminobutryic acid ( GABA ) . Projection neurons of the SNr target pre-motor areas in the ventral thalamus , dorsal midbrain , and tegmentum ( Parent , 1990; Hikosaka , 2007 ) . In addition to these long range targets , nigral projection neurons also elaborate axon collaterals within the SN ( Mailly et al . , 2003; Cebrián et al . , 2005; Deniau et al . , 2007a ) . There are no known interneurons in the SNr ( Deniau et al . , 2007a ) , and thus collaterals of projection neurons are the sole source of intrinsic feedback for the basal ganglia output . Anatomical reconstructions have indicated that the axon collaterals of SNr projection neurons are sparse ( Mailly et al . , 2003 ) . The functional impact of this intranigral microcircuit remains unclear . Antidromic activation of SNr projection neurons in anesthetized animals has been used to infer the presence of inhibition via projection neurons collaterals ( Tepper and Lee , 2007; Brazhnik et al . , 2008 ) ; however , the relative impact , spatial organization and temporal properties of recruitment of feedback inhibition via SNr collateral inhibition remains largely unknown . In this study , we explore the hypothesis that the microcircuit formed by SNr collaterals could implement a critical negative feedback node in the context of a control system for voluntary behavior that is implemented in the extended cortico-basal ganglia circuit . In engineered systems , the functional impact of a negative feedback can be difficult to detect and characterize ( Astrom and Murray , 2008 ) . For example , at steady state or in the absence of change in the state of the system , there may be no obvious effect of appropriately functioning negative feedback . However , sudden transitions in the state of the system can reveal the contribution of negative feedback—altering , for example , the gain and/or the time course of settling around transitions . By analogy to an engineered system , the SNr microcircuit may have little apparent impact in the absence of sudden changes in the state of the population activity . However , changes in behavior and receipt of sensory stimuli are accompanied by phasic transitions , both increases and decreases , in the activity of the SNr population . We thus reasoned that the role of negative feedback , implemented by the SNr microcircuit , could become apparent under such conditions . Recent work has shown that a salient or conditioned stimulus ( CS ) , for example an auditory tone , can lead to phasic changes in the activity of SNr neurons in rodents ( Schmidt et al . , 2013 ) . Moreover , these short-latency modulations of activity are predictive of the initiation of conditioned behavioral responses—that is action selection ( Pan et al . , 2013; Schmidt et al . , 2013 ) . If the basal ganglia act as a control system for behavior , then we would predict that control over the gain or dynamic range of these phasic modulations should be critical for normal voluntary actions . Detailed study of the local inhibitory connections within the SNr has been hampered by the difficulty in isolating and specifically exciting the SNr collaterals independent of afferent inhibitory and excitatory projections ( Hammond et al . , 1983 ) . We overcame this challenge by using cell-type specific expression of channelrhodopsin-2 ( ChR2 ) , a light-gated cation channel ( Boyden et al . , 2005; Zhang et al . , 2006 ) , in SNr GABA neurons . This optogenetic approach allowed us to stimulate SNr GABA neurons with high temporal and spatial resolution without contamination from excitatory afferents , dopaminergic transmission , or afferent inhibitory input from the striatonigral projection both in vitro and in vivo . Consistent with the prediction from anatomical data , we show that inhibition derived from the collaterals of projection neurons in the SNr is sparse and has little to no effect on tonic baseline firing . However , we also observed that activation of the SNr projection neuron population could elicit a large and potent feedback inhibition capable of shaping output activity . Here , we show that this unique combination of effects is the result of a number of distinctive features of the SN microcircuit: ( 1 ) postsynaptic currents resulting from collateral synaptic input provided robust inhibition with a rapid onset during strong activation of the network; ( 2 ) unitary connections are weak with sufficiently low release probability to sustain release during repetitive stimulation; ( 3 ) asynchronous basal inhibition in the tonically active network is effectively compensated for by intrinsic conductances that sustain tonic spiking; ( 4 ) the potency of feedback inhibition is proportional to total activation of the microcircuit due to a sparse , spatially diffuse connectivity . Together , these properties of the intrinsic microcircuit of the SNr implement a robust inhibition that is rapidly and stably recruited in proportion to the sustained activation of the projection neuron population with little effect in the absence of stimulation—in other words , the inhibitory microcircuit of the SNr mediates a divisive gain control on the basal ganglia output . Local axon collaterals of projection neurons provide a source of feedback inhibition proportional to the output of the SNr . For this inhibition to regulate the output of the SNr it must be sufficient to suppress activity even in the presence of strong , phasic activation of the projection neurons . Phasic activation of the SNr population occurs , for example , at the onset of salient sensory cues ( Pan et al . , 2013; Schmidt et al . , 2013; Figure 3—figure supplement 1 ) . Thus , to determine whether local inhibition was sufficient to regulate the gain of the SNr network , we used ChR2 stimulation to drive repetitive somatic spiking in the projection neuron network . This recruits a population of SNr neurons with a time course and distribution of responses similar to that evoked by conditioned stimuli ( Pan et al . , 2013; Figure 3—figure supplement 1 ) . We determined the consequences of local inhibition by comparing activity evoked when inhibition was intact with activity evoked following pharmacological blockade of inhibition . Whole cell current-clamp recordings from individual SNr projection neurons were obtained from brain slices of Thy1-ChR2 mice in the presence of excitatory synaptic transmission blockers ( D-AP5 and NBQX; Figure 3A ) . Wide-field illumination through a 10X objective was used to stimulate activity throughout the SNr network . Direct light-evoked spiking in the recorded neuron was substantially , or in some cases completely , suppressed under control conditions ( Figure 3B , C ) . However , reliable light-evoked spiking was always present following application of the GABAA receptor antagonist gabazine ( Gbz ) to block local inhibition ( Figure 3D ) . The suppression of evoked spiking was consistent across stimulus durations within a cell ( Figure 3E ) , whereas the magnitude of suppression was more idiosyncratic across cells for a given stimulus condition ( Figure 3F ) . 10 . 7554/eLife . 02397 . 005Figure 3 . The local inhibitory microcircuit of the SNr provides feedback gain control . ( A ) Schematic of the experimental configuration . 1-2 SNr GABA neurons were recorded from in the whole-cell current clamp configuration . Wide-field illumination of the slice ( indicated by cyan circle ) was used to photostimulate the SNr network . ( B ) Example recording from an individual SNr GABA neuron during light stimulation ( upper cyan trace ) . Note the stereotyped membrane potential fluctuations during photostimulation ( 10 trials ) . ( C and D ) Example recordings from the same neuron recorded during 10 trials of stimulation ( ‘Stim’; upper cyan trace ) under control conditions ( C; Cntrl; black ) and following pharmacological blockage of inhibition via gabazine application ( D; +Gbz; red ) aligned to stimulus onset . Tick marks indicate spike times for 10 repetitions of the same light stimulus . Lower traces show the intracellular recording from the same neuron overlaid for all trials . ( E ) Raster plot of evoked spiking in control conditions ( left ) and in the presence of Gbz ( right ) for 4 blocks of 10 trails of increasing stimulus durations ( 4 ms , 8 ms , 12 ms , 20 ms; top to bottom ) for a single neuron . ( F ) Raster plots of evoked spiking for 10 trials aligned to the onset of an 8 ms light stimulus for the population of neurons under control conditions ( left ) and in the presence of Gbz ( right ) . ( G ) Normalized response across the population of neurons binned by stimulus duration and grouped by treatment ( black , Cntrl and red , +Gbz ) . Significant effects on both stimulus duration and treatment condition were observed ( Two-way ANOVA , p<0 . 05 ) . Zero stimulus responses ( open symbols ) were estimated from the background firing rate . No significant difference was observed . ( H ) Full width half maximum ( FWHM ) of the peristimulus time histogram ( PSTH ) for evoked spiking in control and following Gbz ( p<0 . 01 ) . ( I ) For paired recordings , the percent inhibition of one neuron in the pair was plotted as a function of the percent inhibition of the other neuron for all stimulus conditions ( black circles ) . A significant positive correlation was found and indicated by the solid black line ( p<0 . 01; two tailed t test ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 00510 . 7554/eLife . 02397 . 006Figure 3—figure supplement 1 . Direct comparison of responses elicited by optogenetic and natural stimulation . Peristimulus time histograms ( PSTH ) for a population of single units recorded from the SN of Thy1-ChR2 mice . ( A ) PSTHs were calculated for responses to optogenetic ( LIGHT ) stimulation through an optical fiber associated with a microwire array and ( B ) a 500 ms auditory stimulus ( 10 kHz pure tone , 500 ms duration ) presented from a speaker positioned in front of the head-fixed animal ( TONE ) . In both cases a strong transient response was observed and is evident in the mean PSTH ( C and D ) . At the lower right panel the mean PSTH for TONE stimulation ( filled black bars ) and a scaled mean PSTH for LIGHT stimulation ( red line ) are overlaid ( D ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 00610 . 7554/eLife . 02397 . 007Figure 3—figure supplement 2 . mIPSC amplitude in SNr GABA neurons . ( A ) Whole-cell recording from SNr neurons in control conditions ( Cntrl ) and following substitution of aCSF Ca2+ with 2 mM Sr2+ ( +Sr2+; Vh 30 mV ) . IPSCs were evoked via photostimulation of SNr collaterals ( cyan ) in slices from Gad2-ChR2 mice . Substitution of Ca2+ for Sr2+ was used to desynchronize release from SNr collateral synapses . ( B and C ) No difference between control sIPSCs and those enriched for SNr collaterals via desynchronized release was observed . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 007 The ability of feedback inhibition to suppress spiking more effectively with increasing stimulus duration ( Figure 3E ) implies a divisive gain control . To quantify the gain effect across the population , we compared the normalized response to photostimulation of increasing duration both in the presence and absence of inhibitory synaptic transmission ( Figure 3G ) . We found that the response of the population showed a significant increase as a function of stimulus duration and the magnitude of the increase was significantly reduced by the presence of feedback inhibition ( two-factor ANOVA; p<0 . 05 ) . Divisive gain control is characterized by a suppression of spiking at large stimuli but little to no effect on the response to weak or absent stimuli . Consistent with a divisive gain control , we found that baseline firing was unaffected by removal of inhibition ( Figure 3G , open circles ) . To contrast a feedback gain control with the effect predicted for subtractive inhibition , we examined recordings from dopamine neurons of the SN that do not express ChR2 , but are strongly inhibited by ChR2 expressing projection neurons ( Pan et al . , 2013 ) . For dopamine neurons , we observed a constant suppression of spiking across the range of stimuli used ( Pan et al . , 2013 ) . If the reduction in spiking observed during stimulation was the result of feedback inhibition one would predict that the inhibition should onset after the onset of the population response and truncate the response present in the absence of feedback . Consistent with this prediction we found that suppression of spiking was characterized by a significant decrease in the duration of the evoked spiking ( Figure 3H ) . This effect on the duration could be observed in many individual neuron responses ( Figure 3F ) . Our in vivo results suggested that the extent of suppression of transient activation in SNr neurons is proportional to the estimated activation of the network ( Figure 1 ) . Anatomical studies indicate that the vast majority of projection neurons elaborate collaterals within the SNr , however , these collaterals can form relatively few ( ∼10 ) putative synaptic contacts ( Mailly et al . , 2003 ) . Moreover , we found that unitary release events produced relatively weak mIPSCs ( ∼150 pS ) ( Figure 3—figure supplement 2 ) . Taken together , these data imply that the inhibition observed results from the activation of approximately 50–100 presynaptic inputs . Given that the GABAergic neurons in the SNr are thought to be exclusively projection neurons ( Deniau et al . , 2007b ) , this is consistent with the finding that projection neuron collaterals form 79 . 4 ± 96 . 1 ( SD ) boutons per neuron within the SNr of the rat ( Mailly et al . , 2003 ) . Assuming a modest or low probability of paired connections , our connectivity estimates imply that the extent of activation across a large population of SNr neurons would determine the extent of feedback inhibition consistent with our observation in vivo . While we do not have a direct measure of the total extent of activation of the SNr by our photostimulation , we note that neurons recorded simultaneously experience the same activation state of the network . Thus , we reasoned that the extent of feedback inhibition in a pair of recorded neurons should be correlated if feedback inhibition is proportional to the total activation of the network . Consistent with this prediction , we found that there was a significant correlation ( Pearson's correlation , p<0 . 01 permutation test ) in paired recordings ( Figure 3I ) . These results suggest that a given SNr projection neuron receives input from a spatially diffuse collection of other SNr projection neurons . The data above are consistent with the claim that a large population of SNr projection neurons must be recruited to fire within a relatively small time window ( 5–20 ms ) in order to achieve robust feedback inhibition and divisive gain effects ( Figure 3G ) . These results were obtained in the Thy1-ChR2 mouse where all neurons in the SNr express ChR2 ( Figure 2 ) . This would suggest that if a local subset of the SNr was expressing ChR2 , the divisive gain effect should be present , but reduced in magnitude analogous to the smaller effects observed when less of the network was recruited in the Thy1-ChR2 preparation ( Figure 3G ) . Indeed , we found that when the same experiment was repeated in ChR2+ nigral neurons from virally infected Gad2-ChR2 mice a divisive gain effect was observed , but reduced in magnitude ( p<0 . 05; two-factor ANOVA; 15% reduction in the saturated response ) . To alter the gain of a response to activation of the network requires a change in the slope of the curve . As described above , we observed that there was a substantial effect of feedback inhibition in the strongly activated SNr circuit , but no effect in the absence of stimulation—resulting in a change in the slope of the response to stimulation . However , it is confusing how a strongly coupled inhibitory network of tonically active neurons could exhibit no effect of feedback even in the absence of stimulation . We first asked whether the rate of spontaneous IPSCs was consistent with our estimate , and a prior anatomical estimate ( Mailly et al . , 2003 ) , of >50 inputs from other SNr projection neurons . The expected rate of spontaneous IPSCs would thus be approximately: ( 1 ) RuIPSCs=Npre×Rpre×Preleasewhere , RuIPSCs is the predicted rate of unitary IPSCs ( uIPSCs ) , Npre is the number of presynaptic inputs ( release sites ) , Rpre is the mean firing rate of presynaptic inputs , and Prelease is the effective release probability across all release sites . Thus , with a relatively low release probability ( <0 . 5 ) , we would predict 75–300 Hz of uIPSCs . This corresponds well to the rate of uIPSCs estimated directly from voltage clamp recordings ( Figure 4A–C ) . Consistent with this estimate , we also show that repetitive stimulation of SNr collaterals fails to completely depress transmission ( Figure 4—figure supplement 2 ) consistent with a vesicular release probability low enough to allow vesicle recycling to keep pace with release . Such a mechanism has been described in detail for Purkinje cell synapses ( Telgkamp et al . , 2004 ) . These observations suggest that there is indeed a substantial background rate of IPSCs that , when pharmacologically blocked , has no significant effect on the tonic firing of SNr projection neurons ( Atherton and Bevan , 2005 ) . 10 . 7554/eLife . 02397 . 008Figure 4 . High background inhibition has little affect on tonic activity of SNr neurons . ( A ) Whole-cell recording of spontaneous IPSCs ( sIPSCs ) onto SNr neurons in control conditions ( Cntrl; black trace ) and following addition of tetrodotoxin ( TTX ) to isolate miniature events ( +TTX; red trace , Vh 0 mV ) . ( B ) Cumulative histogram of IPSC amplitude in control conditions and following addition of TTX ( n = 4 cells ) . ( C ) Box and whisker plot of IPSC amplitude for control and following addition of TTX . ( D ) Spiking output of SNr neurons following addition of high background excitation ( upper ) or inhibition ( lower ) via the dynamic clamp . ( E ) Summary data of change in firing rate of SNr neurons ( n = 11 cells ) following an increasing the relative frequency of inhibitory ( red ) or excitatory conductances ( blue ) . ( F ) The slope of the change in firing rate as a function of change in conductance was significantly greater following increases in excitatory conductance compared to inhibitory conductance . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 00810 . 7554/eLife . 02397 . 009Figure 4—figure supplement 1 . Intrinsic , net inward currents and a positive slope conductance allows feedback gain control of SNr neurons . ( A ) Whole-cell recording of spontaneously spiking SNr neurons in vitro in the presence ( upper black trace; Cntrl: AP5 & NBQX ) and absence ( lower red trace; +Gbz ) of inhibition . Left shows tonic spiking and right shows ∼20 action potential waveforms ( gray ) and average action potential waveforms ( black , Cntrl; red , Gbz ) . ( B ) No significant change in spike frequency was observed following addition of Gbz ( n = 11 cells; n . s . p>0 . 05 , dotted red line represents unitary line ) . ( C ) Phase plot of example average spike waveform from an individual SNr neuron in control conditions ( black trace ) and following removal of inhibition ( red trace ) , lower plot focuses on the perithreshold membrane potential dynamics . ( D and E ) Summary data of perithreshold and threshold slopes measured in control conditions and following pharmacological blockade of inhibition ( +Gbz , n = 11 cells ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 00910 . 7554/eLife . 02397 . 010Figure 4—figure supplement 2 . Low release probability and sustained depression at feedback inhibitory synapse . ( A ) Representative IPSC during first second of photostimulation delivered at a range of frequencies ( 10 , 20 , 50 , 100 Hz , indicated by labels at right of traces and cyan line ) during whole-cell voltage clamp recordings from SNr GABA neuron . ( B ) Average peak IPSC amplitude , normalized to IPSC1 ( left axis; dark blue trace ) and average tonic IPSC amplitude ( right axis; graygray trace ) plotted as a function of stimulus number for 10 , 20 , 50 , and 100 Hz photostimulation ( n = 4 cells; shading represents SEM ) . ( C ) Peak Steady State ( S . S . ) IPSC ratio ( IPSC100/IPSC1 ) and ( D ) S . S . tonic IPSC amplitudes ( IPSC100 ) plotted as a function of stimulation frequency ( n = 4 cells ) . Error bars represent SEM . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 010 The question as we posed it—how can a strongly coupled inhibitory network of tonically active neurons exhibit no effect of feedback under basal conditions ? —implies that tonic spiking is the problem . Alternatively , tonic spiking could be the solution . For a neuron to repetitively fire it must , upon the return from a spike , exhibit a net membrane current that is inward and thus drives the membrane towards spike threshold ( Raman and Bean , 1999 ) . This implies a positive slope conductance combined with a net inward current below threshold ( Nolan et al . , 2003 ) —in other words , the conductances that drive repetitive firing oppose hyperpolarizing currents in the perithreshold regime . Combined with a reduced driving force of inhibition near threshold , these biophysical features suggest that SNr neurons are much less sensitive to inhibition than to excitation in this regime . To test this hypothesis explicitly , we performed dynamic clamp experiments in which we systematically varied the balance between a high background rate of IPSCs and EPSCs ( Figure 4D–F ) . Indeed , we found that the sensitivity of the spike rate to increasing inhibition was much reduced compared to the sensitivity to increasing excitation . Stimulation strongly biased towards an inhibitory conductance often exhibited no effect on the mean spike rate relative to tonic levels . Consistent with the mechanistic model described above , we found that the slope conductance in the perithreshold regime was indeed nonlinear with a sharp positive slope near the inhibitory reversal potential ( Figure 4—figure supplement 1 ) . Moreover , we found that measured biophysical properties ( e . g . , slope conductance , spike threshold ) were unaffected by pharmacological blockade of inhibition ( Figure 4—figure supplement 1 ) . Collateral inhibition resulted in a strong suppression of evoked spiking and was sufficient to truncate evoked responses , often after only a few milliseconds . This suggested that collateral inhibition provided substantial inhibition that onset rapidly following stimulation . However , it is possible that the transient effect could also reflect properties of the photostimulation . To distinguish these possibilities , we examined the kinetics of feedback inhibition and compared it to the main source of feed forward inhibition to the SNr from the striatum . We made intracellular recordings from individual SNr projection neurons in Gad2-ChR2 mice to probe the properties of local feedback inhibition ( Figure 5A , B ) and from Drd1a-cre mice which were injected with a virus expressing a cre-dependent ChR2 transgene into the striatum to target the D1 receptor expressing medium spiny neurons which send axons directly into the SNr ( Gerfen , 1988 ) ( Drd1a-ChR2; Figure 5C ) . Postsynaptic neurons were recorded in the voltage clamp configuration with a holding potential of ∼+20 mV ( reversal potential of the ChR2 current; Figure 2H–J ) to isolate inhibitory postsynaptic currents ( IPSCs ) . Slices were perfused with antagonists of excitatory synaptic transmission . Repeated pulses ( 10 Hz ) of wide-field photostimulation elicited stimulus locked IPSCs in 15/23 SNr GABA neurons in the Gad2-ChR2 mouse line ( Figure 5B ) and 18/24 SNr GABA neurons in the Drd1a-ChR2 mouse ( Figure 5C ) . Outward currents recorded following photostimulation of both inputs were completely abolished by application of Gbz ( Figure 5B , C; p<0 . 001 ) . Evoked IPSCs from feedback nigral collaterals recorded in Gad2-ChR2 mouse exhibited rapid kinetics characterized by short , monosynaptic latencies ( Figure 5D; 1 . 93 ± 0 . 02 ms ) , rapid 10–90% rise times ( Figure 5E; 0 . 53 ± 0 . 01 ms ) and rapid decay time constants ( τ ) ( Figure 5F; 5 . 64 ± 0 . 14 ms ) . In contrast to the intranigral inhibitory synapses , we found that striatonigral IPSCs recorded in Drd1a-ChR2 mouse had significantly longer latencies ( Figure 5D; 2 . 51 ± 0 . 02 ms p<0 . 001 ) , slower 10–90% rise times ( Figure 5E; 0 . 75 ± 0 . 014 ms , p<0 . 001 ) and slower decay τ ( Figure 5F; 9 . 00 ± 0 . 14 ms , p<0 . 001 ) . Perhaps most surprisingly , we observed that with saturating stimulation , intranigral synapses contributed as large or greater inhibition than the major afferent source of inhibitory input , the direct pathway ( Figure 5G ) . 10 . 7554/eLife . 02397 . 011Figure 5 . Feedback inhibition has distinct biophysical properties from feed forward inhibition . ( A ) Schematic showing feedback nigral synapse ( red arrow; Gad2-ChR2 ) and feed forward striatonigral synapse ( blue arrow; Drd1a-ChR2 ) onto SNr GABA neurons . Synaptic properties of feedback inhibition were compared to feed forward inhibition to the SNr using photostimulation of ChR2 expressing SNr and striatal axons respectively . Wide-field 10 Hz photostimulation ( cyan ) to evoked activity of SNr GABA neurons in the Gad2-ChR2 mouse elicited large feedback IPSCs in SNr GABA neurons that were blocked with Gbz ( B; Gad2-ChR2 , n = 6 cells , p<0 . 001 , paired two tailed t test ) . Similarly , photostimulation of striatonigral afferents using Drd1a-ChR2 mouse evoked feed forward IPSCs in SNr GABA neurons that were blocked by Gbz ( n = 5 cells , p<0 . 001 , paired two tailed t test ) . Histograms of measured IPSCs latency ( D ) , rise time ( E ) and decay tau ( F ) for feed forward and feedback inhibition revealed feedback inhibition has significantly faster kinetics compared with feed forward inhibition . ( G ) Average IPSC amplitude as a function of stimulus duration for feed forward and feedback inhibition . ( H ) Fraction of IPSC1 amplitude during a 10 Hz train of photostimulation for feed forward and feedback inhibition . For D–H; maroon traces represent data from Gad2-ChR2 mice measuring feedback inhibition , n = 15 cells; blue traces represent data from Drd1a-cre mice measuring feed forward inhibition , n = 18 cells; for D–H , p<0 . 001 , paired two tailed t test . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 01110 . 7554/eLife . 02397 . 012Figure 5—figure supplement 1 . Feedback inhibition provides fast , transient inhibition . Wide-field illumination of SN elicited large IPSCs in SNr GABA neurons recorded in Gad2-ChR2 ( A , upper ) and Thy1-ChR2 ( B , upper ) mice using a 10 Hz photostimulation protocol ( cyan arrows ) . IPSCs were blocked with application of Gbz ( A; Gad2-ChR2 , n = 6 cells and b; Thy1-ChR2 , n = 8 cells , lower , p<0 . 001 paired two tailed t test ) . IPSC latency ( C ) , rise time ( D ) , and decay time ( E ) were not significantly different between recordings make in Gad2-ChR2 and Thy1-ChR2 mice . ( F ) Average IPSC amplitude as a function of stimulus duration evoked in Gad2-ChR2 mice was significantly reduced compared with IPSCs evoked in Thy1-ChR2 mice ( p<0 . 05 ) . ( G ) Fraction of IPSC1 amplitude during a 10 Hz train of stimulation for neurons recorded from Gad2-ChR2 and Thy1-ChR2 mice . For C–G; maroon lines represent data from Gad2-ChR2 mice , n = 15 cells; green lines represent data from Thy1-ChR2 mice , n = 24 cells . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 012 To probe the short-term plasticity properties of feed forward ( Drd1a-ChR2 ) and feedback ( Gad2-ChR2 ) inhibition to the SNr , we analyzed the amplitude of successive IPSCs . We found that collateral synapses within the SNr exhibited paired pulse ratios ( PPR ) less than 1 ( Figure 5H; PPR = 0 . 91 ± 0 . 03 ) . While , in direct contrast , the striatonigral synapse was modestly facilitating ( Figure 5H; PPR = 1 . 17 ± 0 . 05 ) . This latter observation was consistent with a previous study that used extracellular stimulation of the direct pathway ( Connelly et al . , 2010 ) . The depressing nature of the PPR for local feedback inhibition is unlikely to reflect desensitization of ChR2 as repeated pulses were all suprathreshold under our stimulus conditions ( Figure 2 ) . These results are thus consistent with a rapid onset of feedback inhibition sufficient to truncate sustained activation of the SNr population . We did not find any statistically significant differences between the kinetic properties seen between the IPSCs measured in slices from Gad2-ChR2 mice compared with those measured from Thy1-ChR2 mice ( Figure 5—figure supplement 1 ) . This is consistent with both approaches selectively or predominantly activating SNr projection neurons . By contrast , the amplitude of IPSCs evoked by maximal stimulation was significantly reduced in slices taken from Gad2-ChR2 mice compared to those taken from Thy1-ChR2 mice ( p<0 . 05; unpaired two-tailed t test; Figure 5—figure supplement 1 ) . These differences presumably reflect the non-homogeneous expression of ChR2 in the virally infected SNr of Gad2-ChR2 mice . Our results demonstrate a previously unappreciated potency of feedback inhibition in the SNr ( Figure 5G ) . Collateral synapses provide sufficient inhibition to regulate the gain of the output of the basal ganglia even during strong activation of the network ( Figure 3G ) . However , anatomical reconstruction of individual axons suggests a sparse connectivity within the SNr ( Mailly et al . , 2003 ) . The modest amplitude of individual mIPSCs ( ∼150 pS; Figure 3—figure supplement 2 ) and the low cell density of the SNr ( Gerfen , 2004 ) ( ∼30 , 000 neurons in a ∼4 mm3 volume [Oorschot , 1996] ) imply that feedback inhibition derives from a substantial volume . The maximal amplitude of evoked IPSCs was 10 , 000 pS . If we assume a <1% connection probability then we would predict that inhibition would be derived from neurons in a ∼600 μm radius from a given postsynaptic neuron . Consistent with such a model our wide-field stimulation experiments suggested that feedback inhibition magnitude scaled similarly for pairs of neurons ( Figure 3I ) . While reconstructions of individual axons have been studied in detail , such results cannot be used to reliably infer the convergence of input onto an individual projection neuron . Moreover , light-level anatomy data cannot reliably predict the functional impact of feedback inhibition . Thus , we next adapted the ChR2-assisited circuit mapping ( CRACM ) ( Petreanu et al . , 2007; Wang et al . , 2007 ) method developed for the neocortex to study the interconnectivity within nigral microcircuit . Using a 10X objective , it was possible to contain the entire extent of the SN within a single field of view . We positioned either 81-point or 140-point grids of stimulation sites to cover the SN ( Figure 6A ) . To achieve high spatial resolution of ChR2 activation , we used a focused 470 nm laser beam that could be rapidly re-positioned to each point on the grid in a pseudorandom sequence that avoided nearest neighbors . The duration of the light pulse was gated so as to deliver brief ( <1 ms ) pulses of light at each stimulation site . To obtain reliable and spatially homogeneous expression of ChR2 across nigral projection neurons , we used slices from the Thy1-ChR2 transgenic mouse line . Relative to wide-field stimulation or stimulation targeting axonal fibers , we reduced the maximal power of the laser using neutral density filters and performed calibration experiments to find intensities that would evoke precise , time-locked spikes in a small number of neurons with somatodendritic arbors surrounding the stimulation site ( Figure 6—figure supplement 1 ) . Our data were consistent with a requirement for propagating action potentials to elicit postsynaptic responses and we found no evidence in recorded neurons of direct axonal stimulation under these conditions ( Figure 6—figure supplement 1 ) . Furthermore , we focused on the rising phase and initial peak of IPSCs to bias our analysis towards transmission that resulted from highly reliable , low jitter spikes initiated at each stimulus site . 10 . 7554/eLife . 02397 . 013Figure 6 . Circuit mapping of feedback inhibitory circuitry of SNr . ( A , left ) Schematic of the experimental configuration used for channelrhodopsin-assisted circuit mapping . Whole-cell voltage clamp recordings were obtained from SNr GABA neurons while a focus laser beam was scanned across the SNr to excite SNr neurons with high spatial resolution . ( A , right ) Postsynaptic responses to individual photostimulations ( white ) were aligned to the DIC image of the slice . Stimulation points are indicated by cyan . ( B ) Example of evoked IPSCs from a single recording with a histogram of IPSC latencies for all recordings . Evoked IPSCs were completely inhibited in the presence of Gbz ( B , insert; n = 8 cells; p<0 . 001 , paired two tailed t test ) . ( C ) Cumulative histogram of response magnitude as a function of the distance between the stimulation site and recorded neuron in the Thy1-ChR2 ( green ) and Gad2-ChR2 ( maroon ) preparations . ( D ) Example of IPSC maps for two neurons . The dendritic arbor of each recorded neuron was reconstructed and transformed into the common SN reference frame ( dotted line ) . For each neuron the center of mass ( COM ) of inhibition ( COMIPSC , filled diamond ) , COM of dendritic field ( COMDEND , filled square ) and the isocontour of 50% inhibition ( IS0IPSC , colored line ) were calculated . ( E and F ) The COMIPSC was plotted as a function of the COMDEND for each neuron recorded in coronal ( E; n = 14 cells ) and sagittal ( F; n = 16 cells ) sections and the correlation fit estimated ( blue line ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 01310 . 7554/eLife . 02397 . 014Figure 6—figure supplement 1 . Light-evoked IPSCs result from perisomatic spiking . Calibration experiments were carried out to determine the resolution of photostimulation evoked spiking in individual SNr neurons in Thy1-ChR2 mice . ( A ) Cell-attached recordings from individual SNr neurons were made while a focused laser beam was scanned throughout a pre-defined grid of stimulation points spanning the SNr ( ‘Materials and methods’ ) . SNr neurons are tonically active , firing at ∼10–40 Hz , thus to access which stimulation point evoked reliable spiking , spike traces were averaged over multiple ( >3 ) trials , and voltage responses surrounding stimulation points were superimposed ( A , upper ) . Voltage deflections which exceeded 50% of the maximum amplitude , and which fell within 1 standard deviation ( 1 SD ) of the mean spike latency , were counted as generating reliable spiking . From this a corresponding color map of spiking reliability ( scale bar , 0 = not reliable spiking , and 1 = reliable spiking ) was generated ( A , lower ) . ( B ) Whole cell recordings were then made from the same neuron to access photocurrent amplitude at each stimulation point using the same grid as in A . Cells were held at −70 mV to isolate ChR2 mediated photocurrent . Averaged photocurrents evoked at each stimulation point were superimposed ( B , upper left ) and shown with an expanded time scale ( B , upper right ) and as a corresponding color map of normalized peak amplitudes ( B , right; scale bar , 0 = min amplitude , and 1 = max amplitude ) . For color maps in A and B blue dot indicates cell soma position , white scale bar represents distance between stimulation point . ( C ) The binned , normalized mean photocurrent response as a function of distance away from the soma ( gray bars ) and a fit to the binned probability of reliable spiking over the same distance for SNr GABA neurons ( solid black line , n = 4 cells ) . All recordings were performed in the presence of synaptic transmission blockers . To test whether suprathreshold axonal stimulation of SNr collaterals could evoke synaptic transmission , SNr GABA neurons were voltage-clamped at Vh + 20 mV to isolate IPSCs . Focused photostimulation throughout the SNr evoked IPSCs under control conditions ( D , upper ) and these were completely blocked following the addition of TTX ( D , lower ) . ( E ) Population data showing TTX inhibition of IPSCs ( E , n = 5 cells , p<0 . 001; paired two-tailed t test ) . ( F ) For a subset of mapping experiments , the latency to the detected IPSC is plotted as a function of distance from soma of the stimulation site . Although the spread is rather large due to variations in latency of evoked spikes by ChR2 positive SNr neurons , there is significant slope towards added propagation delays of ∼0 . 5 ms per 1 mm of stimulation distance . This corresponds to roughly 2 m/s conduction velocities , which is comparable to previously obtained estimates ( 1 . 7 m/s ) in rats ( Deniau et al . , 1978 ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 014 To measure the spatial organization of local inhibitory connectivity within the SNr , postsynaptic GABA neurons were clamped to Vh = +20 mV in the presence of glutamate receptor antagonists . The peak amplitude of light-evoked IPSCs was determined for each stimulus position and response maps of IPSC amplitude as a function of stimulus position were generated ( Figure 6A–C ) . While the viral-overexpression preparations do not provide homogeneous expression , we found that the length scale of the intranigral microcircuit estimated using the Gad2-ChR2 mouse was consistent with that from the Thy1-ChR2 mouse ( Figure 6C ) . To characterize the spatial organization of input to individual neurons , we characterized the ‘receptive field’ of feedback inhibition as the center of mass ( COM ) and the 25% isoinhibition contour ( ISO; Figure 6D ) . Finally , the majority of the somatodendritic arbor of each neuron was reconstructed from image stacks acquired on a two-photon microscope ( Figure 6D ) that allowed us to estimate the COM of the dendritic arbor . It has been proposed previously that collateral inhibition may be largely confined to topographic boundaries defined by feed forward input from the striatonigral pathway ( Mailly et al . , 2003 ) . However , if feedback and feed forward input were organized in register , then feedback could not produce a signal proportional to the ‘global’ activation of the network . This would imply a collection of parallel channels each of which could exhibit strong feedback . By contrast , our recording data in vivo ( Figure 1 ) suggested that feedback was proportional to the average activation across a large spatial extent of the SN . Thus , we next asked whether the feedback intranigral inhibition was organized in register with feed forward inhibition from the striatum . If feedback inhibition were organized in register with feed forward inhibition , then we would predict that ( 1 ) the somatodendritic position of the postsynaptic neuron should predict the location of the inhibitory receptive field and ( 2 ) the spatial extent of inhibitory receptive fields should be matched to the topographic boundaries defined by feed forward inhibition . In contrast to the first prediction , we found that there was no correlation between the location of individual neurons and the source of the strongest inhibition within the SNr . In other words , we found that neurons with non-overlapping dendritic arbors could have largely overlapping inhibitory receptive fields ( Figure 6D ) . We found no correlation between relative somatic position and the correlation of the inhibitory receptive fields in neither coronal ( n = 14 ) nor sagittal ( n = 16 ) slices ( Figure 6E–F ) . To assess whether local inhibition in the SNr observed topographic boundaries defined by afferent inhibition , we generated double transgenic mice in which ChR2 was expressed under control of the Thy1 promoter and cre-recombinase was expressed under control of the D1 receptor ( hereafter referred to as Drd1a-cre x Thy1-ChR2 ) . We then made focal injections of a cre-dependent virus expressing a red fluorescent protein into the dorsal striatum 2–3 weeks prior to performing circuit mapping experiments in midbrain slices ( Figure 7—figure supplement 1 ) . Clear axonal labeling could be readily observed in the SNr ( Figure 7A; Figure 7—figure supplement 1 ) . The striatonigral projection exhibited the characteristic ‘dual nature’ that has been observed following focal tracer injections in rats ( Gerfen , 2004 ) . In each slice , we performed circuit mapping for three to six projection neurons at a range of distances from the axonal termination fields ( Figure 7A ) . We found that the correlation in the maps obtained from individual neurons was a monotonically decreasing function of distance between neurons ( Figure 7B ) . However , we found no clear organization between the striatonigral projection and the maps of feedback inhibition ( Figure 7A , C ) . The organization of the local inhibitory circuit in relation to the boundaries of striatonigral axonal tracing suggests that there could be a partial separation between the regions that receive input from the medial and lateral striatum ( Figure 7D , Figure 7—figure supplement 2 ) . However , the consistent fall off of the correlation in inhibitory response maps across neurons ( Figure 7B ) suggests that projection neurons receive feedback inhibition from a diffuse microcircuit and independent of the discontinuous topography of feed forward input . 10 . 7554/eLife . 02397 . 015Figure 7 . Intranigral inhibition is poorly predicted by the organization of the striatonigral pathway . Neurons in the dorsal striatum of Drd1a-cre x Thy1-ChR2 double transgenic mice were infected with cre-dependent AAV that drove the expression of a red fluorescent protein to label striatonigral axons ( tdTomato ) . Bright-field images of the fluorescent axons in the SN were used to estimate the location of labeled axons ( A , bottom layer ) . Estimates of the density of axonal labeling were produced by extracting the axon contour ( quartiles indicated by gray line thickness ) and compared with the localization of local inhibitory input ( thresholded at 20% of maximum response ) for multiple neurons recorded in the same slice ( A , upper 3 layers ) . Individual postsynaptic neurons with proximal dendritic arbors reconstructed are shown in shades of red . The approximate border of the SN is indicated ( cyan dashed line ) . ( B ) The correlation in spatial maps of IPSC amplitudes were computed for all pairwise comparisons between neurons recorded in the same slice ( n = 10 slices; n = 36 cells ) as a function of the distance between somata . Gray circles are individual correlations , red circles are binned means with standard errors , and solid red line is an exponential fit . ( C ) For each slice the correlation between a spatial map of IPSC amplitudes and the axonal density map is shown as a function of the distance between the soma of the recorded neuron and the center of mass of the axon projection . ( D ) For all slices the maximum intensity contrast ( ‘Materials and methods’ ) for the axonal labeling was overlaid with the location of all recorded somata ( red circles ) . The angle and distance to the center of mass of the spatial maps of IPSC amplitudes are indicated by the red arrows . An example projection field from a single infection of the dorso-medial striatum is shown in dark cyan . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 01510 . 7554/eLife . 02397 . 016Figure 7—figure supplement 1 . Mapping striatonigral axonal terminal fields . Focal injections of a cre-dependent AAV expressing tdTomato was targeted to the striatum of Drd1a-cre x Thy1-ChR2 mice to allow both labeling of a subset of striatonigral projecting axonal and subsequent CRACM of the local inhibitory network within the SNr ( n = 5 animals ) . Example slices from four mice with striatal injections ( A , upper ) and their corresponding axonal terminals in the SNr ( B , lower ) . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 01610 . 7554/eLife . 02397 . 017Figure 7—figure supplement 2 . Anatomical organization of the striatonigral pathway . ( Upper panels ) Two cre-dependent AAVs driving the expression of different fluorescent protein transgene ( indicated in top labels ) was injected into the medial ( green ) and lateral ( red ) aspect of the striatum in Drd1a-cre mice . ( Lower panels ) Axonal fibers were found in the SN . The axon termination zones showed strong fluorescence and were largely non-overlapping for injections at the striatal extrema . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 017 The potency and diffuse organization of feedback inhibition suggested that collateral synapses from SNr projection neurons could be a powerful determinant of the activity of the SNr even in the presence of ongoing input from afferent sources . The in vitro preparation presumably reduces or eliminates structured activity in afferent sources of input that could either directly compete with or modulate the consequence of feedback inhibition . Moreover , the length constant of inhibition was similar in the sagittal and coronal planes ( Figure 6E , F ) and thus , the potency of inhibition measured in vitro is , if anything , an underestimate of the impact of intranigral inhibition . We next asked whether we could use a complementary approach to measure the extent of the intranigral microcircuit in the awake mouse . By contrast to the mapping experiments above where we measured at one location ( neuron ) and stimulated many other locations , we next used either a silicon probe electrode array ( Figure 8A ) or wire array ( Figure 8—figure supplement 1 ) with integrated optical fibers to stimulate at one ( somewhat diffuse due to light scattering ) location while measuring the spiking activity of SNr neurons at many neighboring locations . 10 . 7554/eLife . 02397 . 018Figure 8 . Potent and diffuse intranigral inhibition in vivo . ( A ) Schematic of experimental configuration used for in vivo recordings . An optical fiber was affixed to one shank of a silicon probe electrode array . The array was lowered into the SN of awake , head-fixed mice . ( B ) Raster plots of responses to light stimulation for two example single units isolated from such recordings . Spikes are indicated by vertical hash marks , colored and sorted by stimulus duration . Mean PSTHs are shown in lower panels and the average waveform ( ±1 SD ) are shown in the insets . Some units ( e . g . , s06u01 ) exhibited direct excitation by photostimulation followed by suppression . While other units ( e . g . , s03u21 ) located at a more eccentric position on the array exhibited a delayed ( one 5 ms PSTH bin ) suppression of firing . ( C ) The distribution of average firing rates was very similar for single units isolated in vivo ( gray bars ) and on cell spiking rates observed in vitro ( open bars ) . Directly excited units ( red bars ) and units exhibiting inhibition below baseline ( open cyan bars ) are plotted as a function of baseline firing rate . ( D ) The mean response magnitude for units exhibiting short latency activation ( red ) and the most inhibited quartile of the population ( cyan ) are plotted as a function of stimulus duration . ( E ) The spatial arrangement of sites at which direct excitation ( red ) or inhibition below baseline ( cyan ) was observed . Left , individual shanks of the silicon probe array are shown as light gray lines and the shank to which the optical fiber was affixed is shown in darker gray . The position of individual recording sites is represented as black dots . Every significant excitatory and/or inhibitory response is represented as a triangle or circle , respectively . Scale bar: 200 µm , 5 z×ms . The diameter of the symbol reflects the magnitude of the response to stimulation for stimuli of 20 ( upper ) and 50 ( lower ) ms stimuli . Middle , a maximum intensity projection for direct excitation ( red ) and inhibition ( blue ) . Scaling of maximal value is shown in lower left . Right , a cumulative histogram of response magnitude as a function of distance from the focus of excitation ( most strongly activated site on the shank with associated optical fiber ) . Distance calculated based upon 200 µm site spacing on the silicon probe array . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 01810 . 7554/eLife . 02397 . 019Figure 8—figure supplement 1 . Feedback inhibition shapes SNr output in Thy1-ChR2 transgenic mice . ( A ) Schematic of experimental configuration used for in vitro recordings . ( B ) The population response of recorded projection neurons in response to photostimulation of varying durations was sorted by the normalized magnitude of the response ( cyan = −1 , red = +1 ) for control ( Cntrl ) conditions . ( C ) The normalized difference of responses between stimulation under control conditions and stimulation in the presence of gabazine ( Gbz ) . ( D ) Schematic of experimental configuration used for in vivo recordings . ( E ) Normalized PSTHs for single units with baseline firing greater than 9 Hz ( n = 147 cells ) were aligned to stimulus onset . Colors as in B , inhibition and excitation normalized independently . ( F ) Mean PSTHs were calculated for the population of neurons with a dominant excitatory response ( left , ‘Materials and methods’ ) for in vitro ( black ) and in vivo recordings ( 36% of population , red ) or a dominant inhibitory response ( right , ‘Materials and methods’ ) for in vitro ( black ) and in vivo recordings ( 13% of population , cyan ) . ( G ) Raster plots and boxcar averages are shown from example neurons . Examples were selected by finding the single unit PSTH with the highest correlation to the population PSTH for excitation dominant ( left ) and inhibition dominant ( right ) populations . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 019 Following implantation of a 64-site silicon probe array with integrated optical fiber into the SN of Gad2-ChR2 mice ( Figure 8A ) , we observed single units with narrow waveforms and high baseline firing rates characteristic of SNr projection neurons ( Figure 8B ) . The distribution of baseline firing rates obtained in vivo was closely matched to the same distribution obtained using on-cell recordings from identified SNr projection neurons in vitro ( Figure 8C ) . We also confirmed that brief stimulation with a light pulse , like the stimulus used in the gain experiments in Figure 3 , produced population responses similar in time course to those measured during response to natural stimuli ( Figure 3—figure supplement 1 ) . We found that in vivo , as observed in vitro , stimulus pulses of increasing duration produced mixed responses in the SNr population characterized by units with direct excitation and a subsequent delayed inhibition ( Figure 8B , upper example ) as well as units with a pure suppression of firing that onset with a short delay ( ∼5 ms latency; Figure 8B , lower example ) . A relatively constant level of inhibition combined with increasing excitatory drive can produce subtractive effects on the output spiking of a neuron . By contrast , divisive gain effects require that inhibition be recruited in proportion to changes in excitatory drive . Our in vitro data were consistent with a divisive gain effect on nigral output due to intranigral inhibition . To distinguish these two possibilities in vivo , we examined neurons that exhibited apparent direct excitation ( short-latency increase in stimulus evoked firing ) with those that exhibited inhibition ( stimulus evoked suppression of firing below baseline ) . We found that inhibition and excitation were recruited with a similar dependence on stimulus intensity consistent with a model in which intranigral inhibition produces a divisive gain effect on the output of SNr projection neurons ( Figure 8D ) . Finally , our circuit mapping data ( Figures 6 and 7 ) suggested that functional inhibition extended for hundreds of microns within the SNr . To estimate the spatial extent of inhibition in vivo , we next examined the position of excitatory and inhibitory responses on the electrode array . The location of each inhibitory and excitatory response is plotted as a function of position for the 20 ms and 50 ms stimulus conditions as a function of electrode position , maximal intensity projection , or cumulative histogram as a function of distance from the focus of excitation ( Figure 8E ) . Taken together , these data indicate that functional inhibition can extend for hundreds of microns beyond the focus of excitation in vivo as was observed in vitro . In our experiments , we used two approaches to achieve cell-type specific expression of ChR2 in the GABAergic projection neurons of the SNr . Both viral-mediated infection of GABAergic neurons in the SNr and the Thy1-ChR2 transgenic mouse line exhibited cell-type specific expression of ChR2 in GABAergic neurons of the SN , but not in dopamine neurons . GABAergic projection neurons and dopamine neurons are thought to be the only two cell types present in the rodent SN . Importantly , both approaches had indistinguishable properties of inhibition onto both GABAergic projection neurons ( Figure 5—figure supplement 1 ) and onto dopamine neurons ( Pan et al . , 2013 ) . Moreover , viral mediated expression allowed for the best possible quantitative comparison between feed forward and feedback pathways by ensuring that expression of the ChR2 was controlled by a common promoter ( Figure 5 ) . Thus , for these results the only significant differences between the virally-mediated and transgenic expression of ChR2 in the SNr was the fraction of the population found to express ChR2 and the resultant magnitude of feedback inhibition . While viral-mediated expression could be used to estimate the divergence of inhibitory projections from a single stimulation site ( Figure 8 ) , it was not suitable for comprehensive mapping of convergence onto individual SNr neurons ( Figures 6 and 7 ) . The concern in a transgenic mouse line is that ChR2 expression could be present in afferent fibers . To control for this , we blocked excitatory transmission , confirmed that light-activated currents had kinetics expected of ChR2 positive neurons , and confirmed that neurons of the major source of inhibitory input , the striatonigral pathway , did not express ChR2 . Consistent with these observations it has previously been suggested that inhibitory fibers in this Thy1-ChR2 mouse line do not express ChR2 ( Gradinaru et al . , 2009 ) . However , in addition to expression in the SNr , we did also observe that neurons of the external globus pallidus ( GPe ) express ChR2 . The GPe is the source of a projection to the subthalamic nucleus and a more modest projection to the SNr ( Bolam et al . , 2000; Gerfen , 2004; ) . Nonetheless , we found that potent feedback inhibition could be observed in Gad2-ChR2 mice and that stimulation of GPe axons yielded relatively less inhibition in SNr neurons than either the striatonigral or intranigral pathways under our stimulus conditions . Finally , we used laser powers that were attenuated relative to the powers necessary to directly stimulate severed axons in these mice ( data available from the authors on request ) , consistent with our observation that only perisomatic stimulation was sufficient to evoke reliable spiking in SNr neurons ( Figure 6—figure supplement 1 ) . Unlike many other circuits in which gain control has been studied , projection neurons of the SNr are spontaneously active . Divisive gain control requires that there is little effect of inhibition in the absence of stimulation . In the case of gain control mediated by an interneuron this can be achieved in a number of ways , for example , through facilitating inhibitory synaptic transmission ( Silver , 2010; Figure 9 ) . However , in spontaneously active neurons , it is less clear how to prevent feedback inhibition from altering the baseline firing rate as we observe here . One possibility suggested by our data is that a broad distribution of firing rates ( Figure 8 ) combined with relatively weak individual connections ( Figure 4 ) could produce inhibition that is essentially tonic and too small to significantly affect intrinsic currents that drive repetitive spiking ( Figure 4—figure supplement 1 ) and so is counteracted via subthreshold inward currents necessary for repetitive firing . We provided support for such a model by demonstrating that there is , indeed , a high frequency ( 200 Hz ) of spontaneous IPSCs bombarding SNr projection neurons ( Figure 4 ) . Further consistent with this model , simulation of a steady background rate of IPSCs using dynamic clamp revealed a relative insensitivity of spike rate to a tonic net inhibitory background of inputs ( Figure 4 ) . This balance in the tonic firing rate can be disrupted by synchronously recruiting neighboring neurons to produce a rapid inhibition that , combined with the positive feedback produced by inhibition and subsequent disinhibition , overcomes via de-activation ( Nolan et al . , 2003 ) the inward currents that drive the membrane potential towards threshold . This mechanism combining intrinsic properties and synaptic properties is sufficient for a population of spontaneously active inhibitory neurons to implement divisive gain control in the apparent absence of interneurons ( Deniau et al . , 2007b ) . To our knowledge this represents a novel circuit mechanism for divisive gain control ( Silver , 2010; Figure 9 ) . 10 . 7554/eLife . 02397 . 020Figure 9 . Schematic summary of proposed mechanism for divisive gain control in a circuit lacking interneurons . ( A ) Schematic of the canonical basal ganglia circuit with detail showing the anatomical basis for intrinsic feedback control of the basal ganglia output via the intrinsic microcircuitry of the substantia nigra . ( B ) Comparison of candidate mechanisms for gain control described in microcircuits with interneurons ( e . g . , Silver , 2010 ) with the mechanism for divisive gain control in the substantia nigra ( a circuit thought to lack interneurons ) described here . DOI: http://dx . doi . org/10 . 7554/eLife . 02397 . 020 One can think of two possible regimes in which the SNr may operate and each has distinct implications for the function and role of intranigral inhibition . On the one hand , it has been argued for some time that the feed forward pathways of the basal ganglia are topographically organized and largely independent ( Mink , 1996; Haber , 2003; Hikosaka , 2007 ) . Movement is thus thought to occur when a focal population of projection neurons becomes inhibited by feed forward input and thus disinhibits downstream pre-motor structures . From this perspective , the diffuse intranigral microcircuit could act to release neighboring projection neurons from intranigral inhibition and thereby suppress unintended movements . Thus , during focal activation of the SNr , collateral inhibition may be thought of as a mechanism for contrast enhancement akin to the role of lateral inhibition in sensory systems . This mechanism may be reflected in the time locked , bidirectional changes in the firing of SNr projection neurons that are commonly observed prior to and during movement in mice ( Pan et al . , 2013; Fan et al . , 2012 ) and primates ( Turner and Anderson , 1997; Nevet et al . , 2007 ) . On the other hand , while the direct striatonigral pathway is topographically organized there is considerable divergence in the corticostriatal input in mice ( Pan et al . , 2010 ) . In addition , there may be less precise topographic organization of the indirect pathway that enters the SNr via the subthalamic nucleus ( Bolam et al . , 2000 ) . The subthalamic nucleus also receives direct cortical input and ascending input from the midbrain ( Coizet et al . , 2009 ) and hindbrain ( Bevan and Bolam , 1995; Winn , 2006 ) . The topographic organization of these ascending pathways is less well understood . Regardless of the topographic precision , the inputs that arrive at the SNr from the subthalamic nucleus convey a great diversity of information and likely exhibit a diversity of dynamics . While there are mechanisms that could maintain activity within a fixed dynamic range in upstream structures ( Silver , 2010 ) , individual neurons that constitute the output of cortical areas project to multiple subcortical structures ( Kita and Kita , 2012 ) . It is therefore unlikely that the dynamic range of even the cortical output is appropriate for the diverse computations performed in all target structures . To effectively control the basal ganglia output in the presence of such diverse input dynamics and anatomical divergence would seem to require coordinated processing across functional domains . Thus , divisive gain control supplied by a diffuse but potent inhibitory microcircuit could be well suited to ensure that activity remains within a fixed dynamic range . We favor a model in which these contrasting descriptions of the role of the intranigral microcircuit are two aspects of its function that can be engaged in different input regimes . The diffuse organization that could produce lateral inhibition in some regimes under the animal's control ( i . e . , activation of specific well learnt actions ) , but , may be necessary to produce divisive gain control in other regimes ( i . e . , ‘global’ activation of the SNr by salient stimuli ) . While it is not currently possible to monitor nigral dynamics during selective manipulation of feedback , but not feed forward or efferent inhibition , such an approach will be required to definitively test the predictions of our work . Simple geometrical considerations suggest that the intranigral microcircuit integrates functionally distinct information on a large scale . Cortical afferents to the basal ganglia are derived from an estimated 17 million neurons ( Zheng and Wilson , 2002 ) spanning the majority of the roughly 100 mm3 volume of neocortex . These inputs are funneled through the basal ganglia and will ultimately terminate on approximately 30 , 000 projection neurons ( Oorschot , 1996 ) within the roughly 4 mm2 volume of the SN . Dendrites of nigral projection neurons and intranigral inhibition that extends over hundreds of microns ( Grofova et al . , 1982; Mailly et al . , 2001 ) could therefore shape activity derived from cortical inputs separated by several millimeters . Existing functional models focus on the feed forward structure of processing within the intrinsic basal ganglia circuitry ( e . g . , Albin et al . , 1989; DeLong , 2000; Mink , 1996 ) . However , our observation that intranigral inhibition is strong relative to the major source of feed forward input suggests that local processing of diverse streams of information in the SN could be critical for generating dynamics in the basal ganglia output . We hypothesize that the diverse impairments characteristic of pathological disruption of basal ganglia function could reflect , in part , a control system operating outside of a stable regime . Although there is evidence of perturbed dynamics in the SNr in disease models ( Ibanez-Sandoval et al . , 2007; Samadi et al . , 2008; Wang et al . , 2010 ) , the specific contribution of the intranigral microcircuit to the diverse behavioral impairments observed in diseases afflicting the basal ganglia circuit remains unclear . For in vitro experiments , adult transgenic mice ( 10–30 weeks old ) expressing either; ChR2-YFP fusion gene under the control of the mouse thymus cell antigen 1 promoter ( Line 18 , Stock #007612; Jackson Labs , Bar Harbor , Maine , ‘Thy1’ mice ) , cre-recombinase under the control of the glutamic acid decarboxylase 2 gene ( Stock #010802; Jackson Labs , ‘GADcre’ mice ) or cre-recombinase under the control of the dopamine receptor D1A ( GENSAT , Gong et al . , 2010 , Rockefeller University , ‘Drd1a-cre’ mice ) . For in vivo electrophysiology experiments , four adult ( 30 g , 3–6 months old ) Thy1 mice were used . All animals were handled in accordance with guidelines approved by the Institutional Animal Care and Use Committee of Janelia Farm Research Campus . The experimenter was not blinded to genotype . Mice were housed in a temperature- and humidity-controlled room maintained on a reversed 12 hr light/dark cycle . For in vivo physiology experiments , mice were housed individually . For in vitro experiments , mice were group housed ( 1–5 mice per cage ) . We used 3 adeno-associated viruses ( AAV , serotype 2/1 ) to achieve either conditional expression of ChR2 and tdTomato or pan-neuronal expression of eGFP and tdTomato . Viruses were produced at the Molecular Biology Shared resource of Janelia Farm Research Campus . Where indicated similar viruses can be obtained publicly from the Gene Therapy Program at the University of Pennsylvania ( http://www . med . upenn . edu/gtp/vectorcore/Catalogue . shtml ) . Conditional ChR2 expression was achieved with AAV2/1 SYN-FLEX-ChR2-GFP analogous to AV-1-18917P . Conditional tdTomato expression was achieved with AAV2/1 CAG-FLEX-tdTomato-WPRE-bGH available as AV-1-ALL864 . Pan-neuronal expression of tdTomato and eGFP were obtained via AAV2/1 SYN-[tdTomato/eGFP]-WPRE-SV40 available as AV-1-PV1696 . Viruses were injected into the striatum ( STR ) of Drd1a-cre mice , globus pallidus ( GP ) or substantia nigra ( SN ) of GADcre mice , in a fashion similar to that previously described ( Atasoy et al . , 2008 ) . Briefly , under deep anesthesia , a small craniotomy was made over the STR ( 0 . 5 mm anterior-posterior , 1–2 mm medial-lateral , −2 . 5 mm dorso-ventral ) , GP ( −0 . 45 mm anterior-posterior , −1 . 8 mm medial-lateral , −3 . 6 mm dorsal-ventral ) or SN ( −3 mm anterior-posterior , 1 mm medial-lateral , −4 . 2 mm dorso-ventral ) . A glass pipette was used to pressure inject small volumes of virus ( 20–100 nl per injection site ) . Animals were allowed to recover for at least 2 weeks following surgery . Briefly , adult mice were deeply anaesthetized under isoflurane , decapitated , and the brains were dissected out into ice-cold modified artificial cerebral spinal fluid ( aCSF ) ( 52 . 5 mM NaCl , 100 mM sucrose , 26 mM NaHCO3 , 25 mM glucose , 2 . 5 mM KCl , 1 . 25 mM NaH2PO4 , 1 mM CaCl2 , 5 mM MgCl2 , and 100 μM kynurenic acid ) that had been saturated with 95% O2/5% CO2 . 300 μM thick coronal and sagittal slices ( as indicated in the text ) were cut ( Leica VT1200S; Leica Microsystems , Germany ) , transferred to a holding chamber and incubated at 35°C for 30 min in modified aCSF ( 119 mM NaCl , 25 mM NaHCO3 , 28 mM glucose , 2 . 5 mM KCl , 1 . 25 mM NaH2PO4 , 1 . 4 mM CaCl2 , 1 mM MgCl2 , 3 mM sodium pyruvate , 400 μM ascorbate , and 100 μM kynurenic acid , saturated with 95% O2/5% CO2 ) and then stored at 21°C . For recordings , slices were transferred to a recordings chamber perfused with modified aCSF ( 119 mM NaCl , 25 mM NaHCO3 , 11 mM glucose , 2 . 5 mM KCl , 1 . 25 mM NaH2PO4 , 1 . 4 mM CaCl2 , 1 mM MgCl2 , 3 mM sodium pyruvate , 400 μM ascorbate , saturated with 95% O2/5% CO2 ) and maintained at 32–34°C at a flow rate of 2–3 ml min−1 . For mIPSC experiments , extracellular Ca2+ was replaced with 2 mM Sr2+ to desynchronize release . Patch pipettes ( resistance = 5–8 MΩ ) were pulled on a laser micropipette puller ( Model P-2000; Sutter Instrument ) and filled with one of the following intracellular solutions: Current-clamp recordings of spike activity used a potassium gluconate-based intracellular solution ( 137 . 5 mM potassium gluconate , 2 . 5 mM KCl , 10 mM HEPES , 4 mM NaCl , 0 . 3 mM GTP , 4 mM ATP , 10 mM phosphocreatine , pH 7 . 5 ) . Voltage-clamp recordings for IPSC measurements used a CeMeSO4-based intracellular solution ( 114 mM CeMeSO4 , 4 mM NaCl , 10 mM HEPES , 5 mM QX314 . Cl , 0 . 3 mM GTP , 4 mM ATP , 10 mM phosphocreatine , pH 7 . 5 ) . Alexa Fluor 488 or Alexa Fluor 568 was commonly added to intracellular solution to aid cell visualization and post hoc reconstruction . In some experiments the following were added as indicated in the text: 10 μM CNQX or 5 μM NBQX , 50 μM D-AP5 , 10 μM gabazine ( Gbz ) , 0 . 5 μM tetrodotoxin ( TTX ) . All drugs were obtained from Tocris Biosciences . Intracellular recordings were made using a MultiClamp700B amplifier ( Molecular Devices ) interfaced to a computer using an analog to digital converter ( PCI-6259; National Instruments ) controlled by custom written scripts ( to be made available at http://dudmanlab . org/ ) in Igor Pro ( Wavemetrics ) . Photostimulation was carried out using a dual scan head raster scanning confocal microscope and control software developed by Prairie Systems , and incorporated into a BX51 upright microscope ( Olympus America ) . Individual neurons were patched under DIC optics with a water-immersion 40X objective . Spiking was measured in the cell-attached configuration . The spiking frequency and action potential waveform were used to classify neurons as DA or GABA as described previously ( Pan et al . , 2013 ) . Upon break in while diffusion of QX314 was allowed time to progress , negative voltage clamp steps were used to measure the hyperpolarization-activated inward ( Ih ) current . The presence of detectable inward currents was diagnostic for DA neurons . In current clamp recordings lacking QX314 in the internal solution the intracellular spike waveform and spontaneous firing frequency were further used to confirm cell identity . Analysis of postsynaptic currents ( direct photocurrents , sIPSC , mIPSCs and evoke IPSCs ) and spiking was performed using custom written analysis code in Igor Pro ( Wavemetrics ) . Peak current amplitude was measured as the peak synaptic current relative to the baseline holding current preceding each stimulus . Tonic current amplitude was measured as the peak evoked synaptic current relative to the holding current preceding each train . The conductance ( g ) underlying IPSCs were calculated from g = IPSCpeak/ ( Vm − ECI ) , where IPSCpeak is the peak amplitude of the IPSC and Vm is the holding voltage . The equilibrium potential of GABAA current ( ECI ) was estimated at −70 mV from the Nernst equation . Rise time constants of postsynaptic currents were measured by finding the 20–80% slope of the rising phase of the stimulus-evoked current . Decay time constant of postsynaptic currents were measured by fitting a single exponential to the decay phase of the stimulus-evoked currents . Spikes were detected at the threshold of maximum acceleration . Phase plots were constructed by plotting the first derivative of the somatic membrane potential ( dV/dt ) vs the somatic membrane potential for the average spike waveform . The membrane potential at which phase plot slope reached 10 mV∙ms−1 was denoted the voltage threshold and a linear fit was used to calculate the slope . The perithreshold slope was calculated as the slope of the ‘kink’ defined as the slope of dV/dt for 7 ms after the peak of the perithreshold dV/dt . For dynamic clamp experiments , individual postsynaptic conductances were generated using IGOR Pro ( Wavemetrics ) , from the sum of two exponentials with rise tau and decay tau derived from measured IPSC and EPSC rise and decay kinetics ( rise tau = 0 . 5 ms , decay tau = 5 ms ) . The times of individual events were computed by sampling a Poisson distribution in which the rate of IPSC and EPSC events were independently changed from 1000 to 5000 Hz to generate different balances of excitation and inhibition . The convolved waveforms for excitation and inhibition were computed independently and passed to a custom made , digital dynamic clamp ( update rate 30 kHz; to be described elsewhere ) assuming reversal potentials of −70 mV and 0 mV for inhibition and excitation , respectively . The optics were designed to minimize the spread of the laser in the x , y dimensions of the focal plane while accentuating the focus in z by underfilling the back aperture of the objective . Stimulation intensity was controlled by pulse duration ( 0 . 2–1 ms ) . Stimulation typically consisted of 9 × 9 and 10 × 14 maps of stimulation sites with independent stimuli being delivered in a pseudo-random ( non-neighbor ) sequence at an interstimulus interval of ≥150 ms and values reflect the average of 3–4 repetitions of the mapping experiment for each cell . Stimulation strength was modulated by gating the laser at maximal power ( 473 nm , AixiZ or 488 nm , BlueSky Research ) with varying durations using timing signals from an external pulse controller ( PrairieView software ) and the internal power modulation circuitry of the laser or an external Pockels cell ( Conoptics ) with indistinguishable results . Wide-field activation of ChR2 was accomplished using blue LED ( 470 nm , ThorLabs ) transmitted through the fluorescence light path of the BX51 microscope . LED intensity and timing were controlled through a variable current source ( ThorLabs ) . Stimulus families ( input/output curves ) were delivered in a pseudorandom order and repeated 3–10 times per cell . Analysis was performed using custom written routines for Igor Pro ( Wavemetrics ) and Matlab R2011a ( Math Works ) . Analysis of the full field photostimulation was performed using standard analysis metrics as described in the text . To attempt to minimize the variability in estimates of short-term plasticity stimulation was performed at the half-maximum stimulus intensity determined by generation of an input–output function at the beginning of the experiment . The analysis of circuit mapping experiments was more complicated and is described briefly below and demonstrated more explicitly in Figure 6 . Briefly , averages of 3–10 multisite photostimulation experiments were used in all analyses . The moment of photostimulation was determined by thresholding a photodiode signal positioned in a parallel light path to the stimulation light path . Galvonometer position signals were recovered from the PrairieView software and aligned using a transmitted light laser scanning DIC image of the brain slice . Offline analysis routines automatically detected the orientation of the stimulus grid and applied a rotation to put that grid into 0 rotation orientation . The dorsal and medial edges of the grid were manually annotated and used to flip or further rotate all grids into a common reference frame . The average dimensions of the SN are 1 . 67 mm wide by 1 mm tall . We found that the average of all grids had an identical ( 1 . 67–1 ) aspect ratio . There was modest variation ( ∼10% ) in the dimensions of the SNr grids . All grids were linearly stretched or compacted to the same mean aspect ratio and cell positions moved accordingly . Qualitatively similar results were obtained in the presence and absence of warping . Full depth maps were generated by convolving the response amplitude at individual stimulus positions with an empirically-estimated Gaussian response function . An isocontour of the resulting image was generated at the half maximum level using the ‘contour’ function supplied by Matlab . The center of mass ( COM ) was also calculated as the vector average of the Euclidean distance to the stimulus position weighted by response magnitude . Reconstructions of recorded neurons were derived from two-photon fluorescent image stacks using the semi-automated software generously provided by Ting Zhao ( Janelia Farm Research Campus , HHMI ) ( Zhao et al . , 2011 ) . The COM of the dendrites was calculated as the vector average with the weights defined by the width of the dendritic branch segment at the end of the vector position . Data were then loaded into Matlab for display and scaling . Recordings were performed using either a 32-microwire arrays ( CD Neural Technologies ) or a 64-channel silicon probe array ( NeuroNexus Technologies ) . Electrode arrays were stereotaxically implanted under anesthesia ( isoflurane; 1 . 5–2 . 5% in O2 ) in mice that had been previously fitted with a plastic head restraint and held in place by a custom head fixation system ( Osborne and Dudman , 2014 ) . Design files and details about the manufacture and use of our head restraint system are available online ( http://dudmanlab . org/html/rivets . html ) . Electrode arrays were targeted to the SN of the ventral midbrain ( 3 . 0–4 . 5 mm posterior to bregma , 0 . 5–2 . 0 mm lateral to midline and >3 . 5 mm below the surface of skull ) . Electrode arrays were maintained in position by a micromanipulator ( Sutter Instruments or Scientifica ) and connected to the recording systems via a flexible wire coupling and connector . For optogenetic experiments , a 200 μm core multimode fiber ( ThorLabs ) was affixed near the central recording wires of a 32 channel array or to one shank of the silicon probe array as indicated in Figure 8A . The entire array was slowly lowered in to the midbrain . Following >1 hr of recovery single unit recordings were obtained from alert , but quietly resting mice . Single cell isolation was performed offline using Offline Sorter ( Plexon Technologies ) and standard techniques . Analysis of stimulus-evoked responses were calculated and presented using Matlab 2011a . The spike data in Figure 1 are a subset of recording sessions ( all sessions with ≥8 simultaneously recorded , putative GABAergic cells ) from mice performing an auditory trace conditioning task described in detail previously ( Pan et al . , 2013 ) . Briefly , mice were trained to consume sweetened water rewards delivered from a port placed on one wall of a behavior box . A speaker placed behind one wall of the box delivered pure tones ( 10 kHz; 500 ms duration ) as conditioned stimuli ( CS ) . Water rewards were delivered 2 . 5 s following CS onset . This data set included 599 single units recorded across sessions in which 5 to 21 units were recorded simultaneously . For each session , we computed the firing rate of the population of units prior to the onset of the CS and the transient response in the 200 ms following CS onset and subtracted the mean response across all trials . For each trial and all units recorded in a given session , we then computed the population response excepting the ith unit ( PREpopulation ) and the response of the ith unit to the CS ( RESPsingle ) . For individual sessions , we determined the correlation between PREpopulation and RESPsingle for all units in the session . Significance of the correlation was determined using a permutation test . For the entire population , we plotted PREpopulation vs RESPsingle for all units , all trials . The data were binned into 20 equally spaced bins and mean data for all bins with more than five samples was plotted and fit with a sigmoid function using Igor Pro . All statistical tests were performed using the statistics package from Matlab 2011a ( Math Works ) . Paired comparisons were performed using the student's t test ( all results were also confirmed with a non-parametric ranksum test ) . Multiple comparisons were performed using ANOVA . Significance was defined as p<0 . 05 unless otherwise indicated . Averaged data are presented as mean ± standard error of the mean ( SEM ) , unless otherwise specified .	The basal ganglia are a group of nuclei located deep within the brain that are involved in the control of movement . The death of neurons in one particular nucleus—known as the substantia nigra—gives rise to a range of symptoms that are characteristic of Parkinson’s disease , including slowness of movement and tremors . Although the basic anatomy and circuitry of the basal ganglia were worked out many years ago , it is not clear how these structures control voluntary movement . Based on insights from engineering , Brown et al . propose a model in which negative feedback within the substantia nigra—largely overlooked by previous models—regulates the output of the basal ganglia and thus contributes to the control of movement . Most areas of the brain contain projection neurons , which connect to other areas of the brain , and interneurons , which do not form connections beyond the nucleus in which they reside . In these areas , dedicated networks of interneurons use feedback to exert control over the signals that the projection neurons carry to other areas of the brain . However , it is thought that the substantia nigra does not contain interneurons . This led Brown et al . to propose that structures called axon collaterals form a microcircuit that can instead supply such feedback in the substantia nigra . Axons are the nerve fibres that carry signals away from the cell body of a neuron , and axon collaterals are branches of those axons . Data obtained by recording and manipulating electrical activity in the substantia nigra were consistent with this model and further experiments allowed this microcircuit to be mapped in detail . By revealing the circuit mechanisms of negative feedback within the substantia nigra , the work of Brown et al . changes our understanding of the basal ganglia and could have implications for understanding the mechanisms and ultimately the treatment of disorders such as Parkinson’s disease .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2014	The inhibitory microcircuit of the substantia nigra provides feedback gain control of the basal ganglia output
To understand brain circuits it is necessary both to record and manipulate their activity . Transcranial ultrasound stimulation ( TUS ) is a promising non-invasive brain stimulation technique . To date , investigations report short-lived neuromodulatory effects , but to deliver on its full potential for research and therapy , ultrasound protocols are required that induce longer-lasting ‘offline’ changes . Here , we present a TUS protocol that modulates brain activation in macaques for more than one hour after 40 s of stimulation , while circumventing auditory confounds . Normally activity in brain areas reflects activity in interconnected regions but TUS caused stimulated areas to interact more selectively with the rest of the brain . In a within-subject design , we observe regionally specific TUS effects for two medial frontal brain regions – supplementary motor area and frontal polar cortex . Independently of these site-specific effects , TUS also induced signal changes in the meningeal compartment . TUS effects were temporary and not associated with microstructural changes . In neuroscience , to understand brain circuits a two-pronged approach , entailing both recording and manipulating brain activity , is essential . In recent years , there has been extensive progress in this field , which was in part made possible by the availability of new technologies ( Bestmann and Walsh , 2017; Dayan et al . , 2013; Polanía et al . , 2018 ) . While techniques for transiently manipulating activity in rodents , such as microstimulation , optogenetics , and chemogenetics ( Sternson and Roth , 2014; Vanduffel , 2016 ) , are increasingly accessible and applied , techniques for manipulating activity in the primate brain are less widely available and remain accessible to comparatively few researchers in a limited number of research centres worldwide ( Galvan et al . , 2018; Krug et al . , 2015; Vanduffel , 2016 ) . A prominent limitation for many brain stimulation tools for research and therapeutic interventions is the duration of the induced neuromodulatory effects , often not outlasting the stimulation by more than a few seconds or minutes . Here we report on a particular protocol of low intensity pulsed transcranial focused ultrasound stimulation ( TUS ) that we show induces a sustained period of neuromodulation in primates without inducing structural damage . The TUS approach in general is a relatively new one ( Dallapiazza et al . , 2018; Tufail et al . , 2011; Tyler et al . , 2018; Yoo et al . , 2011; Younan et al . , 2013 ) and like transcranial magnetic stimulation ( Dayan et al . , 2013 ) and transcranial electric stimulation ( Polanía et al . , 2018 ) it can be applied in the absence of a craniotomy . In vitro recordings have identified several mechanisms by which ultrasound stimulation could affect neurons . It has been proposed that the sound pressure wave exerts a mechanical effect on neuronal activity through ion channel gating and changes to the membrane capacitance ( Blackmore et al . , 2018; Kubanek et al . , 2018; Kubanek et al . , 2016; Prieto et al . , 2013 ) . While the precise mechanism is being determined , early applications , including the current results , suggest TUS may be suitable as a tool for focal manipulation of activity in many brain areas in primates ( Fomenko et al . , 2018; Lee et al . , 2016a; Legon et al . , 2018; Munoz et al . , 2018; Naor et al . , 2016; Tufail et al . , 2011; Tyler et al . , 2018; Yoo et al . , 2011 ) . In the macaque , its application over the frontal eye field ( FEF ) affects the same aspects of oculomotor behaviour that are compromised by FEF lesion , whilst leaving intact those aspects of oculomotor behaviour that are unaffected by FEF lesion ( Deffieux et al . , 2013 ) . Pioneering work in humans has focused on modulating and eliciting evoked responses by focused ultrasound in both cortical and subcortical sensorimotor regions ( Lee et al . , 2016a; Lee et al . , 2015; Legon et al . , 2018; Legon et al . , 2014 ) . To date ultrasonic applications are primarily focused on direct ‘online’ effects . Nonetheless , some studies have observed neuromodulatory effects outlasting the sonication by several minutes before returning to baseline , with more pronounced effects for protocols delivered at higher intensities ( Isppa >5 W/cm2 ) and higher duty cycles ( >5%; Kim et al . , 2015; Yoo et al . , 2011 ) . For example , following ultrasound protocols with a relatively high rate of acoustic energy deposition , characterized by pulses tens of milliseconds long , repeated at 10 Hz , the neuromodulatory effects outlast the sonication period: for up to 10 min following 43 . 7 ms long 1 . 14MHz pulses repeated at 10 Hz for 40 s ( Dallapiazza et al . , 2018 ) , or for up to 20 min following 30 ms long 350 kHz pulses repeated at 10 Hz for 40 s ( Ahnine et al . , 2018 ) . We have built on such protocols when designing the current experiment . Recent work has highlighted the possibility that in rodents some TUS protocols can evoke a startle response when the stimulation is modulated at audible frequencies ( Guo et al . , 2018; Sato et al . , 2018 ) . Importantly , this auditory effect is limited to the stimulation period and dissipates within 75 ms to 4 s . This work has also emphasized that in more deeply anaesthetized animals , when the intrinsic neural activity of a system is suppressed , some TUS protocols might fail to evoke action potentials at the site of stimulation . This suggests TUS’ actions might be primarily neuromodulatory in nature and/or that they are most prominently observable when they interact with ongoing physiological activity . However , as has been noted ( Airan and Butts Pauly , 2018 ) the impact that any stimulation protocol might have , will be a function of the animal model being used and the precise details of the ultrasound frequencies , pulse shape , protocol , and ongoing brain activity . With the offline protocol and anaesthesia regime we used we control for such potential artefacts and show that TUS has an effect that cannot be attributed to them . Here we focused on the effects of TUS outlasting the stimulation period , investigating the impact of 40 s trains of TUS on measurements of neural activity in three macaque monkeys provided by functional magnetic resonance imaging ( fMRI ) up to 2 hr after stimulation ( Figure 1 , top panel ) . FMRI is one of the most widely used methods for estimating neural activity . Despite limitations in its spatial and temporal resolution , fMRI remains important because it is non-invasive and can often be used to provide information about activity throughout the whole brain . Rather than providing a direct measure of neural activity , however , it provides an estimate of how activity changes in tandem with sensory , cognitive , or motor events or with activity in another brain region . Typically , fMRI-measured activity in any given brain area is a function of activity in other brain areas , especially those with which it is closely interconnected ( Neubert et al . , 2015; O'Reilly et al . , 2013 ) . Although many brain areas may share any given individual connection ( for example both areas A and B may project to C ) , the overall pattern of connections of each area is unique ( Passingham et al . , 2002 ) ; as such , the overall pattern of connections therefore constitutes a ‘connectional fingerprint’ . As a result it is possible to use fMRI measurements of correlations in the blood oxygen level dependent ( BOLD ) signal across brain regions to estimate the connectivity fingerprints of a given brain area ( Figure 1 , bottom right panel; Caspari et al . , 2018; Ghahremani et al . , 2017; Margulies et al . , 2016 , Margulies et al . , 2009; Mars et al . , 2013; Sallet et al . , 2013; Shen , 2015; Shen et al . , 2015 ) . This implies that activity in any given brain area is a function of the activity in the areas with which it is interconnected . We exploited this feature of activity to examine the impact of TUS application to two brain regions in the frontal cortex: supplementary motor area ( SMA; experiment 1 ) and frontal polar cortex ( FPC; experiments 2 and 3 ) . Simulations showed we were able to selectively target these regions ( Figure 2; see Acoustic and thermal modelling for more details ) . These regions have distinct anatomical and functional connections; SMA is most strongly coupled with the sensorimotor system , while the FPC interacts primarily with the prefrontal cortex and only interacts indirectly with the sensorimotor system via SMA ( Petrides and Pandya , 2007; Sallet et al . , 2013 ) . This allows us to test the spatial and connectional specificity of TUS effects . In the control state , each area’s activity is normally a function of the activity in the areas that constitute its connectional fingerprint . If this pattern is altered by TUS in a manner that is dependent on the location of the stimulation , then this will constitute evidence that TUS exerts a spatially selective effect on neural activity . In a within-subject design , the same three animals participated in experiments 1 and 2 , and a control experiment conducted in the absence of TUS ( Figure 1 ) . As such , fMRI was acquired in three conditions for all animals: following SMA TUS , following FPC TUS , and in a control state . In turn , each of the MRI sessions consisted of three consecutive runs . We also validated the results of experiment two in a new set of three different animals ( experiment 3 ) . Ultrasound modulation of SMA and FPC led to focal , area-specific changes in each stimulated region’s connectivity profile . In each area , a region’s activity pattern after TUS application was more a function of its own activity and that of strongly connected regions , but less a function of activity in more remote and weakly interconnected areas . Independent of these specific grey matter signal changes , TUS also interacted with non-neuronal structures , as evidenced by more widespread signal changes observed in the meningeal compartment ( including cerebral spinal fluid and vasculature ) . Finally , in experiment four we demonstrate that TUS application had no observable impact on cortical microstructure apparent on histological examination . In summary , TUS over dorsomedial frontal cortex causes spatially specific sharpening of the stimulated region’s connectivity profile with high efficacy and reproducibility , independent of non-neuronal signal changes . Transcranial focused ultrasound stimulation of SMA induced spatially specific changes to the connectivity profile of SMA . At rest , in the control state , SMA’s activity is coupled with activity throughout sensorimotor regions in frontal and parietal cortex , inferior parietal , prefrontal , and parts of cingulate cortex ( Figure 3a ) . Many of these regions are anatomically connected with SMA ( Dum and Strick , 2005; Geyer et al . , 2000; Strick et al . , 1998 ) . If we change the responsiveness of SMA neurons to such activity in interconnected regions by artificially modulating SMA activity with TUS , then we should see a change in the coupling between SMA activity and activity in other regions . Following ultrasound stimulation , SMA changed its coupling with the sensorimotor system , anterior and posterior cingulate , anterior temporal , inferior parietal , and prefrontal cortex ( Figure 3b ) . This can be seen on the whole brain functional connectivity maps for the SMA region ( Figure 3 , compare panels a and b , representative changes highlighted by dashed black circles ) and on the whole brain differential connectivity maps in Figure 4 ( panels b and c ) . It is also apparent in the illustration of SMA’s connectional fingerprint ( Figure 5a ) . The distance of each coloured line from the centre of the figure ( and hence its proximity to the circumference of the figure ) indicates the strength of activity coupling between SMA and each of the other brain areas indicated on the circumference . Compared to the control state ( blue line ) , after TUS over SMA ( red line ) , SMA’s positive coupling is enhanced with proximal areas in the sensorimotor system but reduced in many long-range connections ( non-parametric permutation test , p=0 . 017 ) . The primary motor cortex ( M1 ) , superior parietal lobe ( SPL ) , and middle cingulate cortex ( MCC ) in the dorsomedial sensorimotor network have been reported to be closely connected with the SMA , whereas prefrontal regions on the dorsomedial ( area 9m and FPC ) , dorsolateral ( areas 9-46d and 8A ) , and ventromedial ( area 11m ) surface , and those in the temporal lobe ( anterior superior temporal gyrus , aSTG; middle superior temporal sulcus , midSTS ) , and parietal cortex ( caudal inferior parietal lobule , IPLc; posterior parietal cortex , PCC ) have been reported to be less closely connected with the SMA ( Dum and Strick , 2005; Geyer et al . , 2000; Strick et al . , 1998 ) . TUS increased positive coupling between the stimulated area and proximal areas normally closely connected with it , while , at the same time , decreasing coupling between the stimulated area and many areas normally less closely connected with it . This pattern not only emerges from the fingerprint analyses , but constitutes a principle evident across the brain , as illustrated by whole-brain differential SMA-connectivity maps of the effect of SMA TUS ( Figure 4 ) . These specific effects of TUS were sustained over the duration of our experiment , lasting up to 2 hr ( Figure 5d–f ) . Disruptive effects of TUS on long-range coupling were especially prominent immediately following the end of TUS application ( Figure 5d ) , and gradually reduced towards the end of our recording session ( Figure 5f ) . The enhancing effects that TUS exerted on SMA’s coupling with adjacent and strongly connected areas had a relatively delayed appearance , arising well after the TUS had ended ( more than 1 hr , Figure 5e ) , but again decreasing towards the end of the recording session . In summary , the most important finding was of a protracted period of connectivity change after TUS . While we are cautious about overinterpreting precise timing differences between long-range connectivity reductions and local connectivity increments , we note that the observed pattern could signify distinct time courses for TUS-induced long-term depression and long-term potentiation . However , the observed pattern is also consistent with the notion that early disruption of long-range input to a network ( Figure 5d ) leaves relatively more signal variance in this network to be explained by remaining local input , driving the observation of subsequent within-network coupling increments ( Figure 5e ) . Both these mechanisms would lead to a sharpening of the stimulated region’s connectivity profile , as observed here . Like SMA , FPC’s activity is coupled with that in interconnected brain regions even when animals are at rest in the control state ( Figure 3d ) . FPC’s activity is positively correlated with activity in a number of adjacent dorsomedial and lateral prefrontal areas and in the central portion of the superior temporal sulcus ( midSTS ) and posterior cingulate cortex with which it is monosynaptically interconnected ( Petrides and Pandya , 2007 ) . By contrast there was , at a rest , a negative relationship with the activity in sensorimotor areas , with which FPC is indirectly connected via regions such as SMA . In comparison to the control state , FPC stimulation induced an enhancement in the normal short-range connectivity between FPC and adjacent dorsomedial ( area 9m ) and lateral prefrontal cortex ( area 9-46d ) with which it is particularly strongly connected . In addition , a similar effect was seen in more distant regions with which it is also strongly connected – the midSTS , IPLc and PCC , together comprising temporal and parietal segments of the primate ‘default mode network’ ( Petrides and Pandya , 2007 ) . By contrast , there was reduced coupling with other areas in prefrontal cortex , including ventromedial ( area 14m ) , subgenual cingulate ( area 25 ) , and lateral orbitofrontal cortex ( area 47-12o ) . These are all areas that FPC is connected to but less strongly ( Petrides and Pandya , 2007 ) . Finally , TUS applied to FPC also led to a change in long-range connectivity between FPC and several motor association regions with which it is not directly connected , especially those in the ventrolateral parieto-frontal sensorimotor network ( areas PF and F4 Petrides and Pandya , 2007 ) . As noted , in the control state , the activity in FPC and these sensorimotor association areas is negatively or anti-correlated , but this anti-correlation was reduced by FPC TUS . These results are apparent in the whole brain functional connectivity maps for the FPC region ( Figure 3 , compare panels d and f , representative changes highlighted by dashed black circles ) and on the whole brain differential connectivity maps in Figure 4 ( panels d and f ) . It is also apparent in the illustration of FPC’s connectional fingerprint ( Figure 5b ) . Here the blue line indicates the strength of activity coupling between FPC and each of the other brain areas indicated on the circumference in the control state . The yellow line shows that FPC’s coupling with each area is changed after FPC TUS ( non-parametric permutation test , p=0 . 027 ) . It is important to test the claim that TUS induces effects that are spatially specific to each sonicated area by directly comparing effects between stimulation sites . Although FPC TUS significantly altered FPC functional connectivity , it had comparatively little impact on SMA’s pattern of functional connectivity; there was no difference in SMA’s functional connectivity between the control state and after FPC TUS ( non-parametric permutation test , p=0 . 231; whole-brain map in Figure 3c and yellow line in connectivity fingerprint in Figure 5a ) . Importantly , the effects of TUS over SMA on SMA’s connectivity were significantly dissociable from the effects of FPC TUS ( non-parametric permutation test , p=0 . 041 ) . Similarly , SMA TUS had some but comparatively little impact on FPC’s pattern of functional connectivity ( non-parametric permutation tests , SMA versus control , p=0 . 047; SMA versus FPC , p=0 . 028; whole-brain map in Figure 3e and red line in connectivity fingerprint in Figure 5b ) . In fact , the most prominent changes in each area’s connectional fingerprint that were induced by stimulation of the other area were the disruption of functional connectivity between FPC and sensorimotor areas when SMA was stimulated ( Figure 3e , encircled ) . This particular result may have occurred because , as already noted , FPC has no direct monosynaptic connections with these sensorimotor areas ( Petrides and Pandya , 2007 ) and so its functional coupling with these areas is likely to be mediated by areas such as SMA and the areas that surround it such as the pre-supplementary motor area and the cingulate motor areas ( Bates and Goldman-Rakic , 1993; Lu et al . , 1994 ) . In fact , these circumscribed exceptions may well confirm the rule that a region’s connectivity pattern is only affected by stimulation of the region itself . Namely , SMA TUS only affects FPC coupling with SMA itself and the regions which are coupled with FPC only through SMA . In conclusion , the effects of TUS in different frontal regions were clearly dissociable . We investigated the reproducibility of TUS effects by examining the impact of TUS to FPC in three additional individuals in a biological replication experiment ( experiment 3 , Figure 3—figure supplement 1 ) . TUS had the same effects as seen in experiment 2: a site-specific sharpening of the stimulated region’s connectional profile . While experiments 1 and 2 followed a within-subject design , for experiment three we conducted a between-subject analysis where the subjects in the two experiments differed in age . We are therefore careful not to draw too strong conclusions on any main effect of subject group but focus on the interaction of the TUS effect with the fingerprint shape . Notwithstanding , when reviewing the simple effects driving the well-matched interaction , we note that the observed sharpening was less prominently related to activity coupling decreases in experiment 3 than in experiment 2 . Putatively , parameters of the general anaesthesia could impact on the effect of TUS . However , subjects used in experiment 2 and 3 did not differ regarding depth of anaesthesia or duration between sedation and fMRI data collection . It has recently been suggested that certain TUS protocols might have a limited efficacy in evoking spiking activity at the stimulation site , but rather exert their influence on the brain through the auditory system , not unlike an auditory startle response ( Guo et al . , 2018; Sato et al . , 2018 ) . Although these online observations in rodents should perhaps not be extrapolated far outside the tested conditions ( for example to our measurements taken tens of minutes after the stimulation ended ) , these observations do argue in favour of performing controlled experiments that address and exclude such confounds ( Airan and Butts Pauly , 2018 ) . Here we consider , first , why neural effects of TUS might be evident in the current study when they were not clear previously . Second , we consider whether neural effects may be due to an auditory artefact . First , it is possible that the efficacy of TUS is a function of both the specifics of the stimulation protocol and of ongoing neural activity . A neuromodulatory technique may fail to elicit spiking activity in deeply anaesthetized rodents ( Guo et al . , 2018; Sato et al . , 2018 ) . However , in this study we specifically test whether it simply modulates ongoing activity , while adopting lighter anaesthesia levels . Importantly , it is known that whole-brain functional connectivity , as measured with the BOLD signal , is preserved at these levels ( Mars et al . , 2013; Neubert et al . , 2015; Neubert et al . , 2014; O'Reilly et al . , 2013; Sallet et al . , 2013; Vincent et al . , 2007 ) . Finally , the experiments were conducted in a primate model as opposed to a rodent model; the importance of species-specific effects in TUS models are currently unknown . Under these distinct conditions , we observed that TUS modulated the activity coupling of each stimulated area in a regionally specific manner . Nevertheless , following these investigations of the effect of TUS on whole-brain connectivity patterns of the stimulated regions , we carried out a second line of investigation and examined the effect of TUS on the signal in the stimulated regions themselves ( Figure 5c ) . While BOLD fMRI cannot provide an absolute measure of neural activity , we can characterize how homogeneous the activation signal is within the stimulated region , as quantified by the coupling strength of the signal at each point in the stimulated region of interest to all other points in that region . This analysis revealed that TUS induced more spatially homogenous activation within the stimulated area , but not in the non-stimulated region ( interaction of TUS x connectivity seed: F ( 1 , 8 ) =1571 . 2 , p=1 . 8044e-10 , d = 11 . 4426 , CI=[1 . 2733 1 . 1333] ) . This effect on spatial homogeneity of the signal was not accompanied by changes to the temporal variance of the BOLD signal fluctuations in the stimulated or other regions ( Figure 5—figure supplement 1 ) . In fact , the standard deviation of the BOLD signal fluctuations over time were strikingly similar between the two stimulation sites ( SMA and FPC , highlighted in Figure 5—figure supplement 1 ) and across the different experimental conditions ( control , SMA TUS , and FPC TUS ) . This suggests that TUS leaves intact basic haemodynamics and neurophysiology and instead has a circumscribed and specific impact on the coupling of the stimulated region with the rest of the brain . Third , the presence of auditory and somatosensory confounds is likely to be a function of the specifics of the TUS protocol . In sonication protocols the ultrasound wave is often pulse modulated at ~1 kHz , well within the audible range of many rodents and primates . At these modulation frequencies auditory stimulation is perhaps not unexpected ( Guo et al . , 2018; Sato et al . , 2018 ) . This is something that we have avoided in our work with macaques: we pulse modulated the 250 kHz ultrasound wave at 10 Hz: as such we ensured that the frequency of both the ultrasound wave and its modulating envelope are well outside of the macaque hearing range . Moreover , here we adopted an offline experimental design where any potential audible stimulation associated with the TUS application was limited to the 40 s sonication period , while the neural activation measures were initiated tens of minutes later . Furthermore , the specificity of our results strengthens the suggestion that it might not be possible to explain away the current findings as the result of an auditory artefact having occurred up to two hours earlier . Nevertheless , we also carried out a fourth line of inquiry and examined the activation in the primary auditory cortex ( A1 ) and its relationship with activity in the rest of the brain ( Figure 5g–i ) . The effects of TUS on SMA and FPC coupling patterns , as quantified in their connectional fingerprints , could not be explained by a potential impact of TUS on A1 activity coupling . In fact , TUS did not affect A1 activity coupling with SMA fingerprint targets ( Figure 5g; non-parametric permutation tests , SMA TUS: p=0 . 8234 , FPC TUS p=0 . 3452 ) , nor its coupling with FPC fingerprint targets ( Figure 5h; non-parametric permutation tests , SMA TUS: p=0 . 5411 , FPC TUS p=0 . 2667 ) . Moreover , neither SMA nor FPC changed its coupling with A1 as a function of TUS ( Figure 5i; main effect of TUS: F ( 1 , 8 ) =0 . 015445 , p=0 . 90416 , d = 0 . 03583 , CI=[−0 . 41483 0 . 46209]; interaction of TUS x connectivity seed: F ( 1 , 8 ) =0 . 18284 , p=0 . 68022 , d = 0 . 1234 , CI=[−0 . 25058 0 . 36466] ) . The ability of ultrasound to reversibly interact with biological tissue is not limited to grey matter . We were aware that our ultrasonic beam , placed over the central midline to target SMA or FPC in both hemispheres simultaneously , was also likely to reach the meningeal compartment in the interhemispheric fissure . In fMRI analyses this region is sometimes referred to as ‘cerebral spinal fluid’ , although in reality it contains the cortical membranes ( dura , arachnoid , and pia mater ) , some cerebral spinal fluid , and important vascular structures , such as the superior sagittal sinus . Ultrasound protocols designed to induce vasodilation or to temporally open the blood-brain-barrier are conventionally markedly distinct from those employed here , for example using higher intensities or supplemented with intravenously injected microbubbles . Nonetheless , we set out to test the influence of ultrasound on what we shall continue to refer to as the ‘meningeal’ signal and the grey matter signal . To do this it was obviously necessary to take a somewhat unconventional rs-fMRI analysis approach that did not remove the meningeal signal that is typically regarded as a confound . A principal component analysis of the signal in the meningeal and grey matter compartments revealed the main components in either compartment explained significantly more variance following TUS compared to control ( Figure 6c; main effect of TUS: F ( 1 , 9 ) =30 . 6 , p=0 . 00036 , d = 0 . 67031 , CI=[2 . 1916 5 . 3343]; in grey matter: F ( 1 , 4 ) =10 . 743 , p=0 . 0306 , d = 1 . 3381 , CI=[0 . 47735 5 . 7655]; in meningeal compartment: F ( 1 , 4 ) =16 . 263 , p=0 . 0157 , d = 1 . 6464 , CI=[1 . 3379 7 . 2512] ) . The fact that this effect is present in both compartments could reflect the tight vascular coupling between grey matter and meningeal signal or be driven by partial-voluming effects ( these are more pronounced when the size of the brain is relatively small , as for monkey fMRI ) . This observation suggests that after TUS the BOLD signal became more homogenous . In a seed-based connectivity analysis , as performed here , this would be reflected in a stronger contribution of global signal coupling . However , the impact of TUS presented above does not seem to exhibit this effect , as illustrated by the specificity of the TUS effects for SMA and FPC ( Figures 3 , 5 ) , and further underpinned by the absence of TUS effects in regions remote from the stimulation sites ( for example , Figure 6a , d illustrates the case of the posterior parietal operculum , POp ) . Importantly , these coupling estimates are obtained when following the conventional rs-fMRI analysis approach to account for global signal confounds by removing WM and meningeal signal contributions before estimating grey matter coupling indices . We hypothesized that if TUS leads to more homogenous global signal , its contribution to the grey matter signal might have been accounted for when removing meningeal signal components in a linear regression framework ( Verhagen , 2012 ) . Accordingly , we have repeated the seed-based connectivity analyses after accounting for global signal confounds based on the white matter compartment alone , excluding the meningeal compartment . In these data , global signal contributions were indeed preserved as evidenced by anatomically implausible global connectivity patterns present in the control state ( compare panels a and b in Figure 6 , e . g . the prefrontal cortex , encircled ) . As such , this procedure allowed us to interrogate the global effects of TUS over dorsomedial frontal regions on BOLD signal ( see Materials and Methods for full details ) . Following this procedure , we observed that SMA stimulation appeared to induce widespread increases in signal coupling compared to the control state . This effect was not limited to the stimulation site but also present in remote regions ( e . g . when seeded in POp , compare panels b and e in Figure 6 ) . This effect can be quantified by considering the strength of local connections for every point in the cortex . It is then apparent that the changes induced by stimulation are global in nature ( Figure 6f ) . In general , when not fully accounting for global confounds , a region’s connectivity profile after TUS could be predicted by considering its profile in the control state and adding a spatially flat constant . This suggests an additive non-neuronal source , captured by signal components in the meningeal compartment , may explain the presence and enhancement of global signal observed following TUS over SMA ( Figure 6e , f , e . g . primary sensorimotor cortex ) . These effects of TUS on widespread coupling mediated by meningeal signal persisted over time for more than 1 hr after stimulation had ended ( Figure 6g ) . Similar effects of TUS on meningeal signal were also observed after FPC TUS ( Figure 6—figure supplement 1 ) . We note that in unconfounded fMRI data FPC TUS did not have a strong impact outside the stimulated region ( illustrated for POp in Figure 6—figure supplement 1 ) . In contrast , in fMRI data confounded by meningeal-driven global nuisance , FPC stimulation led to widespread enhanced signal coupling compared to the control state . This effect of FPC TUS was replicated in a new set of animals in experiment 3 ( Figure 6—figure supplement 1 ) . Although the effect was perhaps not as strong as that seen after SMA TUS , this finding again suggests that TUS over the medial meningeal compartment may produce widespread changes that are non-neuronal in origin . Differences in morphology of the sagittal sinus along the rostro-caudal axis might explain this weakened global effect . Some higher intensity ultrasound stimulation protocols , distinct from those used here , have been shown to induce thermal lesions or haemorrhage following cavitation ( Elias et al . , 2013 ) . Despite the fact that 40 s trains of TUS induced sustained changes in the post-stimulation period in experiments 1 and 2 , no structural changes remotely resembling those seen with higher intensity ultrasound protocols were observed . First of all , we did not observe any indication of TUS-induced oedema when comparing T1w MRI structural scans collected in baseline sessions with T1w scans collected after TUS ( Figure 7 ) . We also did not observe tissue alteration ( e . g . tissue burn ) at the post-mortem examination . Neither were any signs of neuronal alteration or haemorrhage observed in histological analyses of three macaques following pre-SMA TUS ( Figure 8 ) . To quantify the pressure amplitude , peak intensities , spatial distribution , and potential temperature changes in the monkey brain associated with the TUS protocol used in this study we simulated the acoustic wave propagation and its thermal effect in a whole head finite element model based on a high-resolution monkey CT scan . As estimated by these numerical simulations , the maximum spatial-peak pulse-averaged intensity ( Isppa ) at the acoustic focus point was 24 . 1 W/cm2 for the SMA target and 31 . 7 W/cm2 for the FPC target ( spatial peak temporal average intensities , Ispta: 7 . 2 W/cm2 and 9 . 5 W/cm2 for SMA and FPC , respectively ) . Given that the skull is more acoustically absorbing than soft tissue , the highest thermal increase is located in the skull itself , estimated by the simulation to be 2 . 9°C . Given an approximate 0 . 5 mm thickness of the dura ( Galashan et al . , 2011 ) the maximum temperature below the dura was 38 . 0°C . The maximal thermal increase at the geometrical focus of the sonic transducer was less than 0 . 5°C ( Figure 2 ) . In this study we demonstrate a protocol for transcranial focused ultrasound stimulation ( TUS ) that can induce a sustained yet reversible change in neural activity . We focused our investigation on modulations of brain connectivity , following the notion that each brain area’s unique contribution to cognition and behaviour is shaped by how activity in each area is a function of a unique fingerprint of interconnected areas ( Passingham et al . , 2002 ) . We found that each area’s connectional fingerprint was significantly changed by TUS , but only when it was applied to that area itself ( Figures 3 , 5 ) . The changes observed might be summarized as more uniform activation in the stimulated region combined with a sharpening of the normal coupling pattern that each area has even at rest . Activity coupling with strongly interconnected areas , which are often nearby , was increased but activity coupling with less strongly connected regions was reduced . Such changes in connectional fingerprints might constitute the mechanism by which TUS is able to induce regionally specific patterns of behavioural change when applied in awake behaving animals ( Deffieux et al . , 2013; Fouragnan et al . , 2019 ) . The pattern of inputs each area receives from other areas and the influence it wields over other areas are a major determinant of its function and here we have shown that this pattern is altered by TUS . TUS may therefore provide a relatively straightforward method for sustained but reversible manipulation of specific components of neural circuits in the primate brain ( Wattiez et al . , 2017 ) . This may be important for investigating primate brain areas when homologues in non-primate species , such as rodents , are non-existent or disputed ( Preuss , 1995; Wise , 2008 ) . This work paves the way for the development and use of offline TUS protocols in primates , including humans , both as a research tool and as potential clinical intervention . In experiments 1 and 2 , adopting a within-subject design with three animals , we found that TUS application produced different effects when applied to different brain regions: the SMA , a part of motor association cortex , and FPC a part of granular prefrontal cortex ( Figures 3 , 5 ) . However , in each case the TUS effects were prominent within the connectional fingerprint of the area stimulated . The connectional fingerprints of SMA and FPC are distinct ( Johansen-Berg et al . , 2004; Neubert et al . , 2014; Sallet et al . , 2013 ) . The effects of TUS are thus regionally specific . Our results confirm that TUS can be used as a neuromodulatory technique that allows one to non-surgically target cortical and subcortical brain areas with superior spatial specificity and depth of stimulation ( Folloni et al . , 2019 ) compared to other transcranial stimulation approaches ( e . g . TMS and TCS; Bestmann and Walsh , 2017; Dayan et al . , 2013; Polanía et al . , 2018 ) . While successes have been achieved with some invasive techniques , such as electrical microstimulation ( Krug et al . , 2015; Vanduffel , 2016 ) , it is not easy to use them to disrupt activity in all areas especially when they are not somatotopically mapped . Recently it has been reported that some online TUS protocols in rodents induce neural changes as an indirect consequence of the auditory stimulation they entail ( Guo et al . , 2018; Sato et al . , 2018 ) . The spatially specific effects that we observed after TUS cannot , however , be attributed to any common auditory impact that occurs at the time of stimulation . Moreover , in order to avoid both the confounding effect associated with the sound of TUS and interference of the ultrasonic wave field with fMRI measurement , we opted for an ‘offline’ stimulation protocol . Stimulation was a 40 s train that ended at least 20 min before the fMRI data acquisition period . The sustained nature of the train and other features of the stimulation pulses may make the protocol used here more effective for neuromodulation , while ensuring the thermal modulation of the cortex remains limited ( <1°C , Figure 2 ) . Such limited thermal changes are not associated with neuromodulatory effects observed more than 30 min after the stimulation: the thermal rise is short-lived ( Dallapiazza et al . , 2018 ) , not accompanied by tissue damage and below the thermal effects observed with some protocols in rodents ( for a review , see Constans et al . , 2018 ) . In experiment 3 we found that TUS had reproducible effects . When TUS was applied to FPC in three other individuals , it induced spatially specific effects similar to those seen in experiment 2 . The effects of the non-invasive 40 s stimulation protocol used here are sustained and lasted over much of the two-hour period we investigated . These effects are more extended than those produced by other techniques commonly used for offline disruption of cortical activity such as TMS ( Huang et al . , 2005; O'Shea et al . , 2007 ) . Care should therefore be taken in using the technique in human cognitive neuroscience experiments; TUS effects may continue beyond the short periods that participants typically spend within the laboratory . It may therefore be important to carefully characterize the time course of TUS effects in animal models before their use with human participants . Some caution might also be warranted in relation to the potential of ultrasound to cause microstructural damage , especially when stimulating at higher intensities , longer durations , or for more repetitions . While our structural MRI and histological analyses did not reveal any evidence of damage and were comparable to previous studies ( Dallapiazza et al . , 2018; Lee et al . , 2016b ) , it would be of interest in the future to include additional histological indices of apoptotic or inflammatory processes ( Tufail et al . , 2010 ) to further assess the safety of TUS . Similarly , the thermal modelling approach we adopted here was designed to estimate an informed upper-bound on potential thermal effects , but when further developing this protocol additional simulation validations – for example based on phantom measurements – might be informative . In addition to the spatially specific effects of TUS we also observed changes in the BOLD signal originating from the meningeal compartment that were not specific to the area stimulated . Although similar effects were seen each time either the SMA or FPC was stimulated , the effect was generally more pronounced after SMA stimulation than after FPC stimulation . Our preliminary results from TUS of other brain areas suggest that these spatially non-specific effects may be even smaller when TUS is applied elsewhere ( Folloni et al . , 2019 ) . While the precise origin of the non-specific effects was difficult to determine it is possible that they may result from a direct vascular effect of TUS; the sagittal sinus is directly above the midline frontal regions that we targeted and is included in the meningeal compartment during MRI analysis protocols . The presence of such non-specific effects again underlines the need for care in translating the technique to humans; especially when the targeted region is near venous sinuses or cerebral arteries . In addition , they underline the need for comparing the behavioural effects of TUS not just with a non-stimulation sham condition but with TUS application to another control brain region . The absence of marked histological changes in experiment 4 , however , provides one important safety benchmark and confirms previous histological results in lagomorphs ( Yoo et al . , 2011 ) . Combining TUS and fMRI is a promising approach to overcome the restrictions of each of the individual techniques . Here we have shown that TUS has a detectable offline and sustained impact on the distinctive network of connectivity associated with the stimulated brain region – the connectional fingerprint . A brain region’s interactions with other regions – its unique connectional fingerprint or specific pattern of inputs and outputs – are an important determinant of its functional role . The current results are therefore consistent with TUS application exerting regionally specific effects on behaviour ( Deffieux et al . , 2013; Fouragnan et al . , 2019 ) . The fact that fMRI allows the effects of TUS to be studied with a high spatial resolution suggests the TUS-fMRI combination has the potential to become a powerful neuroscientific tool . For this study , six healthy male macaques ( Macaca mulatta , NCBITaxon:9544 ) were stimulated with transcranial focused ultrasound and scanned to acquire resting state functional magnetic resonance images ( rs-fMRI ) and anatomical MR images . Three animals participated in experiment 1 [SMA TUS] ( all males , mean age and weight at time of scan: 5 . 6 years , 10 . 7 kg ) . The same three animals participated in experiment 2 [FPC TUS] ( at time of scan: 6 . 1 years , 11 . 8 kg ) , and the control condition ( at time of scan: 5 . 5 years , 10 . 2 kg ) . Three different animals participated in experiment 3 [FPC TUS validation] ( all males , mean age and weight at time of scan: 10 . 3 years , 13 kg ) . In addition to this set of animals , six animals were included in the histology analysis ( experiment 4 ) : three control animals who did not receive TUS ( two females; mean age and weight at time of perfusion: 9 . 3 years , 9 . 1 kg ) and three pre-SMA TUS animals post stimulation ( all males; mean age and weight at time of perfusion: 8 . 4 years , 13 . 1 kg ) . All procedures were conducted under licenses from the United Kingdom ( UK ) Home Office in accordance with The Animals ( Scientific Procedures ) Act 1986 . In all cases they complied with the European Union guidelines ( EU Directive 2010/63/EU ) . A single element ultrasound transducer ( H115-MR , diameter 64 mm , Sonic Concept , Bothell , WA , USA ) with 51 . 74 mm focal depth was used with a coupling cone filled with degassed water and sealed with a latex membrane ( Durex ) . The resonance frequency of the ultrasonic wave was set at 250 kHz with 30 ms bursts of ultrasound generated every 100 ms , controlled through a digital function generator ( Handyscope HS5 , TiePie engineering , Sneek , The Netherlands ) . The stimulation lasted for 40 s . A 75-Watt amplifier ( 75A250A , Amplifier Research , Souderton , PA ) was used to deliver the required power to the transducer . A TiePie probe ( Handyscope HS5 , TiePie engineering , Sneek , The Netherlands ) connected to an oscilloscope was used to monitor the voltage delivered . The recorded peak-to-peak voltage was kept constant throughout the stimulation . Voltage values per session ranged from 130 to 142 V , corresponding to 1 . 17 to 1 . 35 MPa as measured in water with an in house heterodyne interferometer ( Constans et al . , 2017 ) . Based on numerical simulations ( see Acoustic and thermal modelling below for more details ) , the maximum peak pressure ( Pmax ) and Isppa at the acoustic focus point were estimated to be 0 . 88 MPa and 24 . 1 W/cm2 for the SMA target , and 1 . 01 MPa and 31 . 7 W/cm2 for the FPC target ( Ispta: 7 . 2 W/cm2 and 9 . 5 W/cm2 for SMA and FPC , respectively ) . Each of the areas targeted in experiments 1–4 lie close to the midline . Therefore , we applied a single train over the midline stimulating the target region in both hemispheres simultaneously . In order to direct TUS to the target region , we guided the stimulation using a frameless stereotaxic neuronavigation system ( Rogue Research , Montreal , CA; RRID:SCR_009539 ) set up for each animal individually by registering a T1-weighted MR image to the animal’s head . Positions of both the ultrasound transducer and the head of the animal were tracked continuously with infrared reflectors to inform online and accurate positioning of the transducer over the targeted brain region: SMA in experiment 1 , ( Montreal Neurological Institute ( MNI ) X , Y , and Z coordinates in mm [0 . 1 2 19] ) ; FPC in experiment 2 [0 . 6 24 10]; FPC in experiment 3 [-0 . 7 24 11]; pre-SMA in experiment 4 [0 . 2 11 17] . The ultrasound transducer/coupling cone montage was placed directly onto previously shaved skin prepared with conductive gel ( SignaGel Electrode; Parker Laboratories Inc . ) to ensure ultrasonic coupling between the transducer and the animal's scalp . In the non-stimulation condition ( control ) , all procedures ( anaesthesia , pre-scan preparation , fMRI scan acquisition and timing ) , with the exception of actual TUS , matched the TUS sessions . The acoustic wave propagation of our focused ultrasound protocol ( at 130 V peak-to-peak voltage ) was simulated using a k-space pseudospectral method-based solver , k-Wave ( Cox et al . , 2007 ) to obtain estimates for the pressure amplitude , peak intensity , spatial distribution , and thermal impact at steady state . 3D maps of the skull were extracted from a monkey CT scan ( Kyoto University online database , ID 1478 , 0 . 26 mm isotropic resolution ) . Soft tissues were assumed to be homogeneous , with acoustic values of water ( ρtissue =1000 kg/m3 and ctissue =1500 m/s ) . In the bone , a linear relationship between the Hounsfield Units ( HU ) from the CT scan and the sound speed , as well as the density , was used . The power law model for attenuation is αatt=α1∗ϕβ where the porosity ϕ is defined by ϕ=ρmax−ρρmax−ρwater in the skull ( Aubry et al . , 2003 ) . The attenuation coefficient for the acoustic propagation α1 depends on the frequency: α1=α0fb . We set the parameters to ρmax=2200 kg/m3 , cmax=3100 m/s , β=0 . 5 , α0=8 dB/cm/MHzb , b=1 . 1 ( Constans et al . , 2018 ) . The attenuation coefficient in bone accounts for both absorption and scattering . The propagation simulation was performed at 250 kHz with a 150µs-long pulse signal ( enough to reach a steady state ) . The transducer was modelled as a spherical section ( 63 mm radius of curvature and 64 mm active diameter ) . The simulated pulses were spatially apodized ( r = 0 . 35 ) on the spherical section . Ultrasound propagates first through water before entering the skull cavity with the geometrical focal point located below the surface , inside the brain . Simulations were performed in free water , and the maximum amplitude obtained was used to rescale the results in skull ( the transducer calibration indicates that the maximum amplitude in water at 130V is 1 . 2 MPa ) . The thermal modelling is based on the bio-heat equation ( Pennes , 1948 ) :ρC∂T∂t=κ∇2T+q+wρbCb ( T−Ta ) where T , ρ , C , κ and q are the temperature , density , specific heat , thermal conductivity and rate of heat production respectively . Heat production is defined as q=αabsP²2ρC , αabs being the absorption coefficient and P the peak negative pressure . κ is set to 0 . 528 W . m−1 . K−1 in soft tissue and 0 . 4 W . m−1 . K−1 in the skull; C is set to 3600 J . kg−1 . K−1 in soft tissue and 1300 J . kg−1 . K−1 in the skull ( Duck , 2013 ) . In the tissue , the absorption coefficient was set to αabs tissue=0 . 21 dB/cm/MHzb ( Goss et al . , 1979 ) . In the skull the longitudinal absorption coefficient is proportional to the density with αabs max=a0/3=2 . 7dB/cm/MHzb ( Pinton et al . , 2012 ) . The last term corresponds to the perfusion process: w , ρb , Cb , and Ta correspond to the blood perfusion rate , blood density , blood specific heat and blood ambient temperature respectively . These parameters are assumed homogeneous over the brain , although a more detailed description of the brain cooling processes can be found in the literature ( Wang et al . , 2015 ) . The perfusion parameters are based on previous reports ( Pulkkinen et al . , 2011 ) : w=0 . 008s−1; ρb= 1030 kg . m−3; Cb = 3620 J . kg−1 . K−1 and Ta = 37°C . The bioheat equation is solved by using a 3D finite-difference scheme in MATLAB ( Mathworks , Natick , USA ) with Dirichlet boundary conditions . Initial temperature conditions were 37°C in the brain , skull and tissue , and 24°C in the water coupling cone . Simulations were run over 1 min pre-sonication , followed by 40 s of sonication , and 5 min post-sonication , closely following the experimental procedure . We acquired one MRI session per monkey per condition: in total we performed three sessions per animal across experiments 1 and 2 , and one session per animal in experiment 3 . The ultrasound sonication and subsequent MRI scans were performed under inhalational isoflurane gas anaesthesia using a protocol which has previously proven successful in preserving whole-brain functional connectivity as measured with BOLD signal ( Mars et al . , 2013; Mars et al . , 2011; Neubert et al . , 2015; Neubert et al . , 2014; O'Reilly et al . , 2013; Sallet et al . , 2013; Vincent et al . , 2007 ) . In the case of the TUS conditions , fMRI data collection began only after completion of the TUS train . Anaesthesia was induced using intramuscular injection of ketamine ( 10 mg/kg ) , xylazine ( 0 . 125–0 . 25 mg/kg ) , and midazolam ( 0 . 1 mg/kg ) . Macaques also received injections of atropine ( 0 . 05 mg/kg , intramuscularly ) , meloxicam ( 0 . 2 mg/kg , intravenously ) , and ranitidine ( 0 . 05 mg/kg , intravenously ) . The anaesthetized animals were placed in an MRI-compatible frame ( Crist Instruments ) in a sphinx position and placed in a horizontal 3 T MRI scanner with a full-size bore . Scanning commenced ~2 hr after induction , when the clinical peak of ketamine had passed . Anaesthesia was maintained , in accordance with veterinary recommendation , using the lowest possible concentration of isoflurane to ensure that macaques were anaesthetized . The depth of anaesthesia was assessed and monitored using physiological parameters ( heart rate and blood pressure , as well as clinical checks before the scan for muscle relaxation ) . During the acquisition of the functional data the expired isoflurane concentration was in the range 0 . 6–0 . 8% . Isoflurane was selected for the scans as it was previously demonstrated to preserve rs-fMRI networks ( Mars et al . , 2013; Mars et al . , 2011; Neubert et al . , 2015; Neubert et al . , 2014; O'Reilly et al . , 2013; Sallet et al . , 2013; Vincent et al . , 2007 ) . Macaques were maintained with intermittent positive pressure ventilation to ensure a constant respiration rate during the functional scan , and respiration rate , inspired and expired CO2 , and inspired and expired isoflurane concentration were monitored and recorded using VitalMonitor software ( Vetronic Services Ltd . ) . Core temperature and SpO2 were also constantly monitored throughout the scan . A four-channel phased-array coil was used for data acquisition ( Dr . H . Kolster , Windmiller Kolster Scientific , Fresno , CA , USA ) . FMRI data were collected in experiments 1–3 . In each session whole-brain BOLD fMRI data were collectedfor 3 runs of approximately 26 min each , using the following parameters: 36 axial slices; in-plane resolution , 2 × 2 mm; slice thickness , 2 mm; no slice gap; TR , 2000 ms; TE , 19 ms; 800 volumes per run . A minimum period of 10 days elapsed between sessions . T1-weighted structural MRI scans were collected in experiments 1–3 . A structural scan ( average over up to three T1w images acquired in the same session ) was acquired for each macaque in the same session , using a T1 weighted magnetization-prepared rapid-acquisition gradient echo sequence ( 0 . 5 × 0 . 5×0 . 5 mm voxel resolution ) . The pre-processing and analysis of the MRI data was designed to follow the HCP Minimal Processing Pipeline ( Glasser et al . , 2013 ) , using tools of FSL ( https://fsl . fmrib . ox . ac . uk/fsl/fslwiki; RRID:SCR_002823 ) , HCP Workbench ( https://www . humanconnectome . org/software/connectome-workbench; RRID:SCR_008750 ) , and the Magnetic Resonance Comparative Anatomy Toolbox ( MrCat; https://github . com/neuroecology/MrCat; copy archived at https://github . com/elifesciences-publications/MrCat; Verhagen , 2019 ) . The T1w images were processed in an iterative fashion cycling through brain-extraction , RF bias-field correction , and linear and non-linear template registration to the Macaca mulatta F99 atlas ( Van Essen and Dierker , 2007 ) . The initial skull stripping was performed using a multi-seeded implementation of BET ( Smith , 2002 ) optimized for macaque brains , while subsequent brain extraction was based on a high-fidelity template registered to the F99 macaque space . The RF bias-field was estimated and corrected using a robust implementation of FAST ( Zhang et al . , 2001 ) . Linear and non-linear registration to F99 space was achieved using FLIRT ( Jenkinson et al . , 2002; Jenkinson and Smith , 2001 ) and FNIRT ( Andersson et al . , 2007; Jenkinson et al . , 2012 ) with configurations adjusted to reflect macaque rather than human brain characteristics . The application of a robust and macaque-optimised version of FAST also resulted inestimated compartments for grey matter , white matter , and meninges with cerebral spinal fluid . For each compartment a posterior-probability map was created by integrating a set of prior probability maps based on 112 Macaca mullata individuals ( McLaren et al . , 2009 ) with the dataset-specific evidence provided by FAST . Segmentation of subcortical structures was achieved by registration to the D99 atlas ( Reveley et al . , 2017 ) . The first 5 volumes of the functional EPI datasets were discarded to ensure a steady RF excitation state . EPI timeseries were motion corrected using MCFLIRT . Given that the animals were anaesthetized and their heads were held in a steady position , any apparent image motion , if present at all , is caused by changes to the B0 field and temperature , rather than by head motion . Accordingly , the parameter estimates from MCFLIRT can be considered to be ‘B0-confound parameters’ instead . Each timeseries was checked rigorously for spikes and other artefacts , both visually and using automated algorithms; where applicable slices with spikes were linearly interpolated based on temporally neighbouring slices . Brain extraction , bias-correction , and registration was achieved for the functional EPI datasets in an iterative manner , similar to the pre-processing of the structural images with the only difference that the mean of each functional dataset was registered to its corresponding T1w image using rigid-body boundary-based registration ( FLIRT ) . EPI signal noise was reduced both in the frequency and temporal domain . The functional timeseries were high-pass filtered with a frequency cut-off at 2000 s . Temporally cyclical noise , for example originating from the respiration apparatus , was removed using band-stop filters set dynamically to noise peaks in the frequency domain of the first three principal components of the timeseries . To account for remaining global signal confounds we considered the signal timeseries in white matter ( WM ) and meningeal compartments . Specifically , the WM + meningeal confound timeseries was described by the mean time course and the first five subsequent principal components of the combined WM and meningeal compartment ( considering only voxels with a high posterior probability of belonging to the WM or meningeal compartment , obtained in the T1w image using FAST ) . The principal components of the WM + meningeal signal were estimated using a singular value decomposition approach . The B0 confound parameter estimates obtained from MCFLIRT were expanded as a second degree Volterra series to capture both linear and non-linear B0 effects . Together the WM + meningeal and expanded B0 confound parameters were regressed out of the BOLD signal for each voxel . In a separate analysis ( Figure 6 and Figure 6—figure supplement 1 ) , to assess the contribution of the meningeal compartment signal we repeated the identical procedure as above , with the only difference that the mean and principal components were extracted from signal in the WM compartment alone , excluding the meningeal compartment . Following this confound cleaning step , the timeseries were low-pass filtered with a cut-off at 10 s . The cleaned and filtered signal was projected from the conventional volumetric representation ( 2 mm voxels ) to the F99 cortical surface ( ~1 . 4 mm spaced vertices ) , while maintaining the subcortical volumetric structures . The data were spatially smoothed using a 3 mm FWHM gaussian kernel , while considering the folding of the cortex and the anatomical boundaries of the subcortical structures . Lastly , the data timeseries were demeaned to prepare for functional connectivity analyses . To represent subject effects , the timeseries from the three runs were concatenated to create a single timeseries of 3×800-5=2385 volumes per animal per intervention ( control , SMA TUS , FPC TUS ) . To represent group effects the run-concatenated timeseries of all animals were combined using a group-PCA approach resulting in a series of 200 volumes representing the principal components ( Smith et al . , 2014 ) . We report on the whole-brain functional coupling of the stimulation sites in the control state and after TUS , adopting a seed-based correlation analysis approach . Effects of TUS are quantified based on a limited set of regions , a ‘fingerprint’ , whose strength of interconnection with macaque SMA or FPC is known from anatomical tracing studies ( Bates and Goldman-Rakic , 1993; Dum and Strick , 2005; Geyer et al . , 2000; Lu et al . , 1994; Petrides and Pandya , 2007; Strick et al . , 1998 ) and from studies of fMRI activity coupling under anaesthesia ( Neubert et al . , 2015; Neubert et al . , 2014; Sallet et al . , 2013 ) . For example , SMA is strongly interconnected with core nodes in the sensorimotor network , including M1 , SPL , and MCC . The unique profile of interconnections characterizing a functional area is defined by a pattern of both strong and limited interconnections ( Passingham et al . , 2002 ) . Accordingly , we selected a set of regions based on either distinctively strong or distinctively limited interconnection and fMRI coupling with SMA or FPC . In fact , as a result of the complementary patterns of connections defining SMA and FPC ( Petrides and Pandya , 2007 ) , areas that were included based on their known strong connection and coupling with FPC ( 9m , 9-46d , PCC , IPLc , midSTS ) were also included because of their known limited connections and coupling with SMA . Furthermore , to characterize the effect of TUS targeted at FPC on adjacent prefrontal cortex , we also included additional ventromedial and orbital prefrontal regions with which FPC shared less strong connections . To ensure a balanced distribution of fingerprint targets across the brain and to ensure that both SMA and FPC fingerprints were comprised of the same number of regions we consider areas 11m and aSTG in the SMA fingerprint , both regions with which the SMA is weakly interconnected . Lastly , to test for specificity of the TUS effects , we included SMA as a target for the FPC fingerprint and FPC as a target for the SMA fingerprint . To construct regions-of-interest ( ROI ) for SMA and FPC , circles of 4 mm radius were drawn on the cortical surface around the point closest to the average stimulation coordinate ( Figure 2 ) , in both the left and the right hemisphere . The same procedure was used to define other bilateral cortical regions of interest , based on literature coordinates ( Mars et al . , 2011; Neubert et al . , 2015; Neubert et al . , 2014; Sallet et al . , 2013 ) , to serve as seeds for connectivity analyses ( A1 , Figure 5g–i; POp , Figure 6 , Figure 6—figure supplement 1 ) or targets for the fingerprint analyses ( Figure 5 ) . Apart from the stimulation sites and primary auditory cortex , all ROIs were selected based on their known anatomical connectivity , of relevance because they are known to be distinctively strongly or distinctively weakly connected with the stimulation sites . Coupling between the activity of each region of interest and the rest of the brain was estimated by calculating the correlation coefficient between each point in the ROI and all other data points ( Fisher’s z: inverse hyperbolic tangent of Pearson’s r , bounded at [−2 2] ) . The resulting ‘connectivity-maps’ were averaged across all vertices/voxels in the ROI , and subsequently averaged across hemispheres . Accordingly , the final maps represent the average coupling of a bilateral ROI with the rest of the brain . The fingerprints are obtained by extracting the average coupling with each target ROI and averaging across the two hemispheres . The strength of self-connections within SMA and FPC ROIs was estimated in the same way as the coupling of either SMA or FPC with any remote ROI was determined . Namely , for each point in the ROI the Fisher’s z-transformed correlation between this point’s timeseries and that of each of the other points in the ROI was calculated . Subsequently , for each point the resultant z-values were averaged , describing local coupling at that point , and averaged across the whole ROI to obtain a single estimate of self-coupling per ROI ( Figure 5c ) . To assess the contribution of the meningeal compartment signal , we quantified the effect of TUS for each point on the cortical surface in data that was not corrected for the meningeal signal ( see Macaque rs-fMRI pre-processing ) . For each point , we calculated it’s coupling with the rest of the brain , sorted all Fisher’s z-values in ascending order , and extracted the z-value at the 98th-percentile . Z-values at this level describe relatively strong coupling with the seed point and are often observed in the vicinity of the seed region or other strongly connected regions ( depicted in bright yellow in Figure 6 ) , ensuring that irrelevant connections are ignored . As such , this simple statistic is well-suited to capture any main effects of TUS on coupling strength , including those observed when comparing panels ( b ) and ( e ) in Figure 6 . This statistic allows us to quantify the effect of TUS in a single value for each point on the cortical surface , and to directly contrast the values obtained in the control state with those obtained after TUS ( Figure 6f , Figure 6—figure supplement 1f , i ) . We described the signal variance in the GM and meningeal compartments , or more specifically , the variance explained by the first five principal components ( Figure 6c , Figure 6—figure supplement 1 panel c ) , using the same approach that was used to decompose signal in WM or WM + meningeal compartments in the EPI timeseries cleaning step , as described above . We defined the explained variance as the sum of the first five eigenvalues of the covariance matrix of the compartment signals divided by the sum of all eigenvalues . All suitable animals available at the time of experimentation took part in this study . Accordingly , there was no pre-selection nor restriction for group allocation . No data was excluded for analysis . Sample sizes could not be predetermined statistically in the absence of a prior literature reporting relevant expected effect sizes; instead we adopted sample sizes similar to those reported in previous publications detailing ( interventional ) macaque fMRI studies ( Chau et al . , 2015; O'Reilly et al . , 2013; Papageorgiou et al . , 2017 ) . Data collection and analysis were not performed blind to the conditions of the experiments . We used a group PCA approach to define the centre of group rs-fMRI effects ( see Macaque rs-fMRI analyses ) . For all other effects the centre of the group is defined as the mean of the individual subject effects . To report on the full extent of the TUS effects and the simple effects that drive any observed differences we reproduce the whole-brain functional connectivity maps unthresholded for the control , SMA TUS , and FPC TUS conditions separately . In order to make a statistical comparison of the functional coupling of the SMA and FPC in the control and TUS conditions it is problematic to compare coupling at each and every other point in the brain because there is a risk of false positive effects if multiple comparisons are made . Given the limited sample sizes possible with non-human primate experiments , however , there is a risk of false negative results if stringent correction for multiple comparisons is undertaken at the whole-brain level . Indeed , here we avoid these pitfalls and make inference on a single statistic that combines information from a planned limited set of regions defined by coordinates from previous studies ( Mars et al . , 2011; Neubert et al . , 2015; Neubert et al . , 2014; Sallet et al . , 2013 ) . Importantly , rather than examining activity coupling between the seed area and each of the fingerprint ROIs in turn and risking potential false positive results , we compared the overall pattern of coupling using the method devised by Mars and colleagues ( Mars et al . , 2016 ) ; non-parametric permutation tests were performed on cosine similarity metrics summarizing pairs of fingerprints ( SMA TUS versus control and FPC TUS versus control ) . Each connectional fingerprint can be represented as a multidimensional vector , with the number of dimensions corresponding to the number of target ROIs in the fingerprint . The cosine similarity metric quantifies the angle between two of such multi-dimensional vectors . As such , it takes into account the shape of the fingerprint irrespective of its mean amplitude and results in a single metric per pair of fingerprints , negating the necessity for correcting for multiple comparisons across fingerprint targets . To assess the likelihood of observing this cosine similarity metric under the null-hypothesis of no difference between conditions , it needs to be compared against a distribution of metrics obtained from a random sample . To avoid non-compliance to the requirements for a conventional parametric approach ( e . g . independence of sampling and adherence to a Gaussian normal distribution ) we applied a non-parametric permutation approach , in which the parent distribution is empirically drawn by repeatedly ( randomly ) shuffling the group-membership of data points and re-calculating the metric based on the shuffled labelling ( e . g . data points acquired in the control state might be randomly assigned to the intervention condition , and vice versa ) . In this way the empirical distribution under the null-hypothesis does not rely on assumptions about the shape of the distribution but will acknowledge dependencies between target ROIs in the fingerprint; as such this approach will avoid inflation of type I error ( Maris and Oostenveld , 2007; Winkler et al . , 2014 ) . The relatively small number of TUS sites , animals , and fMRI runs allowed us to exhaustively test all possible permutations ( 24309 ) to obtain the true probability of rejecting the null hypothesis . All other statistical inferences were drawn in the context of generalized linear mixed-effects ( GLME ) models that considered the intercept and factorial design ( including interactions where appropriate ) as fixed effects and the intercept and slope grouped per animal as random effects with possible correlation between them ( as implemented in MATLAB , Mathworks , Natick , USA ) . The models were assumed to adhere to a normal distribution of the data ( not formally tested ) and were fitted using Maximum-Pseudo-Likelihood estimation methods where the covariance of the random effects was approximated using Cholesky parameterization . Statistical significance was set at α = 0 . 05 , two-tailed , and estimated using conventional analyses of variance ( ANOVA ) . For all parametric tests we report the test statistic ( F ) , probability estimate ( p ) , effect size ( Cohen’s d ) , and the lower and upper limits of the 95% confidence interval ( CI ) . For all plots , the central tendency across individual animals is derived directly from the group-PCA approach ( Smith et al . , 2014 ) or described as the mean across all animals ( n = 3 ) . Dispersion is described as the standard-error of the mean ( SE ) . In experiment 4 , prior to histological examination , animals were anaesthetized with sodium pentorbarbitone and perfused with 90% saline and 10% formalin . A post-mortem examination of the surface of the brain was conducted prior to the brain extraction . The brains were then removed and placed in 10% sucrose formalin . The brains were blocked in the coronal plane at the level of the lunate sulcus . Each brain was cut in 50 μm coronal sections . Every tenth section , two adjacent sections were retained for analysis whereby one was stained with Cresyl Violet ( Nissl body staining ) and the other with Haemotoxylin and Eosin ( H and E staining ) , in line with previous studies ( Dallapiazza et al . , 2018; Lee et al . , 2016b ) .	Ultrasound is well known for making visible what is hidden , for example , when giving parents a glimpse of their child before birth . But researchers are now using these high-frequency sound waves – beyond the range of human hearing – for a wholly different purpose: to manipulate the activity of the brain . Conventional brain stimulation techniques use electric currents or magnetic fields to alter brain activity . These techniques , however , have limitations . They can only reach the surface of the brain and are not particularly precise . By contrast , beams of ultrasound can be focused at a millimetre scale , even deep within the brain . Ultrasound thus has the potential to provide new insights into how the brain works . Most studies of ultrasound stimulation have looked at what happens to the brain during the stimulation itself . But could ultrasound also induce longer-lasting changes in brain activity ? Changes that persist after the stimulation has ended would be valuable for research . They would also make it more likely that we could use ultrasound to treat brain disorders by changing brain activity . Verhagen , Gallea et al . used a brain scanner to measure brain activity in macaque monkeys after ultrasound stimulation . The results showed that 40 seconds of repetitive ultrasound changed brain activity for up to two hours . Ultrasound caused the stimulated brain area to interact more selectively with the rest of the brain . Notably , only the stimulated area changed its activity in this way . This helps rule out the possibility that the changes reflect non-specific effects . If the monkeys had been able to hear the ultrasound , for example , it would have changed the activity of the parts of the brain related to hearing . Most important of all , the changes were reversible and did not harm the brain . The results of Verhagen , Gallea et al . show that repetitive ultrasound can induce long-lasting alterations in brain activity . It can target areas deep within the brain , including those that are out of reach with other techniques . If this procedure also shows longer-lasting effects in people , it could yield valuable insights into the links between brain and behaviour . It could also help us develop new treatments for neurological and psychiatric disorders .	[ "Abstract", "Introduction", "Results", "Discussion", "Materials", "and", "methods" ]	[ "neuroscience" ]	2019	Offline impact of transcranial focused ultrasound on cortical activation in primates

Subsets and Splits

No community queries yet

The top public SQL queries from the community will appear here once available.